1 /* 2 * linux/kernel/exit.c 3 * 4 * Copyright (C) 1991, 1992 Linus Torvalds 5 */ 6 7 #include <linux/mm.h> 8 #include <linux/slab.h> 9 #include <linux/interrupt.h> 10 #include <linux/module.h> 11 #include <linux/capability.h> 12 #include <linux/completion.h> 13 #include <linux/personality.h> 14 #include <linux/tty.h> 15 #include <linux/iocontext.h> 16 #include <linux/key.h> 17 #include <linux/security.h> 18 #include <linux/cpu.h> 19 #include <linux/acct.h> 20 #include <linux/tsacct_kern.h> 21 #include <linux/file.h> 22 #include <linux/fdtable.h> 23 #include <linux/freezer.h> 24 #include <linux/binfmts.h> 25 #include <linux/nsproxy.h> 26 #include <linux/pid_namespace.h> 27 #include <linux/ptrace.h> 28 #include <linux/profile.h> 29 #include <linux/mount.h> 30 #include <linux/proc_fs.h> 31 #include <linux/kthread.h> 32 #include <linux/mempolicy.h> 33 #include <linux/taskstats_kern.h> 34 #include <linux/delayacct.h> 35 #include <linux/cgroup.h> 36 #include <linux/syscalls.h> 37 #include <linux/signal.h> 38 #include <linux/posix-timers.h> 39 #include <linux/cn_proc.h> 40 #include <linux/mutex.h> 41 #include <linux/futex.h> 42 #include <linux/pipe_fs_i.h> 43 #include <linux/audit.h> /* for audit_free() */ 44 #include <linux/resource.h> 45 #include <linux/blkdev.h> 46 #include <linux/task_io_accounting_ops.h> 47 #include <linux/tracehook.h> 48 #include <linux/fs_struct.h> 49 #include <linux/init_task.h> 50 #include <linux/perf_event.h> 51 #include <trace/events/sched.h> 52 #include <linux/hw_breakpoint.h> 53 #include <linux/oom.h> 54 #include <linux/writeback.h> 55 #include <linux/shm.h> 56 #include <linux/kcov.h> 57 #include <linux/random.h> 58 #include <linux/rcuwait.h> 59 60 #include <linux/uaccess.h> 61 #include <asm/unistd.h> 62 #include <asm/pgtable.h> 63 #include <asm/mmu_context.h> 64 65 static void __unhash_process(struct task_struct *p, bool group_dead) 66 { 67 nr_threads--; 68 detach_pid(p, PIDTYPE_PID); 69 if (group_dead) { 70 detach_pid(p, PIDTYPE_PGID); 71 detach_pid(p, PIDTYPE_SID); 72 73 list_del_rcu(&p->tasks); 74 list_del_init(&p->sibling); 75 __this_cpu_dec(process_counts); 76 } 77 list_del_rcu(&p->thread_group); 78 list_del_rcu(&p->thread_node); 79 } 80 81 /* 82 * This function expects the tasklist_lock write-locked. 83 */ 84 static void __exit_signal(struct task_struct *tsk) 85 { 86 struct signal_struct *sig = tsk->signal; 87 bool group_dead = thread_group_leader(tsk); 88 struct sighand_struct *sighand; 89 struct tty_struct *uninitialized_var(tty); 90 cputime_t utime, stime; 91 92 sighand = rcu_dereference_check(tsk->sighand, 93 lockdep_tasklist_lock_is_held()); 94 spin_lock(&sighand->siglock); 95 96 #ifdef CONFIG_POSIX_TIMERS 97 posix_cpu_timers_exit(tsk); 98 if (group_dead) { 99 posix_cpu_timers_exit_group(tsk); 100 } else { 101 /* 102 * This can only happen if the caller is de_thread(). 103 * FIXME: this is the temporary hack, we should teach 104 * posix-cpu-timers to handle this case correctly. 105 */ 106 if (unlikely(has_group_leader_pid(tsk))) 107 posix_cpu_timers_exit_group(tsk); 108 } 109 #endif 110 111 if (group_dead) { 112 tty = sig->tty; 113 sig->tty = NULL; 114 } else { 115 /* 116 * If there is any task waiting for the group exit 117 * then notify it: 118 */ 119 if (sig->notify_count > 0 && !--sig->notify_count) 120 wake_up_process(sig->group_exit_task); 121 122 if (tsk == sig->curr_target) 123 sig->curr_target = next_thread(tsk); 124 } 125 126 add_device_randomness((const void*) &tsk->se.sum_exec_runtime, 127 sizeof(unsigned long long)); 128 129 /* 130 * Accumulate here the counters for all threads as they die. We could 131 * skip the group leader because it is the last user of signal_struct, 132 * but we want to avoid the race with thread_group_cputime() which can 133 * see the empty ->thread_head list. 134 */ 135 task_cputime(tsk, &utime, &stime); 136 write_seqlock(&sig->stats_lock); 137 sig->utime += utime; 138 sig->stime += stime; 139 sig->gtime += task_gtime(tsk); 140 sig->min_flt += tsk->min_flt; 141 sig->maj_flt += tsk->maj_flt; 142 sig->nvcsw += tsk->nvcsw; 143 sig->nivcsw += tsk->nivcsw; 144 sig->inblock += task_io_get_inblock(tsk); 145 sig->oublock += task_io_get_oublock(tsk); 146 task_io_accounting_add(&sig->ioac, &tsk->ioac); 147 sig->sum_sched_runtime += tsk->se.sum_exec_runtime; 148 sig->nr_threads--; 149 __unhash_process(tsk, group_dead); 150 write_sequnlock(&sig->stats_lock); 151 152 /* 153 * Do this under ->siglock, we can race with another thread 154 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals. 155 */ 156 flush_sigqueue(&tsk->pending); 157 tsk->sighand = NULL; 158 spin_unlock(&sighand->siglock); 159 160 __cleanup_sighand(sighand); 161 clear_tsk_thread_flag(tsk, TIF_SIGPENDING); 162 if (group_dead) { 163 flush_sigqueue(&sig->shared_pending); 164 tty_kref_put(tty); 165 } 166 } 167 168 static void delayed_put_task_struct(struct rcu_head *rhp) 169 { 170 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu); 171 172 perf_event_delayed_put(tsk); 173 trace_sched_process_free(tsk); 174 put_task_struct(tsk); 175 } 176 177 178 void release_task(struct task_struct *p) 179 { 180 struct task_struct *leader; 181 int zap_leader; 182 repeat: 183 /* don't need to get the RCU readlock here - the process is dead and 184 * can't be modifying its own credentials. But shut RCU-lockdep up */ 185 rcu_read_lock(); 186 atomic_dec(&__task_cred(p)->user->processes); 187 rcu_read_unlock(); 188 189 proc_flush_task(p); 190 191 write_lock_irq(&tasklist_lock); 192 ptrace_release_task(p); 193 __exit_signal(p); 194 195 /* 196 * If we are the last non-leader member of the thread 197 * group, and the leader is zombie, then notify the 198 * group leader's parent process. (if it wants notification.) 199 */ 200 zap_leader = 0; 201 leader = p->group_leader; 202 if (leader != p && thread_group_empty(leader) 203 && leader->exit_state == EXIT_ZOMBIE) { 204 /* 205 * If we were the last child thread and the leader has 206 * exited already, and the leader's parent ignores SIGCHLD, 207 * then we are the one who should release the leader. 208 */ 209 zap_leader = do_notify_parent(leader, leader->exit_signal); 210 if (zap_leader) 211 leader->exit_state = EXIT_DEAD; 212 } 213 214 write_unlock_irq(&tasklist_lock); 215 release_thread(p); 216 call_rcu(&p->rcu, delayed_put_task_struct); 217 218 p = leader; 219 if (unlikely(zap_leader)) 220 goto repeat; 221 } 222 223 /* 224 * Note that if this function returns a valid task_struct pointer (!NULL) 225 * task->usage must remain >0 for the duration of the RCU critical section. 226 */ 227 struct task_struct *task_rcu_dereference(struct task_struct **ptask) 228 { 229 struct sighand_struct *sighand; 230 struct task_struct *task; 231 232 /* 233 * We need to verify that release_task() was not called and thus 234 * delayed_put_task_struct() can't run and drop the last reference 235 * before rcu_read_unlock(). We check task->sighand != NULL, 236 * but we can read the already freed and reused memory. 237 */ 238 retry: 239 task = rcu_dereference(*ptask); 240 if (!task) 241 return NULL; 242 243 probe_kernel_address(&task->sighand, sighand); 244 245 /* 246 * Pairs with atomic_dec_and_test() in put_task_struct(). If this task 247 * was already freed we can not miss the preceding update of this 248 * pointer. 249 */ 250 smp_rmb(); 251 if (unlikely(task != READ_ONCE(*ptask))) 252 goto retry; 253 254 /* 255 * We've re-checked that "task == *ptask", now we have two different 256 * cases: 257 * 258 * 1. This is actually the same task/task_struct. In this case 259 * sighand != NULL tells us it is still alive. 260 * 261 * 2. This is another task which got the same memory for task_struct. 262 * We can't know this of course, and we can not trust 263 * sighand != NULL. 264 * 265 * In this case we actually return a random value, but this is 266 * correct. 267 * 268 * If we return NULL - we can pretend that we actually noticed that 269 * *ptask was updated when the previous task has exited. Or pretend 270 * that probe_slab_address(&sighand) reads NULL. 271 * 272 * If we return the new task (because sighand is not NULL for any 273 * reason) - this is fine too. This (new) task can't go away before 274 * another gp pass. 275 * 276 * And note: We could even eliminate the false positive if re-read 277 * task->sighand once again to avoid the falsely NULL. But this case 278 * is very unlikely so we don't care. 279 */ 280 if (!sighand) 281 return NULL; 282 283 return task; 284 } 285 286 void rcuwait_wake_up(struct rcuwait *w) 287 { 288 struct task_struct *task; 289 290 rcu_read_lock(); 291 292 /* 293 * Order condition vs @task, such that everything prior to the load 294 * of @task is visible. This is the condition as to why the user called 295 * rcuwait_trywake() in the first place. Pairs with set_current_state() 296 * barrier (A) in rcuwait_wait_event(). 297 * 298 * WAIT WAKE 299 * [S] tsk = current [S] cond = true 300 * MB (A) MB (B) 301 * [L] cond [L] tsk 302 */ 303 smp_rmb(); /* (B) */ 304 305 /* 306 * Avoid using task_rcu_dereference() magic as long as we are careful, 307 * see comment in rcuwait_wait_event() regarding ->exit_state. 308 */ 309 task = rcu_dereference(w->task); 310 if (task) 311 wake_up_process(task); 312 rcu_read_unlock(); 313 } 314 315 struct task_struct *try_get_task_struct(struct task_struct **ptask) 316 { 317 struct task_struct *task; 318 319 rcu_read_lock(); 320 task = task_rcu_dereference(ptask); 321 if (task) 322 get_task_struct(task); 323 rcu_read_unlock(); 324 325 return task; 326 } 327 328 /* 329 * Determine if a process group is "orphaned", according to the POSIX 330 * definition in 2.2.2.52. Orphaned process groups are not to be affected 331 * by terminal-generated stop signals. Newly orphaned process groups are 332 * to receive a SIGHUP and a SIGCONT. 333 * 334 * "I ask you, have you ever known what it is to be an orphan?" 335 */ 336 static int will_become_orphaned_pgrp(struct pid *pgrp, 337 struct task_struct *ignored_task) 338 { 339 struct task_struct *p; 340 341 do_each_pid_task(pgrp, PIDTYPE_PGID, p) { 342 if ((p == ignored_task) || 343 (p->exit_state && thread_group_empty(p)) || 344 is_global_init(p->real_parent)) 345 continue; 346 347 if (task_pgrp(p->real_parent) != pgrp && 348 task_session(p->real_parent) == task_session(p)) 349 return 0; 350 } while_each_pid_task(pgrp, PIDTYPE_PGID, p); 351 352 return 1; 353 } 354 355 int is_current_pgrp_orphaned(void) 356 { 357 int retval; 358 359 read_lock(&tasklist_lock); 360 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL); 361 read_unlock(&tasklist_lock); 362 363 return retval; 364 } 365 366 static bool has_stopped_jobs(struct pid *pgrp) 367 { 368 struct task_struct *p; 369 370 do_each_pid_task(pgrp, PIDTYPE_PGID, p) { 371 if (p->signal->flags & SIGNAL_STOP_STOPPED) 372 return true; 373 } while_each_pid_task(pgrp, PIDTYPE_PGID, p); 374 375 return false; 376 } 377 378 /* 379 * Check to see if any process groups have become orphaned as 380 * a result of our exiting, and if they have any stopped jobs, 381 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2) 382 */ 383 static void 384 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent) 385 { 386 struct pid *pgrp = task_pgrp(tsk); 387 struct task_struct *ignored_task = tsk; 388 389 if (!parent) 390 /* exit: our father is in a different pgrp than 391 * we are and we were the only connection outside. 392 */ 393 parent = tsk->real_parent; 394 else 395 /* reparent: our child is in a different pgrp than 396 * we are, and it was the only connection outside. 397 */ 398 ignored_task = NULL; 399 400 if (task_pgrp(parent) != pgrp && 401 task_session(parent) == task_session(tsk) && 402 will_become_orphaned_pgrp(pgrp, ignored_task) && 403 has_stopped_jobs(pgrp)) { 404 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp); 405 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp); 406 } 407 } 408 409 #ifdef CONFIG_MEMCG 410 /* 411 * A task is exiting. If it owned this mm, find a new owner for the mm. 412 */ 413 void mm_update_next_owner(struct mm_struct *mm) 414 { 415 struct task_struct *c, *g, *p = current; 416 417 retry: 418 /* 419 * If the exiting or execing task is not the owner, it's 420 * someone else's problem. 421 */ 422 if (mm->owner != p) 423 return; 424 /* 425 * The current owner is exiting/execing and there are no other 426 * candidates. Do not leave the mm pointing to a possibly 427 * freed task structure. 428 */ 429 if (atomic_read(&mm->mm_users) <= 1) { 430 mm->owner = NULL; 431 return; 432 } 433 434 read_lock(&tasklist_lock); 435 /* 436 * Search in the children 437 */ 438 list_for_each_entry(c, &p->children, sibling) { 439 if (c->mm == mm) 440 goto assign_new_owner; 441 } 442 443 /* 444 * Search in the siblings 445 */ 446 list_for_each_entry(c, &p->real_parent->children, sibling) { 447 if (c->mm == mm) 448 goto assign_new_owner; 449 } 450 451 /* 452 * Search through everything else, we should not get here often. 453 */ 454 for_each_process(g) { 455 if (g->flags & PF_KTHREAD) 456 continue; 457 for_each_thread(g, c) { 458 if (c->mm == mm) 459 goto assign_new_owner; 460 if (c->mm) 461 break; 462 } 463 } 464 read_unlock(&tasklist_lock); 465 /* 466 * We found no owner yet mm_users > 1: this implies that we are 467 * most likely racing with swapoff (try_to_unuse()) or /proc or 468 * ptrace or page migration (get_task_mm()). Mark owner as NULL. 469 */ 470 mm->owner = NULL; 471 return; 472 473 assign_new_owner: 474 BUG_ON(c == p); 475 get_task_struct(c); 476 /* 477 * The task_lock protects c->mm from changing. 478 * We always want mm->owner->mm == mm 479 */ 480 task_lock(c); 481 /* 482 * Delay read_unlock() till we have the task_lock() 483 * to ensure that c does not slip away underneath us 484 */ 485 read_unlock(&tasklist_lock); 486 if (c->mm != mm) { 487 task_unlock(c); 488 put_task_struct(c); 489 goto retry; 490 } 491 mm->owner = c; 492 task_unlock(c); 493 put_task_struct(c); 494 } 495 #endif /* CONFIG_MEMCG */ 496 497 /* 498 * Turn us into a lazy TLB process if we 499 * aren't already.. 500 */ 501 static void exit_mm(void) 502 { 503 struct mm_struct *mm = current->mm; 504 struct core_state *core_state; 505 506 mm_release(current, mm); 507 if (!mm) 508 return; 509 sync_mm_rss(mm); 510 /* 511 * Serialize with any possible pending coredump. 512 * We must hold mmap_sem around checking core_state 513 * and clearing tsk->mm. The core-inducing thread 514 * will increment ->nr_threads for each thread in the 515 * group with ->mm != NULL. 516 */ 517 down_read(&mm->mmap_sem); 518 core_state = mm->core_state; 519 if (core_state) { 520 struct core_thread self; 521 522 up_read(&mm->mmap_sem); 523 524 self.task = current; 525 self.next = xchg(&core_state->dumper.next, &self); 526 /* 527 * Implies mb(), the result of xchg() must be visible 528 * to core_state->dumper. 529 */ 530 if (atomic_dec_and_test(&core_state->nr_threads)) 531 complete(&core_state->startup); 532 533 for (;;) { 534 set_current_state(TASK_UNINTERRUPTIBLE); 535 if (!self.task) /* see coredump_finish() */ 536 break; 537 freezable_schedule(); 538 } 539 __set_current_state(TASK_RUNNING); 540 down_read(&mm->mmap_sem); 541 } 542 atomic_inc(&mm->mm_count); 543 BUG_ON(mm != current->active_mm); 544 /* more a memory barrier than a real lock */ 545 task_lock(current); 546 current->mm = NULL; 547 up_read(&mm->mmap_sem); 548 enter_lazy_tlb(mm, current); 549 task_unlock(current); 550 mm_update_next_owner(mm); 551 mmput(mm); 552 if (test_thread_flag(TIF_MEMDIE)) 553 exit_oom_victim(); 554 } 555 556 static struct task_struct *find_alive_thread(struct task_struct *p) 557 { 558 struct task_struct *t; 559 560 for_each_thread(p, t) { 561 if (!(t->flags & PF_EXITING)) 562 return t; 563 } 564 return NULL; 565 } 566 567 static struct task_struct *find_child_reaper(struct task_struct *father) 568 __releases(&tasklist_lock) 569 __acquires(&tasklist_lock) 570 { 571 struct pid_namespace *pid_ns = task_active_pid_ns(father); 572 struct task_struct *reaper = pid_ns->child_reaper; 573 574 if (likely(reaper != father)) 575 return reaper; 576 577 reaper = find_alive_thread(father); 578 if (reaper) { 579 pid_ns->child_reaper = reaper; 580 return reaper; 581 } 582 583 write_unlock_irq(&tasklist_lock); 584 if (unlikely(pid_ns == &init_pid_ns)) { 585 panic("Attempted to kill init! exitcode=0x%08x\n", 586 father->signal->group_exit_code ?: father->exit_code); 587 } 588 zap_pid_ns_processes(pid_ns); 589 write_lock_irq(&tasklist_lock); 590 591 return father; 592 } 593 594 /* 595 * When we die, we re-parent all our children, and try to: 596 * 1. give them to another thread in our thread group, if such a member exists 597 * 2. give it to the first ancestor process which prctl'd itself as a 598 * child_subreaper for its children (like a service manager) 599 * 3. give it to the init process (PID 1) in our pid namespace 600 */ 601 static struct task_struct *find_new_reaper(struct task_struct *father, 602 struct task_struct *child_reaper) 603 { 604 struct task_struct *thread, *reaper; 605 606 thread = find_alive_thread(father); 607 if (thread) 608 return thread; 609 610 if (father->signal->has_child_subreaper) { 611 /* 612 * Find the first ->is_child_subreaper ancestor in our pid_ns. 613 * We start from father to ensure we can not look into another 614 * namespace, this is safe because all its threads are dead. 615 */ 616 for (reaper = father; 617 !same_thread_group(reaper, child_reaper); 618 reaper = reaper->real_parent) { 619 /* call_usermodehelper() descendants need this check */ 620 if (reaper == &init_task) 621 break; 622 if (!reaper->signal->is_child_subreaper) 623 continue; 624 thread = find_alive_thread(reaper); 625 if (thread) 626 return thread; 627 } 628 } 629 630 return child_reaper; 631 } 632 633 /* 634 * Any that need to be release_task'd are put on the @dead list. 635 */ 636 static void reparent_leader(struct task_struct *father, struct task_struct *p, 637 struct list_head *dead) 638 { 639 if (unlikely(p->exit_state == EXIT_DEAD)) 640 return; 641 642 /* We don't want people slaying init. */ 643 p->exit_signal = SIGCHLD; 644 645 /* If it has exited notify the new parent about this child's death. */ 646 if (!p->ptrace && 647 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) { 648 if (do_notify_parent(p, p->exit_signal)) { 649 p->exit_state = EXIT_DEAD; 650 list_add(&p->ptrace_entry, dead); 651 } 652 } 653 654 kill_orphaned_pgrp(p, father); 655 } 656 657 /* 658 * This does two things: 659 * 660 * A. Make init inherit all the child processes 661 * B. Check to see if any process groups have become orphaned 662 * as a result of our exiting, and if they have any stopped 663 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2) 664 */ 665 static void forget_original_parent(struct task_struct *father, 666 struct list_head *dead) 667 { 668 struct task_struct *p, *t, *reaper; 669 670 if (unlikely(!list_empty(&father->ptraced))) 671 exit_ptrace(father, dead); 672 673 /* Can drop and reacquire tasklist_lock */ 674 reaper = find_child_reaper(father); 675 if (list_empty(&father->children)) 676 return; 677 678 reaper = find_new_reaper(father, reaper); 679 list_for_each_entry(p, &father->children, sibling) { 680 for_each_thread(p, t) { 681 t->real_parent = reaper; 682 BUG_ON((!t->ptrace) != (t->parent == father)); 683 if (likely(!t->ptrace)) 684 t->parent = t->real_parent; 685 if (t->pdeath_signal) 686 group_send_sig_info(t->pdeath_signal, 687 SEND_SIG_NOINFO, t); 688 } 689 /* 690 * If this is a threaded reparent there is no need to 691 * notify anyone anything has happened. 692 */ 693 if (!same_thread_group(reaper, father)) 694 reparent_leader(father, p, dead); 695 } 696 list_splice_tail_init(&father->children, &reaper->children); 697 } 698 699 /* 700 * Send signals to all our closest relatives so that they know 701 * to properly mourn us.. 702 */ 703 static void exit_notify(struct task_struct *tsk, int group_dead) 704 { 705 bool autoreap; 706 struct task_struct *p, *n; 707 LIST_HEAD(dead); 708 709 write_lock_irq(&tasklist_lock); 710 forget_original_parent(tsk, &dead); 711 712 if (group_dead) 713 kill_orphaned_pgrp(tsk->group_leader, NULL); 714 715 if (unlikely(tsk->ptrace)) { 716 int sig = thread_group_leader(tsk) && 717 thread_group_empty(tsk) && 718 !ptrace_reparented(tsk) ? 719 tsk->exit_signal : SIGCHLD; 720 autoreap = do_notify_parent(tsk, sig); 721 } else if (thread_group_leader(tsk)) { 722 autoreap = thread_group_empty(tsk) && 723 do_notify_parent(tsk, tsk->exit_signal); 724 } else { 725 autoreap = true; 726 } 727 728 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE; 729 if (tsk->exit_state == EXIT_DEAD) 730 list_add(&tsk->ptrace_entry, &dead); 731 732 /* mt-exec, de_thread() is waiting for group leader */ 733 if (unlikely(tsk->signal->notify_count < 0)) 734 wake_up_process(tsk->signal->group_exit_task); 735 write_unlock_irq(&tasklist_lock); 736 737 list_for_each_entry_safe(p, n, &dead, ptrace_entry) { 738 list_del_init(&p->ptrace_entry); 739 release_task(p); 740 } 741 } 742 743 #ifdef CONFIG_DEBUG_STACK_USAGE 744 static void check_stack_usage(void) 745 { 746 static DEFINE_SPINLOCK(low_water_lock); 747 static int lowest_to_date = THREAD_SIZE; 748 unsigned long free; 749 750 free = stack_not_used(current); 751 752 if (free >= lowest_to_date) 753 return; 754 755 spin_lock(&low_water_lock); 756 if (free < lowest_to_date) { 757 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n", 758 current->comm, task_pid_nr(current), free); 759 lowest_to_date = free; 760 } 761 spin_unlock(&low_water_lock); 762 } 763 #else 764 static inline void check_stack_usage(void) {} 765 #endif 766 767 void __noreturn do_exit(long code) 768 { 769 struct task_struct *tsk = current; 770 int group_dead; 771 TASKS_RCU(int tasks_rcu_i); 772 773 profile_task_exit(tsk); 774 kcov_task_exit(tsk); 775 776 WARN_ON(blk_needs_flush_plug(tsk)); 777 778 if (unlikely(in_interrupt())) 779 panic("Aiee, killing interrupt handler!"); 780 if (unlikely(!tsk->pid)) 781 panic("Attempted to kill the idle task!"); 782 783 /* 784 * If do_exit is called because this processes oopsed, it's possible 785 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before 786 * continuing. Amongst other possible reasons, this is to prevent 787 * mm_release()->clear_child_tid() from writing to a user-controlled 788 * kernel address. 789 */ 790 set_fs(USER_DS); 791 792 ptrace_event(PTRACE_EVENT_EXIT, code); 793 794 validate_creds_for_do_exit(tsk); 795 796 /* 797 * We're taking recursive faults here in do_exit. Safest is to just 798 * leave this task alone and wait for reboot. 799 */ 800 if (unlikely(tsk->flags & PF_EXITING)) { 801 pr_alert("Fixing recursive fault but reboot is needed!\n"); 802 /* 803 * We can do this unlocked here. The futex code uses 804 * this flag just to verify whether the pi state 805 * cleanup has been done or not. In the worst case it 806 * loops once more. We pretend that the cleanup was 807 * done as there is no way to return. Either the 808 * OWNER_DIED bit is set by now or we push the blocked 809 * task into the wait for ever nirwana as well. 810 */ 811 tsk->flags |= PF_EXITPIDONE; 812 set_current_state(TASK_UNINTERRUPTIBLE); 813 schedule(); 814 } 815 816 exit_signals(tsk); /* sets PF_EXITING */ 817 /* 818 * Ensure that all new tsk->pi_lock acquisitions must observe 819 * PF_EXITING. Serializes against futex.c:attach_to_pi_owner(). 820 */ 821 smp_mb(); 822 /* 823 * Ensure that we must observe the pi_state in exit_mm() -> 824 * mm_release() -> exit_pi_state_list(). 825 */ 826 raw_spin_unlock_wait(&tsk->pi_lock); 827 828 if (unlikely(in_atomic())) { 829 pr_info("note: %s[%d] exited with preempt_count %d\n", 830 current->comm, task_pid_nr(current), 831 preempt_count()); 832 preempt_count_set(PREEMPT_ENABLED); 833 } 834 835 /* sync mm's RSS info before statistics gathering */ 836 if (tsk->mm) 837 sync_mm_rss(tsk->mm); 838 acct_update_integrals(tsk); 839 group_dead = atomic_dec_and_test(&tsk->signal->live); 840 if (group_dead) { 841 #ifdef CONFIG_POSIX_TIMERS 842 hrtimer_cancel(&tsk->signal->real_timer); 843 exit_itimers(tsk->signal); 844 #endif 845 if (tsk->mm) 846 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm); 847 } 848 acct_collect(code, group_dead); 849 if (group_dead) 850 tty_audit_exit(); 851 audit_free(tsk); 852 853 tsk->exit_code = code; 854 taskstats_exit(tsk, group_dead); 855 856 exit_mm(); 857 858 if (group_dead) 859 acct_process(); 860 trace_sched_process_exit(tsk); 861 862 exit_sem(tsk); 863 exit_shm(tsk); 864 exit_files(tsk); 865 exit_fs(tsk); 866 if (group_dead) 867 disassociate_ctty(1); 868 exit_task_namespaces(tsk); 869 exit_task_work(tsk); 870 exit_thread(tsk); 871 872 /* 873 * Flush inherited counters to the parent - before the parent 874 * gets woken up by child-exit notifications. 875 * 876 * because of cgroup mode, must be called before cgroup_exit() 877 */ 878 perf_event_exit_task(tsk); 879 880 sched_autogroup_exit_task(tsk); 881 cgroup_exit(tsk); 882 883 /* 884 * FIXME: do that only when needed, using sched_exit tracepoint 885 */ 886 flush_ptrace_hw_breakpoint(tsk); 887 888 TASKS_RCU(preempt_disable()); 889 TASKS_RCU(tasks_rcu_i = __srcu_read_lock(&tasks_rcu_exit_srcu)); 890 TASKS_RCU(preempt_enable()); 891 exit_notify(tsk, group_dead); 892 proc_exit_connector(tsk); 893 mpol_put_task_policy(tsk); 894 #ifdef CONFIG_FUTEX 895 if (unlikely(current->pi_state_cache)) 896 kfree(current->pi_state_cache); 897 #endif 898 /* 899 * Make sure we are holding no locks: 900 */ 901 debug_check_no_locks_held(); 902 /* 903 * We can do this unlocked here. The futex code uses this flag 904 * just to verify whether the pi state cleanup has been done 905 * or not. In the worst case it loops once more. 906 */ 907 tsk->flags |= PF_EXITPIDONE; 908 909 if (tsk->io_context) 910 exit_io_context(tsk); 911 912 if (tsk->splice_pipe) 913 free_pipe_info(tsk->splice_pipe); 914 915 if (tsk->task_frag.page) 916 put_page(tsk->task_frag.page); 917 918 validate_creds_for_do_exit(tsk); 919 920 check_stack_usage(); 921 preempt_disable(); 922 if (tsk->nr_dirtied) 923 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied); 924 exit_rcu(); 925 TASKS_RCU(__srcu_read_unlock(&tasks_rcu_exit_srcu, tasks_rcu_i)); 926 927 do_task_dead(); 928 } 929 EXPORT_SYMBOL_GPL(do_exit); 930 931 void complete_and_exit(struct completion *comp, long code) 932 { 933 if (comp) 934 complete(comp); 935 936 do_exit(code); 937 } 938 EXPORT_SYMBOL(complete_and_exit); 939 940 SYSCALL_DEFINE1(exit, int, error_code) 941 { 942 do_exit((error_code&0xff)<<8); 943 } 944 945 /* 946 * Take down every thread in the group. This is called by fatal signals 947 * as well as by sys_exit_group (below). 948 */ 949 void 950 do_group_exit(int exit_code) 951 { 952 struct signal_struct *sig = current->signal; 953 954 BUG_ON(exit_code & 0x80); /* core dumps don't get here */ 955 956 if (signal_group_exit(sig)) 957 exit_code = sig->group_exit_code; 958 else if (!thread_group_empty(current)) { 959 struct sighand_struct *const sighand = current->sighand; 960 961 spin_lock_irq(&sighand->siglock); 962 if (signal_group_exit(sig)) 963 /* Another thread got here before we took the lock. */ 964 exit_code = sig->group_exit_code; 965 else { 966 sig->group_exit_code = exit_code; 967 sig->flags = SIGNAL_GROUP_EXIT; 968 zap_other_threads(current); 969 } 970 spin_unlock_irq(&sighand->siglock); 971 } 972 973 do_exit(exit_code); 974 /* NOTREACHED */ 975 } 976 977 /* 978 * this kills every thread in the thread group. Note that any externally 979 * wait4()-ing process will get the correct exit code - even if this 980 * thread is not the thread group leader. 981 */ 982 SYSCALL_DEFINE1(exit_group, int, error_code) 983 { 984 do_group_exit((error_code & 0xff) << 8); 985 /* NOTREACHED */ 986 return 0; 987 } 988 989 struct wait_opts { 990 enum pid_type wo_type; 991 int wo_flags; 992 struct pid *wo_pid; 993 994 struct siginfo __user *wo_info; 995 int __user *wo_stat; 996 struct rusage __user *wo_rusage; 997 998 wait_queue_t child_wait; 999 int notask_error; 1000 }; 1001 1002 static inline 1003 struct pid *task_pid_type(struct task_struct *task, enum pid_type type) 1004 { 1005 if (type != PIDTYPE_PID) 1006 task = task->group_leader; 1007 return task->pids[type].pid; 1008 } 1009 1010 static int eligible_pid(struct wait_opts *wo, struct task_struct *p) 1011 { 1012 return wo->wo_type == PIDTYPE_MAX || 1013 task_pid_type(p, wo->wo_type) == wo->wo_pid; 1014 } 1015 1016 static int 1017 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p) 1018 { 1019 if (!eligible_pid(wo, p)) 1020 return 0; 1021 1022 /* 1023 * Wait for all children (clone and not) if __WALL is set or 1024 * if it is traced by us. 1025 */ 1026 if (ptrace || (wo->wo_flags & __WALL)) 1027 return 1; 1028 1029 /* 1030 * Otherwise, wait for clone children *only* if __WCLONE is set; 1031 * otherwise, wait for non-clone children *only*. 1032 * 1033 * Note: a "clone" child here is one that reports to its parent 1034 * using a signal other than SIGCHLD, or a non-leader thread which 1035 * we can only see if it is traced by us. 1036 */ 1037 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE)) 1038 return 0; 1039 1040 return 1; 1041 } 1042 1043 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p, 1044 pid_t pid, uid_t uid, int why, int status) 1045 { 1046 struct siginfo __user *infop; 1047 int retval = wo->wo_rusage 1048 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0; 1049 1050 put_task_struct(p); 1051 infop = wo->wo_info; 1052 if (infop) { 1053 if (!retval) 1054 retval = put_user(SIGCHLD, &infop->si_signo); 1055 if (!retval) 1056 retval = put_user(0, &infop->si_errno); 1057 if (!retval) 1058 retval = put_user((short)why, &infop->si_code); 1059 if (!retval) 1060 retval = put_user(pid, &infop->si_pid); 1061 if (!retval) 1062 retval = put_user(uid, &infop->si_uid); 1063 if (!retval) 1064 retval = put_user(status, &infop->si_status); 1065 } 1066 if (!retval) 1067 retval = pid; 1068 return retval; 1069 } 1070 1071 /* 1072 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold 1073 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold 1074 * the lock and this task is uninteresting. If we return nonzero, we have 1075 * released the lock and the system call should return. 1076 */ 1077 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p) 1078 { 1079 int state, retval, status; 1080 pid_t pid = task_pid_vnr(p); 1081 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p)); 1082 struct siginfo __user *infop; 1083 1084 if (!likely(wo->wo_flags & WEXITED)) 1085 return 0; 1086 1087 if (unlikely(wo->wo_flags & WNOWAIT)) { 1088 int exit_code = p->exit_code; 1089 int why; 1090 1091 get_task_struct(p); 1092 read_unlock(&tasklist_lock); 1093 sched_annotate_sleep(); 1094 1095 if ((exit_code & 0x7f) == 0) { 1096 why = CLD_EXITED; 1097 status = exit_code >> 8; 1098 } else { 1099 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED; 1100 status = exit_code & 0x7f; 1101 } 1102 return wait_noreap_copyout(wo, p, pid, uid, why, status); 1103 } 1104 /* 1105 * Move the task's state to DEAD/TRACE, only one thread can do this. 1106 */ 1107 state = (ptrace_reparented(p) && thread_group_leader(p)) ? 1108 EXIT_TRACE : EXIT_DEAD; 1109 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE) 1110 return 0; 1111 /* 1112 * We own this thread, nobody else can reap it. 1113 */ 1114 read_unlock(&tasklist_lock); 1115 sched_annotate_sleep(); 1116 1117 /* 1118 * Check thread_group_leader() to exclude the traced sub-threads. 1119 */ 1120 if (state == EXIT_DEAD && thread_group_leader(p)) { 1121 struct signal_struct *sig = p->signal; 1122 struct signal_struct *psig = current->signal; 1123 unsigned long maxrss; 1124 cputime_t tgutime, tgstime; 1125 1126 /* 1127 * The resource counters for the group leader are in its 1128 * own task_struct. Those for dead threads in the group 1129 * are in its signal_struct, as are those for the child 1130 * processes it has previously reaped. All these 1131 * accumulate in the parent's signal_struct c* fields. 1132 * 1133 * We don't bother to take a lock here to protect these 1134 * p->signal fields because the whole thread group is dead 1135 * and nobody can change them. 1136 * 1137 * psig->stats_lock also protects us from our sub-theads 1138 * which can reap other children at the same time. Until 1139 * we change k_getrusage()-like users to rely on this lock 1140 * we have to take ->siglock as well. 1141 * 1142 * We use thread_group_cputime_adjusted() to get times for 1143 * the thread group, which consolidates times for all threads 1144 * in the group including the group leader. 1145 */ 1146 thread_group_cputime_adjusted(p, &tgutime, &tgstime); 1147 spin_lock_irq(¤t->sighand->siglock); 1148 write_seqlock(&psig->stats_lock); 1149 psig->cutime += tgutime + sig->cutime; 1150 psig->cstime += tgstime + sig->cstime; 1151 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime; 1152 psig->cmin_flt += 1153 p->min_flt + sig->min_flt + sig->cmin_flt; 1154 psig->cmaj_flt += 1155 p->maj_flt + sig->maj_flt + sig->cmaj_flt; 1156 psig->cnvcsw += 1157 p->nvcsw + sig->nvcsw + sig->cnvcsw; 1158 psig->cnivcsw += 1159 p->nivcsw + sig->nivcsw + sig->cnivcsw; 1160 psig->cinblock += 1161 task_io_get_inblock(p) + 1162 sig->inblock + sig->cinblock; 1163 psig->coublock += 1164 task_io_get_oublock(p) + 1165 sig->oublock + sig->coublock; 1166 maxrss = max(sig->maxrss, sig->cmaxrss); 1167 if (psig->cmaxrss < maxrss) 1168 psig->cmaxrss = maxrss; 1169 task_io_accounting_add(&psig->ioac, &p->ioac); 1170 task_io_accounting_add(&psig->ioac, &sig->ioac); 1171 write_sequnlock(&psig->stats_lock); 1172 spin_unlock_irq(¤t->sighand->siglock); 1173 } 1174 1175 retval = wo->wo_rusage 1176 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0; 1177 status = (p->signal->flags & SIGNAL_GROUP_EXIT) 1178 ? p->signal->group_exit_code : p->exit_code; 1179 if (!retval && wo->wo_stat) 1180 retval = put_user(status, wo->wo_stat); 1181 1182 infop = wo->wo_info; 1183 if (!retval && infop) 1184 retval = put_user(SIGCHLD, &infop->si_signo); 1185 if (!retval && infop) 1186 retval = put_user(0, &infop->si_errno); 1187 if (!retval && infop) { 1188 int why; 1189 1190 if ((status & 0x7f) == 0) { 1191 why = CLD_EXITED; 1192 status >>= 8; 1193 } else { 1194 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED; 1195 status &= 0x7f; 1196 } 1197 retval = put_user((short)why, &infop->si_code); 1198 if (!retval) 1199 retval = put_user(status, &infop->si_status); 1200 } 1201 if (!retval && infop) 1202 retval = put_user(pid, &infop->si_pid); 1203 if (!retval && infop) 1204 retval = put_user(uid, &infop->si_uid); 1205 if (!retval) 1206 retval = pid; 1207 1208 if (state == EXIT_TRACE) { 1209 write_lock_irq(&tasklist_lock); 1210 /* We dropped tasklist, ptracer could die and untrace */ 1211 ptrace_unlink(p); 1212 1213 /* If parent wants a zombie, don't release it now */ 1214 state = EXIT_ZOMBIE; 1215 if (do_notify_parent(p, p->exit_signal)) 1216 state = EXIT_DEAD; 1217 p->exit_state = state; 1218 write_unlock_irq(&tasklist_lock); 1219 } 1220 if (state == EXIT_DEAD) 1221 release_task(p); 1222 1223 return retval; 1224 } 1225 1226 static int *task_stopped_code(struct task_struct *p, bool ptrace) 1227 { 1228 if (ptrace) { 1229 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING)) 1230 return &p->exit_code; 1231 } else { 1232 if (p->signal->flags & SIGNAL_STOP_STOPPED) 1233 return &p->signal->group_exit_code; 1234 } 1235 return NULL; 1236 } 1237 1238 /** 1239 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED 1240 * @wo: wait options 1241 * @ptrace: is the wait for ptrace 1242 * @p: task to wait for 1243 * 1244 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED. 1245 * 1246 * CONTEXT: 1247 * read_lock(&tasklist_lock), which is released if return value is 1248 * non-zero. Also, grabs and releases @p->sighand->siglock. 1249 * 1250 * RETURNS: 1251 * 0 if wait condition didn't exist and search for other wait conditions 1252 * should continue. Non-zero return, -errno on failure and @p's pid on 1253 * success, implies that tasklist_lock is released and wait condition 1254 * search should terminate. 1255 */ 1256 static int wait_task_stopped(struct wait_opts *wo, 1257 int ptrace, struct task_struct *p) 1258 { 1259 struct siginfo __user *infop; 1260 int retval, exit_code, *p_code, why; 1261 uid_t uid = 0; /* unneeded, required by compiler */ 1262 pid_t pid; 1263 1264 /* 1265 * Traditionally we see ptrace'd stopped tasks regardless of options. 1266 */ 1267 if (!ptrace && !(wo->wo_flags & WUNTRACED)) 1268 return 0; 1269 1270 if (!task_stopped_code(p, ptrace)) 1271 return 0; 1272 1273 exit_code = 0; 1274 spin_lock_irq(&p->sighand->siglock); 1275 1276 p_code = task_stopped_code(p, ptrace); 1277 if (unlikely(!p_code)) 1278 goto unlock_sig; 1279 1280 exit_code = *p_code; 1281 if (!exit_code) 1282 goto unlock_sig; 1283 1284 if (!unlikely(wo->wo_flags & WNOWAIT)) 1285 *p_code = 0; 1286 1287 uid = from_kuid_munged(current_user_ns(), task_uid(p)); 1288 unlock_sig: 1289 spin_unlock_irq(&p->sighand->siglock); 1290 if (!exit_code) 1291 return 0; 1292 1293 /* 1294 * Now we are pretty sure this task is interesting. 1295 * Make sure it doesn't get reaped out from under us while we 1296 * give up the lock and then examine it below. We don't want to 1297 * keep holding onto the tasklist_lock while we call getrusage and 1298 * possibly take page faults for user memory. 1299 */ 1300 get_task_struct(p); 1301 pid = task_pid_vnr(p); 1302 why = ptrace ? CLD_TRAPPED : CLD_STOPPED; 1303 read_unlock(&tasklist_lock); 1304 sched_annotate_sleep(); 1305 1306 if (unlikely(wo->wo_flags & WNOWAIT)) 1307 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code); 1308 1309 retval = wo->wo_rusage 1310 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0; 1311 if (!retval && wo->wo_stat) 1312 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat); 1313 1314 infop = wo->wo_info; 1315 if (!retval && infop) 1316 retval = put_user(SIGCHLD, &infop->si_signo); 1317 if (!retval && infop) 1318 retval = put_user(0, &infop->si_errno); 1319 if (!retval && infop) 1320 retval = put_user((short)why, &infop->si_code); 1321 if (!retval && infop) 1322 retval = put_user(exit_code, &infop->si_status); 1323 if (!retval && infop) 1324 retval = put_user(pid, &infop->si_pid); 1325 if (!retval && infop) 1326 retval = put_user(uid, &infop->si_uid); 1327 if (!retval) 1328 retval = pid; 1329 put_task_struct(p); 1330 1331 BUG_ON(!retval); 1332 return retval; 1333 } 1334 1335 /* 1336 * Handle do_wait work for one task in a live, non-stopped state. 1337 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold 1338 * the lock and this task is uninteresting. If we return nonzero, we have 1339 * released the lock and the system call should return. 1340 */ 1341 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p) 1342 { 1343 int retval; 1344 pid_t pid; 1345 uid_t uid; 1346 1347 if (!unlikely(wo->wo_flags & WCONTINUED)) 1348 return 0; 1349 1350 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) 1351 return 0; 1352 1353 spin_lock_irq(&p->sighand->siglock); 1354 /* Re-check with the lock held. */ 1355 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) { 1356 spin_unlock_irq(&p->sighand->siglock); 1357 return 0; 1358 } 1359 if (!unlikely(wo->wo_flags & WNOWAIT)) 1360 p->signal->flags &= ~SIGNAL_STOP_CONTINUED; 1361 uid = from_kuid_munged(current_user_ns(), task_uid(p)); 1362 spin_unlock_irq(&p->sighand->siglock); 1363 1364 pid = task_pid_vnr(p); 1365 get_task_struct(p); 1366 read_unlock(&tasklist_lock); 1367 sched_annotate_sleep(); 1368 1369 if (!wo->wo_info) { 1370 retval = wo->wo_rusage 1371 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0; 1372 put_task_struct(p); 1373 if (!retval && wo->wo_stat) 1374 retval = put_user(0xffff, wo->wo_stat); 1375 if (!retval) 1376 retval = pid; 1377 } else { 1378 retval = wait_noreap_copyout(wo, p, pid, uid, 1379 CLD_CONTINUED, SIGCONT); 1380 BUG_ON(retval == 0); 1381 } 1382 1383 return retval; 1384 } 1385 1386 /* 1387 * Consider @p for a wait by @parent. 1388 * 1389 * -ECHILD should be in ->notask_error before the first call. 1390 * Returns nonzero for a final return, when we have unlocked tasklist_lock. 1391 * Returns zero if the search for a child should continue; 1392 * then ->notask_error is 0 if @p is an eligible child, 1393 * or another error from security_task_wait(), or still -ECHILD. 1394 */ 1395 static int wait_consider_task(struct wait_opts *wo, int ptrace, 1396 struct task_struct *p) 1397 { 1398 /* 1399 * We can race with wait_task_zombie() from another thread. 1400 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition 1401 * can't confuse the checks below. 1402 */ 1403 int exit_state = ACCESS_ONCE(p->exit_state); 1404 int ret; 1405 1406 if (unlikely(exit_state == EXIT_DEAD)) 1407 return 0; 1408 1409 ret = eligible_child(wo, ptrace, p); 1410 if (!ret) 1411 return ret; 1412 1413 ret = security_task_wait(p); 1414 if (unlikely(ret < 0)) { 1415 /* 1416 * If we have not yet seen any eligible child, 1417 * then let this error code replace -ECHILD. 1418 * A permission error will give the user a clue 1419 * to look for security policy problems, rather 1420 * than for mysterious wait bugs. 1421 */ 1422 if (wo->notask_error) 1423 wo->notask_error = ret; 1424 return 0; 1425 } 1426 1427 if (unlikely(exit_state == EXIT_TRACE)) { 1428 /* 1429 * ptrace == 0 means we are the natural parent. In this case 1430 * we should clear notask_error, debugger will notify us. 1431 */ 1432 if (likely(!ptrace)) 1433 wo->notask_error = 0; 1434 return 0; 1435 } 1436 1437 if (likely(!ptrace) && unlikely(p->ptrace)) { 1438 /* 1439 * If it is traced by its real parent's group, just pretend 1440 * the caller is ptrace_do_wait() and reap this child if it 1441 * is zombie. 1442 * 1443 * This also hides group stop state from real parent; otherwise 1444 * a single stop can be reported twice as group and ptrace stop. 1445 * If a ptracer wants to distinguish these two events for its 1446 * own children it should create a separate process which takes 1447 * the role of real parent. 1448 */ 1449 if (!ptrace_reparented(p)) 1450 ptrace = 1; 1451 } 1452 1453 /* slay zombie? */ 1454 if (exit_state == EXIT_ZOMBIE) { 1455 /* we don't reap group leaders with subthreads */ 1456 if (!delay_group_leader(p)) { 1457 /* 1458 * A zombie ptracee is only visible to its ptracer. 1459 * Notification and reaping will be cascaded to the 1460 * real parent when the ptracer detaches. 1461 */ 1462 if (unlikely(ptrace) || likely(!p->ptrace)) 1463 return wait_task_zombie(wo, p); 1464 } 1465 1466 /* 1467 * Allow access to stopped/continued state via zombie by 1468 * falling through. Clearing of notask_error is complex. 1469 * 1470 * When !@ptrace: 1471 * 1472 * If WEXITED is set, notask_error should naturally be 1473 * cleared. If not, subset of WSTOPPED|WCONTINUED is set, 1474 * so, if there are live subthreads, there are events to 1475 * wait for. If all subthreads are dead, it's still safe 1476 * to clear - this function will be called again in finite 1477 * amount time once all the subthreads are released and 1478 * will then return without clearing. 1479 * 1480 * When @ptrace: 1481 * 1482 * Stopped state is per-task and thus can't change once the 1483 * target task dies. Only continued and exited can happen. 1484 * Clear notask_error if WCONTINUED | WEXITED. 1485 */ 1486 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED))) 1487 wo->notask_error = 0; 1488 } else { 1489 /* 1490 * @p is alive and it's gonna stop, continue or exit, so 1491 * there always is something to wait for. 1492 */ 1493 wo->notask_error = 0; 1494 } 1495 1496 /* 1497 * Wait for stopped. Depending on @ptrace, different stopped state 1498 * is used and the two don't interact with each other. 1499 */ 1500 ret = wait_task_stopped(wo, ptrace, p); 1501 if (ret) 1502 return ret; 1503 1504 /* 1505 * Wait for continued. There's only one continued state and the 1506 * ptracer can consume it which can confuse the real parent. Don't 1507 * use WCONTINUED from ptracer. You don't need or want it. 1508 */ 1509 return wait_task_continued(wo, p); 1510 } 1511 1512 /* 1513 * Do the work of do_wait() for one thread in the group, @tsk. 1514 * 1515 * -ECHILD should be in ->notask_error before the first call. 1516 * Returns nonzero for a final return, when we have unlocked tasklist_lock. 1517 * Returns zero if the search for a child should continue; then 1518 * ->notask_error is 0 if there were any eligible children, 1519 * or another error from security_task_wait(), or still -ECHILD. 1520 */ 1521 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk) 1522 { 1523 struct task_struct *p; 1524 1525 list_for_each_entry(p, &tsk->children, sibling) { 1526 int ret = wait_consider_task(wo, 0, p); 1527 1528 if (ret) 1529 return ret; 1530 } 1531 1532 return 0; 1533 } 1534 1535 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk) 1536 { 1537 struct task_struct *p; 1538 1539 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) { 1540 int ret = wait_consider_task(wo, 1, p); 1541 1542 if (ret) 1543 return ret; 1544 } 1545 1546 return 0; 1547 } 1548 1549 static int child_wait_callback(wait_queue_t *wait, unsigned mode, 1550 int sync, void *key) 1551 { 1552 struct wait_opts *wo = container_of(wait, struct wait_opts, 1553 child_wait); 1554 struct task_struct *p = key; 1555 1556 if (!eligible_pid(wo, p)) 1557 return 0; 1558 1559 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent) 1560 return 0; 1561 1562 return default_wake_function(wait, mode, sync, key); 1563 } 1564 1565 void __wake_up_parent(struct task_struct *p, struct task_struct *parent) 1566 { 1567 __wake_up_sync_key(&parent->signal->wait_chldexit, 1568 TASK_INTERRUPTIBLE, 1, p); 1569 } 1570 1571 static long do_wait(struct wait_opts *wo) 1572 { 1573 struct task_struct *tsk; 1574 int retval; 1575 1576 trace_sched_process_wait(wo->wo_pid); 1577 1578 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback); 1579 wo->child_wait.private = current; 1580 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait); 1581 repeat: 1582 /* 1583 * If there is nothing that can match our criteria, just get out. 1584 * We will clear ->notask_error to zero if we see any child that 1585 * might later match our criteria, even if we are not able to reap 1586 * it yet. 1587 */ 1588 wo->notask_error = -ECHILD; 1589 if ((wo->wo_type < PIDTYPE_MAX) && 1590 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type]))) 1591 goto notask; 1592 1593 set_current_state(TASK_INTERRUPTIBLE); 1594 read_lock(&tasklist_lock); 1595 tsk = current; 1596 do { 1597 retval = do_wait_thread(wo, tsk); 1598 if (retval) 1599 goto end; 1600 1601 retval = ptrace_do_wait(wo, tsk); 1602 if (retval) 1603 goto end; 1604 1605 if (wo->wo_flags & __WNOTHREAD) 1606 break; 1607 } while_each_thread(current, tsk); 1608 read_unlock(&tasklist_lock); 1609 1610 notask: 1611 retval = wo->notask_error; 1612 if (!retval && !(wo->wo_flags & WNOHANG)) { 1613 retval = -ERESTARTSYS; 1614 if (!signal_pending(current)) { 1615 schedule(); 1616 goto repeat; 1617 } 1618 } 1619 end: 1620 __set_current_state(TASK_RUNNING); 1621 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait); 1622 return retval; 1623 } 1624 1625 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *, 1626 infop, int, options, struct rusage __user *, ru) 1627 { 1628 struct wait_opts wo; 1629 struct pid *pid = NULL; 1630 enum pid_type type; 1631 long ret; 1632 1633 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED| 1634 __WNOTHREAD|__WCLONE|__WALL)) 1635 return -EINVAL; 1636 if (!(options & (WEXITED|WSTOPPED|WCONTINUED))) 1637 return -EINVAL; 1638 1639 switch (which) { 1640 case P_ALL: 1641 type = PIDTYPE_MAX; 1642 break; 1643 case P_PID: 1644 type = PIDTYPE_PID; 1645 if (upid <= 0) 1646 return -EINVAL; 1647 break; 1648 case P_PGID: 1649 type = PIDTYPE_PGID; 1650 if (upid <= 0) 1651 return -EINVAL; 1652 break; 1653 default: 1654 return -EINVAL; 1655 } 1656 1657 if (type < PIDTYPE_MAX) 1658 pid = find_get_pid(upid); 1659 1660 wo.wo_type = type; 1661 wo.wo_pid = pid; 1662 wo.wo_flags = options; 1663 wo.wo_info = infop; 1664 wo.wo_stat = NULL; 1665 wo.wo_rusage = ru; 1666 ret = do_wait(&wo); 1667 1668 if (ret > 0) { 1669 ret = 0; 1670 } else if (infop) { 1671 /* 1672 * For a WNOHANG return, clear out all the fields 1673 * we would set so the user can easily tell the 1674 * difference. 1675 */ 1676 if (!ret) 1677 ret = put_user(0, &infop->si_signo); 1678 if (!ret) 1679 ret = put_user(0, &infop->si_errno); 1680 if (!ret) 1681 ret = put_user(0, &infop->si_code); 1682 if (!ret) 1683 ret = put_user(0, &infop->si_pid); 1684 if (!ret) 1685 ret = put_user(0, &infop->si_uid); 1686 if (!ret) 1687 ret = put_user(0, &infop->si_status); 1688 } 1689 1690 put_pid(pid); 1691 return ret; 1692 } 1693 1694 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr, 1695 int, options, struct rusage __user *, ru) 1696 { 1697 struct wait_opts wo; 1698 struct pid *pid = NULL; 1699 enum pid_type type; 1700 long ret; 1701 1702 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED| 1703 __WNOTHREAD|__WCLONE|__WALL)) 1704 return -EINVAL; 1705 1706 if (upid == -1) 1707 type = PIDTYPE_MAX; 1708 else if (upid < 0) { 1709 type = PIDTYPE_PGID; 1710 pid = find_get_pid(-upid); 1711 } else if (upid == 0) { 1712 type = PIDTYPE_PGID; 1713 pid = get_task_pid(current, PIDTYPE_PGID); 1714 } else /* upid > 0 */ { 1715 type = PIDTYPE_PID; 1716 pid = find_get_pid(upid); 1717 } 1718 1719 wo.wo_type = type; 1720 wo.wo_pid = pid; 1721 wo.wo_flags = options | WEXITED; 1722 wo.wo_info = NULL; 1723 wo.wo_stat = stat_addr; 1724 wo.wo_rusage = ru; 1725 ret = do_wait(&wo); 1726 put_pid(pid); 1727 1728 return ret; 1729 } 1730 1731 #ifdef __ARCH_WANT_SYS_WAITPID 1732 1733 /* 1734 * sys_waitpid() remains for compatibility. waitpid() should be 1735 * implemented by calling sys_wait4() from libc.a. 1736 */ 1737 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options) 1738 { 1739 return sys_wait4(pid, stat_addr, options, NULL); 1740 } 1741 1742 #endif 1743