1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * linux/kernel/exit.c 4 * 5 * Copyright (C) 1991, 1992 Linus Torvalds 6 */ 7 8 #include <linux/mm.h> 9 #include <linux/slab.h> 10 #include <linux/sched/autogroup.h> 11 #include <linux/sched/mm.h> 12 #include <linux/sched/stat.h> 13 #include <linux/sched/task.h> 14 #include <linux/sched/task_stack.h> 15 #include <linux/sched/cputime.h> 16 #include <linux/interrupt.h> 17 #include <linux/module.h> 18 #include <linux/capability.h> 19 #include <linux/completion.h> 20 #include <linux/personality.h> 21 #include <linux/tty.h> 22 #include <linux/iocontext.h> 23 #include <linux/key.h> 24 #include <linux/cpu.h> 25 #include <linux/acct.h> 26 #include <linux/tsacct_kern.h> 27 #include <linux/file.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/task_io_accounting_ops.h> 51 #include <linux/blkdev.h> 52 #include <linux/task_work.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/kmsan.h> 63 #include <linux/random.h> 64 #include <linux/rcuwait.h> 65 #include <linux/compat.h> 66 #include <linux/io_uring.h> 67 #include <linux/kprobes.h> 68 #include <linux/rethook.h> 69 #include <linux/sysfs.h> 70 #include <linux/user_events.h> 71 #include <linux/unwind_deferred.h> 72 #include <linux/uaccess.h> 73 #include <linux/pidfs.h> 74 75 #include <uapi/linux/wait.h> 76 77 #include <asm/unistd.h> 78 #include <asm/mmu_context.h> 79 80 #include "exit.h" 81 82 /* 83 * The default value should be high enough to not crash a system that randomly 84 * crashes its kernel from time to time, but low enough to at least not permit 85 * overflowing 32-bit refcounts or the ldsem writer count. 86 */ 87 static unsigned int oops_limit = 10000; 88 89 #ifdef CONFIG_SYSCTL 90 static const struct ctl_table kern_exit_table[] = { 91 { 92 .procname = "oops_limit", 93 .data = &oops_limit, 94 .maxlen = sizeof(oops_limit), 95 .mode = 0644, 96 .proc_handler = proc_douintvec, 97 }, 98 }; 99 100 static __init int kernel_exit_sysctls_init(void) 101 { 102 register_sysctl_init("kernel", kern_exit_table); 103 return 0; 104 } 105 late_initcall(kernel_exit_sysctls_init); 106 #endif 107 108 static atomic_t oops_count = ATOMIC_INIT(0); 109 110 #ifdef CONFIG_SYSFS 111 static ssize_t oops_count_show(struct kobject *kobj, struct kobj_attribute *attr, 112 char *page) 113 { 114 return sysfs_emit(page, "%d\n", atomic_read(&oops_count)); 115 } 116 117 static struct kobj_attribute oops_count_attr = __ATTR_RO(oops_count); 118 119 static __init int kernel_exit_sysfs_init(void) 120 { 121 sysfs_add_file_to_group(kernel_kobj, &oops_count_attr.attr, NULL); 122 return 0; 123 } 124 late_initcall(kernel_exit_sysfs_init); 125 #endif 126 127 /* 128 * For things release_task() would like to do *after* tasklist_lock is released. 129 */ 130 struct release_task_post { 131 struct pid *pids[PIDTYPE_MAX]; 132 }; 133 134 static void __unhash_process(struct release_task_post *post, struct task_struct *p, 135 bool group_dead) 136 { 137 struct pid *pid = task_pid(p); 138 139 nr_threads--; 140 141 detach_pid(post->pids, p, PIDTYPE_PID); 142 wake_up_all(&pid->wait_pidfd); 143 144 if (group_dead) { 145 detach_pid(post->pids, p, PIDTYPE_TGID); 146 detach_pid(post->pids, p, PIDTYPE_PGID); 147 detach_pid(post->pids, p, PIDTYPE_SID); 148 149 list_del_rcu(&p->tasks); 150 list_del_init(&p->sibling); 151 __this_cpu_dec(process_counts); 152 } 153 list_del_rcu(&p->thread_node); 154 } 155 156 /* 157 * This function expects the tasklist_lock write-locked. 158 */ 159 static void __exit_signal(struct release_task_post *post, struct task_struct *tsk) 160 { 161 struct signal_struct *sig = tsk->signal; 162 bool group_dead = thread_group_leader(tsk); 163 struct sighand_struct *sighand; 164 struct tty_struct *tty; 165 u64 utime, stime; 166 167 sighand = rcu_dereference_check(tsk->sighand, 168 lockdep_tasklist_lock_is_held()); 169 spin_lock(&sighand->siglock); 170 171 #ifdef CONFIG_POSIX_TIMERS 172 posix_cpu_timers_exit(tsk); 173 if (group_dead) 174 posix_cpu_timers_exit_group(tsk); 175 #endif 176 177 if (group_dead) { 178 tty = sig->tty; 179 sig->tty = NULL; 180 } else { 181 /* 182 * If there is any task waiting for the group exit 183 * then notify it: 184 */ 185 if (sig->notify_count > 0 && !--sig->notify_count) 186 wake_up_process(sig->group_exec_task); 187 188 if (tsk == sig->curr_target) 189 sig->curr_target = next_thread(tsk); 190 } 191 192 /* 193 * Accumulate here the counters for all threads as they die. We could 194 * skip the group leader because it is the last user of signal_struct, 195 * but we want to avoid the race with thread_group_cputime() which can 196 * see the empty ->thread_head list. 197 */ 198 task_cputime(tsk, &utime, &stime); 199 write_seqlock(&sig->stats_lock); 200 sig->utime += utime; 201 sig->stime += stime; 202 sig->gtime += task_gtime(tsk); 203 sig->min_flt += tsk->min_flt; 204 sig->maj_flt += tsk->maj_flt; 205 sig->nvcsw += tsk->nvcsw; 206 sig->nivcsw += tsk->nivcsw; 207 sig->inblock += task_io_get_inblock(tsk); 208 sig->oublock += task_io_get_oublock(tsk); 209 task_io_accounting_add(&sig->ioac, &tsk->ioac); 210 sig->sum_sched_runtime += tsk->se.sum_exec_runtime; 211 sig->nr_threads--; 212 __unhash_process(post, tsk, group_dead); 213 write_sequnlock(&sig->stats_lock); 214 215 tsk->sighand = NULL; 216 spin_unlock(&sighand->siglock); 217 218 __cleanup_sighand(sighand); 219 if (group_dead) 220 tty_kref_put(tty); 221 } 222 223 static void delayed_put_task_struct(struct rcu_head *rhp) 224 { 225 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu); 226 227 kprobe_flush_task(tsk); 228 rethook_flush_task(tsk); 229 perf_event_delayed_put(tsk); 230 trace_sched_process_free(tsk); 231 put_task_struct(tsk); 232 } 233 234 void put_task_struct_rcu_user(struct task_struct *task) 235 { 236 if (refcount_dec_and_test(&task->rcu_users)) 237 call_rcu(&task->rcu, delayed_put_task_struct); 238 } 239 240 void __weak release_thread(struct task_struct *dead_task) 241 { 242 } 243 244 void release_task(struct task_struct *p) 245 { 246 struct release_task_post post; 247 struct task_struct *leader; 248 struct pid *thread_pid; 249 int zap_leader; 250 repeat: 251 memset(&post, 0, sizeof(post)); 252 253 /* don't need to get the RCU readlock here - the process is dead and 254 * can't be modifying its own credentials. But shut RCU-lockdep up */ 255 rcu_read_lock(); 256 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1); 257 rcu_read_unlock(); 258 259 pidfs_exit(p); 260 cgroup_release(p); 261 262 /* Retrieve @thread_pid before __unhash_process() may set it to NULL. */ 263 thread_pid = task_pid(p); 264 265 write_lock_irq(&tasklist_lock); 266 ptrace_release_task(p); 267 __exit_signal(&post, p); 268 269 /* 270 * If we are the last non-leader member of the thread 271 * group, and the leader is zombie, then notify the 272 * group leader's parent process. (if it wants notification.) 273 */ 274 zap_leader = 0; 275 leader = p->group_leader; 276 if (leader != p && thread_group_empty(leader) 277 && leader->exit_state == EXIT_ZOMBIE) { 278 /* for pidfs_exit() and do_notify_parent() */ 279 if (leader->signal->flags & SIGNAL_GROUP_EXIT) 280 leader->exit_code = leader->signal->group_exit_code; 281 /* 282 * If we were the last child thread and the leader has 283 * exited already, and the leader's parent ignores SIGCHLD, 284 * then we are the one who should release the leader. 285 */ 286 zap_leader = do_notify_parent(leader, leader->exit_signal); 287 if (zap_leader) 288 leader->exit_state = EXIT_DEAD; 289 } 290 291 write_unlock_irq(&tasklist_lock); 292 /* @thread_pid can't go away until free_pids() below */ 293 proc_flush_pid(thread_pid); 294 add_device_randomness(&p->se.sum_exec_runtime, 295 sizeof(p->se.sum_exec_runtime)); 296 free_pids(post.pids); 297 release_thread(p); 298 /* 299 * This task was already removed from the process/thread/pid lists 300 * and lock_task_sighand(p) can't succeed. Nobody else can touch 301 * ->pending or, if group dead, signal->shared_pending. We can call 302 * flush_sigqueue() lockless. 303 */ 304 flush_sigqueue(&p->pending); 305 if (thread_group_leader(p)) 306 flush_sigqueue(&p->signal->shared_pending); 307 308 put_task_struct_rcu_user(p); 309 310 p = leader; 311 if (unlikely(zap_leader)) 312 goto repeat; 313 } 314 315 int rcuwait_wake_up(struct rcuwait *w) 316 { 317 int ret = 0; 318 struct task_struct *task; 319 320 rcu_read_lock(); 321 322 /* 323 * Order condition vs @task, such that everything prior to the load 324 * of @task is visible. This is the condition as to why the user called 325 * rcuwait_wake() in the first place. Pairs with set_current_state() 326 * barrier (A) in rcuwait_wait_event(). 327 * 328 * WAIT WAKE 329 * [S] tsk = current [S] cond = true 330 * MB (A) MB (B) 331 * [L] cond [L] tsk 332 */ 333 smp_mb(); /* (B) */ 334 335 task = rcu_dereference(w->task); 336 if (task) 337 ret = wake_up_process(task); 338 rcu_read_unlock(); 339 340 return ret; 341 } 342 EXPORT_SYMBOL_GPL(rcuwait_wake_up); 343 344 /* 345 * Determine if a process group is "orphaned", according to the POSIX 346 * definition in 2.2.2.52. Orphaned process groups are not to be affected 347 * by terminal-generated stop signals. Newly orphaned process groups are 348 * to receive a SIGHUP and a SIGCONT. 349 * 350 * "I ask you, have you ever known what it is to be an orphan?" 351 */ 352 static int will_become_orphaned_pgrp(struct pid *pgrp, 353 struct task_struct *ignored_task) 354 { 355 struct task_struct *p; 356 357 do_each_pid_task(pgrp, PIDTYPE_PGID, p) { 358 if ((p == ignored_task) || 359 (p->exit_state && thread_group_empty(p)) || 360 is_global_init(p->real_parent)) 361 continue; 362 363 if (task_pgrp(p->real_parent) != pgrp && 364 task_session(p->real_parent) == task_session(p)) 365 return 0; 366 } while_each_pid_task(pgrp, PIDTYPE_PGID, p); 367 368 return 1; 369 } 370 371 int is_current_pgrp_orphaned(void) 372 { 373 int retval; 374 375 read_lock(&tasklist_lock); 376 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL); 377 read_unlock(&tasklist_lock); 378 379 return retval; 380 } 381 382 static bool has_stopped_jobs(struct pid *pgrp) 383 { 384 struct task_struct *p; 385 386 do_each_pid_task(pgrp, PIDTYPE_PGID, p) { 387 if (p->signal->flags & SIGNAL_STOP_STOPPED) 388 return true; 389 } while_each_pid_task(pgrp, PIDTYPE_PGID, p); 390 391 return false; 392 } 393 394 /* 395 * Check to see if any process groups have become orphaned as 396 * a result of our exiting, and if they have any stopped jobs, 397 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2) 398 */ 399 static void 400 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent) 401 { 402 struct pid *pgrp = task_pgrp(tsk); 403 struct task_struct *ignored_task = tsk; 404 405 if (!parent) 406 /* exit: our father is in a different pgrp than 407 * we are and we were the only connection outside. 408 */ 409 parent = tsk->real_parent; 410 else 411 /* reparent: our child is in a different pgrp than 412 * we are, and it was the only connection outside. 413 */ 414 ignored_task = NULL; 415 416 if (task_pgrp(parent) != pgrp && 417 task_session(parent) == task_session(tsk) && 418 will_become_orphaned_pgrp(pgrp, ignored_task) && 419 has_stopped_jobs(pgrp)) { 420 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp); 421 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp); 422 } 423 } 424 425 static void coredump_task_exit(struct task_struct *tsk, 426 struct core_state *core_state) 427 { 428 struct core_thread self; 429 430 self.task = tsk; 431 if (self.task->flags & PF_SIGNALED) 432 self.next = xchg(&core_state->dumper.next, &self); 433 else 434 self.task = NULL; 435 /* 436 * Implies mb(), the result of xchg() must be visible 437 * to core_state->dumper. 438 */ 439 if (atomic_dec_and_test(&core_state->nr_threads)) 440 complete(&core_state->startup); 441 442 for (;;) { 443 set_current_state(TASK_IDLE|TASK_FREEZABLE); 444 if (!self.task) /* see coredump_finish() */ 445 break; 446 schedule(); 447 } 448 __set_current_state(TASK_RUNNING); 449 } 450 451 #ifdef CONFIG_MEMCG 452 /* drops tasklist_lock if succeeds */ 453 static bool __try_to_set_owner(struct task_struct *tsk, struct mm_struct *mm) 454 { 455 bool ret = false; 456 457 task_lock(tsk); 458 if (likely(tsk->mm == mm)) { 459 /* tsk can't pass exit_mm/exec_mmap and exit */ 460 read_unlock(&tasklist_lock); 461 WRITE_ONCE(mm->owner, tsk); 462 lru_gen_migrate_mm(mm); 463 ret = true; 464 } 465 task_unlock(tsk); 466 return ret; 467 } 468 469 static bool try_to_set_owner(struct task_struct *g, struct mm_struct *mm) 470 { 471 struct task_struct *t; 472 473 for_each_thread(g, t) { 474 struct mm_struct *t_mm = READ_ONCE(t->mm); 475 if (t_mm == mm) { 476 if (__try_to_set_owner(t, mm)) 477 return true; 478 } else if (t_mm) 479 break; 480 } 481 482 return false; 483 } 484 485 /* 486 * A task is exiting. If it owned this mm, find a new owner for the mm. 487 */ 488 void mm_update_next_owner(struct mm_struct *mm) 489 { 490 struct task_struct *g, *p = current; 491 492 /* 493 * If the exiting or execing task is not the owner, it's 494 * someone else's problem. 495 */ 496 if (mm->owner != p) 497 return; 498 /* 499 * The current owner is exiting/execing and there are no other 500 * candidates. Do not leave the mm pointing to a possibly 501 * freed task structure. 502 */ 503 if (atomic_read(&mm->mm_users) <= 1) { 504 WRITE_ONCE(mm->owner, NULL); 505 return; 506 } 507 508 read_lock(&tasklist_lock); 509 /* 510 * Search in the children 511 */ 512 list_for_each_entry(g, &p->children, sibling) { 513 if (try_to_set_owner(g, mm)) 514 goto ret; 515 } 516 /* 517 * Search in the siblings 518 */ 519 list_for_each_entry(g, &p->real_parent->children, sibling) { 520 if (try_to_set_owner(g, mm)) 521 goto ret; 522 } 523 /* 524 * Search through everything else, we should not get here often. 525 */ 526 for_each_process(g) { 527 if (atomic_read(&mm->mm_users) <= 1) 528 break; 529 if (g->flags & PF_KTHREAD) 530 continue; 531 if (try_to_set_owner(g, mm)) 532 goto ret; 533 } 534 read_unlock(&tasklist_lock); 535 /* 536 * We found no owner yet mm_users > 1: this implies that we are 537 * most likely racing with swapoff (try_to_unuse()) or /proc or 538 * ptrace or page migration (get_task_mm()). Mark owner as NULL. 539 */ 540 WRITE_ONCE(mm->owner, NULL); 541 ret: 542 return; 543 544 } 545 #endif /* CONFIG_MEMCG */ 546 547 /* 548 * Turn us into a lazy TLB process if we 549 * aren't already.. 550 */ 551 static void exit_mm(void) 552 { 553 struct mm_struct *mm = current->mm; 554 555 exit_mm_release(current, mm); 556 if (!mm) 557 return; 558 mmap_read_lock(mm); 559 mmgrab_lazy_tlb(mm); 560 BUG_ON(mm != current->active_mm); 561 /* more a memory barrier than a real lock */ 562 task_lock(current); 563 /* 564 * When a thread stops operating on an address space, the loop 565 * in membarrier_private_expedited() may not observe that 566 * tsk->mm, and the loop in membarrier_global_expedited() may 567 * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED 568 * rq->membarrier_state, so those would not issue an IPI. 569 * Membarrier requires a memory barrier after accessing 570 * user-space memory, before clearing tsk->mm or the 571 * rq->membarrier_state. 572 */ 573 smp_mb__after_spinlock(); 574 local_irq_disable(); 575 current->mm = NULL; 576 membarrier_update_current_mm(NULL); 577 enter_lazy_tlb(mm, current); 578 local_irq_enable(); 579 task_unlock(current); 580 mmap_read_unlock(mm); 581 mm_update_next_owner(mm); 582 mmput(mm); 583 if (test_thread_flag(TIF_MEMDIE)) 584 exit_oom_victim(); 585 } 586 587 static struct task_struct *find_alive_thread(struct task_struct *p) 588 { 589 struct task_struct *t; 590 591 for_each_thread(p, t) { 592 if (!(t->flags & PF_EXITING)) 593 return t; 594 } 595 return NULL; 596 } 597 598 static struct task_struct *find_child_reaper(struct task_struct *father, 599 struct list_head *dead) 600 __releases(&tasklist_lock) 601 __acquires(&tasklist_lock) 602 { 603 struct pid_namespace *pid_ns = task_active_pid_ns(father); 604 struct task_struct *reaper = pid_ns->child_reaper; 605 struct task_struct *p, *n; 606 607 if (likely(reaper != father)) 608 return reaper; 609 610 reaper = find_alive_thread(father); 611 if (reaper) { 612 pid_ns->child_reaper = reaper; 613 return reaper; 614 } 615 616 write_unlock_irq(&tasklist_lock); 617 618 list_for_each_entry_safe(p, n, dead, ptrace_entry) { 619 list_del_init(&p->ptrace_entry); 620 release_task(p); 621 } 622 623 zap_pid_ns_processes(pid_ns); 624 write_lock_irq(&tasklist_lock); 625 626 return father; 627 } 628 629 /* 630 * When we die, we re-parent all our children, and try to: 631 * 1. give them to another thread in our thread group, if such a member exists 632 * 2. give it to the first ancestor process which prctl'd itself as a 633 * child_subreaper for its children (like a service manager) 634 * 3. give it to the init process (PID 1) in our pid namespace 635 */ 636 static struct task_struct *find_new_reaper(struct task_struct *father, 637 struct task_struct *child_reaper) 638 { 639 struct task_struct *thread, *reaper; 640 641 thread = find_alive_thread(father); 642 if (thread) 643 return thread; 644 645 if (father->signal->has_child_subreaper) { 646 unsigned int ns_level = task_pid(father)->level; 647 /* 648 * Find the first ->is_child_subreaper ancestor in our pid_ns. 649 * We can't check reaper != child_reaper to ensure we do not 650 * cross the namespaces, the exiting parent could be injected 651 * by setns() + fork(). 652 * We check pid->level, this is slightly more efficient than 653 * task_active_pid_ns(reaper) != task_active_pid_ns(father). 654 */ 655 for (reaper = father->real_parent; 656 task_pid(reaper)->level == ns_level; 657 reaper = reaper->real_parent) { 658 if (reaper == &init_task) 659 break; 660 if (!reaper->signal->is_child_subreaper) 661 continue; 662 thread = find_alive_thread(reaper); 663 if (thread) 664 return thread; 665 } 666 } 667 668 return child_reaper; 669 } 670 671 /* 672 * Any that need to be release_task'd are put on the @dead list. 673 */ 674 static void reparent_leader(struct task_struct *father, struct task_struct *p, 675 struct list_head *dead) 676 { 677 if (unlikely(p->exit_state == EXIT_DEAD)) 678 return; 679 680 /* We don't want people slaying init. */ 681 p->exit_signal = SIGCHLD; 682 683 /* If it has exited notify the new parent about this child's death. */ 684 if (!p->ptrace && 685 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) { 686 if (do_notify_parent(p, p->exit_signal)) { 687 p->exit_state = EXIT_DEAD; 688 list_add(&p->ptrace_entry, dead); 689 } 690 } 691 692 kill_orphaned_pgrp(p, father); 693 } 694 695 /* 696 * Make init inherit all the child processes 697 */ 698 static void forget_original_parent(struct task_struct *father, 699 struct list_head *dead) 700 { 701 struct task_struct *p, *t, *reaper; 702 703 if (unlikely(!list_empty(&father->ptraced))) 704 exit_ptrace(father, dead); 705 706 /* Can drop and reacquire tasklist_lock */ 707 reaper = find_child_reaper(father, dead); 708 if (list_empty(&father->children)) 709 return; 710 711 reaper = find_new_reaper(father, reaper); 712 list_for_each_entry(p, &father->children, sibling) { 713 for_each_thread(p, t) { 714 RCU_INIT_POINTER(t->real_parent, reaper); 715 BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father)); 716 if (likely(!t->ptrace)) 717 t->parent = t->real_parent; 718 if (t->pdeath_signal) 719 group_send_sig_info(t->pdeath_signal, 720 SEND_SIG_NOINFO, t, 721 PIDTYPE_TGID); 722 } 723 /* 724 * If this is a threaded reparent there is no need to 725 * notify anyone anything has happened. 726 */ 727 if (!same_thread_group(reaper, father)) 728 reparent_leader(father, p, dead); 729 } 730 list_splice_tail_init(&father->children, &reaper->children); 731 } 732 733 /* 734 * Send signals to all our closest relatives so that they know 735 * to properly mourn us.. 736 */ 737 static void exit_notify(struct task_struct *tsk, int group_dead) 738 { 739 bool autoreap; 740 struct task_struct *p, *n; 741 LIST_HEAD(dead); 742 743 write_lock_irq(&tasklist_lock); 744 forget_original_parent(tsk, &dead); 745 746 if (group_dead) 747 kill_orphaned_pgrp(tsk->group_leader, NULL); 748 749 tsk->exit_state = EXIT_ZOMBIE; 750 751 if (unlikely(tsk->ptrace)) { 752 int sig = thread_group_leader(tsk) && 753 thread_group_empty(tsk) && 754 !ptrace_reparented(tsk) ? 755 tsk->exit_signal : SIGCHLD; 756 autoreap = do_notify_parent(tsk, sig); 757 } else if (thread_group_leader(tsk)) { 758 autoreap = thread_group_empty(tsk) && 759 do_notify_parent(tsk, tsk->exit_signal); 760 } else { 761 autoreap = true; 762 /* untraced sub-thread */ 763 do_notify_pidfd(tsk); 764 } 765 766 if (autoreap) { 767 tsk->exit_state = EXIT_DEAD; 768 list_add(&tsk->ptrace_entry, &dead); 769 } 770 771 /* mt-exec, de_thread() is waiting for group leader */ 772 if (unlikely(tsk->signal->notify_count < 0)) 773 wake_up_process(tsk->signal->group_exec_task); 774 write_unlock_irq(&tasklist_lock); 775 776 list_for_each_entry_safe(p, n, &dead, ptrace_entry) { 777 list_del_init(&p->ptrace_entry); 778 release_task(p); 779 } 780 } 781 782 #ifdef CONFIG_DEBUG_STACK_USAGE 783 unsigned long stack_not_used(struct task_struct *p) 784 { 785 unsigned long *n = end_of_stack(p); 786 787 do { /* Skip over canary */ 788 # ifdef CONFIG_STACK_GROWSUP 789 n--; 790 # else 791 n++; 792 # endif 793 } while (!*n); 794 795 # ifdef CONFIG_STACK_GROWSUP 796 return (unsigned long)end_of_stack(p) - (unsigned long)n; 797 # else 798 return (unsigned long)n - (unsigned long)end_of_stack(p); 799 # endif 800 } 801 802 /* Count the maximum pages reached in kernel stacks */ 803 static inline void kstack_histogram(unsigned long used_stack) 804 { 805 #ifdef CONFIG_VM_EVENT_COUNTERS 806 if (used_stack <= 1024) 807 count_vm_event(KSTACK_1K); 808 #if THREAD_SIZE > 1024 809 else if (used_stack <= 2048) 810 count_vm_event(KSTACK_2K); 811 #endif 812 #if THREAD_SIZE > 2048 813 else if (used_stack <= 4096) 814 count_vm_event(KSTACK_4K); 815 #endif 816 #if THREAD_SIZE > 4096 817 else if (used_stack <= 8192) 818 count_vm_event(KSTACK_8K); 819 #endif 820 #if THREAD_SIZE > 8192 821 else if (used_stack <= 16384) 822 count_vm_event(KSTACK_16K); 823 #endif 824 #if THREAD_SIZE > 16384 825 else if (used_stack <= 32768) 826 count_vm_event(KSTACK_32K); 827 #endif 828 #if THREAD_SIZE > 32768 829 else if (used_stack <= 65536) 830 count_vm_event(KSTACK_64K); 831 #endif 832 #if THREAD_SIZE > 65536 833 else 834 count_vm_event(KSTACK_REST); 835 #endif 836 #endif /* CONFIG_VM_EVENT_COUNTERS */ 837 } 838 839 static void check_stack_usage(void) 840 { 841 static DEFINE_SPINLOCK(low_water_lock); 842 static int lowest_to_date = THREAD_SIZE; 843 unsigned long free; 844 845 free = stack_not_used(current); 846 kstack_histogram(THREAD_SIZE - free); 847 848 if (free >= lowest_to_date) 849 return; 850 851 spin_lock(&low_water_lock); 852 if (free < lowest_to_date) { 853 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n", 854 current->comm, task_pid_nr(current), free); 855 lowest_to_date = free; 856 } 857 spin_unlock(&low_water_lock); 858 } 859 #else 860 static inline void check_stack_usage(void) {} 861 #endif 862 863 static void synchronize_group_exit(struct task_struct *tsk, long code) 864 { 865 struct sighand_struct *sighand = tsk->sighand; 866 struct signal_struct *signal = tsk->signal; 867 struct core_state *core_state; 868 869 spin_lock_irq(&sighand->siglock); 870 signal->quick_threads--; 871 if ((signal->quick_threads == 0) && 872 !(signal->flags & SIGNAL_GROUP_EXIT)) { 873 signal->flags = SIGNAL_GROUP_EXIT; 874 signal->group_exit_code = code; 875 signal->group_stop_count = 0; 876 } 877 /* 878 * Serialize with any possible pending coredump. 879 * We must hold siglock around checking core_state 880 * and setting PF_POSTCOREDUMP. The core-inducing thread 881 * will increment ->nr_threads for each thread in the 882 * group without PF_POSTCOREDUMP set. 883 */ 884 tsk->flags |= PF_POSTCOREDUMP; 885 core_state = signal->core_state; 886 spin_unlock_irq(&sighand->siglock); 887 888 if (unlikely(core_state)) 889 coredump_task_exit(tsk, core_state); 890 } 891 892 void __noreturn do_exit(long code) 893 { 894 struct task_struct *tsk = current; 895 int group_dead; 896 897 WARN_ON(irqs_disabled()); 898 WARN_ON(tsk->plug); 899 900 kcov_task_exit(tsk); 901 kmsan_task_exit(tsk); 902 903 synchronize_group_exit(tsk, code); 904 ptrace_event(PTRACE_EVENT_EXIT, code); 905 user_events_exit(tsk); 906 907 io_uring_files_cancel(); 908 exit_signals(tsk); /* sets PF_EXITING */ 909 910 seccomp_filter_release(tsk); 911 912 acct_update_integrals(tsk); 913 group_dead = atomic_dec_and_test(&tsk->signal->live); 914 if (group_dead) { 915 /* 916 * If the last thread of global init has exited, panic 917 * immediately to get a useable coredump. 918 */ 919 if (unlikely(is_global_init(tsk))) 920 panic("Attempted to kill init! exitcode=0x%08x\n", 921 tsk->signal->group_exit_code ?: (int)code); 922 923 #ifdef CONFIG_POSIX_TIMERS 924 hrtimer_cancel(&tsk->signal->real_timer); 925 exit_itimers(tsk); 926 #endif 927 if (tsk->mm) 928 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm); 929 } 930 acct_collect(code, group_dead); 931 if (group_dead) 932 tty_audit_exit(); 933 audit_free(tsk); 934 935 tsk->exit_code = code; 936 taskstats_exit(tsk, group_dead); 937 unwind_deferred_task_exit(tsk); 938 trace_sched_process_exit(tsk, group_dead); 939 940 /* 941 * Since sampling can touch ->mm, make sure to stop everything before we 942 * tear it down. 943 * 944 * Also flushes inherited counters to the parent - before the parent 945 * gets woken up by child-exit notifications. 946 */ 947 perf_event_exit_task(tsk); 948 949 exit_mm(); 950 951 if (group_dead) 952 acct_process(); 953 954 exit_sem(tsk); 955 exit_shm(tsk); 956 exit_files(tsk); 957 exit_fs(tsk); 958 if (group_dead) 959 disassociate_ctty(1); 960 exit_task_namespaces(tsk); 961 exit_task_work(tsk); 962 exit_thread(tsk); 963 964 sched_autogroup_exit_task(tsk); 965 cgroup_exit(tsk); 966 967 /* 968 * FIXME: do that only when needed, using sched_exit tracepoint 969 */ 970 flush_ptrace_hw_breakpoint(tsk); 971 972 exit_tasks_rcu_start(); 973 exit_notify(tsk, group_dead); 974 proc_exit_connector(tsk); 975 mpol_put_task_policy(tsk); 976 #ifdef CONFIG_FUTEX 977 if (unlikely(current->pi_state_cache)) 978 kfree(current->pi_state_cache); 979 #endif 980 /* 981 * Make sure we are holding no locks: 982 */ 983 debug_check_no_locks_held(); 984 985 if (tsk->io_context) 986 exit_io_context(tsk); 987 988 if (tsk->splice_pipe) 989 free_pipe_info(tsk->splice_pipe); 990 991 if (tsk->task_frag.page) 992 put_page(tsk->task_frag.page); 993 994 exit_task_stack_account(tsk); 995 996 check_stack_usage(); 997 preempt_disable(); 998 if (tsk->nr_dirtied) 999 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied); 1000 exit_rcu(); 1001 exit_tasks_rcu_finish(); 1002 1003 lockdep_free_task(tsk); 1004 do_task_dead(); 1005 } 1006 1007 void __noreturn make_task_dead(int signr) 1008 { 1009 /* 1010 * Take the task off the cpu after something catastrophic has 1011 * happened. 1012 * 1013 * We can get here from a kernel oops, sometimes with preemption off. 1014 * Start by checking for critical errors. 1015 * Then fix up important state like USER_DS and preemption. 1016 * Then do everything else. 1017 */ 1018 struct task_struct *tsk = current; 1019 unsigned int limit; 1020 1021 if (unlikely(in_interrupt())) 1022 panic("Aiee, killing interrupt handler!"); 1023 if (unlikely(!tsk->pid)) 1024 panic("Attempted to kill the idle task!"); 1025 1026 if (unlikely(irqs_disabled())) { 1027 pr_info("note: %s[%d] exited with irqs disabled\n", 1028 current->comm, task_pid_nr(current)); 1029 local_irq_enable(); 1030 } 1031 if (unlikely(in_atomic())) { 1032 pr_info("note: %s[%d] exited with preempt_count %d\n", 1033 current->comm, task_pid_nr(current), 1034 preempt_count()); 1035 preempt_count_set(PREEMPT_ENABLED); 1036 } 1037 1038 /* 1039 * Every time the system oopses, if the oops happens while a reference 1040 * to an object was held, the reference leaks. 1041 * If the oops doesn't also leak memory, repeated oopsing can cause 1042 * reference counters to wrap around (if they're not using refcount_t). 1043 * This means that repeated oopsing can make unexploitable-looking bugs 1044 * exploitable through repeated oopsing. 1045 * To make sure this can't happen, place an upper bound on how often the 1046 * kernel may oops without panic(). 1047 */ 1048 limit = READ_ONCE(oops_limit); 1049 if (atomic_inc_return(&oops_count) >= limit && limit) 1050 panic("Oopsed too often (kernel.oops_limit is %d)", limit); 1051 1052 /* 1053 * We're taking recursive faults here in make_task_dead. Safest is to just 1054 * leave this task alone and wait for reboot. 1055 */ 1056 if (unlikely(tsk->flags & PF_EXITING)) { 1057 pr_alert("Fixing recursive fault but reboot is needed!\n"); 1058 futex_exit_recursive(tsk); 1059 tsk->exit_state = EXIT_DEAD; 1060 refcount_inc(&tsk->rcu_users); 1061 do_task_dead(); 1062 } 1063 1064 do_exit(signr); 1065 } 1066 1067 SYSCALL_DEFINE1(exit, int, error_code) 1068 { 1069 do_exit((error_code&0xff)<<8); 1070 } 1071 1072 /* 1073 * Take down every thread in the group. This is called by fatal signals 1074 * as well as by sys_exit_group (below). 1075 */ 1076 void __noreturn 1077 do_group_exit(int exit_code) 1078 { 1079 struct signal_struct *sig = current->signal; 1080 1081 if (sig->flags & SIGNAL_GROUP_EXIT) 1082 exit_code = sig->group_exit_code; 1083 else if (sig->group_exec_task) 1084 exit_code = 0; 1085 else { 1086 struct sighand_struct *const sighand = current->sighand; 1087 1088 spin_lock_irq(&sighand->siglock); 1089 if (sig->flags & SIGNAL_GROUP_EXIT) 1090 /* Another thread got here before we took the lock. */ 1091 exit_code = sig->group_exit_code; 1092 else if (sig->group_exec_task) 1093 exit_code = 0; 1094 else { 1095 sig->group_exit_code = exit_code; 1096 sig->flags = SIGNAL_GROUP_EXIT; 1097 zap_other_threads(current); 1098 } 1099 spin_unlock_irq(&sighand->siglock); 1100 } 1101 1102 do_exit(exit_code); 1103 /* NOTREACHED */ 1104 } 1105 1106 /* 1107 * this kills every thread in the thread group. Note that any externally 1108 * wait4()-ing process will get the correct exit code - even if this 1109 * thread is not the thread group leader. 1110 */ 1111 SYSCALL_DEFINE1(exit_group, int, error_code) 1112 { 1113 do_group_exit((error_code & 0xff) << 8); 1114 /* NOTREACHED */ 1115 return 0; 1116 } 1117 1118 static int eligible_pid(struct wait_opts *wo, struct task_struct *p) 1119 { 1120 return wo->wo_type == PIDTYPE_MAX || 1121 task_pid_type(p, wo->wo_type) == wo->wo_pid; 1122 } 1123 1124 static int 1125 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p) 1126 { 1127 if (!eligible_pid(wo, p)) 1128 return 0; 1129 1130 /* 1131 * Wait for all children (clone and not) if __WALL is set or 1132 * if it is traced by us. 1133 */ 1134 if (ptrace || (wo->wo_flags & __WALL)) 1135 return 1; 1136 1137 /* 1138 * Otherwise, wait for clone children *only* if __WCLONE is set; 1139 * otherwise, wait for non-clone children *only*. 1140 * 1141 * Note: a "clone" child here is one that reports to its parent 1142 * using a signal other than SIGCHLD, or a non-leader thread which 1143 * we can only see if it is traced by us. 1144 */ 1145 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE)) 1146 return 0; 1147 1148 return 1; 1149 } 1150 1151 /* 1152 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold 1153 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold 1154 * the lock and this task is uninteresting. If we return nonzero, we have 1155 * released the lock and the system call should return. 1156 */ 1157 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p) 1158 { 1159 int state, status; 1160 pid_t pid = task_pid_vnr(p); 1161 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p)); 1162 struct waitid_info *infop; 1163 1164 if (!likely(wo->wo_flags & WEXITED)) 1165 return 0; 1166 1167 if (unlikely(wo->wo_flags & WNOWAIT)) { 1168 status = (p->signal->flags & SIGNAL_GROUP_EXIT) 1169 ? p->signal->group_exit_code : p->exit_code; 1170 get_task_struct(p); 1171 read_unlock(&tasklist_lock); 1172 sched_annotate_sleep(); 1173 if (wo->wo_rusage) 1174 getrusage(p, RUSAGE_BOTH, wo->wo_rusage); 1175 put_task_struct(p); 1176 goto out_info; 1177 } 1178 /* 1179 * Move the task's state to DEAD/TRACE, only one thread can do this. 1180 */ 1181 state = (ptrace_reparented(p) && thread_group_leader(p)) ? 1182 EXIT_TRACE : EXIT_DEAD; 1183 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE) 1184 return 0; 1185 /* 1186 * We own this thread, nobody else can reap it. 1187 */ 1188 read_unlock(&tasklist_lock); 1189 sched_annotate_sleep(); 1190 1191 /* 1192 * Check thread_group_leader() to exclude the traced sub-threads. 1193 */ 1194 if (state == EXIT_DEAD && thread_group_leader(p)) { 1195 struct signal_struct *sig = p->signal; 1196 struct signal_struct *psig = current->signal; 1197 unsigned long maxrss; 1198 u64 tgutime, tgstime; 1199 1200 /* 1201 * The resource counters for the group leader are in its 1202 * own task_struct. Those for dead threads in the group 1203 * are in its signal_struct, as are those for the child 1204 * processes it has previously reaped. All these 1205 * accumulate in the parent's signal_struct c* fields. 1206 * 1207 * We don't bother to take a lock here to protect these 1208 * p->signal fields because the whole thread group is dead 1209 * and nobody can change them. 1210 * 1211 * psig->stats_lock also protects us from our sub-threads 1212 * which can reap other children at the same time. 1213 * 1214 * We use thread_group_cputime_adjusted() to get times for 1215 * the thread group, which consolidates times for all threads 1216 * in the group including the group leader. 1217 */ 1218 thread_group_cputime_adjusted(p, &tgutime, &tgstime); 1219 write_seqlock_irq(&psig->stats_lock); 1220 psig->cutime += tgutime + sig->cutime; 1221 psig->cstime += tgstime + sig->cstime; 1222 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime; 1223 psig->cmin_flt += 1224 p->min_flt + sig->min_flt + sig->cmin_flt; 1225 psig->cmaj_flt += 1226 p->maj_flt + sig->maj_flt + sig->cmaj_flt; 1227 psig->cnvcsw += 1228 p->nvcsw + sig->nvcsw + sig->cnvcsw; 1229 psig->cnivcsw += 1230 p->nivcsw + sig->nivcsw + sig->cnivcsw; 1231 psig->cinblock += 1232 task_io_get_inblock(p) + 1233 sig->inblock + sig->cinblock; 1234 psig->coublock += 1235 task_io_get_oublock(p) + 1236 sig->oublock + sig->coublock; 1237 maxrss = max(sig->maxrss, sig->cmaxrss); 1238 if (psig->cmaxrss < maxrss) 1239 psig->cmaxrss = maxrss; 1240 task_io_accounting_add(&psig->ioac, &p->ioac); 1241 task_io_accounting_add(&psig->ioac, &sig->ioac); 1242 write_sequnlock_irq(&psig->stats_lock); 1243 } 1244 1245 if (wo->wo_rusage) 1246 getrusage(p, RUSAGE_BOTH, wo->wo_rusage); 1247 status = (p->signal->flags & SIGNAL_GROUP_EXIT) 1248 ? p->signal->group_exit_code : p->exit_code; 1249 wo->wo_stat = status; 1250 1251 if (state == EXIT_TRACE) { 1252 write_lock_irq(&tasklist_lock); 1253 /* We dropped tasklist, ptracer could die and untrace */ 1254 ptrace_unlink(p); 1255 1256 /* If parent wants a zombie, don't release it now */ 1257 state = EXIT_ZOMBIE; 1258 if (do_notify_parent(p, p->exit_signal)) 1259 state = EXIT_DEAD; 1260 p->exit_state = state; 1261 write_unlock_irq(&tasklist_lock); 1262 } 1263 if (state == EXIT_DEAD) 1264 release_task(p); 1265 1266 out_info: 1267 infop = wo->wo_info; 1268 if (infop) { 1269 if ((status & 0x7f) == 0) { 1270 infop->cause = CLD_EXITED; 1271 infop->status = status >> 8; 1272 } else { 1273 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED; 1274 infop->status = status & 0x7f; 1275 } 1276 infop->pid = pid; 1277 infop->uid = uid; 1278 } 1279 1280 return pid; 1281 } 1282 1283 static int *task_stopped_code(struct task_struct *p, bool ptrace) 1284 { 1285 if (ptrace) { 1286 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING)) 1287 return &p->exit_code; 1288 } else { 1289 if (p->signal->flags & SIGNAL_STOP_STOPPED) 1290 return &p->signal->group_exit_code; 1291 } 1292 return NULL; 1293 } 1294 1295 /** 1296 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED 1297 * @wo: wait options 1298 * @ptrace: is the wait for ptrace 1299 * @p: task to wait for 1300 * 1301 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED. 1302 * 1303 * CONTEXT: 1304 * read_lock(&tasklist_lock), which is released if return value is 1305 * non-zero. Also, grabs and releases @p->sighand->siglock. 1306 * 1307 * RETURNS: 1308 * 0 if wait condition didn't exist and search for other wait conditions 1309 * should continue. Non-zero return, -errno on failure and @p's pid on 1310 * success, implies that tasklist_lock is released and wait condition 1311 * search should terminate. 1312 */ 1313 static int wait_task_stopped(struct wait_opts *wo, 1314 int ptrace, struct task_struct *p) 1315 { 1316 struct waitid_info *infop; 1317 int exit_code, *p_code, why; 1318 uid_t uid = 0; /* unneeded, required by compiler */ 1319 pid_t pid; 1320 1321 /* 1322 * Traditionally we see ptrace'd stopped tasks regardless of options. 1323 */ 1324 if (!ptrace && !(wo->wo_flags & WUNTRACED)) 1325 return 0; 1326 1327 if (!task_stopped_code(p, ptrace)) 1328 return 0; 1329 1330 exit_code = 0; 1331 spin_lock_irq(&p->sighand->siglock); 1332 1333 p_code = task_stopped_code(p, ptrace); 1334 if (unlikely(!p_code)) 1335 goto unlock_sig; 1336 1337 exit_code = *p_code; 1338 if (!exit_code) 1339 goto unlock_sig; 1340 1341 if (!unlikely(wo->wo_flags & WNOWAIT)) 1342 *p_code = 0; 1343 1344 uid = from_kuid_munged(current_user_ns(), task_uid(p)); 1345 unlock_sig: 1346 spin_unlock_irq(&p->sighand->siglock); 1347 if (!exit_code) 1348 return 0; 1349 1350 /* 1351 * Now we are pretty sure this task is interesting. 1352 * Make sure it doesn't get reaped out from under us while we 1353 * give up the lock and then examine it below. We don't want to 1354 * keep holding onto the tasklist_lock while we call getrusage and 1355 * possibly take page faults for user memory. 1356 */ 1357 get_task_struct(p); 1358 pid = task_pid_vnr(p); 1359 why = ptrace ? CLD_TRAPPED : CLD_STOPPED; 1360 read_unlock(&tasklist_lock); 1361 sched_annotate_sleep(); 1362 if (wo->wo_rusage) 1363 getrusage(p, RUSAGE_BOTH, wo->wo_rusage); 1364 put_task_struct(p); 1365 1366 if (likely(!(wo->wo_flags & WNOWAIT))) 1367 wo->wo_stat = (exit_code << 8) | 0x7f; 1368 1369 infop = wo->wo_info; 1370 if (infop) { 1371 infop->cause = why; 1372 infop->status = exit_code; 1373 infop->pid = pid; 1374 infop->uid = uid; 1375 } 1376 return pid; 1377 } 1378 1379 /* 1380 * Handle do_wait work for one task in a live, non-stopped state. 1381 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold 1382 * the lock and this task is uninteresting. If we return nonzero, we have 1383 * released the lock and the system call should return. 1384 */ 1385 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p) 1386 { 1387 struct waitid_info *infop; 1388 pid_t pid; 1389 uid_t uid; 1390 1391 if (!unlikely(wo->wo_flags & WCONTINUED)) 1392 return 0; 1393 1394 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) 1395 return 0; 1396 1397 spin_lock_irq(&p->sighand->siglock); 1398 /* Re-check with the lock held. */ 1399 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) { 1400 spin_unlock_irq(&p->sighand->siglock); 1401 return 0; 1402 } 1403 if (!unlikely(wo->wo_flags & WNOWAIT)) 1404 p->signal->flags &= ~SIGNAL_STOP_CONTINUED; 1405 uid = from_kuid_munged(current_user_ns(), task_uid(p)); 1406 spin_unlock_irq(&p->sighand->siglock); 1407 1408 pid = task_pid_vnr(p); 1409 get_task_struct(p); 1410 read_unlock(&tasklist_lock); 1411 sched_annotate_sleep(); 1412 if (wo->wo_rusage) 1413 getrusage(p, RUSAGE_BOTH, wo->wo_rusage); 1414 put_task_struct(p); 1415 1416 infop = wo->wo_info; 1417 if (!infop) { 1418 wo->wo_stat = 0xffff; 1419 } else { 1420 infop->cause = CLD_CONTINUED; 1421 infop->pid = pid; 1422 infop->uid = uid; 1423 infop->status = SIGCONT; 1424 } 1425 return pid; 1426 } 1427 1428 /* 1429 * Consider @p for a wait by @parent. 1430 * 1431 * -ECHILD should be in ->notask_error before the first call. 1432 * Returns nonzero for a final return, when we have unlocked tasklist_lock. 1433 * Returns zero if the search for a child should continue; 1434 * then ->notask_error is 0 if @p is an eligible child, 1435 * or still -ECHILD. 1436 */ 1437 static int wait_consider_task(struct wait_opts *wo, int ptrace, 1438 struct task_struct *p) 1439 { 1440 /* 1441 * We can race with wait_task_zombie() from another thread. 1442 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition 1443 * can't confuse the checks below. 1444 */ 1445 int exit_state = READ_ONCE(p->exit_state); 1446 int ret; 1447 1448 if (unlikely(exit_state == EXIT_DEAD)) 1449 return 0; 1450 1451 ret = eligible_child(wo, ptrace, p); 1452 if (!ret) 1453 return ret; 1454 1455 if (unlikely(exit_state == EXIT_TRACE)) { 1456 /* 1457 * ptrace == 0 means we are the natural parent. In this case 1458 * we should clear notask_error, debugger will notify us. 1459 */ 1460 if (likely(!ptrace)) 1461 wo->notask_error = 0; 1462 return 0; 1463 } 1464 1465 if (likely(!ptrace) && unlikely(p->ptrace)) { 1466 /* 1467 * If it is traced by its real parent's group, just pretend 1468 * the caller is ptrace_do_wait() and reap this child if it 1469 * is zombie. 1470 * 1471 * This also hides group stop state from real parent; otherwise 1472 * a single stop can be reported twice as group and ptrace stop. 1473 * If a ptracer wants to distinguish these two events for its 1474 * own children it should create a separate process which takes 1475 * the role of real parent. 1476 */ 1477 if (!ptrace_reparented(p)) 1478 ptrace = 1; 1479 } 1480 1481 /* slay zombie? */ 1482 if (exit_state == EXIT_ZOMBIE) { 1483 /* we don't reap group leaders with subthreads */ 1484 if (!delay_group_leader(p)) { 1485 /* 1486 * A zombie ptracee is only visible to its ptracer. 1487 * Notification and reaping will be cascaded to the 1488 * real parent when the ptracer detaches. 1489 */ 1490 if (unlikely(ptrace) || likely(!p->ptrace)) 1491 return wait_task_zombie(wo, p); 1492 } 1493 1494 /* 1495 * Allow access to stopped/continued state via zombie by 1496 * falling through. Clearing of notask_error is complex. 1497 * 1498 * When !@ptrace: 1499 * 1500 * If WEXITED is set, notask_error should naturally be 1501 * cleared. If not, subset of WSTOPPED|WCONTINUED is set, 1502 * so, if there are live subthreads, there are events to 1503 * wait for. If all subthreads are dead, it's still safe 1504 * to clear - this function will be called again in finite 1505 * amount time once all the subthreads are released and 1506 * will then return without clearing. 1507 * 1508 * When @ptrace: 1509 * 1510 * Stopped state is per-task and thus can't change once the 1511 * target task dies. Only continued and exited can happen. 1512 * Clear notask_error if WCONTINUED | WEXITED. 1513 */ 1514 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED))) 1515 wo->notask_error = 0; 1516 } else { 1517 /* 1518 * @p is alive and it's gonna stop, continue or exit, so 1519 * there always is something to wait for. 1520 */ 1521 wo->notask_error = 0; 1522 } 1523 1524 /* 1525 * Wait for stopped. Depending on @ptrace, different stopped state 1526 * is used and the two don't interact with each other. 1527 */ 1528 ret = wait_task_stopped(wo, ptrace, p); 1529 if (ret) 1530 return ret; 1531 1532 /* 1533 * Wait for continued. There's only one continued state and the 1534 * ptracer can consume it which can confuse the real parent. Don't 1535 * use WCONTINUED from ptracer. You don't need or want it. 1536 */ 1537 return wait_task_continued(wo, p); 1538 } 1539 1540 /* 1541 * Do the work of do_wait() for one thread in the group, @tsk. 1542 * 1543 * -ECHILD should be in ->notask_error before the first call. 1544 * Returns nonzero for a final return, when we have unlocked tasklist_lock. 1545 * Returns zero if the search for a child should continue; then 1546 * ->notask_error is 0 if there were any eligible children, 1547 * or still -ECHILD. 1548 */ 1549 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk) 1550 { 1551 struct task_struct *p; 1552 1553 list_for_each_entry(p, &tsk->children, sibling) { 1554 int ret = wait_consider_task(wo, 0, p); 1555 1556 if (ret) 1557 return ret; 1558 } 1559 1560 return 0; 1561 } 1562 1563 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk) 1564 { 1565 struct task_struct *p; 1566 1567 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) { 1568 int ret = wait_consider_task(wo, 1, p); 1569 1570 if (ret) 1571 return ret; 1572 } 1573 1574 return 0; 1575 } 1576 1577 bool pid_child_should_wake(struct wait_opts *wo, struct task_struct *p) 1578 { 1579 if (!eligible_pid(wo, p)) 1580 return false; 1581 1582 if ((wo->wo_flags & __WNOTHREAD) && wo->child_wait.private != p->parent) 1583 return false; 1584 1585 return true; 1586 } 1587 1588 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode, 1589 int sync, void *key) 1590 { 1591 struct wait_opts *wo = container_of(wait, struct wait_opts, 1592 child_wait); 1593 struct task_struct *p = key; 1594 1595 if (pid_child_should_wake(wo, p)) 1596 return default_wake_function(wait, mode, sync, key); 1597 1598 return 0; 1599 } 1600 1601 void __wake_up_parent(struct task_struct *p, struct task_struct *parent) 1602 { 1603 __wake_up_sync_key(&parent->signal->wait_chldexit, 1604 TASK_INTERRUPTIBLE, p); 1605 } 1606 1607 static bool is_effectively_child(struct wait_opts *wo, bool ptrace, 1608 struct task_struct *target) 1609 { 1610 struct task_struct *parent = 1611 !ptrace ? target->real_parent : target->parent; 1612 1613 return current == parent || (!(wo->wo_flags & __WNOTHREAD) && 1614 same_thread_group(current, parent)); 1615 } 1616 1617 /* 1618 * Optimization for waiting on PIDTYPE_PID. No need to iterate through child 1619 * and tracee lists to find the target task. 1620 */ 1621 static int do_wait_pid(struct wait_opts *wo) 1622 { 1623 bool ptrace; 1624 struct task_struct *target; 1625 int retval; 1626 1627 ptrace = false; 1628 target = pid_task(wo->wo_pid, PIDTYPE_TGID); 1629 if (target && is_effectively_child(wo, ptrace, target)) { 1630 retval = wait_consider_task(wo, ptrace, target); 1631 if (retval) 1632 return retval; 1633 } 1634 1635 ptrace = true; 1636 target = pid_task(wo->wo_pid, PIDTYPE_PID); 1637 if (target && target->ptrace && 1638 is_effectively_child(wo, ptrace, target)) { 1639 retval = wait_consider_task(wo, ptrace, target); 1640 if (retval) 1641 return retval; 1642 } 1643 1644 return 0; 1645 } 1646 1647 long __do_wait(struct wait_opts *wo) 1648 { 1649 long retval; 1650 1651 /* 1652 * If there is nothing that can match our criteria, just get out. 1653 * We will clear ->notask_error to zero if we see any child that 1654 * might later match our criteria, even if we are not able to reap 1655 * it yet. 1656 */ 1657 wo->notask_error = -ECHILD; 1658 if ((wo->wo_type < PIDTYPE_MAX) && 1659 (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type))) 1660 goto notask; 1661 1662 read_lock(&tasklist_lock); 1663 1664 if (wo->wo_type == PIDTYPE_PID) { 1665 retval = do_wait_pid(wo); 1666 if (retval) 1667 return retval; 1668 } else { 1669 struct task_struct *tsk = current; 1670 1671 do { 1672 retval = do_wait_thread(wo, tsk); 1673 if (retval) 1674 return retval; 1675 1676 retval = ptrace_do_wait(wo, tsk); 1677 if (retval) 1678 return retval; 1679 1680 if (wo->wo_flags & __WNOTHREAD) 1681 break; 1682 } while_each_thread(current, tsk); 1683 } 1684 read_unlock(&tasklist_lock); 1685 1686 notask: 1687 retval = wo->notask_error; 1688 if (!retval && !(wo->wo_flags & WNOHANG)) 1689 return -ERESTARTSYS; 1690 1691 return retval; 1692 } 1693 1694 static long do_wait(struct wait_opts *wo) 1695 { 1696 int retval; 1697 1698 trace_sched_process_wait(wo->wo_pid); 1699 1700 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback); 1701 wo->child_wait.private = current; 1702 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait); 1703 1704 do { 1705 set_current_state(TASK_INTERRUPTIBLE); 1706 retval = __do_wait(wo); 1707 if (retval != -ERESTARTSYS) 1708 break; 1709 if (signal_pending(current)) 1710 break; 1711 schedule(); 1712 } while (1); 1713 1714 __set_current_state(TASK_RUNNING); 1715 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait); 1716 return retval; 1717 } 1718 1719 int kernel_waitid_prepare(struct wait_opts *wo, int which, pid_t upid, 1720 struct waitid_info *infop, int options, 1721 struct rusage *ru) 1722 { 1723 unsigned int f_flags = 0; 1724 struct pid *pid = NULL; 1725 enum pid_type type; 1726 1727 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED| 1728 __WNOTHREAD|__WCLONE|__WALL)) 1729 return -EINVAL; 1730 if (!(options & (WEXITED|WSTOPPED|WCONTINUED))) 1731 return -EINVAL; 1732 1733 switch (which) { 1734 case P_ALL: 1735 type = PIDTYPE_MAX; 1736 break; 1737 case P_PID: 1738 type = PIDTYPE_PID; 1739 if (upid <= 0) 1740 return -EINVAL; 1741 1742 pid = find_get_pid(upid); 1743 break; 1744 case P_PGID: 1745 type = PIDTYPE_PGID; 1746 if (upid < 0) 1747 return -EINVAL; 1748 1749 if (upid) 1750 pid = find_get_pid(upid); 1751 else 1752 pid = get_task_pid(current, PIDTYPE_PGID); 1753 break; 1754 case P_PIDFD: 1755 type = PIDTYPE_PID; 1756 if (upid < 0) 1757 return -EINVAL; 1758 1759 pid = pidfd_get_pid(upid, &f_flags); 1760 if (IS_ERR(pid)) 1761 return PTR_ERR(pid); 1762 1763 break; 1764 default: 1765 return -EINVAL; 1766 } 1767 1768 wo->wo_type = type; 1769 wo->wo_pid = pid; 1770 wo->wo_flags = options; 1771 wo->wo_info = infop; 1772 wo->wo_rusage = ru; 1773 if (f_flags & O_NONBLOCK) 1774 wo->wo_flags |= WNOHANG; 1775 1776 return 0; 1777 } 1778 1779 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop, 1780 int options, struct rusage *ru) 1781 { 1782 struct wait_opts wo; 1783 long ret; 1784 1785 ret = kernel_waitid_prepare(&wo, which, upid, infop, options, ru); 1786 if (ret) 1787 return ret; 1788 1789 ret = do_wait(&wo); 1790 if (!ret && !(options & WNOHANG) && (wo.wo_flags & WNOHANG)) 1791 ret = -EAGAIN; 1792 1793 put_pid(wo.wo_pid); 1794 return ret; 1795 } 1796 1797 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *, 1798 infop, int, options, struct rusage __user *, ru) 1799 { 1800 struct rusage r; 1801 struct waitid_info info = {.status = 0}; 1802 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL); 1803 int signo = 0; 1804 1805 if (err > 0) { 1806 signo = SIGCHLD; 1807 err = 0; 1808 if (ru && copy_to_user(ru, &r, sizeof(struct rusage))) 1809 return -EFAULT; 1810 } 1811 if (!infop) 1812 return err; 1813 1814 if (!user_write_access_begin(infop, sizeof(*infop))) 1815 return -EFAULT; 1816 1817 unsafe_put_user(signo, &infop->si_signo, Efault); 1818 unsafe_put_user(0, &infop->si_errno, Efault); 1819 unsafe_put_user(info.cause, &infop->si_code, Efault); 1820 unsafe_put_user(info.pid, &infop->si_pid, Efault); 1821 unsafe_put_user(info.uid, &infop->si_uid, Efault); 1822 unsafe_put_user(info.status, &infop->si_status, Efault); 1823 user_write_access_end(); 1824 return err; 1825 Efault: 1826 user_write_access_end(); 1827 return -EFAULT; 1828 } 1829 1830 long kernel_wait4(pid_t upid, int __user *stat_addr, int options, 1831 struct rusage *ru) 1832 { 1833 struct wait_opts wo; 1834 struct pid *pid = NULL; 1835 enum pid_type type; 1836 long ret; 1837 1838 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED| 1839 __WNOTHREAD|__WCLONE|__WALL)) 1840 return -EINVAL; 1841 1842 /* -INT_MIN is not defined */ 1843 if (upid == INT_MIN) 1844 return -ESRCH; 1845 1846 if (upid == -1) 1847 type = PIDTYPE_MAX; 1848 else if (upid < 0) { 1849 type = PIDTYPE_PGID; 1850 pid = find_get_pid(-upid); 1851 } else if (upid == 0) { 1852 type = PIDTYPE_PGID; 1853 pid = get_task_pid(current, PIDTYPE_PGID); 1854 } else /* upid > 0 */ { 1855 type = PIDTYPE_PID; 1856 pid = find_get_pid(upid); 1857 } 1858 1859 wo.wo_type = type; 1860 wo.wo_pid = pid; 1861 wo.wo_flags = options | WEXITED; 1862 wo.wo_info = NULL; 1863 wo.wo_stat = 0; 1864 wo.wo_rusage = ru; 1865 ret = do_wait(&wo); 1866 put_pid(pid); 1867 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr)) 1868 ret = -EFAULT; 1869 1870 return ret; 1871 } 1872 1873 int kernel_wait(pid_t pid, int *stat) 1874 { 1875 struct wait_opts wo = { 1876 .wo_type = PIDTYPE_PID, 1877 .wo_pid = find_get_pid(pid), 1878 .wo_flags = WEXITED, 1879 }; 1880 int ret; 1881 1882 ret = do_wait(&wo); 1883 if (ret > 0 && wo.wo_stat) 1884 *stat = wo.wo_stat; 1885 put_pid(wo.wo_pid); 1886 return ret; 1887 } 1888 1889 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr, 1890 int, options, struct rusage __user *, ru) 1891 { 1892 struct rusage r; 1893 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL); 1894 1895 if (err > 0) { 1896 if (ru && copy_to_user(ru, &r, sizeof(struct rusage))) 1897 return -EFAULT; 1898 } 1899 return err; 1900 } 1901 1902 #ifdef __ARCH_WANT_SYS_WAITPID 1903 1904 /* 1905 * sys_waitpid() remains for compatibility. waitpid() should be 1906 * implemented by calling sys_wait4() from libc.a. 1907 */ 1908 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options) 1909 { 1910 return kernel_wait4(pid, stat_addr, options, NULL); 1911 } 1912 1913 #endif 1914 1915 #ifdef CONFIG_COMPAT 1916 COMPAT_SYSCALL_DEFINE4(wait4, 1917 compat_pid_t, pid, 1918 compat_uint_t __user *, stat_addr, 1919 int, options, 1920 struct compat_rusage __user *, ru) 1921 { 1922 struct rusage r; 1923 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL); 1924 if (err > 0) { 1925 if (ru && put_compat_rusage(&r, ru)) 1926 return -EFAULT; 1927 } 1928 return err; 1929 } 1930 1931 COMPAT_SYSCALL_DEFINE5(waitid, 1932 int, which, compat_pid_t, pid, 1933 struct compat_siginfo __user *, infop, int, options, 1934 struct compat_rusage __user *, uru) 1935 { 1936 struct rusage ru; 1937 struct waitid_info info = {.status = 0}; 1938 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL); 1939 int signo = 0; 1940 if (err > 0) { 1941 signo = SIGCHLD; 1942 err = 0; 1943 if (uru) { 1944 /* kernel_waitid() overwrites everything in ru */ 1945 if (COMPAT_USE_64BIT_TIME) 1946 err = copy_to_user(uru, &ru, sizeof(ru)); 1947 else 1948 err = put_compat_rusage(&ru, uru); 1949 if (err) 1950 return -EFAULT; 1951 } 1952 } 1953 1954 if (!infop) 1955 return err; 1956 1957 if (!user_write_access_begin(infop, sizeof(*infop))) 1958 return -EFAULT; 1959 1960 unsafe_put_user(signo, &infop->si_signo, Efault); 1961 unsafe_put_user(0, &infop->si_errno, Efault); 1962 unsafe_put_user(info.cause, &infop->si_code, Efault); 1963 unsafe_put_user(info.pid, &infop->si_pid, Efault); 1964 unsafe_put_user(info.uid, &infop->si_uid, Efault); 1965 unsafe_put_user(info.status, &infop->si_status, Efault); 1966 user_write_access_end(); 1967 return err; 1968 Efault: 1969 user_write_access_end(); 1970 return -EFAULT; 1971 } 1972 #endif 1973 1974 /* 1975 * This needs to be __function_aligned as GCC implicitly makes any 1976 * implementation of abort() cold and drops alignment specified by 1977 * -falign-functions=N. 1978 * 1979 * See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88345#c11 1980 */ 1981 __weak __function_aligned void abort(void) 1982 { 1983 BUG(); 1984 1985 /* if that doesn't kill us, halt */ 1986 panic("Oops failed to kill thread"); 1987 } 1988 EXPORT_SYMBOL(abort); 1989