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 * This does two things: 697 * 698 * A. Make init inherit all the child processes 699 * B. Check to see if any process groups have become orphaned 700 * as a result of our exiting, and if they have any stopped 701 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2) 702 */ 703 static void forget_original_parent(struct task_struct *father, 704 struct list_head *dead) 705 { 706 struct task_struct *p, *t, *reaper; 707 708 if (unlikely(!list_empty(&father->ptraced))) 709 exit_ptrace(father, dead); 710 711 /* Can drop and reacquire tasklist_lock */ 712 reaper = find_child_reaper(father, dead); 713 if (list_empty(&father->children)) 714 return; 715 716 reaper = find_new_reaper(father, reaper); 717 list_for_each_entry(p, &father->children, sibling) { 718 for_each_thread(p, t) { 719 RCU_INIT_POINTER(t->real_parent, reaper); 720 BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father)); 721 if (likely(!t->ptrace)) 722 t->parent = t->real_parent; 723 if (t->pdeath_signal) 724 group_send_sig_info(t->pdeath_signal, 725 SEND_SIG_NOINFO, t, 726 PIDTYPE_TGID); 727 } 728 /* 729 * If this is a threaded reparent there is no need to 730 * notify anyone anything has happened. 731 */ 732 if (!same_thread_group(reaper, father)) 733 reparent_leader(father, p, dead); 734 } 735 list_splice_tail_init(&father->children, &reaper->children); 736 } 737 738 /* 739 * Send signals to all our closest relatives so that they know 740 * to properly mourn us.. 741 */ 742 static void exit_notify(struct task_struct *tsk, int group_dead) 743 { 744 bool autoreap; 745 struct task_struct *p, *n; 746 LIST_HEAD(dead); 747 748 write_lock_irq(&tasklist_lock); 749 forget_original_parent(tsk, &dead); 750 751 if (group_dead) 752 kill_orphaned_pgrp(tsk->group_leader, NULL); 753 754 tsk->exit_state = EXIT_ZOMBIE; 755 756 if (unlikely(tsk->ptrace)) { 757 int sig = thread_group_leader(tsk) && 758 thread_group_empty(tsk) && 759 !ptrace_reparented(tsk) ? 760 tsk->exit_signal : SIGCHLD; 761 autoreap = do_notify_parent(tsk, sig); 762 } else if (thread_group_leader(tsk)) { 763 autoreap = thread_group_empty(tsk) && 764 do_notify_parent(tsk, tsk->exit_signal); 765 } else { 766 autoreap = true; 767 /* untraced sub-thread */ 768 do_notify_pidfd(tsk); 769 } 770 771 if (autoreap) { 772 tsk->exit_state = EXIT_DEAD; 773 list_add(&tsk->ptrace_entry, &dead); 774 } 775 776 /* mt-exec, de_thread() is waiting for group leader */ 777 if (unlikely(tsk->signal->notify_count < 0)) 778 wake_up_process(tsk->signal->group_exec_task); 779 write_unlock_irq(&tasklist_lock); 780 781 list_for_each_entry_safe(p, n, &dead, ptrace_entry) { 782 list_del_init(&p->ptrace_entry); 783 release_task(p); 784 } 785 } 786 787 #ifdef CONFIG_DEBUG_STACK_USAGE 788 unsigned long stack_not_used(struct task_struct *p) 789 { 790 unsigned long *n = end_of_stack(p); 791 792 do { /* Skip over canary */ 793 # ifdef CONFIG_STACK_GROWSUP 794 n--; 795 # else 796 n++; 797 # endif 798 } while (!*n); 799 800 # ifdef CONFIG_STACK_GROWSUP 801 return (unsigned long)end_of_stack(p) - (unsigned long)n; 802 # else 803 return (unsigned long)n - (unsigned long)end_of_stack(p); 804 # endif 805 } 806 807 /* Count the maximum pages reached in kernel stacks */ 808 static inline void kstack_histogram(unsigned long used_stack) 809 { 810 #ifdef CONFIG_VM_EVENT_COUNTERS 811 if (used_stack <= 1024) 812 count_vm_event(KSTACK_1K); 813 #if THREAD_SIZE > 1024 814 else if (used_stack <= 2048) 815 count_vm_event(KSTACK_2K); 816 #endif 817 #if THREAD_SIZE > 2048 818 else if (used_stack <= 4096) 819 count_vm_event(KSTACK_4K); 820 #endif 821 #if THREAD_SIZE > 4096 822 else if (used_stack <= 8192) 823 count_vm_event(KSTACK_8K); 824 #endif 825 #if THREAD_SIZE > 8192 826 else if (used_stack <= 16384) 827 count_vm_event(KSTACK_16K); 828 #endif 829 #if THREAD_SIZE > 16384 830 else if (used_stack <= 32768) 831 count_vm_event(KSTACK_32K); 832 #endif 833 #if THREAD_SIZE > 32768 834 else if (used_stack <= 65536) 835 count_vm_event(KSTACK_64K); 836 #endif 837 #if THREAD_SIZE > 65536 838 else 839 count_vm_event(KSTACK_REST); 840 #endif 841 #endif /* CONFIG_VM_EVENT_COUNTERS */ 842 } 843 844 static void check_stack_usage(void) 845 { 846 static DEFINE_SPINLOCK(low_water_lock); 847 static int lowest_to_date = THREAD_SIZE; 848 unsigned long free; 849 850 free = stack_not_used(current); 851 kstack_histogram(THREAD_SIZE - free); 852 853 if (free >= lowest_to_date) 854 return; 855 856 spin_lock(&low_water_lock); 857 if (free < lowest_to_date) { 858 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n", 859 current->comm, task_pid_nr(current), free); 860 lowest_to_date = free; 861 } 862 spin_unlock(&low_water_lock); 863 } 864 #else 865 static inline void check_stack_usage(void) {} 866 #endif 867 868 static void synchronize_group_exit(struct task_struct *tsk, long code) 869 { 870 struct sighand_struct *sighand = tsk->sighand; 871 struct signal_struct *signal = tsk->signal; 872 struct core_state *core_state; 873 874 spin_lock_irq(&sighand->siglock); 875 signal->quick_threads--; 876 if ((signal->quick_threads == 0) && 877 !(signal->flags & SIGNAL_GROUP_EXIT)) { 878 signal->flags = SIGNAL_GROUP_EXIT; 879 signal->group_exit_code = code; 880 signal->group_stop_count = 0; 881 } 882 /* 883 * Serialize with any possible pending coredump. 884 * We must hold siglock around checking core_state 885 * and setting PF_POSTCOREDUMP. The core-inducing thread 886 * will increment ->nr_threads for each thread in the 887 * group without PF_POSTCOREDUMP set. 888 */ 889 tsk->flags |= PF_POSTCOREDUMP; 890 core_state = signal->core_state; 891 spin_unlock_irq(&sighand->siglock); 892 893 if (unlikely(core_state)) 894 coredump_task_exit(tsk, core_state); 895 } 896 897 void __noreturn do_exit(long code) 898 { 899 struct task_struct *tsk = current; 900 int group_dead; 901 902 WARN_ON(irqs_disabled()); 903 WARN_ON(tsk->plug); 904 905 kcov_task_exit(tsk); 906 kmsan_task_exit(tsk); 907 908 synchronize_group_exit(tsk, code); 909 ptrace_event(PTRACE_EVENT_EXIT, code); 910 user_events_exit(tsk); 911 912 io_uring_files_cancel(); 913 exit_signals(tsk); /* sets PF_EXITING */ 914 915 seccomp_filter_release(tsk); 916 917 acct_update_integrals(tsk); 918 group_dead = atomic_dec_and_test(&tsk->signal->live); 919 if (group_dead) { 920 /* 921 * If the last thread of global init has exited, panic 922 * immediately to get a useable coredump. 923 */ 924 if (unlikely(is_global_init(tsk))) 925 panic("Attempted to kill init! exitcode=0x%08x\n", 926 tsk->signal->group_exit_code ?: (int)code); 927 928 #ifdef CONFIG_POSIX_TIMERS 929 hrtimer_cancel(&tsk->signal->real_timer); 930 exit_itimers(tsk); 931 #endif 932 if (tsk->mm) 933 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm); 934 } 935 acct_collect(code, group_dead); 936 if (group_dead) 937 tty_audit_exit(); 938 audit_free(tsk); 939 940 tsk->exit_code = code; 941 taskstats_exit(tsk, group_dead); 942 unwind_deferred_task_exit(tsk); 943 trace_sched_process_exit(tsk, group_dead); 944 945 /* 946 * Since sampling can touch ->mm, make sure to stop everything before we 947 * tear it down. 948 * 949 * Also flushes inherited counters to the parent - before the parent 950 * gets woken up by child-exit notifications. 951 */ 952 perf_event_exit_task(tsk); 953 954 exit_mm(); 955 956 if (group_dead) 957 acct_process(); 958 959 exit_sem(tsk); 960 exit_shm(tsk); 961 exit_files(tsk); 962 exit_fs(tsk); 963 if (group_dead) 964 disassociate_ctty(1); 965 exit_task_namespaces(tsk); 966 exit_task_work(tsk); 967 exit_thread(tsk); 968 969 sched_autogroup_exit_task(tsk); 970 cgroup_exit(tsk); 971 972 /* 973 * FIXME: do that only when needed, using sched_exit tracepoint 974 */ 975 flush_ptrace_hw_breakpoint(tsk); 976 977 exit_tasks_rcu_start(); 978 exit_notify(tsk, group_dead); 979 proc_exit_connector(tsk); 980 mpol_put_task_policy(tsk); 981 #ifdef CONFIG_FUTEX 982 if (unlikely(current->pi_state_cache)) 983 kfree(current->pi_state_cache); 984 #endif 985 /* 986 * Make sure we are holding no locks: 987 */ 988 debug_check_no_locks_held(); 989 990 if (tsk->io_context) 991 exit_io_context(tsk); 992 993 if (tsk->splice_pipe) 994 free_pipe_info(tsk->splice_pipe); 995 996 if (tsk->task_frag.page) 997 put_page(tsk->task_frag.page); 998 999 exit_task_stack_account(tsk); 1000 1001 check_stack_usage(); 1002 preempt_disable(); 1003 if (tsk->nr_dirtied) 1004 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied); 1005 exit_rcu(); 1006 exit_tasks_rcu_finish(); 1007 1008 lockdep_free_task(tsk); 1009 do_task_dead(); 1010 } 1011 1012 void __noreturn make_task_dead(int signr) 1013 { 1014 /* 1015 * Take the task off the cpu after something catastrophic has 1016 * happened. 1017 * 1018 * We can get here from a kernel oops, sometimes with preemption off. 1019 * Start by checking for critical errors. 1020 * Then fix up important state like USER_DS and preemption. 1021 * Then do everything else. 1022 */ 1023 struct task_struct *tsk = current; 1024 unsigned int limit; 1025 1026 if (unlikely(in_interrupt())) 1027 panic("Aiee, killing interrupt handler!"); 1028 if (unlikely(!tsk->pid)) 1029 panic("Attempted to kill the idle task!"); 1030 1031 if (unlikely(irqs_disabled())) { 1032 pr_info("note: %s[%d] exited with irqs disabled\n", 1033 current->comm, task_pid_nr(current)); 1034 local_irq_enable(); 1035 } 1036 if (unlikely(in_atomic())) { 1037 pr_info("note: %s[%d] exited with preempt_count %d\n", 1038 current->comm, task_pid_nr(current), 1039 preempt_count()); 1040 preempt_count_set(PREEMPT_ENABLED); 1041 } 1042 1043 /* 1044 * Every time the system oopses, if the oops happens while a reference 1045 * to an object was held, the reference leaks. 1046 * If the oops doesn't also leak memory, repeated oopsing can cause 1047 * reference counters to wrap around (if they're not using refcount_t). 1048 * This means that repeated oopsing can make unexploitable-looking bugs 1049 * exploitable through repeated oopsing. 1050 * To make sure this can't happen, place an upper bound on how often the 1051 * kernel may oops without panic(). 1052 */ 1053 limit = READ_ONCE(oops_limit); 1054 if (atomic_inc_return(&oops_count) >= limit && limit) 1055 panic("Oopsed too often (kernel.oops_limit is %d)", limit); 1056 1057 /* 1058 * We're taking recursive faults here in make_task_dead. Safest is to just 1059 * leave this task alone and wait for reboot. 1060 */ 1061 if (unlikely(tsk->flags & PF_EXITING)) { 1062 pr_alert("Fixing recursive fault but reboot is needed!\n"); 1063 futex_exit_recursive(tsk); 1064 tsk->exit_state = EXIT_DEAD; 1065 refcount_inc(&tsk->rcu_users); 1066 do_task_dead(); 1067 } 1068 1069 do_exit(signr); 1070 } 1071 1072 SYSCALL_DEFINE1(exit, int, error_code) 1073 { 1074 do_exit((error_code&0xff)<<8); 1075 } 1076 1077 /* 1078 * Take down every thread in the group. This is called by fatal signals 1079 * as well as by sys_exit_group (below). 1080 */ 1081 void __noreturn 1082 do_group_exit(int exit_code) 1083 { 1084 struct signal_struct *sig = current->signal; 1085 1086 if (sig->flags & SIGNAL_GROUP_EXIT) 1087 exit_code = sig->group_exit_code; 1088 else if (sig->group_exec_task) 1089 exit_code = 0; 1090 else { 1091 struct sighand_struct *const sighand = current->sighand; 1092 1093 spin_lock_irq(&sighand->siglock); 1094 if (sig->flags & SIGNAL_GROUP_EXIT) 1095 /* Another thread got here before we took the lock. */ 1096 exit_code = sig->group_exit_code; 1097 else if (sig->group_exec_task) 1098 exit_code = 0; 1099 else { 1100 sig->group_exit_code = exit_code; 1101 sig->flags = SIGNAL_GROUP_EXIT; 1102 zap_other_threads(current); 1103 } 1104 spin_unlock_irq(&sighand->siglock); 1105 } 1106 1107 do_exit(exit_code); 1108 /* NOTREACHED */ 1109 } 1110 1111 /* 1112 * this kills every thread in the thread group. Note that any externally 1113 * wait4()-ing process will get the correct exit code - even if this 1114 * thread is not the thread group leader. 1115 */ 1116 SYSCALL_DEFINE1(exit_group, int, error_code) 1117 { 1118 do_group_exit((error_code & 0xff) << 8); 1119 /* NOTREACHED */ 1120 return 0; 1121 } 1122 1123 static int eligible_pid(struct wait_opts *wo, struct task_struct *p) 1124 { 1125 return wo->wo_type == PIDTYPE_MAX || 1126 task_pid_type(p, wo->wo_type) == wo->wo_pid; 1127 } 1128 1129 static int 1130 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p) 1131 { 1132 if (!eligible_pid(wo, p)) 1133 return 0; 1134 1135 /* 1136 * Wait for all children (clone and not) if __WALL is set or 1137 * if it is traced by us. 1138 */ 1139 if (ptrace || (wo->wo_flags & __WALL)) 1140 return 1; 1141 1142 /* 1143 * Otherwise, wait for clone children *only* if __WCLONE is set; 1144 * otherwise, wait for non-clone children *only*. 1145 * 1146 * Note: a "clone" child here is one that reports to its parent 1147 * using a signal other than SIGCHLD, or a non-leader thread which 1148 * we can only see if it is traced by us. 1149 */ 1150 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE)) 1151 return 0; 1152 1153 return 1; 1154 } 1155 1156 /* 1157 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold 1158 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold 1159 * the lock and this task is uninteresting. If we return nonzero, we have 1160 * released the lock and the system call should return. 1161 */ 1162 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p) 1163 { 1164 int state, status; 1165 pid_t pid = task_pid_vnr(p); 1166 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p)); 1167 struct waitid_info *infop; 1168 1169 if (!likely(wo->wo_flags & WEXITED)) 1170 return 0; 1171 1172 if (unlikely(wo->wo_flags & WNOWAIT)) { 1173 status = (p->signal->flags & SIGNAL_GROUP_EXIT) 1174 ? p->signal->group_exit_code : p->exit_code; 1175 get_task_struct(p); 1176 read_unlock(&tasklist_lock); 1177 sched_annotate_sleep(); 1178 if (wo->wo_rusage) 1179 getrusage(p, RUSAGE_BOTH, wo->wo_rusage); 1180 put_task_struct(p); 1181 goto out_info; 1182 } 1183 /* 1184 * Move the task's state to DEAD/TRACE, only one thread can do this. 1185 */ 1186 state = (ptrace_reparented(p) && thread_group_leader(p)) ? 1187 EXIT_TRACE : EXIT_DEAD; 1188 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE) 1189 return 0; 1190 /* 1191 * We own this thread, nobody else can reap it. 1192 */ 1193 read_unlock(&tasklist_lock); 1194 sched_annotate_sleep(); 1195 1196 /* 1197 * Check thread_group_leader() to exclude the traced sub-threads. 1198 */ 1199 if (state == EXIT_DEAD && thread_group_leader(p)) { 1200 struct signal_struct *sig = p->signal; 1201 struct signal_struct *psig = current->signal; 1202 unsigned long maxrss; 1203 u64 tgutime, tgstime; 1204 1205 /* 1206 * The resource counters for the group leader are in its 1207 * own task_struct. Those for dead threads in the group 1208 * are in its signal_struct, as are those for the child 1209 * processes it has previously reaped. All these 1210 * accumulate in the parent's signal_struct c* fields. 1211 * 1212 * We don't bother to take a lock here to protect these 1213 * p->signal fields because the whole thread group is dead 1214 * and nobody can change them. 1215 * 1216 * psig->stats_lock also protects us from our sub-threads 1217 * which can reap other children at the same time. 1218 * 1219 * We use thread_group_cputime_adjusted() to get times for 1220 * the thread group, which consolidates times for all threads 1221 * in the group including the group leader. 1222 */ 1223 thread_group_cputime_adjusted(p, &tgutime, &tgstime); 1224 write_seqlock_irq(&psig->stats_lock); 1225 psig->cutime += tgutime + sig->cutime; 1226 psig->cstime += tgstime + sig->cstime; 1227 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime; 1228 psig->cmin_flt += 1229 p->min_flt + sig->min_flt + sig->cmin_flt; 1230 psig->cmaj_flt += 1231 p->maj_flt + sig->maj_flt + sig->cmaj_flt; 1232 psig->cnvcsw += 1233 p->nvcsw + sig->nvcsw + sig->cnvcsw; 1234 psig->cnivcsw += 1235 p->nivcsw + sig->nivcsw + sig->cnivcsw; 1236 psig->cinblock += 1237 task_io_get_inblock(p) + 1238 sig->inblock + sig->cinblock; 1239 psig->coublock += 1240 task_io_get_oublock(p) + 1241 sig->oublock + sig->coublock; 1242 maxrss = max(sig->maxrss, sig->cmaxrss); 1243 if (psig->cmaxrss < maxrss) 1244 psig->cmaxrss = maxrss; 1245 task_io_accounting_add(&psig->ioac, &p->ioac); 1246 task_io_accounting_add(&psig->ioac, &sig->ioac); 1247 write_sequnlock_irq(&psig->stats_lock); 1248 } 1249 1250 if (wo->wo_rusage) 1251 getrusage(p, RUSAGE_BOTH, wo->wo_rusage); 1252 status = (p->signal->flags & SIGNAL_GROUP_EXIT) 1253 ? p->signal->group_exit_code : p->exit_code; 1254 wo->wo_stat = status; 1255 1256 if (state == EXIT_TRACE) { 1257 write_lock_irq(&tasklist_lock); 1258 /* We dropped tasklist, ptracer could die and untrace */ 1259 ptrace_unlink(p); 1260 1261 /* If parent wants a zombie, don't release it now */ 1262 state = EXIT_ZOMBIE; 1263 if (do_notify_parent(p, p->exit_signal)) 1264 state = EXIT_DEAD; 1265 p->exit_state = state; 1266 write_unlock_irq(&tasklist_lock); 1267 } 1268 if (state == EXIT_DEAD) 1269 release_task(p); 1270 1271 out_info: 1272 infop = wo->wo_info; 1273 if (infop) { 1274 if ((status & 0x7f) == 0) { 1275 infop->cause = CLD_EXITED; 1276 infop->status = status >> 8; 1277 } else { 1278 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED; 1279 infop->status = status & 0x7f; 1280 } 1281 infop->pid = pid; 1282 infop->uid = uid; 1283 } 1284 1285 return pid; 1286 } 1287 1288 static int *task_stopped_code(struct task_struct *p, bool ptrace) 1289 { 1290 if (ptrace) { 1291 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING)) 1292 return &p->exit_code; 1293 } else { 1294 if (p->signal->flags & SIGNAL_STOP_STOPPED) 1295 return &p->signal->group_exit_code; 1296 } 1297 return NULL; 1298 } 1299 1300 /** 1301 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED 1302 * @wo: wait options 1303 * @ptrace: is the wait for ptrace 1304 * @p: task to wait for 1305 * 1306 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED. 1307 * 1308 * CONTEXT: 1309 * read_lock(&tasklist_lock), which is released if return value is 1310 * non-zero. Also, grabs and releases @p->sighand->siglock. 1311 * 1312 * RETURNS: 1313 * 0 if wait condition didn't exist and search for other wait conditions 1314 * should continue. Non-zero return, -errno on failure and @p's pid on 1315 * success, implies that tasklist_lock is released and wait condition 1316 * search should terminate. 1317 */ 1318 static int wait_task_stopped(struct wait_opts *wo, 1319 int ptrace, struct task_struct *p) 1320 { 1321 struct waitid_info *infop; 1322 int exit_code, *p_code, why; 1323 uid_t uid = 0; /* unneeded, required by compiler */ 1324 pid_t pid; 1325 1326 /* 1327 * Traditionally we see ptrace'd stopped tasks regardless of options. 1328 */ 1329 if (!ptrace && !(wo->wo_flags & WUNTRACED)) 1330 return 0; 1331 1332 if (!task_stopped_code(p, ptrace)) 1333 return 0; 1334 1335 exit_code = 0; 1336 spin_lock_irq(&p->sighand->siglock); 1337 1338 p_code = task_stopped_code(p, ptrace); 1339 if (unlikely(!p_code)) 1340 goto unlock_sig; 1341 1342 exit_code = *p_code; 1343 if (!exit_code) 1344 goto unlock_sig; 1345 1346 if (!unlikely(wo->wo_flags & WNOWAIT)) 1347 *p_code = 0; 1348 1349 uid = from_kuid_munged(current_user_ns(), task_uid(p)); 1350 unlock_sig: 1351 spin_unlock_irq(&p->sighand->siglock); 1352 if (!exit_code) 1353 return 0; 1354 1355 /* 1356 * Now we are pretty sure this task is interesting. 1357 * Make sure it doesn't get reaped out from under us while we 1358 * give up the lock and then examine it below. We don't want to 1359 * keep holding onto the tasklist_lock while we call getrusage and 1360 * possibly take page faults for user memory. 1361 */ 1362 get_task_struct(p); 1363 pid = task_pid_vnr(p); 1364 why = ptrace ? CLD_TRAPPED : CLD_STOPPED; 1365 read_unlock(&tasklist_lock); 1366 sched_annotate_sleep(); 1367 if (wo->wo_rusage) 1368 getrusage(p, RUSAGE_BOTH, wo->wo_rusage); 1369 put_task_struct(p); 1370 1371 if (likely(!(wo->wo_flags & WNOWAIT))) 1372 wo->wo_stat = (exit_code << 8) | 0x7f; 1373 1374 infop = wo->wo_info; 1375 if (infop) { 1376 infop->cause = why; 1377 infop->status = exit_code; 1378 infop->pid = pid; 1379 infop->uid = uid; 1380 } 1381 return pid; 1382 } 1383 1384 /* 1385 * Handle do_wait work for one task in a live, non-stopped state. 1386 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold 1387 * the lock and this task is uninteresting. If we return nonzero, we have 1388 * released the lock and the system call should return. 1389 */ 1390 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p) 1391 { 1392 struct waitid_info *infop; 1393 pid_t pid; 1394 uid_t uid; 1395 1396 if (!unlikely(wo->wo_flags & WCONTINUED)) 1397 return 0; 1398 1399 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) 1400 return 0; 1401 1402 spin_lock_irq(&p->sighand->siglock); 1403 /* Re-check with the lock held. */ 1404 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) { 1405 spin_unlock_irq(&p->sighand->siglock); 1406 return 0; 1407 } 1408 if (!unlikely(wo->wo_flags & WNOWAIT)) 1409 p->signal->flags &= ~SIGNAL_STOP_CONTINUED; 1410 uid = from_kuid_munged(current_user_ns(), task_uid(p)); 1411 spin_unlock_irq(&p->sighand->siglock); 1412 1413 pid = task_pid_vnr(p); 1414 get_task_struct(p); 1415 read_unlock(&tasklist_lock); 1416 sched_annotate_sleep(); 1417 if (wo->wo_rusage) 1418 getrusage(p, RUSAGE_BOTH, wo->wo_rusage); 1419 put_task_struct(p); 1420 1421 infop = wo->wo_info; 1422 if (!infop) { 1423 wo->wo_stat = 0xffff; 1424 } else { 1425 infop->cause = CLD_CONTINUED; 1426 infop->pid = pid; 1427 infop->uid = uid; 1428 infop->status = SIGCONT; 1429 } 1430 return pid; 1431 } 1432 1433 /* 1434 * Consider @p for a wait by @parent. 1435 * 1436 * -ECHILD should be in ->notask_error before the first call. 1437 * Returns nonzero for a final return, when we have unlocked tasklist_lock. 1438 * Returns zero if the search for a child should continue; 1439 * then ->notask_error is 0 if @p is an eligible child, 1440 * or still -ECHILD. 1441 */ 1442 static int wait_consider_task(struct wait_opts *wo, int ptrace, 1443 struct task_struct *p) 1444 { 1445 /* 1446 * We can race with wait_task_zombie() from another thread. 1447 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition 1448 * can't confuse the checks below. 1449 */ 1450 int exit_state = READ_ONCE(p->exit_state); 1451 int ret; 1452 1453 if (unlikely(exit_state == EXIT_DEAD)) 1454 return 0; 1455 1456 ret = eligible_child(wo, ptrace, p); 1457 if (!ret) 1458 return ret; 1459 1460 if (unlikely(exit_state == EXIT_TRACE)) { 1461 /* 1462 * ptrace == 0 means we are the natural parent. In this case 1463 * we should clear notask_error, debugger will notify us. 1464 */ 1465 if (likely(!ptrace)) 1466 wo->notask_error = 0; 1467 return 0; 1468 } 1469 1470 if (likely(!ptrace) && unlikely(p->ptrace)) { 1471 /* 1472 * If it is traced by its real parent's group, just pretend 1473 * the caller is ptrace_do_wait() and reap this child if it 1474 * is zombie. 1475 * 1476 * This also hides group stop state from real parent; otherwise 1477 * a single stop can be reported twice as group and ptrace stop. 1478 * If a ptracer wants to distinguish these two events for its 1479 * own children it should create a separate process which takes 1480 * the role of real parent. 1481 */ 1482 if (!ptrace_reparented(p)) 1483 ptrace = 1; 1484 } 1485 1486 /* slay zombie? */ 1487 if (exit_state == EXIT_ZOMBIE) { 1488 /* we don't reap group leaders with subthreads */ 1489 if (!delay_group_leader(p)) { 1490 /* 1491 * A zombie ptracee is only visible to its ptracer. 1492 * Notification and reaping will be cascaded to the 1493 * real parent when the ptracer detaches. 1494 */ 1495 if (unlikely(ptrace) || likely(!p->ptrace)) 1496 return wait_task_zombie(wo, p); 1497 } 1498 1499 /* 1500 * Allow access to stopped/continued state via zombie by 1501 * falling through. Clearing of notask_error is complex. 1502 * 1503 * When !@ptrace: 1504 * 1505 * If WEXITED is set, notask_error should naturally be 1506 * cleared. If not, subset of WSTOPPED|WCONTINUED is set, 1507 * so, if there are live subthreads, there are events to 1508 * wait for. If all subthreads are dead, it's still safe 1509 * to clear - this function will be called again in finite 1510 * amount time once all the subthreads are released and 1511 * will then return without clearing. 1512 * 1513 * When @ptrace: 1514 * 1515 * Stopped state is per-task and thus can't change once the 1516 * target task dies. Only continued and exited can happen. 1517 * Clear notask_error if WCONTINUED | WEXITED. 1518 */ 1519 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED))) 1520 wo->notask_error = 0; 1521 } else { 1522 /* 1523 * @p is alive and it's gonna stop, continue or exit, so 1524 * there always is something to wait for. 1525 */ 1526 wo->notask_error = 0; 1527 } 1528 1529 /* 1530 * Wait for stopped. Depending on @ptrace, different stopped state 1531 * is used and the two don't interact with each other. 1532 */ 1533 ret = wait_task_stopped(wo, ptrace, p); 1534 if (ret) 1535 return ret; 1536 1537 /* 1538 * Wait for continued. There's only one continued state and the 1539 * ptracer can consume it which can confuse the real parent. Don't 1540 * use WCONTINUED from ptracer. You don't need or want it. 1541 */ 1542 return wait_task_continued(wo, p); 1543 } 1544 1545 /* 1546 * Do the work of do_wait() for one thread in the group, @tsk. 1547 * 1548 * -ECHILD should be in ->notask_error before the first call. 1549 * Returns nonzero for a final return, when we have unlocked tasklist_lock. 1550 * Returns zero if the search for a child should continue; then 1551 * ->notask_error is 0 if there were any eligible children, 1552 * or still -ECHILD. 1553 */ 1554 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk) 1555 { 1556 struct task_struct *p; 1557 1558 list_for_each_entry(p, &tsk->children, sibling) { 1559 int ret = wait_consider_task(wo, 0, p); 1560 1561 if (ret) 1562 return ret; 1563 } 1564 1565 return 0; 1566 } 1567 1568 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk) 1569 { 1570 struct task_struct *p; 1571 1572 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) { 1573 int ret = wait_consider_task(wo, 1, p); 1574 1575 if (ret) 1576 return ret; 1577 } 1578 1579 return 0; 1580 } 1581 1582 bool pid_child_should_wake(struct wait_opts *wo, struct task_struct *p) 1583 { 1584 if (!eligible_pid(wo, p)) 1585 return false; 1586 1587 if ((wo->wo_flags & __WNOTHREAD) && wo->child_wait.private != p->parent) 1588 return false; 1589 1590 return true; 1591 } 1592 1593 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode, 1594 int sync, void *key) 1595 { 1596 struct wait_opts *wo = container_of(wait, struct wait_opts, 1597 child_wait); 1598 struct task_struct *p = key; 1599 1600 if (pid_child_should_wake(wo, p)) 1601 return default_wake_function(wait, mode, sync, key); 1602 1603 return 0; 1604 } 1605 1606 void __wake_up_parent(struct task_struct *p, struct task_struct *parent) 1607 { 1608 __wake_up_sync_key(&parent->signal->wait_chldexit, 1609 TASK_INTERRUPTIBLE, p); 1610 } 1611 1612 static bool is_effectively_child(struct wait_opts *wo, bool ptrace, 1613 struct task_struct *target) 1614 { 1615 struct task_struct *parent = 1616 !ptrace ? target->real_parent : target->parent; 1617 1618 return current == parent || (!(wo->wo_flags & __WNOTHREAD) && 1619 same_thread_group(current, parent)); 1620 } 1621 1622 /* 1623 * Optimization for waiting on PIDTYPE_PID. No need to iterate through child 1624 * and tracee lists to find the target task. 1625 */ 1626 static int do_wait_pid(struct wait_opts *wo) 1627 { 1628 bool ptrace; 1629 struct task_struct *target; 1630 int retval; 1631 1632 ptrace = false; 1633 target = pid_task(wo->wo_pid, PIDTYPE_TGID); 1634 if (target && is_effectively_child(wo, ptrace, target)) { 1635 retval = wait_consider_task(wo, ptrace, target); 1636 if (retval) 1637 return retval; 1638 } 1639 1640 ptrace = true; 1641 target = pid_task(wo->wo_pid, PIDTYPE_PID); 1642 if (target && target->ptrace && 1643 is_effectively_child(wo, ptrace, target)) { 1644 retval = wait_consider_task(wo, ptrace, target); 1645 if (retval) 1646 return retval; 1647 } 1648 1649 return 0; 1650 } 1651 1652 long __do_wait(struct wait_opts *wo) 1653 { 1654 long retval; 1655 1656 /* 1657 * If there is nothing that can match our criteria, just get out. 1658 * We will clear ->notask_error to zero if we see any child that 1659 * might later match our criteria, even if we are not able to reap 1660 * it yet. 1661 */ 1662 wo->notask_error = -ECHILD; 1663 if ((wo->wo_type < PIDTYPE_MAX) && 1664 (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type))) 1665 goto notask; 1666 1667 read_lock(&tasklist_lock); 1668 1669 if (wo->wo_type == PIDTYPE_PID) { 1670 retval = do_wait_pid(wo); 1671 if (retval) 1672 return retval; 1673 } else { 1674 struct task_struct *tsk = current; 1675 1676 do { 1677 retval = do_wait_thread(wo, tsk); 1678 if (retval) 1679 return retval; 1680 1681 retval = ptrace_do_wait(wo, tsk); 1682 if (retval) 1683 return retval; 1684 1685 if (wo->wo_flags & __WNOTHREAD) 1686 break; 1687 } while_each_thread(current, tsk); 1688 } 1689 read_unlock(&tasklist_lock); 1690 1691 notask: 1692 retval = wo->notask_error; 1693 if (!retval && !(wo->wo_flags & WNOHANG)) 1694 return -ERESTARTSYS; 1695 1696 return retval; 1697 } 1698 1699 static long do_wait(struct wait_opts *wo) 1700 { 1701 int retval; 1702 1703 trace_sched_process_wait(wo->wo_pid); 1704 1705 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback); 1706 wo->child_wait.private = current; 1707 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait); 1708 1709 do { 1710 set_current_state(TASK_INTERRUPTIBLE); 1711 retval = __do_wait(wo); 1712 if (retval != -ERESTARTSYS) 1713 break; 1714 if (signal_pending(current)) 1715 break; 1716 schedule(); 1717 } while (1); 1718 1719 __set_current_state(TASK_RUNNING); 1720 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait); 1721 return retval; 1722 } 1723 1724 int kernel_waitid_prepare(struct wait_opts *wo, int which, pid_t upid, 1725 struct waitid_info *infop, int options, 1726 struct rusage *ru) 1727 { 1728 unsigned int f_flags = 0; 1729 struct pid *pid = NULL; 1730 enum pid_type type; 1731 1732 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED| 1733 __WNOTHREAD|__WCLONE|__WALL)) 1734 return -EINVAL; 1735 if (!(options & (WEXITED|WSTOPPED|WCONTINUED))) 1736 return -EINVAL; 1737 1738 switch (which) { 1739 case P_ALL: 1740 type = PIDTYPE_MAX; 1741 break; 1742 case P_PID: 1743 type = PIDTYPE_PID; 1744 if (upid <= 0) 1745 return -EINVAL; 1746 1747 pid = find_get_pid(upid); 1748 break; 1749 case P_PGID: 1750 type = PIDTYPE_PGID; 1751 if (upid < 0) 1752 return -EINVAL; 1753 1754 if (upid) 1755 pid = find_get_pid(upid); 1756 else 1757 pid = get_task_pid(current, PIDTYPE_PGID); 1758 break; 1759 case P_PIDFD: 1760 type = PIDTYPE_PID; 1761 if (upid < 0) 1762 return -EINVAL; 1763 1764 pid = pidfd_get_pid(upid, &f_flags); 1765 if (IS_ERR(pid)) 1766 return PTR_ERR(pid); 1767 1768 break; 1769 default: 1770 return -EINVAL; 1771 } 1772 1773 wo->wo_type = type; 1774 wo->wo_pid = pid; 1775 wo->wo_flags = options; 1776 wo->wo_info = infop; 1777 wo->wo_rusage = ru; 1778 if (f_flags & O_NONBLOCK) 1779 wo->wo_flags |= WNOHANG; 1780 1781 return 0; 1782 } 1783 1784 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop, 1785 int options, struct rusage *ru) 1786 { 1787 struct wait_opts wo; 1788 long ret; 1789 1790 ret = kernel_waitid_prepare(&wo, which, upid, infop, options, ru); 1791 if (ret) 1792 return ret; 1793 1794 ret = do_wait(&wo); 1795 if (!ret && !(options & WNOHANG) && (wo.wo_flags & WNOHANG)) 1796 ret = -EAGAIN; 1797 1798 put_pid(wo.wo_pid); 1799 return ret; 1800 } 1801 1802 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *, 1803 infop, int, options, struct rusage __user *, ru) 1804 { 1805 struct rusage r; 1806 struct waitid_info info = {.status = 0}; 1807 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL); 1808 int signo = 0; 1809 1810 if (err > 0) { 1811 signo = SIGCHLD; 1812 err = 0; 1813 if (ru && copy_to_user(ru, &r, sizeof(struct rusage))) 1814 return -EFAULT; 1815 } 1816 if (!infop) 1817 return err; 1818 1819 if (!user_write_access_begin(infop, sizeof(*infop))) 1820 return -EFAULT; 1821 1822 unsafe_put_user(signo, &infop->si_signo, Efault); 1823 unsafe_put_user(0, &infop->si_errno, Efault); 1824 unsafe_put_user(info.cause, &infop->si_code, Efault); 1825 unsafe_put_user(info.pid, &infop->si_pid, Efault); 1826 unsafe_put_user(info.uid, &infop->si_uid, Efault); 1827 unsafe_put_user(info.status, &infop->si_status, Efault); 1828 user_write_access_end(); 1829 return err; 1830 Efault: 1831 user_write_access_end(); 1832 return -EFAULT; 1833 } 1834 1835 long kernel_wait4(pid_t upid, int __user *stat_addr, int options, 1836 struct rusage *ru) 1837 { 1838 struct wait_opts wo; 1839 struct pid *pid = NULL; 1840 enum pid_type type; 1841 long ret; 1842 1843 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED| 1844 __WNOTHREAD|__WCLONE|__WALL)) 1845 return -EINVAL; 1846 1847 /* -INT_MIN is not defined */ 1848 if (upid == INT_MIN) 1849 return -ESRCH; 1850 1851 if (upid == -1) 1852 type = PIDTYPE_MAX; 1853 else if (upid < 0) { 1854 type = PIDTYPE_PGID; 1855 pid = find_get_pid(-upid); 1856 } else if (upid == 0) { 1857 type = PIDTYPE_PGID; 1858 pid = get_task_pid(current, PIDTYPE_PGID); 1859 } else /* upid > 0 */ { 1860 type = PIDTYPE_PID; 1861 pid = find_get_pid(upid); 1862 } 1863 1864 wo.wo_type = type; 1865 wo.wo_pid = pid; 1866 wo.wo_flags = options | WEXITED; 1867 wo.wo_info = NULL; 1868 wo.wo_stat = 0; 1869 wo.wo_rusage = ru; 1870 ret = do_wait(&wo); 1871 put_pid(pid); 1872 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr)) 1873 ret = -EFAULT; 1874 1875 return ret; 1876 } 1877 1878 int kernel_wait(pid_t pid, int *stat) 1879 { 1880 struct wait_opts wo = { 1881 .wo_type = PIDTYPE_PID, 1882 .wo_pid = find_get_pid(pid), 1883 .wo_flags = WEXITED, 1884 }; 1885 int ret; 1886 1887 ret = do_wait(&wo); 1888 if (ret > 0 && wo.wo_stat) 1889 *stat = wo.wo_stat; 1890 put_pid(wo.wo_pid); 1891 return ret; 1892 } 1893 1894 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr, 1895 int, options, struct rusage __user *, ru) 1896 { 1897 struct rusage r; 1898 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL); 1899 1900 if (err > 0) { 1901 if (ru && copy_to_user(ru, &r, sizeof(struct rusage))) 1902 return -EFAULT; 1903 } 1904 return err; 1905 } 1906 1907 #ifdef __ARCH_WANT_SYS_WAITPID 1908 1909 /* 1910 * sys_waitpid() remains for compatibility. waitpid() should be 1911 * implemented by calling sys_wait4() from libc.a. 1912 */ 1913 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options) 1914 { 1915 return kernel_wait4(pid, stat_addr, options, NULL); 1916 } 1917 1918 #endif 1919 1920 #ifdef CONFIG_COMPAT 1921 COMPAT_SYSCALL_DEFINE4(wait4, 1922 compat_pid_t, pid, 1923 compat_uint_t __user *, stat_addr, 1924 int, options, 1925 struct compat_rusage __user *, ru) 1926 { 1927 struct rusage r; 1928 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL); 1929 if (err > 0) { 1930 if (ru && put_compat_rusage(&r, ru)) 1931 return -EFAULT; 1932 } 1933 return err; 1934 } 1935 1936 COMPAT_SYSCALL_DEFINE5(waitid, 1937 int, which, compat_pid_t, pid, 1938 struct compat_siginfo __user *, infop, int, options, 1939 struct compat_rusage __user *, uru) 1940 { 1941 struct rusage ru; 1942 struct waitid_info info = {.status = 0}; 1943 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL); 1944 int signo = 0; 1945 if (err > 0) { 1946 signo = SIGCHLD; 1947 err = 0; 1948 if (uru) { 1949 /* kernel_waitid() overwrites everything in ru */ 1950 if (COMPAT_USE_64BIT_TIME) 1951 err = copy_to_user(uru, &ru, sizeof(ru)); 1952 else 1953 err = put_compat_rusage(&ru, uru); 1954 if (err) 1955 return -EFAULT; 1956 } 1957 } 1958 1959 if (!infop) 1960 return err; 1961 1962 if (!user_write_access_begin(infop, sizeof(*infop))) 1963 return -EFAULT; 1964 1965 unsafe_put_user(signo, &infop->si_signo, Efault); 1966 unsafe_put_user(0, &infop->si_errno, Efault); 1967 unsafe_put_user(info.cause, &infop->si_code, Efault); 1968 unsafe_put_user(info.pid, &infop->si_pid, Efault); 1969 unsafe_put_user(info.uid, &infop->si_uid, Efault); 1970 unsafe_put_user(info.status, &infop->si_status, Efault); 1971 user_write_access_end(); 1972 return err; 1973 Efault: 1974 user_write_access_end(); 1975 return -EFAULT; 1976 } 1977 #endif 1978 1979 /* 1980 * This needs to be __function_aligned as GCC implicitly makes any 1981 * implementation of abort() cold and drops alignment specified by 1982 * -falign-functions=N. 1983 * 1984 * See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88345#c11 1985 */ 1986 __weak __function_aligned void abort(void) 1987 { 1988 BUG(); 1989 1990 /* if that doesn't kill us, halt */ 1991 panic("Oops failed to kill thread"); 1992 } 1993 EXPORT_SYMBOL(abort); 1994