Linux Audio

Check our new training course

Loading...
v6.8
   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/coredump.h>
  20#include <linux/sched/user.h>
  21#include <linux/sched/numa_balancing.h>
  22#include <linux/sched/stat.h>
  23#include <linux/sched/task.h>
  24#include <linux/sched/task_stack.h>
  25#include <linux/sched/cputime.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/nsproxy.h>
  48#include <linux/capability.h>
  49#include <linux/cpu.h>
  50#include <linux/cgroup.h>
  51#include <linux/security.h>
  52#include <linux/hugetlb.h>
  53#include <linux/seccomp.h>
  54#include <linux/swap.h>
  55#include <linux/syscalls.h>
  56#include <linux/syscall_user_dispatch.h>
  57#include <linux/jiffies.h>
  58#include <linux/futex.h>
  59#include <linux/compat.h>
  60#include <linux/kthread.h>
  61#include <linux/task_io_accounting_ops.h>
  62#include <linux/rcupdate.h>
  63#include <linux/ptrace.h>
  64#include <linux/mount.h>
  65#include <linux/audit.h>
  66#include <linux/memcontrol.h>
  67#include <linux/ftrace.h>
  68#include <linux/proc_fs.h>
  69#include <linux/profile.h>
  70#include <linux/rmap.h>
  71#include <linux/ksm.h>
  72#include <linux/acct.h>
  73#include <linux/userfaultfd_k.h>
  74#include <linux/tsacct_kern.h>
  75#include <linux/cn_proc.h>
  76#include <linux/freezer.h>
  77#include <linux/delayacct.h>
  78#include <linux/taskstats_kern.h>
 
  79#include <linux/tty.h>
 
  80#include <linux/fs_struct.h>
  81#include <linux/magic.h>
  82#include <linux/perf_event.h>
  83#include <linux/posix-timers.h>
  84#include <linux/user-return-notifier.h>
  85#include <linux/oom.h>
  86#include <linux/khugepaged.h>
  87#include <linux/signalfd.h>
  88#include <linux/uprobes.h>
  89#include <linux/aio.h>
  90#include <linux/compiler.h>
  91#include <linux/sysctl.h>
  92#include <linux/kcov.h>
  93#include <linux/livepatch.h>
  94#include <linux/thread_info.h>
  95#include <linux/stackleak.h>
  96#include <linux/kasan.h>
  97#include <linux/scs.h>
  98#include <linux/io_uring.h>
  99#include <linux/bpf.h>
 100#include <linux/stackprotector.h>
 101#include <linux/user_events.h>
 102#include <linux/iommu.h>
 103#include <linux/rseq.h>
 104
 105#include <asm/pgalloc.h>
 106#include <linux/uaccess.h>
 107#include <asm/mmu_context.h>
 108#include <asm/cacheflush.h>
 109#include <asm/tlbflush.h>
 110
 111#include <trace/events/sched.h>
 112
 113#define CREATE_TRACE_POINTS
 114#include <trace/events/task.h>
 115
 116/*
 117 * Minimum number of threads to boot the kernel
 118 */
 119#define MIN_THREADS 20
 120
 121/*
 122 * Maximum number of threads
 123 */
 124#define MAX_THREADS FUTEX_TID_MASK
 125
 126/*
 127 * Protected counters by write_lock_irq(&tasklist_lock)
 128 */
 129unsigned long total_forks;	/* Handle normal Linux uptimes. */
 130int nr_threads;			/* The idle threads do not count.. */
 131
 132static int max_threads;		/* tunable limit on nr_threads */
 133
 134#define NAMED_ARRAY_INDEX(x)	[x] = __stringify(x)
 135
 136static const char * const resident_page_types[] = {
 137	NAMED_ARRAY_INDEX(MM_FILEPAGES),
 138	NAMED_ARRAY_INDEX(MM_ANONPAGES),
 139	NAMED_ARRAY_INDEX(MM_SWAPENTS),
 140	NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
 141};
 142
 143DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 144
 145__cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 146
 147#ifdef CONFIG_PROVE_RCU
 148int lockdep_tasklist_lock_is_held(void)
 149{
 150	return lockdep_is_held(&tasklist_lock);
 151}
 152EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 153#endif /* #ifdef CONFIG_PROVE_RCU */
 154
 155int nr_processes(void)
 156{
 157	int cpu;
 158	int total = 0;
 159
 160	for_each_possible_cpu(cpu)
 161		total += per_cpu(process_counts, cpu);
 162
 163	return total;
 164}
 165
 166void __weak arch_release_task_struct(struct task_struct *tsk)
 167{
 168}
 169
 
 170static struct kmem_cache *task_struct_cachep;
 171
 172static inline struct task_struct *alloc_task_struct_node(int node)
 173{
 174	return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 175}
 176
 177static inline void free_task_struct(struct task_struct *tsk)
 178{
 179	kmem_cache_free(task_struct_cachep, tsk);
 180}
 
 
 
 181
 182/*
 183 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 184 * kmemcache based allocator.
 185 */
 186# if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 187
 188#  ifdef CONFIG_VMAP_STACK
 189/*
 190 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 191 * flush.  Try to minimize the number of calls by caching stacks.
 192 */
 193#define NR_CACHED_STACKS 2
 194static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 195
 196struct vm_stack {
 197	struct rcu_head rcu;
 198	struct vm_struct *stack_vm_area;
 199};
 200
 201static bool try_release_thread_stack_to_cache(struct vm_struct *vm)
 202{
 203	unsigned int i;
 204
 205	for (i = 0; i < NR_CACHED_STACKS; i++) {
 206		if (this_cpu_cmpxchg(cached_stacks[i], NULL, vm) != NULL)
 207			continue;
 208		return true;
 209	}
 210	return false;
 211}
 212
 213static void thread_stack_free_rcu(struct rcu_head *rh)
 214{
 215	struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
 216
 217	if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
 218		return;
 219
 220	vfree(vm_stack);
 221}
 222
 223static void thread_stack_delayed_free(struct task_struct *tsk)
 224{
 225	struct vm_stack *vm_stack = tsk->stack;
 226
 227	vm_stack->stack_vm_area = tsk->stack_vm_area;
 228	call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
 229}
 230
 231static int free_vm_stack_cache(unsigned int cpu)
 232{
 233	struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 234	int i;
 235
 236	for (i = 0; i < NR_CACHED_STACKS; i++) {
 237		struct vm_struct *vm_stack = cached_vm_stacks[i];
 238
 239		if (!vm_stack)
 240			continue;
 241
 242		vfree(vm_stack->addr);
 243		cached_vm_stacks[i] = NULL;
 244	}
 245
 246	return 0;
 247}
 
 248
 249static int memcg_charge_kernel_stack(struct vm_struct *vm)
 250{
 251	int i;
 252	int ret;
 253	int nr_charged = 0;
 254
 255	BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 256
 257	for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 258		ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0);
 259		if (ret)
 260			goto err;
 261		nr_charged++;
 262	}
 263	return 0;
 264err:
 265	for (i = 0; i < nr_charged; i++)
 266		memcg_kmem_uncharge_page(vm->pages[i], 0);
 267	return ret;
 268}
 269
 270static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 271{
 272	struct vm_struct *vm;
 273	void *stack;
 274	int i;
 275
 276	for (i = 0; i < NR_CACHED_STACKS; i++) {
 277		struct vm_struct *s;
 278
 279		s = this_cpu_xchg(cached_stacks[i], NULL);
 280
 281		if (!s)
 282			continue;
 283
 284		/* Reset stack metadata. */
 285		kasan_unpoison_range(s->addr, THREAD_SIZE);
 286
 287		stack = kasan_reset_tag(s->addr);
 288
 289		/* Clear stale pointers from reused stack. */
 290		memset(stack, 0, THREAD_SIZE);
 291
 292		if (memcg_charge_kernel_stack(s)) {
 293			vfree(s->addr);
 294			return -ENOMEM;
 295		}
 296
 297		tsk->stack_vm_area = s;
 298		tsk->stack = stack;
 299		return 0;
 300	}
 301
 302	/*
 303	 * Allocated stacks are cached and later reused by new threads,
 304	 * so memcg accounting is performed manually on assigning/releasing
 305	 * stacks to tasks. Drop __GFP_ACCOUNT.
 306	 */
 307	stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 308				     VMALLOC_START, VMALLOC_END,
 309				     THREADINFO_GFP & ~__GFP_ACCOUNT,
 310				     PAGE_KERNEL,
 311				     0, node, __builtin_return_address(0));
 312	if (!stack)
 313		return -ENOMEM;
 314
 315	vm = find_vm_area(stack);
 316	if (memcg_charge_kernel_stack(vm)) {
 317		vfree(stack);
 318		return -ENOMEM;
 319	}
 320	/*
 321	 * We can't call find_vm_area() in interrupt context, and
 322	 * free_thread_stack() can be called in interrupt context,
 323	 * so cache the vm_struct.
 324	 */
 325	tsk->stack_vm_area = vm;
 326	stack = kasan_reset_tag(stack);
 327	tsk->stack = stack;
 328	return 0;
 329}
 330
 331static void free_thread_stack(struct task_struct *tsk)
 332{
 333	if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
 334		thread_stack_delayed_free(tsk);
 335
 336	tsk->stack = NULL;
 337	tsk->stack_vm_area = NULL;
 338}
 339
 340#  else /* !CONFIG_VMAP_STACK */
 341
 342static void thread_stack_free_rcu(struct rcu_head *rh)
 343{
 344	__free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
 345}
 346
 347static void thread_stack_delayed_free(struct task_struct *tsk)
 348{
 349	struct rcu_head *rh = tsk->stack;
 350
 351	call_rcu(rh, thread_stack_free_rcu);
 352}
 353
 354static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 355{
 356	struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 357					     THREAD_SIZE_ORDER);
 358
 359	if (likely(page)) {
 360		tsk->stack = kasan_reset_tag(page_address(page));
 361		return 0;
 362	}
 363	return -ENOMEM;
 
 364}
 365
 366static void free_thread_stack(struct task_struct *tsk)
 367{
 368	thread_stack_delayed_free(tsk);
 369	tsk->stack = NULL;
 370}
 371
 372#  endif /* CONFIG_VMAP_STACK */
 373# else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
 374
 375static struct kmem_cache *thread_stack_cache;
 
 376
 377static void thread_stack_free_rcu(struct rcu_head *rh)
 378{
 379	kmem_cache_free(thread_stack_cache, rh);
 380}
 381
 382static void thread_stack_delayed_free(struct task_struct *tsk)
 383{
 384	struct rcu_head *rh = tsk->stack;
 
 
 
 
 385
 386	call_rcu(rh, thread_stack_free_rcu);
 387}
 
 
 388
 389static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 
 390{
 391	unsigned long *stack;
 392	stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 393	stack = kasan_reset_tag(stack);
 394	tsk->stack = stack;
 395	return stack ? 0 : -ENOMEM;
 396}
 397
 398static void free_thread_stack(struct task_struct *tsk)
 399{
 400	thread_stack_delayed_free(tsk);
 401	tsk->stack = NULL;
 402}
 403
 404void thread_stack_cache_init(void)
 405{
 406	thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 407					THREAD_SIZE, THREAD_SIZE, 0, 0,
 408					THREAD_SIZE, NULL);
 409	BUG_ON(thread_stack_cache == NULL);
 410}
 411
 412# endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
 413
 414/* SLAB cache for signal_struct structures (tsk->signal) */
 415static struct kmem_cache *signal_cachep;
 416
 417/* SLAB cache for sighand_struct structures (tsk->sighand) */
 418struct kmem_cache *sighand_cachep;
 419
 420/* SLAB cache for files_struct structures (tsk->files) */
 421struct kmem_cache *files_cachep;
 422
 423/* SLAB cache for fs_struct structures (tsk->fs) */
 424struct kmem_cache *fs_cachep;
 425
 426/* SLAB cache for vm_area_struct structures */
 427static struct kmem_cache *vm_area_cachep;
 428
 429/* SLAB cache for mm_struct structures (tsk->mm) */
 430static struct kmem_cache *mm_cachep;
 431
 432#ifdef CONFIG_PER_VMA_LOCK
 433
 434/* SLAB cache for vm_area_struct.lock */
 435static struct kmem_cache *vma_lock_cachep;
 436
 437static bool vma_lock_alloc(struct vm_area_struct *vma)
 438{
 439	vma->vm_lock = kmem_cache_alloc(vma_lock_cachep, GFP_KERNEL);
 440	if (!vma->vm_lock)
 441		return false;
 442
 443	init_rwsem(&vma->vm_lock->lock);
 444	vma->vm_lock_seq = -1;
 445
 446	return true;
 447}
 448
 449static inline void vma_lock_free(struct vm_area_struct *vma)
 450{
 451	kmem_cache_free(vma_lock_cachep, vma->vm_lock);
 452}
 453
 454#else /* CONFIG_PER_VMA_LOCK */
 455
 456static inline bool vma_lock_alloc(struct vm_area_struct *vma) { return true; }
 457static inline void vma_lock_free(struct vm_area_struct *vma) {}
 458
 459#endif /* CONFIG_PER_VMA_LOCK */
 460
 461struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 462{
 463	struct vm_area_struct *vma;
 464
 465	vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 466	if (!vma)
 467		return NULL;
 468
 469	vma_init(vma, mm);
 470	if (!vma_lock_alloc(vma)) {
 471		kmem_cache_free(vm_area_cachep, vma);
 472		return NULL;
 473	}
 474
 475	return vma;
 476}
 477
 478struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 479{
 480	struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 481
 482	if (!new)
 483		return NULL;
 484
 485	ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
 486	ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
 487	/*
 488	 * orig->shared.rb may be modified concurrently, but the clone
 489	 * will be reinitialized.
 490	 */
 491	data_race(memcpy(new, orig, sizeof(*new)));
 492	if (!vma_lock_alloc(new)) {
 493		kmem_cache_free(vm_area_cachep, new);
 494		return NULL;
 495	}
 496	INIT_LIST_HEAD(&new->anon_vma_chain);
 497	vma_numab_state_init(new);
 498	dup_anon_vma_name(orig, new);
 499
 500	return new;
 501}
 502
 503void __vm_area_free(struct vm_area_struct *vma)
 504{
 505	vma_numab_state_free(vma);
 506	free_anon_vma_name(vma);
 507	vma_lock_free(vma);
 508	kmem_cache_free(vm_area_cachep, vma);
 509}
 510
 511#ifdef CONFIG_PER_VMA_LOCK
 512static void vm_area_free_rcu_cb(struct rcu_head *head)
 513{
 514	struct vm_area_struct *vma = container_of(head, struct vm_area_struct,
 515						  vm_rcu);
 516
 517	/* The vma should not be locked while being destroyed. */
 518	VM_BUG_ON_VMA(rwsem_is_locked(&vma->vm_lock->lock), vma);
 519	__vm_area_free(vma);
 520}
 521#endif
 522
 523void vm_area_free(struct vm_area_struct *vma)
 524{
 525#ifdef CONFIG_PER_VMA_LOCK
 526	call_rcu(&vma->vm_rcu, vm_area_free_rcu_cb);
 527#else
 528	__vm_area_free(vma);
 529#endif
 530}
 531
 532static void account_kernel_stack(struct task_struct *tsk, int account)
 533{
 534	if (IS_ENABLED(CONFIG_VMAP_STACK)) {
 535		struct vm_struct *vm = task_stack_vm_area(tsk);
 536		int i;
 537
 538		for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 539			mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
 540					      account * (PAGE_SIZE / 1024));
 541	} else {
 542		void *stack = task_stack_page(tsk);
 543
 544		/* All stack pages are in the same node. */
 545		mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
 
 
 
 
 546				      account * (THREAD_SIZE / 1024));
 547	}
 548}
 549
 550void exit_task_stack_account(struct task_struct *tsk)
 551{
 552	account_kernel_stack(tsk, -1);
 
 
 
 
 553
 554	if (IS_ENABLED(CONFIG_VMAP_STACK)) {
 555		struct vm_struct *vm;
 556		int i;
 557
 558		vm = task_stack_vm_area(tsk);
 559		for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 560			memcg_kmem_uncharge_page(vm->pages[i], 0);
 
 
 
 
 
 
 
 
 
 
 561	}
 
 
 562}
 563
 564static void release_task_stack(struct task_struct *tsk)
 565{
 566	if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
 567		return;  /* Better to leak the stack than to free prematurely */
 568
 
 569	free_thread_stack(tsk);
 
 
 
 
 570}
 571
 572#ifdef CONFIG_THREAD_INFO_IN_TASK
 573void put_task_stack(struct task_struct *tsk)
 574{
 575	if (refcount_dec_and_test(&tsk->stack_refcount))
 576		release_task_stack(tsk);
 577}
 578#endif
 579
 580void free_task(struct task_struct *tsk)
 581{
 582#ifdef CONFIG_SECCOMP
 583	WARN_ON_ONCE(tsk->seccomp.filter);
 584#endif
 585	release_user_cpus_ptr(tsk);
 586	scs_release(tsk);
 587
 588#ifndef CONFIG_THREAD_INFO_IN_TASK
 589	/*
 590	 * The task is finally done with both the stack and thread_info,
 591	 * so free both.
 592	 */
 593	release_task_stack(tsk);
 594#else
 595	/*
 596	 * If the task had a separate stack allocation, it should be gone
 597	 * by now.
 598	 */
 599	WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
 600#endif
 601	rt_mutex_debug_task_free(tsk);
 602	ftrace_graph_exit_task(tsk);
 603	arch_release_task_struct(tsk);
 604	if (tsk->flags & PF_KTHREAD)
 605		free_kthread_struct(tsk);
 606	bpf_task_storage_free(tsk);
 607	free_task_struct(tsk);
 608}
 609EXPORT_SYMBOL(free_task);
 610
 611static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
 612{
 613	struct file *exe_file;
 614
 615	exe_file = get_mm_exe_file(oldmm);
 616	RCU_INIT_POINTER(mm->exe_file, exe_file);
 617	/*
 618	 * We depend on the oldmm having properly denied write access to the
 619	 * exe_file already.
 620	 */
 621	if (exe_file && deny_write_access(exe_file))
 622		pr_warn_once("deny_write_access() failed in %s\n", __func__);
 623}
 624
 625#ifdef CONFIG_MMU
 626static __latent_entropy int dup_mmap(struct mm_struct *mm,
 627					struct mm_struct *oldmm)
 628{
 629	struct vm_area_struct *mpnt, *tmp;
 
 630	int retval;
 631	unsigned long charge = 0;
 632	LIST_HEAD(uf);
 633	VMA_ITERATOR(vmi, mm, 0);
 634
 635	uprobe_start_dup_mmap();
 636	if (mmap_write_lock_killable(oldmm)) {
 637		retval = -EINTR;
 638		goto fail_uprobe_end;
 639	}
 640	flush_cache_dup_mm(oldmm);
 641	uprobe_dup_mmap(oldmm, mm);
 642	/*
 643	 * Not linked in yet - no deadlock potential:
 644	 */
 645	mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
 646
 647	/* No ordering required: file already has been exposed. */
 648	dup_mm_exe_file(mm, oldmm);
 649
 650	mm->total_vm = oldmm->total_vm;
 651	mm->data_vm = oldmm->data_vm;
 652	mm->exec_vm = oldmm->exec_vm;
 653	mm->stack_vm = oldmm->stack_vm;
 654
 
 
 
 655	retval = ksm_fork(mm, oldmm);
 656	if (retval)
 657		goto out;
 658	khugepaged_fork(mm, oldmm);
 659
 660	/* Use __mt_dup() to efficiently build an identical maple tree. */
 661	retval = __mt_dup(&oldmm->mm_mt, &mm->mm_mt, GFP_KERNEL);
 662	if (unlikely(retval))
 663		goto out;
 664
 665	mt_clear_in_rcu(vmi.mas.tree);
 666	for_each_vma(vmi, mpnt) {
 667		struct file *file;
 668
 669		vma_start_write(mpnt);
 670		if (mpnt->vm_flags & VM_DONTCOPY) {
 671			retval = vma_iter_clear_gfp(&vmi, mpnt->vm_start,
 672						    mpnt->vm_end, GFP_KERNEL);
 673			if (retval)
 674				goto loop_out;
 675
 676			vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 677			continue;
 678		}
 679		charge = 0;
 680		/*
 681		 * Don't duplicate many vmas if we've been oom-killed (for
 682		 * example)
 683		 */
 684		if (fatal_signal_pending(current)) {
 685			retval = -EINTR;
 686			goto loop_out;
 687		}
 688		if (mpnt->vm_flags & VM_ACCOUNT) {
 689			unsigned long len = vma_pages(mpnt);
 690
 691			if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 692				goto fail_nomem;
 693			charge = len;
 694		}
 695		tmp = vm_area_dup(mpnt);
 696		if (!tmp)
 697			goto fail_nomem;
 698		retval = vma_dup_policy(mpnt, tmp);
 699		if (retval)
 700			goto fail_nomem_policy;
 701		tmp->vm_mm = mm;
 702		retval = dup_userfaultfd(tmp, &uf);
 703		if (retval)
 704			goto fail_nomem_anon_vma_fork;
 705		if (tmp->vm_flags & VM_WIPEONFORK) {
 706			/*
 707			 * VM_WIPEONFORK gets a clean slate in the child.
 708			 * Don't prepare anon_vma until fault since we don't
 709			 * copy page for current vma.
 710			 */
 711			tmp->anon_vma = NULL;
 712		} else if (anon_vma_fork(tmp, mpnt))
 713			goto fail_nomem_anon_vma_fork;
 714		vm_flags_clear(tmp, VM_LOCKED_MASK);
 715		file = tmp->vm_file;
 716		if (file) {
 
 717			struct address_space *mapping = file->f_mapping;
 718
 719			get_file(file);
 
 
 720			i_mmap_lock_write(mapping);
 721			if (vma_is_shared_maywrite(tmp))
 722				mapping_allow_writable(mapping);
 723			flush_dcache_mmap_lock(mapping);
 724			/* insert tmp into the share list, just after mpnt */
 725			vma_interval_tree_insert_after(tmp, mpnt,
 726					&mapping->i_mmap);
 727			flush_dcache_mmap_unlock(mapping);
 728			i_mmap_unlock_write(mapping);
 729		}
 730
 731		/*
 732		 * Copy/update hugetlb private vma information.
 
 
 733		 */
 734		if (is_vm_hugetlb_page(tmp))
 735			hugetlb_dup_vma_private(tmp);
 736
 737		/*
 738		 * Link the vma into the MT. After using __mt_dup(), memory
 739		 * allocation is not necessary here, so it cannot fail.
 740		 */
 741		vma_iter_bulk_store(&vmi, tmp);
 
 
 
 
 
 
 
 742
 743		mm->map_count++;
 744		if (!(tmp->vm_flags & VM_WIPEONFORK))
 745			retval = copy_page_range(tmp, mpnt);
 746
 747		if (tmp->vm_ops && tmp->vm_ops->open)
 748			tmp->vm_ops->open(tmp);
 749
 750		if (retval) {
 751			mpnt = vma_next(&vmi);
 752			goto loop_out;
 753		}
 754	}
 755	/* a new mm has just been created */
 756	retval = arch_dup_mmap(oldmm, mm);
 757loop_out:
 758	vma_iter_free(&vmi);
 759	if (!retval) {
 760		mt_set_in_rcu(vmi.mas.tree);
 761	} else if (mpnt) {
 762		/*
 763		 * The entire maple tree has already been duplicated. If the
 764		 * mmap duplication fails, mark the failure point with
 765		 * XA_ZERO_ENTRY. In exit_mmap(), if this marker is encountered,
 766		 * stop releasing VMAs that have not been duplicated after this
 767		 * point.
 768		 */
 769		mas_set_range(&vmi.mas, mpnt->vm_start, mpnt->vm_end - 1);
 770		mas_store(&vmi.mas, XA_ZERO_ENTRY);
 771	}
 772out:
 773	mmap_write_unlock(mm);
 774	flush_tlb_mm(oldmm);
 775	mmap_write_unlock(oldmm);
 776	dup_userfaultfd_complete(&uf);
 777fail_uprobe_end:
 778	uprobe_end_dup_mmap();
 779	return retval;
 780
 781fail_nomem_anon_vma_fork:
 782	mpol_put(vma_policy(tmp));
 783fail_nomem_policy:
 784	vm_area_free(tmp);
 785fail_nomem:
 786	retval = -ENOMEM;
 787	vm_unacct_memory(charge);
 788	goto loop_out;
 789}
 790
 791static inline int mm_alloc_pgd(struct mm_struct *mm)
 792{
 793	mm->pgd = pgd_alloc(mm);
 794	if (unlikely(!mm->pgd))
 795		return -ENOMEM;
 796	return 0;
 797}
 798
 799static inline void mm_free_pgd(struct mm_struct *mm)
 800{
 801	pgd_free(mm, mm->pgd);
 802}
 803#else
 804static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 805{
 806	mmap_write_lock(oldmm);
 807	dup_mm_exe_file(mm, oldmm);
 808	mmap_write_unlock(oldmm);
 809	return 0;
 810}
 811#define mm_alloc_pgd(mm)	(0)
 812#define mm_free_pgd(mm)
 813#endif /* CONFIG_MMU */
 814
 815static void check_mm(struct mm_struct *mm)
 816{
 817	int i;
 818
 819	BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
 820			 "Please make sure 'struct resident_page_types[]' is updated as well");
 821
 822	for (i = 0; i < NR_MM_COUNTERS; i++) {
 823		long x = percpu_counter_sum(&mm->rss_stat[i]);
 824
 825		if (unlikely(x))
 826			pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
 827				 mm, resident_page_types[i], x);
 828	}
 829
 830	if (mm_pgtables_bytes(mm))
 831		pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 832				mm_pgtables_bytes(mm));
 833
 834#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 835	VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 836#endif
 837}
 838
 839#define allocate_mm()	(kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 840#define free_mm(mm)	(kmem_cache_free(mm_cachep, (mm)))
 841
 842static void do_check_lazy_tlb(void *arg)
 843{
 844	struct mm_struct *mm = arg;
 845
 846	WARN_ON_ONCE(current->active_mm == mm);
 847}
 848
 849static void do_shoot_lazy_tlb(void *arg)
 850{
 851	struct mm_struct *mm = arg;
 852
 853	if (current->active_mm == mm) {
 854		WARN_ON_ONCE(current->mm);
 855		current->active_mm = &init_mm;
 856		switch_mm(mm, &init_mm, current);
 857	}
 858}
 859
 860static void cleanup_lazy_tlbs(struct mm_struct *mm)
 861{
 862	if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) {
 863		/*
 864		 * In this case, lazy tlb mms are refounted and would not reach
 865		 * __mmdrop until all CPUs have switched away and mmdrop()ed.
 866		 */
 867		return;
 868	}
 869
 870	/*
 871	 * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
 872	 * requires lazy mm users to switch to another mm when the refcount
 873	 * drops to zero, before the mm is freed. This requires IPIs here to
 874	 * switch kernel threads to init_mm.
 875	 *
 876	 * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
 877	 * switch with the final userspace teardown TLB flush which leaves the
 878	 * mm lazy on this CPU but no others, reducing the need for additional
 879	 * IPIs here. There are cases where a final IPI is still required here,
 880	 * such as the final mmdrop being performed on a different CPU than the
 881	 * one exiting, or kernel threads using the mm when userspace exits.
 882	 *
 883	 * IPI overheads have not found to be expensive, but they could be
 884	 * reduced in a number of possible ways, for example (roughly
 885	 * increasing order of complexity):
 886	 * - The last lazy reference created by exit_mm() could instead switch
 887	 *   to init_mm, however it's probable this will run on the same CPU
 888	 *   immediately afterwards, so this may not reduce IPIs much.
 889	 * - A batch of mms requiring IPIs could be gathered and freed at once.
 890	 * - CPUs store active_mm where it can be remotely checked without a
 891	 *   lock, to filter out false-positives in the cpumask.
 892	 * - After mm_users or mm_count reaches zero, switching away from the
 893	 *   mm could clear mm_cpumask to reduce some IPIs, perhaps together
 894	 *   with some batching or delaying of the final IPIs.
 895	 * - A delayed freeing and RCU-like quiescing sequence based on mm
 896	 *   switching to avoid IPIs completely.
 897	 */
 898	on_each_cpu_mask(mm_cpumask(mm), do_shoot_lazy_tlb, (void *)mm, 1);
 899	if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES))
 900		on_each_cpu(do_check_lazy_tlb, (void *)mm, 1);
 901}
 902
 903/*
 904 * Called when the last reference to the mm
 905 * is dropped: either by a lazy thread or by
 906 * mmput. Free the page directory and the mm.
 907 */
 908void __mmdrop(struct mm_struct *mm)
 909{
 910	BUG_ON(mm == &init_mm);
 911	WARN_ON_ONCE(mm == current->mm);
 912
 913	/* Ensure no CPUs are using this as their lazy tlb mm */
 914	cleanup_lazy_tlbs(mm);
 915
 916	WARN_ON_ONCE(mm == current->active_mm);
 917	mm_free_pgd(mm);
 918	destroy_context(mm);
 919	mmu_notifier_subscriptions_destroy(mm);
 920	check_mm(mm);
 921	put_user_ns(mm->user_ns);
 922	mm_pasid_drop(mm);
 923	mm_destroy_cid(mm);
 924	percpu_counter_destroy_many(mm->rss_stat, NR_MM_COUNTERS);
 925
 926	free_mm(mm);
 927}
 928EXPORT_SYMBOL_GPL(__mmdrop);
 929
 930static void mmdrop_async_fn(struct work_struct *work)
 931{
 932	struct mm_struct *mm;
 933
 934	mm = container_of(work, struct mm_struct, async_put_work);
 935	__mmdrop(mm);
 936}
 937
 938static void mmdrop_async(struct mm_struct *mm)
 939{
 940	if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 941		INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 942		schedule_work(&mm->async_put_work);
 943	}
 944}
 945
 946static inline void free_signal_struct(struct signal_struct *sig)
 947{
 948	taskstats_tgid_free(sig);
 949	sched_autogroup_exit(sig);
 950	/*
 951	 * __mmdrop is not safe to call from softirq context on x86 due to
 952	 * pgd_dtor so postpone it to the async context
 953	 */
 954	if (sig->oom_mm)
 955		mmdrop_async(sig->oom_mm);
 956	kmem_cache_free(signal_cachep, sig);
 957}
 958
 959static inline void put_signal_struct(struct signal_struct *sig)
 960{
 961	if (refcount_dec_and_test(&sig->sigcnt))
 962		free_signal_struct(sig);
 963}
 964
 965void __put_task_struct(struct task_struct *tsk)
 966{
 967	WARN_ON(!tsk->exit_state);
 968	WARN_ON(refcount_read(&tsk->usage));
 969	WARN_ON(tsk == current);
 970
 971	io_uring_free(tsk);
 972	cgroup_free(tsk);
 973	task_numa_free(tsk, true);
 974	security_task_free(tsk);
 975	exit_creds(tsk);
 976	delayacct_tsk_free(tsk);
 977	put_signal_struct(tsk->signal);
 978	sched_core_free(tsk);
 979	free_task(tsk);
 980}
 981EXPORT_SYMBOL_GPL(__put_task_struct);
 982
 983void __put_task_struct_rcu_cb(struct rcu_head *rhp)
 984{
 985	struct task_struct *task = container_of(rhp, struct task_struct, rcu);
 986
 987	__put_task_struct(task);
 988}
 989EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb);
 990
 991void __init __weak arch_task_cache_init(void) { }
 992
 993/*
 994 * set_max_threads
 995 */
 996static void set_max_threads(unsigned int max_threads_suggested)
 997{
 998	u64 threads;
 999	unsigned long nr_pages = totalram_pages();
1000
1001	/*
1002	 * The number of threads shall be limited such that the thread
1003	 * structures may only consume a small part of the available memory.
1004	 */
1005	if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
1006		threads = MAX_THREADS;
1007	else
1008		threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
1009				    (u64) THREAD_SIZE * 8UL);
1010
1011	if (threads > max_threads_suggested)
1012		threads = max_threads_suggested;
1013
1014	max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
1015}
1016
1017#ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
1018/* Initialized by the architecture: */
1019int arch_task_struct_size __read_mostly;
1020#endif
1021
 
1022static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
1023{
1024	/* Fetch thread_struct whitelist for the architecture. */
1025	arch_thread_struct_whitelist(offset, size);
1026
1027	/*
1028	 * Handle zero-sized whitelist or empty thread_struct, otherwise
1029	 * adjust offset to position of thread_struct in task_struct.
1030	 */
1031	if (unlikely(*size == 0))
1032		*offset = 0;
1033	else
1034		*offset += offsetof(struct task_struct, thread);
1035}
 
1036
1037void __init fork_init(void)
1038{
1039	int i;
 
1040#ifndef ARCH_MIN_TASKALIGN
1041#define ARCH_MIN_TASKALIGN	0
1042#endif
1043	int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
1044	unsigned long useroffset, usersize;
1045
1046	/* create a slab on which task_structs can be allocated */
1047	task_struct_whitelist(&useroffset, &usersize);
1048	task_struct_cachep = kmem_cache_create_usercopy("task_struct",
1049			arch_task_struct_size, align,
1050			SLAB_PANIC|SLAB_ACCOUNT,
1051			useroffset, usersize, NULL);
 
1052
1053	/* do the arch specific task caches init */
1054	arch_task_cache_init();
1055
1056	set_max_threads(MAX_THREADS);
1057
1058	init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
1059	init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
1060	init_task.signal->rlim[RLIMIT_SIGPENDING] =
1061		init_task.signal->rlim[RLIMIT_NPROC];
1062
1063	for (i = 0; i < UCOUNT_COUNTS; i++)
1064		init_user_ns.ucount_max[i] = max_threads/2;
1065
1066	set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC,      RLIM_INFINITY);
1067	set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE,   RLIM_INFINITY);
1068	set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
1069	set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK,    RLIM_INFINITY);
1070
1071#ifdef CONFIG_VMAP_STACK
1072	cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
1073			  NULL, free_vm_stack_cache);
1074#endif
1075
1076	scs_init();
1077
1078	lockdep_init_task(&init_task);
1079	uprobes_init();
1080}
1081
1082int __weak arch_dup_task_struct(struct task_struct *dst,
1083					       struct task_struct *src)
1084{
1085	*dst = *src;
1086	return 0;
1087}
1088
1089void set_task_stack_end_magic(struct task_struct *tsk)
1090{
1091	unsigned long *stackend;
1092
1093	stackend = end_of_stack(tsk);
1094	*stackend = STACK_END_MAGIC;	/* for overflow detection */
1095}
1096
1097static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
1098{
1099	struct task_struct *tsk;
 
 
1100	int err;
1101
1102	if (node == NUMA_NO_NODE)
1103		node = tsk_fork_get_node(orig);
1104	tsk = alloc_task_struct_node(node);
1105	if (!tsk)
1106		return NULL;
1107
1108	err = arch_dup_task_struct(tsk, orig);
1109	if (err)
1110		goto free_tsk;
1111
1112	err = alloc_thread_stack_node(tsk, node);
1113	if (err)
1114		goto free_tsk;
1115
 
 
 
 
 
 
 
 
 
 
 
 
 
1116#ifdef CONFIG_THREAD_INFO_IN_TASK
1117	refcount_set(&tsk->stack_refcount, 1);
1118#endif
1119	account_kernel_stack(tsk, 1);
 
 
1120
1121	err = scs_prepare(tsk, node);
1122	if (err)
1123		goto free_stack;
1124
1125#ifdef CONFIG_SECCOMP
1126	/*
1127	 * We must handle setting up seccomp filters once we're under
1128	 * the sighand lock in case orig has changed between now and
1129	 * then. Until then, filter must be NULL to avoid messing up
1130	 * the usage counts on the error path calling free_task.
1131	 */
1132	tsk->seccomp.filter = NULL;
1133#endif
1134
1135	setup_thread_stack(tsk, orig);
1136	clear_user_return_notifier(tsk);
1137	clear_tsk_need_resched(tsk);
1138	set_task_stack_end_magic(tsk);
1139	clear_syscall_work_syscall_user_dispatch(tsk);
1140
1141#ifdef CONFIG_STACKPROTECTOR
1142	tsk->stack_canary = get_random_canary();
1143#endif
1144	if (orig->cpus_ptr == &orig->cpus_mask)
1145		tsk->cpus_ptr = &tsk->cpus_mask;
1146	dup_user_cpus_ptr(tsk, orig, node);
1147
1148	/*
1149	 * One for the user space visible state that goes away when reaped.
1150	 * One for the scheduler.
1151	 */
1152	refcount_set(&tsk->rcu_users, 2);
1153	/* One for the rcu users */
1154	refcount_set(&tsk->usage, 1);
1155#ifdef CONFIG_BLK_DEV_IO_TRACE
1156	tsk->btrace_seq = 0;
1157#endif
1158	tsk->splice_pipe = NULL;
1159	tsk->task_frag.page = NULL;
1160	tsk->wake_q.next = NULL;
1161	tsk->worker_private = NULL;
 
1162
1163	kcov_task_init(tsk);
1164	kmsan_task_create(tsk);
1165	kmap_local_fork(tsk);
1166
1167#ifdef CONFIG_FAULT_INJECTION
1168	tsk->fail_nth = 0;
1169#endif
1170
1171#ifdef CONFIG_BLK_CGROUP
1172	tsk->throttle_disk = NULL;
1173	tsk->use_memdelay = 0;
1174#endif
1175
1176#ifdef CONFIG_ARCH_HAS_CPU_PASID
1177	tsk->pasid_activated = 0;
1178#endif
1179
1180#ifdef CONFIG_MEMCG
1181	tsk->active_memcg = NULL;
1182#endif
1183
1184#ifdef CONFIG_CPU_SUP_INTEL
1185	tsk->reported_split_lock = 0;
1186#endif
1187
1188#ifdef CONFIG_SCHED_MM_CID
1189	tsk->mm_cid = -1;
1190	tsk->last_mm_cid = -1;
1191	tsk->mm_cid_active = 0;
1192	tsk->migrate_from_cpu = -1;
1193#endif
1194	return tsk;
1195
1196free_stack:
1197	exit_task_stack_account(tsk);
1198	free_thread_stack(tsk);
1199free_tsk:
1200	free_task_struct(tsk);
1201	return NULL;
1202}
1203
1204__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1205
1206static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1207
1208static int __init coredump_filter_setup(char *s)
1209{
1210	default_dump_filter =
1211		(simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1212		MMF_DUMP_FILTER_MASK;
1213	return 1;
1214}
1215
1216__setup("coredump_filter=", coredump_filter_setup);
1217
1218#include <linux/init_task.h>
1219
1220static void mm_init_aio(struct mm_struct *mm)
1221{
1222#ifdef CONFIG_AIO
1223	spin_lock_init(&mm->ioctx_lock);
1224	mm->ioctx_table = NULL;
1225#endif
1226}
1227
1228static __always_inline void mm_clear_owner(struct mm_struct *mm,
1229					   struct task_struct *p)
1230{
1231#ifdef CONFIG_MEMCG
1232	if (mm->owner == p)
1233		WRITE_ONCE(mm->owner, NULL);
1234#endif
1235}
1236
1237static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1238{
1239#ifdef CONFIG_MEMCG
1240	mm->owner = p;
1241#endif
1242}
1243
1244static void mm_init_uprobes_state(struct mm_struct *mm)
1245{
1246#ifdef CONFIG_UPROBES
1247	mm->uprobes_state.xol_area = NULL;
1248#endif
1249}
1250
1251static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1252	struct user_namespace *user_ns)
1253{
1254	mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1255	mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
 
1256	atomic_set(&mm->mm_users, 1);
1257	atomic_set(&mm->mm_count, 1);
1258	seqcount_init(&mm->write_protect_seq);
1259	mmap_init_lock(mm);
1260	INIT_LIST_HEAD(&mm->mmlist);
1261#ifdef CONFIG_PER_VMA_LOCK
1262	mm->mm_lock_seq = 0;
1263#endif
1264	mm_pgtables_bytes_init(mm);
1265	mm->map_count = 0;
1266	mm->locked_vm = 0;
 
1267	atomic64_set(&mm->pinned_vm, 0);
1268	memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1269	spin_lock_init(&mm->page_table_lock);
1270	spin_lock_init(&mm->arg_lock);
1271	mm_init_cpumask(mm);
1272	mm_init_aio(mm);
1273	mm_init_owner(mm, p);
1274	mm_pasid_init(mm);
1275	RCU_INIT_POINTER(mm->exe_file, NULL);
1276	mmu_notifier_subscriptions_init(mm);
1277	init_tlb_flush_pending(mm);
1278#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1279	mm->pmd_huge_pte = NULL;
1280#endif
1281	mm_init_uprobes_state(mm);
1282	hugetlb_count_init(mm);
1283
1284	if (current->mm) {
1285		mm->flags = mmf_init_flags(current->mm->flags);
1286		mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1287	} else {
1288		mm->flags = default_dump_filter;
1289		mm->def_flags = 0;
1290	}
1291
1292	if (mm_alloc_pgd(mm))
1293		goto fail_nopgd;
1294
1295	if (init_new_context(p, mm))
1296		goto fail_nocontext;
1297
1298	if (mm_alloc_cid(mm))
1299		goto fail_cid;
1300
1301	if (percpu_counter_init_many(mm->rss_stat, 0, GFP_KERNEL_ACCOUNT,
1302				     NR_MM_COUNTERS))
1303		goto fail_pcpu;
1304
1305	mm->user_ns = get_user_ns(user_ns);
1306	lru_gen_init_mm(mm);
1307	return mm;
1308
1309fail_pcpu:
1310	mm_destroy_cid(mm);
1311fail_cid:
1312	destroy_context(mm);
1313fail_nocontext:
1314	mm_free_pgd(mm);
1315fail_nopgd:
1316	free_mm(mm);
1317	return NULL;
1318}
1319
1320/*
1321 * Allocate and initialize an mm_struct.
1322 */
1323struct mm_struct *mm_alloc(void)
1324{
1325	struct mm_struct *mm;
1326
1327	mm = allocate_mm();
1328	if (!mm)
1329		return NULL;
1330
1331	memset(mm, 0, sizeof(*mm));
1332	return mm_init(mm, current, current_user_ns());
1333}
1334
1335static inline void __mmput(struct mm_struct *mm)
1336{
1337	VM_BUG_ON(atomic_read(&mm->mm_users));
1338
1339	uprobe_clear_state(mm);
1340	exit_aio(mm);
1341	ksm_exit(mm);
1342	khugepaged_exit(mm); /* must run before exit_mmap */
1343	exit_mmap(mm);
1344	mm_put_huge_zero_page(mm);
1345	set_mm_exe_file(mm, NULL);
1346	if (!list_empty(&mm->mmlist)) {
1347		spin_lock(&mmlist_lock);
1348		list_del(&mm->mmlist);
1349		spin_unlock(&mmlist_lock);
1350	}
1351	if (mm->binfmt)
1352		module_put(mm->binfmt->module);
1353	lru_gen_del_mm(mm);
1354	mmdrop(mm);
1355}
1356
1357/*
1358 * Decrement the use count and release all resources for an mm.
1359 */
1360void mmput(struct mm_struct *mm)
1361{
1362	might_sleep();
1363
1364	if (atomic_dec_and_test(&mm->mm_users))
1365		__mmput(mm);
1366}
1367EXPORT_SYMBOL_GPL(mmput);
1368
1369#ifdef CONFIG_MMU
1370static void mmput_async_fn(struct work_struct *work)
1371{
1372	struct mm_struct *mm = container_of(work, struct mm_struct,
1373					    async_put_work);
1374
1375	__mmput(mm);
1376}
1377
1378void mmput_async(struct mm_struct *mm)
1379{
1380	if (atomic_dec_and_test(&mm->mm_users)) {
1381		INIT_WORK(&mm->async_put_work, mmput_async_fn);
1382		schedule_work(&mm->async_put_work);
1383	}
1384}
1385EXPORT_SYMBOL_GPL(mmput_async);
1386#endif
1387
1388/**
1389 * set_mm_exe_file - change a reference to the mm's executable file
1390 * @mm: The mm to change.
1391 * @new_exe_file: The new file to use.
1392 *
1393 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1394 *
1395 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1396 * invocations: in mmput() nobody alive left, in execve it happens before
1397 * the new mm is made visible to anyone.
1398 *
1399 * Can only fail if new_exe_file != NULL.
1400 */
1401int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1402{
1403	struct file *old_exe_file;
1404
1405	/*
1406	 * It is safe to dereference the exe_file without RCU as
1407	 * this function is only called if nobody else can access
1408	 * this mm -- see comment above for justification.
1409	 */
1410	old_exe_file = rcu_dereference_raw(mm->exe_file);
1411
1412	if (new_exe_file) {
1413		/*
1414		 * We expect the caller (i.e., sys_execve) to already denied
1415		 * write access, so this is unlikely to fail.
1416		 */
1417		if (unlikely(deny_write_access(new_exe_file)))
1418			return -EACCES;
1419		get_file(new_exe_file);
1420	}
1421	rcu_assign_pointer(mm->exe_file, new_exe_file);
1422	if (old_exe_file) {
1423		allow_write_access(old_exe_file);
1424		fput(old_exe_file);
1425	}
1426	return 0;
1427}
1428
1429/**
1430 * replace_mm_exe_file - replace a reference to the mm's executable file
1431 * @mm: The mm to change.
1432 * @new_exe_file: The new file to use.
1433 *
1434 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1435 *
1436 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1437 */
1438int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1439{
1440	struct vm_area_struct *vma;
1441	struct file *old_exe_file;
1442	int ret = 0;
1443
1444	/* Forbid mm->exe_file change if old file still mapped. */
1445	old_exe_file = get_mm_exe_file(mm);
1446	if (old_exe_file) {
1447		VMA_ITERATOR(vmi, mm, 0);
1448		mmap_read_lock(mm);
1449		for_each_vma(vmi, vma) {
1450			if (!vma->vm_file)
1451				continue;
1452			if (path_equal(&vma->vm_file->f_path,
1453				       &old_exe_file->f_path)) {
1454				ret = -EBUSY;
1455				break;
1456			}
1457		}
1458		mmap_read_unlock(mm);
1459		fput(old_exe_file);
1460		if (ret)
1461			return ret;
1462	}
1463
1464	ret = deny_write_access(new_exe_file);
1465	if (ret)
1466		return -EACCES;
1467	get_file(new_exe_file);
1468
1469	/* set the new file */
1470	mmap_write_lock(mm);
1471	old_exe_file = rcu_dereference_raw(mm->exe_file);
1472	rcu_assign_pointer(mm->exe_file, new_exe_file);
1473	mmap_write_unlock(mm);
1474
1475	if (old_exe_file) {
1476		allow_write_access(old_exe_file);
1477		fput(old_exe_file);
1478	}
1479	return 0;
1480}
1481
1482/**
1483 * get_mm_exe_file - acquire a reference to the mm's executable file
1484 * @mm: The mm of interest.
1485 *
1486 * Returns %NULL if mm has no associated executable file.
1487 * User must release file via fput().
1488 */
1489struct file *get_mm_exe_file(struct mm_struct *mm)
1490{
1491	struct file *exe_file;
1492
1493	rcu_read_lock();
1494	exe_file = get_file_rcu(&mm->exe_file);
 
 
1495	rcu_read_unlock();
1496	return exe_file;
1497}
 
1498
1499/**
1500 * get_task_exe_file - acquire a reference to the task's executable file
1501 * @task: The task.
1502 *
1503 * Returns %NULL if task's mm (if any) has no associated executable file or
1504 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1505 * User must release file via fput().
1506 */
1507struct file *get_task_exe_file(struct task_struct *task)
1508{
1509	struct file *exe_file = NULL;
1510	struct mm_struct *mm;
1511
1512	task_lock(task);
1513	mm = task->mm;
1514	if (mm) {
1515		if (!(task->flags & PF_KTHREAD))
1516			exe_file = get_mm_exe_file(mm);
1517	}
1518	task_unlock(task);
1519	return exe_file;
1520}
 
1521
1522/**
1523 * get_task_mm - acquire a reference to the task's mm
1524 * @task: The task.
1525 *
1526 * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1527 * this kernel workthread has transiently adopted a user mm with use_mm,
1528 * to do its AIO) is not set and if so returns a reference to it, after
1529 * bumping up the use count.  User must release the mm via mmput()
1530 * after use.  Typically used by /proc and ptrace.
1531 */
1532struct mm_struct *get_task_mm(struct task_struct *task)
1533{
1534	struct mm_struct *mm;
1535
1536	task_lock(task);
1537	mm = task->mm;
1538	if (mm) {
1539		if (task->flags & PF_KTHREAD)
1540			mm = NULL;
1541		else
1542			mmget(mm);
1543	}
1544	task_unlock(task);
1545	return mm;
1546}
1547EXPORT_SYMBOL_GPL(get_task_mm);
1548
1549struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1550{
1551	struct mm_struct *mm;
1552	int err;
1553
1554	err =  down_read_killable(&task->signal->exec_update_lock);
1555	if (err)
1556		return ERR_PTR(err);
1557
1558	mm = get_task_mm(task);
1559	if (mm && mm != current->mm &&
1560			!ptrace_may_access(task, mode)) {
1561		mmput(mm);
1562		mm = ERR_PTR(-EACCES);
1563	}
1564	up_read(&task->signal->exec_update_lock);
1565
1566	return mm;
1567}
1568
1569static void complete_vfork_done(struct task_struct *tsk)
1570{
1571	struct completion *vfork;
1572
1573	task_lock(tsk);
1574	vfork = tsk->vfork_done;
1575	if (likely(vfork)) {
1576		tsk->vfork_done = NULL;
1577		complete(vfork);
1578	}
1579	task_unlock(tsk);
1580}
1581
1582static int wait_for_vfork_done(struct task_struct *child,
1583				struct completion *vfork)
1584{
1585	unsigned int state = TASK_KILLABLE|TASK_FREEZABLE;
1586	int killed;
1587
 
1588	cgroup_enter_frozen();
1589	killed = wait_for_completion_state(vfork, state);
1590	cgroup_leave_frozen(false);
 
1591
1592	if (killed) {
1593		task_lock(child);
1594		child->vfork_done = NULL;
1595		task_unlock(child);
1596	}
1597
1598	put_task_struct(child);
1599	return killed;
1600}
1601
1602/* Please note the differences between mmput and mm_release.
1603 * mmput is called whenever we stop holding onto a mm_struct,
1604 * error success whatever.
1605 *
1606 * mm_release is called after a mm_struct has been removed
1607 * from the current process.
1608 *
1609 * This difference is important for error handling, when we
1610 * only half set up a mm_struct for a new process and need to restore
1611 * the old one.  Because we mmput the new mm_struct before
1612 * restoring the old one. . .
1613 * Eric Biederman 10 January 1998
1614 */
1615static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1616{
1617	uprobe_free_utask(tsk);
1618
1619	/* Get rid of any cached register state */
1620	deactivate_mm(tsk, mm);
1621
1622	/*
1623	 * Signal userspace if we're not exiting with a core dump
1624	 * because we want to leave the value intact for debugging
1625	 * purposes.
1626	 */
1627	if (tsk->clear_child_tid) {
1628		if (atomic_read(&mm->mm_users) > 1) {
 
1629			/*
1630			 * We don't check the error code - if userspace has
1631			 * not set up a proper pointer then tough luck.
1632			 */
1633			put_user(0, tsk->clear_child_tid);
1634			do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1635					1, NULL, NULL, 0, 0);
1636		}
1637		tsk->clear_child_tid = NULL;
1638	}
1639
1640	/*
1641	 * All done, finally we can wake up parent and return this mm to him.
1642	 * Also kthread_stop() uses this completion for synchronization.
1643	 */
1644	if (tsk->vfork_done)
1645		complete_vfork_done(tsk);
1646}
1647
1648void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1649{
1650	futex_exit_release(tsk);
1651	mm_release(tsk, mm);
1652}
1653
1654void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1655{
1656	futex_exec_release(tsk);
1657	mm_release(tsk, mm);
1658}
1659
1660/**
1661 * dup_mm() - duplicates an existing mm structure
1662 * @tsk: the task_struct with which the new mm will be associated.
1663 * @oldmm: the mm to duplicate.
1664 *
1665 * Allocates a new mm structure and duplicates the provided @oldmm structure
1666 * content into it.
1667 *
1668 * Return: the duplicated mm or NULL on failure.
1669 */
1670static struct mm_struct *dup_mm(struct task_struct *tsk,
1671				struct mm_struct *oldmm)
1672{
1673	struct mm_struct *mm;
1674	int err;
1675
1676	mm = allocate_mm();
1677	if (!mm)
1678		goto fail_nomem;
1679
1680	memcpy(mm, oldmm, sizeof(*mm));
1681
1682	if (!mm_init(mm, tsk, mm->user_ns))
1683		goto fail_nomem;
1684
1685	err = dup_mmap(mm, oldmm);
1686	if (err)
1687		goto free_pt;
1688
1689	mm->hiwater_rss = get_mm_rss(mm);
1690	mm->hiwater_vm = mm->total_vm;
1691
1692	if (mm->binfmt && !try_module_get(mm->binfmt->module))
1693		goto free_pt;
1694
1695	return mm;
1696
1697free_pt:
1698	/* don't put binfmt in mmput, we haven't got module yet */
1699	mm->binfmt = NULL;
1700	mm_init_owner(mm, NULL);
1701	mmput(mm);
1702
1703fail_nomem:
1704	return NULL;
1705}
1706
1707static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1708{
1709	struct mm_struct *mm, *oldmm;
 
1710
1711	tsk->min_flt = tsk->maj_flt = 0;
1712	tsk->nvcsw = tsk->nivcsw = 0;
1713#ifdef CONFIG_DETECT_HUNG_TASK
1714	tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1715	tsk->last_switch_time = 0;
1716#endif
1717
1718	tsk->mm = NULL;
1719	tsk->active_mm = NULL;
1720
1721	/*
1722	 * Are we cloning a kernel thread?
1723	 *
1724	 * We need to steal a active VM for that..
1725	 */
1726	oldmm = current->mm;
1727	if (!oldmm)
1728		return 0;
1729
 
 
 
1730	if (clone_flags & CLONE_VM) {
1731		mmget(oldmm);
1732		mm = oldmm;
1733	} else {
1734		mm = dup_mm(tsk, current->mm);
1735		if (!mm)
1736			return -ENOMEM;
1737	}
1738
 
 
 
 
 
 
1739	tsk->mm = mm;
1740	tsk->active_mm = mm;
1741	sched_mm_cid_fork(tsk);
1742	return 0;
 
 
 
1743}
1744
1745static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1746{
1747	struct fs_struct *fs = current->fs;
1748	if (clone_flags & CLONE_FS) {
1749		/* tsk->fs is already what we want */
1750		spin_lock(&fs->lock);
1751		/* "users" and "in_exec" locked for check_unsafe_exec() */
1752		if (fs->in_exec) {
1753			spin_unlock(&fs->lock);
1754			return -EAGAIN;
1755		}
1756		fs->users++;
1757		spin_unlock(&fs->lock);
1758		return 0;
1759	}
1760	tsk->fs = copy_fs_struct(fs);
1761	if (!tsk->fs)
1762		return -ENOMEM;
1763	return 0;
1764}
1765
1766static int copy_files(unsigned long clone_flags, struct task_struct *tsk,
1767		      int no_files)
1768{
1769	struct files_struct *oldf, *newf;
1770	int error = 0;
1771
1772	/*
1773	 * A background process may not have any files ...
1774	 */
1775	oldf = current->files;
1776	if (!oldf)
1777		goto out;
1778
1779	if (no_files) {
1780		tsk->files = NULL;
1781		goto out;
1782	}
1783
1784	if (clone_flags & CLONE_FILES) {
1785		atomic_inc(&oldf->count);
1786		goto out;
1787	}
1788
1789	newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1790	if (!newf)
1791		goto out;
1792
1793	tsk->files = newf;
1794	error = 0;
1795out:
1796	return error;
1797}
1798
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1799static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1800{
1801	struct sighand_struct *sig;
1802
1803	if (clone_flags & CLONE_SIGHAND) {
1804		refcount_inc(&current->sighand->count);
1805		return 0;
1806	}
1807	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1808	RCU_INIT_POINTER(tsk->sighand, sig);
1809	if (!sig)
1810		return -ENOMEM;
1811
1812	refcount_set(&sig->count, 1);
1813	spin_lock_irq(&current->sighand->siglock);
1814	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1815	spin_unlock_irq(&current->sighand->siglock);
1816
1817	/* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1818	if (clone_flags & CLONE_CLEAR_SIGHAND)
1819		flush_signal_handlers(tsk, 0);
1820
1821	return 0;
1822}
1823
1824void __cleanup_sighand(struct sighand_struct *sighand)
1825{
1826	if (refcount_dec_and_test(&sighand->count)) {
1827		signalfd_cleanup(sighand);
1828		/*
1829		 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1830		 * without an RCU grace period, see __lock_task_sighand().
1831		 */
1832		kmem_cache_free(sighand_cachep, sighand);
1833	}
1834}
1835
1836/*
1837 * Initialize POSIX timer handling for a thread group.
1838 */
1839static void posix_cpu_timers_init_group(struct signal_struct *sig)
1840{
1841	struct posix_cputimers *pct = &sig->posix_cputimers;
1842	unsigned long cpu_limit;
1843
1844	cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1845	posix_cputimers_group_init(pct, cpu_limit);
1846}
1847
1848static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1849{
1850	struct signal_struct *sig;
1851
1852	if (clone_flags & CLONE_THREAD)
1853		return 0;
1854
1855	sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1856	tsk->signal = sig;
1857	if (!sig)
1858		return -ENOMEM;
1859
1860	sig->nr_threads = 1;
1861	sig->quick_threads = 1;
1862	atomic_set(&sig->live, 1);
1863	refcount_set(&sig->sigcnt, 1);
1864
1865	/* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1866	sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1867	tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1868
1869	init_waitqueue_head(&sig->wait_chldexit);
1870	sig->curr_target = tsk;
1871	init_sigpending(&sig->shared_pending);
1872	INIT_HLIST_HEAD(&sig->multiprocess);
1873	seqlock_init(&sig->stats_lock);
1874	prev_cputime_init(&sig->prev_cputime);
1875
1876#ifdef CONFIG_POSIX_TIMERS
1877	INIT_LIST_HEAD(&sig->posix_timers);
1878	hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1879	sig->real_timer.function = it_real_fn;
1880#endif
1881
1882	task_lock(current->group_leader);
1883	memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1884	task_unlock(current->group_leader);
1885
1886	posix_cpu_timers_init_group(sig);
1887
1888	tty_audit_fork(sig);
1889	sched_autogroup_fork(sig);
1890
1891	sig->oom_score_adj = current->signal->oom_score_adj;
1892	sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1893
1894	mutex_init(&sig->cred_guard_mutex);
1895	init_rwsem(&sig->exec_update_lock);
1896
1897	return 0;
1898}
1899
1900static void copy_seccomp(struct task_struct *p)
1901{
1902#ifdef CONFIG_SECCOMP
1903	/*
1904	 * Must be called with sighand->lock held, which is common to
1905	 * all threads in the group. Holding cred_guard_mutex is not
1906	 * needed because this new task is not yet running and cannot
1907	 * be racing exec.
1908	 */
1909	assert_spin_locked(&current->sighand->siglock);
1910
1911	/* Ref-count the new filter user, and assign it. */
1912	get_seccomp_filter(current);
1913	p->seccomp = current->seccomp;
1914
1915	/*
1916	 * Explicitly enable no_new_privs here in case it got set
1917	 * between the task_struct being duplicated and holding the
1918	 * sighand lock. The seccomp state and nnp must be in sync.
1919	 */
1920	if (task_no_new_privs(current))
1921		task_set_no_new_privs(p);
1922
1923	/*
1924	 * If the parent gained a seccomp mode after copying thread
1925	 * flags and between before we held the sighand lock, we have
1926	 * to manually enable the seccomp thread flag here.
1927	 */
1928	if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1929		set_task_syscall_work(p, SECCOMP);
1930#endif
1931}
1932
1933SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1934{
1935	current->clear_child_tid = tidptr;
1936
1937	return task_pid_vnr(current);
1938}
1939
1940static void rt_mutex_init_task(struct task_struct *p)
1941{
1942	raw_spin_lock_init(&p->pi_lock);
1943#ifdef CONFIG_RT_MUTEXES
1944	p->pi_waiters = RB_ROOT_CACHED;
1945	p->pi_top_task = NULL;
1946	p->pi_blocked_on = NULL;
1947#endif
1948}
1949
1950static inline void init_task_pid_links(struct task_struct *task)
1951{
1952	enum pid_type type;
1953
1954	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1955		INIT_HLIST_NODE(&task->pid_links[type]);
 
1956}
1957
1958static inline void
1959init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1960{
1961	if (type == PIDTYPE_PID)
1962		task->thread_pid = pid;
1963	else
1964		task->signal->pids[type] = pid;
1965}
1966
1967static inline void rcu_copy_process(struct task_struct *p)
1968{
1969#ifdef CONFIG_PREEMPT_RCU
1970	p->rcu_read_lock_nesting = 0;
1971	p->rcu_read_unlock_special.s = 0;
1972	p->rcu_blocked_node = NULL;
1973	INIT_LIST_HEAD(&p->rcu_node_entry);
1974#endif /* #ifdef CONFIG_PREEMPT_RCU */
1975#ifdef CONFIG_TASKS_RCU
1976	p->rcu_tasks_holdout = false;
1977	INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1978	p->rcu_tasks_idle_cpu = -1;
1979#endif /* #ifdef CONFIG_TASKS_RCU */
1980#ifdef CONFIG_TASKS_TRACE_RCU
1981	p->trc_reader_nesting = 0;
1982	p->trc_reader_special.s = 0;
1983	INIT_LIST_HEAD(&p->trc_holdout_list);
1984	INIT_LIST_HEAD(&p->trc_blkd_node);
1985#endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1986}
1987
1988struct pid *pidfd_pid(const struct file *file)
1989{
1990	if (file->f_op == &pidfd_fops)
1991		return file->private_data;
1992
1993	return ERR_PTR(-EBADF);
1994}
1995
1996static int pidfd_release(struct inode *inode, struct file *file)
1997{
1998	struct pid *pid = file->private_data;
1999
2000	file->private_data = NULL;
2001	put_pid(pid);
2002	return 0;
2003}
2004
2005#ifdef CONFIG_PROC_FS
2006/**
2007 * pidfd_show_fdinfo - print information about a pidfd
2008 * @m: proc fdinfo file
2009 * @f: file referencing a pidfd
2010 *
2011 * Pid:
2012 * This function will print the pid that a given pidfd refers to in the
2013 * pid namespace of the procfs instance.
2014 * If the pid namespace of the process is not a descendant of the pid
2015 * namespace of the procfs instance 0 will be shown as its pid. This is
2016 * similar to calling getppid() on a process whose parent is outside of
2017 * its pid namespace.
2018 *
2019 * NSpid:
2020 * If pid namespaces are supported then this function will also print
2021 * the pid of a given pidfd refers to for all descendant pid namespaces
2022 * starting from the current pid namespace of the instance, i.e. the
2023 * Pid field and the first entry in the NSpid field will be identical.
2024 * If the pid namespace of the process is not a descendant of the pid
2025 * namespace of the procfs instance 0 will be shown as its first NSpid
2026 * entry and no others will be shown.
2027 * Note that this differs from the Pid and NSpid fields in
2028 * /proc/<pid>/status where Pid and NSpid are always shown relative to
2029 * the  pid namespace of the procfs instance. The difference becomes
2030 * obvious when sending around a pidfd between pid namespaces from a
2031 * different branch of the tree, i.e. where no ancestral relation is
2032 * present between the pid namespaces:
2033 * - create two new pid namespaces ns1 and ns2 in the initial pid
2034 *   namespace (also take care to create new mount namespaces in the
2035 *   new pid namespace and mount procfs)
2036 * - create a process with a pidfd in ns1
2037 * - send pidfd from ns1 to ns2
2038 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
2039 *   have exactly one entry, which is 0
2040 */
2041static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
2042{
2043	struct pid *pid = f->private_data;
2044	struct pid_namespace *ns;
2045	pid_t nr = -1;
2046
2047	if (likely(pid_has_task(pid, PIDTYPE_PID))) {
2048		ns = proc_pid_ns(file_inode(m->file)->i_sb);
2049		nr = pid_nr_ns(pid, ns);
2050	}
2051
2052	seq_put_decimal_ll(m, "Pid:\t", nr);
2053
2054#ifdef CONFIG_PID_NS
2055	seq_put_decimal_ll(m, "\nNSpid:\t", nr);
2056	if (nr > 0) {
2057		int i;
2058
2059		/* If nr is non-zero it means that 'pid' is valid and that
2060		 * ns, i.e. the pid namespace associated with the procfs
2061		 * instance, is in the pid namespace hierarchy of pid.
2062		 * Start at one below the already printed level.
2063		 */
2064		for (i = ns->level + 1; i <= pid->level; i++)
2065			seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
2066	}
2067#endif
2068	seq_putc(m, '\n');
2069}
2070#endif
2071
2072/*
2073 * Poll support for process exit notification.
2074 */
2075static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
2076{
2077	struct pid *pid = file->private_data;
2078	__poll_t poll_flags = 0;
2079
2080	poll_wait(file, &pid->wait_pidfd, pts);
2081
2082	/*
2083	 * Inform pollers only when the whole thread group exits.
2084	 * If the thread group leader exits before all other threads in the
2085	 * group, then poll(2) should block, similar to the wait(2) family.
2086	 */
2087	if (thread_group_exited(pid))
2088		poll_flags = EPOLLIN | EPOLLRDNORM;
2089
2090	return poll_flags;
2091}
2092
2093const struct file_operations pidfd_fops = {
2094	.release = pidfd_release,
2095	.poll = pidfd_poll,
2096#ifdef CONFIG_PROC_FS
2097	.show_fdinfo = pidfd_show_fdinfo,
2098#endif
2099};
2100
2101/**
2102 * __pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2103 * @pid:   the struct pid for which to create a pidfd
2104 * @flags: flags of the new @pidfd
2105 * @ret: Where to return the file for the pidfd.
2106 *
2107 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2108 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2109 *
2110 * The helper doesn't perform checks on @pid which makes it useful for pidfds
2111 * created via CLONE_PIDFD where @pid has no task attached when the pidfd and
2112 * pidfd file are prepared.
2113 *
2114 * If this function returns successfully the caller is responsible to either
2115 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2116 * order to install the pidfd into its file descriptor table or they must use
2117 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2118 * respectively.
2119 *
2120 * This function is useful when a pidfd must already be reserved but there
2121 * might still be points of failure afterwards and the caller wants to ensure
2122 * that no pidfd is leaked into its file descriptor table.
2123 *
2124 * Return: On success, a reserved pidfd is returned from the function and a new
2125 *         pidfd file is returned in the last argument to the function. On
2126 *         error, a negative error code is returned from the function and the
2127 *         last argument remains unchanged.
2128 */
2129static int __pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2130{
2131	int pidfd;
2132	struct file *pidfd_file;
2133
2134	if (flags & ~(O_NONBLOCK | O_RDWR | O_CLOEXEC))
2135		return -EINVAL;
2136
2137	pidfd = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2138	if (pidfd < 0)
2139		return pidfd;
2140
2141	pidfd_file = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2142					flags | O_RDWR | O_CLOEXEC);
2143	if (IS_ERR(pidfd_file)) {
2144		put_unused_fd(pidfd);
2145		return PTR_ERR(pidfd_file);
2146	}
2147	get_pid(pid); /* held by pidfd_file now */
2148	*ret = pidfd_file;
2149	return pidfd;
2150}
2151
2152/**
2153 * pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2154 * @pid:   the struct pid for which to create a pidfd
2155 * @flags: flags of the new @pidfd
2156 * @ret: Where to return the pidfd.
2157 *
2158 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2159 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2160 *
2161 * The helper verifies that @pid is used as a thread group leader.
2162 *
2163 * If this function returns successfully the caller is responsible to either
2164 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2165 * order to install the pidfd into its file descriptor table or they must use
2166 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2167 * respectively.
2168 *
2169 * This function is useful when a pidfd must already be reserved but there
2170 * might still be points of failure afterwards and the caller wants to ensure
2171 * that no pidfd is leaked into its file descriptor table.
2172 *
2173 * Return: On success, a reserved pidfd is returned from the function and a new
2174 *         pidfd file is returned in the last argument to the function. On
2175 *         error, a negative error code is returned from the function and the
2176 *         last argument remains unchanged.
2177 */
2178int pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2179{
2180	if (!pid || !pid_has_task(pid, PIDTYPE_TGID))
2181		return -EINVAL;
2182
2183	return __pidfd_prepare(pid, flags, ret);
2184}
2185
2186static void __delayed_free_task(struct rcu_head *rhp)
2187{
2188	struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
2189
2190	free_task(tsk);
2191}
2192
2193static __always_inline void delayed_free_task(struct task_struct *tsk)
2194{
2195	if (IS_ENABLED(CONFIG_MEMCG))
2196		call_rcu(&tsk->rcu, __delayed_free_task);
2197	else
2198		free_task(tsk);
2199}
2200
2201static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
2202{
2203	/* Skip if kernel thread */
2204	if (!tsk->mm)
2205		return;
2206
2207	/* Skip if spawning a thread or using vfork */
2208	if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
2209		return;
2210
2211	/* We need to synchronize with __set_oom_adj */
2212	mutex_lock(&oom_adj_mutex);
2213	set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
2214	/* Update the values in case they were changed after copy_signal */
2215	tsk->signal->oom_score_adj = current->signal->oom_score_adj;
2216	tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
2217	mutex_unlock(&oom_adj_mutex);
2218}
2219
2220#ifdef CONFIG_RV
2221static void rv_task_fork(struct task_struct *p)
2222{
2223	int i;
2224
2225	for (i = 0; i < RV_PER_TASK_MONITORS; i++)
2226		p->rv[i].da_mon.monitoring = false;
2227}
2228#else
2229#define rv_task_fork(p) do {} while (0)
2230#endif
2231
2232/*
2233 * This creates a new process as a copy of the old one,
2234 * but does not actually start it yet.
2235 *
2236 * It copies the registers, and all the appropriate
2237 * parts of the process environment (as per the clone
2238 * flags). The actual kick-off is left to the caller.
2239 */
2240__latent_entropy struct task_struct *copy_process(
2241					struct pid *pid,
2242					int trace,
2243					int node,
2244					struct kernel_clone_args *args)
2245{
2246	int pidfd = -1, retval;
2247	struct task_struct *p;
2248	struct multiprocess_signals delayed;
2249	struct file *pidfile = NULL;
2250	const u64 clone_flags = args->flags;
2251	struct nsproxy *nsp = current->nsproxy;
2252
2253	/*
2254	 * Don't allow sharing the root directory with processes in a different
2255	 * namespace
2256	 */
2257	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2258		return ERR_PTR(-EINVAL);
2259
2260	if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2261		return ERR_PTR(-EINVAL);
2262
2263	/*
2264	 * Thread groups must share signals as well, and detached threads
2265	 * can only be started up within the thread group.
2266	 */
2267	if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2268		return ERR_PTR(-EINVAL);
2269
2270	/*
2271	 * Shared signal handlers imply shared VM. By way of the above,
2272	 * thread groups also imply shared VM. Blocking this case allows
2273	 * for various simplifications in other code.
2274	 */
2275	if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2276		return ERR_PTR(-EINVAL);
2277
2278	/*
2279	 * Siblings of global init remain as zombies on exit since they are
2280	 * not reaped by their parent (swapper). To solve this and to avoid
2281	 * multi-rooted process trees, prevent global and container-inits
2282	 * from creating siblings.
2283	 */
2284	if ((clone_flags & CLONE_PARENT) &&
2285				current->signal->flags & SIGNAL_UNKILLABLE)
2286		return ERR_PTR(-EINVAL);
2287
2288	/*
2289	 * If the new process will be in a different pid or user namespace
2290	 * do not allow it to share a thread group with the forking task.
2291	 */
2292	if (clone_flags & CLONE_THREAD) {
2293		if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2294		    (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2295			return ERR_PTR(-EINVAL);
2296	}
2297
 
 
 
 
 
 
 
 
 
2298	if (clone_flags & CLONE_PIDFD) {
2299		/*
2300		 * - CLONE_DETACHED is blocked so that we can potentially
2301		 *   reuse it later for CLONE_PIDFD.
2302		 * - CLONE_THREAD is blocked until someone really needs it.
2303		 */
2304		if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
2305			return ERR_PTR(-EINVAL);
2306	}
2307
2308	/*
2309	 * Force any signals received before this point to be delivered
2310	 * before the fork happens.  Collect up signals sent to multiple
2311	 * processes that happen during the fork and delay them so that
2312	 * they appear to happen after the fork.
2313	 */
2314	sigemptyset(&delayed.signal);
2315	INIT_HLIST_NODE(&delayed.node);
2316
2317	spin_lock_irq(&current->sighand->siglock);
2318	if (!(clone_flags & CLONE_THREAD))
2319		hlist_add_head(&delayed.node, &current->signal->multiprocess);
2320	recalc_sigpending();
2321	spin_unlock_irq(&current->sighand->siglock);
2322	retval = -ERESTARTNOINTR;
2323	if (task_sigpending(current))
2324		goto fork_out;
2325
2326	retval = -ENOMEM;
2327	p = dup_task_struct(current, node);
2328	if (!p)
2329		goto fork_out;
2330	p->flags &= ~PF_KTHREAD;
2331	if (args->kthread)
2332		p->flags |= PF_KTHREAD;
2333	if (args->user_worker) {
2334		/*
2335		 * Mark us a user worker, and block any signal that isn't
2336		 * fatal or STOP
2337		 */
2338		p->flags |= PF_USER_WORKER;
2339		siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2340	}
2341	if (args->io_thread)
2342		p->flags |= PF_IO_WORKER;
2343
2344	if (args->name)
2345		strscpy_pad(p->comm, args->name, sizeof(p->comm));
2346
 
 
 
 
 
 
2347	p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2348	/*
2349	 * Clear TID on mm_release()?
2350	 */
2351	p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2352
2353	ftrace_graph_init_task(p);
2354
2355	rt_mutex_init_task(p);
2356
2357	lockdep_assert_irqs_enabled();
2358#ifdef CONFIG_PROVE_LOCKING
2359	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2360#endif
2361	retval = copy_creds(p, clone_flags);
2362	if (retval < 0)
2363		goto bad_fork_free;
2364
2365	retval = -EAGAIN;
2366	if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
 
2367		if (p->real_cred->user != INIT_USER &&
2368		    !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2369			goto bad_fork_cleanup_count;
2370	}
2371	current->flags &= ~PF_NPROC_EXCEEDED;
2372
 
 
 
 
2373	/*
2374	 * If multiple threads are within copy_process(), then this check
2375	 * triggers too late. This doesn't hurt, the check is only there
2376	 * to stop root fork bombs.
2377	 */
2378	retval = -EAGAIN;
2379	if (data_race(nr_threads >= max_threads))
2380		goto bad_fork_cleanup_count;
2381
2382	delayacct_tsk_init(p);	/* Must remain after dup_task_struct() */
2383	p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2384	p->flags |= PF_FORKNOEXEC;
2385	INIT_LIST_HEAD(&p->children);
2386	INIT_LIST_HEAD(&p->sibling);
2387	rcu_copy_process(p);
2388	p->vfork_done = NULL;
2389	spin_lock_init(&p->alloc_lock);
2390
2391	init_sigpending(&p->pending);
2392
2393	p->utime = p->stime = p->gtime = 0;
2394#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2395	p->utimescaled = p->stimescaled = 0;
2396#endif
2397	prev_cputime_init(&p->prev_cputime);
2398
2399#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2400	seqcount_init(&p->vtime.seqcount);
2401	p->vtime.starttime = 0;
2402	p->vtime.state = VTIME_INACTIVE;
2403#endif
2404
2405#ifdef CONFIG_IO_URING
2406	p->io_uring = NULL;
2407#endif
2408
2409	p->default_timer_slack_ns = current->timer_slack_ns;
2410
2411#ifdef CONFIG_PSI
2412	p->psi_flags = 0;
2413#endif
2414
2415	task_io_accounting_init(&p->ioac);
2416	acct_clear_integrals(p);
2417
2418	posix_cputimers_init(&p->posix_cputimers);
2419
2420	p->io_context = NULL;
2421	audit_set_context(p, NULL);
2422	cgroup_fork(p);
2423	if (args->kthread) {
2424		if (!set_kthread_struct(p))
2425			goto bad_fork_cleanup_delayacct;
2426	}
2427#ifdef CONFIG_NUMA
2428	p->mempolicy = mpol_dup(p->mempolicy);
2429	if (IS_ERR(p->mempolicy)) {
2430		retval = PTR_ERR(p->mempolicy);
2431		p->mempolicy = NULL;
2432		goto bad_fork_cleanup_delayacct;
2433	}
2434#endif
2435#ifdef CONFIG_CPUSETS
2436	p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2437	p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2438	seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2439#endif
2440#ifdef CONFIG_TRACE_IRQFLAGS
2441	memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2442	p->irqtrace.hardirq_disable_ip	= _THIS_IP_;
2443	p->irqtrace.softirq_enable_ip	= _THIS_IP_;
2444	p->softirqs_enabled		= 1;
2445	p->softirq_context		= 0;
2446#endif
2447
2448	p->pagefault_disabled = 0;
2449
2450#ifdef CONFIG_LOCKDEP
2451	lockdep_init_task(p);
2452#endif
2453
2454#ifdef CONFIG_DEBUG_MUTEXES
2455	p->blocked_on = NULL; /* not blocked yet */
2456#endif
2457#ifdef CONFIG_BCACHE
2458	p->sequential_io	= 0;
2459	p->sequential_io_avg	= 0;
2460#endif
2461#ifdef CONFIG_BPF_SYSCALL
2462	RCU_INIT_POINTER(p->bpf_storage, NULL);
2463	p->bpf_ctx = NULL;
2464#endif
2465
2466	/* Perform scheduler related setup. Assign this task to a CPU. */
2467	retval = sched_fork(clone_flags, p);
2468	if (retval)
2469		goto bad_fork_cleanup_policy;
2470
2471	retval = perf_event_init_task(p, clone_flags);
2472	if (retval)
2473		goto bad_fork_cleanup_policy;
2474	retval = audit_alloc(p);
2475	if (retval)
2476		goto bad_fork_cleanup_perf;
2477	/* copy all the process information */
2478	shm_init_task(p);
2479	retval = security_task_alloc(p, clone_flags);
2480	if (retval)
2481		goto bad_fork_cleanup_audit;
2482	retval = copy_semundo(clone_flags, p);
2483	if (retval)
2484		goto bad_fork_cleanup_security;
2485	retval = copy_files(clone_flags, p, args->no_files);
2486	if (retval)
2487		goto bad_fork_cleanup_semundo;
2488	retval = copy_fs(clone_flags, p);
2489	if (retval)
2490		goto bad_fork_cleanup_files;
2491	retval = copy_sighand(clone_flags, p);
2492	if (retval)
2493		goto bad_fork_cleanup_fs;
2494	retval = copy_signal(clone_flags, p);
2495	if (retval)
2496		goto bad_fork_cleanup_sighand;
2497	retval = copy_mm(clone_flags, p);
2498	if (retval)
2499		goto bad_fork_cleanup_signal;
2500	retval = copy_namespaces(clone_flags, p);
2501	if (retval)
2502		goto bad_fork_cleanup_mm;
2503	retval = copy_io(clone_flags, p);
2504	if (retval)
2505		goto bad_fork_cleanup_namespaces;
2506	retval = copy_thread(p, args);
2507	if (retval)
2508		goto bad_fork_cleanup_io;
2509
2510	stackleak_task_init(p);
2511
2512	if (pid != &init_struct_pid) {
2513		pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2514				args->set_tid_size);
2515		if (IS_ERR(pid)) {
2516			retval = PTR_ERR(pid);
2517			goto bad_fork_cleanup_thread;
2518		}
2519	}
2520
2521	/*
2522	 * This has to happen after we've potentially unshared the file
2523	 * descriptor table (so that the pidfd doesn't leak into the child
2524	 * if the fd table isn't shared).
2525	 */
2526	if (clone_flags & CLONE_PIDFD) {
2527		/* Note that no task has been attached to @pid yet. */
2528		retval = __pidfd_prepare(pid, O_RDWR | O_CLOEXEC, &pidfile);
2529		if (retval < 0)
2530			goto bad_fork_free_pid;
 
2531		pidfd = retval;
2532
 
 
 
 
 
 
 
 
 
2533		retval = put_user(pidfd, args->pidfd);
2534		if (retval)
2535			goto bad_fork_put_pidfd;
2536	}
2537
2538#ifdef CONFIG_BLOCK
2539	p->plug = NULL;
2540#endif
2541	futex_init_task(p);
2542
2543	/*
2544	 * sigaltstack should be cleared when sharing the same VM
2545	 */
2546	if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2547		sas_ss_reset(p);
2548
2549	/*
2550	 * Syscall tracing and stepping should be turned off in the
2551	 * child regardless of CLONE_PTRACE.
2552	 */
2553	user_disable_single_step(p);
2554	clear_task_syscall_work(p, SYSCALL_TRACE);
2555#if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2556	clear_task_syscall_work(p, SYSCALL_EMU);
2557#endif
2558	clear_tsk_latency_tracing(p);
2559
2560	/* ok, now we should be set up.. */
2561	p->pid = pid_nr(pid);
2562	if (clone_flags & CLONE_THREAD) {
 
2563		p->group_leader = current->group_leader;
2564		p->tgid = current->tgid;
2565	} else {
 
 
 
 
2566		p->group_leader = p;
2567		p->tgid = p->pid;
2568	}
2569
2570	p->nr_dirtied = 0;
2571	p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2572	p->dirty_paused_when = 0;
2573
2574	p->pdeath_signal = 0;
 
2575	p->task_works = NULL;
2576	clear_posix_cputimers_work(p);
2577
2578#ifdef CONFIG_KRETPROBES
2579	p->kretprobe_instances.first = NULL;
2580#endif
2581#ifdef CONFIG_RETHOOK
2582	p->rethooks.first = NULL;
2583#endif
2584
2585	/*
2586	 * Ensure that the cgroup subsystem policies allow the new process to be
2587	 * forked. It should be noted that the new process's css_set can be changed
2588	 * between here and cgroup_post_fork() if an organisation operation is in
2589	 * progress.
2590	 */
2591	retval = cgroup_can_fork(p, args);
2592	if (retval)
2593		goto bad_fork_put_pidfd;
2594
2595	/*
2596	 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2597	 * the new task on the correct runqueue. All this *before* the task
2598	 * becomes visible.
2599	 *
2600	 * This isn't part of ->can_fork() because while the re-cloning is
2601	 * cgroup specific, it unconditionally needs to place the task on a
2602	 * runqueue.
2603	 */
2604	sched_cgroup_fork(p, args);
2605
2606	/*
2607	 * From this point on we must avoid any synchronous user-space
2608	 * communication until we take the tasklist-lock. In particular, we do
2609	 * not want user-space to be able to predict the process start-time by
2610	 * stalling fork(2) after we recorded the start_time but before it is
2611	 * visible to the system.
2612	 */
2613
2614	p->start_time = ktime_get_ns();
2615	p->start_boottime = ktime_get_boottime_ns();
2616
2617	/*
2618	 * Make it visible to the rest of the system, but dont wake it up yet.
2619	 * Need tasklist lock for parent etc handling!
2620	 */
2621	write_lock_irq(&tasklist_lock);
2622
2623	/* CLONE_PARENT re-uses the old parent */
2624	if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2625		p->real_parent = current->real_parent;
2626		p->parent_exec_id = current->parent_exec_id;
2627		if (clone_flags & CLONE_THREAD)
2628			p->exit_signal = -1;
2629		else
2630			p->exit_signal = current->group_leader->exit_signal;
2631	} else {
2632		p->real_parent = current;
2633		p->parent_exec_id = current->self_exec_id;
2634		p->exit_signal = args->exit_signal;
2635	}
2636
2637	klp_copy_process(p);
2638
2639	sched_core_fork(p);
2640
2641	spin_lock(&current->sighand->siglock);
2642
2643	rv_task_fork(p);
 
 
 
 
2644
2645	rseq_fork(p, clone_flags);
2646
2647	/* Don't start children in a dying pid namespace */
2648	if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2649		retval = -ENOMEM;
2650		goto bad_fork_cancel_cgroup;
2651	}
2652
2653	/* Let kill terminate clone/fork in the middle */
2654	if (fatal_signal_pending(current)) {
2655		retval = -EINTR;
2656		goto bad_fork_cancel_cgroup;
2657	}
2658
2659	/* No more failure paths after this point. */
2660
2661	/*
2662	 * Copy seccomp details explicitly here, in case they were changed
2663	 * before holding sighand lock.
2664	 */
2665	copy_seccomp(p);
2666
2667	init_task_pid_links(p);
2668	if (likely(p->pid)) {
2669		ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2670
2671		init_task_pid(p, PIDTYPE_PID, pid);
2672		if (thread_group_leader(p)) {
2673			init_task_pid(p, PIDTYPE_TGID, pid);
2674			init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2675			init_task_pid(p, PIDTYPE_SID, task_session(current));
2676
2677			if (is_child_reaper(pid)) {
2678				ns_of_pid(pid)->child_reaper = p;
2679				p->signal->flags |= SIGNAL_UNKILLABLE;
2680			}
2681			p->signal->shared_pending.signal = delayed.signal;
2682			p->signal->tty = tty_kref_get(current->signal->tty);
2683			/*
2684			 * Inherit has_child_subreaper flag under the same
2685			 * tasklist_lock with adding child to the process tree
2686			 * for propagate_has_child_subreaper optimization.
2687			 */
2688			p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2689							 p->real_parent->signal->is_child_subreaper;
2690			list_add_tail(&p->sibling, &p->real_parent->children);
2691			list_add_tail_rcu(&p->tasks, &init_task.tasks);
2692			attach_pid(p, PIDTYPE_TGID);
2693			attach_pid(p, PIDTYPE_PGID);
2694			attach_pid(p, PIDTYPE_SID);
2695			__this_cpu_inc(process_counts);
2696		} else {
2697			current->signal->nr_threads++;
2698			current->signal->quick_threads++;
2699			atomic_inc(&current->signal->live);
2700			refcount_inc(&current->signal->sigcnt);
2701			task_join_group_stop(p);
 
 
2702			list_add_tail_rcu(&p->thread_node,
2703					  &p->signal->thread_head);
2704		}
2705		attach_pid(p, PIDTYPE_PID);
2706		nr_threads++;
2707	}
2708	total_forks++;
2709	hlist_del_init(&delayed.node);
2710	spin_unlock(&current->sighand->siglock);
2711	syscall_tracepoint_update(p);
2712	write_unlock_irq(&tasklist_lock);
2713
2714	if (pidfile)
2715		fd_install(pidfd, pidfile);
2716
2717	proc_fork_connector(p);
2718	sched_post_fork(p);
2719	cgroup_post_fork(p, args);
2720	perf_event_fork(p);
2721
2722	trace_task_newtask(p, clone_flags);
2723	uprobe_copy_process(p, clone_flags);
2724	user_events_fork(p, clone_flags);
2725
2726	copy_oom_score_adj(clone_flags, p);
2727
2728	return p;
2729
2730bad_fork_cancel_cgroup:
2731	sched_core_free(p);
2732	spin_unlock(&current->sighand->siglock);
2733	write_unlock_irq(&tasklist_lock);
2734	cgroup_cancel_fork(p, args);
2735bad_fork_put_pidfd:
2736	if (clone_flags & CLONE_PIDFD) {
2737		fput(pidfile);
2738		put_unused_fd(pidfd);
2739	}
2740bad_fork_free_pid:
2741	if (pid != &init_struct_pid)
2742		free_pid(pid);
2743bad_fork_cleanup_thread:
2744	exit_thread(p);
2745bad_fork_cleanup_io:
2746	if (p->io_context)
2747		exit_io_context(p);
2748bad_fork_cleanup_namespaces:
2749	exit_task_namespaces(p);
2750bad_fork_cleanup_mm:
2751	if (p->mm) {
2752		mm_clear_owner(p->mm, p);
2753		mmput(p->mm);
2754	}
2755bad_fork_cleanup_signal:
2756	if (!(clone_flags & CLONE_THREAD))
2757		free_signal_struct(p->signal);
2758bad_fork_cleanup_sighand:
2759	__cleanup_sighand(p->sighand);
2760bad_fork_cleanup_fs:
2761	exit_fs(p); /* blocking */
2762bad_fork_cleanup_files:
2763	exit_files(p); /* blocking */
2764bad_fork_cleanup_semundo:
2765	exit_sem(p);
2766bad_fork_cleanup_security:
2767	security_task_free(p);
2768bad_fork_cleanup_audit:
2769	audit_free(p);
2770bad_fork_cleanup_perf:
2771	perf_event_free_task(p);
2772bad_fork_cleanup_policy:
2773	lockdep_free_task(p);
2774#ifdef CONFIG_NUMA
2775	mpol_put(p->mempolicy);
 
2776#endif
2777bad_fork_cleanup_delayacct:
2778	delayacct_tsk_free(p);
2779bad_fork_cleanup_count:
2780	dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2781	exit_creds(p);
2782bad_fork_free:
2783	WRITE_ONCE(p->__state, TASK_DEAD);
2784	exit_task_stack_account(p);
2785	put_task_stack(p);
2786	delayed_free_task(p);
2787fork_out:
2788	spin_lock_irq(&current->sighand->siglock);
2789	hlist_del_init(&delayed.node);
2790	spin_unlock_irq(&current->sighand->siglock);
2791	return ERR_PTR(retval);
2792}
2793
2794static inline void init_idle_pids(struct task_struct *idle)
2795{
2796	enum pid_type type;
2797
2798	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2799		INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2800		init_task_pid(idle, type, &init_struct_pid);
2801	}
2802}
2803
2804static int idle_dummy(void *dummy)
2805{
2806	/* This function is never called */
2807	return 0;
2808}
2809
2810struct task_struct * __init fork_idle(int cpu)
2811{
2812	struct task_struct *task;
2813	struct kernel_clone_args args = {
2814		.flags		= CLONE_VM,
2815		.fn		= &idle_dummy,
2816		.fn_arg		= NULL,
2817		.kthread	= 1,
2818		.idle		= 1,
2819	};
2820
2821	task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2822	if (!IS_ERR(task)) {
2823		init_idle_pids(task);
2824		init_idle(task, cpu);
2825	}
2826
2827	return task;
2828}
2829
2830/*
2831 * This is like kernel_clone(), but shaved down and tailored to just
2832 * creating io_uring workers. It returns a created task, or an error pointer.
2833 * The returned task is inactive, and the caller must fire it up through
2834 * wake_up_new_task(p). All signals are blocked in the created task.
2835 */
2836struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2837{
2838	unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2839				CLONE_IO;
2840	struct kernel_clone_args args = {
2841		.flags		= ((lower_32_bits(flags) | CLONE_VM |
2842				    CLONE_UNTRACED) & ~CSIGNAL),
2843		.exit_signal	= (lower_32_bits(flags) & CSIGNAL),
2844		.fn		= fn,
2845		.fn_arg		= arg,
2846		.io_thread	= 1,
2847		.user_worker	= 1,
2848	};
2849
2850	return copy_process(NULL, 0, node, &args);
2851}
2852
2853/*
2854 *  Ok, this is the main fork-routine.
2855 *
2856 * It copies the process, and if successful kick-starts
2857 * it and waits for it to finish using the VM if required.
2858 *
2859 * args->exit_signal is expected to be checked for sanity by the caller.
2860 */
2861pid_t kernel_clone(struct kernel_clone_args *args)
2862{
2863	u64 clone_flags = args->flags;
2864	struct completion vfork;
2865	struct pid *pid;
2866	struct task_struct *p;
2867	int trace = 0;
2868	pid_t nr;
2869
2870	/*
2871	 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2872	 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2873	 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2874	 * field in struct clone_args and it still doesn't make sense to have
2875	 * them both point at the same memory location. Performing this check
2876	 * here has the advantage that we don't need to have a separate helper
2877	 * to check for legacy clone().
2878	 */
2879	if ((args->flags & CLONE_PIDFD) &&
2880	    (args->flags & CLONE_PARENT_SETTID) &&
2881	    (args->pidfd == args->parent_tid))
2882		return -EINVAL;
2883
2884	/*
2885	 * Determine whether and which event to report to ptracer.  When
2886	 * called from kernel_thread or CLONE_UNTRACED is explicitly
2887	 * requested, no event is reported; otherwise, report if the event
2888	 * for the type of forking is enabled.
2889	 */
2890	if (!(clone_flags & CLONE_UNTRACED)) {
2891		if (clone_flags & CLONE_VFORK)
2892			trace = PTRACE_EVENT_VFORK;
2893		else if (args->exit_signal != SIGCHLD)
2894			trace = PTRACE_EVENT_CLONE;
2895		else
2896			trace = PTRACE_EVENT_FORK;
2897
2898		if (likely(!ptrace_event_enabled(current, trace)))
2899			trace = 0;
2900	}
2901
2902	p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2903	add_latent_entropy();
2904
2905	if (IS_ERR(p))
2906		return PTR_ERR(p);
2907
2908	/*
2909	 * Do this prior waking up the new thread - the thread pointer
2910	 * might get invalid after that point, if the thread exits quickly.
2911	 */
2912	trace_sched_process_fork(current, p);
2913
2914	pid = get_task_pid(p, PIDTYPE_PID);
2915	nr = pid_vnr(pid);
2916
2917	if (clone_flags & CLONE_PARENT_SETTID)
2918		put_user(nr, args->parent_tid);
2919
2920	if (clone_flags & CLONE_VFORK) {
2921		p->vfork_done = &vfork;
2922		init_completion(&vfork);
2923		get_task_struct(p);
2924	}
2925
2926	if (IS_ENABLED(CONFIG_LRU_GEN_WALKS_MMU) && !(clone_flags & CLONE_VM)) {
2927		/* lock the task to synchronize with memcg migration */
2928		task_lock(p);
2929		lru_gen_add_mm(p->mm);
2930		task_unlock(p);
2931	}
2932
2933	wake_up_new_task(p);
2934
2935	/* forking complete and child started to run, tell ptracer */
2936	if (unlikely(trace))
2937		ptrace_event_pid(trace, pid);
2938
2939	if (clone_flags & CLONE_VFORK) {
2940		if (!wait_for_vfork_done(p, &vfork))
2941			ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2942	}
2943
2944	put_pid(pid);
2945	return nr;
2946}
2947
2948/*
2949 * Create a kernel thread.
2950 */
2951pid_t kernel_thread(int (*fn)(void *), void *arg, const char *name,
2952		    unsigned long flags)
2953{
2954	struct kernel_clone_args args = {
2955		.flags		= ((lower_32_bits(flags) | CLONE_VM |
2956				    CLONE_UNTRACED) & ~CSIGNAL),
2957		.exit_signal	= (lower_32_bits(flags) & CSIGNAL),
2958		.fn		= fn,
2959		.fn_arg		= arg,
2960		.name		= name,
2961		.kthread	= 1,
2962	};
2963
2964	return kernel_clone(&args);
2965}
2966
2967/*
2968 * Create a user mode thread.
2969 */
2970pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2971{
2972	struct kernel_clone_args args = {
2973		.flags		= ((lower_32_bits(flags) | CLONE_VM |
2974				    CLONE_UNTRACED) & ~CSIGNAL),
2975		.exit_signal	= (lower_32_bits(flags) & CSIGNAL),
2976		.fn		= fn,
2977		.fn_arg		= arg,
2978	};
2979
2980	return kernel_clone(&args);
2981}
2982
2983#ifdef __ARCH_WANT_SYS_FORK
2984SYSCALL_DEFINE0(fork)
2985{
2986#ifdef CONFIG_MMU
2987	struct kernel_clone_args args = {
2988		.exit_signal = SIGCHLD,
2989	};
2990
2991	return kernel_clone(&args);
2992#else
2993	/* can not support in nommu mode */
2994	return -EINVAL;
2995#endif
2996}
2997#endif
2998
2999#ifdef __ARCH_WANT_SYS_VFORK
3000SYSCALL_DEFINE0(vfork)
3001{
3002	struct kernel_clone_args args = {
3003		.flags		= CLONE_VFORK | CLONE_VM,
3004		.exit_signal	= SIGCHLD,
3005	};
3006
3007	return kernel_clone(&args);
3008}
3009#endif
3010
3011#ifdef __ARCH_WANT_SYS_CLONE
3012#ifdef CONFIG_CLONE_BACKWARDS
3013SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
3014		 int __user *, parent_tidptr,
3015		 unsigned long, tls,
3016		 int __user *, child_tidptr)
3017#elif defined(CONFIG_CLONE_BACKWARDS2)
3018SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
3019		 int __user *, parent_tidptr,
3020		 int __user *, child_tidptr,
3021		 unsigned long, tls)
3022#elif defined(CONFIG_CLONE_BACKWARDS3)
3023SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
3024		int, stack_size,
3025		int __user *, parent_tidptr,
3026		int __user *, child_tidptr,
3027		unsigned long, tls)
3028#else
3029SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
3030		 int __user *, parent_tidptr,
3031		 int __user *, child_tidptr,
3032		 unsigned long, tls)
3033#endif
3034{
3035	struct kernel_clone_args args = {
3036		.flags		= (lower_32_bits(clone_flags) & ~CSIGNAL),
3037		.pidfd		= parent_tidptr,
3038		.child_tid	= child_tidptr,
3039		.parent_tid	= parent_tidptr,
3040		.exit_signal	= (lower_32_bits(clone_flags) & CSIGNAL),
3041		.stack		= newsp,
3042		.tls		= tls,
3043	};
3044
3045	return kernel_clone(&args);
3046}
3047#endif
3048
3049#ifdef __ARCH_WANT_SYS_CLONE3
3050
3051noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
3052					      struct clone_args __user *uargs,
3053					      size_t usize)
3054{
3055	int err;
3056	struct clone_args args;
3057	pid_t *kset_tid = kargs->set_tid;
3058
3059	BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
3060		     CLONE_ARGS_SIZE_VER0);
3061	BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
3062		     CLONE_ARGS_SIZE_VER1);
3063	BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
3064		     CLONE_ARGS_SIZE_VER2);
3065	BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
3066
3067	if (unlikely(usize > PAGE_SIZE))
3068		return -E2BIG;
3069	if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
3070		return -EINVAL;
3071
3072	err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
3073	if (err)
3074		return err;
3075
3076	if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
3077		return -EINVAL;
3078
3079	if (unlikely(!args.set_tid && args.set_tid_size > 0))
3080		return -EINVAL;
3081
3082	if (unlikely(args.set_tid && args.set_tid_size == 0))
3083		return -EINVAL;
3084
3085	/*
3086	 * Verify that higher 32bits of exit_signal are unset and that
3087	 * it is a valid signal
3088	 */
3089	if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
3090		     !valid_signal(args.exit_signal)))
3091		return -EINVAL;
3092
3093	if ((args.flags & CLONE_INTO_CGROUP) &&
3094	    (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
3095		return -EINVAL;
3096
3097	*kargs = (struct kernel_clone_args){
3098		.flags		= args.flags,
3099		.pidfd		= u64_to_user_ptr(args.pidfd),
3100		.child_tid	= u64_to_user_ptr(args.child_tid),
3101		.parent_tid	= u64_to_user_ptr(args.parent_tid),
3102		.exit_signal	= args.exit_signal,
3103		.stack		= args.stack,
3104		.stack_size	= args.stack_size,
3105		.tls		= args.tls,
3106		.set_tid_size	= args.set_tid_size,
3107		.cgroup		= args.cgroup,
3108	};
3109
3110	if (args.set_tid &&
3111		copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
3112			(kargs->set_tid_size * sizeof(pid_t))))
3113		return -EFAULT;
3114
3115	kargs->set_tid = kset_tid;
3116
3117	return 0;
3118}
3119
3120/**
3121 * clone3_stack_valid - check and prepare stack
3122 * @kargs: kernel clone args
3123 *
3124 * Verify that the stack arguments userspace gave us are sane.
3125 * In addition, set the stack direction for userspace since it's easy for us to
3126 * determine.
3127 */
3128static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
3129{
3130	if (kargs->stack == 0) {
3131		if (kargs->stack_size > 0)
3132			return false;
3133	} else {
3134		if (kargs->stack_size == 0)
3135			return false;
3136
3137		if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
3138			return false;
3139
3140#if !defined(CONFIG_STACK_GROWSUP)
3141		kargs->stack += kargs->stack_size;
3142#endif
3143	}
3144
3145	return true;
3146}
3147
3148static bool clone3_args_valid(struct kernel_clone_args *kargs)
3149{
3150	/* Verify that no unknown flags are passed along. */
3151	if (kargs->flags &
3152	    ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
3153		return false;
3154
3155	/*
3156	 * - make the CLONE_DETACHED bit reusable for clone3
3157	 * - make the CSIGNAL bits reusable for clone3
3158	 */
3159	if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
3160		return false;
3161
3162	if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
3163	    (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
3164		return false;
3165
3166	if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
3167	    kargs->exit_signal)
3168		return false;
3169
3170	if (!clone3_stack_valid(kargs))
3171		return false;
3172
3173	return true;
3174}
3175
3176/**
3177 * sys_clone3 - create a new process with specific properties
3178 * @uargs: argument structure
3179 * @size:  size of @uargs
3180 *
3181 * clone3() is the extensible successor to clone()/clone2().
3182 * It takes a struct as argument that is versioned by its size.
3183 *
3184 * Return: On success, a positive PID for the child process.
3185 *         On error, a negative errno number.
3186 */
3187SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
3188{
3189	int err;
3190
3191	struct kernel_clone_args kargs;
3192	pid_t set_tid[MAX_PID_NS_LEVEL];
3193
3194	kargs.set_tid = set_tid;
3195
3196	err = copy_clone_args_from_user(&kargs, uargs, size);
3197	if (err)
3198		return err;
3199
3200	if (!clone3_args_valid(&kargs))
3201		return -EINVAL;
3202
3203	return kernel_clone(&kargs);
3204}
3205#endif
3206
3207void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
3208{
3209	struct task_struct *leader, *parent, *child;
3210	int res;
3211
3212	read_lock(&tasklist_lock);
3213	leader = top = top->group_leader;
3214down:
3215	for_each_thread(leader, parent) {
3216		list_for_each_entry(child, &parent->children, sibling) {
3217			res = visitor(child, data);
3218			if (res) {
3219				if (res < 0)
3220					goto out;
3221				leader = child;
3222				goto down;
3223			}
3224up:
3225			;
3226		}
3227	}
3228
3229	if (leader != top) {
3230		child = leader;
3231		parent = child->real_parent;
3232		leader = parent->group_leader;
3233		goto up;
3234	}
3235out:
3236	read_unlock(&tasklist_lock);
3237}
3238
3239#ifndef ARCH_MIN_MMSTRUCT_ALIGN
3240#define ARCH_MIN_MMSTRUCT_ALIGN 0
3241#endif
3242
3243static void sighand_ctor(void *data)
3244{
3245	struct sighand_struct *sighand = data;
3246
3247	spin_lock_init(&sighand->siglock);
3248	init_waitqueue_head(&sighand->signalfd_wqh);
3249}
3250
3251void __init mm_cache_init(void)
3252{
3253	unsigned int mm_size;
3254
3255	/*
3256	 * The mm_cpumask is located at the end of mm_struct, and is
3257	 * dynamically sized based on the maximum CPU number this system
3258	 * can have, taking hotplug into account (nr_cpu_ids).
3259	 */
3260	mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
3261
3262	mm_cachep = kmem_cache_create_usercopy("mm_struct",
3263			mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3264			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3265			offsetof(struct mm_struct, saved_auxv),
3266			sizeof_field(struct mm_struct, saved_auxv),
3267			NULL);
3268}
3269
3270void __init proc_caches_init(void)
3271{
3272	sighand_cachep = kmem_cache_create("sighand_cache",
3273			sizeof(struct sighand_struct), 0,
3274			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3275			SLAB_ACCOUNT, sighand_ctor);
3276	signal_cachep = kmem_cache_create("signal_cache",
3277			sizeof(struct signal_struct), 0,
3278			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3279			NULL);
3280	files_cachep = kmem_cache_create("files_cache",
3281			sizeof(struct files_struct), 0,
3282			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3283			NULL);
3284	fs_cachep = kmem_cache_create("fs_cache",
3285			sizeof(struct fs_struct), 0,
3286			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3287			NULL);
3288
 
 
 
 
 
 
 
 
 
 
 
 
 
3289	vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3290#ifdef CONFIG_PER_VMA_LOCK
3291	vma_lock_cachep = KMEM_CACHE(vma_lock, SLAB_PANIC|SLAB_ACCOUNT);
3292#endif
3293	mmap_init();
3294	nsproxy_cache_init();
3295}
3296
3297/*
3298 * Check constraints on flags passed to the unshare system call.
3299 */
3300static int check_unshare_flags(unsigned long unshare_flags)
3301{
3302	if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3303				CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3304				CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3305				CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3306				CLONE_NEWTIME))
3307		return -EINVAL;
3308	/*
3309	 * Not implemented, but pretend it works if there is nothing
3310	 * to unshare.  Note that unsharing the address space or the
3311	 * signal handlers also need to unshare the signal queues (aka
3312	 * CLONE_THREAD).
3313	 */
3314	if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3315		if (!thread_group_empty(current))
3316			return -EINVAL;
3317	}
3318	if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3319		if (refcount_read(&current->sighand->count) > 1)
3320			return -EINVAL;
3321	}
3322	if (unshare_flags & CLONE_VM) {
3323		if (!current_is_single_threaded())
3324			return -EINVAL;
3325	}
3326
3327	return 0;
3328}
3329
3330/*
3331 * Unshare the filesystem structure if it is being shared
3332 */
3333static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3334{
3335	struct fs_struct *fs = current->fs;
3336
3337	if (!(unshare_flags & CLONE_FS) || !fs)
3338		return 0;
3339
3340	/* don't need lock here; in the worst case we'll do useless copy */
3341	if (fs->users == 1)
3342		return 0;
3343
3344	*new_fsp = copy_fs_struct(fs);
3345	if (!*new_fsp)
3346		return -ENOMEM;
3347
3348	return 0;
3349}
3350
3351/*
3352 * Unshare file descriptor table if it is being shared
3353 */
3354int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3355	       struct files_struct **new_fdp)
3356{
3357	struct files_struct *fd = current->files;
3358	int error = 0;
3359
3360	if ((unshare_flags & CLONE_FILES) &&
3361	    (fd && atomic_read(&fd->count) > 1)) {
3362		*new_fdp = dup_fd(fd, max_fds, &error);
3363		if (!*new_fdp)
3364			return error;
3365	}
3366
3367	return 0;
3368}
3369
3370/*
3371 * unshare allows a process to 'unshare' part of the process
3372 * context which was originally shared using clone.  copy_*
3373 * functions used by kernel_clone() cannot be used here directly
3374 * because they modify an inactive task_struct that is being
3375 * constructed. Here we are modifying the current, active,
3376 * task_struct.
3377 */
3378int ksys_unshare(unsigned long unshare_flags)
3379{
3380	struct fs_struct *fs, *new_fs = NULL;
3381	struct files_struct *new_fd = NULL;
3382	struct cred *new_cred = NULL;
3383	struct nsproxy *new_nsproxy = NULL;
3384	int do_sysvsem = 0;
3385	int err;
3386
3387	/*
3388	 * If unsharing a user namespace must also unshare the thread group
3389	 * and unshare the filesystem root and working directories.
3390	 */
3391	if (unshare_flags & CLONE_NEWUSER)
3392		unshare_flags |= CLONE_THREAD | CLONE_FS;
3393	/*
3394	 * If unsharing vm, must also unshare signal handlers.
3395	 */
3396	if (unshare_flags & CLONE_VM)
3397		unshare_flags |= CLONE_SIGHAND;
3398	/*
3399	 * If unsharing a signal handlers, must also unshare the signal queues.
3400	 */
3401	if (unshare_flags & CLONE_SIGHAND)
3402		unshare_flags |= CLONE_THREAD;
3403	/*
3404	 * If unsharing namespace, must also unshare filesystem information.
3405	 */
3406	if (unshare_flags & CLONE_NEWNS)
3407		unshare_flags |= CLONE_FS;
3408
3409	err = check_unshare_flags(unshare_flags);
3410	if (err)
3411		goto bad_unshare_out;
3412	/*
3413	 * CLONE_NEWIPC must also detach from the undolist: after switching
3414	 * to a new ipc namespace, the semaphore arrays from the old
3415	 * namespace are unreachable.
3416	 */
3417	if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3418		do_sysvsem = 1;
3419	err = unshare_fs(unshare_flags, &new_fs);
3420	if (err)
3421		goto bad_unshare_out;
3422	err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3423	if (err)
3424		goto bad_unshare_cleanup_fs;
3425	err = unshare_userns(unshare_flags, &new_cred);
3426	if (err)
3427		goto bad_unshare_cleanup_fd;
3428	err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3429					 new_cred, new_fs);
3430	if (err)
3431		goto bad_unshare_cleanup_cred;
3432
3433	if (new_cred) {
3434		err = set_cred_ucounts(new_cred);
3435		if (err)
3436			goto bad_unshare_cleanup_cred;
3437	}
3438
3439	if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3440		if (do_sysvsem) {
3441			/*
3442			 * CLONE_SYSVSEM is equivalent to sys_exit().
3443			 */
3444			exit_sem(current);
3445		}
3446		if (unshare_flags & CLONE_NEWIPC) {
3447			/* Orphan segments in old ns (see sem above). */
3448			exit_shm(current);
3449			shm_init_task(current);
3450		}
3451
3452		if (new_nsproxy)
3453			switch_task_namespaces(current, new_nsproxy);
3454
3455		task_lock(current);
3456
3457		if (new_fs) {
3458			fs = current->fs;
3459			spin_lock(&fs->lock);
3460			current->fs = new_fs;
3461			if (--fs->users)
3462				new_fs = NULL;
3463			else
3464				new_fs = fs;
3465			spin_unlock(&fs->lock);
3466		}
3467
3468		if (new_fd)
3469			swap(current->files, new_fd);
 
 
 
3470
3471		task_unlock(current);
3472
3473		if (new_cred) {
3474			/* Install the new user namespace */
3475			commit_creds(new_cred);
3476			new_cred = NULL;
3477		}
3478	}
3479
3480	perf_event_namespaces(current);
3481
3482bad_unshare_cleanup_cred:
3483	if (new_cred)
3484		put_cred(new_cred);
3485bad_unshare_cleanup_fd:
3486	if (new_fd)
3487		put_files_struct(new_fd);
3488
3489bad_unshare_cleanup_fs:
3490	if (new_fs)
3491		free_fs_struct(new_fs);
3492
3493bad_unshare_out:
3494	return err;
3495}
3496
3497SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3498{
3499	return ksys_unshare(unshare_flags);
3500}
3501
3502/*
3503 *	Helper to unshare the files of the current task.
3504 *	We don't want to expose copy_files internals to
3505 *	the exec layer of the kernel.
3506 */
3507
3508int unshare_files(void)
3509{
3510	struct task_struct *task = current;
3511	struct files_struct *old, *copy = NULL;
3512	int error;
3513
3514	error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, &copy);
3515	if (error || !copy)
 
3516		return error;
3517
3518	old = task->files;
3519	task_lock(task);
3520	task->files = copy;
3521	task_unlock(task);
3522	put_files_struct(old);
3523	return 0;
3524}
3525
3526int sysctl_max_threads(struct ctl_table *table, int write,
3527		       void *buffer, size_t *lenp, loff_t *ppos)
3528{
3529	struct ctl_table t;
3530	int ret;
3531	int threads = max_threads;
3532	int min = 1;
3533	int max = MAX_THREADS;
3534
3535	t = *table;
3536	t.data = &threads;
3537	t.extra1 = &min;
3538	t.extra2 = &max;
3539
3540	ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3541	if (ret || !write)
3542		return ret;
3543
3544	max_threads = threads;
3545
3546	return 0;
3547}
v5.9
   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/coredump.h>
  20#include <linux/sched/user.h>
  21#include <linux/sched/numa_balancing.h>
  22#include <linux/sched/stat.h>
  23#include <linux/sched/task.h>
  24#include <linux/sched/task_stack.h>
  25#include <linux/sched/cputime.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/binfmts.h>
  41#include <linux/mman.h>
  42#include <linux/mmu_notifier.h>
  43#include <linux/fs.h>
  44#include <linux/mm.h>
  45#include <linux/vmacache.h>
  46#include <linux/nsproxy.h>
  47#include <linux/capability.h>
  48#include <linux/cpu.h>
  49#include <linux/cgroup.h>
  50#include <linux/security.h>
  51#include <linux/hugetlb.h>
  52#include <linux/seccomp.h>
  53#include <linux/swap.h>
  54#include <linux/syscalls.h>
 
  55#include <linux/jiffies.h>
  56#include <linux/futex.h>
  57#include <linux/compat.h>
  58#include <linux/kthread.h>
  59#include <linux/task_io_accounting_ops.h>
  60#include <linux/rcupdate.h>
  61#include <linux/ptrace.h>
  62#include <linux/mount.h>
  63#include <linux/audit.h>
  64#include <linux/memcontrol.h>
  65#include <linux/ftrace.h>
  66#include <linux/proc_fs.h>
  67#include <linux/profile.h>
  68#include <linux/rmap.h>
  69#include <linux/ksm.h>
  70#include <linux/acct.h>
  71#include <linux/userfaultfd_k.h>
  72#include <linux/tsacct_kern.h>
  73#include <linux/cn_proc.h>
  74#include <linux/freezer.h>
  75#include <linux/delayacct.h>
  76#include <linux/taskstats_kern.h>
  77#include <linux/random.h>
  78#include <linux/tty.h>
  79#include <linux/blkdev.h>
  80#include <linux/fs_struct.h>
  81#include <linux/magic.h>
  82#include <linux/perf_event.h>
  83#include <linux/posix-timers.h>
  84#include <linux/user-return-notifier.h>
  85#include <linux/oom.h>
  86#include <linux/khugepaged.h>
  87#include <linux/signalfd.h>
  88#include <linux/uprobes.h>
  89#include <linux/aio.h>
  90#include <linux/compiler.h>
  91#include <linux/sysctl.h>
  92#include <linux/kcov.h>
  93#include <linux/livepatch.h>
  94#include <linux/thread_info.h>
  95#include <linux/stackleak.h>
  96#include <linux/kasan.h>
  97#include <linux/scs.h>
 
 
 
 
 
 
  98
  99#include <asm/pgalloc.h>
 100#include <linux/uaccess.h>
 101#include <asm/mmu_context.h>
 102#include <asm/cacheflush.h>
 103#include <asm/tlbflush.h>
 104
 105#include <trace/events/sched.h>
 106
 107#define CREATE_TRACE_POINTS
 108#include <trace/events/task.h>
 109
 110/*
 111 * Minimum number of threads to boot the kernel
 112 */
 113#define MIN_THREADS 20
 114
 115/*
 116 * Maximum number of threads
 117 */
 118#define MAX_THREADS FUTEX_TID_MASK
 119
 120/*
 121 * Protected counters by write_lock_irq(&tasklist_lock)
 122 */
 123unsigned long total_forks;	/* Handle normal Linux uptimes. */
 124int nr_threads;			/* The idle threads do not count.. */
 125
 126static int max_threads;		/* tunable limit on nr_threads */
 127
 128#define NAMED_ARRAY_INDEX(x)	[x] = __stringify(x)
 129
 130static const char * const resident_page_types[] = {
 131	NAMED_ARRAY_INDEX(MM_FILEPAGES),
 132	NAMED_ARRAY_INDEX(MM_ANONPAGES),
 133	NAMED_ARRAY_INDEX(MM_SWAPENTS),
 134	NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
 135};
 136
 137DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 138
 139__cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 140
 141#ifdef CONFIG_PROVE_RCU
 142int lockdep_tasklist_lock_is_held(void)
 143{
 144	return lockdep_is_held(&tasklist_lock);
 145}
 146EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 147#endif /* #ifdef CONFIG_PROVE_RCU */
 148
 149int nr_processes(void)
 150{
 151	int cpu;
 152	int total = 0;
 153
 154	for_each_possible_cpu(cpu)
 155		total += per_cpu(process_counts, cpu);
 156
 157	return total;
 158}
 159
 160void __weak arch_release_task_struct(struct task_struct *tsk)
 161{
 162}
 163
 164#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 165static struct kmem_cache *task_struct_cachep;
 166
 167static inline struct task_struct *alloc_task_struct_node(int node)
 168{
 169	return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 170}
 171
 172static inline void free_task_struct(struct task_struct *tsk)
 173{
 174	kmem_cache_free(task_struct_cachep, tsk);
 175}
 176#endif
 177
 178#ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 179
 180/*
 181 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 182 * kmemcache based allocator.
 183 */
 184# if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 185
 186#ifdef CONFIG_VMAP_STACK
 187/*
 188 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 189 * flush.  Try to minimize the number of calls by caching stacks.
 190 */
 191#define NR_CACHED_STACKS 2
 192static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 193
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 194static int free_vm_stack_cache(unsigned int cpu)
 195{
 196	struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 197	int i;
 198
 199	for (i = 0; i < NR_CACHED_STACKS; i++) {
 200		struct vm_struct *vm_stack = cached_vm_stacks[i];
 201
 202		if (!vm_stack)
 203			continue;
 204
 205		vfree(vm_stack->addr);
 206		cached_vm_stacks[i] = NULL;
 207	}
 208
 209	return 0;
 210}
 211#endif
 212
 213static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 214{
 215#ifdef CONFIG_VMAP_STACK
 216	void *stack;
 217	int i;
 218
 219	for (i = 0; i < NR_CACHED_STACKS; i++) {
 220		struct vm_struct *s;
 221
 222		s = this_cpu_xchg(cached_stacks[i], NULL);
 223
 224		if (!s)
 225			continue;
 226
 227		/* Clear the KASAN shadow of the stack. */
 228		kasan_unpoison_shadow(s->addr, THREAD_SIZE);
 
 
 229
 230		/* Clear stale pointers from reused stack. */
 231		memset(s->addr, 0, THREAD_SIZE);
 
 
 
 
 
 232
 233		tsk->stack_vm_area = s;
 234		tsk->stack = s->addr;
 235		return s->addr;
 236	}
 237
 238	/*
 239	 * Allocated stacks are cached and later reused by new threads,
 240	 * so memcg accounting is performed manually on assigning/releasing
 241	 * stacks to tasks. Drop __GFP_ACCOUNT.
 242	 */
 243	stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 244				     VMALLOC_START, VMALLOC_END,
 245				     THREADINFO_GFP & ~__GFP_ACCOUNT,
 246				     PAGE_KERNEL,
 247				     0, node, __builtin_return_address(0));
 
 
 248
 
 
 
 
 
 249	/*
 250	 * We can't call find_vm_area() in interrupt context, and
 251	 * free_thread_stack() can be called in interrupt context,
 252	 * so cache the vm_struct.
 253	 */
 254	if (stack) {
 255		tsk->stack_vm_area = find_vm_area(stack);
 256		tsk->stack = stack;
 257	}
 258	return stack;
 259#else
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 260	struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 261					     THREAD_SIZE_ORDER);
 262
 263	if (likely(page)) {
 264		tsk->stack = kasan_reset_tag(page_address(page));
 265		return tsk->stack;
 266	}
 267	return NULL;
 268#endif
 269}
 270
 271static inline void free_thread_stack(struct task_struct *tsk)
 272{
 273#ifdef CONFIG_VMAP_STACK
 274	struct vm_struct *vm = task_stack_vm_area(tsk);
 
 275
 276	if (vm) {
 277		int i;
 278
 279		for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 280			memcg_kmem_uncharge_page(vm->pages[i], 0);
 281
 282		for (i = 0; i < NR_CACHED_STACKS; i++) {
 283			if (this_cpu_cmpxchg(cached_stacks[i],
 284					NULL, tsk->stack_vm_area) != NULL)
 285				continue;
 286
 287			return;
 288		}
 289
 290		vfree_atomic(tsk->stack);
 291		return;
 292	}
 293#endif
 294
 295	__free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 296}
 297# else
 298static struct kmem_cache *thread_stack_cache;
 299
 300static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 301						  int node)
 302{
 303	unsigned long *stack;
 304	stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 305	stack = kasan_reset_tag(stack);
 306	tsk->stack = stack;
 307	return stack;
 308}
 309
 310static void free_thread_stack(struct task_struct *tsk)
 311{
 312	kmem_cache_free(thread_stack_cache, tsk->stack);
 
 313}
 314
 315void thread_stack_cache_init(void)
 316{
 317	thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 318					THREAD_SIZE, THREAD_SIZE, 0, 0,
 319					THREAD_SIZE, NULL);
 320	BUG_ON(thread_stack_cache == NULL);
 321}
 322# endif
 323#endif
 324
 325/* SLAB cache for signal_struct structures (tsk->signal) */
 326static struct kmem_cache *signal_cachep;
 327
 328/* SLAB cache for sighand_struct structures (tsk->sighand) */
 329struct kmem_cache *sighand_cachep;
 330
 331/* SLAB cache for files_struct structures (tsk->files) */
 332struct kmem_cache *files_cachep;
 333
 334/* SLAB cache for fs_struct structures (tsk->fs) */
 335struct kmem_cache *fs_cachep;
 336
 337/* SLAB cache for vm_area_struct structures */
 338static struct kmem_cache *vm_area_cachep;
 339
 340/* SLAB cache for mm_struct structures (tsk->mm) */
 341static struct kmem_cache *mm_cachep;
 342
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 343struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 344{
 345	struct vm_area_struct *vma;
 346
 347	vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 348	if (vma)
 349		vma_init(vma, mm);
 
 
 
 
 
 
 
 350	return vma;
 351}
 352
 353struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 354{
 355	struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 356
 357	if (new) {
 358		ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
 359		ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
 360		/*
 361		 * orig->shared.rb may be modified concurrently, but the clone
 362		 * will be reinitialized.
 363		 */
 364		*new = data_race(*orig);
 365		INIT_LIST_HEAD(&new->anon_vma_chain);
 366		new->vm_next = new->vm_prev = NULL;
 
 
 
 367	}
 
 
 
 
 368	return new;
 369}
 370
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 371void vm_area_free(struct vm_area_struct *vma)
 372{
 373	kmem_cache_free(vm_area_cachep, vma);
 
 
 
 
 374}
 375
 376static void account_kernel_stack(struct task_struct *tsk, int account)
 377{
 378	void *stack = task_stack_page(tsk);
 379	struct vm_struct *vm = task_stack_vm_area(tsk);
 
 380
 
 
 
 
 
 381
 382	/* All stack pages are in the same node. */
 383	if (vm)
 384		mod_lruvec_page_state(vm->pages[0], NR_KERNEL_STACK_KB,
 385				      account * (THREAD_SIZE / 1024));
 386	else
 387		mod_lruvec_slab_state(stack, NR_KERNEL_STACK_KB,
 388				      account * (THREAD_SIZE / 1024));
 
 389}
 390
 391static int memcg_charge_kernel_stack(struct task_struct *tsk)
 392{
 393#ifdef CONFIG_VMAP_STACK
 394	struct vm_struct *vm = task_stack_vm_area(tsk);
 395	int ret;
 396
 397	BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 398
 399	if (vm) {
 
 400		int i;
 401
 402		BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 403
 404		for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 405			/*
 406			 * If memcg_kmem_charge_page() fails, page->mem_cgroup
 407			 * pointer is NULL, and memcg_kmem_uncharge_page() in
 408			 * free_thread_stack() will ignore this page.
 409			 */
 410			ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL,
 411						     0);
 412			if (ret)
 413				return ret;
 414		}
 415	}
 416#endif
 417	return 0;
 418}
 419
 420static void release_task_stack(struct task_struct *tsk)
 421{
 422	if (WARN_ON(tsk->state != TASK_DEAD))
 423		return;  /* Better to leak the stack than to free prematurely */
 424
 425	account_kernel_stack(tsk, -1);
 426	free_thread_stack(tsk);
 427	tsk->stack = NULL;
 428#ifdef CONFIG_VMAP_STACK
 429	tsk->stack_vm_area = NULL;
 430#endif
 431}
 432
 433#ifdef CONFIG_THREAD_INFO_IN_TASK
 434void put_task_stack(struct task_struct *tsk)
 435{
 436	if (refcount_dec_and_test(&tsk->stack_refcount))
 437		release_task_stack(tsk);
 438}
 439#endif
 440
 441void free_task(struct task_struct *tsk)
 442{
 
 
 
 
 443	scs_release(tsk);
 444
 445#ifndef CONFIG_THREAD_INFO_IN_TASK
 446	/*
 447	 * The task is finally done with both the stack and thread_info,
 448	 * so free both.
 449	 */
 450	release_task_stack(tsk);
 451#else
 452	/*
 453	 * If the task had a separate stack allocation, it should be gone
 454	 * by now.
 455	 */
 456	WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
 457#endif
 458	rt_mutex_debug_task_free(tsk);
 459	ftrace_graph_exit_task(tsk);
 460	arch_release_task_struct(tsk);
 461	if (tsk->flags & PF_KTHREAD)
 462		free_kthread_struct(tsk);
 
 463	free_task_struct(tsk);
 464}
 465EXPORT_SYMBOL(free_task);
 466
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 467#ifdef CONFIG_MMU
 468static __latent_entropy int dup_mmap(struct mm_struct *mm,
 469					struct mm_struct *oldmm)
 470{
 471	struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 472	struct rb_node **rb_link, *rb_parent;
 473	int retval;
 474	unsigned long charge;
 475	LIST_HEAD(uf);
 
 476
 477	uprobe_start_dup_mmap();
 478	if (mmap_write_lock_killable(oldmm)) {
 479		retval = -EINTR;
 480		goto fail_uprobe_end;
 481	}
 482	flush_cache_dup_mm(oldmm);
 483	uprobe_dup_mmap(oldmm, mm);
 484	/*
 485	 * Not linked in yet - no deadlock potential:
 486	 */
 487	mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
 488
 489	/* No ordering required: file already has been exposed. */
 490	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 491
 492	mm->total_vm = oldmm->total_vm;
 493	mm->data_vm = oldmm->data_vm;
 494	mm->exec_vm = oldmm->exec_vm;
 495	mm->stack_vm = oldmm->stack_vm;
 496
 497	rb_link = &mm->mm_rb.rb_node;
 498	rb_parent = NULL;
 499	pprev = &mm->mmap;
 500	retval = ksm_fork(mm, oldmm);
 501	if (retval)
 502		goto out;
 503	retval = khugepaged_fork(mm, oldmm);
 504	if (retval)
 
 
 
 505		goto out;
 506
 507	prev = NULL;
 508	for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 509		struct file *file;
 510
 
 511		if (mpnt->vm_flags & VM_DONTCOPY) {
 
 
 
 
 
 512			vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 513			continue;
 514		}
 515		charge = 0;
 516		/*
 517		 * Don't duplicate many vmas if we've been oom-killed (for
 518		 * example)
 519		 */
 520		if (fatal_signal_pending(current)) {
 521			retval = -EINTR;
 522			goto out;
 523		}
 524		if (mpnt->vm_flags & VM_ACCOUNT) {
 525			unsigned long len = vma_pages(mpnt);
 526
 527			if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 528				goto fail_nomem;
 529			charge = len;
 530		}
 531		tmp = vm_area_dup(mpnt);
 532		if (!tmp)
 533			goto fail_nomem;
 534		retval = vma_dup_policy(mpnt, tmp);
 535		if (retval)
 536			goto fail_nomem_policy;
 537		tmp->vm_mm = mm;
 538		retval = dup_userfaultfd(tmp, &uf);
 539		if (retval)
 540			goto fail_nomem_anon_vma_fork;
 541		if (tmp->vm_flags & VM_WIPEONFORK) {
 542			/*
 543			 * VM_WIPEONFORK gets a clean slate in the child.
 544			 * Don't prepare anon_vma until fault since we don't
 545			 * copy page for current vma.
 546			 */
 547			tmp->anon_vma = NULL;
 548		} else if (anon_vma_fork(tmp, mpnt))
 549			goto fail_nomem_anon_vma_fork;
 550		tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 551		file = tmp->vm_file;
 552		if (file) {
 553			struct inode *inode = file_inode(file);
 554			struct address_space *mapping = file->f_mapping;
 555
 556			get_file(file);
 557			if (tmp->vm_flags & VM_DENYWRITE)
 558				atomic_dec(&inode->i_writecount);
 559			i_mmap_lock_write(mapping);
 560			if (tmp->vm_flags & VM_SHARED)
 561				atomic_inc(&mapping->i_mmap_writable);
 562			flush_dcache_mmap_lock(mapping);
 563			/* insert tmp into the share list, just after mpnt */
 564			vma_interval_tree_insert_after(tmp, mpnt,
 565					&mapping->i_mmap);
 566			flush_dcache_mmap_unlock(mapping);
 567			i_mmap_unlock_write(mapping);
 568		}
 569
 570		/*
 571		 * Clear hugetlb-related page reserves for children. This only
 572		 * affects MAP_PRIVATE mappings. Faults generated by the child
 573		 * are not guaranteed to succeed, even if read-only
 574		 */
 575		if (is_vm_hugetlb_page(tmp))
 576			reset_vma_resv_huge_pages(tmp);
 577
 578		/*
 579		 * Link in the new vma and copy the page table entries.
 
 580		 */
 581		*pprev = tmp;
 582		pprev = &tmp->vm_next;
 583		tmp->vm_prev = prev;
 584		prev = tmp;
 585
 586		__vma_link_rb(mm, tmp, rb_link, rb_parent);
 587		rb_link = &tmp->vm_rb.rb_right;
 588		rb_parent = &tmp->vm_rb;
 589
 590		mm->map_count++;
 591		if (!(tmp->vm_flags & VM_WIPEONFORK))
 592			retval = copy_page_range(mm, oldmm, mpnt, tmp);
 593
 594		if (tmp->vm_ops && tmp->vm_ops->open)
 595			tmp->vm_ops->open(tmp);
 596
 597		if (retval)
 598			goto out;
 
 
 599	}
 600	/* a new mm has just been created */
 601	retval = arch_dup_mmap(oldmm, mm);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 602out:
 603	mmap_write_unlock(mm);
 604	flush_tlb_mm(oldmm);
 605	mmap_write_unlock(oldmm);
 606	dup_userfaultfd_complete(&uf);
 607fail_uprobe_end:
 608	uprobe_end_dup_mmap();
 609	return retval;
 
 610fail_nomem_anon_vma_fork:
 611	mpol_put(vma_policy(tmp));
 612fail_nomem_policy:
 613	vm_area_free(tmp);
 614fail_nomem:
 615	retval = -ENOMEM;
 616	vm_unacct_memory(charge);
 617	goto out;
 618}
 619
 620static inline int mm_alloc_pgd(struct mm_struct *mm)
 621{
 622	mm->pgd = pgd_alloc(mm);
 623	if (unlikely(!mm->pgd))
 624		return -ENOMEM;
 625	return 0;
 626}
 627
 628static inline void mm_free_pgd(struct mm_struct *mm)
 629{
 630	pgd_free(mm, mm->pgd);
 631}
 632#else
 633static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 634{
 635	mmap_write_lock(oldmm);
 636	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 637	mmap_write_unlock(oldmm);
 638	return 0;
 639}
 640#define mm_alloc_pgd(mm)	(0)
 641#define mm_free_pgd(mm)
 642#endif /* CONFIG_MMU */
 643
 644static void check_mm(struct mm_struct *mm)
 645{
 646	int i;
 647
 648	BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
 649			 "Please make sure 'struct resident_page_types[]' is updated as well");
 650
 651	for (i = 0; i < NR_MM_COUNTERS; i++) {
 652		long x = atomic_long_read(&mm->rss_stat.count[i]);
 653
 654		if (unlikely(x))
 655			pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
 656				 mm, resident_page_types[i], x);
 657	}
 658
 659	if (mm_pgtables_bytes(mm))
 660		pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 661				mm_pgtables_bytes(mm));
 662
 663#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 664	VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 665#endif
 666}
 667
 668#define allocate_mm()	(kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 669#define free_mm(mm)	(kmem_cache_free(mm_cachep, (mm)))
 670
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 671/*
 672 * Called when the last reference to the mm
 673 * is dropped: either by a lazy thread or by
 674 * mmput. Free the page directory and the mm.
 675 */
 676void __mmdrop(struct mm_struct *mm)
 677{
 678	BUG_ON(mm == &init_mm);
 679	WARN_ON_ONCE(mm == current->mm);
 
 
 
 
 680	WARN_ON_ONCE(mm == current->active_mm);
 681	mm_free_pgd(mm);
 682	destroy_context(mm);
 683	mmu_notifier_subscriptions_destroy(mm);
 684	check_mm(mm);
 685	put_user_ns(mm->user_ns);
 
 
 
 
 686	free_mm(mm);
 687}
 688EXPORT_SYMBOL_GPL(__mmdrop);
 689
 690static void mmdrop_async_fn(struct work_struct *work)
 691{
 692	struct mm_struct *mm;
 693
 694	mm = container_of(work, struct mm_struct, async_put_work);
 695	__mmdrop(mm);
 696}
 697
 698static void mmdrop_async(struct mm_struct *mm)
 699{
 700	if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 701		INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 702		schedule_work(&mm->async_put_work);
 703	}
 704}
 705
 706static inline void free_signal_struct(struct signal_struct *sig)
 707{
 708	taskstats_tgid_free(sig);
 709	sched_autogroup_exit(sig);
 710	/*
 711	 * __mmdrop is not safe to call from softirq context on x86 due to
 712	 * pgd_dtor so postpone it to the async context
 713	 */
 714	if (sig->oom_mm)
 715		mmdrop_async(sig->oom_mm);
 716	kmem_cache_free(signal_cachep, sig);
 717}
 718
 719static inline void put_signal_struct(struct signal_struct *sig)
 720{
 721	if (refcount_dec_and_test(&sig->sigcnt))
 722		free_signal_struct(sig);
 723}
 724
 725void __put_task_struct(struct task_struct *tsk)
 726{
 727	WARN_ON(!tsk->exit_state);
 728	WARN_ON(refcount_read(&tsk->usage));
 729	WARN_ON(tsk == current);
 730
 
 731	cgroup_free(tsk);
 732	task_numa_free(tsk, true);
 733	security_task_free(tsk);
 734	exit_creds(tsk);
 735	delayacct_tsk_free(tsk);
 736	put_signal_struct(tsk->signal);
 
 
 
 
 737
 738	if (!profile_handoff_task(tsk))
 739		free_task(tsk);
 
 
 
 740}
 741EXPORT_SYMBOL_GPL(__put_task_struct);
 742
 743void __init __weak arch_task_cache_init(void) { }
 744
 745/*
 746 * set_max_threads
 747 */
 748static void set_max_threads(unsigned int max_threads_suggested)
 749{
 750	u64 threads;
 751	unsigned long nr_pages = totalram_pages();
 752
 753	/*
 754	 * The number of threads shall be limited such that the thread
 755	 * structures may only consume a small part of the available memory.
 756	 */
 757	if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
 758		threads = MAX_THREADS;
 759	else
 760		threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
 761				    (u64) THREAD_SIZE * 8UL);
 762
 763	if (threads > max_threads_suggested)
 764		threads = max_threads_suggested;
 765
 766	max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 767}
 768
 769#ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 770/* Initialized by the architecture: */
 771int arch_task_struct_size __read_mostly;
 772#endif
 773
 774#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 775static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 776{
 777	/* Fetch thread_struct whitelist for the architecture. */
 778	arch_thread_struct_whitelist(offset, size);
 779
 780	/*
 781	 * Handle zero-sized whitelist or empty thread_struct, otherwise
 782	 * adjust offset to position of thread_struct in task_struct.
 783	 */
 784	if (unlikely(*size == 0))
 785		*offset = 0;
 786	else
 787		*offset += offsetof(struct task_struct, thread);
 788}
 789#endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
 790
 791void __init fork_init(void)
 792{
 793	int i;
 794#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 795#ifndef ARCH_MIN_TASKALIGN
 796#define ARCH_MIN_TASKALIGN	0
 797#endif
 798	int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 799	unsigned long useroffset, usersize;
 800
 801	/* create a slab on which task_structs can be allocated */
 802	task_struct_whitelist(&useroffset, &usersize);
 803	task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 804			arch_task_struct_size, align,
 805			SLAB_PANIC|SLAB_ACCOUNT,
 806			useroffset, usersize, NULL);
 807#endif
 808
 809	/* do the arch specific task caches init */
 810	arch_task_cache_init();
 811
 812	set_max_threads(MAX_THREADS);
 813
 814	init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 815	init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 816	init_task.signal->rlim[RLIMIT_SIGPENDING] =
 817		init_task.signal->rlim[RLIMIT_NPROC];
 818
 819	for (i = 0; i < UCOUNT_COUNTS; i++) {
 820		init_user_ns.ucount_max[i] = max_threads/2;
 821	}
 
 
 
 
 822
 823#ifdef CONFIG_VMAP_STACK
 824	cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 825			  NULL, free_vm_stack_cache);
 826#endif
 827
 828	scs_init();
 829
 830	lockdep_init_task(&init_task);
 831	uprobes_init();
 832}
 833
 834int __weak arch_dup_task_struct(struct task_struct *dst,
 835					       struct task_struct *src)
 836{
 837	*dst = *src;
 838	return 0;
 839}
 840
 841void set_task_stack_end_magic(struct task_struct *tsk)
 842{
 843	unsigned long *stackend;
 844
 845	stackend = end_of_stack(tsk);
 846	*stackend = STACK_END_MAGIC;	/* for overflow detection */
 847}
 848
 849static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 850{
 851	struct task_struct *tsk;
 852	unsigned long *stack;
 853	struct vm_struct *stack_vm_area __maybe_unused;
 854	int err;
 855
 856	if (node == NUMA_NO_NODE)
 857		node = tsk_fork_get_node(orig);
 858	tsk = alloc_task_struct_node(node);
 859	if (!tsk)
 860		return NULL;
 861
 862	stack = alloc_thread_stack_node(tsk, node);
 863	if (!stack)
 864		goto free_tsk;
 865
 866	if (memcg_charge_kernel_stack(tsk))
 867		goto free_stack;
 
 868
 869	stack_vm_area = task_stack_vm_area(tsk);
 870
 871	err = arch_dup_task_struct(tsk, orig);
 872
 873	/*
 874	 * arch_dup_task_struct() clobbers the stack-related fields.  Make
 875	 * sure they're properly initialized before using any stack-related
 876	 * functions again.
 877	 */
 878	tsk->stack = stack;
 879#ifdef CONFIG_VMAP_STACK
 880	tsk->stack_vm_area = stack_vm_area;
 881#endif
 882#ifdef CONFIG_THREAD_INFO_IN_TASK
 883	refcount_set(&tsk->stack_refcount, 1);
 884#endif
 885
 886	if (err)
 887		goto free_stack;
 888
 889	err = scs_prepare(tsk, node);
 890	if (err)
 891		goto free_stack;
 892
 893#ifdef CONFIG_SECCOMP
 894	/*
 895	 * We must handle setting up seccomp filters once we're under
 896	 * the sighand lock in case orig has changed between now and
 897	 * then. Until then, filter must be NULL to avoid messing up
 898	 * the usage counts on the error path calling free_task.
 899	 */
 900	tsk->seccomp.filter = NULL;
 901#endif
 902
 903	setup_thread_stack(tsk, orig);
 904	clear_user_return_notifier(tsk);
 905	clear_tsk_need_resched(tsk);
 906	set_task_stack_end_magic(tsk);
 
 907
 908#ifdef CONFIG_STACKPROTECTOR
 909	tsk->stack_canary = get_random_canary();
 910#endif
 911	if (orig->cpus_ptr == &orig->cpus_mask)
 912		tsk->cpus_ptr = &tsk->cpus_mask;
 
 913
 914	/*
 915	 * One for the user space visible state that goes away when reaped.
 916	 * One for the scheduler.
 917	 */
 918	refcount_set(&tsk->rcu_users, 2);
 919	/* One for the rcu users */
 920	refcount_set(&tsk->usage, 1);
 921#ifdef CONFIG_BLK_DEV_IO_TRACE
 922	tsk->btrace_seq = 0;
 923#endif
 924	tsk->splice_pipe = NULL;
 925	tsk->task_frag.page = NULL;
 926	tsk->wake_q.next = NULL;
 927
 928	account_kernel_stack(tsk, 1);
 929
 930	kcov_task_init(tsk);
 
 
 931
 932#ifdef CONFIG_FAULT_INJECTION
 933	tsk->fail_nth = 0;
 934#endif
 935
 936#ifdef CONFIG_BLK_CGROUP
 937	tsk->throttle_queue = NULL;
 938	tsk->use_memdelay = 0;
 939#endif
 940
 
 
 
 
 941#ifdef CONFIG_MEMCG
 942	tsk->active_memcg = NULL;
 943#endif
 
 
 
 
 
 
 
 
 
 
 
 944	return tsk;
 945
 946free_stack:
 
 947	free_thread_stack(tsk);
 948free_tsk:
 949	free_task_struct(tsk);
 950	return NULL;
 951}
 952
 953__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 954
 955static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 956
 957static int __init coredump_filter_setup(char *s)
 958{
 959	default_dump_filter =
 960		(simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 961		MMF_DUMP_FILTER_MASK;
 962	return 1;
 963}
 964
 965__setup("coredump_filter=", coredump_filter_setup);
 966
 967#include <linux/init_task.h>
 968
 969static void mm_init_aio(struct mm_struct *mm)
 970{
 971#ifdef CONFIG_AIO
 972	spin_lock_init(&mm->ioctx_lock);
 973	mm->ioctx_table = NULL;
 974#endif
 975}
 976
 977static __always_inline void mm_clear_owner(struct mm_struct *mm,
 978					   struct task_struct *p)
 979{
 980#ifdef CONFIG_MEMCG
 981	if (mm->owner == p)
 982		WRITE_ONCE(mm->owner, NULL);
 983#endif
 984}
 985
 986static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 987{
 988#ifdef CONFIG_MEMCG
 989	mm->owner = p;
 990#endif
 991}
 992
 993static void mm_init_uprobes_state(struct mm_struct *mm)
 994{
 995#ifdef CONFIG_UPROBES
 996	mm->uprobes_state.xol_area = NULL;
 997#endif
 998}
 999
1000static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1001	struct user_namespace *user_ns)
1002{
1003	mm->mmap = NULL;
1004	mm->mm_rb = RB_ROOT;
1005	mm->vmacache_seqnum = 0;
1006	atomic_set(&mm->mm_users, 1);
1007	atomic_set(&mm->mm_count, 1);
 
1008	mmap_init_lock(mm);
1009	INIT_LIST_HEAD(&mm->mmlist);
1010	mm->core_state = NULL;
 
 
1011	mm_pgtables_bytes_init(mm);
1012	mm->map_count = 0;
1013	mm->locked_vm = 0;
1014	atomic_set(&mm->has_pinned, 0);
1015	atomic64_set(&mm->pinned_vm, 0);
1016	memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1017	spin_lock_init(&mm->page_table_lock);
1018	spin_lock_init(&mm->arg_lock);
1019	mm_init_cpumask(mm);
1020	mm_init_aio(mm);
1021	mm_init_owner(mm, p);
 
1022	RCU_INIT_POINTER(mm->exe_file, NULL);
1023	mmu_notifier_subscriptions_init(mm);
1024	init_tlb_flush_pending(mm);
1025#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1026	mm->pmd_huge_pte = NULL;
1027#endif
1028	mm_init_uprobes_state(mm);
 
1029
1030	if (current->mm) {
1031		mm->flags = current->mm->flags & MMF_INIT_MASK;
1032		mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1033	} else {
1034		mm->flags = default_dump_filter;
1035		mm->def_flags = 0;
1036	}
1037
1038	if (mm_alloc_pgd(mm))
1039		goto fail_nopgd;
1040
1041	if (init_new_context(p, mm))
1042		goto fail_nocontext;
1043
 
 
 
 
 
 
 
1044	mm->user_ns = get_user_ns(user_ns);
 
1045	return mm;
1046
 
 
 
 
1047fail_nocontext:
1048	mm_free_pgd(mm);
1049fail_nopgd:
1050	free_mm(mm);
1051	return NULL;
1052}
1053
1054/*
1055 * Allocate and initialize an mm_struct.
1056 */
1057struct mm_struct *mm_alloc(void)
1058{
1059	struct mm_struct *mm;
1060
1061	mm = allocate_mm();
1062	if (!mm)
1063		return NULL;
1064
1065	memset(mm, 0, sizeof(*mm));
1066	return mm_init(mm, current, current_user_ns());
1067}
1068
1069static inline void __mmput(struct mm_struct *mm)
1070{
1071	VM_BUG_ON(atomic_read(&mm->mm_users));
1072
1073	uprobe_clear_state(mm);
1074	exit_aio(mm);
1075	ksm_exit(mm);
1076	khugepaged_exit(mm); /* must run before exit_mmap */
1077	exit_mmap(mm);
1078	mm_put_huge_zero_page(mm);
1079	set_mm_exe_file(mm, NULL);
1080	if (!list_empty(&mm->mmlist)) {
1081		spin_lock(&mmlist_lock);
1082		list_del(&mm->mmlist);
1083		spin_unlock(&mmlist_lock);
1084	}
1085	if (mm->binfmt)
1086		module_put(mm->binfmt->module);
 
1087	mmdrop(mm);
1088}
1089
1090/*
1091 * Decrement the use count and release all resources for an mm.
1092 */
1093void mmput(struct mm_struct *mm)
1094{
1095	might_sleep();
1096
1097	if (atomic_dec_and_test(&mm->mm_users))
1098		__mmput(mm);
1099}
1100EXPORT_SYMBOL_GPL(mmput);
1101
1102#ifdef CONFIG_MMU
1103static void mmput_async_fn(struct work_struct *work)
1104{
1105	struct mm_struct *mm = container_of(work, struct mm_struct,
1106					    async_put_work);
1107
1108	__mmput(mm);
1109}
1110
1111void mmput_async(struct mm_struct *mm)
1112{
1113	if (atomic_dec_and_test(&mm->mm_users)) {
1114		INIT_WORK(&mm->async_put_work, mmput_async_fn);
1115		schedule_work(&mm->async_put_work);
1116	}
1117}
 
1118#endif
1119
1120/**
1121 * set_mm_exe_file - change a reference to the mm's executable file
 
 
1122 *
1123 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1124 *
1125 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1126 * invocations: in mmput() nobody alive left, in execve task is single
1127 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1128 * mm->exe_file, but does so without using set_mm_exe_file() in order
1129 * to do avoid the need for any locks.
1130 */
1131void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1132{
1133	struct file *old_exe_file;
1134
1135	/*
1136	 * It is safe to dereference the exe_file without RCU as
1137	 * this function is only called if nobody else can access
1138	 * this mm -- see comment above for justification.
1139	 */
1140	old_exe_file = rcu_dereference_raw(mm->exe_file);
1141
1142	if (new_exe_file)
 
 
 
 
 
 
1143		get_file(new_exe_file);
 
1144	rcu_assign_pointer(mm->exe_file, new_exe_file);
1145	if (old_exe_file)
 
1146		fput(old_exe_file);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1147}
1148
1149/**
1150 * get_mm_exe_file - acquire a reference to the mm's executable file
 
1151 *
1152 * Returns %NULL if mm has no associated executable file.
1153 * User must release file via fput().
1154 */
1155struct file *get_mm_exe_file(struct mm_struct *mm)
1156{
1157	struct file *exe_file;
1158
1159	rcu_read_lock();
1160	exe_file = rcu_dereference(mm->exe_file);
1161	if (exe_file && !get_file_rcu(exe_file))
1162		exe_file = NULL;
1163	rcu_read_unlock();
1164	return exe_file;
1165}
1166EXPORT_SYMBOL(get_mm_exe_file);
1167
1168/**
1169 * get_task_exe_file - acquire a reference to the task's executable file
 
1170 *
1171 * Returns %NULL if task's mm (if any) has no associated executable file or
1172 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1173 * User must release file via fput().
1174 */
1175struct file *get_task_exe_file(struct task_struct *task)
1176{
1177	struct file *exe_file = NULL;
1178	struct mm_struct *mm;
1179
1180	task_lock(task);
1181	mm = task->mm;
1182	if (mm) {
1183		if (!(task->flags & PF_KTHREAD))
1184			exe_file = get_mm_exe_file(mm);
1185	}
1186	task_unlock(task);
1187	return exe_file;
1188}
1189EXPORT_SYMBOL(get_task_exe_file);
1190
1191/**
1192 * get_task_mm - acquire a reference to the task's mm
 
1193 *
1194 * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1195 * this kernel workthread has transiently adopted a user mm with use_mm,
1196 * to do its AIO) is not set and if so returns a reference to it, after
1197 * bumping up the use count.  User must release the mm via mmput()
1198 * after use.  Typically used by /proc and ptrace.
1199 */
1200struct mm_struct *get_task_mm(struct task_struct *task)
1201{
1202	struct mm_struct *mm;
1203
1204	task_lock(task);
1205	mm = task->mm;
1206	if (mm) {
1207		if (task->flags & PF_KTHREAD)
1208			mm = NULL;
1209		else
1210			mmget(mm);
1211	}
1212	task_unlock(task);
1213	return mm;
1214}
1215EXPORT_SYMBOL_GPL(get_task_mm);
1216
1217struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1218{
1219	struct mm_struct *mm;
1220	int err;
1221
1222	err =  mutex_lock_killable(&task->signal->exec_update_mutex);
1223	if (err)
1224		return ERR_PTR(err);
1225
1226	mm = get_task_mm(task);
1227	if (mm && mm != current->mm &&
1228			!ptrace_may_access(task, mode)) {
1229		mmput(mm);
1230		mm = ERR_PTR(-EACCES);
1231	}
1232	mutex_unlock(&task->signal->exec_update_mutex);
1233
1234	return mm;
1235}
1236
1237static void complete_vfork_done(struct task_struct *tsk)
1238{
1239	struct completion *vfork;
1240
1241	task_lock(tsk);
1242	vfork = tsk->vfork_done;
1243	if (likely(vfork)) {
1244		tsk->vfork_done = NULL;
1245		complete(vfork);
1246	}
1247	task_unlock(tsk);
1248}
1249
1250static int wait_for_vfork_done(struct task_struct *child,
1251				struct completion *vfork)
1252{
 
1253	int killed;
1254
1255	freezer_do_not_count();
1256	cgroup_enter_frozen();
1257	killed = wait_for_completion_killable(vfork);
1258	cgroup_leave_frozen(false);
1259	freezer_count();
1260
1261	if (killed) {
1262		task_lock(child);
1263		child->vfork_done = NULL;
1264		task_unlock(child);
1265	}
1266
1267	put_task_struct(child);
1268	return killed;
1269}
1270
1271/* Please note the differences between mmput and mm_release.
1272 * mmput is called whenever we stop holding onto a mm_struct,
1273 * error success whatever.
1274 *
1275 * mm_release is called after a mm_struct has been removed
1276 * from the current process.
1277 *
1278 * This difference is important for error handling, when we
1279 * only half set up a mm_struct for a new process and need to restore
1280 * the old one.  Because we mmput the new mm_struct before
1281 * restoring the old one. . .
1282 * Eric Biederman 10 January 1998
1283 */
1284static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1285{
1286	uprobe_free_utask(tsk);
1287
1288	/* Get rid of any cached register state */
1289	deactivate_mm(tsk, mm);
1290
1291	/*
1292	 * Signal userspace if we're not exiting with a core dump
1293	 * because we want to leave the value intact for debugging
1294	 * purposes.
1295	 */
1296	if (tsk->clear_child_tid) {
1297		if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1298		    atomic_read(&mm->mm_users) > 1) {
1299			/*
1300			 * We don't check the error code - if userspace has
1301			 * not set up a proper pointer then tough luck.
1302			 */
1303			put_user(0, tsk->clear_child_tid);
1304			do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1305					1, NULL, NULL, 0, 0);
1306		}
1307		tsk->clear_child_tid = NULL;
1308	}
1309
1310	/*
1311	 * All done, finally we can wake up parent and return this mm to him.
1312	 * Also kthread_stop() uses this completion for synchronization.
1313	 */
1314	if (tsk->vfork_done)
1315		complete_vfork_done(tsk);
1316}
1317
1318void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1319{
1320	futex_exit_release(tsk);
1321	mm_release(tsk, mm);
1322}
1323
1324void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1325{
1326	futex_exec_release(tsk);
1327	mm_release(tsk, mm);
1328}
1329
1330/**
1331 * dup_mm() - duplicates an existing mm structure
1332 * @tsk: the task_struct with which the new mm will be associated.
1333 * @oldmm: the mm to duplicate.
1334 *
1335 * Allocates a new mm structure and duplicates the provided @oldmm structure
1336 * content into it.
1337 *
1338 * Return: the duplicated mm or NULL on failure.
1339 */
1340static struct mm_struct *dup_mm(struct task_struct *tsk,
1341				struct mm_struct *oldmm)
1342{
1343	struct mm_struct *mm;
1344	int err;
1345
1346	mm = allocate_mm();
1347	if (!mm)
1348		goto fail_nomem;
1349
1350	memcpy(mm, oldmm, sizeof(*mm));
1351
1352	if (!mm_init(mm, tsk, mm->user_ns))
1353		goto fail_nomem;
1354
1355	err = dup_mmap(mm, oldmm);
1356	if (err)
1357		goto free_pt;
1358
1359	mm->hiwater_rss = get_mm_rss(mm);
1360	mm->hiwater_vm = mm->total_vm;
1361
1362	if (mm->binfmt && !try_module_get(mm->binfmt->module))
1363		goto free_pt;
1364
1365	return mm;
1366
1367free_pt:
1368	/* don't put binfmt in mmput, we haven't got module yet */
1369	mm->binfmt = NULL;
1370	mm_init_owner(mm, NULL);
1371	mmput(mm);
1372
1373fail_nomem:
1374	return NULL;
1375}
1376
1377static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1378{
1379	struct mm_struct *mm, *oldmm;
1380	int retval;
1381
1382	tsk->min_flt = tsk->maj_flt = 0;
1383	tsk->nvcsw = tsk->nivcsw = 0;
1384#ifdef CONFIG_DETECT_HUNG_TASK
1385	tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1386	tsk->last_switch_time = 0;
1387#endif
1388
1389	tsk->mm = NULL;
1390	tsk->active_mm = NULL;
1391
1392	/*
1393	 * Are we cloning a kernel thread?
1394	 *
1395	 * We need to steal a active VM for that..
1396	 */
1397	oldmm = current->mm;
1398	if (!oldmm)
1399		return 0;
1400
1401	/* initialize the new vmacache entries */
1402	vmacache_flush(tsk);
1403
1404	if (clone_flags & CLONE_VM) {
1405		mmget(oldmm);
1406		mm = oldmm;
1407		goto good_mm;
 
 
 
1408	}
1409
1410	retval = -ENOMEM;
1411	mm = dup_mm(tsk, current->mm);
1412	if (!mm)
1413		goto fail_nomem;
1414
1415good_mm:
1416	tsk->mm = mm;
1417	tsk->active_mm = mm;
 
1418	return 0;
1419
1420fail_nomem:
1421	return retval;
1422}
1423
1424static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1425{
1426	struct fs_struct *fs = current->fs;
1427	if (clone_flags & CLONE_FS) {
1428		/* tsk->fs is already what we want */
1429		spin_lock(&fs->lock);
 
1430		if (fs->in_exec) {
1431			spin_unlock(&fs->lock);
1432			return -EAGAIN;
1433		}
1434		fs->users++;
1435		spin_unlock(&fs->lock);
1436		return 0;
1437	}
1438	tsk->fs = copy_fs_struct(fs);
1439	if (!tsk->fs)
1440		return -ENOMEM;
1441	return 0;
1442}
1443
1444static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 
1445{
1446	struct files_struct *oldf, *newf;
1447	int error = 0;
1448
1449	/*
1450	 * A background process may not have any files ...
1451	 */
1452	oldf = current->files;
1453	if (!oldf)
1454		goto out;
1455
 
 
 
 
 
1456	if (clone_flags & CLONE_FILES) {
1457		atomic_inc(&oldf->count);
1458		goto out;
1459	}
1460
1461	newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1462	if (!newf)
1463		goto out;
1464
1465	tsk->files = newf;
1466	error = 0;
1467out:
1468	return error;
1469}
1470
1471static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1472{
1473#ifdef CONFIG_BLOCK
1474	struct io_context *ioc = current->io_context;
1475	struct io_context *new_ioc;
1476
1477	if (!ioc)
1478		return 0;
1479	/*
1480	 * Share io context with parent, if CLONE_IO is set
1481	 */
1482	if (clone_flags & CLONE_IO) {
1483		ioc_task_link(ioc);
1484		tsk->io_context = ioc;
1485	} else if (ioprio_valid(ioc->ioprio)) {
1486		new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1487		if (unlikely(!new_ioc))
1488			return -ENOMEM;
1489
1490		new_ioc->ioprio = ioc->ioprio;
1491		put_io_context(new_ioc);
1492	}
1493#endif
1494	return 0;
1495}
1496
1497static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1498{
1499	struct sighand_struct *sig;
1500
1501	if (clone_flags & CLONE_SIGHAND) {
1502		refcount_inc(&current->sighand->count);
1503		return 0;
1504	}
1505	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1506	RCU_INIT_POINTER(tsk->sighand, sig);
1507	if (!sig)
1508		return -ENOMEM;
1509
1510	refcount_set(&sig->count, 1);
1511	spin_lock_irq(&current->sighand->siglock);
1512	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1513	spin_unlock_irq(&current->sighand->siglock);
1514
1515	/* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1516	if (clone_flags & CLONE_CLEAR_SIGHAND)
1517		flush_signal_handlers(tsk, 0);
1518
1519	return 0;
1520}
1521
1522void __cleanup_sighand(struct sighand_struct *sighand)
1523{
1524	if (refcount_dec_and_test(&sighand->count)) {
1525		signalfd_cleanup(sighand);
1526		/*
1527		 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1528		 * without an RCU grace period, see __lock_task_sighand().
1529		 */
1530		kmem_cache_free(sighand_cachep, sighand);
1531	}
1532}
1533
1534/*
1535 * Initialize POSIX timer handling for a thread group.
1536 */
1537static void posix_cpu_timers_init_group(struct signal_struct *sig)
1538{
1539	struct posix_cputimers *pct = &sig->posix_cputimers;
1540	unsigned long cpu_limit;
1541
1542	cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1543	posix_cputimers_group_init(pct, cpu_limit);
1544}
1545
1546static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1547{
1548	struct signal_struct *sig;
1549
1550	if (clone_flags & CLONE_THREAD)
1551		return 0;
1552
1553	sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1554	tsk->signal = sig;
1555	if (!sig)
1556		return -ENOMEM;
1557
1558	sig->nr_threads = 1;
 
1559	atomic_set(&sig->live, 1);
1560	refcount_set(&sig->sigcnt, 1);
1561
1562	/* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1563	sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1564	tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1565
1566	init_waitqueue_head(&sig->wait_chldexit);
1567	sig->curr_target = tsk;
1568	init_sigpending(&sig->shared_pending);
1569	INIT_HLIST_HEAD(&sig->multiprocess);
1570	seqlock_init(&sig->stats_lock);
1571	prev_cputime_init(&sig->prev_cputime);
1572
1573#ifdef CONFIG_POSIX_TIMERS
1574	INIT_LIST_HEAD(&sig->posix_timers);
1575	hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1576	sig->real_timer.function = it_real_fn;
1577#endif
1578
1579	task_lock(current->group_leader);
1580	memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1581	task_unlock(current->group_leader);
1582
1583	posix_cpu_timers_init_group(sig);
1584
1585	tty_audit_fork(sig);
1586	sched_autogroup_fork(sig);
1587
1588	sig->oom_score_adj = current->signal->oom_score_adj;
1589	sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1590
1591	mutex_init(&sig->cred_guard_mutex);
1592	mutex_init(&sig->exec_update_mutex);
1593
1594	return 0;
1595}
1596
1597static void copy_seccomp(struct task_struct *p)
1598{
1599#ifdef CONFIG_SECCOMP
1600	/*
1601	 * Must be called with sighand->lock held, which is common to
1602	 * all threads in the group. Holding cred_guard_mutex is not
1603	 * needed because this new task is not yet running and cannot
1604	 * be racing exec.
1605	 */
1606	assert_spin_locked(&current->sighand->siglock);
1607
1608	/* Ref-count the new filter user, and assign it. */
1609	get_seccomp_filter(current);
1610	p->seccomp = current->seccomp;
1611
1612	/*
1613	 * Explicitly enable no_new_privs here in case it got set
1614	 * between the task_struct being duplicated and holding the
1615	 * sighand lock. The seccomp state and nnp must be in sync.
1616	 */
1617	if (task_no_new_privs(current))
1618		task_set_no_new_privs(p);
1619
1620	/*
1621	 * If the parent gained a seccomp mode after copying thread
1622	 * flags and between before we held the sighand lock, we have
1623	 * to manually enable the seccomp thread flag here.
1624	 */
1625	if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1626		set_tsk_thread_flag(p, TIF_SECCOMP);
1627#endif
1628}
1629
1630SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1631{
1632	current->clear_child_tid = tidptr;
1633
1634	return task_pid_vnr(current);
1635}
1636
1637static void rt_mutex_init_task(struct task_struct *p)
1638{
1639	raw_spin_lock_init(&p->pi_lock);
1640#ifdef CONFIG_RT_MUTEXES
1641	p->pi_waiters = RB_ROOT_CACHED;
1642	p->pi_top_task = NULL;
1643	p->pi_blocked_on = NULL;
1644#endif
1645}
1646
1647static inline void init_task_pid_links(struct task_struct *task)
1648{
1649	enum pid_type type;
1650
1651	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1652		INIT_HLIST_NODE(&task->pid_links[type]);
1653	}
1654}
1655
1656static inline void
1657init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1658{
1659	if (type == PIDTYPE_PID)
1660		task->thread_pid = pid;
1661	else
1662		task->signal->pids[type] = pid;
1663}
1664
1665static inline void rcu_copy_process(struct task_struct *p)
1666{
1667#ifdef CONFIG_PREEMPT_RCU
1668	p->rcu_read_lock_nesting = 0;
1669	p->rcu_read_unlock_special.s = 0;
1670	p->rcu_blocked_node = NULL;
1671	INIT_LIST_HEAD(&p->rcu_node_entry);
1672#endif /* #ifdef CONFIG_PREEMPT_RCU */
1673#ifdef CONFIG_TASKS_RCU
1674	p->rcu_tasks_holdout = false;
1675	INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1676	p->rcu_tasks_idle_cpu = -1;
1677#endif /* #ifdef CONFIG_TASKS_RCU */
1678#ifdef CONFIG_TASKS_TRACE_RCU
1679	p->trc_reader_nesting = 0;
1680	p->trc_reader_special.s = 0;
1681	INIT_LIST_HEAD(&p->trc_holdout_list);
 
1682#endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1683}
1684
1685struct pid *pidfd_pid(const struct file *file)
1686{
1687	if (file->f_op == &pidfd_fops)
1688		return file->private_data;
1689
1690	return ERR_PTR(-EBADF);
1691}
1692
1693static int pidfd_release(struct inode *inode, struct file *file)
1694{
1695	struct pid *pid = file->private_data;
1696
1697	file->private_data = NULL;
1698	put_pid(pid);
1699	return 0;
1700}
1701
1702#ifdef CONFIG_PROC_FS
1703/**
1704 * pidfd_show_fdinfo - print information about a pidfd
1705 * @m: proc fdinfo file
1706 * @f: file referencing a pidfd
1707 *
1708 * Pid:
1709 * This function will print the pid that a given pidfd refers to in the
1710 * pid namespace of the procfs instance.
1711 * If the pid namespace of the process is not a descendant of the pid
1712 * namespace of the procfs instance 0 will be shown as its pid. This is
1713 * similar to calling getppid() on a process whose parent is outside of
1714 * its pid namespace.
1715 *
1716 * NSpid:
1717 * If pid namespaces are supported then this function will also print
1718 * the pid of a given pidfd refers to for all descendant pid namespaces
1719 * starting from the current pid namespace of the instance, i.e. the
1720 * Pid field and the first entry in the NSpid field will be identical.
1721 * If the pid namespace of the process is not a descendant of the pid
1722 * namespace of the procfs instance 0 will be shown as its first NSpid
1723 * entry and no others will be shown.
1724 * Note that this differs from the Pid and NSpid fields in
1725 * /proc/<pid>/status where Pid and NSpid are always shown relative to
1726 * the  pid namespace of the procfs instance. The difference becomes
1727 * obvious when sending around a pidfd between pid namespaces from a
1728 * different branch of the tree, i.e. where no ancestoral relation is
1729 * present between the pid namespaces:
1730 * - create two new pid namespaces ns1 and ns2 in the initial pid
1731 *   namespace (also take care to create new mount namespaces in the
1732 *   new pid namespace and mount procfs)
1733 * - create a process with a pidfd in ns1
1734 * - send pidfd from ns1 to ns2
1735 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1736 *   have exactly one entry, which is 0
1737 */
1738static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1739{
1740	struct pid *pid = f->private_data;
1741	struct pid_namespace *ns;
1742	pid_t nr = -1;
1743
1744	if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1745		ns = proc_pid_ns(file_inode(m->file)->i_sb);
1746		nr = pid_nr_ns(pid, ns);
1747	}
1748
1749	seq_put_decimal_ll(m, "Pid:\t", nr);
1750
1751#ifdef CONFIG_PID_NS
1752	seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1753	if (nr > 0) {
1754		int i;
1755
1756		/* If nr is non-zero it means that 'pid' is valid and that
1757		 * ns, i.e. the pid namespace associated with the procfs
1758		 * instance, is in the pid namespace hierarchy of pid.
1759		 * Start at one below the already printed level.
1760		 */
1761		for (i = ns->level + 1; i <= pid->level; i++)
1762			seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1763	}
1764#endif
1765	seq_putc(m, '\n');
1766}
1767#endif
1768
1769/*
1770 * Poll support for process exit notification.
1771 */
1772static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1773{
1774	struct pid *pid = file->private_data;
1775	__poll_t poll_flags = 0;
1776
1777	poll_wait(file, &pid->wait_pidfd, pts);
1778
1779	/*
1780	 * Inform pollers only when the whole thread group exits.
1781	 * If the thread group leader exits before all other threads in the
1782	 * group, then poll(2) should block, similar to the wait(2) family.
1783	 */
1784	if (thread_group_exited(pid))
1785		poll_flags = EPOLLIN | EPOLLRDNORM;
1786
1787	return poll_flags;
1788}
1789
1790const struct file_operations pidfd_fops = {
1791	.release = pidfd_release,
1792	.poll = pidfd_poll,
1793#ifdef CONFIG_PROC_FS
1794	.show_fdinfo = pidfd_show_fdinfo,
1795#endif
1796};
1797
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1798static void __delayed_free_task(struct rcu_head *rhp)
1799{
1800	struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1801
1802	free_task(tsk);
1803}
1804
1805static __always_inline void delayed_free_task(struct task_struct *tsk)
1806{
1807	if (IS_ENABLED(CONFIG_MEMCG))
1808		call_rcu(&tsk->rcu, __delayed_free_task);
1809	else
1810		free_task(tsk);
1811}
1812
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1813/*
1814 * This creates a new process as a copy of the old one,
1815 * but does not actually start it yet.
1816 *
1817 * It copies the registers, and all the appropriate
1818 * parts of the process environment (as per the clone
1819 * flags). The actual kick-off is left to the caller.
1820 */
1821static __latent_entropy struct task_struct *copy_process(
1822					struct pid *pid,
1823					int trace,
1824					int node,
1825					struct kernel_clone_args *args)
1826{
1827	int pidfd = -1, retval;
1828	struct task_struct *p;
1829	struct multiprocess_signals delayed;
1830	struct file *pidfile = NULL;
1831	u64 clone_flags = args->flags;
1832	struct nsproxy *nsp = current->nsproxy;
1833
1834	/*
1835	 * Don't allow sharing the root directory with processes in a different
1836	 * namespace
1837	 */
1838	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1839		return ERR_PTR(-EINVAL);
1840
1841	if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1842		return ERR_PTR(-EINVAL);
1843
1844	/*
1845	 * Thread groups must share signals as well, and detached threads
1846	 * can only be started up within the thread group.
1847	 */
1848	if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1849		return ERR_PTR(-EINVAL);
1850
1851	/*
1852	 * Shared signal handlers imply shared VM. By way of the above,
1853	 * thread groups also imply shared VM. Blocking this case allows
1854	 * for various simplifications in other code.
1855	 */
1856	if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1857		return ERR_PTR(-EINVAL);
1858
1859	/*
1860	 * Siblings of global init remain as zombies on exit since they are
1861	 * not reaped by their parent (swapper). To solve this and to avoid
1862	 * multi-rooted process trees, prevent global and container-inits
1863	 * from creating siblings.
1864	 */
1865	if ((clone_flags & CLONE_PARENT) &&
1866				current->signal->flags & SIGNAL_UNKILLABLE)
1867		return ERR_PTR(-EINVAL);
1868
1869	/*
1870	 * If the new process will be in a different pid or user namespace
1871	 * do not allow it to share a thread group with the forking task.
1872	 */
1873	if (clone_flags & CLONE_THREAD) {
1874		if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1875		    (task_active_pid_ns(current) != nsp->pid_ns_for_children))
1876			return ERR_PTR(-EINVAL);
1877	}
1878
1879	/*
1880	 * If the new process will be in a different time namespace
1881	 * do not allow it to share VM or a thread group with the forking task.
1882	 */
1883	if (clone_flags & (CLONE_THREAD | CLONE_VM)) {
1884		if (nsp->time_ns != nsp->time_ns_for_children)
1885			return ERR_PTR(-EINVAL);
1886	}
1887
1888	if (clone_flags & CLONE_PIDFD) {
1889		/*
1890		 * - CLONE_DETACHED is blocked so that we can potentially
1891		 *   reuse it later for CLONE_PIDFD.
1892		 * - CLONE_THREAD is blocked until someone really needs it.
1893		 */
1894		if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1895			return ERR_PTR(-EINVAL);
1896	}
1897
1898	/*
1899	 * Force any signals received before this point to be delivered
1900	 * before the fork happens.  Collect up signals sent to multiple
1901	 * processes that happen during the fork and delay them so that
1902	 * they appear to happen after the fork.
1903	 */
1904	sigemptyset(&delayed.signal);
1905	INIT_HLIST_NODE(&delayed.node);
1906
1907	spin_lock_irq(&current->sighand->siglock);
1908	if (!(clone_flags & CLONE_THREAD))
1909		hlist_add_head(&delayed.node, &current->signal->multiprocess);
1910	recalc_sigpending();
1911	spin_unlock_irq(&current->sighand->siglock);
1912	retval = -ERESTARTNOINTR;
1913	if (signal_pending(current))
1914		goto fork_out;
1915
1916	retval = -ENOMEM;
1917	p = dup_task_struct(current, node);
1918	if (!p)
1919		goto fork_out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1920
1921	/*
1922	 * This _must_ happen before we call free_task(), i.e. before we jump
1923	 * to any of the bad_fork_* labels. This is to avoid freeing
1924	 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1925	 * kernel threads (PF_KTHREAD).
1926	 */
1927	p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1928	/*
1929	 * Clear TID on mm_release()?
1930	 */
1931	p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1932
1933	ftrace_graph_init_task(p);
1934
1935	rt_mutex_init_task(p);
1936
1937	lockdep_assert_irqs_enabled();
1938#ifdef CONFIG_PROVE_LOCKING
1939	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1940#endif
 
 
 
 
1941	retval = -EAGAIN;
1942	if (atomic_read(&p->real_cred->user->processes) >=
1943			task_rlimit(p, RLIMIT_NPROC)) {
1944		if (p->real_cred->user != INIT_USER &&
1945		    !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1946			goto bad_fork_free;
1947	}
1948	current->flags &= ~PF_NPROC_EXCEEDED;
1949
1950	retval = copy_creds(p, clone_flags);
1951	if (retval < 0)
1952		goto bad_fork_free;
1953
1954	/*
1955	 * If multiple threads are within copy_process(), then this check
1956	 * triggers too late. This doesn't hurt, the check is only there
1957	 * to stop root fork bombs.
1958	 */
1959	retval = -EAGAIN;
1960	if (data_race(nr_threads >= max_threads))
1961		goto bad_fork_cleanup_count;
1962
1963	delayacct_tsk_init(p);	/* Must remain after dup_task_struct() */
1964	p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1965	p->flags |= PF_FORKNOEXEC;
1966	INIT_LIST_HEAD(&p->children);
1967	INIT_LIST_HEAD(&p->sibling);
1968	rcu_copy_process(p);
1969	p->vfork_done = NULL;
1970	spin_lock_init(&p->alloc_lock);
1971
1972	init_sigpending(&p->pending);
1973
1974	p->utime = p->stime = p->gtime = 0;
1975#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1976	p->utimescaled = p->stimescaled = 0;
1977#endif
1978	prev_cputime_init(&p->prev_cputime);
1979
1980#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1981	seqcount_init(&p->vtime.seqcount);
1982	p->vtime.starttime = 0;
1983	p->vtime.state = VTIME_INACTIVE;
1984#endif
1985
1986#if defined(SPLIT_RSS_COUNTING)
1987	memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1988#endif
1989
1990	p->default_timer_slack_ns = current->timer_slack_ns;
1991
1992#ifdef CONFIG_PSI
1993	p->psi_flags = 0;
1994#endif
1995
1996	task_io_accounting_init(&p->ioac);
1997	acct_clear_integrals(p);
1998
1999	posix_cputimers_init(&p->posix_cputimers);
2000
2001	p->io_context = NULL;
2002	audit_set_context(p, NULL);
2003	cgroup_fork(p);
 
 
 
 
2004#ifdef CONFIG_NUMA
2005	p->mempolicy = mpol_dup(p->mempolicy);
2006	if (IS_ERR(p->mempolicy)) {
2007		retval = PTR_ERR(p->mempolicy);
2008		p->mempolicy = NULL;
2009		goto bad_fork_cleanup_threadgroup_lock;
2010	}
2011#endif
2012#ifdef CONFIG_CPUSETS
2013	p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2014	p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2015	seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2016#endif
2017#ifdef CONFIG_TRACE_IRQFLAGS
2018	memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2019	p->irqtrace.hardirq_disable_ip	= _THIS_IP_;
2020	p->irqtrace.softirq_enable_ip	= _THIS_IP_;
2021	p->softirqs_enabled		= 1;
2022	p->softirq_context		= 0;
2023#endif
2024
2025	p->pagefault_disabled = 0;
2026
2027#ifdef CONFIG_LOCKDEP
2028	lockdep_init_task(p);
2029#endif
2030
2031#ifdef CONFIG_DEBUG_MUTEXES
2032	p->blocked_on = NULL; /* not blocked yet */
2033#endif
2034#ifdef CONFIG_BCACHE
2035	p->sequential_io	= 0;
2036	p->sequential_io_avg	= 0;
2037#endif
 
 
 
 
2038
2039	/* Perform scheduler related setup. Assign this task to a CPU. */
2040	retval = sched_fork(clone_flags, p);
2041	if (retval)
2042		goto bad_fork_cleanup_policy;
2043
2044	retval = perf_event_init_task(p);
2045	if (retval)
2046		goto bad_fork_cleanup_policy;
2047	retval = audit_alloc(p);
2048	if (retval)
2049		goto bad_fork_cleanup_perf;
2050	/* copy all the process information */
2051	shm_init_task(p);
2052	retval = security_task_alloc(p, clone_flags);
2053	if (retval)
2054		goto bad_fork_cleanup_audit;
2055	retval = copy_semundo(clone_flags, p);
2056	if (retval)
2057		goto bad_fork_cleanup_security;
2058	retval = copy_files(clone_flags, p);
2059	if (retval)
2060		goto bad_fork_cleanup_semundo;
2061	retval = copy_fs(clone_flags, p);
2062	if (retval)
2063		goto bad_fork_cleanup_files;
2064	retval = copy_sighand(clone_flags, p);
2065	if (retval)
2066		goto bad_fork_cleanup_fs;
2067	retval = copy_signal(clone_flags, p);
2068	if (retval)
2069		goto bad_fork_cleanup_sighand;
2070	retval = copy_mm(clone_flags, p);
2071	if (retval)
2072		goto bad_fork_cleanup_signal;
2073	retval = copy_namespaces(clone_flags, p);
2074	if (retval)
2075		goto bad_fork_cleanup_mm;
2076	retval = copy_io(clone_flags, p);
2077	if (retval)
2078		goto bad_fork_cleanup_namespaces;
2079	retval = copy_thread(clone_flags, args->stack, args->stack_size, p, args->tls);
2080	if (retval)
2081		goto bad_fork_cleanup_io;
2082
2083	stackleak_task_init(p);
2084
2085	if (pid != &init_struct_pid) {
2086		pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2087				args->set_tid_size);
2088		if (IS_ERR(pid)) {
2089			retval = PTR_ERR(pid);
2090			goto bad_fork_cleanup_thread;
2091		}
2092	}
2093
2094	/*
2095	 * This has to happen after we've potentially unshared the file
2096	 * descriptor table (so that the pidfd doesn't leak into the child
2097	 * if the fd table isn't shared).
2098	 */
2099	if (clone_flags & CLONE_PIDFD) {
2100		retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
 
2101		if (retval < 0)
2102			goto bad_fork_free_pid;
2103
2104		pidfd = retval;
2105
2106		pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2107					      O_RDWR | O_CLOEXEC);
2108		if (IS_ERR(pidfile)) {
2109			put_unused_fd(pidfd);
2110			retval = PTR_ERR(pidfile);
2111			goto bad_fork_free_pid;
2112		}
2113		get_pid(pid);	/* held by pidfile now */
2114
2115		retval = put_user(pidfd, args->pidfd);
2116		if (retval)
2117			goto bad_fork_put_pidfd;
2118	}
2119
2120#ifdef CONFIG_BLOCK
2121	p->plug = NULL;
2122#endif
2123	futex_init_task(p);
2124
2125	/*
2126	 * sigaltstack should be cleared when sharing the same VM
2127	 */
2128	if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2129		sas_ss_reset(p);
2130
2131	/*
2132	 * Syscall tracing and stepping should be turned off in the
2133	 * child regardless of CLONE_PTRACE.
2134	 */
2135	user_disable_single_step(p);
2136	clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
2137#ifdef TIF_SYSCALL_EMU
2138	clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
2139#endif
2140	clear_tsk_latency_tracing(p);
2141
2142	/* ok, now we should be set up.. */
2143	p->pid = pid_nr(pid);
2144	if (clone_flags & CLONE_THREAD) {
2145		p->exit_signal = -1;
2146		p->group_leader = current->group_leader;
2147		p->tgid = current->tgid;
2148	} else {
2149		if (clone_flags & CLONE_PARENT)
2150			p->exit_signal = current->group_leader->exit_signal;
2151		else
2152			p->exit_signal = args->exit_signal;
2153		p->group_leader = p;
2154		p->tgid = p->pid;
2155	}
2156
2157	p->nr_dirtied = 0;
2158	p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2159	p->dirty_paused_when = 0;
2160
2161	p->pdeath_signal = 0;
2162	INIT_LIST_HEAD(&p->thread_group);
2163	p->task_works = NULL;
 
 
 
 
 
 
 
 
2164
2165	/*
2166	 * Ensure that the cgroup subsystem policies allow the new process to be
2167	 * forked. It should be noted the the new process's css_set can be changed
2168	 * between here and cgroup_post_fork() if an organisation operation is in
2169	 * progress.
2170	 */
2171	retval = cgroup_can_fork(p, args);
2172	if (retval)
2173		goto bad_fork_put_pidfd;
2174
2175	/*
 
 
 
 
 
 
 
 
 
 
 
2176	 * From this point on we must avoid any synchronous user-space
2177	 * communication until we take the tasklist-lock. In particular, we do
2178	 * not want user-space to be able to predict the process start-time by
2179	 * stalling fork(2) after we recorded the start_time but before it is
2180	 * visible to the system.
2181	 */
2182
2183	p->start_time = ktime_get_ns();
2184	p->start_boottime = ktime_get_boottime_ns();
2185
2186	/*
2187	 * Make it visible to the rest of the system, but dont wake it up yet.
2188	 * Need tasklist lock for parent etc handling!
2189	 */
2190	write_lock_irq(&tasklist_lock);
2191
2192	/* CLONE_PARENT re-uses the old parent */
2193	if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2194		p->real_parent = current->real_parent;
2195		p->parent_exec_id = current->parent_exec_id;
 
 
 
 
2196	} else {
2197		p->real_parent = current;
2198		p->parent_exec_id = current->self_exec_id;
 
2199	}
2200
2201	klp_copy_process(p);
2202
 
 
2203	spin_lock(&current->sighand->siglock);
2204
2205	/*
2206	 * Copy seccomp details explicitly here, in case they were changed
2207	 * before holding sighand lock.
2208	 */
2209	copy_seccomp(p);
2210
2211	rseq_fork(p, clone_flags);
2212
2213	/* Don't start children in a dying pid namespace */
2214	if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2215		retval = -ENOMEM;
2216		goto bad_fork_cancel_cgroup;
2217	}
2218
2219	/* Let kill terminate clone/fork in the middle */
2220	if (fatal_signal_pending(current)) {
2221		retval = -EINTR;
2222		goto bad_fork_cancel_cgroup;
2223	}
2224
2225	/* past the last point of failure */
2226	if (pidfile)
2227		fd_install(pidfd, pidfile);
 
 
 
 
2228
2229	init_task_pid_links(p);
2230	if (likely(p->pid)) {
2231		ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2232
2233		init_task_pid(p, PIDTYPE_PID, pid);
2234		if (thread_group_leader(p)) {
2235			init_task_pid(p, PIDTYPE_TGID, pid);
2236			init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2237			init_task_pid(p, PIDTYPE_SID, task_session(current));
2238
2239			if (is_child_reaper(pid)) {
2240				ns_of_pid(pid)->child_reaper = p;
2241				p->signal->flags |= SIGNAL_UNKILLABLE;
2242			}
2243			p->signal->shared_pending.signal = delayed.signal;
2244			p->signal->tty = tty_kref_get(current->signal->tty);
2245			/*
2246			 * Inherit has_child_subreaper flag under the same
2247			 * tasklist_lock with adding child to the process tree
2248			 * for propagate_has_child_subreaper optimization.
2249			 */
2250			p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2251							 p->real_parent->signal->is_child_subreaper;
2252			list_add_tail(&p->sibling, &p->real_parent->children);
2253			list_add_tail_rcu(&p->tasks, &init_task.tasks);
2254			attach_pid(p, PIDTYPE_TGID);
2255			attach_pid(p, PIDTYPE_PGID);
2256			attach_pid(p, PIDTYPE_SID);
2257			__this_cpu_inc(process_counts);
2258		} else {
2259			current->signal->nr_threads++;
 
2260			atomic_inc(&current->signal->live);
2261			refcount_inc(&current->signal->sigcnt);
2262			task_join_group_stop(p);
2263			list_add_tail_rcu(&p->thread_group,
2264					  &p->group_leader->thread_group);
2265			list_add_tail_rcu(&p->thread_node,
2266					  &p->signal->thread_head);
2267		}
2268		attach_pid(p, PIDTYPE_PID);
2269		nr_threads++;
2270	}
2271	total_forks++;
2272	hlist_del_init(&delayed.node);
2273	spin_unlock(&current->sighand->siglock);
2274	syscall_tracepoint_update(p);
2275	write_unlock_irq(&tasklist_lock);
2276
 
 
 
2277	proc_fork_connector(p);
2278	sched_post_fork(p);
2279	cgroup_post_fork(p, args);
2280	perf_event_fork(p);
2281
2282	trace_task_newtask(p, clone_flags);
2283	uprobe_copy_process(p, clone_flags);
 
 
 
2284
2285	return p;
2286
2287bad_fork_cancel_cgroup:
 
2288	spin_unlock(&current->sighand->siglock);
2289	write_unlock_irq(&tasklist_lock);
2290	cgroup_cancel_fork(p, args);
2291bad_fork_put_pidfd:
2292	if (clone_flags & CLONE_PIDFD) {
2293		fput(pidfile);
2294		put_unused_fd(pidfd);
2295	}
2296bad_fork_free_pid:
2297	if (pid != &init_struct_pid)
2298		free_pid(pid);
2299bad_fork_cleanup_thread:
2300	exit_thread(p);
2301bad_fork_cleanup_io:
2302	if (p->io_context)
2303		exit_io_context(p);
2304bad_fork_cleanup_namespaces:
2305	exit_task_namespaces(p);
2306bad_fork_cleanup_mm:
2307	if (p->mm) {
2308		mm_clear_owner(p->mm, p);
2309		mmput(p->mm);
2310	}
2311bad_fork_cleanup_signal:
2312	if (!(clone_flags & CLONE_THREAD))
2313		free_signal_struct(p->signal);
2314bad_fork_cleanup_sighand:
2315	__cleanup_sighand(p->sighand);
2316bad_fork_cleanup_fs:
2317	exit_fs(p); /* blocking */
2318bad_fork_cleanup_files:
2319	exit_files(p); /* blocking */
2320bad_fork_cleanup_semundo:
2321	exit_sem(p);
2322bad_fork_cleanup_security:
2323	security_task_free(p);
2324bad_fork_cleanup_audit:
2325	audit_free(p);
2326bad_fork_cleanup_perf:
2327	perf_event_free_task(p);
2328bad_fork_cleanup_policy:
2329	lockdep_free_task(p);
2330#ifdef CONFIG_NUMA
2331	mpol_put(p->mempolicy);
2332bad_fork_cleanup_threadgroup_lock:
2333#endif
 
2334	delayacct_tsk_free(p);
2335bad_fork_cleanup_count:
2336	atomic_dec(&p->cred->user->processes);
2337	exit_creds(p);
2338bad_fork_free:
2339	p->state = TASK_DEAD;
 
2340	put_task_stack(p);
2341	delayed_free_task(p);
2342fork_out:
2343	spin_lock_irq(&current->sighand->siglock);
2344	hlist_del_init(&delayed.node);
2345	spin_unlock_irq(&current->sighand->siglock);
2346	return ERR_PTR(retval);
2347}
2348
2349static inline void init_idle_pids(struct task_struct *idle)
2350{
2351	enum pid_type type;
2352
2353	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2354		INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2355		init_task_pid(idle, type, &init_struct_pid);
2356	}
2357}
2358
2359struct task_struct *fork_idle(int cpu)
 
 
 
 
 
 
2360{
2361	struct task_struct *task;
2362	struct kernel_clone_args args = {
2363		.flags = CLONE_VM,
 
 
 
 
2364	};
2365
2366	task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2367	if (!IS_ERR(task)) {
2368		init_idle_pids(task);
2369		init_idle(task, cpu);
2370	}
2371
2372	return task;
2373}
2374
2375struct mm_struct *copy_init_mm(void)
 
 
 
 
 
 
2376{
2377	return dup_mm(NULL, &init_mm);
 
 
 
 
 
 
 
 
 
 
 
 
2378}
2379
2380/*
2381 *  Ok, this is the main fork-routine.
2382 *
2383 * It copies the process, and if successful kick-starts
2384 * it and waits for it to finish using the VM if required.
2385 *
2386 * args->exit_signal is expected to be checked for sanity by the caller.
2387 */
2388long _do_fork(struct kernel_clone_args *args)
2389{
2390	u64 clone_flags = args->flags;
2391	struct completion vfork;
2392	struct pid *pid;
2393	struct task_struct *p;
2394	int trace = 0;
2395	long nr;
2396
2397	/*
2398	 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2399	 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2400	 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2401	 * field in struct clone_args and it still doesn't make sense to have
2402	 * them both point at the same memory location. Performing this check
2403	 * here has the advantage that we don't need to have a separate helper
2404	 * to check for legacy clone().
2405	 */
2406	if ((args->flags & CLONE_PIDFD) &&
2407	    (args->flags & CLONE_PARENT_SETTID) &&
2408	    (args->pidfd == args->parent_tid))
2409		return -EINVAL;
2410
2411	/*
2412	 * Determine whether and which event to report to ptracer.  When
2413	 * called from kernel_thread or CLONE_UNTRACED is explicitly
2414	 * requested, no event is reported; otherwise, report if the event
2415	 * for the type of forking is enabled.
2416	 */
2417	if (!(clone_flags & CLONE_UNTRACED)) {
2418		if (clone_flags & CLONE_VFORK)
2419			trace = PTRACE_EVENT_VFORK;
2420		else if (args->exit_signal != SIGCHLD)
2421			trace = PTRACE_EVENT_CLONE;
2422		else
2423			trace = PTRACE_EVENT_FORK;
2424
2425		if (likely(!ptrace_event_enabled(current, trace)))
2426			trace = 0;
2427	}
2428
2429	p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2430	add_latent_entropy();
2431
2432	if (IS_ERR(p))
2433		return PTR_ERR(p);
2434
2435	/*
2436	 * Do this prior waking up the new thread - the thread pointer
2437	 * might get invalid after that point, if the thread exits quickly.
2438	 */
2439	trace_sched_process_fork(current, p);
2440
2441	pid = get_task_pid(p, PIDTYPE_PID);
2442	nr = pid_vnr(pid);
2443
2444	if (clone_flags & CLONE_PARENT_SETTID)
2445		put_user(nr, args->parent_tid);
2446
2447	if (clone_flags & CLONE_VFORK) {
2448		p->vfork_done = &vfork;
2449		init_completion(&vfork);
2450		get_task_struct(p);
2451	}
2452
 
 
 
 
 
 
 
2453	wake_up_new_task(p);
2454
2455	/* forking complete and child started to run, tell ptracer */
2456	if (unlikely(trace))
2457		ptrace_event_pid(trace, pid);
2458
2459	if (clone_flags & CLONE_VFORK) {
2460		if (!wait_for_vfork_done(p, &vfork))
2461			ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2462	}
2463
2464	put_pid(pid);
2465	return nr;
2466}
2467
2468/*
2469 * Create a kernel thread.
2470 */
2471pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2472{
2473	struct kernel_clone_args args = {
2474		.flags		= ((lower_32_bits(flags) | CLONE_VM |
2475				    CLONE_UNTRACED) & ~CSIGNAL),
2476		.exit_signal	= (lower_32_bits(flags) & CSIGNAL),
2477		.stack		= (unsigned long)fn,
2478		.stack_size	= (unsigned long)arg,
2479	};
2480
2481	return _do_fork(&args);
2482}
2483
2484#ifdef __ARCH_WANT_SYS_FORK
2485SYSCALL_DEFINE0(fork)
2486{
2487#ifdef CONFIG_MMU
2488	struct kernel_clone_args args = {
2489		.exit_signal = SIGCHLD,
2490	};
2491
2492	return _do_fork(&args);
2493#else
2494	/* can not support in nommu mode */
2495	return -EINVAL;
2496#endif
2497}
2498#endif
2499
2500#ifdef __ARCH_WANT_SYS_VFORK
2501SYSCALL_DEFINE0(vfork)
2502{
2503	struct kernel_clone_args args = {
2504		.flags		= CLONE_VFORK | CLONE_VM,
2505		.exit_signal	= SIGCHLD,
2506	};
2507
2508	return _do_fork(&args);
2509}
2510#endif
2511
2512#ifdef __ARCH_WANT_SYS_CLONE
2513#ifdef CONFIG_CLONE_BACKWARDS
2514SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2515		 int __user *, parent_tidptr,
2516		 unsigned long, tls,
2517		 int __user *, child_tidptr)
2518#elif defined(CONFIG_CLONE_BACKWARDS2)
2519SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2520		 int __user *, parent_tidptr,
2521		 int __user *, child_tidptr,
2522		 unsigned long, tls)
2523#elif defined(CONFIG_CLONE_BACKWARDS3)
2524SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2525		int, stack_size,
2526		int __user *, parent_tidptr,
2527		int __user *, child_tidptr,
2528		unsigned long, tls)
2529#else
2530SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2531		 int __user *, parent_tidptr,
2532		 int __user *, child_tidptr,
2533		 unsigned long, tls)
2534#endif
2535{
2536	struct kernel_clone_args args = {
2537		.flags		= (lower_32_bits(clone_flags) & ~CSIGNAL),
2538		.pidfd		= parent_tidptr,
2539		.child_tid	= child_tidptr,
2540		.parent_tid	= parent_tidptr,
2541		.exit_signal	= (lower_32_bits(clone_flags) & CSIGNAL),
2542		.stack		= newsp,
2543		.tls		= tls,
2544	};
2545
2546	return _do_fork(&args);
2547}
2548#endif
2549
2550#ifdef __ARCH_WANT_SYS_CLONE3
2551
2552noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2553					      struct clone_args __user *uargs,
2554					      size_t usize)
2555{
2556	int err;
2557	struct clone_args args;
2558	pid_t *kset_tid = kargs->set_tid;
2559
2560	BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2561		     CLONE_ARGS_SIZE_VER0);
2562	BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2563		     CLONE_ARGS_SIZE_VER1);
2564	BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2565		     CLONE_ARGS_SIZE_VER2);
2566	BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2567
2568	if (unlikely(usize > PAGE_SIZE))
2569		return -E2BIG;
2570	if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2571		return -EINVAL;
2572
2573	err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2574	if (err)
2575		return err;
2576
2577	if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2578		return -EINVAL;
2579
2580	if (unlikely(!args.set_tid && args.set_tid_size > 0))
2581		return -EINVAL;
2582
2583	if (unlikely(args.set_tid && args.set_tid_size == 0))
2584		return -EINVAL;
2585
2586	/*
2587	 * Verify that higher 32bits of exit_signal are unset and that
2588	 * it is a valid signal
2589	 */
2590	if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2591		     !valid_signal(args.exit_signal)))
2592		return -EINVAL;
2593
2594	if ((args.flags & CLONE_INTO_CGROUP) &&
2595	    (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2596		return -EINVAL;
2597
2598	*kargs = (struct kernel_clone_args){
2599		.flags		= args.flags,
2600		.pidfd		= u64_to_user_ptr(args.pidfd),
2601		.child_tid	= u64_to_user_ptr(args.child_tid),
2602		.parent_tid	= u64_to_user_ptr(args.parent_tid),
2603		.exit_signal	= args.exit_signal,
2604		.stack		= args.stack,
2605		.stack_size	= args.stack_size,
2606		.tls		= args.tls,
2607		.set_tid_size	= args.set_tid_size,
2608		.cgroup		= args.cgroup,
2609	};
2610
2611	if (args.set_tid &&
2612		copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2613			(kargs->set_tid_size * sizeof(pid_t))))
2614		return -EFAULT;
2615
2616	kargs->set_tid = kset_tid;
2617
2618	return 0;
2619}
2620
2621/**
2622 * clone3_stack_valid - check and prepare stack
2623 * @kargs: kernel clone args
2624 *
2625 * Verify that the stack arguments userspace gave us are sane.
2626 * In addition, set the stack direction for userspace since it's easy for us to
2627 * determine.
2628 */
2629static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2630{
2631	if (kargs->stack == 0) {
2632		if (kargs->stack_size > 0)
2633			return false;
2634	} else {
2635		if (kargs->stack_size == 0)
2636			return false;
2637
2638		if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2639			return false;
2640
2641#if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2642		kargs->stack += kargs->stack_size;
2643#endif
2644	}
2645
2646	return true;
2647}
2648
2649static bool clone3_args_valid(struct kernel_clone_args *kargs)
2650{
2651	/* Verify that no unknown flags are passed along. */
2652	if (kargs->flags &
2653	    ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2654		return false;
2655
2656	/*
2657	 * - make the CLONE_DETACHED bit reuseable for clone3
2658	 * - make the CSIGNAL bits reuseable for clone3
2659	 */
2660	if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2661		return false;
2662
2663	if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2664	    (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2665		return false;
2666
2667	if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2668	    kargs->exit_signal)
2669		return false;
2670
2671	if (!clone3_stack_valid(kargs))
2672		return false;
2673
2674	return true;
2675}
2676
2677/**
2678 * clone3 - create a new process with specific properties
2679 * @uargs: argument structure
2680 * @size:  size of @uargs
2681 *
2682 * clone3() is the extensible successor to clone()/clone2().
2683 * It takes a struct as argument that is versioned by its size.
2684 *
2685 * Return: On success, a positive PID for the child process.
2686 *         On error, a negative errno number.
2687 */
2688SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2689{
2690	int err;
2691
2692	struct kernel_clone_args kargs;
2693	pid_t set_tid[MAX_PID_NS_LEVEL];
2694
2695	kargs.set_tid = set_tid;
2696
2697	err = copy_clone_args_from_user(&kargs, uargs, size);
2698	if (err)
2699		return err;
2700
2701	if (!clone3_args_valid(&kargs))
2702		return -EINVAL;
2703
2704	return _do_fork(&kargs);
2705}
2706#endif
2707
2708void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2709{
2710	struct task_struct *leader, *parent, *child;
2711	int res;
2712
2713	read_lock(&tasklist_lock);
2714	leader = top = top->group_leader;
2715down:
2716	for_each_thread(leader, parent) {
2717		list_for_each_entry(child, &parent->children, sibling) {
2718			res = visitor(child, data);
2719			if (res) {
2720				if (res < 0)
2721					goto out;
2722				leader = child;
2723				goto down;
2724			}
2725up:
2726			;
2727		}
2728	}
2729
2730	if (leader != top) {
2731		child = leader;
2732		parent = child->real_parent;
2733		leader = parent->group_leader;
2734		goto up;
2735	}
2736out:
2737	read_unlock(&tasklist_lock);
2738}
2739
2740#ifndef ARCH_MIN_MMSTRUCT_ALIGN
2741#define ARCH_MIN_MMSTRUCT_ALIGN 0
2742#endif
2743
2744static void sighand_ctor(void *data)
2745{
2746	struct sighand_struct *sighand = data;
2747
2748	spin_lock_init(&sighand->siglock);
2749	init_waitqueue_head(&sighand->signalfd_wqh);
2750}
2751
2752void __init proc_caches_init(void)
2753{
2754	unsigned int mm_size;
2755
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2756	sighand_cachep = kmem_cache_create("sighand_cache",
2757			sizeof(struct sighand_struct), 0,
2758			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2759			SLAB_ACCOUNT, sighand_ctor);
2760	signal_cachep = kmem_cache_create("signal_cache",
2761			sizeof(struct signal_struct), 0,
2762			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2763			NULL);
2764	files_cachep = kmem_cache_create("files_cache",
2765			sizeof(struct files_struct), 0,
2766			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2767			NULL);
2768	fs_cachep = kmem_cache_create("fs_cache",
2769			sizeof(struct fs_struct), 0,
2770			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2771			NULL);
2772
2773	/*
2774	 * The mm_cpumask is located at the end of mm_struct, and is
2775	 * dynamically sized based on the maximum CPU number this system
2776	 * can have, taking hotplug into account (nr_cpu_ids).
2777	 */
2778	mm_size = sizeof(struct mm_struct) + cpumask_size();
2779
2780	mm_cachep = kmem_cache_create_usercopy("mm_struct",
2781			mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2782			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2783			offsetof(struct mm_struct, saved_auxv),
2784			sizeof_field(struct mm_struct, saved_auxv),
2785			NULL);
2786	vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
 
 
 
2787	mmap_init();
2788	nsproxy_cache_init();
2789}
2790
2791/*
2792 * Check constraints on flags passed to the unshare system call.
2793 */
2794static int check_unshare_flags(unsigned long unshare_flags)
2795{
2796	if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2797				CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2798				CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2799				CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
2800				CLONE_NEWTIME))
2801		return -EINVAL;
2802	/*
2803	 * Not implemented, but pretend it works if there is nothing
2804	 * to unshare.  Note that unsharing the address space or the
2805	 * signal handlers also need to unshare the signal queues (aka
2806	 * CLONE_THREAD).
2807	 */
2808	if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2809		if (!thread_group_empty(current))
2810			return -EINVAL;
2811	}
2812	if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2813		if (refcount_read(&current->sighand->count) > 1)
2814			return -EINVAL;
2815	}
2816	if (unshare_flags & CLONE_VM) {
2817		if (!current_is_single_threaded())
2818			return -EINVAL;
2819	}
2820
2821	return 0;
2822}
2823
2824/*
2825 * Unshare the filesystem structure if it is being shared
2826 */
2827static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2828{
2829	struct fs_struct *fs = current->fs;
2830
2831	if (!(unshare_flags & CLONE_FS) || !fs)
2832		return 0;
2833
2834	/* don't need lock here; in the worst case we'll do useless copy */
2835	if (fs->users == 1)
2836		return 0;
2837
2838	*new_fsp = copy_fs_struct(fs);
2839	if (!*new_fsp)
2840		return -ENOMEM;
2841
2842	return 0;
2843}
2844
2845/*
2846 * Unshare file descriptor table if it is being shared
2847 */
2848int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
2849	       struct files_struct **new_fdp)
2850{
2851	struct files_struct *fd = current->files;
2852	int error = 0;
2853
2854	if ((unshare_flags & CLONE_FILES) &&
2855	    (fd && atomic_read(&fd->count) > 1)) {
2856		*new_fdp = dup_fd(fd, max_fds, &error);
2857		if (!*new_fdp)
2858			return error;
2859	}
2860
2861	return 0;
2862}
2863
2864/*
2865 * unshare allows a process to 'unshare' part of the process
2866 * context which was originally shared using clone.  copy_*
2867 * functions used by _do_fork() cannot be used here directly
2868 * because they modify an inactive task_struct that is being
2869 * constructed. Here we are modifying the current, active,
2870 * task_struct.
2871 */
2872int ksys_unshare(unsigned long unshare_flags)
2873{
2874	struct fs_struct *fs, *new_fs = NULL;
2875	struct files_struct *fd, *new_fd = NULL;
2876	struct cred *new_cred = NULL;
2877	struct nsproxy *new_nsproxy = NULL;
2878	int do_sysvsem = 0;
2879	int err;
2880
2881	/*
2882	 * If unsharing a user namespace must also unshare the thread group
2883	 * and unshare the filesystem root and working directories.
2884	 */
2885	if (unshare_flags & CLONE_NEWUSER)
2886		unshare_flags |= CLONE_THREAD | CLONE_FS;
2887	/*
2888	 * If unsharing vm, must also unshare signal handlers.
2889	 */
2890	if (unshare_flags & CLONE_VM)
2891		unshare_flags |= CLONE_SIGHAND;
2892	/*
2893	 * If unsharing a signal handlers, must also unshare the signal queues.
2894	 */
2895	if (unshare_flags & CLONE_SIGHAND)
2896		unshare_flags |= CLONE_THREAD;
2897	/*
2898	 * If unsharing namespace, must also unshare filesystem information.
2899	 */
2900	if (unshare_flags & CLONE_NEWNS)
2901		unshare_flags |= CLONE_FS;
2902
2903	err = check_unshare_flags(unshare_flags);
2904	if (err)
2905		goto bad_unshare_out;
2906	/*
2907	 * CLONE_NEWIPC must also detach from the undolist: after switching
2908	 * to a new ipc namespace, the semaphore arrays from the old
2909	 * namespace are unreachable.
2910	 */
2911	if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2912		do_sysvsem = 1;
2913	err = unshare_fs(unshare_flags, &new_fs);
2914	if (err)
2915		goto bad_unshare_out;
2916	err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
2917	if (err)
2918		goto bad_unshare_cleanup_fs;
2919	err = unshare_userns(unshare_flags, &new_cred);
2920	if (err)
2921		goto bad_unshare_cleanup_fd;
2922	err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2923					 new_cred, new_fs);
2924	if (err)
2925		goto bad_unshare_cleanup_cred;
2926
 
 
 
 
 
 
2927	if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2928		if (do_sysvsem) {
2929			/*
2930			 * CLONE_SYSVSEM is equivalent to sys_exit().
2931			 */
2932			exit_sem(current);
2933		}
2934		if (unshare_flags & CLONE_NEWIPC) {
2935			/* Orphan segments in old ns (see sem above). */
2936			exit_shm(current);
2937			shm_init_task(current);
2938		}
2939
2940		if (new_nsproxy)
2941			switch_task_namespaces(current, new_nsproxy);
2942
2943		task_lock(current);
2944
2945		if (new_fs) {
2946			fs = current->fs;
2947			spin_lock(&fs->lock);
2948			current->fs = new_fs;
2949			if (--fs->users)
2950				new_fs = NULL;
2951			else
2952				new_fs = fs;
2953			spin_unlock(&fs->lock);
2954		}
2955
2956		if (new_fd) {
2957			fd = current->files;
2958			current->files = new_fd;
2959			new_fd = fd;
2960		}
2961
2962		task_unlock(current);
2963
2964		if (new_cred) {
2965			/* Install the new user namespace */
2966			commit_creds(new_cred);
2967			new_cred = NULL;
2968		}
2969	}
2970
2971	perf_event_namespaces(current);
2972
2973bad_unshare_cleanup_cred:
2974	if (new_cred)
2975		put_cred(new_cred);
2976bad_unshare_cleanup_fd:
2977	if (new_fd)
2978		put_files_struct(new_fd);
2979
2980bad_unshare_cleanup_fs:
2981	if (new_fs)
2982		free_fs_struct(new_fs);
2983
2984bad_unshare_out:
2985	return err;
2986}
2987
2988SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2989{
2990	return ksys_unshare(unshare_flags);
2991}
2992
2993/*
2994 *	Helper to unshare the files of the current task.
2995 *	We don't want to expose copy_files internals to
2996 *	the exec layer of the kernel.
2997 */
2998
2999int unshare_files(struct files_struct **displaced)
3000{
3001	struct task_struct *task = current;
3002	struct files_struct *copy = NULL;
3003	int error;
3004
3005	error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, &copy);
3006	if (error || !copy) {
3007		*displaced = NULL;
3008		return error;
3009	}
3010	*displaced = task->files;
3011	task_lock(task);
3012	task->files = copy;
3013	task_unlock(task);
 
3014	return 0;
3015}
3016
3017int sysctl_max_threads(struct ctl_table *table, int write,
3018		       void *buffer, size_t *lenp, loff_t *ppos)
3019{
3020	struct ctl_table t;
3021	int ret;
3022	int threads = max_threads;
3023	int min = 1;
3024	int max = MAX_THREADS;
3025
3026	t = *table;
3027	t.data = &threads;
3028	t.extra1 = &min;
3029	t.extra2 = &max;
3030
3031	ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3032	if (ret || !write)
3033		return ret;
3034
3035	max_threads = threads;
3036
3037	return 0;
3038}