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/sched/autogroup.h> 10 #include <linux/sched/mm.h> 11 #include <linux/sched/stat.h> 12 #include <linux/sched/task.h> 13 #include <linux/sched/task_stack.h> 14 #include <linux/sched/cputime.h> 15 #include <linux/interrupt.h> 16 #include <linux/module.h> 17 #include <linux/capability.h> 18 #include <linux/completion.h> 19 #include <linux/personality.h> 20 #include <linux/tty.h> 21 #include <linux/iocontext.h> 22 #include <linux/key.h> 23 #include <linux/cpu.h> 24 #include <linux/acct.h> 25 #include <linux/tsacct_kern.h> 26 #include <linux/file.h> 27 #include <linux/fdtable.h> 28 #include <linux/freezer.h> 29 #include <linux/binfmts.h> 30 #include <linux/nsproxy.h> 31 #include <linux/pid_namespace.h> 32 #include <linux/ptrace.h> 33 #include <linux/profile.h> 34 #include <linux/mount.h> 35 #include <linux/proc_fs.h> 36 #include <linux/kthread.h> 37 #include <linux/mempolicy.h> 38 #include <linux/taskstats_kern.h> 39 #include <linux/delayacct.h> 40 #include <linux/cgroup.h> 41 #include <linux/syscalls.h> 42 #include <linux/signal.h> 43 #include <linux/posix-timers.h> 44 #include <linux/cn_proc.h> 45 #include <linux/mutex.h> 46 #include <linux/futex.h> 47 #include <linux/pipe_fs_i.h> 48 #include <linux/audit.h> /* for audit_free() */ 49 #include <linux/resource.h> 50 #include <linux/blkdev.h> 51 #include <linux/task_io_accounting_ops.h> 52 #include <linux/tracehook.h> 53 #include <linux/fs_struct.h> 54 #include <linux/init_task.h> 55 #include <linux/perf_event.h> 56 #include <trace/events/sched.h> 57 #include <linux/hw_breakpoint.h> 58 #include <linux/oom.h> 59 #include <linux/writeback.h> 60 #include <linux/shm.h> 61 #include <linux/kcov.h> 62 #include <linux/random.h> 63 #include <linux/rcuwait.h> 64 #include <linux/compat.h> 65 66 #include <linux/uaccess.h> 67 #include <asm/unistd.h> 68 #include <asm/pgtable.h> 69 #include <asm/mmu_context.h> 70 71 static void __unhash_process(struct task_struct *p, bool group_dead) 72 { 73 nr_threads--; 74 detach_pid(p, PIDTYPE_PID); 75 if (group_dead) { 76 detach_pid(p, PIDTYPE_TGID); 77 detach_pid(p, PIDTYPE_PGID); 78 detach_pid(p, PIDTYPE_SID); 79 80 list_del_rcu(&p->tasks); 81 list_del_init(&p->sibling); 82 __this_cpu_dec(process_counts); 83 } 84 list_del_rcu(&p->thread_group); 85 list_del_rcu(&p->thread_node); 86 } 87 88 /* 89 * This function expects the tasklist_lock write-locked. 90 */ 91 static void __exit_signal(struct task_struct *tsk) 92 { 93 struct signal_struct *sig = tsk->signal; 94 bool group_dead = thread_group_leader(tsk); 95 struct sighand_struct *sighand; 96 struct tty_struct *uninitialized_var(tty); 97 u64 utime, stime; 98 99 sighand = rcu_dereference_check(tsk->sighand, 100 lockdep_tasklist_lock_is_held()); 101 spin_lock(&sighand->siglock); 102 103 #ifdef CONFIG_POSIX_TIMERS 104 posix_cpu_timers_exit(tsk); 105 if (group_dead) { 106 posix_cpu_timers_exit_group(tsk); 107 } else { 108 /* 109 * This can only happen if the caller is de_thread(). 110 * FIXME: this is the temporary hack, we should teach 111 * posix-cpu-timers to handle this case correctly. 112 */ 113 if (unlikely(has_group_leader_pid(tsk))) 114 posix_cpu_timers_exit_group(tsk); 115 } 116 #endif 117 118 if (group_dead) { 119 tty = sig->tty; 120 sig->tty = NULL; 121 } else { 122 /* 123 * If there is any task waiting for the group exit 124 * then notify it: 125 */ 126 if (sig->notify_count > 0 && !--sig->notify_count) 127 wake_up_process(sig->group_exit_task); 128 129 if (tsk == sig->curr_target) 130 sig->curr_target = next_thread(tsk); 131 } 132 133 add_device_randomness((const void*) &tsk->se.sum_exec_runtime, 134 sizeof(unsigned long long)); 135 136 /* 137 * Accumulate here the counters for all threads as they die. We could 138 * skip the group leader because it is the last user of signal_struct, 139 * but we want to avoid the race with thread_group_cputime() which can 140 * see the empty ->thread_head list. 141 */ 142 task_cputime(tsk, &utime, &stime); 143 write_seqlock(&sig->stats_lock); 144 sig->utime += utime; 145 sig->stime += stime; 146 sig->gtime += task_gtime(tsk); 147 sig->min_flt += tsk->min_flt; 148 sig->maj_flt += tsk->maj_flt; 149 sig->nvcsw += tsk->nvcsw; 150 sig->nivcsw += tsk->nivcsw; 151 sig->inblock += task_io_get_inblock(tsk); 152 sig->oublock += task_io_get_oublock(tsk); 153 task_io_accounting_add(&sig->ioac, &tsk->ioac); 154 sig->sum_sched_runtime += tsk->se.sum_exec_runtime; 155 sig->nr_threads--; 156 __unhash_process(tsk, group_dead); 157 write_sequnlock(&sig->stats_lock); 158 159 /* 160 * Do this under ->siglock, we can race with another thread 161 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals. 162 */ 163 flush_sigqueue(&tsk->pending); 164 tsk->sighand = NULL; 165 spin_unlock(&sighand->siglock); 166 167 __cleanup_sighand(sighand); 168 clear_tsk_thread_flag(tsk, TIF_SIGPENDING); 169 if (group_dead) { 170 flush_sigqueue(&sig->shared_pending); 171 tty_kref_put(tty); 172 } 173 } 174 175 static void delayed_put_task_struct(struct rcu_head *rhp) 176 { 177 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu); 178 179 perf_event_delayed_put(tsk); 180 trace_sched_process_free(tsk); 181 put_task_struct(tsk); 182 } 183 184 185 void release_task(struct task_struct *p) 186 { 187 struct task_struct *leader; 188 int zap_leader; 189 repeat: 190 /* don't need to get the RCU readlock here - the process is dead and 191 * can't be modifying its own credentials. But shut RCU-lockdep up */ 192 rcu_read_lock(); 193 atomic_dec(&__task_cred(p)->user->processes); 194 rcu_read_unlock(); 195 196 proc_flush_task(p); 197 198 write_lock_irq(&tasklist_lock); 199 ptrace_release_task(p); 200 __exit_signal(p); 201 202 /* 203 * If we are the last non-leader member of the thread 204 * group, and the leader is zombie, then notify the 205 * group leader's parent process. (if it wants notification.) 206 */ 207 zap_leader = 0; 208 leader = p->group_leader; 209 if (leader != p && thread_group_empty(leader) 210 && leader->exit_state == EXIT_ZOMBIE) { 211 /* 212 * If we were the last child thread and the leader has 213 * exited already, and the leader's parent ignores SIGCHLD, 214 * then we are the one who should release the leader. 215 */ 216 zap_leader = do_notify_parent(leader, leader->exit_signal); 217 if (zap_leader) 218 leader->exit_state = EXIT_DEAD; 219 } 220 221 write_unlock_irq(&tasklist_lock); 222 release_thread(p); 223 call_rcu(&p->rcu, delayed_put_task_struct); 224 225 p = leader; 226 if (unlikely(zap_leader)) 227 goto repeat; 228 } 229 230 /* 231 * Note that if this function returns a valid task_struct pointer (!NULL) 232 * task->usage must remain >0 for the duration of the RCU critical section. 233 */ 234 struct task_struct *task_rcu_dereference(struct task_struct **ptask) 235 { 236 struct sighand_struct *sighand; 237 struct task_struct *task; 238 239 /* 240 * We need to verify that release_task() was not called and thus 241 * delayed_put_task_struct() can't run and drop the last reference 242 * before rcu_read_unlock(). We check task->sighand != NULL, 243 * but we can read the already freed and reused memory. 244 */ 245 retry: 246 task = rcu_dereference(*ptask); 247 if (!task) 248 return NULL; 249 250 probe_kernel_address(&task->sighand, sighand); 251 252 /* 253 * Pairs with atomic_dec_and_test() in put_task_struct(). If this task 254 * was already freed we can not miss the preceding update of this 255 * pointer. 256 */ 257 smp_rmb(); 258 if (unlikely(task != READ_ONCE(*ptask))) 259 goto retry; 260 261 /* 262 * We've re-checked that "task == *ptask", now we have two different 263 * cases: 264 * 265 * 1. This is actually the same task/task_struct. In this case 266 * sighand != NULL tells us it is still alive. 267 * 268 * 2. This is another task which got the same memory for task_struct. 269 * We can't know this of course, and we can not trust 270 * sighand != NULL. 271 * 272 * In this case we actually return a random value, but this is 273 * correct. 274 * 275 * If we return NULL - we can pretend that we actually noticed that 276 * *ptask was updated when the previous task has exited. Or pretend 277 * that probe_slab_address(&sighand) reads NULL. 278 * 279 * If we return the new task (because sighand is not NULL for any 280 * reason) - this is fine too. This (new) task can't go away before 281 * another gp pass. 282 * 283 * And note: We could even eliminate the false positive if re-read 284 * task->sighand once again to avoid the falsely NULL. But this case 285 * is very unlikely so we don't care. 286 */ 287 if (!sighand) 288 return NULL; 289 290 return task; 291 } 292 293 void rcuwait_wake_up(struct rcuwait *w) 294 { 295 struct task_struct *task; 296 297 rcu_read_lock(); 298 299 /* 300 * Order condition vs @task, such that everything prior to the load 301 * of @task is visible. This is the condition as to why the user called 302 * rcuwait_trywake() in the first place. Pairs with set_current_state() 303 * barrier (A) in rcuwait_wait_event(). 304 * 305 * WAIT WAKE 306 * [S] tsk = current [S] cond = true 307 * MB (A) MB (B) 308 * [L] cond [L] tsk 309 */ 310 smp_mb(); /* (B) */ 311 312 /* 313 * Avoid using task_rcu_dereference() magic as long as we are careful, 314 * see comment in rcuwait_wait_event() regarding ->exit_state. 315 */ 316 task = rcu_dereference(w->task); 317 if (task) 318 wake_up_process(task); 319 rcu_read_unlock(); 320 } 321 322 /* 323 * Determine if a process group is "orphaned", according to the POSIX 324 * definition in 2.2.2.52. Orphaned process groups are not to be affected 325 * by terminal-generated stop signals. Newly orphaned process groups are 326 * to receive a SIGHUP and a SIGCONT. 327 * 328 * "I ask you, have you ever known what it is to be an orphan?" 329 */ 330 static int will_become_orphaned_pgrp(struct pid *pgrp, 331 struct task_struct *ignored_task) 332 { 333 struct task_struct *p; 334 335 do_each_pid_task(pgrp, PIDTYPE_PGID, p) { 336 if ((p == ignored_task) || 337 (p->exit_state && thread_group_empty(p)) || 338 is_global_init(p->real_parent)) 339 continue; 340 341 if (task_pgrp(p->real_parent) != pgrp && 342 task_session(p->real_parent) == task_session(p)) 343 return 0; 344 } while_each_pid_task(pgrp, PIDTYPE_PGID, p); 345 346 return 1; 347 } 348 349 int is_current_pgrp_orphaned(void) 350 { 351 int retval; 352 353 read_lock(&tasklist_lock); 354 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL); 355 read_unlock(&tasklist_lock); 356 357 return retval; 358 } 359 360 static bool has_stopped_jobs(struct pid *pgrp) 361 { 362 struct task_struct *p; 363 364 do_each_pid_task(pgrp, PIDTYPE_PGID, p) { 365 if (p->signal->flags & SIGNAL_STOP_STOPPED) 366 return true; 367 } while_each_pid_task(pgrp, PIDTYPE_PGID, p); 368 369 return false; 370 } 371 372 /* 373 * Check to see if any process groups have become orphaned as 374 * a result of our exiting, and if they have any stopped jobs, 375 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2) 376 */ 377 static void 378 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent) 379 { 380 struct pid *pgrp = task_pgrp(tsk); 381 struct task_struct *ignored_task = tsk; 382 383 if (!parent) 384 /* exit: our father is in a different pgrp than 385 * we are and we were the only connection outside. 386 */ 387 parent = tsk->real_parent; 388 else 389 /* reparent: our child is in a different pgrp than 390 * we are, and it was the only connection outside. 391 */ 392 ignored_task = NULL; 393 394 if (task_pgrp(parent) != pgrp && 395 task_session(parent) == task_session(tsk) && 396 will_become_orphaned_pgrp(pgrp, ignored_task) && 397 has_stopped_jobs(pgrp)) { 398 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp); 399 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp); 400 } 401 } 402 403 #ifdef CONFIG_MEMCG 404 /* 405 * A task is exiting. If it owned this mm, find a new owner for the mm. 406 */ 407 void mm_update_next_owner(struct mm_struct *mm) 408 { 409 struct task_struct *c, *g, *p = current; 410 411 retry: 412 /* 413 * If the exiting or execing task is not the owner, it's 414 * someone else's problem. 415 */ 416 if (mm->owner != p) 417 return; 418 /* 419 * The current owner is exiting/execing and there are no other 420 * candidates. Do not leave the mm pointing to a possibly 421 * freed task structure. 422 */ 423 if (atomic_read(&mm->mm_users) <= 1) { 424 mm->owner = NULL; 425 return; 426 } 427 428 read_lock(&tasklist_lock); 429 /* 430 * Search in the children 431 */ 432 list_for_each_entry(c, &p->children, sibling) { 433 if (c->mm == mm) 434 goto assign_new_owner; 435 } 436 437 /* 438 * Search in the siblings 439 */ 440 list_for_each_entry(c, &p->real_parent->children, sibling) { 441 if (c->mm == mm) 442 goto assign_new_owner; 443 } 444 445 /* 446 * Search through everything else, we should not get here often. 447 */ 448 for_each_process(g) { 449 if (g->flags & PF_KTHREAD) 450 continue; 451 for_each_thread(g, c) { 452 if (c->mm == mm) 453 goto assign_new_owner; 454 if (c->mm) 455 break; 456 } 457 } 458 read_unlock(&tasklist_lock); 459 /* 460 * We found no owner yet mm_users > 1: this implies that we are 461 * most likely racing with swapoff (try_to_unuse()) or /proc or 462 * ptrace or page migration (get_task_mm()). Mark owner as NULL. 463 */ 464 mm->owner = NULL; 465 return; 466 467 assign_new_owner: 468 BUG_ON(c == p); 469 get_task_struct(c); 470 /* 471 * The task_lock protects c->mm from changing. 472 * We always want mm->owner->mm == mm 473 */ 474 task_lock(c); 475 /* 476 * Delay read_unlock() till we have the task_lock() 477 * to ensure that c does not slip away underneath us 478 */ 479 read_unlock(&tasklist_lock); 480 if (c->mm != mm) { 481 task_unlock(c); 482 put_task_struct(c); 483 goto retry; 484 } 485 mm->owner = c; 486 task_unlock(c); 487 put_task_struct(c); 488 } 489 #endif /* CONFIG_MEMCG */ 490 491 /* 492 * Turn us into a lazy TLB process if we 493 * aren't already.. 494 */ 495 static void exit_mm(void) 496 { 497 struct mm_struct *mm = current->mm; 498 struct core_state *core_state; 499 500 mm_release(current, mm); 501 if (!mm) 502 return; 503 sync_mm_rss(mm); 504 /* 505 * Serialize with any possible pending coredump. 506 * We must hold mmap_sem around checking core_state 507 * and clearing tsk->mm. The core-inducing thread 508 * will increment ->nr_threads for each thread in the 509 * group with ->mm != NULL. 510 */ 511 down_read(&mm->mmap_sem); 512 core_state = mm->core_state; 513 if (core_state) { 514 struct core_thread self; 515 516 up_read(&mm->mmap_sem); 517 518 self.task = current; 519 self.next = xchg(&core_state->dumper.next, &self); 520 /* 521 * Implies mb(), the result of xchg() must be visible 522 * to core_state->dumper. 523 */ 524 if (atomic_dec_and_test(&core_state->nr_threads)) 525 complete(&core_state->startup); 526 527 for (;;) { 528 set_current_state(TASK_UNINTERRUPTIBLE); 529 if (!self.task) /* see coredump_finish() */ 530 break; 531 freezable_schedule(); 532 } 533 __set_current_state(TASK_RUNNING); 534 down_read(&mm->mmap_sem); 535 } 536 mmgrab(mm); 537 BUG_ON(mm != current->active_mm); 538 /* more a memory barrier than a real lock */ 539 task_lock(current); 540 current->mm = NULL; 541 up_read(&mm->mmap_sem); 542 enter_lazy_tlb(mm, current); 543 task_unlock(current); 544 mm_update_next_owner(mm); 545 mmput(mm); 546 if (test_thread_flag(TIF_MEMDIE)) 547 exit_oom_victim(); 548 } 549 550 static struct task_struct *find_alive_thread(struct task_struct *p) 551 { 552 struct task_struct *t; 553 554 for_each_thread(p, t) { 555 if (!(t->flags & PF_EXITING)) 556 return t; 557 } 558 return NULL; 559 } 560 561 static struct task_struct *find_child_reaper(struct task_struct *father, 562 struct list_head *dead) 563 __releases(&tasklist_lock) 564 __acquires(&tasklist_lock) 565 { 566 struct pid_namespace *pid_ns = task_active_pid_ns(father); 567 struct task_struct *reaper = pid_ns->child_reaper; 568 struct task_struct *p, *n; 569 570 if (likely(reaper != father)) 571 return reaper; 572 573 reaper = find_alive_thread(father); 574 if (reaper) { 575 pid_ns->child_reaper = reaper; 576 return reaper; 577 } 578 579 write_unlock_irq(&tasklist_lock); 580 if (unlikely(pid_ns == &init_pid_ns)) { 581 panic("Attempted to kill init! exitcode=0x%08x\n", 582 father->signal->group_exit_code ?: father->exit_code); 583 } 584 585 list_for_each_entry_safe(p, n, dead, ptrace_entry) { 586 list_del_init(&p->ptrace_entry); 587 release_task(p); 588 } 589 590 zap_pid_ns_processes(pid_ns); 591 write_lock_irq(&tasklist_lock); 592 593 return father; 594 } 595 596 /* 597 * When we die, we re-parent all our children, and try to: 598 * 1. give them to another thread in our thread group, if such a member exists 599 * 2. give it to the first ancestor process which prctl'd itself as a 600 * child_subreaper for its children (like a service manager) 601 * 3. give it to the init process (PID 1) in our pid namespace 602 */ 603 static struct task_struct *find_new_reaper(struct task_struct *father, 604 struct task_struct *child_reaper) 605 { 606 struct task_struct *thread, *reaper; 607 608 thread = find_alive_thread(father); 609 if (thread) 610 return thread; 611 612 if (father->signal->has_child_subreaper) { 613 unsigned int ns_level = task_pid(father)->level; 614 /* 615 * Find the first ->is_child_subreaper ancestor in our pid_ns. 616 * We can't check reaper != child_reaper to ensure we do not 617 * cross the namespaces, the exiting parent could be injected 618 * by setns() + fork(). 619 * We check pid->level, this is slightly more efficient than 620 * task_active_pid_ns(reaper) != task_active_pid_ns(father). 621 */ 622 for (reaper = father->real_parent; 623 task_pid(reaper)->level == ns_level; 624 reaper = reaper->real_parent) { 625 if (reaper == &init_task) 626 break; 627 if (!reaper->signal->is_child_subreaper) 628 continue; 629 thread = find_alive_thread(reaper); 630 if (thread) 631 return thread; 632 } 633 } 634 635 return child_reaper; 636 } 637 638 /* 639 * Any that need to be release_task'd are put on the @dead list. 640 */ 641 static void reparent_leader(struct task_struct *father, struct task_struct *p, 642 struct list_head *dead) 643 { 644 if (unlikely(p->exit_state == EXIT_DEAD)) 645 return; 646 647 /* We don't want people slaying init. */ 648 p->exit_signal = SIGCHLD; 649 650 /* If it has exited notify the new parent about this child's death. */ 651 if (!p->ptrace && 652 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) { 653 if (do_notify_parent(p, p->exit_signal)) { 654 p->exit_state = EXIT_DEAD; 655 list_add(&p->ptrace_entry, dead); 656 } 657 } 658 659 kill_orphaned_pgrp(p, father); 660 } 661 662 /* 663 * This does two things: 664 * 665 * A. Make init inherit all the child processes 666 * B. Check to see if any process groups have become orphaned 667 * as a result of our exiting, and if they have any stopped 668 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2) 669 */ 670 static void forget_original_parent(struct task_struct *father, 671 struct list_head *dead) 672 { 673 struct task_struct *p, *t, *reaper; 674 675 if (unlikely(!list_empty(&father->ptraced))) 676 exit_ptrace(father, dead); 677 678 /* Can drop and reacquire tasklist_lock */ 679 reaper = find_child_reaper(father, dead); 680 if (list_empty(&father->children)) 681 return; 682 683 reaper = find_new_reaper(father, reaper); 684 list_for_each_entry(p, &father->children, sibling) { 685 for_each_thread(p, t) { 686 t->real_parent = reaper; 687 BUG_ON((!t->ptrace) != (t->parent == father)); 688 if (likely(!t->ptrace)) 689 t->parent = t->real_parent; 690 if (t->pdeath_signal) 691 group_send_sig_info(t->pdeath_signal, 692 SEND_SIG_NOINFO, t, 693 PIDTYPE_TGID); 694 } 695 /* 696 * If this is a threaded reparent there is no need to 697 * notify anyone anything has happened. 698 */ 699 if (!same_thread_group(reaper, father)) 700 reparent_leader(father, p, dead); 701 } 702 list_splice_tail_init(&father->children, &reaper->children); 703 } 704 705 /* 706 * Send signals to all our closest relatives so that they know 707 * to properly mourn us.. 708 */ 709 static void exit_notify(struct task_struct *tsk, int group_dead) 710 { 711 bool autoreap; 712 struct task_struct *p, *n; 713 LIST_HEAD(dead); 714 715 write_lock_irq(&tasklist_lock); 716 forget_original_parent(tsk, &dead); 717 718 if (group_dead) 719 kill_orphaned_pgrp(tsk->group_leader, NULL); 720 721 if (unlikely(tsk->ptrace)) { 722 int sig = thread_group_leader(tsk) && 723 thread_group_empty(tsk) && 724 !ptrace_reparented(tsk) ? 725 tsk->exit_signal : SIGCHLD; 726 autoreap = do_notify_parent(tsk, sig); 727 } else if (thread_group_leader(tsk)) { 728 autoreap = thread_group_empty(tsk) && 729 do_notify_parent(tsk, tsk->exit_signal); 730 } else { 731 autoreap = true; 732 } 733 734 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE; 735 if (tsk->exit_state == EXIT_DEAD) 736 list_add(&tsk->ptrace_entry, &dead); 737 738 /* mt-exec, de_thread() is waiting for group leader */ 739 if (unlikely(tsk->signal->notify_count < 0)) 740 wake_up_process(tsk->signal->group_exit_task); 741 write_unlock_irq(&tasklist_lock); 742 743 list_for_each_entry_safe(p, n, &dead, ptrace_entry) { 744 list_del_init(&p->ptrace_entry); 745 release_task(p); 746 } 747 } 748 749 #ifdef CONFIG_DEBUG_STACK_USAGE 750 static void check_stack_usage(void) 751 { 752 static DEFINE_SPINLOCK(low_water_lock); 753 static int lowest_to_date = THREAD_SIZE; 754 unsigned long free; 755 756 free = stack_not_used(current); 757 758 if (free >= lowest_to_date) 759 return; 760 761 spin_lock(&low_water_lock); 762 if (free < lowest_to_date) { 763 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n", 764 current->comm, task_pid_nr(current), free); 765 lowest_to_date = free; 766 } 767 spin_unlock(&low_water_lock); 768 } 769 #else 770 static inline void check_stack_usage(void) {} 771 #endif 772 773 void __noreturn do_exit(long code) 774 { 775 struct task_struct *tsk = current; 776 int group_dead; 777 778 profile_task_exit(tsk); 779 kcov_task_exit(tsk); 780 781 WARN_ON(blk_needs_flush_plug(tsk)); 782 783 if (unlikely(in_interrupt())) 784 panic("Aiee, killing interrupt handler!"); 785 if (unlikely(!tsk->pid)) 786 panic("Attempted to kill the idle task!"); 787 788 /* 789 * If do_exit is called because this processes oopsed, it's possible 790 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before 791 * continuing. Amongst other possible reasons, this is to prevent 792 * mm_release()->clear_child_tid() from writing to a user-controlled 793 * kernel address. 794 */ 795 set_fs(USER_DS); 796 797 ptrace_event(PTRACE_EVENT_EXIT, code); 798 799 validate_creds_for_do_exit(tsk); 800 801 /* 802 * We're taking recursive faults here in do_exit. Safest is to just 803 * leave this task alone and wait for reboot. 804 */ 805 if (unlikely(tsk->flags & PF_EXITING)) { 806 pr_alert("Fixing recursive fault but reboot is needed!\n"); 807 /* 808 * We can do this unlocked here. The futex code uses 809 * this flag just to verify whether the pi state 810 * cleanup has been done or not. In the worst case it 811 * loops once more. We pretend that the cleanup was 812 * done as there is no way to return. Either the 813 * OWNER_DIED bit is set by now or we push the blocked 814 * task into the wait for ever nirwana as well. 815 */ 816 tsk->flags |= PF_EXITPIDONE; 817 set_current_state(TASK_UNINTERRUPTIBLE); 818 schedule(); 819 } 820 821 exit_signals(tsk); /* sets PF_EXITING */ 822 /* 823 * Ensure that all new tsk->pi_lock acquisitions must observe 824 * PF_EXITING. Serializes against futex.c:attach_to_pi_owner(). 825 */ 826 smp_mb(); 827 /* 828 * Ensure that we must observe the pi_state in exit_mm() -> 829 * mm_release() -> exit_pi_state_list(). 830 */ 831 raw_spin_lock_irq(&tsk->pi_lock); 832 raw_spin_unlock_irq(&tsk->pi_lock); 833 834 if (unlikely(in_atomic())) { 835 pr_info("note: %s[%d] exited with preempt_count %d\n", 836 current->comm, task_pid_nr(current), 837 preempt_count()); 838 preempt_count_set(PREEMPT_ENABLED); 839 } 840 841 /* sync mm's RSS info before statistics gathering */ 842 if (tsk->mm) 843 sync_mm_rss(tsk->mm); 844 acct_update_integrals(tsk); 845 group_dead = atomic_dec_and_test(&tsk->signal->live); 846 if (group_dead) { 847 #ifdef CONFIG_POSIX_TIMERS 848 hrtimer_cancel(&tsk->signal->real_timer); 849 exit_itimers(tsk->signal); 850 #endif 851 if (tsk->mm) 852 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm); 853 } 854 acct_collect(code, group_dead); 855 if (group_dead) 856 tty_audit_exit(); 857 audit_free(tsk); 858 859 tsk->exit_code = code; 860 taskstats_exit(tsk, group_dead); 861 862 exit_mm(); 863 864 if (group_dead) 865 acct_process(); 866 trace_sched_process_exit(tsk); 867 868 exit_sem(tsk); 869 exit_shm(tsk); 870 exit_files(tsk); 871 exit_fs(tsk); 872 if (group_dead) 873 disassociate_ctty(1); 874 exit_task_namespaces(tsk); 875 exit_task_work(tsk); 876 exit_thread(tsk); 877 exit_umh(tsk); 878 879 /* 880 * Flush inherited counters to the parent - before the parent 881 * gets woken up by child-exit notifications. 882 * 883 * because of cgroup mode, must be called before cgroup_exit() 884 */ 885 perf_event_exit_task(tsk); 886 887 sched_autogroup_exit_task(tsk); 888 cgroup_exit(tsk); 889 890 /* 891 * FIXME: do that only when needed, using sched_exit tracepoint 892 */ 893 flush_ptrace_hw_breakpoint(tsk); 894 895 exit_tasks_rcu_start(); 896 exit_notify(tsk, group_dead); 897 proc_exit_connector(tsk); 898 mpol_put_task_policy(tsk); 899 #ifdef CONFIG_FUTEX 900 if (unlikely(current->pi_state_cache)) 901 kfree(current->pi_state_cache); 902 #endif 903 /* 904 * Make sure we are holding no locks: 905 */ 906 debug_check_no_locks_held(); 907 /* 908 * We can do this unlocked here. The futex code uses this flag 909 * just to verify whether the pi state cleanup has been done 910 * or not. In the worst case it loops once more. 911 */ 912 tsk->flags |= PF_EXITPIDONE; 913 914 if (tsk->io_context) 915 exit_io_context(tsk); 916 917 if (tsk->splice_pipe) 918 free_pipe_info(tsk->splice_pipe); 919 920 if (tsk->task_frag.page) 921 put_page(tsk->task_frag.page); 922 923 validate_creds_for_do_exit(tsk); 924 925 check_stack_usage(); 926 preempt_disable(); 927 if (tsk->nr_dirtied) 928 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied); 929 exit_rcu(); 930 exit_tasks_rcu_finish(); 931 932 lockdep_free_task(tsk); 933 do_task_dead(); 934 } 935 EXPORT_SYMBOL_GPL(do_exit); 936 937 void complete_and_exit(struct completion *comp, long code) 938 { 939 if (comp) 940 complete(comp); 941 942 do_exit(code); 943 } 944 EXPORT_SYMBOL(complete_and_exit); 945 946 SYSCALL_DEFINE1(exit, int, error_code) 947 { 948 do_exit((error_code&0xff)<<8); 949 } 950 951 /* 952 * Take down every thread in the group. This is called by fatal signals 953 * as well as by sys_exit_group (below). 954 */ 955 void 956 do_group_exit(int exit_code) 957 { 958 struct signal_struct *sig = current->signal; 959 960 BUG_ON(exit_code & 0x80); /* core dumps don't get here */ 961 962 if (signal_group_exit(sig)) 963 exit_code = sig->group_exit_code; 964 else if (!thread_group_empty(current)) { 965 struct sighand_struct *const sighand = current->sighand; 966 967 spin_lock_irq(&sighand->siglock); 968 if (signal_group_exit(sig)) 969 /* Another thread got here before we took the lock. */ 970 exit_code = sig->group_exit_code; 971 else { 972 sig->group_exit_code = exit_code; 973 sig->flags = SIGNAL_GROUP_EXIT; 974 zap_other_threads(current); 975 } 976 spin_unlock_irq(&sighand->siglock); 977 } 978 979 do_exit(exit_code); 980 /* NOTREACHED */ 981 } 982 983 /* 984 * this kills every thread in the thread group. Note that any externally 985 * wait4()-ing process will get the correct exit code - even if this 986 * thread is not the thread group leader. 987 */ 988 SYSCALL_DEFINE1(exit_group, int, error_code) 989 { 990 do_group_exit((error_code & 0xff) << 8); 991 /* NOTREACHED */ 992 return 0; 993 } 994 995 struct waitid_info { 996 pid_t pid; 997 uid_t uid; 998 int status; 999 int cause; 1000 }; 1001 1002 struct wait_opts { 1003 enum pid_type wo_type; 1004 int wo_flags; 1005 struct pid *wo_pid; 1006 1007 struct waitid_info *wo_info; 1008 int wo_stat; 1009 struct rusage *wo_rusage; 1010 1011 wait_queue_entry_t child_wait; 1012 int notask_error; 1013 }; 1014 1015 static int eligible_pid(struct wait_opts *wo, struct task_struct *p) 1016 { 1017 return wo->wo_type == PIDTYPE_MAX || 1018 task_pid_type(p, wo->wo_type) == wo->wo_pid; 1019 } 1020 1021 static int 1022 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p) 1023 { 1024 if (!eligible_pid(wo, p)) 1025 return 0; 1026 1027 /* 1028 * Wait for all children (clone and not) if __WALL is set or 1029 * if it is traced by us. 1030 */ 1031 if (ptrace || (wo->wo_flags & __WALL)) 1032 return 1; 1033 1034 /* 1035 * Otherwise, wait for clone children *only* if __WCLONE is set; 1036 * otherwise, wait for non-clone children *only*. 1037 * 1038 * Note: a "clone" child here is one that reports to its parent 1039 * using a signal other than SIGCHLD, or a non-leader thread which 1040 * we can only see if it is traced by us. 1041 */ 1042 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE)) 1043 return 0; 1044 1045 return 1; 1046 } 1047 1048 /* 1049 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold 1050 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold 1051 * the lock and this task is uninteresting. If we return nonzero, we have 1052 * released the lock and the system call should return. 1053 */ 1054 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p) 1055 { 1056 int state, status; 1057 pid_t pid = task_pid_vnr(p); 1058 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p)); 1059 struct waitid_info *infop; 1060 1061 if (!likely(wo->wo_flags & WEXITED)) 1062 return 0; 1063 1064 if (unlikely(wo->wo_flags & WNOWAIT)) { 1065 status = p->exit_code; 1066 get_task_struct(p); 1067 read_unlock(&tasklist_lock); 1068 sched_annotate_sleep(); 1069 if (wo->wo_rusage) 1070 getrusage(p, RUSAGE_BOTH, wo->wo_rusage); 1071 put_task_struct(p); 1072 goto out_info; 1073 } 1074 /* 1075 * Move the task's state to DEAD/TRACE, only one thread can do this. 1076 */ 1077 state = (ptrace_reparented(p) && thread_group_leader(p)) ? 1078 EXIT_TRACE : EXIT_DEAD; 1079 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE) 1080 return 0; 1081 /* 1082 * We own this thread, nobody else can reap it. 1083 */ 1084 read_unlock(&tasklist_lock); 1085 sched_annotate_sleep(); 1086 1087 /* 1088 * Check thread_group_leader() to exclude the traced sub-threads. 1089 */ 1090 if (state == EXIT_DEAD && thread_group_leader(p)) { 1091 struct signal_struct *sig = p->signal; 1092 struct signal_struct *psig = current->signal; 1093 unsigned long maxrss; 1094 u64 tgutime, tgstime; 1095 1096 /* 1097 * The resource counters for the group leader are in its 1098 * own task_struct. Those for dead threads in the group 1099 * are in its signal_struct, as are those for the child 1100 * processes it has previously reaped. All these 1101 * accumulate in the parent's signal_struct c* fields. 1102 * 1103 * We don't bother to take a lock here to protect these 1104 * p->signal fields because the whole thread group is dead 1105 * and nobody can change them. 1106 * 1107 * psig->stats_lock also protects us from our sub-theads 1108 * which can reap other children at the same time. Until 1109 * we change k_getrusage()-like users to rely on this lock 1110 * we have to take ->siglock as well. 1111 * 1112 * We use thread_group_cputime_adjusted() to get times for 1113 * the thread group, which consolidates times for all threads 1114 * in the group including the group leader. 1115 */ 1116 thread_group_cputime_adjusted(p, &tgutime, &tgstime); 1117 spin_lock_irq(¤t->sighand->siglock); 1118 write_seqlock(&psig->stats_lock); 1119 psig->cutime += tgutime + sig->cutime; 1120 psig->cstime += tgstime + sig->cstime; 1121 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime; 1122 psig->cmin_flt += 1123 p->min_flt + sig->min_flt + sig->cmin_flt; 1124 psig->cmaj_flt += 1125 p->maj_flt + sig->maj_flt + sig->cmaj_flt; 1126 psig->cnvcsw += 1127 p->nvcsw + sig->nvcsw + sig->cnvcsw; 1128 psig->cnivcsw += 1129 p->nivcsw + sig->nivcsw + sig->cnivcsw; 1130 psig->cinblock += 1131 task_io_get_inblock(p) + 1132 sig->inblock + sig->cinblock; 1133 psig->coublock += 1134 task_io_get_oublock(p) + 1135 sig->oublock + sig->coublock; 1136 maxrss = max(sig->maxrss, sig->cmaxrss); 1137 if (psig->cmaxrss < maxrss) 1138 psig->cmaxrss = maxrss; 1139 task_io_accounting_add(&psig->ioac, &p->ioac); 1140 task_io_accounting_add(&psig->ioac, &sig->ioac); 1141 write_sequnlock(&psig->stats_lock); 1142 spin_unlock_irq(¤t->sighand->siglock); 1143 } 1144 1145 if (wo->wo_rusage) 1146 getrusage(p, RUSAGE_BOTH, wo->wo_rusage); 1147 status = (p->signal->flags & SIGNAL_GROUP_EXIT) 1148 ? p->signal->group_exit_code : p->exit_code; 1149 wo->wo_stat = status; 1150 1151 if (state == EXIT_TRACE) { 1152 write_lock_irq(&tasklist_lock); 1153 /* We dropped tasklist, ptracer could die and untrace */ 1154 ptrace_unlink(p); 1155 1156 /* If parent wants a zombie, don't release it now */ 1157 state = EXIT_ZOMBIE; 1158 if (do_notify_parent(p, p->exit_signal)) 1159 state = EXIT_DEAD; 1160 p->exit_state = state; 1161 write_unlock_irq(&tasklist_lock); 1162 } 1163 if (state == EXIT_DEAD) 1164 release_task(p); 1165 1166 out_info: 1167 infop = wo->wo_info; 1168 if (infop) { 1169 if ((status & 0x7f) == 0) { 1170 infop->cause = CLD_EXITED; 1171 infop->status = status >> 8; 1172 } else { 1173 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED; 1174 infop->status = status & 0x7f; 1175 } 1176 infop->pid = pid; 1177 infop->uid = uid; 1178 } 1179 1180 return pid; 1181 } 1182 1183 static int *task_stopped_code(struct task_struct *p, bool ptrace) 1184 { 1185 if (ptrace) { 1186 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING)) 1187 return &p->exit_code; 1188 } else { 1189 if (p->signal->flags & SIGNAL_STOP_STOPPED) 1190 return &p->signal->group_exit_code; 1191 } 1192 return NULL; 1193 } 1194 1195 /** 1196 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED 1197 * @wo: wait options 1198 * @ptrace: is the wait for ptrace 1199 * @p: task to wait for 1200 * 1201 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED. 1202 * 1203 * CONTEXT: 1204 * read_lock(&tasklist_lock), which is released if return value is 1205 * non-zero. Also, grabs and releases @p->sighand->siglock. 1206 * 1207 * RETURNS: 1208 * 0 if wait condition didn't exist and search for other wait conditions 1209 * should continue. Non-zero return, -errno on failure and @p's pid on 1210 * success, implies that tasklist_lock is released and wait condition 1211 * search should terminate. 1212 */ 1213 static int wait_task_stopped(struct wait_opts *wo, 1214 int ptrace, struct task_struct *p) 1215 { 1216 struct waitid_info *infop; 1217 int exit_code, *p_code, why; 1218 uid_t uid = 0; /* unneeded, required by compiler */ 1219 pid_t pid; 1220 1221 /* 1222 * Traditionally we see ptrace'd stopped tasks regardless of options. 1223 */ 1224 if (!ptrace && !(wo->wo_flags & WUNTRACED)) 1225 return 0; 1226 1227 if (!task_stopped_code(p, ptrace)) 1228 return 0; 1229 1230 exit_code = 0; 1231 spin_lock_irq(&p->sighand->siglock); 1232 1233 p_code = task_stopped_code(p, ptrace); 1234 if (unlikely(!p_code)) 1235 goto unlock_sig; 1236 1237 exit_code = *p_code; 1238 if (!exit_code) 1239 goto unlock_sig; 1240 1241 if (!unlikely(wo->wo_flags & WNOWAIT)) 1242 *p_code = 0; 1243 1244 uid = from_kuid_munged(current_user_ns(), task_uid(p)); 1245 unlock_sig: 1246 spin_unlock_irq(&p->sighand->siglock); 1247 if (!exit_code) 1248 return 0; 1249 1250 /* 1251 * Now we are pretty sure this task is interesting. 1252 * Make sure it doesn't get reaped out from under us while we 1253 * give up the lock and then examine it below. We don't want to 1254 * keep holding onto the tasklist_lock while we call getrusage and 1255 * possibly take page faults for user memory. 1256 */ 1257 get_task_struct(p); 1258 pid = task_pid_vnr(p); 1259 why = ptrace ? CLD_TRAPPED : CLD_STOPPED; 1260 read_unlock(&tasklist_lock); 1261 sched_annotate_sleep(); 1262 if (wo->wo_rusage) 1263 getrusage(p, RUSAGE_BOTH, wo->wo_rusage); 1264 put_task_struct(p); 1265 1266 if (likely(!(wo->wo_flags & WNOWAIT))) 1267 wo->wo_stat = (exit_code << 8) | 0x7f; 1268 1269 infop = wo->wo_info; 1270 if (infop) { 1271 infop->cause = why; 1272 infop->status = exit_code; 1273 infop->pid = pid; 1274 infop->uid = uid; 1275 } 1276 return pid; 1277 } 1278 1279 /* 1280 * Handle do_wait work for one task in a live, non-stopped state. 1281 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold 1282 * the lock and this task is uninteresting. If we return nonzero, we have 1283 * released the lock and the system call should return. 1284 */ 1285 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p) 1286 { 1287 struct waitid_info *infop; 1288 pid_t pid; 1289 uid_t uid; 1290 1291 if (!unlikely(wo->wo_flags & WCONTINUED)) 1292 return 0; 1293 1294 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) 1295 return 0; 1296 1297 spin_lock_irq(&p->sighand->siglock); 1298 /* Re-check with the lock held. */ 1299 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) { 1300 spin_unlock_irq(&p->sighand->siglock); 1301 return 0; 1302 } 1303 if (!unlikely(wo->wo_flags & WNOWAIT)) 1304 p->signal->flags &= ~SIGNAL_STOP_CONTINUED; 1305 uid = from_kuid_munged(current_user_ns(), task_uid(p)); 1306 spin_unlock_irq(&p->sighand->siglock); 1307 1308 pid = task_pid_vnr(p); 1309 get_task_struct(p); 1310 read_unlock(&tasklist_lock); 1311 sched_annotate_sleep(); 1312 if (wo->wo_rusage) 1313 getrusage(p, RUSAGE_BOTH, wo->wo_rusage); 1314 put_task_struct(p); 1315 1316 infop = wo->wo_info; 1317 if (!infop) { 1318 wo->wo_stat = 0xffff; 1319 } else { 1320 infop->cause = CLD_CONTINUED; 1321 infop->pid = pid; 1322 infop->uid = uid; 1323 infop->status = SIGCONT; 1324 } 1325 return pid; 1326 } 1327 1328 /* 1329 * Consider @p for a wait by @parent. 1330 * 1331 * -ECHILD should be in ->notask_error before the first call. 1332 * Returns nonzero for a final return, when we have unlocked tasklist_lock. 1333 * Returns zero if the search for a child should continue; 1334 * then ->notask_error is 0 if @p is an eligible child, 1335 * or still -ECHILD. 1336 */ 1337 static int wait_consider_task(struct wait_opts *wo, int ptrace, 1338 struct task_struct *p) 1339 { 1340 /* 1341 * We can race with wait_task_zombie() from another thread. 1342 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition 1343 * can't confuse the checks below. 1344 */ 1345 int exit_state = READ_ONCE(p->exit_state); 1346 int ret; 1347 1348 if (unlikely(exit_state == EXIT_DEAD)) 1349 return 0; 1350 1351 ret = eligible_child(wo, ptrace, p); 1352 if (!ret) 1353 return ret; 1354 1355 if (unlikely(exit_state == EXIT_TRACE)) { 1356 /* 1357 * ptrace == 0 means we are the natural parent. In this case 1358 * we should clear notask_error, debugger will notify us. 1359 */ 1360 if (likely(!ptrace)) 1361 wo->notask_error = 0; 1362 return 0; 1363 } 1364 1365 if (likely(!ptrace) && unlikely(p->ptrace)) { 1366 /* 1367 * If it is traced by its real parent's group, just pretend 1368 * the caller is ptrace_do_wait() and reap this child if it 1369 * is zombie. 1370 * 1371 * This also hides group stop state from real parent; otherwise 1372 * a single stop can be reported twice as group and ptrace stop. 1373 * If a ptracer wants to distinguish these two events for its 1374 * own children it should create a separate process which takes 1375 * the role of real parent. 1376 */ 1377 if (!ptrace_reparented(p)) 1378 ptrace = 1; 1379 } 1380 1381 /* slay zombie? */ 1382 if (exit_state == EXIT_ZOMBIE) { 1383 /* we don't reap group leaders with subthreads */ 1384 if (!delay_group_leader(p)) { 1385 /* 1386 * A zombie ptracee is only visible to its ptracer. 1387 * Notification and reaping will be cascaded to the 1388 * real parent when the ptracer detaches. 1389 */ 1390 if (unlikely(ptrace) || likely(!p->ptrace)) 1391 return wait_task_zombie(wo, p); 1392 } 1393 1394 /* 1395 * Allow access to stopped/continued state via zombie by 1396 * falling through. Clearing of notask_error is complex. 1397 * 1398 * When !@ptrace: 1399 * 1400 * If WEXITED is set, notask_error should naturally be 1401 * cleared. If not, subset of WSTOPPED|WCONTINUED is set, 1402 * so, if there are live subthreads, there are events to 1403 * wait for. If all subthreads are dead, it's still safe 1404 * to clear - this function will be called again in finite 1405 * amount time once all the subthreads are released and 1406 * will then return without clearing. 1407 * 1408 * When @ptrace: 1409 * 1410 * Stopped state is per-task and thus can't change once the 1411 * target task dies. Only continued and exited can happen. 1412 * Clear notask_error if WCONTINUED | WEXITED. 1413 */ 1414 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED))) 1415 wo->notask_error = 0; 1416 } else { 1417 /* 1418 * @p is alive and it's gonna stop, continue or exit, so 1419 * there always is something to wait for. 1420 */ 1421 wo->notask_error = 0; 1422 } 1423 1424 /* 1425 * Wait for stopped. Depending on @ptrace, different stopped state 1426 * is used and the two don't interact with each other. 1427 */ 1428 ret = wait_task_stopped(wo, ptrace, p); 1429 if (ret) 1430 return ret; 1431 1432 /* 1433 * Wait for continued. There's only one continued state and the 1434 * ptracer can consume it which can confuse the real parent. Don't 1435 * use WCONTINUED from ptracer. You don't need or want it. 1436 */ 1437 return wait_task_continued(wo, p); 1438 } 1439 1440 /* 1441 * Do the work of do_wait() for one thread in the group, @tsk. 1442 * 1443 * -ECHILD should be in ->notask_error before the first call. 1444 * Returns nonzero for a final return, when we have unlocked tasklist_lock. 1445 * Returns zero if the search for a child should continue; then 1446 * ->notask_error is 0 if there were any eligible children, 1447 * or still -ECHILD. 1448 */ 1449 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk) 1450 { 1451 struct task_struct *p; 1452 1453 list_for_each_entry(p, &tsk->children, sibling) { 1454 int ret = wait_consider_task(wo, 0, p); 1455 1456 if (ret) 1457 return ret; 1458 } 1459 1460 return 0; 1461 } 1462 1463 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk) 1464 { 1465 struct task_struct *p; 1466 1467 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) { 1468 int ret = wait_consider_task(wo, 1, p); 1469 1470 if (ret) 1471 return ret; 1472 } 1473 1474 return 0; 1475 } 1476 1477 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode, 1478 int sync, void *key) 1479 { 1480 struct wait_opts *wo = container_of(wait, struct wait_opts, 1481 child_wait); 1482 struct task_struct *p = key; 1483 1484 if (!eligible_pid(wo, p)) 1485 return 0; 1486 1487 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent) 1488 return 0; 1489 1490 return default_wake_function(wait, mode, sync, key); 1491 } 1492 1493 void __wake_up_parent(struct task_struct *p, struct task_struct *parent) 1494 { 1495 __wake_up_sync_key(&parent->signal->wait_chldexit, 1496 TASK_INTERRUPTIBLE, 1, p); 1497 } 1498 1499 static long do_wait(struct wait_opts *wo) 1500 { 1501 struct task_struct *tsk; 1502 int retval; 1503 1504 trace_sched_process_wait(wo->wo_pid); 1505 1506 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback); 1507 wo->child_wait.private = current; 1508 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait); 1509 repeat: 1510 /* 1511 * If there is nothing that can match our criteria, just get out. 1512 * We will clear ->notask_error to zero if we see any child that 1513 * might later match our criteria, even if we are not able to reap 1514 * it yet. 1515 */ 1516 wo->notask_error = -ECHILD; 1517 if ((wo->wo_type < PIDTYPE_MAX) && 1518 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type]))) 1519 goto notask; 1520 1521 set_current_state(TASK_INTERRUPTIBLE); 1522 read_lock(&tasklist_lock); 1523 tsk = current; 1524 do { 1525 retval = do_wait_thread(wo, tsk); 1526 if (retval) 1527 goto end; 1528 1529 retval = ptrace_do_wait(wo, tsk); 1530 if (retval) 1531 goto end; 1532 1533 if (wo->wo_flags & __WNOTHREAD) 1534 break; 1535 } while_each_thread(current, tsk); 1536 read_unlock(&tasklist_lock); 1537 1538 notask: 1539 retval = wo->notask_error; 1540 if (!retval && !(wo->wo_flags & WNOHANG)) { 1541 retval = -ERESTARTSYS; 1542 if (!signal_pending(current)) { 1543 schedule(); 1544 goto repeat; 1545 } 1546 } 1547 end: 1548 __set_current_state(TASK_RUNNING); 1549 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait); 1550 return retval; 1551 } 1552 1553 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop, 1554 int options, struct rusage *ru) 1555 { 1556 struct wait_opts wo; 1557 struct pid *pid = NULL; 1558 enum pid_type type; 1559 long ret; 1560 1561 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED| 1562 __WNOTHREAD|__WCLONE|__WALL)) 1563 return -EINVAL; 1564 if (!(options & (WEXITED|WSTOPPED|WCONTINUED))) 1565 return -EINVAL; 1566 1567 switch (which) { 1568 case P_ALL: 1569 type = PIDTYPE_MAX; 1570 break; 1571 case P_PID: 1572 type = PIDTYPE_PID; 1573 if (upid <= 0) 1574 return -EINVAL; 1575 break; 1576 case P_PGID: 1577 type = PIDTYPE_PGID; 1578 if (upid <= 0) 1579 return -EINVAL; 1580 break; 1581 default: 1582 return -EINVAL; 1583 } 1584 1585 if (type < PIDTYPE_MAX) 1586 pid = find_get_pid(upid); 1587 1588 wo.wo_type = type; 1589 wo.wo_pid = pid; 1590 wo.wo_flags = options; 1591 wo.wo_info = infop; 1592 wo.wo_rusage = ru; 1593 ret = do_wait(&wo); 1594 1595 put_pid(pid); 1596 return ret; 1597 } 1598 1599 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *, 1600 infop, int, options, struct rusage __user *, ru) 1601 { 1602 struct rusage r; 1603 struct waitid_info info = {.status = 0}; 1604 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL); 1605 int signo = 0; 1606 1607 if (err > 0) { 1608 signo = SIGCHLD; 1609 err = 0; 1610 if (ru && copy_to_user(ru, &r, sizeof(struct rusage))) 1611 return -EFAULT; 1612 } 1613 if (!infop) 1614 return err; 1615 1616 if (!user_access_begin(infop, sizeof(*infop))) 1617 return -EFAULT; 1618 1619 unsafe_put_user(signo, &infop->si_signo, Efault); 1620 unsafe_put_user(0, &infop->si_errno, Efault); 1621 unsafe_put_user(info.cause, &infop->si_code, Efault); 1622 unsafe_put_user(info.pid, &infop->si_pid, Efault); 1623 unsafe_put_user(info.uid, &infop->si_uid, Efault); 1624 unsafe_put_user(info.status, &infop->si_status, Efault); 1625 user_access_end(); 1626 return err; 1627 Efault: 1628 user_access_end(); 1629 return -EFAULT; 1630 } 1631 1632 long kernel_wait4(pid_t upid, int __user *stat_addr, int options, 1633 struct rusage *ru) 1634 { 1635 struct wait_opts wo; 1636 struct pid *pid = NULL; 1637 enum pid_type type; 1638 long ret; 1639 1640 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED| 1641 __WNOTHREAD|__WCLONE|__WALL)) 1642 return -EINVAL; 1643 1644 /* -INT_MIN is not defined */ 1645 if (upid == INT_MIN) 1646 return -ESRCH; 1647 1648 if (upid == -1) 1649 type = PIDTYPE_MAX; 1650 else if (upid < 0) { 1651 type = PIDTYPE_PGID; 1652 pid = find_get_pid(-upid); 1653 } else if (upid == 0) { 1654 type = PIDTYPE_PGID; 1655 pid = get_task_pid(current, PIDTYPE_PGID); 1656 } else /* upid > 0 */ { 1657 type = PIDTYPE_PID; 1658 pid = find_get_pid(upid); 1659 } 1660 1661 wo.wo_type = type; 1662 wo.wo_pid = pid; 1663 wo.wo_flags = options | WEXITED; 1664 wo.wo_info = NULL; 1665 wo.wo_stat = 0; 1666 wo.wo_rusage = ru; 1667 ret = do_wait(&wo); 1668 put_pid(pid); 1669 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr)) 1670 ret = -EFAULT; 1671 1672 return ret; 1673 } 1674 1675 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr, 1676 int, options, struct rusage __user *, ru) 1677 { 1678 struct rusage r; 1679 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL); 1680 1681 if (err > 0) { 1682 if (ru && copy_to_user(ru, &r, sizeof(struct rusage))) 1683 return -EFAULT; 1684 } 1685 return err; 1686 } 1687 1688 #ifdef __ARCH_WANT_SYS_WAITPID 1689 1690 /* 1691 * sys_waitpid() remains for compatibility. waitpid() should be 1692 * implemented by calling sys_wait4() from libc.a. 1693 */ 1694 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options) 1695 { 1696 return kernel_wait4(pid, stat_addr, options, NULL); 1697 } 1698 1699 #endif 1700 1701 #ifdef CONFIG_COMPAT 1702 COMPAT_SYSCALL_DEFINE4(wait4, 1703 compat_pid_t, pid, 1704 compat_uint_t __user *, stat_addr, 1705 int, options, 1706 struct compat_rusage __user *, ru) 1707 { 1708 struct rusage r; 1709 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL); 1710 if (err > 0) { 1711 if (ru && put_compat_rusage(&r, ru)) 1712 return -EFAULT; 1713 } 1714 return err; 1715 } 1716 1717 COMPAT_SYSCALL_DEFINE5(waitid, 1718 int, which, compat_pid_t, pid, 1719 struct compat_siginfo __user *, infop, int, options, 1720 struct compat_rusage __user *, uru) 1721 { 1722 struct rusage ru; 1723 struct waitid_info info = {.status = 0}; 1724 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL); 1725 int signo = 0; 1726 if (err > 0) { 1727 signo = SIGCHLD; 1728 err = 0; 1729 if (uru) { 1730 /* kernel_waitid() overwrites everything in ru */ 1731 if (COMPAT_USE_64BIT_TIME) 1732 err = copy_to_user(uru, &ru, sizeof(ru)); 1733 else 1734 err = put_compat_rusage(&ru, uru); 1735 if (err) 1736 return -EFAULT; 1737 } 1738 } 1739 1740 if (!infop) 1741 return err; 1742 1743 if (!user_access_begin(infop, sizeof(*infop))) 1744 return -EFAULT; 1745 1746 unsafe_put_user(signo, &infop->si_signo, Efault); 1747 unsafe_put_user(0, &infop->si_errno, Efault); 1748 unsafe_put_user(info.cause, &infop->si_code, Efault); 1749 unsafe_put_user(info.pid, &infop->si_pid, Efault); 1750 unsafe_put_user(info.uid, &infop->si_uid, Efault); 1751 unsafe_put_user(info.status, &infop->si_status, Efault); 1752 user_access_end(); 1753 return err; 1754 Efault: 1755 user_access_end(); 1756 return -EFAULT; 1757 } 1758 #endif 1759 1760 __weak void abort(void) 1761 { 1762 BUG(); 1763 1764 /* if that doesn't kill us, halt */ 1765 panic("Oops failed to kill thread"); 1766 } 1767 EXPORT_SYMBOL(abort); 1768