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/fdtable.h> 29#include <linux/freezer.h> 30#include <linux/binfmts.h> 31#include <linux/nsproxy.h> 32#include <linux/pid_namespace.h> 33#include <linux/ptrace.h> 34#include <linux/profile.h> 35#include <linux/mount.h> 36#include <linux/proc_fs.h> 37#include <linux/kthread.h> 38#include <linux/mempolicy.h> 39#include <linux/taskstats_kern.h> 40#include <linux/delayacct.h> 41#include <linux/cgroup.h> 42#include <linux/syscalls.h> 43#include <linux/signal.h> 44#include <linux/posix-timers.h> 45#include <linux/cn_proc.h> 46#include <linux/mutex.h> 47#include <linux/futex.h> 48#include <linux/pipe_fs_i.h> 49#include <linux/audit.h> /* for audit_free() */ 50#include <linux/resource.h> 51#include <linux/task_io_accounting_ops.h> 52#include <linux/blkdev.h> 53#include <linux/task_work.h> 54#include <linux/fs_struct.h> 55#include <linux/init_task.h> 56#include <linux/perf_event.h> 57#include <trace/events/sched.h> 58#include <linux/hw_breakpoint.h> 59#include <linux/oom.h> 60#include <linux/writeback.h> 61#include <linux/shm.h> 62#include <linux/kcov.h> 63#include <linux/kmsan.h> 64#include <linux/random.h> 65#include <linux/rcuwait.h> 66#include <linux/compat.h> 67#include <linux/io_uring.h> 68#include <linux/kprobes.h> 69#include <linux/rethook.h> 70#include <linux/sysfs.h> 71#include <linux/user_events.h> 72#include <linux/uaccess.h> 73 74#include <uapi/linux/wait.h> 75 76#include <asm/unistd.h> 77#include <asm/mmu_context.h> 78 79#include "exit.h" 80 81/* 82 * The default value should be high enough to not crash a system that randomly 83 * crashes its kernel from time to time, but low enough to at least not permit 84 * overflowing 32-bit refcounts or the ldsem writer count. 85 */ 86static unsigned int oops_limit = 10000; 87 88#ifdef CONFIG_SYSCTL 89static struct ctl_table kern_exit_table[] = { 90 { 91 .procname = "oops_limit", 92 .data = &oops_limit, 93 .maxlen = sizeof(oops_limit), 94 .mode = 0644, 95 .proc_handler = proc_douintvec, 96 }, 97}; 98 99static __init int kernel_exit_sysctls_init(void) 100{ 101 register_sysctl_init("kernel", kern_exit_table); 102 return 0; 103} 104late_initcall(kernel_exit_sysctls_init); 105#endif 106 107static atomic_t oops_count = ATOMIC_INIT(0); 108 109#ifdef CONFIG_SYSFS 110static ssize_t oops_count_show(struct kobject *kobj, struct kobj_attribute *attr, 111 char *page) 112{ 113 return sysfs_emit(page, "%d\n", atomic_read(&oops_count)); 114} 115 116static struct kobj_attribute oops_count_attr = __ATTR_RO(oops_count); 117 118static __init int kernel_exit_sysfs_init(void) 119{ 120 sysfs_add_file_to_group(kernel_kobj, &oops_count_attr.attr, NULL); 121 return 0; 122} 123late_initcall(kernel_exit_sysfs_init); 124#endif 125 126static void __unhash_process(struct task_struct *p, bool group_dead) 127{ 128 nr_threads--; 129 detach_pid(p, PIDTYPE_PID); 130 if (group_dead) { 131 detach_pid(p, PIDTYPE_TGID); 132 detach_pid(p, PIDTYPE_PGID); 133 detach_pid(p, PIDTYPE_SID); 134 135 list_del_rcu(&p->tasks); 136 list_del_init(&p->sibling); 137 __this_cpu_dec(process_counts); 138 } 139 list_del_rcu(&p->thread_node); 140} 141 142/* 143 * This function expects the tasklist_lock write-locked. 144 */ 145static void __exit_signal(struct task_struct *tsk) 146{ 147 struct signal_struct *sig = tsk->signal; 148 bool group_dead = thread_group_leader(tsk); 149 struct sighand_struct *sighand; 150 struct tty_struct *tty; 151 u64 utime, stime; 152 153 sighand = rcu_dereference_check(tsk->sighand, 154 lockdep_tasklist_lock_is_held()); 155 spin_lock(&sighand->siglock); 156 157#ifdef CONFIG_POSIX_TIMERS 158 posix_cpu_timers_exit(tsk); 159 if (group_dead) 160 posix_cpu_timers_exit_group(tsk); 161#endif 162 163 if (group_dead) { 164 tty = sig->tty; 165 sig->tty = NULL; 166 } else { 167 /* 168 * If there is any task waiting for the group exit 169 * then notify it: 170 */ 171 if (sig->notify_count > 0 && !--sig->notify_count) 172 wake_up_process(sig->group_exec_task); 173 174 if (tsk == sig->curr_target) 175 sig->curr_target = next_thread(tsk); 176 } 177 178 add_device_randomness((const void*) &tsk->se.sum_exec_runtime, 179 sizeof(unsigned long long)); 180 181 /* 182 * Accumulate here the counters for all threads as they die. We could 183 * skip the group leader because it is the last user of signal_struct, 184 * but we want to avoid the race with thread_group_cputime() which can 185 * see the empty ->thread_head list. 186 */ 187 task_cputime(tsk, &utime, &stime); 188 write_seqlock(&sig->stats_lock); 189 sig->utime += utime; 190 sig->stime += stime; 191 sig->gtime += task_gtime(tsk); 192 sig->min_flt += tsk->min_flt; 193 sig->maj_flt += tsk->maj_flt; 194 sig->nvcsw += tsk->nvcsw; 195 sig->nivcsw += tsk->nivcsw; 196 sig->inblock += task_io_get_inblock(tsk); 197 sig->oublock += task_io_get_oublock(tsk); 198 task_io_accounting_add(&sig->ioac, &tsk->ioac); 199 sig->sum_sched_runtime += tsk->se.sum_exec_runtime; 200 sig->nr_threads--; 201 __unhash_process(tsk, group_dead); 202 write_sequnlock(&sig->stats_lock); 203 204 /* 205 * Do this under ->siglock, we can race with another thread 206 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals. 207 */ 208 flush_sigqueue(&tsk->pending); 209 tsk->sighand = NULL; 210 spin_unlock(&sighand->siglock); 211 212 __cleanup_sighand(sighand); 213 clear_tsk_thread_flag(tsk, TIF_SIGPENDING); 214 if (group_dead) { 215 flush_sigqueue(&sig->shared_pending); 216 tty_kref_put(tty); 217 } 218} 219 220static void delayed_put_task_struct(struct rcu_head *rhp) 221{ 222 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu); 223 224 kprobe_flush_task(tsk); 225 rethook_flush_task(tsk); 226 perf_event_delayed_put(tsk); 227 trace_sched_process_free(tsk); 228 put_task_struct(tsk); 229} 230 231void put_task_struct_rcu_user(struct task_struct *task) 232{ 233 if (refcount_dec_and_test(&task->rcu_users)) 234 call_rcu(&task->rcu, delayed_put_task_struct); 235} 236 237void __weak release_thread(struct task_struct *dead_task) 238{ 239} 240 241void release_task(struct task_struct *p) 242{ 243 struct task_struct *leader; 244 struct pid *thread_pid; 245 int zap_leader; 246repeat: 247 /* don't need to get the RCU readlock here - the process is dead and 248 * can't be modifying its own credentials. But shut RCU-lockdep up */ 249 rcu_read_lock(); 250 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1); 251 rcu_read_unlock(); 252 253 cgroup_release(p); 254 255 write_lock_irq(&tasklist_lock); 256 ptrace_release_task(p); 257 thread_pid = get_pid(p->thread_pid); 258 __exit_signal(p); 259 260 /* 261 * If we are the last non-leader member of the thread 262 * group, and the leader is zombie, then notify the 263 * group leader's parent process. (if it wants notification.) 264 */ 265 zap_leader = 0; 266 leader = p->group_leader; 267 if (leader != p && thread_group_empty(leader) 268 && leader->exit_state == EXIT_ZOMBIE) { 269 /* 270 * If we were the last child thread and the leader has 271 * exited already, and the leader's parent ignores SIGCHLD, 272 * then we are the one who should release the leader. 273 */ 274 zap_leader = do_notify_parent(leader, leader->exit_signal); 275 if (zap_leader) 276 leader->exit_state = EXIT_DEAD; 277 } 278 279 write_unlock_irq(&tasklist_lock); 280 seccomp_filter_release(p); 281 proc_flush_pid(thread_pid); 282 put_pid(thread_pid); 283 release_thread(p); 284 put_task_struct_rcu_user(p); 285 286 p = leader; 287 if (unlikely(zap_leader)) 288 goto repeat; 289} 290 291int rcuwait_wake_up(struct rcuwait *w) 292{ 293 int ret = 0; 294 struct task_struct *task; 295 296 rcu_read_lock(); 297 298 /* 299 * Order condition vs @task, such that everything prior to the load 300 * of @task is visible. This is the condition as to why the user called 301 * rcuwait_wake() in the first place. Pairs with set_current_state() 302 * barrier (A) in rcuwait_wait_event(). 303 * 304 * WAIT WAKE 305 * [S] tsk = current [S] cond = true 306 * MB (A) MB (B) 307 * [L] cond [L] tsk 308 */ 309 smp_mb(); /* (B) */ 310 311 task = rcu_dereference(w->task); 312 if (task) 313 ret = wake_up_process(task); 314 rcu_read_unlock(); 315 316 return ret; 317} 318EXPORT_SYMBOL_GPL(rcuwait_wake_up); 319 320/* 321 * Determine if a process group is "orphaned", according to the POSIX 322 * definition in 2.2.2.52. Orphaned process groups are not to be affected 323 * by terminal-generated stop signals. Newly orphaned process groups are 324 * to receive a SIGHUP and a SIGCONT. 325 * 326 * "I ask you, have you ever known what it is to be an orphan?" 327 */ 328static int will_become_orphaned_pgrp(struct pid *pgrp, 329 struct task_struct *ignored_task) 330{ 331 struct task_struct *p; 332 333 do_each_pid_task(pgrp, PIDTYPE_PGID, p) { 334 if ((p == ignored_task) || 335 (p->exit_state && thread_group_empty(p)) || 336 is_global_init(p->real_parent)) 337 continue; 338 339 if (task_pgrp(p->real_parent) != pgrp && 340 task_session(p->real_parent) == task_session(p)) 341 return 0; 342 } while_each_pid_task(pgrp, PIDTYPE_PGID, p); 343 344 return 1; 345} 346 347int is_current_pgrp_orphaned(void) 348{ 349 int retval; 350 351 read_lock(&tasklist_lock); 352 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL); 353 read_unlock(&tasklist_lock); 354 355 return retval; 356} 357 358static bool has_stopped_jobs(struct pid *pgrp) 359{ 360 struct task_struct *p; 361 362 do_each_pid_task(pgrp, PIDTYPE_PGID, p) { 363 if (p->signal->flags & SIGNAL_STOP_STOPPED) 364 return true; 365 } while_each_pid_task(pgrp, PIDTYPE_PGID, p); 366 367 return false; 368} 369 370/* 371 * Check to see if any process groups have become orphaned as 372 * a result of our exiting, and if they have any stopped jobs, 373 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2) 374 */ 375static void 376kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent) 377{ 378 struct pid *pgrp = task_pgrp(tsk); 379 struct task_struct *ignored_task = tsk; 380 381 if (!parent) 382 /* exit: our father is in a different pgrp than 383 * we are and we were the only connection outside. 384 */ 385 parent = tsk->real_parent; 386 else 387 /* reparent: our child is in a different pgrp than 388 * we are, and it was the only connection outside. 389 */ 390 ignored_task = NULL; 391 392 if (task_pgrp(parent) != pgrp && 393 task_session(parent) == task_session(tsk) && 394 will_become_orphaned_pgrp(pgrp, ignored_task) && 395 has_stopped_jobs(pgrp)) { 396 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp); 397 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp); 398 } 399} 400 401static void coredump_task_exit(struct task_struct *tsk) 402{ 403 struct core_state *core_state; 404 405 /* 406 * Serialize with any possible pending coredump. 407 * We must hold siglock around checking core_state 408 * and setting PF_POSTCOREDUMP. The core-inducing thread 409 * will increment ->nr_threads for each thread in the 410 * group without PF_POSTCOREDUMP set. 411 */ 412 spin_lock_irq(&tsk->sighand->siglock); 413 tsk->flags |= PF_POSTCOREDUMP; 414 core_state = tsk->signal->core_state; 415 spin_unlock_irq(&tsk->sighand->siglock); 416 417 /* The vhost_worker does not particpate in coredumps */ 418 if (core_state && 419 ((tsk->flags & (PF_IO_WORKER | PF_USER_WORKER)) != PF_USER_WORKER)) { 420 struct core_thread self; 421 422 self.task = current; 423 if (self.task->flags & PF_SIGNALED) 424 self.next = xchg(&core_state->dumper.next, &self); 425 else 426 self.task = NULL; 427 /* 428 * Implies mb(), the result of xchg() must be visible 429 * to core_state->dumper. 430 */ 431 if (atomic_dec_and_test(&core_state->nr_threads)) 432 complete(&core_state->startup); 433 434 for (;;) { 435 set_current_state(TASK_UNINTERRUPTIBLE|TASK_FREEZABLE); 436 if (!self.task) /* see coredump_finish() */ 437 break; 438 schedule(); 439 } 440 __set_current_state(TASK_RUNNING); 441 } 442} 443 444#ifdef CONFIG_MEMCG 445/* 446 * A task is exiting. If it owned this mm, find a new owner for the mm. 447 */ 448void mm_update_next_owner(struct mm_struct *mm) 449{ 450 struct task_struct *c, *g, *p = current; 451 452retry: 453 /* 454 * If the exiting or execing task is not the owner, it's 455 * someone else's problem. 456 */ 457 if (mm->owner != p) 458 return; 459 /* 460 * The current owner is exiting/execing and there are no other 461 * candidates. Do not leave the mm pointing to a possibly 462 * freed task structure. 463 */ 464 if (atomic_read(&mm->mm_users) <= 1) { 465 WRITE_ONCE(mm->owner, NULL); 466 return; 467 } 468 469 read_lock(&tasklist_lock); 470 /* 471 * Search in the children 472 */ 473 list_for_each_entry(c, &p->children, sibling) { 474 if (c->mm == mm) 475 goto assign_new_owner; 476 } 477 478 /* 479 * Search in the siblings 480 */ 481 list_for_each_entry(c, &p->real_parent->children, sibling) { 482 if (c->mm == mm) 483 goto assign_new_owner; 484 } 485 486 /* 487 * Search through everything else, we should not get here often. 488 */ 489 for_each_process(g) { 490 if (g->flags & PF_KTHREAD) 491 continue; 492 for_each_thread(g, c) { 493 if (c->mm == mm) 494 goto assign_new_owner; 495 if (c->mm) 496 break; 497 } 498 } 499 read_unlock(&tasklist_lock); 500 /* 501 * We found no owner yet mm_users > 1: this implies that we are 502 * most likely racing with swapoff (try_to_unuse()) or /proc or 503 * ptrace or page migration (get_task_mm()). Mark owner as NULL. 504 */ 505 WRITE_ONCE(mm->owner, NULL); 506 return; 507 508assign_new_owner: 509 BUG_ON(c == p); 510 get_task_struct(c); 511 /* 512 * The task_lock protects c->mm from changing. 513 * We always want mm->owner->mm == mm 514 */ 515 task_lock(c); 516 /* 517 * Delay read_unlock() till we have the task_lock() 518 * to ensure that c does not slip away underneath us 519 */ 520 read_unlock(&tasklist_lock); 521 if (c->mm != mm) { 522 task_unlock(c); 523 put_task_struct(c); 524 goto retry; 525 } 526 WRITE_ONCE(mm->owner, c); 527 lru_gen_migrate_mm(mm); 528 task_unlock(c); 529 put_task_struct(c); 530} 531#endif /* CONFIG_MEMCG */ 532 533/* 534 * Turn us into a lazy TLB process if we 535 * aren't already.. 536 */ 537static void exit_mm(void) 538{ 539 struct mm_struct *mm = current->mm; 540 541 exit_mm_release(current, mm); 542 if (!mm) 543 return; 544 mmap_read_lock(mm); 545 mmgrab_lazy_tlb(mm); 546 BUG_ON(mm != current->active_mm); 547 /* more a memory barrier than a real lock */ 548 task_lock(current); 549 /* 550 * When a thread stops operating on an address space, the loop 551 * in membarrier_private_expedited() may not observe that 552 * tsk->mm, and the loop in membarrier_global_expedited() may 553 * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED 554 * rq->membarrier_state, so those would not issue an IPI. 555 * Membarrier requires a memory barrier after accessing 556 * user-space memory, before clearing tsk->mm or the 557 * rq->membarrier_state. 558 */ 559 smp_mb__after_spinlock(); 560 local_irq_disable(); 561 current->mm = NULL; 562 membarrier_update_current_mm(NULL); 563 enter_lazy_tlb(mm, current); 564 local_irq_enable(); 565 task_unlock(current); 566 mmap_read_unlock(mm); 567 mm_update_next_owner(mm); 568 mmput(mm); 569 if (test_thread_flag(TIF_MEMDIE)) 570 exit_oom_victim(); 571} 572 573static struct task_struct *find_alive_thread(struct task_struct *p) 574{ 575 struct task_struct *t; 576 577 for_each_thread(p, t) { 578 if (!(t->flags & PF_EXITING)) 579 return t; 580 } 581 return NULL; 582} 583 584static struct task_struct *find_child_reaper(struct task_struct *father, 585 struct list_head *dead) 586 __releases(&tasklist_lock) 587 __acquires(&tasklist_lock) 588{ 589 struct pid_namespace *pid_ns = task_active_pid_ns(father); 590 struct task_struct *reaper = pid_ns->child_reaper; 591 struct task_struct *p, *n; 592 593 if (likely(reaper != father)) 594 return reaper; 595 596 reaper = find_alive_thread(father); 597 if (reaper) { 598 pid_ns->child_reaper = reaper; 599 return reaper; 600 } 601 602 write_unlock_irq(&tasklist_lock); 603 604 list_for_each_entry_safe(p, n, dead, ptrace_entry) { 605 list_del_init(&p->ptrace_entry); 606 release_task(p); 607 } 608 609 zap_pid_ns_processes(pid_ns); 610 write_lock_irq(&tasklist_lock); 611 612 return father; 613} 614 615/* 616 * When we die, we re-parent all our children, and try to: 617 * 1. give them to another thread in our thread group, if such a member exists 618 * 2. give it to the first ancestor process which prctl'd itself as a 619 * child_subreaper for its children (like a service manager) 620 * 3. give it to the init process (PID 1) in our pid namespace 621 */ 622static struct task_struct *find_new_reaper(struct task_struct *father, 623 struct task_struct *child_reaper) 624{ 625 struct task_struct *thread, *reaper; 626 627 thread = find_alive_thread(father); 628 if (thread) 629 return thread; 630 631 if (father->signal->has_child_subreaper) { 632 unsigned int ns_level = task_pid(father)->level; 633 /* 634 * Find the first ->is_child_subreaper ancestor in our pid_ns. 635 * We can't check reaper != child_reaper to ensure we do not 636 * cross the namespaces, the exiting parent could be injected 637 * by setns() + fork(). 638 * We check pid->level, this is slightly more efficient than 639 * task_active_pid_ns(reaper) != task_active_pid_ns(father). 640 */ 641 for (reaper = father->real_parent; 642 task_pid(reaper)->level == ns_level; 643 reaper = reaper->real_parent) { 644 if (reaper == &init_task) 645 break; 646 if (!reaper->signal->is_child_subreaper) 647 continue; 648 thread = find_alive_thread(reaper); 649 if (thread) 650 return thread; 651 } 652 } 653 654 return child_reaper; 655} 656 657/* 658* Any that need to be release_task'd are put on the @dead list. 659 */ 660static void reparent_leader(struct task_struct *father, struct task_struct *p, 661 struct list_head *dead) 662{ 663 if (unlikely(p->exit_state == EXIT_DEAD)) 664 return; 665 666 /* We don't want people slaying init. */ 667 p->exit_signal = SIGCHLD; 668 669 /* If it has exited notify the new parent about this child's death. */ 670 if (!p->ptrace && 671 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) { 672 if (do_notify_parent(p, p->exit_signal)) { 673 p->exit_state = EXIT_DEAD; 674 list_add(&p->ptrace_entry, dead); 675 } 676 } 677 678 kill_orphaned_pgrp(p, father); 679} 680 681/* 682 * This does two things: 683 * 684 * A. Make init inherit all the child processes 685 * B. Check to see if any process groups have become orphaned 686 * as a result of our exiting, and if they have any stopped 687 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2) 688 */ 689static void forget_original_parent(struct task_struct *father, 690 struct list_head *dead) 691{ 692 struct task_struct *p, *t, *reaper; 693 694 if (unlikely(!list_empty(&father->ptraced))) 695 exit_ptrace(father, dead); 696 697 /* Can drop and reacquire tasklist_lock */ 698 reaper = find_child_reaper(father, dead); 699 if (list_empty(&father->children)) 700 return; 701 702 reaper = find_new_reaper(father, reaper); 703 list_for_each_entry(p, &father->children, sibling) { 704 for_each_thread(p, t) { 705 RCU_INIT_POINTER(t->real_parent, reaper); 706 BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father)); 707 if (likely(!t->ptrace)) 708 t->parent = t->real_parent; 709 if (t->pdeath_signal) 710 group_send_sig_info(t->pdeath_signal, 711 SEND_SIG_NOINFO, t, 712 PIDTYPE_TGID); 713 } 714 /* 715 * If this is a threaded reparent there is no need to 716 * notify anyone anything has happened. 717 */ 718 if (!same_thread_group(reaper, father)) 719 reparent_leader(father, p, dead); 720 } 721 list_splice_tail_init(&father->children, &reaper->children); 722} 723 724/* 725 * Send signals to all our closest relatives so that they know 726 * to properly mourn us.. 727 */ 728static void exit_notify(struct task_struct *tsk, int group_dead) 729{ 730 bool autoreap; 731 struct task_struct *p, *n; 732 LIST_HEAD(dead); 733 734 write_lock_irq(&tasklist_lock); 735 forget_original_parent(tsk, &dead); 736 737 if (group_dead) 738 kill_orphaned_pgrp(tsk->group_leader, NULL); 739 740 tsk->exit_state = EXIT_ZOMBIE; 741 /* 742 * sub-thread or delay_group_leader(), wake up the 743 * PIDFD_THREAD waiters. 744 */ 745 if (!thread_group_empty(tsk)) 746 do_notify_pidfd(tsk); 747 748 if (unlikely(tsk->ptrace)) { 749 int sig = thread_group_leader(tsk) && 750 thread_group_empty(tsk) && 751 !ptrace_reparented(tsk) ? 752 tsk->exit_signal : SIGCHLD; 753 autoreap = do_notify_parent(tsk, sig); 754 } else if (thread_group_leader(tsk)) { 755 autoreap = thread_group_empty(tsk) && 756 do_notify_parent(tsk, tsk->exit_signal); 757 } else { 758 autoreap = true; 759 } 760 761 if (autoreap) { 762 tsk->exit_state = EXIT_DEAD; 763 list_add(&tsk->ptrace_entry, &dead); 764 } 765 766 /* mt-exec, de_thread() is waiting for group leader */ 767 if (unlikely(tsk->signal->notify_count < 0)) 768 wake_up_process(tsk->signal->group_exec_task); 769 write_unlock_irq(&tasklist_lock); 770 771 list_for_each_entry_safe(p, n, &dead, ptrace_entry) { 772 list_del_init(&p->ptrace_entry); 773 release_task(p); 774 } 775} 776 777#ifdef CONFIG_DEBUG_STACK_USAGE 778static void check_stack_usage(void) 779{ 780 static DEFINE_SPINLOCK(low_water_lock); 781 static int lowest_to_date = THREAD_SIZE; 782 unsigned long free; 783 784 free = stack_not_used(current); 785 786 if (free >= lowest_to_date) 787 return; 788 789 spin_lock(&low_water_lock); 790 if (free < lowest_to_date) { 791 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n", 792 current->comm, task_pid_nr(current), free); 793 lowest_to_date = free; 794 } 795 spin_unlock(&low_water_lock); 796} 797#else 798static inline void check_stack_usage(void) {} 799#endif 800 801static void synchronize_group_exit(struct task_struct *tsk, long code) 802{ 803 struct sighand_struct *sighand = tsk->sighand; 804 struct signal_struct *signal = tsk->signal; 805 806 spin_lock_irq(&sighand->siglock); 807 signal->quick_threads--; 808 if ((signal->quick_threads == 0) && 809 !(signal->flags & SIGNAL_GROUP_EXIT)) { 810 signal->flags = SIGNAL_GROUP_EXIT; 811 signal->group_exit_code = code; 812 signal->group_stop_count = 0; 813 } 814 spin_unlock_irq(&sighand->siglock); 815} 816 817void __noreturn do_exit(long code) 818{ 819 struct task_struct *tsk = current; 820 int group_dead; 821 822 WARN_ON(irqs_disabled()); 823 824 synchronize_group_exit(tsk, code); 825 826 WARN_ON(tsk->plug); 827 828 kcov_task_exit(tsk); 829 kmsan_task_exit(tsk); 830 831 coredump_task_exit(tsk); 832 ptrace_event(PTRACE_EVENT_EXIT, code); 833 user_events_exit(tsk); 834 835 io_uring_files_cancel(); 836 exit_signals(tsk); /* sets PF_EXITING */ 837 838 acct_update_integrals(tsk); 839 group_dead = atomic_dec_and_test(&tsk->signal->live); 840 if (group_dead) { 841 /* 842 * If the last thread of global init has exited, panic 843 * immediately to get a useable coredump. 844 */ 845 if (unlikely(is_global_init(tsk))) 846 panic("Attempted to kill init! exitcode=0x%08x\n", 847 tsk->signal->group_exit_code ?: (int)code); 848 849#ifdef CONFIG_POSIX_TIMERS 850 hrtimer_cancel(&tsk->signal->real_timer); 851 exit_itimers(tsk); 852#endif 853 if (tsk->mm) 854 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm); 855 } 856 acct_collect(code, group_dead); 857 if (group_dead) 858 tty_audit_exit(); 859 audit_free(tsk); 860 861 tsk->exit_code = code; 862 taskstats_exit(tsk, group_dead); 863 864 exit_mm(); 865 866 if (group_dead) 867 acct_process(); 868 trace_sched_process_exit(tsk); 869 870 exit_sem(tsk); 871 exit_shm(tsk); 872 exit_files(tsk); 873 exit_fs(tsk); 874 if (group_dead) 875 disassociate_ctty(1); 876 exit_task_namespaces(tsk); 877 exit_task_work(tsk); 878 exit_thread(tsk); 879 880 /* 881 * Flush inherited counters to the parent - before the parent 882 * gets woken up by child-exit notifications. 883 * 884 * because of cgroup mode, must be called before cgroup_exit() 885 */ 886 perf_event_exit_task(tsk); 887 888 sched_autogroup_exit_task(tsk); 889 cgroup_exit(tsk); 890 891 /* 892 * FIXME: do that only when needed, using sched_exit tracepoint 893 */ 894 flush_ptrace_hw_breakpoint(tsk); 895 896 exit_tasks_rcu_start(); 897 exit_notify(tsk, group_dead); 898 proc_exit_connector(tsk); 899 mpol_put_task_policy(tsk); 900#ifdef CONFIG_FUTEX 901 if (unlikely(current->pi_state_cache)) 902 kfree(current->pi_state_cache); 903#endif 904 /* 905 * Make sure we are holding no locks: 906 */ 907 debug_check_no_locks_held(); 908 909 if (tsk->io_context) 910 exit_io_context(tsk); 911 912 if (tsk->splice_pipe) 913 free_pipe_info(tsk->splice_pipe); 914 915 if (tsk->task_frag.page) 916 put_page(tsk->task_frag.page); 917 918 exit_task_stack_account(tsk); 919 920 check_stack_usage(); 921 preempt_disable(); 922 if (tsk->nr_dirtied) 923 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied); 924 exit_rcu(); 925 exit_tasks_rcu_finish(); 926 927 lockdep_free_task(tsk); 928 do_task_dead(); 929} 930 931void __noreturn make_task_dead(int signr) 932{ 933 /* 934 * Take the task off the cpu after something catastrophic has 935 * happened. 936 * 937 * We can get here from a kernel oops, sometimes with preemption off. 938 * Start by checking for critical errors. 939 * Then fix up important state like USER_DS and preemption. 940 * Then do everything else. 941 */ 942 struct task_struct *tsk = current; 943 unsigned int limit; 944 945 if (unlikely(in_interrupt())) 946 panic("Aiee, killing interrupt handler!"); 947 if (unlikely(!tsk->pid)) 948 panic("Attempted to kill the idle task!"); 949 950 if (unlikely(irqs_disabled())) { 951 pr_info("note: %s[%d] exited with irqs disabled\n", 952 current->comm, task_pid_nr(current)); 953 local_irq_enable(); 954 } 955 if (unlikely(in_atomic())) { 956 pr_info("note: %s[%d] exited with preempt_count %d\n", 957 current->comm, task_pid_nr(current), 958 preempt_count()); 959 preempt_count_set(PREEMPT_ENABLED); 960 } 961 962 /* 963 * Every time the system oopses, if the oops happens while a reference 964 * to an object was held, the reference leaks. 965 * If the oops doesn't also leak memory, repeated oopsing can cause 966 * reference counters to wrap around (if they're not using refcount_t). 967 * This means that repeated oopsing can make unexploitable-looking bugs 968 * exploitable through repeated oopsing. 969 * To make sure this can't happen, place an upper bound on how often the 970 * kernel may oops without panic(). 971 */ 972 limit = READ_ONCE(oops_limit); 973 if (atomic_inc_return(&oops_count) >= limit && limit) 974 panic("Oopsed too often (kernel.oops_limit is %d)", limit); 975 976 /* 977 * We're taking recursive faults here in make_task_dead. Safest is to just 978 * leave this task alone and wait for reboot. 979 */ 980 if (unlikely(tsk->flags & PF_EXITING)) { 981 pr_alert("Fixing recursive fault but reboot is needed!\n"); 982 futex_exit_recursive(tsk); 983 tsk->exit_state = EXIT_DEAD; 984 refcount_inc(&tsk->rcu_users); 985 do_task_dead(); 986 } 987 988 do_exit(signr); 989} 990 991SYSCALL_DEFINE1(exit, int, error_code) 992{ 993 do_exit((error_code&0xff)<<8); 994} 995 996/* 997 * Take down every thread in the group. This is called by fatal signals 998 * as well as by sys_exit_group (below). 999 */ 1000void __noreturn 1001do_group_exit(int exit_code) 1002{ 1003 struct signal_struct *sig = current->signal; 1004 1005 if (sig->flags & SIGNAL_GROUP_EXIT) 1006 exit_code = sig->group_exit_code; 1007 else if (sig->group_exec_task) 1008 exit_code = 0; 1009 else { 1010 struct sighand_struct *const sighand = current->sighand; 1011 1012 spin_lock_irq(&sighand->siglock); 1013 if (sig->flags & SIGNAL_GROUP_EXIT) 1014 /* Another thread got here before we took the lock. */ 1015 exit_code = sig->group_exit_code; 1016 else if (sig->group_exec_task) 1017 exit_code = 0; 1018 else { 1019 sig->group_exit_code = exit_code; 1020 sig->flags = SIGNAL_GROUP_EXIT; 1021 zap_other_threads(current); 1022 } 1023 spin_unlock_irq(&sighand->siglock); 1024 } 1025 1026 do_exit(exit_code); 1027 /* NOTREACHED */ 1028} 1029 1030/* 1031 * this kills every thread in the thread group. Note that any externally 1032 * wait4()-ing process will get the correct exit code - even if this 1033 * thread is not the thread group leader. 1034 */ 1035SYSCALL_DEFINE1(exit_group, int, error_code) 1036{ 1037 do_group_exit((error_code & 0xff) << 8); 1038 /* NOTREACHED */ 1039 return 0; 1040} 1041 1042static int eligible_pid(struct wait_opts *wo, struct task_struct *p) 1043{ 1044 return wo->wo_type == PIDTYPE_MAX || 1045 task_pid_type(p, wo->wo_type) == wo->wo_pid; 1046} 1047 1048static int 1049eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p) 1050{ 1051 if (!eligible_pid(wo, p)) 1052 return 0; 1053 1054 /* 1055 * Wait for all children (clone and not) if __WALL is set or 1056 * if it is traced by us. 1057 */ 1058 if (ptrace || (wo->wo_flags & __WALL)) 1059 return 1; 1060 1061 /* 1062 * Otherwise, wait for clone children *only* if __WCLONE is set; 1063 * otherwise, wait for non-clone children *only*. 1064 * 1065 * Note: a "clone" child here is one that reports to its parent 1066 * using a signal other than SIGCHLD, or a non-leader thread which 1067 * we can only see if it is traced by us. 1068 */ 1069 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE)) 1070 return 0; 1071 1072 return 1; 1073} 1074 1075/* 1076 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold 1077 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold 1078 * the lock and this task is uninteresting. If we return nonzero, we have 1079 * released the lock and the system call should return. 1080 */ 1081static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p) 1082{ 1083 int state, status; 1084 pid_t pid = task_pid_vnr(p); 1085 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p)); 1086 struct waitid_info *infop; 1087 1088 if (!likely(wo->wo_flags & WEXITED)) 1089 return 0; 1090 1091 if (unlikely(wo->wo_flags & WNOWAIT)) { 1092 status = (p->signal->flags & SIGNAL_GROUP_EXIT) 1093 ? p->signal->group_exit_code : p->exit_code; 1094 get_task_struct(p); 1095 read_unlock(&tasklist_lock); 1096 sched_annotate_sleep(); 1097 if (wo->wo_rusage) 1098 getrusage(p, RUSAGE_BOTH, wo->wo_rusage); 1099 put_task_struct(p); 1100 goto out_info; 1101 } 1102 /* 1103 * Move the task's state to DEAD/TRACE, only one thread can do this. 1104 */ 1105 state = (ptrace_reparented(p) && thread_group_leader(p)) ? 1106 EXIT_TRACE : EXIT_DEAD; 1107 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE) 1108 return 0; 1109 /* 1110 * We own this thread, nobody else can reap it. 1111 */ 1112 read_unlock(&tasklist_lock); 1113 sched_annotate_sleep(); 1114 1115 /* 1116 * Check thread_group_leader() to exclude the traced sub-threads. 1117 */ 1118 if (state == EXIT_DEAD && thread_group_leader(p)) { 1119 struct signal_struct *sig = p->signal; 1120 struct signal_struct *psig = current->signal; 1121 unsigned long maxrss; 1122 u64 tgutime, tgstime; 1123 1124 /* 1125 * The resource counters for the group leader are in its 1126 * own task_struct. Those for dead threads in the group 1127 * are in its signal_struct, as are those for the child 1128 * processes it has previously reaped. All these 1129 * accumulate in the parent's signal_struct c* fields. 1130 * 1131 * We don't bother to take a lock here to protect these 1132 * p->signal fields because the whole thread group is dead 1133 * and nobody can change them. 1134 * 1135 * psig->stats_lock also protects us from our sub-threads 1136 * which can reap other children at the same time. 1137 * 1138 * We use thread_group_cputime_adjusted() to get times for 1139 * the thread group, which consolidates times for all threads 1140 * in the group including the group leader. 1141 */ 1142 thread_group_cputime_adjusted(p, &tgutime, &tgstime); 1143 write_seqlock_irq(&psig->stats_lock); 1144 psig->cutime += tgutime + sig->cutime; 1145 psig->cstime += tgstime + sig->cstime; 1146 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime; 1147 psig->cmin_flt += 1148 p->min_flt + sig->min_flt + sig->cmin_flt; 1149 psig->cmaj_flt += 1150 p->maj_flt + sig->maj_flt + sig->cmaj_flt; 1151 psig->cnvcsw += 1152 p->nvcsw + sig->nvcsw + sig->cnvcsw; 1153 psig->cnivcsw += 1154 p->nivcsw + sig->nivcsw + sig->cnivcsw; 1155 psig->cinblock += 1156 task_io_get_inblock(p) + 1157 sig->inblock + sig->cinblock; 1158 psig->coublock += 1159 task_io_get_oublock(p) + 1160 sig->oublock + sig->coublock; 1161 maxrss = max(sig->maxrss, sig->cmaxrss); 1162 if (psig->cmaxrss < maxrss) 1163 psig->cmaxrss = maxrss; 1164 task_io_accounting_add(&psig->ioac, &p->ioac); 1165 task_io_accounting_add(&psig->ioac, &sig->ioac); 1166 write_sequnlock_irq(&psig->stats_lock); 1167 } 1168 1169 if (wo->wo_rusage) 1170 getrusage(p, RUSAGE_BOTH, wo->wo_rusage); 1171 status = (p->signal->flags & SIGNAL_GROUP_EXIT) 1172 ? p->signal->group_exit_code : p->exit_code; 1173 wo->wo_stat = status; 1174 1175 if (state == EXIT_TRACE) { 1176 write_lock_irq(&tasklist_lock); 1177 /* We dropped tasklist, ptracer could die and untrace */ 1178 ptrace_unlink(p); 1179 1180 /* If parent wants a zombie, don't release it now */ 1181 state = EXIT_ZOMBIE; 1182 if (do_notify_parent(p, p->exit_signal)) 1183 state = EXIT_DEAD; 1184 p->exit_state = state; 1185 write_unlock_irq(&tasklist_lock); 1186 } 1187 if (state == EXIT_DEAD) 1188 release_task(p); 1189 1190out_info: 1191 infop = wo->wo_info; 1192 if (infop) { 1193 if ((status & 0x7f) == 0) { 1194 infop->cause = CLD_EXITED; 1195 infop->status = status >> 8; 1196 } else { 1197 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED; 1198 infop->status = status & 0x7f; 1199 } 1200 infop->pid = pid; 1201 infop->uid = uid; 1202 } 1203 1204 return pid; 1205} 1206 1207static int *task_stopped_code(struct task_struct *p, bool ptrace) 1208{ 1209 if (ptrace) { 1210 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING)) 1211 return &p->exit_code; 1212 } else { 1213 if (p->signal->flags & SIGNAL_STOP_STOPPED) 1214 return &p->signal->group_exit_code; 1215 } 1216 return NULL; 1217} 1218 1219/** 1220 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED 1221 * @wo: wait options 1222 * @ptrace: is the wait for ptrace 1223 * @p: task to wait for 1224 * 1225 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED. 1226 * 1227 * CONTEXT: 1228 * read_lock(&tasklist_lock), which is released if return value is 1229 * non-zero. Also, grabs and releases @p->sighand->siglock. 1230 * 1231 * RETURNS: 1232 * 0 if wait condition didn't exist and search for other wait conditions 1233 * should continue. Non-zero return, -errno on failure and @p's pid on 1234 * success, implies that tasklist_lock is released and wait condition 1235 * search should terminate. 1236 */ 1237static int wait_task_stopped(struct wait_opts *wo, 1238 int ptrace, struct task_struct *p) 1239{ 1240 struct waitid_info *infop; 1241 int exit_code, *p_code, why; 1242 uid_t uid = 0; /* unneeded, required by compiler */ 1243 pid_t pid; 1244 1245 /* 1246 * Traditionally we see ptrace'd stopped tasks regardless of options. 1247 */ 1248 if (!ptrace && !(wo->wo_flags & WUNTRACED)) 1249 return 0; 1250 1251 if (!task_stopped_code(p, ptrace)) 1252 return 0; 1253 1254 exit_code = 0; 1255 spin_lock_irq(&p->sighand->siglock); 1256 1257 p_code = task_stopped_code(p, ptrace); 1258 if (unlikely(!p_code)) 1259 goto unlock_sig; 1260 1261 exit_code = *p_code; 1262 if (!exit_code) 1263 goto unlock_sig; 1264 1265 if (!unlikely(wo->wo_flags & WNOWAIT)) 1266 *p_code = 0; 1267 1268 uid = from_kuid_munged(current_user_ns(), task_uid(p)); 1269unlock_sig: 1270 spin_unlock_irq(&p->sighand->siglock); 1271 if (!exit_code) 1272 return 0; 1273 1274 /* 1275 * Now we are pretty sure this task is interesting. 1276 * Make sure it doesn't get reaped out from under us while we 1277 * give up the lock and then examine it below. We don't want to 1278 * keep holding onto the tasklist_lock while we call getrusage and 1279 * possibly take page faults for user memory. 1280 */ 1281 get_task_struct(p); 1282 pid = task_pid_vnr(p); 1283 why = ptrace ? CLD_TRAPPED : CLD_STOPPED; 1284 read_unlock(&tasklist_lock); 1285 sched_annotate_sleep(); 1286 if (wo->wo_rusage) 1287 getrusage(p, RUSAGE_BOTH, wo->wo_rusage); 1288 put_task_struct(p); 1289 1290 if (likely(!(wo->wo_flags & WNOWAIT))) 1291 wo->wo_stat = (exit_code << 8) | 0x7f; 1292 1293 infop = wo->wo_info; 1294 if (infop) { 1295 infop->cause = why; 1296 infop->status = exit_code; 1297 infop->pid = pid; 1298 infop->uid = uid; 1299 } 1300 return pid; 1301} 1302 1303/* 1304 * Handle do_wait work for one task in a live, non-stopped state. 1305 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold 1306 * the lock and this task is uninteresting. If we return nonzero, we have 1307 * released the lock and the system call should return. 1308 */ 1309static int wait_task_continued(struct wait_opts *wo, struct task_struct *p) 1310{ 1311 struct waitid_info *infop; 1312 pid_t pid; 1313 uid_t uid; 1314 1315 if (!unlikely(wo->wo_flags & WCONTINUED)) 1316 return 0; 1317 1318 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) 1319 return 0; 1320 1321 spin_lock_irq(&p->sighand->siglock); 1322 /* Re-check with the lock held. */ 1323 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) { 1324 spin_unlock_irq(&p->sighand->siglock); 1325 return 0; 1326 } 1327 if (!unlikely(wo->wo_flags & WNOWAIT)) 1328 p->signal->flags &= ~SIGNAL_STOP_CONTINUED; 1329 uid = from_kuid_munged(current_user_ns(), task_uid(p)); 1330 spin_unlock_irq(&p->sighand->siglock); 1331 1332 pid = task_pid_vnr(p); 1333 get_task_struct(p); 1334 read_unlock(&tasklist_lock); 1335 sched_annotate_sleep(); 1336 if (wo->wo_rusage) 1337 getrusage(p, RUSAGE_BOTH, wo->wo_rusage); 1338 put_task_struct(p); 1339 1340 infop = wo->wo_info; 1341 if (!infop) { 1342 wo->wo_stat = 0xffff; 1343 } else { 1344 infop->cause = CLD_CONTINUED; 1345 infop->pid = pid; 1346 infop->uid = uid; 1347 infop->status = SIGCONT; 1348 } 1349 return pid; 1350} 1351 1352/* 1353 * Consider @p for a wait by @parent. 1354 * 1355 * -ECHILD should be in ->notask_error before the first call. 1356 * Returns nonzero for a final return, when we have unlocked tasklist_lock. 1357 * Returns zero if the search for a child should continue; 1358 * then ->notask_error is 0 if @p is an eligible child, 1359 * or still -ECHILD. 1360 */ 1361static int wait_consider_task(struct wait_opts *wo, int ptrace, 1362 struct task_struct *p) 1363{ 1364 /* 1365 * We can race with wait_task_zombie() from another thread. 1366 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition 1367 * can't confuse the checks below. 1368 */ 1369 int exit_state = READ_ONCE(p->exit_state); 1370 int ret; 1371 1372 if (unlikely(exit_state == EXIT_DEAD)) 1373 return 0; 1374 1375 ret = eligible_child(wo, ptrace, p); 1376 if (!ret) 1377 return ret; 1378 1379 if (unlikely(exit_state == EXIT_TRACE)) { 1380 /* 1381 * ptrace == 0 means we are the natural parent. In this case 1382 * we should clear notask_error, debugger will notify us. 1383 */ 1384 if (likely(!ptrace)) 1385 wo->notask_error = 0; 1386 return 0; 1387 } 1388 1389 if (likely(!ptrace) && unlikely(p->ptrace)) { 1390 /* 1391 * If it is traced by its real parent's group, just pretend 1392 * the caller is ptrace_do_wait() and reap this child if it 1393 * is zombie. 1394 * 1395 * This also hides group stop state from real parent; otherwise 1396 * a single stop can be reported twice as group and ptrace stop. 1397 * If a ptracer wants to distinguish these two events for its 1398 * own children it should create a separate process which takes 1399 * the role of real parent. 1400 */ 1401 if (!ptrace_reparented(p)) 1402 ptrace = 1; 1403 } 1404 1405 /* slay zombie? */ 1406 if (exit_state == EXIT_ZOMBIE) { 1407 /* we don't reap group leaders with subthreads */ 1408 if (!delay_group_leader(p)) { 1409 /* 1410 * A zombie ptracee is only visible to its ptracer. 1411 * Notification and reaping will be cascaded to the 1412 * real parent when the ptracer detaches. 1413 */ 1414 if (unlikely(ptrace) || likely(!p->ptrace)) 1415 return wait_task_zombie(wo, p); 1416 } 1417 1418 /* 1419 * Allow access to stopped/continued state via zombie by 1420 * falling through. Clearing of notask_error is complex. 1421 * 1422 * When !@ptrace: 1423 * 1424 * If WEXITED is set, notask_error should naturally be 1425 * cleared. If not, subset of WSTOPPED|WCONTINUED is set, 1426 * so, if there are live subthreads, there are events to 1427 * wait for. If all subthreads are dead, it's still safe 1428 * to clear - this function will be called again in finite 1429 * amount time once all the subthreads are released and 1430 * will then return without clearing. 1431 * 1432 * When @ptrace: 1433 * 1434 * Stopped state is per-task and thus can't change once the 1435 * target task dies. Only continued and exited can happen. 1436 * Clear notask_error if WCONTINUED | WEXITED. 1437 */ 1438 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED))) 1439 wo->notask_error = 0; 1440 } else { 1441 /* 1442 * @p is alive and it's gonna stop, continue or exit, so 1443 * there always is something to wait for. 1444 */ 1445 wo->notask_error = 0; 1446 } 1447 1448 /* 1449 * Wait for stopped. Depending on @ptrace, different stopped state 1450 * is used and the two don't interact with each other. 1451 */ 1452 ret = wait_task_stopped(wo, ptrace, p); 1453 if (ret) 1454 return ret; 1455 1456 /* 1457 * Wait for continued. There's only one continued state and the 1458 * ptracer can consume it which can confuse the real parent. Don't 1459 * use WCONTINUED from ptracer. You don't need or want it. 1460 */ 1461 return wait_task_continued(wo, p); 1462} 1463 1464/* 1465 * Do the work of do_wait() for one thread in the group, @tsk. 1466 * 1467 * -ECHILD should be in ->notask_error before the first call. 1468 * Returns nonzero for a final return, when we have unlocked tasklist_lock. 1469 * Returns zero if the search for a child should continue; then 1470 * ->notask_error is 0 if there were any eligible children, 1471 * or still -ECHILD. 1472 */ 1473static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk) 1474{ 1475 struct task_struct *p; 1476 1477 list_for_each_entry(p, &tsk->children, sibling) { 1478 int ret = wait_consider_task(wo, 0, p); 1479 1480 if (ret) 1481 return ret; 1482 } 1483 1484 return 0; 1485} 1486 1487static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk) 1488{ 1489 struct task_struct *p; 1490 1491 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) { 1492 int ret = wait_consider_task(wo, 1, p); 1493 1494 if (ret) 1495 return ret; 1496 } 1497 1498 return 0; 1499} 1500 1501bool pid_child_should_wake(struct wait_opts *wo, struct task_struct *p) 1502{ 1503 if (!eligible_pid(wo, p)) 1504 return false; 1505 1506 if ((wo->wo_flags & __WNOTHREAD) && wo->child_wait.private != p->parent) 1507 return false; 1508 1509 return true; 1510} 1511 1512static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode, 1513 int sync, void *key) 1514{ 1515 struct wait_opts *wo = container_of(wait, struct wait_opts, 1516 child_wait); 1517 struct task_struct *p = key; 1518 1519 if (pid_child_should_wake(wo, p)) 1520 return default_wake_function(wait, mode, sync, key); 1521 1522 return 0; 1523} 1524 1525void __wake_up_parent(struct task_struct *p, struct task_struct *parent) 1526{ 1527 __wake_up_sync_key(&parent->signal->wait_chldexit, 1528 TASK_INTERRUPTIBLE, p); 1529} 1530 1531static bool is_effectively_child(struct wait_opts *wo, bool ptrace, 1532 struct task_struct *target) 1533{ 1534 struct task_struct *parent = 1535 !ptrace ? target->real_parent : target->parent; 1536 1537 return current == parent || (!(wo->wo_flags & __WNOTHREAD) && 1538 same_thread_group(current, parent)); 1539} 1540 1541/* 1542 * Optimization for waiting on PIDTYPE_PID. No need to iterate through child 1543 * and tracee lists to find the target task. 1544 */ 1545static int do_wait_pid(struct wait_opts *wo) 1546{ 1547 bool ptrace; 1548 struct task_struct *target; 1549 int retval; 1550 1551 ptrace = false; 1552 target = pid_task(wo->wo_pid, PIDTYPE_TGID); 1553 if (target && is_effectively_child(wo, ptrace, target)) { 1554 retval = wait_consider_task(wo, ptrace, target); 1555 if (retval) 1556 return retval; 1557 } 1558 1559 ptrace = true; 1560 target = pid_task(wo->wo_pid, PIDTYPE_PID); 1561 if (target && target->ptrace && 1562 is_effectively_child(wo, ptrace, target)) { 1563 retval = wait_consider_task(wo, ptrace, target); 1564 if (retval) 1565 return retval; 1566 } 1567 1568 return 0; 1569} 1570 1571long __do_wait(struct wait_opts *wo) 1572{ 1573 long retval; 1574 1575 /* 1576 * If there is nothing that can match our criteria, just get out. 1577 * We will clear ->notask_error to zero if we see any child that 1578 * might later match our criteria, even if we are not able to reap 1579 * it yet. 1580 */ 1581 wo->notask_error = -ECHILD; 1582 if ((wo->wo_type < PIDTYPE_MAX) && 1583 (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type))) 1584 goto notask; 1585 1586 read_lock(&tasklist_lock); 1587 1588 if (wo->wo_type == PIDTYPE_PID) { 1589 retval = do_wait_pid(wo); 1590 if (retval) 1591 return retval; 1592 } else { 1593 struct task_struct *tsk = current; 1594 1595 do { 1596 retval = do_wait_thread(wo, tsk); 1597 if (retval) 1598 return retval; 1599 1600 retval = ptrace_do_wait(wo, tsk); 1601 if (retval) 1602 return retval; 1603 1604 if (wo->wo_flags & __WNOTHREAD) 1605 break; 1606 } while_each_thread(current, tsk); 1607 } 1608 read_unlock(&tasklist_lock); 1609 1610notask: 1611 retval = wo->notask_error; 1612 if (!retval && !(wo->wo_flags & WNOHANG)) 1613 return -ERESTARTSYS; 1614 1615 return retval; 1616} 1617 1618static long do_wait(struct wait_opts *wo) 1619{ 1620 int retval; 1621 1622 trace_sched_process_wait(wo->wo_pid); 1623 1624 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback); 1625 wo->child_wait.private = current; 1626 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait); 1627 1628 do { 1629 set_current_state(TASK_INTERRUPTIBLE); 1630 retval = __do_wait(wo); 1631 if (retval != -ERESTARTSYS) 1632 break; 1633 if (signal_pending(current)) 1634 break; 1635 schedule(); 1636 } while (1); 1637 1638 __set_current_state(TASK_RUNNING); 1639 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait); 1640 return retval; 1641} 1642 1643int kernel_waitid_prepare(struct wait_opts *wo, int which, pid_t upid, 1644 struct waitid_info *infop, int options, 1645 struct rusage *ru) 1646{ 1647 unsigned int f_flags = 0; 1648 struct pid *pid = NULL; 1649 enum pid_type type; 1650 1651 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED| 1652 __WNOTHREAD|__WCLONE|__WALL)) 1653 return -EINVAL; 1654 if (!(options & (WEXITED|WSTOPPED|WCONTINUED))) 1655 return -EINVAL; 1656 1657 switch (which) { 1658 case P_ALL: 1659 type = PIDTYPE_MAX; 1660 break; 1661 case P_PID: 1662 type = PIDTYPE_PID; 1663 if (upid <= 0) 1664 return -EINVAL; 1665 1666 pid = find_get_pid(upid); 1667 break; 1668 case P_PGID: 1669 type = PIDTYPE_PGID; 1670 if (upid < 0) 1671 return -EINVAL; 1672 1673 if (upid) 1674 pid = find_get_pid(upid); 1675 else 1676 pid = get_task_pid(current, PIDTYPE_PGID); 1677 break; 1678 case P_PIDFD: 1679 type = PIDTYPE_PID; 1680 if (upid < 0) 1681 return -EINVAL; 1682 1683 pid = pidfd_get_pid(upid, &f_flags); 1684 if (IS_ERR(pid)) 1685 return PTR_ERR(pid); 1686 1687 break; 1688 default: 1689 return -EINVAL; 1690 } 1691 1692 wo->wo_type = type; 1693 wo->wo_pid = pid; 1694 wo->wo_flags = options; 1695 wo->wo_info = infop; 1696 wo->wo_rusage = ru; 1697 if (f_flags & O_NONBLOCK) 1698 wo->wo_flags |= WNOHANG; 1699 1700 return 0; 1701} 1702 1703static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop, 1704 int options, struct rusage *ru) 1705{ 1706 struct wait_opts wo; 1707 long ret; 1708 1709 ret = kernel_waitid_prepare(&wo, which, upid, infop, options, ru); 1710 if (ret) 1711 return ret; 1712 1713 ret = do_wait(&wo); 1714 if (!ret && !(options & WNOHANG) && (wo.wo_flags & WNOHANG)) 1715 ret = -EAGAIN; 1716 1717 put_pid(wo.wo_pid); 1718 return ret; 1719} 1720 1721SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *, 1722 infop, int, options, struct rusage __user *, ru) 1723{ 1724 struct rusage r; 1725 struct waitid_info info = {.status = 0}; 1726 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL); 1727 int signo = 0; 1728 1729 if (err > 0) { 1730 signo = SIGCHLD; 1731 err = 0; 1732 if (ru && copy_to_user(ru, &r, sizeof(struct rusage))) 1733 return -EFAULT; 1734 } 1735 if (!infop) 1736 return err; 1737 1738 if (!user_write_access_begin(infop, sizeof(*infop))) 1739 return -EFAULT; 1740 1741 unsafe_put_user(signo, &infop->si_signo, Efault); 1742 unsafe_put_user(0, &infop->si_errno, Efault); 1743 unsafe_put_user(info.cause, &infop->si_code, Efault); 1744 unsafe_put_user(info.pid, &infop->si_pid, Efault); 1745 unsafe_put_user(info.uid, &infop->si_uid, Efault); 1746 unsafe_put_user(info.status, &infop->si_status, Efault); 1747 user_write_access_end(); 1748 return err; 1749Efault: 1750 user_write_access_end(); 1751 return -EFAULT; 1752} 1753 1754long kernel_wait4(pid_t upid, int __user *stat_addr, int options, 1755 struct rusage *ru) 1756{ 1757 struct wait_opts wo; 1758 struct pid *pid = NULL; 1759 enum pid_type type; 1760 long ret; 1761 1762 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED| 1763 __WNOTHREAD|__WCLONE|__WALL)) 1764 return -EINVAL; 1765 1766 /* -INT_MIN is not defined */ 1767 if (upid == INT_MIN) 1768 return -ESRCH; 1769 1770 if (upid == -1) 1771 type = PIDTYPE_MAX; 1772 else if (upid < 0) { 1773 type = PIDTYPE_PGID; 1774 pid = find_get_pid(-upid); 1775 } else if (upid == 0) { 1776 type = PIDTYPE_PGID; 1777 pid = get_task_pid(current, PIDTYPE_PGID); 1778 } else /* upid > 0 */ { 1779 type = PIDTYPE_PID; 1780 pid = find_get_pid(upid); 1781 } 1782 1783 wo.wo_type = type; 1784 wo.wo_pid = pid; 1785 wo.wo_flags = options | WEXITED; 1786 wo.wo_info = NULL; 1787 wo.wo_stat = 0; 1788 wo.wo_rusage = ru; 1789 ret = do_wait(&wo); 1790 put_pid(pid); 1791 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr)) 1792 ret = -EFAULT; 1793 1794 return ret; 1795} 1796 1797int kernel_wait(pid_t pid, int *stat) 1798{ 1799 struct wait_opts wo = { 1800 .wo_type = PIDTYPE_PID, 1801 .wo_pid = find_get_pid(pid), 1802 .wo_flags = WEXITED, 1803 }; 1804 int ret; 1805 1806 ret = do_wait(&wo); 1807 if (ret > 0 && wo.wo_stat) 1808 *stat = wo.wo_stat; 1809 put_pid(wo.wo_pid); 1810 return ret; 1811} 1812 1813SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr, 1814 int, options, struct rusage __user *, ru) 1815{ 1816 struct rusage r; 1817 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL); 1818 1819 if (err > 0) { 1820 if (ru && copy_to_user(ru, &r, sizeof(struct rusage))) 1821 return -EFAULT; 1822 } 1823 return err; 1824} 1825 1826#ifdef __ARCH_WANT_SYS_WAITPID 1827 1828/* 1829 * sys_waitpid() remains for compatibility. waitpid() should be 1830 * implemented by calling sys_wait4() from libc.a. 1831 */ 1832SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options) 1833{ 1834 return kernel_wait4(pid, stat_addr, options, NULL); 1835} 1836 1837#endif 1838 1839#ifdef CONFIG_COMPAT 1840COMPAT_SYSCALL_DEFINE4(wait4, 1841 compat_pid_t, pid, 1842 compat_uint_t __user *, stat_addr, 1843 int, options, 1844 struct compat_rusage __user *, ru) 1845{ 1846 struct rusage r; 1847 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL); 1848 if (err > 0) { 1849 if (ru && put_compat_rusage(&r, ru)) 1850 return -EFAULT; 1851 } 1852 return err; 1853} 1854 1855COMPAT_SYSCALL_DEFINE5(waitid, 1856 int, which, compat_pid_t, pid, 1857 struct compat_siginfo __user *, infop, int, options, 1858 struct compat_rusage __user *, uru) 1859{ 1860 struct rusage ru; 1861 struct waitid_info info = {.status = 0}; 1862 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL); 1863 int signo = 0; 1864 if (err > 0) { 1865 signo = SIGCHLD; 1866 err = 0; 1867 if (uru) { 1868 /* kernel_waitid() overwrites everything in ru */ 1869 if (COMPAT_USE_64BIT_TIME) 1870 err = copy_to_user(uru, &ru, sizeof(ru)); 1871 else 1872 err = put_compat_rusage(&ru, uru); 1873 if (err) 1874 return -EFAULT; 1875 } 1876 } 1877 1878 if (!infop) 1879 return err; 1880 1881 if (!user_write_access_begin(infop, sizeof(*infop))) 1882 return -EFAULT; 1883 1884 unsafe_put_user(signo, &infop->si_signo, Efault); 1885 unsafe_put_user(0, &infop->si_errno, Efault); 1886 unsafe_put_user(info.cause, &infop->si_code, Efault); 1887 unsafe_put_user(info.pid, &infop->si_pid, Efault); 1888 unsafe_put_user(info.uid, &infop->si_uid, Efault); 1889 unsafe_put_user(info.status, &infop->si_status, Efault); 1890 user_write_access_end(); 1891 return err; 1892Efault: 1893 user_write_access_end(); 1894 return -EFAULT; 1895} 1896#endif 1897 1898/* 1899 * This needs to be __function_aligned as GCC implicitly makes any 1900 * implementation of abort() cold and drops alignment specified by 1901 * -falign-functions=N. 1902 * 1903 * See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88345#c11 1904 */ 1905__weak __function_aligned void abort(void) 1906{ 1907 BUG(); 1908 1909 /* if that doesn't kill us, halt */ 1910 panic("Oops failed to kill thread"); 1911} 1912EXPORT_SYMBOL(abort); 1913