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