fork.c

   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *  linux/kernel/fork.c
   4 *
   5 *  Copyright (C) 1991, 1992  Linus Torvalds
   6 */
   7
   8/*
   9 *  'fork.c' contains the help-routines for the 'fork' system call
  10 * (see also entry.S and others).
  11 * Fork is rather simple, once you get the hang of it, but the memory
  12 * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  13 */
  14
  15#include <linux/anon_inodes.h>
  16#include <linux/slab.h>
  17#include <linux/sched/autogroup.h>
  18#include <linux/sched/mm.h>
  19#include <linux/sched/user.h>
  20#include <linux/sched/numa_balancing.h>
  21#include <linux/sched/stat.h>
  22#include <linux/sched/task.h>
  23#include <linux/sched/task_stack.h>
  24#include <linux/sched/cputime.h>
  25#include <linux/sched/ext.h>
  26#include <linux/seq_file.h>
  27#include <linux/rtmutex.h>
  28#include <linux/init.h>
  29#include <linux/unistd.h>
  30#include <linux/module.h>
  31#include <linux/vmalloc.h>
  32#include <linux/completion.h>
  33#include <linux/personality.h>
  34#include <linux/mempolicy.h>
  35#include <linux/sem.h>
  36#include <linux/file.h>
  37#include <linux/fdtable.h>
  38#include <linux/iocontext.h>
  39#include <linux/key.h>
  40#include <linux/kmsan.h>
  41#include <linux/binfmts.h>
  42#include <linux/mman.h>
  43#include <linux/mmu_notifier.h>
  44#include <linux/fs.h>
  45#include <linux/mm.h>
  46#include <linux/mm_inline.h>
  47#include <linux/memblock.h>
  48#include <linux/nsproxy.h>
  49#include <linux/capability.h>
  50#include <linux/cpu.h>
  51#include <linux/cgroup.h>
  52#include <linux/security.h>
  53#include <linux/hugetlb.h>
  54#include <linux/seccomp.h>
  55#include <linux/swap.h>
  56#include <linux/syscalls.h>
  57#include <linux/syscall_user_dispatch.h>
  58#include <linux/jiffies.h>
  59#include <linux/futex.h>
  60#include <linux/compat.h>
  61#include <linux/kthread.h>
  62#include <linux/task_io_accounting_ops.h>
  63#include <linux/rcupdate.h>
  64#include <linux/ptrace.h>
  65#include <linux/mount.h>
  66#include <linux/audit.h>
  67#include <linux/memcontrol.h>
  68#include <linux/ftrace.h>
  69#include <linux/proc_fs.h>
  70#include <linux/profile.h>
  71#include <linux/rmap.h>
  72#include <linux/ksm.h>
  73#include <linux/acct.h>
  74#include <linux/userfaultfd_k.h>
  75#include <linux/tsacct_kern.h>
  76#include <linux/cn_proc.h>
  77#include <linux/freezer.h>
  78#include <linux/delayacct.h>
  79#include <linux/taskstats_kern.h>
  80#include <linux/tty.h>
  81#include <linux/fs_struct.h>
  82#include <linux/magic.h>
  83#include <linux/perf_event.h>
  84#include <linux/posix-timers.h>
  85#include <linux/user-return-notifier.h>
  86#include <linux/oom.h>
  87#include <linux/khugepaged.h>
  88#include <linux/signalfd.h>
  89#include <linux/uprobes.h>
  90#include <linux/aio.h>
  91#include <linux/compiler.h>
  92#include <linux/sysctl.h>
  93#include <linux/kcov.h>
  94#include <linux/livepatch.h>
  95#include <linux/thread_info.h>
  96#include <linux/kstack_erase.h>
  97#include <linux/kasan.h>
  98#include <linux/scs.h>
  99#include <linux/io_uring.h>
 100#include <linux/bpf.h>
 101#include <linux/stackprotector.h>
 102#include <linux/user_events.h>
 103#include <linux/iommu.h>
 104#include <linux/rseq.h>
 105#include <uapi/linux/pidfd.h>
 106#include <linux/pidfs.h>
 107#include <linux/tick.h>
 108#include <linux/unwind_deferred.h>
 109#include <linux/pgalloc.h>
 110#include <linux/uaccess.h>
 111
 112#include <asm/mmu_context.h>
 113#include <asm/cacheflush.h>
 114#include <asm/tlbflush.h>
 115
 116/* For dup_mmap(). */
 117#include "../mm/internal.h"
 118
 119#include <trace/events/sched.h>
 120
 121#define CREATE_TRACE_POINTS
 122#include <trace/events/task.h>
 123
 124#include <kunit/visibility.h>
 125
 126/*
 127 * Minimum number of threads to boot the kernel
 128 */
 129#define MIN_THREADS 20
 130
 131/*
 132 * Maximum number of threads
 133 */
 134#define MAX_THREADS FUTEX_TID_MASK
 135
 136/*
 137 * Protected counters by write_lock_irq(&tasklist_lock)
 138 */
 139unsigned long total_forks;	/* Handle normal Linux uptimes. */
 140int nr_threads;			/* The idle threads do not count.. */
 141
 142static int max_threads;		/* tunable limit on nr_threads */
 143
 144#define NAMED_ARRAY_INDEX(x)	[x] = __stringify(x)
 145
 146static const char * const resident_page_types[] = {
 147	NAMED_ARRAY_INDEX(MM_FILEPAGES),
 148	NAMED_ARRAY_INDEX(MM_ANONPAGES),
 149	NAMED_ARRAY_INDEX(MM_SWAPENTS),
 150	NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
 151};
 152
 153DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 154
 155__cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 156
 157#ifdef CONFIG_PROVE_RCU
 158int lockdep_tasklist_lock_is_held(void)
 159{
 160	return lockdep_is_held(&tasklist_lock);
 161}
 162EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 163#endif /* #ifdef CONFIG_PROVE_RCU */
 164
 165int nr_processes(void)
 166{
 167	int cpu;
 168	int total = 0;
 169
 170	for_each_possible_cpu(cpu)
 171		total += per_cpu(process_counts, cpu);
 172
 173	return total;
 174}
 175
 176void __weak arch_release_task_struct(struct task_struct *tsk)
 177{
 178}
 179
 180static struct kmem_cache *task_struct_cachep;
 181
 182static inline struct task_struct *alloc_task_struct_node(int node)
 183{
 184	return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 185}
 186
 187static inline void free_task_struct(struct task_struct *tsk)
 188{
 189	kmem_cache_free(task_struct_cachep, tsk);
 190}
 191
 192#ifdef CONFIG_VMAP_STACK
 193/*
 194 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 195 * flush.  Try to minimize the number of calls by caching stacks.
 196 */
 197#define NR_CACHED_STACKS 2
 198static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 199/*
 200 * Allocated stacks are cached and later reused by new threads, so memcg
 201 * accounting is performed by the code assigning/releasing stacks to tasks.
 202 * We need a zeroed memory without __GFP_ACCOUNT.
 203 */
 204#define GFP_VMAP_STACK (GFP_KERNEL | __GFP_ZERO)
 205
 206struct vm_stack {
 207	struct rcu_head rcu;
 208	struct vm_struct *stack_vm_area;
 209};
 210
 211static struct vm_struct *alloc_thread_stack_node_from_cache(struct task_struct *tsk, int node)
 212{
 213	struct vm_struct *vm_area;
 214	unsigned int i;
 215
 216	/*
 217	 * If the node has memory, we are guaranteed the stacks are backed by local pages.
 218	 * Otherwise the pages are arbitrary.
 219	 *
 220	 * Note that depending on cpuset it is possible we will get migrated to a different
 221	 * node immediately after allocating here, so this does *not* guarantee locality for
 222	 * arbitrary callers.
 223	 */
 224	scoped_guard(preempt) {
 225		if (node != NUMA_NO_NODE && numa_node_id() != node)
 226			return NULL;
 227
 228		for (i = 0; i < NR_CACHED_STACKS; i++) {
 229			vm_area = this_cpu_xchg(cached_stacks[i], NULL);
 230			if (vm_area)
 231				return vm_area;
 232		}
 233	}
 234
 235	return NULL;
 236}
 237
 238static bool try_release_thread_stack_to_cache(struct vm_struct *vm_area)
 239{
 240	unsigned int i;
 241	int nid;
 242
 243	/*
 244	 * Don't cache stacks if any of the pages don't match the local domain, unless
 245	 * there is no local memory to begin with.
 246	 *
 247	 * Note that lack of local memory does not automatically mean it makes no difference
 248	 * performance-wise which other domain backs the stack. In this case we are merely
 249	 * trying to avoid constantly going to vmalloc.
 250	 */
 251	scoped_guard(preempt) {
 252		nid = numa_node_id();
 253		if (node_state(nid, N_MEMORY)) {
 254			for (i = 0; i < vm_area->nr_pages; i++) {
 255				struct page *page = vm_area->pages[i];
 256				if (page_to_nid(page) != nid)
 257					return false;
 258			}
 259		}
 260
 261		for (i = 0; i < NR_CACHED_STACKS; i++) {
 262			struct vm_struct *tmp = NULL;
 263
 264			if (this_cpu_try_cmpxchg(cached_stacks[i], &tmp, vm_area))
 265				return true;
 266		}
 267	}
 268	return false;
 269}
 270
 271static void thread_stack_free_rcu(struct rcu_head *rh)
 272{
 273	struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
 274	struct vm_struct *vm_area = vm_stack->stack_vm_area;
 275
 276	if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
 277		return;
 278
 279	vfree(vm_area->addr);
 280}
 281
 282static void thread_stack_delayed_free(struct task_struct *tsk)
 283{
 284	struct vm_stack *vm_stack = tsk->stack;
 285
 286	vm_stack->stack_vm_area = tsk->stack_vm_area;
 287	call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
 288}
 289
 290static int free_vm_stack_cache(unsigned int cpu)
 291{
 292	struct vm_struct **cached_vm_stack_areas = per_cpu_ptr(cached_stacks, cpu);
 293	int i;
 294
 295	for (i = 0; i < NR_CACHED_STACKS; i++) {
 296		struct vm_struct *vm_area = cached_vm_stack_areas[i];
 297
 298		if (!vm_area)
 299			continue;
 300
 301		vfree(vm_area->addr);
 302		cached_vm_stack_areas[i] = NULL;
 303	}
 304
 305	return 0;
 306}
 307
 308static int memcg_charge_kernel_stack(struct vm_struct *vm_area)
 309{
 310	int i;
 311	int ret;
 312	int nr_charged = 0;
 313
 314	BUG_ON(vm_area->nr_pages != THREAD_SIZE / PAGE_SIZE);
 315
 316	for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 317		ret = memcg_kmem_charge_page(vm_area->pages[i], GFP_KERNEL, 0);
 318		if (ret)
 319			goto err;
 320		nr_charged++;
 321	}
 322	return 0;
 323err:
 324	for (i = 0; i < nr_charged; i++)
 325		memcg_kmem_uncharge_page(vm_area->pages[i], 0);
 326	return ret;
 327}
 328
 329static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 330{
 331	struct vm_struct *vm_area;
 332	void *stack;
 333
 334	vm_area = alloc_thread_stack_node_from_cache(tsk, node);
 335	if (vm_area) {
 336		if (memcg_charge_kernel_stack(vm_area)) {
 337			vfree(vm_area->addr);
 338			return -ENOMEM;
 339		}
 340
 341		/* Reset stack metadata. */
 342		kasan_unpoison_range(vm_area->addr, THREAD_SIZE);
 343
 344		stack = kasan_reset_tag(vm_area->addr);
 345
 346		/* Clear stale pointers from reused stack. */
 347		memset(stack, 0, THREAD_SIZE);
 348
 349		tsk->stack_vm_area = vm_area;
 350		tsk->stack = stack;
 351		return 0;
 352	}
 353
 354	stack = __vmalloc_node(THREAD_SIZE, THREAD_ALIGN,
 355				     GFP_VMAP_STACK,
 356				     node, __builtin_return_address(0));
 357	if (!stack)
 358		return -ENOMEM;
 359
 360	vm_area = find_vm_area(stack);
 361	if (memcg_charge_kernel_stack(vm_area)) {
 362		vfree(stack);
 363		return -ENOMEM;
 364	}
 365	/*
 366	 * We can't call find_vm_area() in interrupt context, and
 367	 * free_thread_stack() can be called in interrupt context,
 368	 * so cache the vm_struct.
 369	 */
 370	tsk->stack_vm_area = vm_area;
 371	stack = kasan_reset_tag(stack);
 372	tsk->stack = stack;
 373	return 0;
 374}
 375
 376static void free_thread_stack(struct task_struct *tsk)
 377{
 378	if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
 379		thread_stack_delayed_free(tsk);
 380
 381	tsk->stack = NULL;
 382	tsk->stack_vm_area = NULL;
 383}
 384
 385#else /* !CONFIG_VMAP_STACK */
 386
 387/*
 388 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 389 * kmemcache based allocator.
 390 */
 391#if THREAD_SIZE >= PAGE_SIZE
 392
 393static void thread_stack_free_rcu(struct rcu_head *rh)
 394{
 395	__free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
 396}
 397
 398static void thread_stack_delayed_free(struct task_struct *tsk)
 399{
 400	struct rcu_head *rh = tsk->stack;
 401
 402	call_rcu(rh, thread_stack_free_rcu);
 403}
 404
 405static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 406{
 407	struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 408					     THREAD_SIZE_ORDER);
 409
 410	if (likely(page)) {
 411		tsk->stack = kasan_reset_tag(page_address(page));
 412		return 0;
 413	}
 414	return -ENOMEM;
 415}
 416
 417static void free_thread_stack(struct task_struct *tsk)
 418{
 419	thread_stack_delayed_free(tsk);
 420	tsk->stack = NULL;
 421}
 422
 423#else /* !(THREAD_SIZE >= PAGE_SIZE) */
 424
 425static struct kmem_cache *thread_stack_cache;
 426
 427static void thread_stack_free_rcu(struct rcu_head *rh)
 428{
 429	kmem_cache_free(thread_stack_cache, rh);
 430}
 431
 432static void thread_stack_delayed_free(struct task_struct *tsk)
 433{
 434	struct rcu_head *rh = tsk->stack;
 435
 436	call_rcu(rh, thread_stack_free_rcu);
 437}
 438
 439static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 440{
 441	unsigned long *stack;
 442	stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 443	stack = kasan_reset_tag(stack);
 444	tsk->stack = stack;
 445	return stack ? 0 : -ENOMEM;
 446}
 447
 448static void free_thread_stack(struct task_struct *tsk)
 449{
 450	thread_stack_delayed_free(tsk);
 451	tsk->stack = NULL;
 452}
 453
 454void thread_stack_cache_init(void)
 455{
 456	thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 457					THREAD_SIZE, THREAD_SIZE, 0, 0,
 458					THREAD_SIZE, NULL);
 459	BUG_ON(thread_stack_cache == NULL);
 460}
 461
 462#endif /* THREAD_SIZE >= PAGE_SIZE */
 463#endif /* CONFIG_VMAP_STACK */
 464
 465/* SLAB cache for signal_struct structures (tsk->signal) */
 466static struct kmem_cache *signal_cachep;
 467
 468/* SLAB cache for sighand_struct structures (tsk->sighand) */
 469struct kmem_cache *sighand_cachep;
 470
 471/* SLAB cache for files_struct structures (tsk->files) */
 472struct kmem_cache *files_cachep;
 473
 474/* SLAB cache for fs_struct structures (tsk->fs) */
 475struct kmem_cache *fs_cachep;
 476
 477/* SLAB cache for mm_struct structures (tsk->mm) */
 478static struct kmem_cache *mm_cachep;
 479
 480static void account_kernel_stack(struct task_struct *tsk, int account)
 481{
 482	if (IS_ENABLED(CONFIG_VMAP_STACK)) {
 483		struct vm_struct *vm_area = task_stack_vm_area(tsk);
 484		int i;
 485
 486		for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 487			mod_lruvec_page_state(vm_area->pages[i], NR_KERNEL_STACK_KB,
 488					      account * (PAGE_SIZE / 1024));
 489	} else {
 490		void *stack = task_stack_page(tsk);
 491
 492		/* All stack pages are in the same node. */
 493		mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
 494				      account * (THREAD_SIZE / 1024));
 495	}
 496}
 497
 498void exit_task_stack_account(struct task_struct *tsk)
 499{
 500	account_kernel_stack(tsk, -1);
 501
 502	if (IS_ENABLED(CONFIG_VMAP_STACK)) {
 503		struct vm_struct *vm_area;
 504		int i;
 505
 506		vm_area = task_stack_vm_area(tsk);
 507		for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 508			memcg_kmem_uncharge_page(vm_area->pages[i], 0);
 509	}
 510}
 511
 512static void release_task_stack(struct task_struct *tsk)
 513{
 514	if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
 515		return;  /* Better to leak the stack than to free prematurely */
 516
 517	free_thread_stack(tsk);
 518}
 519
 520#ifdef CONFIG_THREAD_INFO_IN_TASK
 521void put_task_stack(struct task_struct *tsk)
 522{
 523	if (refcount_dec_and_test(&tsk->stack_refcount))
 524		release_task_stack(tsk);
 525}
 526#endif
 527
 528void free_task(struct task_struct *tsk)
 529{
 530#ifdef CONFIG_SECCOMP
 531	WARN_ON_ONCE(tsk->seccomp.filter);
 532#endif
 533	release_user_cpus_ptr(tsk);
 534	scs_release(tsk);
 535
 536#ifndef CONFIG_THREAD_INFO_IN_TASK
 537	/*
 538	 * The task is finally done with both the stack and thread_info,
 539	 * so free both.
 540	 */
 541	release_task_stack(tsk);
 542#else
 543	/*
 544	 * If the task had a separate stack allocation, it should be gone
 545	 * by now.
 546	 */
 547	WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
 548#endif
 549	rt_mutex_debug_task_free(tsk);
 550	ftrace_graph_exit_task(tsk);
 551	arch_release_task_struct(tsk);
 552	if (tsk->flags & PF_KTHREAD)
 553		free_kthread_struct(tsk);
 554	bpf_task_storage_free(tsk);
 555	free_task_struct(tsk);
 556}
 557EXPORT_SYMBOL(free_task);
 558
 559void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
 560{
 561	struct file *exe_file;
 562
 563	exe_file = get_mm_exe_file(oldmm);
 564	RCU_INIT_POINTER(mm->exe_file, exe_file);
 565	/*
 566	 * We depend on the oldmm having properly denied write access to the
 567	 * exe_file already.
 568	 */
 569	if (exe_file && exe_file_deny_write_access(exe_file))
 570		pr_warn_once("exe_file_deny_write_access() failed in %s\n", __func__);
 571}
 572
 573#ifdef CONFIG_MMU
 574static inline int mm_alloc_pgd(struct mm_struct *mm)
 575{
 576	mm->pgd = pgd_alloc(mm);
 577	if (unlikely(!mm->pgd))
 578		return -ENOMEM;
 579	return 0;
 580}
 581
 582static inline void mm_free_pgd(struct mm_struct *mm)
 583{
 584	pgd_free(mm, mm->pgd);
 585}
 586#else
 587#define mm_alloc_pgd(mm)	(0)
 588#define mm_free_pgd(mm)
 589#endif /* CONFIG_MMU */
 590
 591#ifdef CONFIG_MM_ID
 592static DEFINE_IDA(mm_ida);
 593
 594static inline int mm_alloc_id(struct mm_struct *mm)
 595{
 596	int ret;
 597
 598	ret = ida_alloc_range(&mm_ida, MM_ID_MIN, MM_ID_MAX, GFP_KERNEL);
 599	if (ret < 0)
 600		return ret;
 601	mm->mm_id = ret;
 602	return 0;
 603}
 604
 605static inline void mm_free_id(struct mm_struct *mm)
 606{
 607	const mm_id_t id = mm->mm_id;
 608
 609	mm->mm_id = MM_ID_DUMMY;
 610	if (id == MM_ID_DUMMY)
 611		return;
 612	if (WARN_ON_ONCE(id < MM_ID_MIN || id > MM_ID_MAX))
 613		return;
 614	ida_free(&mm_ida, id);
 615}
 616#else /* !CONFIG_MM_ID */
 617static inline int mm_alloc_id(struct mm_struct *mm) { return 0; }
 618static inline void mm_free_id(struct mm_struct *mm) {}
 619#endif /* CONFIG_MM_ID */
 620
 621static void check_mm(struct mm_struct *mm)
 622{
 623	int i;
 624
 625	BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
 626			 "Please make sure 'struct resident_page_types[]' is updated as well");
 627
 628	for (i = 0; i < NR_MM_COUNTERS; i++) {
 629		long x = percpu_counter_sum(&mm->rss_stat[i]);
 630
 631		if (unlikely(x)) {
 632			pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld Comm:%s Pid:%d\n",
 633				 mm, resident_page_types[i], x,
 634				 current->comm,
 635				 task_pid_nr(current));
 636		}
 637	}
 638
 639	if (mm_pgtables_bytes(mm))
 640		pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 641				mm_pgtables_bytes(mm));
 642
 643#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !defined(CONFIG_SPLIT_PMD_PTLOCKS)
 644	VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 645#endif
 646}
 647
 648#define allocate_mm()	(kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 649#define free_mm(mm)	(kmem_cache_free(mm_cachep, (mm)))
 650
 651static void do_check_lazy_tlb(void *arg)
 652{
 653	struct mm_struct *mm = arg;
 654
 655	WARN_ON_ONCE(current->active_mm == mm);
 656}
 657
 658static void do_shoot_lazy_tlb(void *arg)
 659{
 660	struct mm_struct *mm = arg;
 661
 662	if (current->active_mm == mm) {
 663		WARN_ON_ONCE(current->mm);
 664		current->active_mm = &init_mm;
 665		switch_mm(mm, &init_mm, current);
 666	}
 667}
 668
 669static void cleanup_lazy_tlbs(struct mm_struct *mm)
 670{
 671	if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) {
 672		/*
 673		 * In this case, lazy tlb mms are refounted and would not reach
 674		 * __mmdrop until all CPUs have switched away and mmdrop()ed.
 675		 */
 676		return;
 677	}
 678
 679	/*
 680	 * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
 681	 * requires lazy mm users to switch to another mm when the refcount
 682	 * drops to zero, before the mm is freed. This requires IPIs here to
 683	 * switch kernel threads to init_mm.
 684	 *
 685	 * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
 686	 * switch with the final userspace teardown TLB flush which leaves the
 687	 * mm lazy on this CPU but no others, reducing the need for additional
 688	 * IPIs here. There are cases where a final IPI is still required here,
 689	 * such as the final mmdrop being performed on a different CPU than the
 690	 * one exiting, or kernel threads using the mm when userspace exits.
 691	 *
 692	 * IPI overheads have not found to be expensive, but they could be
 693	 * reduced in a number of possible ways, for example (roughly
 694	 * increasing order of complexity):
 695	 * - The last lazy reference created by exit_mm() could instead switch
 696	 *   to init_mm, however it's probable this will run on the same CPU
 697	 *   immediately afterwards, so this may not reduce IPIs much.
 698	 * - A batch of mms requiring IPIs could be gathered and freed at once.
 699	 * - CPUs store active_mm where it can be remotely checked without a
 700	 *   lock, to filter out false-positives in the cpumask.
 701	 * - After mm_users or mm_count reaches zero, switching away from the
 702	 *   mm could clear mm_cpumask to reduce some IPIs, perhaps together
 703	 *   with some batching or delaying of the final IPIs.
 704	 * - A delayed freeing and RCU-like quiescing sequence based on mm
 705	 *   switching to avoid IPIs completely.
 706	 */
 707	on_each_cpu_mask(mm_cpumask(mm), do_shoot_lazy_tlb, (void *)mm, 1);
 708	if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES))
 709		on_each_cpu(do_check_lazy_tlb, (void *)mm, 1);
 710}
 711
 712/*
 713 * Called when the last reference to the mm
 714 * is dropped: either by a lazy thread or by
 715 * mmput. Free the page directory and the mm.
 716 */
 717void __mmdrop(struct mm_struct *mm)
 718{
 719	BUG_ON(mm == &init_mm);
 720	WARN_ON_ONCE(mm == current->mm);
 721
 722	/* Ensure no CPUs are using this as their lazy tlb mm */
 723	cleanup_lazy_tlbs(mm);
 724
 725	WARN_ON_ONCE(mm == current->active_mm);
 726	mm_free_pgd(mm);
 727	mm_free_id(mm);
 728	destroy_context(mm);
 729	mmu_notifier_subscriptions_destroy(mm);
 730	check_mm(mm);
 731	put_user_ns(mm->user_ns);
 732	mm_pasid_drop(mm);
 733	mm_destroy_cid(mm);
 734	percpu_counter_destroy_many(mm->rss_stat, NR_MM_COUNTERS);
 735
 736	free_mm(mm);
 737}
 738EXPORT_SYMBOL_GPL(__mmdrop);
 739
 740static void mmdrop_async_fn(struct work_struct *work)
 741{
 742	struct mm_struct *mm;
 743
 744	mm = container_of(work, struct mm_struct, async_put_work);
 745	__mmdrop(mm);
 746}
 747
 748static void mmdrop_async(struct mm_struct *mm)
 749{
 750	if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 751		INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 752		schedule_work(&mm->async_put_work);
 753	}
 754}
 755
 756static inline void free_signal_struct(struct signal_struct *sig)
 757{
 758	taskstats_tgid_free(sig);
 759	sched_autogroup_exit(sig);
 760	/*
 761	 * __mmdrop is not safe to call from softirq context on x86 due to
 762	 * pgd_dtor so postpone it to the async context
 763	 */
 764	if (sig->oom_mm)
 765		mmdrop_async(sig->oom_mm);
 766	kmem_cache_free(signal_cachep, sig);
 767}
 768
 769static inline void put_signal_struct(struct signal_struct *sig)
 770{
 771	if (refcount_dec_and_test(&sig->sigcnt))
 772		free_signal_struct(sig);
 773}
 774
 775void __put_task_struct(struct task_struct *tsk)
 776{
 777	WARN_ON(!tsk->exit_state);
 778	WARN_ON(refcount_read(&tsk->usage));
 779	WARN_ON(tsk == current);
 780
 781	unwind_task_free(tsk);
 782	io_uring_free(tsk);
 783	cgroup_task_free(tsk);
 784	task_numa_free(tsk, true);
 785	security_task_free(tsk);
 786	exit_creds(tsk);
 787	delayacct_tsk_free(tsk);
 788	put_signal_struct(tsk->signal);
 789	sched_core_free(tsk);
 790	free_task(tsk);
 791}
 792EXPORT_SYMBOL_GPL(__put_task_struct);
 793
 794void __put_task_struct_rcu_cb(struct rcu_head *rhp)
 795{
 796	struct task_struct *task = container_of(rhp, struct task_struct, rcu);
 797
 798	__put_task_struct(task);
 799}
 800EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb);
 801
 802void __init __weak arch_task_cache_init(void) { }
 803
 804/*
 805 * set_max_threads
 806 */
 807static void __init set_max_threads(unsigned int max_threads_suggested)
 808{
 809	u64 threads;
 810	unsigned long nr_pages = memblock_estimated_nr_free_pages();
 811
 812	/*
 813	 * The number of threads shall be limited such that the thread
 814	 * structures may only consume a small part of the available memory.
 815	 */
 816	if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
 817		threads = MAX_THREADS;
 818	else
 819		threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
 820				    (u64) THREAD_SIZE * 8UL);
 821
 822	if (threads > max_threads_suggested)
 823		threads = max_threads_suggested;
 824
 825	max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 826}
 827
 828#ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 829/* Initialized by the architecture: */
 830int arch_task_struct_size __read_mostly;
 831#endif
 832
 833static void __init task_struct_whitelist(unsigned long *offset, unsigned long *size)
 834{
 835	/* Fetch thread_struct whitelist for the architecture. */
 836	arch_thread_struct_whitelist(offset, size);
 837
 838	/*
 839	 * Handle zero-sized whitelist or empty thread_struct, otherwise
 840	 * adjust offset to position of thread_struct in task_struct.
 841	 */
 842	if (unlikely(*size == 0))
 843		*offset = 0;
 844	else
 845		*offset += offsetof(struct task_struct, thread);
 846}
 847
 848void __init fork_init(void)
 849{
 850	int i;
 851#ifndef ARCH_MIN_TASKALIGN
 852#define ARCH_MIN_TASKALIGN	0
 853#endif
 854	int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 855	unsigned long useroffset, usersize;
 856
 857	/* create a slab on which task_structs can be allocated */
 858	task_struct_whitelist(&useroffset, &usersize);
 859	task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 860			arch_task_struct_size, align,
 861			SLAB_PANIC|SLAB_ACCOUNT,
 862			useroffset, usersize, NULL);
 863
 864	/* do the arch specific task caches init */
 865	arch_task_cache_init();
 866
 867	set_max_threads(MAX_THREADS);
 868
 869	init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 870	init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 871	init_task.signal->rlim[RLIMIT_SIGPENDING] =
 872		init_task.signal->rlim[RLIMIT_NPROC];
 873
 874	for (i = 0; i < UCOUNT_COUNTS; i++)
 875		init_user_ns.ucount_max[i] = max_threads/2;
 876
 877	set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC,      RLIM_INFINITY);
 878	set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE,   RLIM_INFINITY);
 879	set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
 880	set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK,    RLIM_INFINITY);
 881
 882#ifdef CONFIG_VMAP_STACK
 883	cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 884			  NULL, free_vm_stack_cache);
 885#endif
 886
 887	scs_init();
 888
 889	lockdep_init_task(&init_task);
 890	uprobes_init();
 891}
 892
 893int __weak arch_dup_task_struct(struct task_struct *dst,
 894					       struct task_struct *src)
 895{
 896	*dst = *src;
 897	return 0;
 898}
 899
 900void set_task_stack_end_magic(struct task_struct *tsk)
 901{
 902	unsigned long *stackend;
 903
 904	stackend = end_of_stack(tsk);
 905	*stackend = STACK_END_MAGIC;	/* for overflow detection */
 906}
 907
 908static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 909{
 910	struct task_struct *tsk;
 911	int err;
 912
 913	if (node == NUMA_NO_NODE)
 914		node = tsk_fork_get_node(orig);
 915	tsk = alloc_task_struct_node(node);
 916	if (!tsk)
 917		return NULL;
 918
 919	err = arch_dup_task_struct(tsk, orig);
 920	if (err)
 921		goto free_tsk;
 922
 923	err = alloc_thread_stack_node(tsk, node);
 924	if (err)
 925		goto free_tsk;
 926
 927#ifdef CONFIG_THREAD_INFO_IN_TASK
 928	refcount_set(&tsk->stack_refcount, 1);
 929#endif
 930	account_kernel_stack(tsk, 1);
 931
 932	err = scs_prepare(tsk, node);
 933	if (err)
 934		goto free_stack;
 935
 936#ifdef CONFIG_SECCOMP
 937	/*
 938	 * We must handle setting up seccomp filters once we're under
 939	 * the sighand lock in case orig has changed between now and
 940	 * then. Until then, filter must be NULL to avoid messing up
 941	 * the usage counts on the error path calling free_task.
 942	 */
 943	tsk->seccomp.filter = NULL;
 944#endif
 945
 946	setup_thread_stack(tsk, orig);
 947	clear_user_return_notifier(tsk);
 948	clear_tsk_need_resched(tsk);
 949	set_task_stack_end_magic(tsk);
 950	clear_syscall_work_syscall_user_dispatch(tsk);
 951
 952#ifdef CONFIG_STACKPROTECTOR
 953	tsk->stack_canary = get_random_canary();
 954#endif
 955	if (orig->cpus_ptr == &orig->cpus_mask)
 956		tsk->cpus_ptr = &tsk->cpus_mask;
 957	dup_user_cpus_ptr(tsk, orig, node);
 958
 959	/*
 960	 * One for the user space visible state that goes away when reaped.
 961	 * One for the scheduler.
 962	 */
 963	refcount_set(&tsk->rcu_users, 2);
 964	/* One for the rcu users */
 965	refcount_set(&tsk->usage, 1);
 966#ifdef CONFIG_BLK_DEV_IO_TRACE
 967	tsk->btrace_seq = 0;
 968#endif
 969	tsk->splice_pipe = NULL;
 970	tsk->task_frag.page = NULL;
 971	tsk->wake_q.next = NULL;
 972	tsk->worker_private = NULL;
 973
 974	kcov_task_init(tsk);
 975	kmsan_task_create(tsk);
 976	kmap_local_fork(tsk);
 977
 978#ifdef CONFIG_FAULT_INJECTION
 979	tsk->fail_nth = 0;
 980#endif
 981
 982#ifdef CONFIG_BLK_CGROUP
 983	tsk->throttle_disk = NULL;
 984	tsk->use_memdelay = 0;
 985#endif
 986
 987#ifdef CONFIG_ARCH_HAS_CPU_PASID
 988	tsk->pasid_activated = 0;
 989#endif
 990
 991#ifdef CONFIG_MEMCG
 992	tsk->active_memcg = NULL;
 993#endif
 994
 995#ifdef CONFIG_X86_BUS_LOCK_DETECT
 996	tsk->reported_split_lock = 0;
 997#endif
 998
 999#ifdef CONFIG_SCHED_MM_CID
1000	tsk->mm_cid.cid = MM_CID_UNSET;
1001	tsk->mm_cid.active = 0;
1002	INIT_HLIST_NODE(&tsk->mm_cid.node);
1003#endif
1004	return tsk;
1005
1006free_stack:
1007	exit_task_stack_account(tsk);
1008	free_thread_stack(tsk);
1009free_tsk:
1010	free_task_struct(tsk);
1011	return NULL;
1012}
1013
1014__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1015
1016static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1017
1018static int __init coredump_filter_setup(char *s)
1019{
1020	default_dump_filter =
1021		(simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1022		MMF_DUMP_FILTER_MASK;
1023	return 1;
1024}
1025
1026__setup("coredump_filter=", coredump_filter_setup);
1027
1028#include <linux/init_task.h>
1029
1030static void mm_init_aio(struct mm_struct *mm)
1031{
1032#ifdef CONFIG_AIO
1033	spin_lock_init(&mm->ioctx_lock);
1034	mm->ioctx_table = NULL;
1035#endif
1036}
1037
1038static __always_inline void mm_clear_owner(struct mm_struct *mm,
1039					   struct task_struct *p)
1040{
1041#ifdef CONFIG_MEMCG
1042	if (mm->owner == p)
1043		WRITE_ONCE(mm->owner, NULL);
1044#endif
1045}
1046
1047static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1048{
1049#ifdef CONFIG_MEMCG
1050	mm->owner = p;
1051#endif
1052}
1053
1054static void mm_init_uprobes_state(struct mm_struct *mm)
1055{
1056#ifdef CONFIG_UPROBES
1057	mm->uprobes_state.xol_area = NULL;
1058	arch_uprobe_init_state(mm);
1059#endif
1060}
1061
1062static void mmap_init_lock(struct mm_struct *mm)
1063{
1064	init_rwsem(&mm->mmap_lock);
1065	mm_lock_seqcount_init(mm);
1066#ifdef CONFIG_PER_VMA_LOCK
1067	rcuwait_init(&mm->vma_writer_wait);
1068#endif
1069}
1070
1071static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1072	struct user_namespace *user_ns)
1073{
1074	mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1075	mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1076	atomic_set(&mm->mm_users, 1);
1077	atomic_set(&mm->mm_count, 1);
1078	seqcount_init(&mm->write_protect_seq);
1079	mmap_init_lock(mm);
1080	INIT_LIST_HEAD(&mm->mmlist);
1081	mm_pgtables_bytes_init(mm);
1082	mm->map_count = 0;
1083	mm->locked_vm = 0;
1084	atomic64_set(&mm->pinned_vm, 0);
1085	memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1086	spin_lock_init(&mm->page_table_lock);
1087	spin_lock_init(&mm->arg_lock);
1088	mm_init_cpumask(mm);
1089	mm_init_aio(mm);
1090	mm_init_owner(mm, p);
1091	mm_pasid_init(mm);
1092	RCU_INIT_POINTER(mm->exe_file, NULL);
1093	mmu_notifier_subscriptions_init(mm);
1094	init_tlb_flush_pending(mm);
1095#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !defined(CONFIG_SPLIT_PMD_PTLOCKS)
1096	mm->pmd_huge_pte = NULL;
1097#endif
1098	mm_init_uprobes_state(mm);
1099	hugetlb_count_init(mm);
1100
1101	mm_flags_clear_all(mm);
1102	if (current->mm) {
1103		unsigned long flags = __mm_flags_get_word(current->mm);
1104
1105		__mm_flags_overwrite_word(mm, mmf_init_legacy_flags(flags));
1106		mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1107	} else {
1108		__mm_flags_overwrite_word(mm, default_dump_filter);
1109		mm->def_flags = 0;
1110	}
1111
1112	if (futex_mm_init(mm))
1113		goto fail_mm_init;
1114
1115	if (mm_alloc_pgd(mm))
1116		goto fail_nopgd;
1117
1118	if (mm_alloc_id(mm))
1119		goto fail_noid;
1120
1121	if (init_new_context(p, mm))
1122		goto fail_nocontext;
1123
1124	if (mm_alloc_cid(mm, p))
1125		goto fail_cid;
1126
1127	if (percpu_counter_init_many(mm->rss_stat, 0, GFP_KERNEL_ACCOUNT,
1128				     NR_MM_COUNTERS))
1129		goto fail_pcpu;
1130
1131	mm->user_ns = get_user_ns(user_ns);
1132	lru_gen_init_mm(mm);
1133	return mm;
1134
1135fail_pcpu:
1136	mm_destroy_cid(mm);
1137fail_cid:
1138	destroy_context(mm);
1139fail_nocontext:
1140	mm_free_id(mm);
1141fail_noid:
1142	mm_free_pgd(mm);
1143fail_nopgd:
1144	futex_hash_free(mm);
1145fail_mm_init:
1146	free_mm(mm);
1147	return NULL;
1148}
1149
1150/*
1151 * Allocate and initialize an mm_struct.
1152 */
1153struct mm_struct *mm_alloc(void)
1154{
1155	struct mm_struct *mm;
1156
1157	mm = allocate_mm();
1158	if (!mm)
1159		return NULL;
1160
1161	memset(mm, 0, sizeof(*mm));
1162	return mm_init(mm, current, current_user_ns());
1163}
1164EXPORT_SYMBOL_IF_KUNIT(mm_alloc);
1165
1166static inline void __mmput(struct mm_struct *mm)
1167{
1168	VM_BUG_ON(atomic_read(&mm->mm_users));
1169
1170	uprobe_clear_state(mm);
1171	exit_aio(mm);
1172	ksm_exit(mm);
1173	khugepaged_exit(mm); /* must run before exit_mmap */
1174	exit_mmap(mm);
1175	mm_put_huge_zero_folio(mm);
1176	set_mm_exe_file(mm, NULL);
1177	if (!list_empty(&mm->mmlist)) {
1178		spin_lock(&mmlist_lock);
1179		list_del(&mm->mmlist);
1180		spin_unlock(&mmlist_lock);
1181	}
1182	if (mm->binfmt)
1183		module_put(mm->binfmt->module);
1184	lru_gen_del_mm(mm);
1185	futex_hash_free(mm);
1186	mmdrop(mm);
1187}
1188
1189/*
1190 * Decrement the use count and release all resources for an mm.
1191 */
1192void mmput(struct mm_struct *mm)
1193{
1194	might_sleep();
1195
1196	if (atomic_dec_and_test(&mm->mm_users))
1197		__mmput(mm);
1198}
1199EXPORT_SYMBOL_GPL(mmput);
1200
1201#if defined(CONFIG_MMU) || defined(CONFIG_FUTEX_PRIVATE_HASH)
1202static void mmput_async_fn(struct work_struct *work)
1203{
1204	struct mm_struct *mm = container_of(work, struct mm_struct,
1205					    async_put_work);
1206
1207	__mmput(mm);
1208}
1209
1210void mmput_async(struct mm_struct *mm)
1211{
1212	if (atomic_dec_and_test(&mm->mm_users)) {
1213		INIT_WORK(&mm->async_put_work, mmput_async_fn);
1214		schedule_work(&mm->async_put_work);
1215	}
1216}
1217EXPORT_SYMBOL_GPL(mmput_async);
1218#endif
1219
1220/**
1221 * set_mm_exe_file - change a reference to the mm's executable file
1222 * @mm: The mm to change.
1223 * @new_exe_file: The new file to use.
1224 *
1225 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1226 *
1227 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1228 * invocations: in mmput() nobody alive left, in execve it happens before
1229 * the new mm is made visible to anyone.
1230 *
1231 * Can only fail if new_exe_file != NULL.
1232 */
1233int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1234{
1235	struct file *old_exe_file;
1236
1237	/*
1238	 * It is safe to dereference the exe_file without RCU as
1239	 * this function is only called if nobody else can access
1240	 * this mm -- see comment above for justification.
1241	 */
1242	old_exe_file = rcu_dereference_raw(mm->exe_file);
1243
1244	if (new_exe_file) {
1245		/*
1246		 * We expect the caller (i.e., sys_execve) to already denied
1247		 * write access, so this is unlikely to fail.
1248		 */
1249		if (unlikely(exe_file_deny_write_access(new_exe_file)))
1250			return -EACCES;
1251		get_file(new_exe_file);
1252	}
1253	rcu_assign_pointer(mm->exe_file, new_exe_file);
1254	if (old_exe_file) {
1255		exe_file_allow_write_access(old_exe_file);
1256		fput(old_exe_file);
1257	}
1258	return 0;
1259}
1260
1261/**
1262 * replace_mm_exe_file - replace a reference to the mm's executable file
1263 * @mm: The mm to change.
1264 * @new_exe_file: The new file to use.
1265 *
1266 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1267 *
1268 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1269 */
1270int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1271{
1272	struct vm_area_struct *vma;
1273	struct file *old_exe_file;
1274	int ret = 0;
1275
1276	/* Forbid mm->exe_file change if old file still mapped. */
1277	old_exe_file = get_mm_exe_file(mm);
1278	if (old_exe_file) {
1279		VMA_ITERATOR(vmi, mm, 0);
1280		mmap_read_lock(mm);
1281		for_each_vma(vmi, vma) {
1282			if (!vma->vm_file)
1283				continue;
1284			if (path_equal(&vma->vm_file->f_path,
1285				       &old_exe_file->f_path)) {
1286				ret = -EBUSY;
1287				break;
1288			}
1289		}
1290		mmap_read_unlock(mm);
1291		fput(old_exe_file);
1292		if (ret)
1293			return ret;
1294	}
1295
1296	ret = exe_file_deny_write_access(new_exe_file);
1297	if (ret)
1298		return -EACCES;
1299	get_file(new_exe_file);
1300
1301	/* set the new file */
1302	mmap_write_lock(mm);
1303	old_exe_file = rcu_dereference_raw(mm->exe_file);
1304	rcu_assign_pointer(mm->exe_file, new_exe_file);
1305	mmap_write_unlock(mm);
1306
1307	if (old_exe_file) {
1308		exe_file_allow_write_access(old_exe_file);
1309		fput(old_exe_file);
1310	}
1311	return 0;
1312}
1313
1314/**
1315 * get_mm_exe_file - acquire a reference to the mm's executable file
1316 * @mm: The mm of interest.
1317 *
1318 * Returns %NULL if mm has no associated executable file.
1319 * User must release file via fput().
1320 */
1321struct file *get_mm_exe_file(struct mm_struct *mm)
1322{
1323	struct file *exe_file;
1324
1325	rcu_read_lock();
1326	exe_file = get_file_rcu(&mm->exe_file);
1327	rcu_read_unlock();
1328	return exe_file;
1329}
1330
1331/**
1332 * get_task_exe_file - acquire a reference to the task's executable file
1333 * @task: The task.
1334 *
1335 * Returns %NULL if task's mm (if any) has no associated executable file or
1336 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1337 * User must release file via fput().
1338 */
1339struct file *get_task_exe_file(struct task_struct *task)
1340{
1341	struct file *exe_file = NULL;
1342	struct mm_struct *mm;
1343
1344	if (task->flags & PF_KTHREAD)
1345		return NULL;
1346
1347	task_lock(task);
1348	mm = task->mm;
1349	if (mm)
1350		exe_file = get_mm_exe_file(mm);
1351	task_unlock(task);
1352	return exe_file;
1353}
1354
1355/**
1356 * get_task_mm - acquire a reference to the task's mm
1357 * @task: The task.
1358 *
1359 * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1360 * this kernel workthread has transiently adopted a user mm with use_mm,
1361 * to do its AIO) is not set and if so returns a reference to it, after
1362 * bumping up the use count.  User must release the mm via mmput()
1363 * after use.  Typically used by /proc and ptrace.
1364 */
1365struct mm_struct *get_task_mm(struct task_struct *task)
1366{
1367	struct mm_struct *mm;
1368
1369	if (task->flags & PF_KTHREAD)
1370		return NULL;
1371
1372	task_lock(task);
1373	mm = task->mm;
1374	if (mm)
1375		mmget(mm);
1376	task_unlock(task);
1377	return mm;
1378}
1379EXPORT_SYMBOL_GPL(get_task_mm);
1380
1381static bool may_access_mm(struct mm_struct *mm, struct task_struct *task, unsigned int mode)
1382{
1383	if (mm == current->mm)
1384		return true;
1385	if (ptrace_may_access(task, mode))
1386		return true;
1387	if ((mode & PTRACE_MODE_READ) && perfmon_capable())
1388		return true;
1389	return false;
1390}
1391
1392struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1393{
1394	struct mm_struct *mm;
1395	int err;
1396
1397	err =  down_read_killable(&task->signal->exec_update_lock);
1398	if (err)
1399		return ERR_PTR(err);
1400
1401	mm = get_task_mm(task);
1402	if (!mm) {
1403		mm = ERR_PTR(-ESRCH);
1404	} else if (!may_access_mm(mm, task, mode)) {
1405		mmput(mm);
1406		mm = ERR_PTR(-EACCES);
1407	}
1408	up_read(&task->signal->exec_update_lock);
1409
1410	return mm;
1411}
1412
1413static void complete_vfork_done(struct task_struct *tsk)
1414{
1415	struct completion *vfork;
1416
1417	task_lock(tsk);
1418	vfork = tsk->vfork_done;
1419	if (likely(vfork)) {
1420		tsk->vfork_done = NULL;
1421		complete(vfork);
1422	}
1423	task_unlock(tsk);
1424}
1425
1426static int wait_for_vfork_done(struct task_struct *child,
1427				struct completion *vfork)
1428{
1429	unsigned int state = TASK_KILLABLE|TASK_FREEZABLE;
1430	int killed;
1431
1432	cgroup_enter_frozen();
1433	killed = wait_for_completion_state(vfork, state);
1434	cgroup_leave_frozen(false);
1435
1436	if (killed) {
1437		task_lock(child);
1438		child->vfork_done = NULL;
1439		task_unlock(child);
1440	}
1441
1442	put_task_struct(child);
1443	return killed;
1444}
1445
1446/* Please note the differences between mmput and mm_release.
1447 * mmput is called whenever we stop holding onto a mm_struct,
1448 * error success whatever.
1449 *
1450 * mm_release is called after a mm_struct has been removed
1451 * from the current process.
1452 *
1453 * This difference is important for error handling, when we
1454 * only half set up a mm_struct for a new process and need to restore
1455 * the old one.  Because we mmput the new mm_struct before
1456 * restoring the old one. . .
1457 * Eric Biederman 10 January 1998
1458 */
1459static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1460{
1461	uprobe_free_utask(tsk);
1462
1463	/* Get rid of any cached register state */
1464	deactivate_mm(tsk, mm);
1465
1466	/*
1467	 * Signal userspace if we're not exiting with a core dump
1468	 * because we want to leave the value intact for debugging
1469	 * purposes.
1470	 */
1471	if (tsk->clear_child_tid) {
1472		if (atomic_read(&mm->mm_users) > 1) {
1473			/*
1474			 * We don't check the error code - if userspace has
1475			 * not set up a proper pointer then tough luck.
1476			 */
1477			put_user(0, tsk->clear_child_tid);
1478			do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1479					1, NULL, NULL, 0, 0);
1480		}
1481		tsk->clear_child_tid = NULL;
1482	}
1483
1484	/*
1485	 * All done, finally we can wake up parent and return this mm to him.
1486	 * Also kthread_stop() uses this completion for synchronization.
1487	 */
1488	if (tsk->vfork_done)
1489		complete_vfork_done(tsk);
1490}
1491
1492void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1493{
1494	futex_exit_release(tsk);
1495	mm_release(tsk, mm);
1496}
1497
1498void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1499{
1500	futex_exec_release(tsk);
1501	mm_release(tsk, mm);
1502}
1503
1504/**
1505 * dup_mm() - duplicates an existing mm structure
1506 * @tsk: the task_struct with which the new mm will be associated.
1507 * @oldmm: the mm to duplicate.
1508 *
1509 * Allocates a new mm structure and duplicates the provided @oldmm structure
1510 * content into it.
1511 *
1512 * Return: the duplicated mm or NULL on failure.
1513 */
1514static struct mm_struct *dup_mm(struct task_struct *tsk,
1515				struct mm_struct *oldmm)
1516{
1517	struct mm_struct *mm;
1518	int err;
1519
1520	mm = allocate_mm();
1521	if (!mm)
1522		goto fail_nomem;
1523
1524	memcpy(mm, oldmm, sizeof(*mm));
1525
1526	if (!mm_init(mm, tsk, mm->user_ns))
1527		goto fail_nomem;
1528
1529	uprobe_start_dup_mmap();
1530	err = dup_mmap(mm, oldmm);
1531	if (err)
1532		goto free_pt;
1533	uprobe_end_dup_mmap();
1534
1535	mm->hiwater_rss = get_mm_rss(mm);
1536	mm->hiwater_vm = mm->total_vm;
1537
1538	if (mm->binfmt && !try_module_get(mm->binfmt->module))
1539		goto free_pt;
1540
1541	return mm;
1542
1543free_pt:
1544	/* don't put binfmt in mmput, we haven't got module yet */
1545	mm->binfmt = NULL;
1546	mm_init_owner(mm, NULL);
1547	mmput(mm);
1548	if (err)
1549		uprobe_end_dup_mmap();
1550
1551fail_nomem:
1552	return NULL;
1553}
1554
1555static int copy_mm(u64 clone_flags, struct task_struct *tsk)
1556{
1557	struct mm_struct *mm, *oldmm;
1558
1559	tsk->min_flt = tsk->maj_flt = 0;
1560	tsk->nvcsw = tsk->nivcsw = 0;
1561#ifdef CONFIG_DETECT_HUNG_TASK
1562	tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1563	tsk->last_switch_time = 0;
1564#endif
1565
1566	tsk->mm = NULL;
1567	tsk->active_mm = NULL;
1568
1569	/*
1570	 * Are we cloning a kernel thread?
1571	 *
1572	 * We need to steal a active VM for that..
1573	 */
1574	oldmm = current->mm;
1575	if (!oldmm)
1576		return 0;
1577
1578	if (clone_flags & CLONE_VM) {
1579		mmget(oldmm);
1580		mm = oldmm;
1581	} else {
1582		mm = dup_mm(tsk, current->mm);
1583		if (!mm)
1584			return -ENOMEM;
1585	}
1586
1587	tsk->mm = mm;
1588	tsk->active_mm = mm;
1589	return 0;
1590}
1591
1592static int copy_fs(u64 clone_flags, struct task_struct *tsk)
1593{
1594	struct fs_struct *fs = current->fs;
1595	if (clone_flags & CLONE_FS) {
1596		/* tsk->fs is already what we want */
1597		read_seqlock_excl(&fs->seq);
1598		/* "users" and "in_exec" locked for check_unsafe_exec() */
1599		if (fs->in_exec) {
1600			read_sequnlock_excl(&fs->seq);
1601			return -EAGAIN;
1602		}
1603		fs->users++;
1604		read_sequnlock_excl(&fs->seq);
1605		return 0;
1606	}
1607	tsk->fs = copy_fs_struct(fs);
1608	if (!tsk->fs)
1609		return -ENOMEM;
1610	return 0;
1611}
1612
1613static int copy_files(u64 clone_flags, struct task_struct *tsk,
1614		      int no_files)
1615{
1616	struct files_struct *oldf, *newf;
1617
1618	/*
1619	 * A background process may not have any files ...
1620	 */
1621	oldf = current->files;
1622	if (!oldf)
1623		return 0;
1624
1625	if (no_files) {
1626		tsk->files = NULL;
1627		return 0;
1628	}
1629
1630	if (clone_flags & CLONE_FILES) {
1631		atomic_inc(&oldf->count);
1632		return 0;
1633	}
1634
1635	newf = dup_fd(oldf, NULL);
1636	if (IS_ERR(newf))
1637		return PTR_ERR(newf);
1638
1639	tsk->files = newf;
1640	return 0;
1641}
1642
1643static int copy_sighand(u64 clone_flags, struct task_struct *tsk)
1644{
1645	struct sighand_struct *sig;
1646
1647	if (clone_flags & CLONE_SIGHAND) {
1648		refcount_inc(&current->sighand->count);
1649		return 0;
1650	}
1651	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1652	RCU_INIT_POINTER(tsk->sighand, sig);
1653	if (!sig)
1654		return -ENOMEM;
1655
1656	refcount_set(&sig->count, 1);
1657	spin_lock_irq(&current->sighand->siglock);
1658	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1659	spin_unlock_irq(&current->sighand->siglock);
1660
1661	/* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1662	if (clone_flags & CLONE_CLEAR_SIGHAND)
1663		flush_signal_handlers(tsk, 0);
1664
1665	return 0;
1666}
1667
1668void __cleanup_sighand(struct sighand_struct *sighand)
1669{
1670	if (refcount_dec_and_test(&sighand->count)) {
1671		signalfd_cleanup(sighand);
1672		/*
1673		 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1674		 * without an RCU grace period, see __lock_task_sighand().
1675		 */
1676		kmem_cache_free(sighand_cachep, sighand);
1677	}
1678}
1679
1680/*
1681 * Initialize POSIX timer handling for a thread group.
1682 */
1683static void posix_cpu_timers_init_group(struct signal_struct *sig)
1684{
1685	struct posix_cputimers *pct = &sig->posix_cputimers;
1686	unsigned long cpu_limit;
1687
1688	cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1689	posix_cputimers_group_init(pct, cpu_limit);
1690}
1691
1692static int copy_signal(u64 clone_flags, struct task_struct *tsk)
1693{
1694	struct signal_struct *sig;
1695
1696	if (clone_flags & CLONE_THREAD)
1697		return 0;
1698
1699	sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1700	tsk->signal = sig;
1701	if (!sig)
1702		return -ENOMEM;
1703
1704	sig->nr_threads = 1;
1705	sig->quick_threads = 1;
1706	atomic_set(&sig->live, 1);
1707	refcount_set(&sig->sigcnt, 1);
1708
1709	/* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1710	sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1711	tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1712
1713	init_waitqueue_head(&sig->wait_chldexit);
1714	sig->curr_target = tsk;
1715	init_sigpending(&sig->shared_pending);
1716	INIT_HLIST_HEAD(&sig->multiprocess);
1717	seqlock_init(&sig->stats_lock);
1718	prev_cputime_init(&sig->prev_cputime);
1719
1720#ifdef CONFIG_POSIX_TIMERS
1721	INIT_HLIST_HEAD(&sig->posix_timers);
1722	INIT_HLIST_HEAD(&sig->ignored_posix_timers);
1723	hrtimer_setup(&sig->real_timer, it_real_fn, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1724#endif
1725
1726	task_lock(current->group_leader);
1727	memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1728	task_unlock(current->group_leader);
1729
1730	posix_cpu_timers_init_group(sig);
1731
1732	tty_audit_fork(sig);
1733	sched_autogroup_fork(sig);
1734
1735#ifdef CONFIG_CGROUPS
1736	init_rwsem(&sig->cgroup_threadgroup_rwsem);
1737#endif
1738
1739	sig->oom_score_adj = current->signal->oom_score_adj;
1740	sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1741
1742	mutex_init(&sig->cred_guard_mutex);
1743	init_rwsem(&sig->exec_update_lock);
1744
1745	return 0;
1746}
1747
1748static void copy_seccomp(struct task_struct *p)
1749{
1750#ifdef CONFIG_SECCOMP
1751	/*
1752	 * Must be called with sighand->lock held, which is common to
1753	 * all threads in the group. Holding cred_guard_mutex is not
1754	 * needed because this new task is not yet running and cannot
1755	 * be racing exec.
1756	 */
1757	assert_spin_locked(&current->sighand->siglock);
1758
1759	/* Ref-count the new filter user, and assign it. */
1760	get_seccomp_filter(current);
1761	p->seccomp = current->seccomp;
1762
1763	/*
1764	 * Explicitly enable no_new_privs here in case it got set
1765	 * between the task_struct being duplicated and holding the
1766	 * sighand lock. The seccomp state and nnp must be in sync.
1767	 */
1768	if (task_no_new_privs(current))
1769		task_set_no_new_privs(p);
1770
1771	/*
1772	 * If the parent gained a seccomp mode after copying thread
1773	 * flags and between before we held the sighand lock, we have
1774	 * to manually enable the seccomp thread flag here.
1775	 */
1776	if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1777		set_task_syscall_work(p, SECCOMP);
1778#endif
1779}
1780
1781SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1782{
1783	current->clear_child_tid = tidptr;
1784
1785	return task_pid_vnr(current);
1786}
1787
1788static void rt_mutex_init_task(struct task_struct *p)
1789{
1790	raw_spin_lock_init(&p->pi_lock);
1791#ifdef CONFIG_RT_MUTEXES
1792	p->pi_waiters = RB_ROOT_CACHED;
1793	p->pi_top_task = NULL;
1794	p->pi_blocked_on = NULL;
1795#endif
1796}
1797
1798static inline void init_task_pid_links(struct task_struct *task)
1799{
1800	enum pid_type type;
1801
1802	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1803		INIT_HLIST_NODE(&task->pid_links[type]);
1804}
1805
1806static inline void
1807init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1808{
1809	if (type == PIDTYPE_PID)
1810		task->thread_pid = pid;
1811	else
1812		task->signal->pids[type] = pid;
1813}
1814
1815static inline void rcu_copy_process(struct task_struct *p)
1816{
1817#ifdef CONFIG_PREEMPT_RCU
1818	p->rcu_read_lock_nesting = 0;
1819	p->rcu_read_unlock_special.s = 0;
1820	p->rcu_blocked_node = NULL;
1821	INIT_LIST_HEAD(&p->rcu_node_entry);
1822#endif /* #ifdef CONFIG_PREEMPT_RCU */
1823#ifdef CONFIG_TASKS_RCU
1824	p->rcu_tasks_holdout = false;
1825	INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1826	p->rcu_tasks_idle_cpu = -1;
1827	INIT_LIST_HEAD(&p->rcu_tasks_exit_list);
1828#endif /* #ifdef CONFIG_TASKS_RCU */
1829#ifdef CONFIG_TASKS_TRACE_RCU
1830	p->trc_reader_nesting = 0;
1831	p->trc_reader_special.s = 0;
1832	INIT_LIST_HEAD(&p->trc_holdout_list);
1833	INIT_LIST_HEAD(&p->trc_blkd_node);
1834#endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1835}
1836
1837/**
1838 * pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
1839 * @pid:   the struct pid for which to create a pidfd
1840 * @flags: flags of the new @pidfd
1841 * @ret_file: return the new pidfs file
1842 *
1843 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
1844 * caller's file descriptor table. The pidfd is reserved but not installed yet.
1845 *
1846 * The helper verifies that @pid is still in use, without PIDFD_THREAD the
1847 * task identified by @pid must be a thread-group leader.
1848 *
1849 * If this function returns successfully the caller is responsible to either
1850 * call fd_install() passing the returned pidfd and pidfd file as arguments in
1851 * order to install the pidfd into its file descriptor table or they must use
1852 * put_unused_fd() and fput() on the returned pidfd and pidfd file
1853 * respectively.
1854 *
1855 * This function is useful when a pidfd must already be reserved but there
1856 * might still be points of failure afterwards and the caller wants to ensure
1857 * that no pidfd is leaked into its file descriptor table.
1858 *
1859 * Return: On success, a reserved pidfd is returned from the function and a new
1860 *         pidfd file is returned in the last argument to the function. On
1861 *         error, a negative error code is returned from the function and the
1862 *         last argument remains unchanged.
1863 */
1864int pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret_file)
1865{
1866	struct file *pidfs_file;
1867
1868	/*
1869	 * PIDFD_STALE is only allowed to be passed if the caller knows
1870	 * that @pid is already registered in pidfs and thus
1871	 * PIDFD_INFO_EXIT information is guaranteed to be available.
1872	 */
1873	if (!(flags & PIDFD_STALE)) {
1874		/*
1875		 * While holding the pidfd waitqueue lock removing the
1876		 * task linkage for the thread-group leader pid
1877		 * (PIDTYPE_TGID) isn't possible. Thus, if there's still
1878		 * task linkage for PIDTYPE_PID not having thread-group
1879		 * leader linkage for the pid means it wasn't a
1880		 * thread-group leader in the first place.
1881		 */
1882		guard(spinlock_irq)(&pid->wait_pidfd.lock);
1883
1884		/* Task has already been reaped. */
1885		if (!pid_has_task(pid, PIDTYPE_PID))
1886			return -ESRCH;
1887		/*
1888		 * If this struct pid isn't used as a thread-group
1889		 * leader but the caller requested to create a
1890		 * thread-group leader pidfd then report ENOENT.
1891		 */
1892		if (!(flags & PIDFD_THREAD) && !pid_has_task(pid, PIDTYPE_TGID))
1893			return -ENOENT;
1894	}
1895
1896	CLASS(get_unused_fd, pidfd)(O_CLOEXEC);
1897	if (pidfd < 0)
1898		return pidfd;
1899
1900	pidfs_file = pidfs_alloc_file(pid, flags | O_RDWR);
1901	if (IS_ERR(pidfs_file))
1902		return PTR_ERR(pidfs_file);
1903
1904	*ret_file = pidfs_file;
1905	return take_fd(pidfd);
1906}
1907
1908static void __delayed_free_task(struct rcu_head *rhp)
1909{
1910	struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1911
1912	free_task(tsk);
1913}
1914
1915static __always_inline void delayed_free_task(struct task_struct *tsk)
1916{
1917	if (IS_ENABLED(CONFIG_MEMCG))
1918		call_rcu(&tsk->rcu, __delayed_free_task);
1919	else
1920		free_task(tsk);
1921}
1922
1923static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1924{
1925	/* Skip if kernel thread */
1926	if (!tsk->mm)
1927		return;
1928
1929	/* Skip if spawning a thread or using vfork */
1930	if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1931		return;
1932
1933	/* We need to synchronize with __set_oom_adj */
1934	mutex_lock(&oom_adj_mutex);
1935	mm_flags_set(MMF_MULTIPROCESS, tsk->mm);
1936	/* Update the values in case they were changed after copy_signal */
1937	tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1938	tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1939	mutex_unlock(&oom_adj_mutex);
1940}
1941
1942#ifdef CONFIG_RV
1943static void rv_task_fork(struct task_struct *p)
1944{
1945	memset(&p->rv, 0, sizeof(p->rv));
1946}
1947#else
1948#define rv_task_fork(p) do {} while (0)
1949#endif
1950
1951static bool need_futex_hash_allocate_default(u64 clone_flags)
1952{
1953	if ((clone_flags & (CLONE_THREAD | CLONE_VM)) != (CLONE_THREAD | CLONE_VM))
1954		return false;
1955	return true;
1956}
1957
1958/*
1959 * This creates a new process as a copy of the old one,
1960 * but does not actually start it yet.
1961 *
1962 * It copies the registers, and all the appropriate
1963 * parts of the process environment (as per the clone
1964 * flags). The actual kick-off is left to the caller.
1965 */
1966__latent_entropy struct task_struct *copy_process(
1967					struct pid *pid,
1968					int trace,
1969					int node,
1970					struct kernel_clone_args *args)
1971{
1972	int pidfd = -1, retval;
1973	struct task_struct *p;
1974	struct multiprocess_signals delayed;
1975	struct file *pidfile = NULL;
1976	const u64 clone_flags = args->flags;
1977	struct nsproxy *nsp = current->nsproxy;
1978
1979	/*
1980	 * Don't allow sharing the root directory with processes in a different
1981	 * namespace
1982	 */
1983	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1984		return ERR_PTR(-EINVAL);
1985
1986	if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1987		return ERR_PTR(-EINVAL);
1988
1989	/*
1990	 * Thread groups must share signals as well, and detached threads
1991	 * can only be started up within the thread group.
1992	 */
1993	if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1994		return ERR_PTR(-EINVAL);
1995
1996	/*
1997	 * Shared signal handlers imply shared VM. By way of the above,
1998	 * thread groups also imply shared VM. Blocking this case allows
1999	 * for various simplifications in other code.
2000	 */
2001	if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2002		return ERR_PTR(-EINVAL);
2003
2004	/*
2005	 * Siblings of global init remain as zombies on exit since they are
2006	 * not reaped by their parent (swapper). To solve this and to avoid
2007	 * multi-rooted process trees, prevent global and container-inits
2008	 * from creating siblings.
2009	 */
2010	if ((clone_flags & CLONE_PARENT) &&
2011				current->signal->flags & SIGNAL_UNKILLABLE)
2012		return ERR_PTR(-EINVAL);
2013
2014	/*
2015	 * If the new process will be in a different pid or user namespace
2016	 * do not allow it to share a thread group with the forking task.
2017	 */
2018	if (clone_flags & CLONE_THREAD) {
2019		if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2020		    (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2021			return ERR_PTR(-EINVAL);
2022	}
2023
2024	if (clone_flags & CLONE_PIDFD) {
2025		/*
2026		 * - CLONE_DETACHED is blocked so that we can potentially
2027		 *   reuse it later for CLONE_PIDFD.
2028		 */
2029		if (clone_flags & CLONE_DETACHED)
2030			return ERR_PTR(-EINVAL);
2031	}
2032
2033	/*
2034	 * Force any signals received before this point to be delivered
2035	 * before the fork happens.  Collect up signals sent to multiple
2036	 * processes that happen during the fork and delay them so that
2037	 * they appear to happen after the fork.
2038	 */
2039	sigemptyset(&delayed.signal);
2040	INIT_HLIST_NODE(&delayed.node);
2041
2042	spin_lock_irq(&current->sighand->siglock);
2043	if (!(clone_flags & CLONE_THREAD))
2044		hlist_add_head(&delayed.node, &current->signal->multiprocess);
2045	recalc_sigpending();
2046	spin_unlock_irq(&current->sighand->siglock);
2047	retval = -ERESTARTNOINTR;
2048	if (task_sigpending(current))
2049		goto fork_out;
2050
2051	retval = -ENOMEM;
2052	p = dup_task_struct(current, node);
2053	if (!p)
2054		goto fork_out;
2055	p->flags &= ~PF_KTHREAD;
2056	if (args->kthread)
2057		p->flags |= PF_KTHREAD;
2058	if (args->user_worker) {
2059		/*
2060		 * Mark us a user worker, and block any signal that isn't
2061		 * fatal or STOP
2062		 */
2063		p->flags |= PF_USER_WORKER;
2064		siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2065	}
2066	if (args->io_thread)
2067		p->flags |= PF_IO_WORKER;
2068
2069	if (args->name)
2070		strscpy_pad(p->comm, args->name, sizeof(p->comm));
2071
2072	p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2073	/*
2074	 * Clear TID on mm_release()?
2075	 */
2076	p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2077
2078	ftrace_graph_init_task(p);
2079
2080	rt_mutex_init_task(p);
2081
2082	lockdep_assert_irqs_enabled();
2083#ifdef CONFIG_PROVE_LOCKING
2084	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2085#endif
2086	retval = copy_creds(p, clone_flags);
2087	if (retval < 0)
2088		goto bad_fork_free;
2089
2090	retval = -EAGAIN;
2091	if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2092		if (p->real_cred->user != INIT_USER &&
2093		    !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2094			goto bad_fork_cleanup_count;
2095	}
2096	current->flags &= ~PF_NPROC_EXCEEDED;
2097
2098	/*
2099	 * If multiple threads are within copy_process(), then this check
2100	 * triggers too late. This doesn't hurt, the check is only there
2101	 * to stop root fork bombs.
2102	 */
2103	retval = -EAGAIN;
2104	if (data_race(nr_threads >= max_threads))
2105		goto bad_fork_cleanup_count;
2106
2107	delayacct_tsk_init(p);	/* Must remain after dup_task_struct() */
2108	p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2109	p->flags |= PF_FORKNOEXEC;
2110	INIT_LIST_HEAD(&p->children);
2111	INIT_LIST_HEAD(&p->sibling);
2112	rcu_copy_process(p);
2113	p->vfork_done = NULL;
2114	spin_lock_init(&p->alloc_lock);
2115
2116	init_sigpending(&p->pending);
2117
2118	p->utime = p->stime = p->gtime = 0;
2119#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2120	p->utimescaled = p->stimescaled = 0;
2121#endif
2122	prev_cputime_init(&p->prev_cputime);
2123
2124#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2125	seqcount_init(&p->vtime.seqcount);
2126	p->vtime.starttime = 0;
2127	p->vtime.state = VTIME_INACTIVE;
2128#endif
2129
2130#ifdef CONFIG_IO_URING
2131	p->io_uring = NULL;
2132#endif
2133
2134	p->default_timer_slack_ns = current->timer_slack_ns;
2135
2136#ifdef CONFIG_PSI
2137	p->psi_flags = 0;
2138#endif
2139
2140	task_io_accounting_init(&p->ioac);
2141	acct_clear_integrals(p);
2142
2143	posix_cputimers_init(&p->posix_cputimers);
2144	tick_dep_init_task(p);
2145
2146	p->io_context = NULL;
2147	audit_set_context(p, NULL);
2148	cgroup_fork(p);
2149	if (args->kthread) {
2150		if (!set_kthread_struct(p))
2151			goto bad_fork_cleanup_delayacct;
2152	}
2153#ifdef CONFIG_NUMA
2154	p->mempolicy = mpol_dup(p->mempolicy);
2155	if (IS_ERR(p->mempolicy)) {
2156		retval = PTR_ERR(p->mempolicy);
2157		p->mempolicy = NULL;
2158		goto bad_fork_cleanup_delayacct;
2159	}
2160#endif
2161#ifdef CONFIG_CPUSETS
2162	p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2163	seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2164#endif
2165#ifdef CONFIG_TRACE_IRQFLAGS
2166	memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2167	p->irqtrace.hardirq_disable_ip	= _THIS_IP_;
2168	p->irqtrace.softirq_enable_ip	= _THIS_IP_;
2169	p->softirqs_enabled		= 1;
2170	p->softirq_context		= 0;
2171#endif
2172
2173	p->pagefault_disabled = 0;
2174
2175	lockdep_init_task(p);
2176
2177	p->blocked_on = NULL; /* not blocked yet */
2178
2179#ifdef CONFIG_BCACHE
2180	p->sequential_io	= 0;
2181	p->sequential_io_avg	= 0;
2182#endif
2183#ifdef CONFIG_BPF_SYSCALL
2184	RCU_INIT_POINTER(p->bpf_storage, NULL);
2185	p->bpf_ctx = NULL;
2186#endif
2187
2188	unwind_task_init(p);
2189
2190	/* Perform scheduler related setup. Assign this task to a CPU. */
2191	retval = sched_fork(clone_flags, p);
2192	if (retval)
2193		goto bad_fork_cleanup_policy;
2194
2195	retval = perf_event_init_task(p, clone_flags);
2196	if (retval)
2197		goto bad_fork_sched_cancel_fork;
2198	retval = audit_alloc(p);
2199	if (retval)
2200		goto bad_fork_cleanup_perf;
2201	/* copy all the process information */
2202	shm_init_task(p);
2203	retval = security_task_alloc(p, clone_flags);
2204	if (retval)
2205		goto bad_fork_cleanup_audit;
2206	retval = copy_semundo(clone_flags, p);
2207	if (retval)
2208		goto bad_fork_cleanup_security;
2209	retval = copy_files(clone_flags, p, args->no_files);
2210	if (retval)
2211		goto bad_fork_cleanup_semundo;
2212	retval = copy_fs(clone_flags, p);
2213	if (retval)
2214		goto bad_fork_cleanup_files;
2215	retval = copy_sighand(clone_flags, p);
2216	if (retval)
2217		goto bad_fork_cleanup_fs;
2218	retval = copy_signal(clone_flags, p);
2219	if (retval)
2220		goto bad_fork_cleanup_sighand;
2221	retval = copy_mm(clone_flags, p);
2222	if (retval)
2223		goto bad_fork_cleanup_signal;
2224	retval = copy_namespaces(clone_flags, p);
2225	if (retval)
2226		goto bad_fork_cleanup_mm;
2227	retval = copy_io(clone_flags, p);
2228	if (retval)
2229		goto bad_fork_cleanup_namespaces;
2230	retval = copy_thread(p, args);
2231	if (retval)
2232		goto bad_fork_cleanup_io;
2233
2234	stackleak_task_init(p);
2235
2236	if (pid != &init_struct_pid) {
2237		pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2238				args->set_tid_size);
2239		if (IS_ERR(pid)) {
2240			retval = PTR_ERR(pid);
2241			goto bad_fork_cleanup_thread;
2242		}
2243	}
2244
2245	/*
2246	 * This has to happen after we've potentially unshared the file
2247	 * descriptor table (so that the pidfd doesn't leak into the child
2248	 * if the fd table isn't shared).
2249	 */
2250	if (clone_flags & CLONE_PIDFD) {
2251		int flags = (clone_flags & CLONE_THREAD) ? PIDFD_THREAD : 0;
2252
2253		/*
2254		 * Note that no task has been attached to @pid yet indicate
2255		 * that via CLONE_PIDFD.
2256		 */
2257		retval = pidfd_prepare(pid, flags | PIDFD_STALE, &pidfile);
2258		if (retval < 0)
2259			goto bad_fork_free_pid;
2260		pidfd = retval;
2261
2262		retval = put_user(pidfd, args->pidfd);
2263		if (retval)
2264			goto bad_fork_put_pidfd;
2265	}
2266
2267#ifdef CONFIG_BLOCK
2268	p->plug = NULL;
2269#endif
2270	futex_init_task(p);
2271
2272	/*
2273	 * sigaltstack should be cleared when sharing the same VM
2274	 */
2275	if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2276		sas_ss_reset(p);
2277
2278	/*
2279	 * Syscall tracing and stepping should be turned off in the
2280	 * child regardless of CLONE_PTRACE.
2281	 */
2282	user_disable_single_step(p);
2283	clear_task_syscall_work(p, SYSCALL_TRACE);
2284#if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2285	clear_task_syscall_work(p, SYSCALL_EMU);
2286#endif
2287	clear_tsk_latency_tracing(p);
2288
2289	/* ok, now we should be set up.. */
2290	p->pid = pid_nr(pid);
2291	if (clone_flags & CLONE_THREAD) {
2292		p->group_leader = current->group_leader;
2293		p->tgid = current->tgid;
2294	} else {
2295		p->group_leader = p;
2296		p->tgid = p->pid;
2297	}
2298
2299	p->nr_dirtied = 0;
2300	p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2301	p->dirty_paused_when = 0;
2302
2303	p->pdeath_signal = 0;
2304	p->task_works = NULL;
2305	clear_posix_cputimers_work(p);
2306
2307#ifdef CONFIG_KRETPROBES
2308	p->kretprobe_instances.first = NULL;
2309#endif
2310#ifdef CONFIG_RETHOOK
2311	p->rethooks.first = NULL;
2312#endif
2313
2314	/*
2315	 * Ensure that the cgroup subsystem policies allow the new process to be
2316	 * forked. It should be noted that the new process's css_set can be changed
2317	 * between here and cgroup_post_fork() if an organisation operation is in
2318	 * progress.
2319	 */
2320	retval = cgroup_can_fork(p, args);
2321	if (retval)
2322		goto bad_fork_put_pidfd;
2323
2324	/*
2325	 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2326	 * the new task on the correct runqueue. All this *before* the task
2327	 * becomes visible.
2328	 *
2329	 * This isn't part of ->can_fork() because while the re-cloning is
2330	 * cgroup specific, it unconditionally needs to place the task on a
2331	 * runqueue.
2332	 */
2333	retval = sched_cgroup_fork(p, args);
2334	if (retval)
2335		goto bad_fork_cancel_cgroup;
2336
2337	/*
2338	 * Allocate a default futex hash for the user process once the first
2339	 * thread spawns.
2340	 */
2341	if (need_futex_hash_allocate_default(clone_flags)) {
2342		retval = futex_hash_allocate_default();
2343		if (retval)
2344			goto bad_fork_cancel_cgroup;
2345		/*
2346		 * If we fail beyond this point we don't free the allocated
2347		 * futex hash map. We assume that another thread will be created
2348		 * and makes use of it. The hash map will be freed once the main
2349		 * thread terminates.
2350		 */
2351	}
2352	/*
2353	 * From this point on we must avoid any synchronous user-space
2354	 * communication until we take the tasklist-lock. In particular, we do
2355	 * not want user-space to be able to predict the process start-time by
2356	 * stalling fork(2) after we recorded the start_time but before it is
2357	 * visible to the system.
2358	 */
2359
2360	p->start_time = ktime_get_ns();
2361	p->start_boottime = ktime_get_boottime_ns();
2362
2363	/*
2364	 * Make it visible to the rest of the system, but dont wake it up yet.
2365	 * Need tasklist lock for parent etc handling!
2366	 */
2367	write_lock_irq(&tasklist_lock);
2368
2369	/* CLONE_PARENT re-uses the old parent */
2370	if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2371		p->real_parent = current->real_parent;
2372		p->parent_exec_id = current->parent_exec_id;
2373		if (clone_flags & CLONE_THREAD)
2374			p->exit_signal = -1;
2375		else
2376			p->exit_signal = current->group_leader->exit_signal;
2377	} else {
2378		p->real_parent = current;
2379		p->parent_exec_id = current->self_exec_id;
2380		p->exit_signal = args->exit_signal;
2381	}
2382
2383	klp_copy_process(p);
2384
2385	sched_core_fork(p);
2386
2387	spin_lock(&current->sighand->siglock);
2388
2389	rv_task_fork(p);
2390
2391	rseq_fork(p, clone_flags);
2392
2393	/* Don't start children in a dying pid namespace */
2394	if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2395		retval = -ENOMEM;
2396		goto bad_fork_core_free;
2397	}
2398
2399	/* Let kill terminate clone/fork in the middle */
2400	if (fatal_signal_pending(current)) {
2401		retval = -EINTR;
2402		goto bad_fork_core_free;
2403	}
2404
2405	/* No more failure paths after this point. */
2406
2407	/*
2408	 * Copy seccomp details explicitly here, in case they were changed
2409	 * before holding sighand lock.
2410	 */
2411	copy_seccomp(p);
2412
2413	init_task_pid_links(p);
2414	if (likely(p->pid)) {
2415		ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2416
2417		init_task_pid(p, PIDTYPE_PID, pid);
2418		if (thread_group_leader(p)) {
2419			init_task_pid(p, PIDTYPE_TGID, pid);
2420			init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2421			init_task_pid(p, PIDTYPE_SID, task_session(current));
2422
2423			if (is_child_reaper(pid)) {
2424				ns_of_pid(pid)->child_reaper = p;
2425				p->signal->flags |= SIGNAL_UNKILLABLE;
2426			}
2427			p->signal->shared_pending.signal = delayed.signal;
2428			p->signal->tty = tty_kref_get(current->signal->tty);
2429			/*
2430			 * Inherit has_child_subreaper flag under the same
2431			 * tasklist_lock with adding child to the process tree
2432			 * for propagate_has_child_subreaper optimization.
2433			 */
2434			p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2435							 p->real_parent->signal->is_child_subreaper;
2436			list_add_tail(&p->sibling, &p->real_parent->children);
2437			list_add_tail_rcu(&p->tasks, &init_task.tasks);
2438			attach_pid(p, PIDTYPE_TGID);
2439			attach_pid(p, PIDTYPE_PGID);
2440			attach_pid(p, PIDTYPE_SID);
2441			__this_cpu_inc(process_counts);
2442		} else {
2443			current->signal->nr_threads++;
2444			current->signal->quick_threads++;
2445			atomic_inc(&current->signal->live);
2446			refcount_inc(&current->signal->sigcnt);
2447			task_join_group_stop(p);
2448			list_add_tail_rcu(&p->thread_node,
2449					  &p->signal->thread_head);
2450		}
2451		attach_pid(p, PIDTYPE_PID);
2452		nr_threads++;
2453	}
2454	total_forks++;
2455	hlist_del_init(&delayed.node);
2456	spin_unlock(&current->sighand->siglock);
2457	syscall_tracepoint_update(p);
2458	write_unlock_irq(&tasklist_lock);
2459
2460	if (pidfile)
2461		fd_install(pidfd, pidfile);
2462
2463	proc_fork_connector(p);
2464	sched_post_fork(p);
2465	cgroup_post_fork(p, args);
2466	perf_event_fork(p);
2467
2468	trace_task_newtask(p, clone_flags);
2469	uprobe_copy_process(p, clone_flags);
2470	user_events_fork(p, clone_flags);
2471
2472	copy_oom_score_adj(clone_flags, p);
2473
2474	return p;
2475
2476bad_fork_core_free:
2477	sched_core_free(p);
2478	spin_unlock(&current->sighand->siglock);
2479	write_unlock_irq(&tasklist_lock);
2480bad_fork_cancel_cgroup:
2481	cgroup_cancel_fork(p, args);
2482bad_fork_put_pidfd:
2483	if (clone_flags & CLONE_PIDFD) {
2484		fput(pidfile);
2485		put_unused_fd(pidfd);
2486	}
2487bad_fork_free_pid:
2488	if (pid != &init_struct_pid)
2489		free_pid(pid);
2490bad_fork_cleanup_thread:
2491	exit_thread(p);
2492bad_fork_cleanup_io:
2493	if (p->io_context)
2494		exit_io_context(p);
2495bad_fork_cleanup_namespaces:
2496	exit_nsproxy_namespaces(p);
2497bad_fork_cleanup_mm:
2498	if (p->mm) {
2499		mm_clear_owner(p->mm, p);
2500		mmput(p->mm);
2501	}
2502bad_fork_cleanup_signal:
2503	if (!(clone_flags & CLONE_THREAD))
2504		free_signal_struct(p->signal);
2505bad_fork_cleanup_sighand:
2506	__cleanup_sighand(p->sighand);
2507bad_fork_cleanup_fs:
2508	exit_fs(p); /* blocking */
2509bad_fork_cleanup_files:
2510	exit_files(p); /* blocking */
2511bad_fork_cleanup_semundo:
2512	exit_sem(p);
2513bad_fork_cleanup_security:
2514	security_task_free(p);
2515bad_fork_cleanup_audit:
2516	audit_free(p);
2517bad_fork_cleanup_perf:
2518	perf_event_free_task(p);
2519bad_fork_sched_cancel_fork:
2520	sched_cancel_fork(p);
2521bad_fork_cleanup_policy:
2522	lockdep_free_task(p);
2523#ifdef CONFIG_NUMA
2524	mpol_put(p->mempolicy);
2525#endif
2526bad_fork_cleanup_delayacct:
2527	delayacct_tsk_free(p);
2528bad_fork_cleanup_count:
2529	dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2530	exit_cred_namespaces(p);
2531	exit_creds(p);
2532bad_fork_free:
2533	WRITE_ONCE(p->__state, TASK_DEAD);
2534	exit_task_stack_account(p);
2535	put_task_stack(p);
2536	delayed_free_task(p);
2537fork_out:
2538	spin_lock_irq(&current->sighand->siglock);
2539	hlist_del_init(&delayed.node);
2540	spin_unlock_irq(&current->sighand->siglock);
2541	return ERR_PTR(retval);
2542}
2543
2544static inline void init_idle_pids(struct task_struct *idle)
2545{
2546	enum pid_type type;
2547
2548	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2549		INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2550		init_task_pid(idle, type, &init_struct_pid);
2551	}
2552}
2553
2554static int idle_dummy(void *dummy)
2555{
2556	/* This function is never called */
2557	return 0;
2558}
2559
2560struct task_struct * __init fork_idle(int cpu)
2561{
2562	struct task_struct *task;
2563	struct kernel_clone_args args = {
2564		.flags		= CLONE_VM,
2565		.fn		= &idle_dummy,
2566		.fn_arg		= NULL,
2567		.kthread	= 1,
2568		.idle		= 1,
2569	};
2570
2571	task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2572	if (!IS_ERR(task)) {
2573		init_idle_pids(task);
2574		init_idle(task, cpu);
2575	}
2576
2577	return task;
2578}
2579
2580/*
2581 * This is like kernel_clone(), but shaved down and tailored to just
2582 * creating io_uring workers. It returns a created task, or an error pointer.
2583 * The returned task is inactive, and the caller must fire it up through
2584 * wake_up_new_task(p). All signals are blocked in the created task.
2585 */
2586struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2587{
2588	unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2589			      CLONE_IO|CLONE_VM|CLONE_UNTRACED;
2590	struct kernel_clone_args args = {
2591		.flags		= flags,
2592		.fn		= fn,
2593		.fn_arg		= arg,
2594		.io_thread	= 1,
2595		.user_worker	= 1,
2596	};
2597
2598	return copy_process(NULL, 0, node, &args);
2599}
2600
2601/*
2602 *  Ok, this is the main fork-routine.
2603 *
2604 * It copies the process, and if successful kick-starts
2605 * it and waits for it to finish using the VM if required.
2606 *
2607 * args->exit_signal is expected to be checked for sanity by the caller.
2608 */
2609pid_t kernel_clone(struct kernel_clone_args *args)
2610{
2611	u64 clone_flags = args->flags;
2612	struct completion vfork;
2613	struct pid *pid;
2614	struct task_struct *p;
2615	int trace = 0;
2616	pid_t nr;
2617
2618	/*
2619	 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2620	 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2621	 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2622	 * field in struct clone_args and it still doesn't make sense to have
2623	 * them both point at the same memory location. Performing this check
2624	 * here has the advantage that we don't need to have a separate helper
2625	 * to check for legacy clone().
2626	 */
2627	if ((clone_flags & CLONE_PIDFD) &&
2628	    (clone_flags & CLONE_PARENT_SETTID) &&
2629	    (args->pidfd == args->parent_tid))
2630		return -EINVAL;
2631
2632	/*
2633	 * Determine whether and which event to report to ptracer.  When
2634	 * called from kernel_thread or CLONE_UNTRACED is explicitly
2635	 * requested, no event is reported; otherwise, report if the event
2636	 * for the type of forking is enabled.
2637	 */
2638	if (!(clone_flags & CLONE_UNTRACED)) {
2639		if (clone_flags & CLONE_VFORK)
2640			trace = PTRACE_EVENT_VFORK;
2641		else if (args->exit_signal != SIGCHLD)
2642			trace = PTRACE_EVENT_CLONE;
2643		else
2644			trace = PTRACE_EVENT_FORK;
2645
2646		if (likely(!ptrace_event_enabled(current, trace)))
2647			trace = 0;
2648	}
2649
2650	p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2651	add_latent_entropy();
2652
2653	if (IS_ERR(p))
2654		return PTR_ERR(p);
2655
2656	/*
2657	 * Do this prior waking up the new thread - the thread pointer
2658	 * might get invalid after that point, if the thread exits quickly.
2659	 */
2660	trace_sched_process_fork(current, p);
2661
2662	pid = get_task_pid(p, PIDTYPE_PID);
2663	nr = pid_vnr(pid);
2664
2665	if (clone_flags & CLONE_PARENT_SETTID)
2666		put_user(nr, args->parent_tid);
2667
2668	if (clone_flags & CLONE_VFORK) {
2669		p->vfork_done = &vfork;
2670		init_completion(&vfork);
2671		get_task_struct(p);
2672	}
2673
2674	if (IS_ENABLED(CONFIG_LRU_GEN_WALKS_MMU) && !(clone_flags & CLONE_VM)) {
2675		/* lock the task to synchronize with memcg migration */
2676		task_lock(p);
2677		lru_gen_add_mm(p->mm);
2678		task_unlock(p);
2679	}
2680
2681	wake_up_new_task(p);
2682
2683	/* forking complete and child started to run, tell ptracer */
2684	if (unlikely(trace))
2685		ptrace_event_pid(trace, pid);
2686
2687	if (clone_flags & CLONE_VFORK) {
2688		if (!wait_for_vfork_done(p, &vfork))
2689			ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2690	}
2691
2692	put_pid(pid);
2693	return nr;
2694}
2695
2696/*
2697 * Create a kernel thread.
2698 */
2699pid_t kernel_thread(int (*fn)(void *), void *arg, const char *name,
2700		    unsigned long flags)
2701{
2702	struct kernel_clone_args args = {
2703		.flags		= ((flags | CLONE_VM | CLONE_UNTRACED) & ~CSIGNAL),
2704		.exit_signal	= (flags & CSIGNAL),
2705		.fn		= fn,
2706		.fn_arg		= arg,
2707		.name		= name,
2708		.kthread	= 1,
2709	};
2710
2711	return kernel_clone(&args);
2712}
2713
2714/*
2715 * Create a user mode thread.
2716 */
2717pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2718{
2719	struct kernel_clone_args args = {
2720		.flags		= ((flags | CLONE_VM | CLONE_UNTRACED) & ~CSIGNAL),
2721		.exit_signal	= (flags & CSIGNAL),
2722		.fn		= fn,
2723		.fn_arg		= arg,
2724	};
2725
2726	return kernel_clone(&args);
2727}
2728
2729#ifdef __ARCH_WANT_SYS_FORK
2730SYSCALL_DEFINE0(fork)
2731{
2732#ifdef CONFIG_MMU
2733	struct kernel_clone_args args = {
2734		.exit_signal = SIGCHLD,
2735	};
2736
2737	return kernel_clone(&args);
2738#else
2739	/* can not support in nommu mode */
2740	return -EINVAL;
2741#endif
2742}
2743#endif
2744
2745#ifdef __ARCH_WANT_SYS_VFORK
2746SYSCALL_DEFINE0(vfork)
2747{
2748	struct kernel_clone_args args = {
2749		.flags		= CLONE_VFORK | CLONE_VM,
2750		.exit_signal	= SIGCHLD,
2751	};
2752
2753	return kernel_clone(&args);
2754}
2755#endif
2756
2757#ifdef __ARCH_WANT_SYS_CLONE
2758#ifdef CONFIG_CLONE_BACKWARDS
2759SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2760		 int __user *, parent_tidptr,
2761		 unsigned long, tls,
2762		 int __user *, child_tidptr)
2763#elif defined(CONFIG_CLONE_BACKWARDS2)
2764SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2765		 int __user *, parent_tidptr,
2766		 int __user *, child_tidptr,
2767		 unsigned long, tls)
2768#elif defined(CONFIG_CLONE_BACKWARDS3)
2769SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2770		int, stack_size,
2771		int __user *, parent_tidptr,
2772		int __user *, child_tidptr,
2773		unsigned long, tls)
2774#else
2775SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2776		 int __user *, parent_tidptr,
2777		 int __user *, child_tidptr,
2778		 unsigned long, tls)
2779#endif
2780{
2781	struct kernel_clone_args args = {
2782		.flags		= (lower_32_bits(clone_flags) & ~CSIGNAL),
2783		.pidfd		= parent_tidptr,
2784		.child_tid	= child_tidptr,
2785		.parent_tid	= parent_tidptr,
2786		.exit_signal	= (lower_32_bits(clone_flags) & CSIGNAL),
2787		.stack		= newsp,
2788		.tls		= tls,
2789	};
2790
2791	return kernel_clone(&args);
2792}
2793#endif
2794
2795static noinline int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2796					      struct clone_args __user *uargs,
2797					      size_t usize)
2798{
2799	int err;
2800	struct clone_args args;
2801	pid_t *kset_tid = kargs->set_tid;
2802
2803	BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2804		     CLONE_ARGS_SIZE_VER0);
2805	BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2806		     CLONE_ARGS_SIZE_VER1);
2807	BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2808		     CLONE_ARGS_SIZE_VER2);
2809	BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2810
2811	if (unlikely(usize > PAGE_SIZE))
2812		return -E2BIG;
2813	if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2814		return -EINVAL;
2815
2816	err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2817	if (err)
2818		return err;
2819
2820	if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2821		return -EINVAL;
2822
2823	if (unlikely(!args.set_tid && args.set_tid_size > 0))
2824		return -EINVAL;
2825
2826	if (unlikely(args.set_tid && args.set_tid_size == 0))
2827		return -EINVAL;
2828
2829	/*
2830	 * Verify that higher 32bits of exit_signal are unset and that
2831	 * it is a valid signal
2832	 */
2833	if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2834		     !valid_signal(args.exit_signal)))
2835		return -EINVAL;
2836
2837	if ((args.flags & CLONE_INTO_CGROUP) &&
2838	    (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2839		return -EINVAL;
2840
2841	*kargs = (struct kernel_clone_args){
2842		.flags		= args.flags,
2843		.pidfd		= u64_to_user_ptr(args.pidfd),
2844		.child_tid	= u64_to_user_ptr(args.child_tid),
2845		.parent_tid	= u64_to_user_ptr(args.parent_tid),
2846		.exit_signal	= args.exit_signal,
2847		.stack		= args.stack,
2848		.stack_size	= args.stack_size,
2849		.tls		= args.tls,
2850		.set_tid_size	= args.set_tid_size,
2851		.cgroup		= args.cgroup,
2852	};
2853
2854	if (args.set_tid &&
2855		copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2856			(kargs->set_tid_size * sizeof(pid_t))))
2857		return -EFAULT;
2858
2859	kargs->set_tid = kset_tid;
2860
2861	return 0;
2862}
2863
2864/**
2865 * clone3_stack_valid - check and prepare stack
2866 * @kargs: kernel clone args
2867 *
2868 * Verify that the stack arguments userspace gave us are sane.
2869 * In addition, set the stack direction for userspace since it's easy for us to
2870 * determine.
2871 */
2872static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2873{
2874	if (kargs->stack == 0) {
2875		if (kargs->stack_size > 0)
2876			return false;
2877	} else {
2878		if (kargs->stack_size == 0)
2879			return false;
2880
2881		if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2882			return false;
2883
2884#if !defined(CONFIG_STACK_GROWSUP)
2885		kargs->stack += kargs->stack_size;
2886#endif
2887	}
2888
2889	return true;
2890}
2891
2892static bool clone3_args_valid(struct kernel_clone_args *kargs)
2893{
2894	/* Verify that no unknown flags are passed along. */
2895	if (kargs->flags &
2896	    ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2897		return false;
2898
2899	/*
2900	 * - make the CLONE_DETACHED bit reusable for clone3
2901	 * - make the CSIGNAL bits reusable for clone3
2902	 */
2903	if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
2904		return false;
2905
2906	if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2907	    (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2908		return false;
2909
2910	if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2911	    kargs->exit_signal)
2912		return false;
2913
2914	if (!clone3_stack_valid(kargs))
2915		return false;
2916
2917	return true;
2918}
2919
2920/**
2921 * sys_clone3 - create a new process with specific properties
2922 * @uargs: argument structure
2923 * @size:  size of @uargs
2924 *
2925 * clone3() is the extensible successor to clone()/clone2().
2926 * It takes a struct as argument that is versioned by its size.
2927 *
2928 * Return: On success, a positive PID for the child process.
2929 *         On error, a negative errno number.
2930 */
2931SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2932{
2933	int err;
2934
2935	struct kernel_clone_args kargs;
2936	pid_t set_tid[MAX_PID_NS_LEVEL];
2937
2938#ifdef __ARCH_BROKEN_SYS_CLONE3
2939#warning clone3() entry point is missing, please fix
2940	return -ENOSYS;
2941#endif
2942
2943	kargs.set_tid = set_tid;
2944
2945	err = copy_clone_args_from_user(&kargs, uargs, size);
2946	if (err)
2947		return err;
2948
2949	if (!clone3_args_valid(&kargs))
2950		return -EINVAL;
2951
2952	return kernel_clone(&kargs);
2953}
2954
2955void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2956{
2957	struct task_struct *leader, *parent, *child;
2958	int res;
2959
2960	read_lock(&tasklist_lock);
2961	leader = top = top->group_leader;
2962down:
2963	for_each_thread(leader, parent) {
2964		list_for_each_entry(child, &parent->children, sibling) {
2965			res = visitor(child, data);
2966			if (res) {
2967				if (res < 0)
2968					goto out;
2969				leader = child;
2970				goto down;
2971			}
2972up:
2973			;
2974		}
2975	}
2976
2977	if (leader != top) {
2978		child = leader;
2979		parent = child->real_parent;
2980		leader = parent->group_leader;
2981		goto up;
2982	}
2983out:
2984	read_unlock(&tasklist_lock);
2985}
2986
2987#ifndef ARCH_MIN_MMSTRUCT_ALIGN
2988#define ARCH_MIN_MMSTRUCT_ALIGN 0
2989#endif
2990
2991static void sighand_ctor(void *data)
2992{
2993	struct sighand_struct *sighand = data;
2994
2995	spin_lock_init(&sighand->siglock);
2996	init_waitqueue_head(&sighand->signalfd_wqh);
2997}
2998
2999void __init mm_cache_init(void)
3000{
3001	unsigned int mm_size;
3002
3003	/*
3004	 * The mm_cpumask is located at the end of mm_struct, and is
3005	 * dynamically sized based on the maximum CPU number this system
3006	 * can have, taking hotplug into account (nr_cpu_ids).
3007	 */
3008	mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
3009
3010	mm_cachep = kmem_cache_create_usercopy("mm_struct",
3011			mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3012			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3013			offsetof(struct mm_struct, saved_auxv),
3014			sizeof_field(struct mm_struct, saved_auxv),
3015			NULL);
3016}
3017
3018void __init proc_caches_init(void)
3019{
3020	sighand_cachep = kmem_cache_create("sighand_cache",
3021			sizeof(struct sighand_struct), 0,
3022			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3023			SLAB_ACCOUNT, sighand_ctor);
3024	signal_cachep = kmem_cache_create("signal_cache",
3025			sizeof(struct signal_struct), 0,
3026			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3027			NULL);
3028	files_cachep = kmem_cache_create("files_cache",
3029			sizeof(struct files_struct), 0,
3030			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3031			NULL);
3032	fs_cachep = kmem_cache_create("fs_cache",
3033			sizeof(struct fs_struct), 0,
3034			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3035			NULL);
3036	mmap_init();
3037	nsproxy_cache_init();
3038}
3039
3040/*
3041 * Check constraints on flags passed to the unshare system call.
3042 */
3043static int check_unshare_flags(unsigned long unshare_flags)
3044{
3045	if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3046				CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3047				CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3048				CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3049				CLONE_NEWTIME))
3050		return -EINVAL;
3051	/*
3052	 * Not implemented, but pretend it works if there is nothing
3053	 * to unshare.  Note that unsharing the address space or the
3054	 * signal handlers also need to unshare the signal queues (aka
3055	 * CLONE_THREAD).
3056	 */
3057	if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3058		if (!thread_group_empty(current))
3059			return -EINVAL;
3060	}
3061	if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3062		if (refcount_read(&current->sighand->count) > 1)
3063			return -EINVAL;
3064	}
3065	if (unshare_flags & CLONE_VM) {
3066		if (!current_is_single_threaded())
3067			return -EINVAL;
3068	}
3069
3070	return 0;
3071}
3072
3073/*
3074 * Unshare the filesystem structure if it is being shared
3075 */
3076static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3077{
3078	struct fs_struct *fs = current->fs;
3079
3080	if (!(unshare_flags & CLONE_FS) || !fs)
3081		return 0;
3082
3083	/* don't need lock here; in the worst case we'll do useless copy */
3084	if (!(unshare_flags & CLONE_NEWNS) && fs->users == 1)
3085		return 0;
3086
3087	*new_fsp = copy_fs_struct(fs);
3088	if (!*new_fsp)
3089		return -ENOMEM;
3090
3091	return 0;
3092}
3093
3094/*
3095 * Unshare file descriptor table if it is being shared
3096 */
3097static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
3098{
3099	struct files_struct *fd = current->files;
3100
3101	if ((unshare_flags & CLONE_FILES) &&
3102	    (fd && atomic_read(&fd->count) > 1)) {
3103		fd = dup_fd(fd, NULL);
3104		if (IS_ERR(fd))
3105			return PTR_ERR(fd);
3106		*new_fdp = fd;
3107	}
3108
3109	return 0;
3110}
3111
3112/*
3113 * unshare allows a process to 'unshare' part of the process
3114 * context which was originally shared using clone.  copy_*
3115 * functions used by kernel_clone() cannot be used here directly
3116 * because they modify an inactive task_struct that is being
3117 * constructed. Here we are modifying the current, active,
3118 * task_struct.
3119 */
3120int ksys_unshare(unsigned long unshare_flags)
3121{
3122	struct fs_struct *fs, *new_fs = NULL;
3123	struct files_struct *new_fd = NULL;
3124	struct cred *new_cred = NULL;
3125	struct nsproxy *new_nsproxy = NULL;
3126	int do_sysvsem = 0;
3127	int err;
3128
3129	/*
3130	 * If unsharing a user namespace must also unshare the thread group
3131	 * and unshare the filesystem root and working directories.
3132	 */
3133	if (unshare_flags & CLONE_NEWUSER)
3134		unshare_flags |= CLONE_THREAD | CLONE_FS;
3135	/*
3136	 * If unsharing vm, must also unshare signal handlers.
3137	 */
3138	if (unshare_flags & CLONE_VM)
3139		unshare_flags |= CLONE_SIGHAND;
3140	/*
3141	 * If unsharing a signal handlers, must also unshare the signal queues.
3142	 */
3143	if (unshare_flags & CLONE_SIGHAND)
3144		unshare_flags |= CLONE_THREAD;
3145	/*
3146	 * If unsharing namespace, must also unshare filesystem information.
3147	 */
3148	if (unshare_flags & CLONE_NEWNS)
3149		unshare_flags |= CLONE_FS;
3150
3151	err = check_unshare_flags(unshare_flags);
3152	if (err)
3153		goto bad_unshare_out;
3154	/*
3155	 * CLONE_NEWIPC must also detach from the undolist: after switching
3156	 * to a new ipc namespace, the semaphore arrays from the old
3157	 * namespace are unreachable.
3158	 */
3159	if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3160		do_sysvsem = 1;
3161	err = unshare_fs(unshare_flags, &new_fs);
3162	if (err)
3163		goto bad_unshare_out;
3164	err = unshare_fd(unshare_flags, &new_fd);
3165	if (err)
3166		goto bad_unshare_cleanup_fs;
3167	err = unshare_userns(unshare_flags, &new_cred);
3168	if (err)
3169		goto bad_unshare_cleanup_fd;
3170	err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3171					 new_cred, new_fs);
3172	if (err)
3173		goto bad_unshare_cleanup_cred;
3174
3175	if (new_cred) {
3176		err = set_cred_ucounts(new_cred);
3177		if (err)
3178			goto bad_unshare_cleanup_cred;
3179	}
3180
3181	if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3182		if (do_sysvsem) {
3183			/*
3184			 * CLONE_SYSVSEM is equivalent to sys_exit().
3185			 */
3186			exit_sem(current);
3187		}
3188		if (unshare_flags & CLONE_NEWIPC) {
3189			/* Orphan segments in old ns (see sem above). */
3190			exit_shm(current);
3191			shm_init_task(current);
3192		}
3193
3194		if (new_nsproxy)
3195			switch_task_namespaces(current, new_nsproxy);
3196
3197		task_lock(current);
3198
3199		if (new_fs) {
3200			fs = current->fs;
3201			read_seqlock_excl(&fs->seq);
3202			current->fs = new_fs;
3203			if (--fs->users)
3204				new_fs = NULL;
3205			else
3206				new_fs = fs;
3207			read_sequnlock_excl(&fs->seq);
3208		}
3209
3210		if (new_fd)
3211			swap(current->files, new_fd);
3212
3213		task_unlock(current);
3214
3215		if (new_cred) {
3216			/* Install the new user namespace */
3217			commit_creds(new_cred);
3218			new_cred = NULL;
3219		}
3220	}
3221
3222	perf_event_namespaces(current);
3223
3224bad_unshare_cleanup_cred:
3225	if (new_cred)
3226		put_cred(new_cred);
3227bad_unshare_cleanup_fd:
3228	if (new_fd)
3229		put_files_struct(new_fd);
3230
3231bad_unshare_cleanup_fs:
3232	if (new_fs)
3233		free_fs_struct(new_fs);
3234
3235bad_unshare_out:
3236	return err;
3237}
3238
3239SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3240{
3241	return ksys_unshare(unshare_flags);
3242}
3243
3244/*
3245 *	Helper to unshare the files of the current task.
3246 *	We don't want to expose copy_files internals to
3247 *	the exec layer of the kernel.
3248 */
3249
3250int unshare_files(void)
3251{
3252	struct task_struct *task = current;
3253	struct files_struct *old, *copy = NULL;
3254	int error;
3255
3256	error = unshare_fd(CLONE_FILES, &copy);
3257	if (error || !copy)
3258		return error;
3259
3260	old = task->files;
3261	task_lock(task);
3262	task->files = copy;
3263	task_unlock(task);
3264	put_files_struct(old);
3265	return 0;
3266}
3267
3268static int sysctl_max_threads(const struct ctl_table *table, int write,
3269		       void *buffer, size_t *lenp, loff_t *ppos)
3270{
3271	struct ctl_table t;
3272	int ret;
3273	int threads = max_threads;
3274	int min = 1;
3275	int max = MAX_THREADS;
3276
3277	t = *table;
3278	t.data = &threads;
3279	t.extra1 = &min;
3280	t.extra2 = &max;
3281
3282	ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3283	if (ret || !write)
3284		return ret;
3285
3286	max_threads = threads;
3287
3288	return 0;
3289}
3290
3291static const struct ctl_table fork_sysctl_table[] = {
3292	{
3293		.procname	= "threads-max",
3294		.data		= NULL,
3295		.maxlen		= sizeof(int),
3296		.mode		= 0644,
3297		.proc_handler	= sysctl_max_threads,
3298	},
3299};
3300
3301static int __init init_fork_sysctl(void)
3302{
3303	register_sysctl_init("kernel", fork_sysctl_table);
3304	return 0;
3305}
3306
3307subsys_initcall(init_fork_sysctl);