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