kernel/fork.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/kernel/fork.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  */
   7
   8 /*
   9  *  'fork.c' contains the help-routines for the 'fork' system call
  10  * (see also entry.S and others).
  11  * Fork is rather simple, once you get the hang of it, but the memory
  12  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  13  */
  14
  15 #include <linux/anon_inodes.h>
  16 #include <linux/slab.h>
  17 #include <linux/sched/autogroup.h>
  18 #include <linux/sched/mm.h>
  19 #include <linux/sched/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/jiffies.h>
  57 #include <linux/futex.h>
  58 #include <linux/compat.h>
  59 #include <linux/kthread.h>
  60 #include <linux/task_io_accounting_ops.h>
  61 #include <linux/rcupdate.h>
  62 #include <linux/ptrace.h>
  63 #include <linux/mount.h>
  64 #include <linux/audit.h>
  65 #include <linux/memcontrol.h>
  66 #include <linux/ftrace.h>
  67 #include <linux/proc_fs.h>
  68 #include <linux/profile.h>
  69 #include <linux/rmap.h>
  70 #include <linux/ksm.h>
  71 #include <linux/acct.h>
  72 #include <linux/userfaultfd_k.h>
  73 #include <linux/tsacct_kern.h>
  74 #include <linux/cn_proc.h>
  75 #include <linux/freezer.h>
  76 #include <linux/delayacct.h>
  77 #include <linux/taskstats_kern.h>
  78 #include <linux/random.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/sched/mm.h>
 101
 102 #include <asm/pgalloc.h>
 103 #include <linux/uaccess.h>
 104 #include <asm/mmu_context.h>
 105 #include <asm/cacheflush.h>
 106 #include <asm/tlbflush.h>
 107
 108 #include <trace/events/sched.h>
 109
 110 #define CREATE_TRACE_POINTS
 111 #include <trace/events/task.h>
 112
 113 /*
 114  * Minimum number of threads to boot the kernel
 115  */
 116 #define MIN_THREADS 20
 117
 118 /*
 119  * Maximum number of threads
 120  */
 121 #define MAX_THREADS FUTEX_TID_MASK
 122
 123 /*
 124  * Protected counters by write_lock_irq(&tasklist_lock)
 125  */
 126 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 127 int nr_threads;                 /* The idle threads do not count.. */
 128
 129 static int max_threads;         /* tunable limit on nr_threads */
 130
 131 #define NAMED_ARRAY_INDEX(x)    [x] = __stringify(x)
 132
 133 static const char * const resident_page_types[] = {
 134         NAMED_ARRAY_INDEX(MM_FILEPAGES),
 135         NAMED_ARRAY_INDEX(MM_ANONPAGES),
 136         NAMED_ARRAY_INDEX(MM_SWAPENTS),
 137         NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
 138 };
 139
 140 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 141
 142 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 143
 144 #ifdef CONFIG_PROVE_RCU
 145 int lockdep_tasklist_lock_is_held(void)
 146 {
 147         return lockdep_is_held(&tasklist_lock);
 148 }
 149 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 150 #endif /* #ifdef CONFIG_PROVE_RCU */
 151
 152 int nr_processes(void)
 153 {
 154         int cpu;
 155         int total = 0;
 156
 157         for_each_possible_cpu(cpu)
 158                 total += per_cpu(process_counts, cpu);
 159
 160         return total;
 161 }
 162
 163 void __weak arch_release_task_struct(struct task_struct *tsk)
 164 {
 165 }
 166
 167 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 168 static struct kmem_cache *task_struct_cachep;
 169
 170 static inline struct task_struct *alloc_task_struct_node(int node)
 171 {
 172         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 173 }
 174
 175 static inline void free_task_struct(struct task_struct *tsk)
 176 {
 177         kmem_cache_free(task_struct_cachep, tsk);
 178 }
 179 #endif
 180
 181 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 182
 183 /*
 184  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 185  * kmemcache based allocator.
 186  */
 187 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 188
 189 #  ifdef CONFIG_VMAP_STACK
 190 /*
 191  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 192  * flush.  Try to minimize the number of calls by caching stacks.
 193  */
 194 #define NR_CACHED_STACKS 2
 195 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 196
 197 struct vm_stack {
 198         struct rcu_head rcu;
 199         struct vm_struct *stack_vm_area;
 200 };
 201
 202 static bool try_release_thread_stack_to_cache(struct vm_struct *vm)
 203 {
 204         unsigned int i;
 205
 206         for (i = 0; i < NR_CACHED_STACKS; i++) {
 207                 if (this_cpu_cmpxchg(cached_stacks[i], NULL, vm) != NULL)
 208                         continue;
 209                 return true;
 210         }
 211         return false;
 212 }
 213
 214 static void thread_stack_free_rcu(struct rcu_head *rh)
 215 {
 216         struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
 217
 218         if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
 219                 return;
 220
 221         vfree(vm_stack);
 222 }
 223
 224 static void thread_stack_delayed_free(struct task_struct *tsk)
 225 {
 226         struct vm_stack *vm_stack = tsk->stack;
 227
 228         vm_stack->stack_vm_area = tsk->stack_vm_area;
 229         call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
 230 }
 231
 232 static int free_vm_stack_cache(unsigned int cpu)
 233 {
 234         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 235         int i;
 236
 237         for (i = 0; i < NR_CACHED_STACKS; i++) {
 238                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 239
 240                 if (!vm_stack)
 241                         continue;
 242
 243                 vfree(vm_stack->addr);
 244                 cached_vm_stacks[i] = NULL;
 245         }
 246
 247         return 0;
 248 }
 249
 250 static int memcg_charge_kernel_stack(struct vm_struct *vm)
 251 {
 252         int i;
 253         int ret;
 254
 255         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 256         BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 257
 258         for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 259                 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0);
 260                 if (ret)
 261                         goto err;
 262         }
 263         return 0;
 264 err:
 265         /*
 266          * If memcg_kmem_charge_page() fails, page's memory cgroup pointer is
 267          * NULL, and memcg_kmem_uncharge_page() in free_thread_stack() will
 268          * ignore this page.
 269          */
 270         for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 271                 memcg_kmem_uncharge_page(vm->pages[i], 0);
 272         return ret;
 273 }
 274
 275 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 276 {
 277         struct vm_struct *vm;
 278         void *stack;
 279         int i;
 280
 281         for (i = 0; i < NR_CACHED_STACKS; i++) {
 282                 struct vm_struct *s;
 283
 284                 s = this_cpu_xchg(cached_stacks[i], NULL);
 285
 286                 if (!s)
 287                         continue;
 288
 289                 /* Reset stack metadata. */
 290                 kasan_unpoison_range(s->addr, THREAD_SIZE);
 291
 292                 stack = kasan_reset_tag(s->addr);
 293
 294                 /* Clear stale pointers from reused stack. */
 295                 memset(stack, 0, THREAD_SIZE);
 296
 297                 if (memcg_charge_kernel_stack(s)) {
 298                         vfree(s->addr);
 299                         return -ENOMEM;
 300                 }
 301
 302                 tsk->stack_vm_area = s;
 303                 tsk->stack = stack;
 304                 return 0;
 305         }
 306
 307         /*
 308          * Allocated stacks are cached and later reused by new threads,
 309          * so memcg accounting is performed manually on assigning/releasing
 310          * stacks to tasks. Drop __GFP_ACCOUNT.
 311          */
 312         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 313                                      VMALLOC_START, VMALLOC_END,
 314                                      THREADINFO_GFP & ~__GFP_ACCOUNT,
 315                                      PAGE_KERNEL,
 316                                      0, node, __builtin_return_address(0));
 317         if (!stack)
 318                 return -ENOMEM;
 319
 320         vm = find_vm_area(stack);
 321         if (memcg_charge_kernel_stack(vm)) {
 322                 vfree(stack);
 323                 return -ENOMEM;
 324         }
 325         /*
 326          * We can't call find_vm_area() in interrupt context, and
 327          * free_thread_stack() can be called in interrupt context,
 328          * so cache the vm_struct.
 329          */
 330         tsk->stack_vm_area = vm;
 331         stack = kasan_reset_tag(stack);
 332         tsk->stack = stack;
 333         return 0;
 334 }
 335
 336 static void free_thread_stack(struct task_struct *tsk)
 337 {
 338         if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
 339                 thread_stack_delayed_free(tsk);
 340
 341         tsk->stack = NULL;
 342         tsk->stack_vm_area = NULL;
 343 }
 344
 345 #  else /* !CONFIG_VMAP_STACK */
 346
 347 static void thread_stack_free_rcu(struct rcu_head *rh)
 348 {
 349         __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
 350 }
 351
 352 static void thread_stack_delayed_free(struct task_struct *tsk)
 353 {
 354         struct rcu_head *rh = tsk->stack;
 355
 356         call_rcu(rh, thread_stack_free_rcu);
 357 }
 358
 359 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 360 {
 361         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 362                                              THREAD_SIZE_ORDER);
 363
 364         if (likely(page)) {
 365                 tsk->stack = kasan_reset_tag(page_address(page));
 366                 return 0;
 367         }
 368         return -ENOMEM;
 369 }
 370
 371 static void free_thread_stack(struct task_struct *tsk)
 372 {
 373         thread_stack_delayed_free(tsk);
 374         tsk->stack = NULL;
 375 }
 376
 377 #  endif /* CONFIG_VMAP_STACK */
 378 # else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
 379
 380 static struct kmem_cache *thread_stack_cache;
 381
 382 static void thread_stack_free_rcu(struct rcu_head *rh)
 383 {
 384         kmem_cache_free(thread_stack_cache, rh);
 385 }
 386
 387 static void thread_stack_delayed_free(struct task_struct *tsk)
 388 {
 389         struct rcu_head *rh = tsk->stack;
 390
 391         call_rcu(rh, thread_stack_free_rcu);
 392 }
 393
 394 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 395 {
 396         unsigned long *stack;
 397         stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 398         stack = kasan_reset_tag(stack);
 399         tsk->stack = stack;
 400         return stack ? 0 : -ENOMEM;
 401 }
 402
 403 static void free_thread_stack(struct task_struct *tsk)
 404 {
 405         thread_stack_delayed_free(tsk);
 406         tsk->stack = NULL;
 407 }
 408
 409 void thread_stack_cache_init(void)
 410 {
 411         thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 412                                         THREAD_SIZE, THREAD_SIZE, 0, 0,
 413                                         THREAD_SIZE, NULL);
 414         BUG_ON(thread_stack_cache == NULL);
 415 }
 416
 417 # endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
 418 #else /* CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
 419
 420 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 421 {
 422         unsigned long *stack;
 423
 424         stack = arch_alloc_thread_stack_node(tsk, node);
 425         tsk->stack = stack;
 426         return stack ? 0 : -ENOMEM;
 427 }
 428
 429 static void free_thread_stack(struct task_struct *tsk)
 430 {
 431         arch_free_thread_stack(tsk);
 432         tsk->stack = NULL;
 433 }
 434
 435 #endif /* !CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
 436
 437 /* SLAB cache for signal_struct structures (tsk->signal) */
 438 static struct kmem_cache *signal_cachep;
 439
 440 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 441 struct kmem_cache *sighand_cachep;
 442
 443 /* SLAB cache for files_struct structures (tsk->files) */
 444 struct kmem_cache *files_cachep;
 445
 446 /* SLAB cache for fs_struct structures (tsk->fs) */
 447 struct kmem_cache *fs_cachep;
 448
 449 /* SLAB cache for vm_area_struct structures */
 450 static struct kmem_cache *vm_area_cachep;
 451
 452 /* SLAB cache for mm_struct structures (tsk->mm) */
 453 static struct kmem_cache *mm_cachep;
 454
 455 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 456 {
 457         struct vm_area_struct *vma;
 458
 459         vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 460         if (vma)
 461                 vma_init(vma, mm);
 462         return vma;
 463 }
 464
 465 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 466 {
 467         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 468
 469         if (new) {
 470                 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
 471                 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
 472                 /*
 473                  * orig->shared.rb may be modified concurrently, but the clone
 474                  * will be reinitialized.
 475                  */
 476                 *new = data_race(*orig);
 477                 INIT_LIST_HEAD(&new->anon_vma_chain);
 478                 dup_anon_vma_name(orig, new);
 479         }
 480         return new;
 481 }
 482
 483 void vm_area_free(struct vm_area_struct *vma)
 484 {
 485         free_anon_vma_name(vma);
 486         kmem_cache_free(vm_area_cachep, vma);
 487 }
 488
 489 static void account_kernel_stack(struct task_struct *tsk, int account)
 490 {
 491         if (IS_ENABLED(CONFIG_VMAP_STACK)) {
 492                 struct vm_struct *vm = task_stack_vm_area(tsk);
 493                 int i;
 494
 495                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 496                         mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
 497                                               account * (PAGE_SIZE / 1024));
 498         } else {
 499                 void *stack = task_stack_page(tsk);
 500
 501                 /* All stack pages are in the same node. */
 502                 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
 503                                       account * (THREAD_SIZE / 1024));
 504         }
 505 }
 506
 507 void exit_task_stack_account(struct task_struct *tsk)
 508 {
 509         account_kernel_stack(tsk, -1);
 510
 511         if (IS_ENABLED(CONFIG_VMAP_STACK)) {
 512                 struct vm_struct *vm;
 513                 int i;
 514
 515                 vm = task_stack_vm_area(tsk);
 516                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 517                         memcg_kmem_uncharge_page(vm->pages[i], 0);
 518         }
 519 }
 520
 521 static void release_task_stack(struct task_struct *tsk)
 522 {
 523         if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
 524                 return;  /* Better to leak the stack than to free prematurely */
 525
 526         free_thread_stack(tsk);
 527 }
 528
 529 #ifdef CONFIG_THREAD_INFO_IN_TASK
 530 void put_task_stack(struct task_struct *tsk)
 531 {
 532         if (refcount_dec_and_test(&tsk->stack_refcount))
 533                 release_task_stack(tsk);
 534 }
 535 #endif
 536
 537 void free_task(struct task_struct *tsk)
 538 {
 539         release_user_cpus_ptr(tsk);
 540         scs_release(tsk);
 541
 542 #ifndef CONFIG_THREAD_INFO_IN_TASK
 543         /*
 544          * The task is finally done with both the stack and thread_info,
 545          * so free both.
 546          */
 547         release_task_stack(tsk);
 548 #else
 549         /*
 550          * If the task had a separate stack allocation, it should be gone
 551          * by now.
 552          */
 553         WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
 554 #endif
 555         rt_mutex_debug_task_free(tsk);
 556         ftrace_graph_exit_task(tsk);
 557         arch_release_task_struct(tsk);
 558         if (tsk->flags & PF_KTHREAD)
 559                 free_kthread_struct(tsk);
 560         free_task_struct(tsk);
 561 }
 562 EXPORT_SYMBOL(free_task);
 563
 564 static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
 565 {
 566         struct file *exe_file;
 567
 568         exe_file = get_mm_exe_file(oldmm);
 569         RCU_INIT_POINTER(mm->exe_file, exe_file);
 570         /*
 571          * We depend on the oldmm having properly denied write access to the
 572          * exe_file already.
 573          */
 574         if (exe_file && deny_write_access(exe_file))
 575                 pr_warn_once("deny_write_access() failed in %s\n", __func__);
 576 }
 577
 578 #ifdef CONFIG_MMU
 579 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 580                                         struct mm_struct *oldmm)
 581 {
 582         struct vm_area_struct *mpnt, *tmp;
 583         int retval;
 584         unsigned long charge = 0;
 585         LIST_HEAD(uf);
 586         MA_STATE(old_mas, &oldmm->mm_mt, 0, 0);
 587         MA_STATE(mas, &mm->mm_mt, 0, 0);
 588
 589         uprobe_start_dup_mmap();
 590         if (mmap_write_lock_killable(oldmm)) {
 591                 retval = -EINTR;
 592                 goto fail_uprobe_end;
 593         }
 594         flush_cache_dup_mm(oldmm);
 595         uprobe_dup_mmap(oldmm, mm);
 596         /*
 597          * Not linked in yet - no deadlock potential:
 598          */
 599         mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
 600
 601         /* No ordering required: file already has been exposed. */
 602         dup_mm_exe_file(mm, oldmm);
 603
 604         mm->total_vm = oldmm->total_vm;
 605         mm->data_vm = oldmm->data_vm;
 606         mm->exec_vm = oldmm->exec_vm;
 607         mm->stack_vm = oldmm->stack_vm;
 608
 609         retval = ksm_fork(mm, oldmm);
 610         if (retval)
 611                 goto out;
 612         khugepaged_fork(mm, oldmm);
 613
 614         retval = mas_expected_entries(&mas, oldmm->map_count);
 615         if (retval)
 616                 goto out;
 617
 618         mas_for_each(&old_mas, mpnt, ULONG_MAX) {
 619                 struct file *file;
 620
 621                 if (mpnt->vm_flags & VM_DONTCOPY) {
 622                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 623                         continue;
 624                 }
 625                 charge = 0;
 626                 /*
 627                  * Don't duplicate many vmas if we've been oom-killed (for
 628                  * example)
 629                  */
 630                 if (fatal_signal_pending(current)) {
 631                         retval = -EINTR;
 632                         goto loop_out;
 633                 }
 634                 if (mpnt->vm_flags & VM_ACCOUNT) {
 635                         unsigned long len = vma_pages(mpnt);
 636
 637                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 638                                 goto fail_nomem;
 639                         charge = len;
 640                 }
 641                 tmp = vm_area_dup(mpnt);
 642                 if (!tmp)
 643                         goto fail_nomem;
 644                 retval = vma_dup_policy(mpnt, tmp);
 645                 if (retval)
 646                         goto fail_nomem_policy;
 647                 tmp->vm_mm = mm;
 648                 retval = dup_userfaultfd(tmp, &uf);
 649                 if (retval)
 650                         goto fail_nomem_anon_vma_fork;
 651                 if (tmp->vm_flags & VM_WIPEONFORK) {
 652                         /*
 653                          * VM_WIPEONFORK gets a clean slate in the child.
 654                          * Don't prepare anon_vma until fault since we don't
 655                          * copy page for current vma.
 656                          */
 657                         tmp->anon_vma = NULL;
 658                 } else if (anon_vma_fork(tmp, mpnt))
 659                         goto fail_nomem_anon_vma_fork;
 660                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 661                 file = tmp->vm_file;
 662                 if (file) {
 663                         struct address_space *mapping = file->f_mapping;
 664
 665                         get_file(file);
 666                         i_mmap_lock_write(mapping);
 667                         if (tmp->vm_flags & VM_SHARED)
 668                                 mapping_allow_writable(mapping);
 669                         flush_dcache_mmap_lock(mapping);
 670                         /* insert tmp into the share list, just after mpnt */
 671                         vma_interval_tree_insert_after(tmp, mpnt,
 672                                         &mapping->i_mmap);
 673                         flush_dcache_mmap_unlock(mapping);
 674                         i_mmap_unlock_write(mapping);
 675                 }
 676
 677                 /*
 678                  * Copy/update hugetlb private vma information.
 679                  */
 680                 if (is_vm_hugetlb_page(tmp))
 681                         hugetlb_dup_vma_private(tmp);
 682
 683                 /* Link the vma into the MT */
 684                 mas.index = tmp->vm_start;
 685                 mas.last = tmp->vm_end - 1;
 686                 mas_store(&mas, tmp);
 687                 if (mas_is_err(&mas))
 688                         goto fail_nomem_mas_store;
 689
 690                 mm->map_count++;
 691                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 692                         retval = copy_page_range(tmp, mpnt);
 693
 694                 if (tmp->vm_ops && tmp->vm_ops->open)
 695                         tmp->vm_ops->open(tmp);
 696
 697                 if (retval)
 698                         goto loop_out;
 699         }
 700         /* a new mm has just been created */
 701         retval = arch_dup_mmap(oldmm, mm);
 702 loop_out:
 703         mas_destroy(&mas);
 704 out:
 705         mmap_write_unlock(mm);
 706         flush_tlb_mm(oldmm);
 707         mmap_write_unlock(oldmm);
 708         dup_userfaultfd_complete(&uf);
 709 fail_uprobe_end:
 710         uprobe_end_dup_mmap();
 711         return retval;
 712
 713 fail_nomem_mas_store:
 714         unlink_anon_vmas(tmp);
 715 fail_nomem_anon_vma_fork:
 716         mpol_put(vma_policy(tmp));
 717 fail_nomem_policy:
 718         vm_area_free(tmp);
 719 fail_nomem:
 720         retval = -ENOMEM;
 721         vm_unacct_memory(charge);
 722         goto loop_out;
 723 }
 724
 725 static inline int mm_alloc_pgd(struct mm_struct *mm)
 726 {
 727         mm->pgd = pgd_alloc(mm);
 728         if (unlikely(!mm->pgd))
 729                 return -ENOMEM;
 730         return 0;
 731 }
 732
 733 static inline void mm_free_pgd(struct mm_struct *mm)
 734 {
 735         pgd_free(mm, mm->pgd);
 736 }
 737 #else
 738 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 739 {
 740         mmap_write_lock(oldmm);
 741         dup_mm_exe_file(mm, oldmm);
 742         mmap_write_unlock(oldmm);
 743         return 0;
 744 }
 745 #define mm_alloc_pgd(mm)        (0)
 746 #define mm_free_pgd(mm)
 747 #endif /* CONFIG_MMU */
 748
 749 static void check_mm(struct mm_struct *mm)
 750 {
 751         int i;
 752
 753         BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
 754                          "Please make sure 'struct resident_page_types[]' is updated as well");
 755
 756         for (i = 0; i < NR_MM_COUNTERS; i++) {
 757                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 758
 759                 if (unlikely(x))
 760                         pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
 761                                  mm, resident_page_types[i], x);
 762         }
 763
 764         if (mm_pgtables_bytes(mm))
 765                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 766                                 mm_pgtables_bytes(mm));
 767
 768 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 769         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 770 #endif
 771 }
 772
 773 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 774 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 775
 776 /*
 777  * Called when the last reference to the mm
 778  * is dropped: either by a lazy thread or by
 779  * mmput. Free the page directory and the mm.
 780  */
 781 void __mmdrop(struct mm_struct *mm)
 782 {
 783         BUG_ON(mm == &init_mm);
 784         WARN_ON_ONCE(mm == current->mm);
 785         WARN_ON_ONCE(mm == current->active_mm);
 786         mm_free_pgd(mm);
 787         destroy_context(mm);
 788         mmu_notifier_subscriptions_destroy(mm);
 789         check_mm(mm);
 790         put_user_ns(mm->user_ns);
 791         mm_pasid_drop(mm);
 792         free_mm(mm);
 793 }
 794 EXPORT_SYMBOL_GPL(__mmdrop);
 795
 796 static void mmdrop_async_fn(struct work_struct *work)
 797 {
 798         struct mm_struct *mm;
 799
 800         mm = container_of(work, struct mm_struct, async_put_work);
 801         __mmdrop(mm);
 802 }
 803
 804 static void mmdrop_async(struct mm_struct *mm)
 805 {
 806         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 807                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 808                 schedule_work(&mm->async_put_work);
 809         }
 810 }
 811
 812 static inline void free_signal_struct(struct signal_struct *sig)
 813 {
 814         taskstats_tgid_free(sig);
 815         sched_autogroup_exit(sig);
 816         /*
 817          * __mmdrop is not safe to call from softirq context on x86 due to
 818          * pgd_dtor so postpone it to the async context
 819          */
 820         if (sig->oom_mm)
 821                 mmdrop_async(sig->oom_mm);
 822         kmem_cache_free(signal_cachep, sig);
 823 }
 824
 825 static inline void put_signal_struct(struct signal_struct *sig)
 826 {
 827         if (refcount_dec_and_test(&sig->sigcnt))
 828                 free_signal_struct(sig);
 829 }
 830
 831 void __put_task_struct(struct task_struct *tsk)
 832 {
 833         WARN_ON(!tsk->exit_state);
 834         WARN_ON(refcount_read(&tsk->usage));
 835         WARN_ON(tsk == current);
 836
 837         io_uring_free(tsk);
 838         cgroup_free(tsk);
 839         task_numa_free(tsk, true);
 840         security_task_free(tsk);
 841         bpf_task_storage_free(tsk);
 842         exit_creds(tsk);
 843         delayacct_tsk_free(tsk);
 844         put_signal_struct(tsk->signal);
 845         sched_core_free(tsk);
 846         free_task(tsk);
 847 }
 848 EXPORT_SYMBOL_GPL(__put_task_struct);
 849
 850 void __init __weak arch_task_cache_init(void) { }
 851
 852 /*
 853  * set_max_threads
 854  */
 855 static void set_max_threads(unsigned int max_threads_suggested)
 856 {
 857         u64 threads;
 858         unsigned long nr_pages = totalram_pages();
 859
 860         /*
 861          * The number of threads shall be limited such that the thread
 862          * structures may only consume a small part of the available memory.
 863          */
 864         if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
 865                 threads = MAX_THREADS;
 866         else
 867                 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
 868                                     (u64) THREAD_SIZE * 8UL);
 869
 870         if (threads > max_threads_suggested)
 871                 threads = max_threads_suggested;
 872
 873         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 874 }
 875
 876 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 877 /* Initialized by the architecture: */
 878 int arch_task_struct_size __read_mostly;
 879 #endif
 880
 881 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 882 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 883 {
 884         /* Fetch thread_struct whitelist for the architecture. */
 885         arch_thread_struct_whitelist(offset, size);
 886
 887         /*
 888          * Handle zero-sized whitelist or empty thread_struct, otherwise
 889          * adjust offset to position of thread_struct in task_struct.
 890          */
 891         if (unlikely(*size == 0))
 892                 *offset = 0;
 893         else
 894                 *offset += offsetof(struct task_struct, thread);
 895 }
 896 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
 897
 898 void __init fork_init(void)
 899 {
 900         int i;
 901 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 902 #ifndef ARCH_MIN_TASKALIGN
 903 #define ARCH_MIN_TASKALIGN      0
 904 #endif
 905         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 906         unsigned long useroffset, usersize;
 907
 908         /* create a slab on which task_structs can be allocated */
 909         task_struct_whitelist(&useroffset, &usersize);
 910         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 911                         arch_task_struct_size, align,
 912                         SLAB_PANIC|SLAB_ACCOUNT,
 913                         useroffset, usersize, NULL);
 914 #endif
 915
 916         /* do the arch specific task caches init */
 917         arch_task_cache_init();
 918
 919         set_max_threads(MAX_THREADS);
 920
 921         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 922         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 923         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 924                 init_task.signal->rlim[RLIMIT_NPROC];
 925
 926         for (i = 0; i < MAX_PER_NAMESPACE_UCOUNTS; i++)
 927                 init_user_ns.ucount_max[i] = max_threads/2;
 928
 929         set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_NPROC,      RLIM_INFINITY);
 930         set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE,   RLIM_INFINITY);
 931         set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
 932         set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK,    RLIM_INFINITY);
 933
 934 #ifdef CONFIG_VMAP_STACK
 935         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 936                           NULL, free_vm_stack_cache);
 937 #endif
 938
 939         scs_init();
 940
 941         lockdep_init_task(&init_task);
 942         uprobes_init();
 943 }
 944
 945 int __weak arch_dup_task_struct(struct task_struct *dst,
 946                                                struct task_struct *src)
 947 {
 948         *dst = *src;
 949         return 0;
 950 }
 951
 952 void set_task_stack_end_magic(struct task_struct *tsk)
 953 {
 954         unsigned long *stackend;
 955
 956         stackend = end_of_stack(tsk);
 957         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 958 }
 959
 960 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 961 {
 962         struct task_struct *tsk;
 963         int err;
 964
 965         if (node == NUMA_NO_NODE)
 966                 node = tsk_fork_get_node(orig);
 967         tsk = alloc_task_struct_node(node);
 968         if (!tsk)
 969                 return NULL;
 970
 971         err = arch_dup_task_struct(tsk, orig);
 972         if (err)
 973                 goto free_tsk;
 974
 975         err = alloc_thread_stack_node(tsk, node);
 976         if (err)
 977                 goto free_tsk;
 978
 979 #ifdef CONFIG_THREAD_INFO_IN_TASK
 980         refcount_set(&tsk->stack_refcount, 1);
 981 #endif
 982         account_kernel_stack(tsk, 1);
 983
 984         err = scs_prepare(tsk, node);
 985         if (err)
 986                 goto free_stack;
 987
 988 #ifdef CONFIG_SECCOMP
 989         /*
 990          * We must handle setting up seccomp filters once we're under
 991          * the sighand lock in case orig has changed between now and
 992          * then. Until then, filter must be NULL to avoid messing up
 993          * the usage counts on the error path calling free_task.
 994          */
 995         tsk->seccomp.filter = NULL;
 996 #endif
 997
 998         setup_thread_stack(tsk, orig);
 999         clear_user_return_notifier(tsk);
1000         clear_tsk_need_resched(tsk);
1001         set_task_stack_end_magic(tsk);
1002         clear_syscall_work_syscall_user_dispatch(tsk);
1003
1004 #ifdef CONFIG_STACKPROTECTOR
1005         tsk->stack_canary = get_random_canary();
1006 #endif
1007         if (orig->cpus_ptr == &orig->cpus_mask)
1008                 tsk->cpus_ptr = &tsk->cpus_mask;
1009         dup_user_cpus_ptr(tsk, orig, node);
1010
1011         /*
1012          * One for the user space visible state that goes away when reaped.
1013          * One for the scheduler.
1014          */
1015         refcount_set(&tsk->rcu_users, 2);
1016         /* One for the rcu users */
1017         refcount_set(&tsk->usage, 1);
1018 #ifdef CONFIG_BLK_DEV_IO_TRACE
1019         tsk->btrace_seq = 0;
1020 #endif
1021         tsk->splice_pipe = NULL;
1022         tsk->task_frag.page = NULL;
1023         tsk->wake_q.next = NULL;
1024         tsk->worker_private = NULL;
1025
1026         kcov_task_init(tsk);
1027         kmsan_task_create(tsk);
1028         kmap_local_fork(tsk);
1029
1030 #ifdef CONFIG_FAULT_INJECTION
1031         tsk->fail_nth = 0;
1032 #endif
1033
1034 #ifdef CONFIG_BLK_CGROUP
1035         tsk->throttle_queue = NULL;
1036         tsk->use_memdelay = 0;
1037 #endif
1038
1039 #ifdef CONFIG_IOMMU_SVA
1040         tsk->pasid_activated = 0;
1041 #endif
1042
1043 #ifdef CONFIG_MEMCG
1044         tsk->active_memcg = NULL;
1045 #endif
1046
1047 #ifdef CONFIG_CPU_SUP_INTEL
1048         tsk->reported_split_lock = 0;
1049 #endif
1050
1051         return tsk;
1052
1053 free_stack:
1054         exit_task_stack_account(tsk);
1055         free_thread_stack(tsk);
1056 free_tsk:
1057         free_task_struct(tsk);
1058         return NULL;
1059 }
1060
1061 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1062
1063 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1064
1065 static int __init coredump_filter_setup(char *s)
1066 {
1067         default_dump_filter =
1068                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1069                 MMF_DUMP_FILTER_MASK;
1070         return 1;
1071 }
1072
1073 __setup("coredump_filter=", coredump_filter_setup);
1074
1075 #include <linux/init_task.h>
1076
1077 static void mm_init_aio(struct mm_struct *mm)
1078 {
1079 #ifdef CONFIG_AIO
1080         spin_lock_init(&mm->ioctx_lock);
1081         mm->ioctx_table = NULL;
1082 #endif
1083 }
1084
1085 static __always_inline void mm_clear_owner(struct mm_struct *mm,
1086                                            struct task_struct *p)
1087 {
1088 #ifdef CONFIG_MEMCG
1089         if (mm->owner == p)
1090                 WRITE_ONCE(mm->owner, NULL);
1091 #endif
1092 }
1093
1094 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1095 {
1096 #ifdef CONFIG_MEMCG
1097         mm->owner = p;
1098 #endif
1099 }
1100
1101 static void mm_init_uprobes_state(struct mm_struct *mm)
1102 {
1103 #ifdef CONFIG_UPROBES
1104         mm->uprobes_state.xol_area = NULL;
1105 #endif
1106 }
1107
1108 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1109         struct user_namespace *user_ns)
1110 {
1111         mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1112         mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1113         atomic_set(&mm->mm_users, 1);
1114         atomic_set(&mm->mm_count, 1);
1115         seqcount_init(&mm->write_protect_seq);
1116         mmap_init_lock(mm);
1117         INIT_LIST_HEAD(&mm->mmlist);
1118         mm_pgtables_bytes_init(mm);
1119         mm->map_count = 0;
1120         mm->locked_vm = 0;
1121         atomic64_set(&mm->pinned_vm, 0);
1122         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1123         spin_lock_init(&mm->page_table_lock);
1124         spin_lock_init(&mm->arg_lock);
1125         mm_init_cpumask(mm);
1126         mm_init_aio(mm);
1127         mm_init_owner(mm, p);
1128         mm_pasid_init(mm);
1129         RCU_INIT_POINTER(mm->exe_file, NULL);
1130         mmu_notifier_subscriptions_init(mm);
1131         init_tlb_flush_pending(mm);
1132 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1133         mm->pmd_huge_pte = NULL;
1134 #endif
1135         mm_init_uprobes_state(mm);
1136         hugetlb_count_init(mm);
1137
1138         if (current->mm) {
1139                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1140                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1141         } else {
1142                 mm->flags = default_dump_filter;
1143                 mm->def_flags = 0;
1144         }
1145
1146         if (mm_alloc_pgd(mm))
1147                 goto fail_nopgd;
1148
1149         if (init_new_context(p, mm))
1150                 goto fail_nocontext;
1151
1152         mm->user_ns = get_user_ns(user_ns);
1153         lru_gen_init_mm(mm);
1154         return mm;
1155
1156 fail_nocontext:
1157         mm_free_pgd(mm);
1158 fail_nopgd:
1159         free_mm(mm);
1160         return NULL;
1161 }
1162
1163 /*
1164  * Allocate and initialize an mm_struct.
1165  */
1166 struct mm_struct *mm_alloc(void)
1167 {
1168         struct mm_struct *mm;
1169
1170         mm = allocate_mm();
1171         if (!mm)
1172                 return NULL;
1173
1174         memset(mm, 0, sizeof(*mm));
1175         return mm_init(mm, current, current_user_ns());
1176 }
1177
1178 static inline void __mmput(struct mm_struct *mm)
1179 {
1180         VM_BUG_ON(atomic_read(&mm->mm_users));
1181
1182         uprobe_clear_state(mm);
1183         exit_aio(mm);
1184         ksm_exit(mm);
1185         khugepaged_exit(mm); /* must run before exit_mmap */
1186         exit_mmap(mm);
1187         mm_put_huge_zero_page(mm);
1188         set_mm_exe_file(mm, NULL);
1189         if (!list_empty(&mm->mmlist)) {
1190                 spin_lock(&mmlist_lock);
1191                 list_del(&mm->mmlist);
1192                 spin_unlock(&mmlist_lock);
1193         }
1194         if (mm->binfmt)
1195                 module_put(mm->binfmt->module);
1196         lru_gen_del_mm(mm);
1197         mmdrop(mm);
1198 }
1199
1200 /*
1201  * Decrement the use count and release all resources for an mm.
1202  */
1203 void mmput(struct mm_struct *mm)
1204 {
1205         might_sleep();
1206
1207         if (atomic_dec_and_test(&mm->mm_users))
1208                 __mmput(mm);
1209 }
1210 EXPORT_SYMBOL_GPL(mmput);
1211
1212 #ifdef CONFIG_MMU
1213 static void mmput_async_fn(struct work_struct *work)
1214 {
1215         struct mm_struct *mm = container_of(work, struct mm_struct,
1216                                             async_put_work);
1217
1218         __mmput(mm);
1219 }
1220
1221 void mmput_async(struct mm_struct *mm)
1222 {
1223         if (atomic_dec_and_test(&mm->mm_users)) {
1224                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1225                 schedule_work(&mm->async_put_work);
1226         }
1227 }
1228 #endif
1229
1230 /**
1231  * set_mm_exe_file - change a reference to the mm's executable file
1232  *
1233  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1234  *
1235  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1236  * invocations: in mmput() nobody alive left, in execve task is single
1237  * threaded.
1238  *
1239  * Can only fail if new_exe_file != NULL.
1240  */
1241 int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1242 {
1243         struct file *old_exe_file;
1244
1245         /*
1246          * It is safe to dereference the exe_file without RCU as
1247          * this function is only called if nobody else can access
1248          * this mm -- see comment above for justification.
1249          */
1250         old_exe_file = rcu_dereference_raw(mm->exe_file);
1251
1252         if (new_exe_file) {
1253                 /*
1254                  * We expect the caller (i.e., sys_execve) to already denied
1255                  * write access, so this is unlikely to fail.
1256                  */
1257                 if (unlikely(deny_write_access(new_exe_file)))
1258                         return -EACCES;
1259                 get_file(new_exe_file);
1260         }
1261         rcu_assign_pointer(mm->exe_file, new_exe_file);
1262         if (old_exe_file) {
1263                 allow_write_access(old_exe_file);
1264                 fput(old_exe_file);
1265         }
1266         return 0;
1267 }
1268
1269 /**
1270  * replace_mm_exe_file - replace a reference to the mm's executable file
1271  *
1272  * This changes mm's executable file (shown as symlink /proc/[pid]/exe),
1273  * dealing with concurrent invocation and without grabbing the mmap lock in
1274  * write mode.
1275  *
1276  * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1277  */
1278 int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1279 {
1280         struct vm_area_struct *vma;
1281         struct file *old_exe_file;
1282         int ret = 0;
1283
1284         /* Forbid mm->exe_file change if old file still mapped. */
1285         old_exe_file = get_mm_exe_file(mm);
1286         if (old_exe_file) {
1287                 VMA_ITERATOR(vmi, mm, 0);
1288                 mmap_read_lock(mm);
1289                 for_each_vma(vmi, vma) {
1290                         if (!vma->vm_file)
1291                                 continue;
1292                         if (path_equal(&vma->vm_file->f_path,
1293                                        &old_exe_file->f_path)) {
1294                                 ret = -EBUSY;
1295                                 break;
1296                         }
1297                 }
1298                 mmap_read_unlock(mm);
1299                 fput(old_exe_file);
1300                 if (ret)
1301                         return ret;
1302         }
1303
1304         /* set the new file, lockless */
1305         ret = deny_write_access(new_exe_file);
1306         if (ret)
1307                 return -EACCES;
1308         get_file(new_exe_file);
1309
1310         old_exe_file = xchg(&mm->exe_file, new_exe_file);
1311         if (old_exe_file) {
1312                 /*
1313                  * Don't race with dup_mmap() getting the file and disallowing
1314                  * write access while someone might open the file writable.
1315                  */
1316                 mmap_read_lock(mm);
1317                 allow_write_access(old_exe_file);
1318                 fput(old_exe_file);
1319                 mmap_read_unlock(mm);
1320         }
1321         return 0;
1322 }
1323
1324 /**
1325  * get_mm_exe_file - acquire a reference to the mm's executable file
1326  *
1327  * Returns %NULL if mm has no associated executable file.
1328  * User must release file via fput().
1329  */
1330 struct file *get_mm_exe_file(struct mm_struct *mm)
1331 {
1332         struct file *exe_file;
1333
1334         rcu_read_lock();
1335         exe_file = rcu_dereference(mm->exe_file);
1336         if (exe_file && !get_file_rcu(exe_file))
1337                 exe_file = NULL;
1338         rcu_read_unlock();
1339         return exe_file;
1340 }
1341
1342 /**
1343  * get_task_exe_file - acquire a reference to the task's executable file
1344  *
1345  * Returns %NULL if task's mm (if any) has no associated executable file or
1346  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1347  * User must release file via fput().
1348  */
1349 struct file *get_task_exe_file(struct task_struct *task)
1350 {
1351         struct file *exe_file = NULL;
1352         struct mm_struct *mm;
1353
1354         task_lock(task);
1355         mm = task->mm;
1356         if (mm) {
1357                 if (!(task->flags & PF_KTHREAD))
1358                         exe_file = get_mm_exe_file(mm);
1359         }
1360         task_unlock(task);
1361         return exe_file;
1362 }
1363
1364 /**
1365  * get_task_mm - acquire a reference to the task's mm
1366  *
1367  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1368  * this kernel workthread has transiently adopted a user mm with use_mm,
1369  * to do its AIO) is not set and if so returns a reference to it, after
1370  * bumping up the use count.  User must release the mm via mmput()
1371  * after use.  Typically used by /proc and ptrace.
1372  */
1373 struct mm_struct *get_task_mm(struct task_struct *task)
1374 {
1375         struct mm_struct *mm;
1376
1377         task_lock(task);
1378         mm = task->mm;
1379         if (mm) {
1380                 if (task->flags & PF_KTHREAD)
1381                         mm = NULL;
1382                 else
1383                         mmget(mm);
1384         }
1385         task_unlock(task);
1386         return mm;
1387 }
1388 EXPORT_SYMBOL_GPL(get_task_mm);
1389
1390 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1391 {
1392         struct mm_struct *mm;
1393         int err;
1394
1395         err =  down_read_killable(&task->signal->exec_update_lock);
1396         if (err)
1397                 return ERR_PTR(err);
1398
1399         mm = get_task_mm(task);
1400         if (mm && mm != current->mm &&
1401                         !ptrace_may_access(task, mode)) {
1402                 mmput(mm);
1403                 mm = ERR_PTR(-EACCES);
1404         }
1405         up_read(&task->signal->exec_update_lock);
1406
1407         return mm;
1408 }
1409
1410 static void complete_vfork_done(struct task_struct *tsk)
1411 {
1412         struct completion *vfork;
1413
1414         task_lock(tsk);
1415         vfork = tsk->vfork_done;
1416         if (likely(vfork)) {
1417                 tsk->vfork_done = NULL;
1418                 complete(vfork);
1419         }
1420         task_unlock(tsk);
1421 }
1422
1423 static int wait_for_vfork_done(struct task_struct *child,
1424                                 struct completion *vfork)
1425 {
1426         int killed;
1427
1428         freezer_do_not_count();
1429         cgroup_enter_frozen();
1430         killed = wait_for_completion_killable(vfork);
1431         cgroup_leave_frozen(false);
1432         freezer_count();
1433
1434         if (killed) {
1435                 task_lock(child);
1436                 child->vfork_done = NULL;
1437                 task_unlock(child);
1438         }
1439
1440         put_task_struct(child);
1441         return killed;
1442 }
1443
1444 /* Please note the differences between mmput and mm_release.
1445  * mmput is called whenever we stop holding onto a mm_struct,
1446  * error success whatever.
1447  *
1448  * mm_release is called after a mm_struct has been removed
1449  * from the current process.
1450  *
1451  * This difference is important for error handling, when we
1452  * only half set up a mm_struct for a new process and need to restore
1453  * the old one.  Because we mmput the new mm_struct before
1454  * restoring the old one. . .
1455  * Eric Biederman 10 January 1998
1456  */
1457 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1458 {
1459         uprobe_free_utask(tsk);
1460
1461         /* Get rid of any cached register state */
1462         deactivate_mm(tsk, mm);
1463
1464         /*
1465          * Signal userspace if we're not exiting with a core dump
1466          * because we want to leave the value intact for debugging
1467          * purposes.
1468          */
1469         if (tsk->clear_child_tid) {
1470                 if (atomic_read(&mm->mm_users) > 1) {
1471                         /*
1472                          * We don't check the error code - if userspace has
1473                          * not set up a proper pointer then tough luck.
1474                          */
1475                         put_user(0, tsk->clear_child_tid);
1476                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1477                                         1, NULL, NULL, 0, 0);
1478                 }
1479                 tsk->clear_child_tid = NULL;
1480         }
1481
1482         /*
1483          * All done, finally we can wake up parent and return this mm to him.
1484          * Also kthread_stop() uses this completion for synchronization.
1485          */
1486         if (tsk->vfork_done)
1487                 complete_vfork_done(tsk);
1488 }
1489
1490 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1491 {
1492         futex_exit_release(tsk);
1493         mm_release(tsk, mm);
1494 }
1495
1496 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1497 {
1498         futex_exec_release(tsk);
1499         mm_release(tsk, mm);
1500 }
1501
1502 /**
1503  * dup_mm() - duplicates an existing mm structure
1504  * @tsk: the task_struct with which the new mm will be associated.
1505  * @oldmm: the mm to duplicate.
1506  *
1507  * Allocates a new mm structure and duplicates the provided @oldmm structure
1508  * content into it.
1509  *
1510  * Return: the duplicated mm or NULL on failure.
1511  */
1512 static struct mm_struct *dup_mm(struct task_struct *tsk,
1513                                 struct mm_struct *oldmm)
1514 {
1515         struct mm_struct *mm;
1516         int err;
1517
1518         mm = allocate_mm();
1519         if (!mm)
1520                 goto fail_nomem;
1521
1522         memcpy(mm, oldmm, sizeof(*mm));
1523
1524         if (!mm_init(mm, tsk, mm->user_ns))
1525                 goto fail_nomem;
1526
1527         err = dup_mmap(mm, oldmm);
1528         if (err)
1529                 goto free_pt;
1530
1531         mm->hiwater_rss = get_mm_rss(mm);
1532         mm->hiwater_vm = mm->total_vm;
1533
1534         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1535                 goto free_pt;
1536
1537         return mm;
1538
1539 free_pt:
1540         /* don't put binfmt in mmput, we haven't got module yet */
1541         mm->binfmt = NULL;
1542         mm_init_owner(mm, NULL);
1543         mmput(mm);
1544
1545 fail_nomem:
1546         return NULL;
1547 }
1548
1549 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1550 {
1551         struct mm_struct *mm, *oldmm;
1552
1553         tsk->min_flt = tsk->maj_flt = 0;
1554         tsk->nvcsw = tsk->nivcsw = 0;
1555 #ifdef CONFIG_DETECT_HUNG_TASK
1556         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1557         tsk->last_switch_time = 0;
1558 #endif
1559
1560         tsk->mm = NULL;
1561         tsk->active_mm = NULL;
1562
1563         /*
1564          * Are we cloning a kernel thread?
1565          *
1566          * We need to steal a active VM for that..
1567          */
1568         oldmm = current->mm;
1569         if (!oldmm)
1570                 return 0;
1571
1572         if (clone_flags & CLONE_VM) {
1573                 mmget(oldmm);
1574                 mm = oldmm;
1575         } else {
1576                 mm = dup_mm(tsk, current->mm);
1577                 if (!mm)
1578                         return -ENOMEM;
1579         }
1580
1581         tsk->mm = mm;
1582         tsk->active_mm = mm;
1583         return 0;
1584 }
1585
1586 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1587 {
1588         struct fs_struct *fs = current->fs;
1589         if (clone_flags & CLONE_FS) {
1590                 /* tsk->fs is already what we want */
1591                 spin_lock(&fs->lock);
1592                 if (fs->in_exec) {
1593                         spin_unlock(&fs->lock);
1594                         return -EAGAIN;
1595                 }
1596                 fs->users++;
1597                 spin_unlock(&fs->lock);
1598                 return 0;
1599         }
1600         tsk->fs = copy_fs_struct(fs);
1601         if (!tsk->fs)
1602                 return -ENOMEM;
1603         return 0;
1604 }
1605
1606 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1607 {
1608         struct files_struct *oldf, *newf;
1609         int error = 0;
1610
1611         /*
1612          * A background process may not have any files ...
1613          */
1614         oldf = current->files;
1615         if (!oldf)
1616                 goto out;
1617
1618         if (clone_flags & CLONE_FILES) {
1619                 atomic_inc(&oldf->count);
1620                 goto out;
1621         }
1622
1623         newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1624         if (!newf)
1625                 goto out;
1626
1627         tsk->files = newf;
1628         error = 0;
1629 out:
1630         return error;
1631 }
1632
1633 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1634 {
1635         struct sighand_struct *sig;
1636
1637         if (clone_flags & CLONE_SIGHAND) {
1638                 refcount_inc(&current->sighand->count);
1639                 return 0;
1640         }
1641         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1642         RCU_INIT_POINTER(tsk->sighand, sig);
1643         if (!sig)
1644                 return -ENOMEM;
1645
1646         refcount_set(&sig->count, 1);
1647         spin_lock_irq(&current->sighand->siglock);
1648         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1649         spin_unlock_irq(&current->sighand->siglock);
1650
1651         /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1652         if (clone_flags & CLONE_CLEAR_SIGHAND)
1653                 flush_signal_handlers(tsk, 0);
1654
1655         return 0;
1656 }
1657
1658 void __cleanup_sighand(struct sighand_struct *sighand)
1659 {
1660         if (refcount_dec_and_test(&sighand->count)) {
1661                 signalfd_cleanup(sighand);
1662                 /*
1663                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1664                  * without an RCU grace period, see __lock_task_sighand().
1665                  */
1666                 kmem_cache_free(sighand_cachep, sighand);
1667         }
1668 }
1669
1670 /*
1671  * Initialize POSIX timer handling for a thread group.
1672  */
1673 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1674 {
1675         struct posix_cputimers *pct = &sig->posix_cputimers;
1676         unsigned long cpu_limit;
1677
1678         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1679         posix_cputimers_group_init(pct, cpu_limit);
1680 }
1681
1682 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1683 {
1684         struct signal_struct *sig;
1685
1686         if (clone_flags & CLONE_THREAD)
1687                 return 0;
1688
1689         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1690         tsk->signal = sig;
1691         if (!sig)
1692                 return -ENOMEM;
1693
1694         sig->nr_threads = 1;
1695         atomic_set(&sig->live, 1);
1696         refcount_set(&sig->sigcnt, 1);
1697
1698         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1699         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1700         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1701
1702         init_waitqueue_head(&sig->wait_chldexit);
1703         sig->curr_target = tsk;
1704         init_sigpending(&sig->shared_pending);
1705         INIT_HLIST_HEAD(&sig->multiprocess);
1706         seqlock_init(&sig->stats_lock);
1707         prev_cputime_init(&sig->prev_cputime);
1708
1709 #ifdef CONFIG_POSIX_TIMERS
1710         INIT_LIST_HEAD(&sig->posix_timers);
1711         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1712         sig->real_timer.function = it_real_fn;
1713 #endif
1714
1715         task_lock(current->group_leader);
1716         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1717         task_unlock(current->group_leader);
1718
1719         posix_cpu_timers_init_group(sig);
1720
1721         tty_audit_fork(sig);
1722         sched_autogroup_fork(sig);
1723
1724         sig->oom_score_adj = current->signal->oom_score_adj;
1725         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1726
1727         mutex_init(&sig->cred_guard_mutex);
1728         init_rwsem(&sig->exec_update_lock);
1729
1730         return 0;
1731 }
1732
1733 static void copy_seccomp(struct task_struct *p)
1734 {
1735 #ifdef CONFIG_SECCOMP
1736         /*
1737          * Must be called with sighand->lock held, which is common to
1738          * all threads in the group. Holding cred_guard_mutex is not
1739          * needed because this new task is not yet running and cannot
1740          * be racing exec.
1741          */
1742         assert_spin_locked(&current->sighand->siglock);
1743
1744         /* Ref-count the new filter user, and assign it. */
1745         get_seccomp_filter(current);
1746         p->seccomp = current->seccomp;
1747
1748         /*
1749          * Explicitly enable no_new_privs here in case it got set
1750          * between the task_struct being duplicated and holding the
1751          * sighand lock. The seccomp state and nnp must be in sync.
1752          */
1753         if (task_no_new_privs(current))
1754                 task_set_no_new_privs(p);
1755
1756         /*
1757          * If the parent gained a seccomp mode after copying thread
1758          * flags and between before we held the sighand lock, we have
1759          * to manually enable the seccomp thread flag here.
1760          */
1761         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1762                 set_task_syscall_work(p, SECCOMP);
1763 #endif
1764 }
1765
1766 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1767 {
1768         current->clear_child_tid = tidptr;
1769
1770         return task_pid_vnr(current);
1771 }
1772
1773 static void rt_mutex_init_task(struct task_struct *p)
1774 {
1775         raw_spin_lock_init(&p->pi_lock);
1776 #ifdef CONFIG_RT_MUTEXES
1777         p->pi_waiters = RB_ROOT_CACHED;
1778         p->pi_top_task = NULL;
1779         p->pi_blocked_on = NULL;
1780 #endif
1781 }
1782
1783 static inline void init_task_pid_links(struct task_struct *task)
1784 {
1785         enum pid_type type;
1786
1787         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1788                 INIT_HLIST_NODE(&task->pid_links[type]);
1789 }
1790
1791 static inline void
1792 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1793 {
1794         if (type == PIDTYPE_PID)
1795                 task->thread_pid = pid;
1796         else
1797                 task->signal->pids[type] = pid;
1798 }
1799
1800 static inline void rcu_copy_process(struct task_struct *p)
1801 {
1802 #ifdef CONFIG_PREEMPT_RCU
1803         p->rcu_read_lock_nesting = 0;
1804         p->rcu_read_unlock_special.s = 0;
1805         p->rcu_blocked_node = NULL;
1806         INIT_LIST_HEAD(&p->rcu_node_entry);
1807 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1808 #ifdef CONFIG_TASKS_RCU
1809         p->rcu_tasks_holdout = false;
1810         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1811         p->rcu_tasks_idle_cpu = -1;
1812 #endif /* #ifdef CONFIG_TASKS_RCU */
1813 #ifdef CONFIG_TASKS_TRACE_RCU
1814         p->trc_reader_nesting = 0;
1815         p->trc_reader_special.s = 0;
1816         INIT_LIST_HEAD(&p->trc_holdout_list);
1817         INIT_LIST_HEAD(&p->trc_blkd_node);
1818 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1819 }
1820
1821 struct pid *pidfd_pid(const struct file *file)
1822 {
1823         if (file->f_op == &pidfd_fops)
1824                 return file->private_data;
1825
1826         return ERR_PTR(-EBADF);
1827 }
1828
1829 static int pidfd_release(struct inode *inode, struct file *file)
1830 {
1831         struct pid *pid = file->private_data;
1832
1833         file->private_data = NULL;
1834         put_pid(pid);
1835         return 0;
1836 }
1837
1838 #ifdef CONFIG_PROC_FS
1839 /**
1840  * pidfd_show_fdinfo - print information about a pidfd
1841  * @m: proc fdinfo file
1842  * @f: file referencing a pidfd
1843  *
1844  * Pid:
1845  * This function will print the pid that a given pidfd refers to in the
1846  * pid namespace of the procfs instance.
1847  * If the pid namespace of the process is not a descendant of the pid
1848  * namespace of the procfs instance 0 will be shown as its pid. This is
1849  * similar to calling getppid() on a process whose parent is outside of
1850  * its pid namespace.
1851  *
1852  * NSpid:
1853  * If pid namespaces are supported then this function will also print
1854  * the pid of a given pidfd refers to for all descendant pid namespaces
1855  * starting from the current pid namespace of the instance, i.e. the
1856  * Pid field and the first entry in the NSpid field will be identical.
1857  * If the pid namespace of the process is not a descendant of the pid
1858  * namespace of the procfs instance 0 will be shown as its first NSpid
1859  * entry and no others will be shown.
1860  * Note that this differs from the Pid and NSpid fields in
1861  * /proc/<pid>/status where Pid and NSpid are always shown relative to
1862  * the  pid namespace of the procfs instance. The difference becomes
1863  * obvious when sending around a pidfd between pid namespaces from a
1864  * different branch of the tree, i.e. where no ancestral relation is
1865  * present between the pid namespaces:
1866  * - create two new pid namespaces ns1 and ns2 in the initial pid
1867  *   namespace (also take care to create new mount namespaces in the
1868  *   new pid namespace and mount procfs)
1869  * - create a process with a pidfd in ns1
1870  * - send pidfd from ns1 to ns2
1871  * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1872  *   have exactly one entry, which is 0
1873  */
1874 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1875 {
1876         struct pid *pid = f->private_data;
1877         struct pid_namespace *ns;
1878         pid_t nr = -1;
1879
1880         if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1881                 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1882                 nr = pid_nr_ns(pid, ns);
1883         }
1884
1885         seq_put_decimal_ll(m, "Pid:\t", nr);
1886
1887 #ifdef CONFIG_PID_NS
1888         seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1889         if (nr > 0) {
1890                 int i;
1891
1892                 /* If nr is non-zero it means that 'pid' is valid and that
1893                  * ns, i.e. the pid namespace associated with the procfs
1894                  * instance, is in the pid namespace hierarchy of pid.
1895                  * Start at one below the already printed level.
1896                  */
1897                 for (i = ns->level + 1; i <= pid->level; i++)
1898                         seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1899         }
1900 #endif
1901         seq_putc(m, '\n');
1902 }
1903 #endif
1904
1905 /*
1906  * Poll support for process exit notification.
1907  */
1908 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1909 {
1910         struct pid *pid = file->private_data;
1911         __poll_t poll_flags = 0;
1912
1913         poll_wait(file, &pid->wait_pidfd, pts);
1914
1915         /*
1916          * Inform pollers only when the whole thread group exits.
1917          * If the thread group leader exits before all other threads in the
1918          * group, then poll(2) should block, similar to the wait(2) family.
1919          */
1920         if (thread_group_exited(pid))
1921                 poll_flags = EPOLLIN | EPOLLRDNORM;
1922
1923         return poll_flags;
1924 }
1925
1926 const struct file_operations pidfd_fops = {
1927         .release = pidfd_release,
1928         .poll = pidfd_poll,
1929 #ifdef CONFIG_PROC_FS
1930         .show_fdinfo = pidfd_show_fdinfo,
1931 #endif
1932 };
1933
1934 static void __delayed_free_task(struct rcu_head *rhp)
1935 {
1936         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1937
1938         free_task(tsk);
1939 }
1940
1941 static __always_inline void delayed_free_task(struct task_struct *tsk)
1942 {
1943         if (IS_ENABLED(CONFIG_MEMCG))
1944                 call_rcu(&tsk->rcu, __delayed_free_task);
1945         else
1946                 free_task(tsk);
1947 }
1948
1949 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1950 {
1951         /* Skip if kernel thread */
1952         if (!tsk->mm)
1953                 return;
1954
1955         /* Skip if spawning a thread or using vfork */
1956         if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1957                 return;
1958
1959         /* We need to synchronize with __set_oom_adj */
1960         mutex_lock(&oom_adj_mutex);
1961         set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1962         /* Update the values in case they were changed after copy_signal */
1963         tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1964         tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1965         mutex_unlock(&oom_adj_mutex);
1966 }
1967
1968 #ifdef CONFIG_RV
1969 static void rv_task_fork(struct task_struct *p)
1970 {
1971         int i;
1972
1973         for (i = 0; i < RV_PER_TASK_MONITORS; i++)
1974                 p->rv[i].da_mon.monitoring = false;
1975 }
1976 #else
1977 #define rv_task_fork(p) do {} while (0)
1978 #endif
1979
1980 /*
1981  * This creates a new process as a copy of the old one,
1982  * but does not actually start it yet.
1983  *
1984  * It copies the registers, and all the appropriate
1985  * parts of the process environment (as per the clone
1986  * flags). The actual kick-off is left to the caller.
1987  */
1988 static __latent_entropy struct task_struct *copy_process(
1989                                         struct pid *pid,
1990                                         int trace,
1991                                         int node,
1992                                         struct kernel_clone_args *args)
1993 {
1994         int pidfd = -1, retval;
1995         struct task_struct *p;
1996         struct multiprocess_signals delayed;
1997         struct file *pidfile = NULL;
1998         const u64 clone_flags = args->flags;
1999         struct nsproxy *nsp = current->nsproxy;
2000
2001         /*
2002          * Don't allow sharing the root directory with processes in a different
2003          * namespace
2004          */
2005         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2006                 return ERR_PTR(-EINVAL);
2007
2008         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2009                 return ERR_PTR(-EINVAL);
2010
2011         /*
2012          * Thread groups must share signals as well, and detached threads
2013          * can only be started up within the thread group.
2014          */
2015         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2016                 return ERR_PTR(-EINVAL);
2017
2018         /*
2019          * Shared signal handlers imply shared VM. By way of the above,
2020          * thread groups also imply shared VM. Blocking this case allows
2021          * for various simplifications in other code.
2022          */
2023         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2024                 return ERR_PTR(-EINVAL);
2025
2026         /*
2027          * Siblings of global init remain as zombies on exit since they are
2028          * not reaped by their parent (swapper). To solve this and to avoid
2029          * multi-rooted process trees, prevent global and container-inits
2030          * from creating siblings.
2031          */
2032         if ((clone_flags & CLONE_PARENT) &&
2033                                 current->signal->flags & SIGNAL_UNKILLABLE)
2034                 return ERR_PTR(-EINVAL);
2035
2036         /*
2037          * If the new process will be in a different pid or user namespace
2038          * do not allow it to share a thread group with the forking task.
2039          */
2040         if (clone_flags & CLONE_THREAD) {
2041                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2042                     (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2043                         return ERR_PTR(-EINVAL);
2044         }
2045
2046         /*
2047          * If the new process will be in a different time namespace
2048          * do not allow it to share VM or a thread group with the forking task.
2049          *
2050          * On vfork, the child process enters the target time namespace only
2051          * after exec.
2052          */
2053         if ((clone_flags & (CLONE_VM | CLONE_VFORK)) == CLONE_VM) {
2054                 if (nsp->time_ns != nsp->time_ns_for_children)
2055                         return ERR_PTR(-EINVAL);
2056         }
2057
2058         if (clone_flags & CLONE_PIDFD) {
2059                 /*
2060                  * - CLONE_DETACHED is blocked so that we can potentially
2061                  *   reuse it later for CLONE_PIDFD.
2062                  * - CLONE_THREAD is blocked until someone really needs it.
2063                  */
2064                 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
2065                         return ERR_PTR(-EINVAL);
2066         }
2067
2068         /*
2069          * Force any signals received before this point to be delivered
2070          * before the fork happens.  Collect up signals sent to multiple
2071          * processes that happen during the fork and delay them so that
2072          * they appear to happen after the fork.
2073          */
2074         sigemptyset(&delayed.signal);
2075         INIT_HLIST_NODE(&delayed.node);
2076
2077         spin_lock_irq(&current->sighand->siglock);
2078         if (!(clone_flags & CLONE_THREAD))
2079                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
2080         recalc_sigpending();
2081         spin_unlock_irq(&current->sighand->siglock);
2082         retval = -ERESTARTNOINTR;
2083         if (task_sigpending(current))
2084                 goto fork_out;
2085
2086         retval = -ENOMEM;
2087         p = dup_task_struct(current, node);
2088         if (!p)
2089                 goto fork_out;
2090         p->flags &= ~PF_KTHREAD;
2091         if (args->kthread)
2092                 p->flags |= PF_KTHREAD;
2093         if (args->io_thread) {
2094                 /*
2095                  * Mark us an IO worker, and block any signal that isn't
2096                  * fatal or STOP
2097                  */
2098                 p->flags |= PF_IO_WORKER;
2099                 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2100         }
2101
2102         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2103         /*
2104          * Clear TID on mm_release()?
2105          */
2106         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2107
2108         ftrace_graph_init_task(p);
2109
2110         rt_mutex_init_task(p);
2111
2112         lockdep_assert_irqs_enabled();
2113 #ifdef CONFIG_PROVE_LOCKING
2114         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2115 #endif
2116         retval = copy_creds(p, clone_flags);
2117         if (retval < 0)
2118                 goto bad_fork_free;
2119
2120         retval = -EAGAIN;
2121         if (is_ucounts_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2122                 if (p->real_cred->user != INIT_USER &&
2123                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2124                         goto bad_fork_cleanup_count;
2125         }
2126         current->flags &= ~PF_NPROC_EXCEEDED;
2127
2128         /*
2129          * If multiple threads are within copy_process(), then this check
2130          * triggers too late. This doesn't hurt, the check is only there
2131          * to stop root fork bombs.
2132          */
2133         retval = -EAGAIN;
2134         if (data_race(nr_threads >= max_threads))
2135                 goto bad_fork_cleanup_count;
2136
2137         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
2138         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2139         p->flags |= PF_FORKNOEXEC;
2140         INIT_LIST_HEAD(&p->children);
2141         INIT_LIST_HEAD(&p->sibling);
2142         rcu_copy_process(p);
2143         p->vfork_done = NULL;
2144         spin_lock_init(&p->alloc_lock);
2145
2146         init_sigpending(&p->pending);
2147
2148         p->utime = p->stime = p->gtime = 0;
2149 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2150         p->utimescaled = p->stimescaled = 0;
2151 #endif
2152         prev_cputime_init(&p->prev_cputime);
2153
2154 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2155         seqcount_init(&p->vtime.seqcount);
2156         p->vtime.starttime = 0;
2157         p->vtime.state = VTIME_INACTIVE;
2158 #endif
2159
2160 #ifdef CONFIG_IO_URING
2161         p->io_uring = NULL;
2162 #endif
2163
2164 #if defined(SPLIT_RSS_COUNTING)
2165         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2166 #endif
2167
2168         p->default_timer_slack_ns = current->timer_slack_ns;
2169
2170 #ifdef CONFIG_PSI
2171         p->psi_flags = 0;
2172 #endif
2173
2174         task_io_accounting_init(&p->ioac);
2175         acct_clear_integrals(p);
2176
2177         posix_cputimers_init(&p->posix_cputimers);
2178
2179         p->io_context = NULL;
2180         audit_set_context(p, NULL);
2181         cgroup_fork(p);
2182         if (args->kthread) {
2183                 if (!set_kthread_struct(p))
2184                         goto bad_fork_cleanup_delayacct;
2185         }
2186 #ifdef CONFIG_NUMA
2187         p->mempolicy = mpol_dup(p->mempolicy);
2188         if (IS_ERR(p->mempolicy)) {
2189                 retval = PTR_ERR(p->mempolicy);
2190                 p->mempolicy = NULL;
2191                 goto bad_fork_cleanup_delayacct;
2192         }
2193 #endif
2194 #ifdef CONFIG_CPUSETS
2195         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2196         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2197         seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2198 #endif
2199 #ifdef CONFIG_TRACE_IRQFLAGS
2200         memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2201         p->irqtrace.hardirq_disable_ip  = _THIS_IP_;
2202         p->irqtrace.softirq_enable_ip   = _THIS_IP_;
2203         p->softirqs_enabled             = 1;
2204         p->softirq_context              = 0;
2205 #endif
2206
2207         p->pagefault_disabled = 0;
2208
2209 #ifdef CONFIG_LOCKDEP
2210         lockdep_init_task(p);
2211 #endif
2212
2213 #ifdef CONFIG_DEBUG_MUTEXES
2214         p->blocked_on = NULL; /* not blocked yet */
2215 #endif
2216 #ifdef CONFIG_BCACHE
2217         p->sequential_io        = 0;
2218         p->sequential_io_avg    = 0;
2219 #endif
2220 #ifdef CONFIG_BPF_SYSCALL
2221         RCU_INIT_POINTER(p->bpf_storage, NULL);
2222         p->bpf_ctx = NULL;
2223 #endif
2224
2225         /* Perform scheduler related setup. Assign this task to a CPU. */
2226         retval = sched_fork(clone_flags, p);
2227         if (retval)
2228                 goto bad_fork_cleanup_policy;
2229
2230         retval = perf_event_init_task(p, clone_flags);
2231         if (retval)
2232                 goto bad_fork_cleanup_policy;
2233         retval = audit_alloc(p);
2234         if (retval)
2235                 goto bad_fork_cleanup_perf;
2236         /* copy all the process information */
2237         shm_init_task(p);
2238         retval = security_task_alloc(p, clone_flags);
2239         if (retval)
2240                 goto bad_fork_cleanup_audit;
2241         retval = copy_semundo(clone_flags, p);
2242         if (retval)
2243                 goto bad_fork_cleanup_security;
2244         retval = copy_files(clone_flags, p);
2245         if (retval)
2246                 goto bad_fork_cleanup_semundo;
2247         retval = copy_fs(clone_flags, p);
2248         if (retval)
2249                 goto bad_fork_cleanup_files;
2250         retval = copy_sighand(clone_flags, p);
2251         if (retval)
2252                 goto bad_fork_cleanup_fs;
2253         retval = copy_signal(clone_flags, p);
2254         if (retval)
2255                 goto bad_fork_cleanup_sighand;
2256         retval = copy_mm(clone_flags, p);
2257         if (retval)
2258                 goto bad_fork_cleanup_signal;
2259         retval = copy_namespaces(clone_flags, p);
2260         if (retval)
2261                 goto bad_fork_cleanup_mm;
2262         retval = copy_io(clone_flags, p);
2263         if (retval)
2264                 goto bad_fork_cleanup_namespaces;
2265         retval = copy_thread(p, args);
2266         if (retval)
2267                 goto bad_fork_cleanup_io;
2268
2269         stackleak_task_init(p);
2270
2271         if (pid != &init_struct_pid) {
2272                 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2273                                 args->set_tid_size);
2274                 if (IS_ERR(pid)) {
2275                         retval = PTR_ERR(pid);
2276                         goto bad_fork_cleanup_thread;
2277                 }
2278         }
2279
2280         /*
2281          * This has to happen after we've potentially unshared the file
2282          * descriptor table (so that the pidfd doesn't leak into the child
2283          * if the fd table isn't shared).
2284          */
2285         if (clone_flags & CLONE_PIDFD) {
2286                 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2287                 if (retval < 0)
2288                         goto bad_fork_free_pid;
2289
2290                 pidfd = retval;
2291
2292                 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2293                                               O_RDWR | O_CLOEXEC);
2294                 if (IS_ERR(pidfile)) {
2295                         put_unused_fd(pidfd);
2296                         retval = PTR_ERR(pidfile);
2297                         goto bad_fork_free_pid;
2298                 }
2299                 get_pid(pid);   /* held by pidfile now */
2300
2301                 retval = put_user(pidfd, args->pidfd);
2302                 if (retval)
2303                         goto bad_fork_put_pidfd;
2304         }
2305
2306 #ifdef CONFIG_BLOCK
2307         p->plug = NULL;
2308 #endif
2309         futex_init_task(p);
2310
2311         /*
2312          * sigaltstack should be cleared when sharing the same VM
2313          */
2314         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2315                 sas_ss_reset(p);
2316
2317         /*
2318          * Syscall tracing and stepping should be turned off in the
2319          * child regardless of CLONE_PTRACE.
2320          */
2321         user_disable_single_step(p);
2322         clear_task_syscall_work(p, SYSCALL_TRACE);
2323 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2324         clear_task_syscall_work(p, SYSCALL_EMU);
2325 #endif
2326         clear_tsk_latency_tracing(p);
2327
2328         /* ok, now we should be set up.. */
2329         p->pid = pid_nr(pid);
2330         if (clone_flags & CLONE_THREAD) {
2331                 p->group_leader = current->group_leader;
2332                 p->tgid = current->tgid;
2333         } else {
2334                 p->group_leader = p;
2335                 p->tgid = p->pid;
2336         }
2337
2338         p->nr_dirtied = 0;
2339         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2340         p->dirty_paused_when = 0;
2341
2342         p->pdeath_signal = 0;
2343         INIT_LIST_HEAD(&p->thread_group);
2344         p->task_works = NULL;
2345         clear_posix_cputimers_work(p);
2346
2347 #ifdef CONFIG_KRETPROBES
2348         p->kretprobe_instances.first = NULL;
2349 #endif
2350 #ifdef CONFIG_RETHOOK
2351         p->rethooks.first = NULL;
2352 #endif
2353
2354         /*
2355          * Ensure that the cgroup subsystem policies allow the new process to be
2356          * forked. It should be noted that the new process's css_set can be changed
2357          * between here and cgroup_post_fork() if an organisation operation is in
2358          * progress.
2359          */
2360         retval = cgroup_can_fork(p, args);
2361         if (retval)
2362                 goto bad_fork_put_pidfd;
2363
2364         /*
2365          * Now that the cgroups are pinned, re-clone the parent cgroup and put
2366          * the new task on the correct runqueue. All this *before* the task
2367          * becomes visible.
2368          *
2369          * This isn't part of ->can_fork() because while the re-cloning is
2370          * cgroup specific, it unconditionally needs to place the task on a
2371          * runqueue.
2372          */
2373         sched_cgroup_fork(p, args);
2374
2375         /*
2376          * From this point on we must avoid any synchronous user-space
2377          * communication until we take the tasklist-lock. In particular, we do
2378          * not want user-space to be able to predict the process start-time by
2379          * stalling fork(2) after we recorded the start_time but before it is
2380          * visible to the system.
2381          */
2382
2383         p->start_time = ktime_get_ns();
2384         p->start_boottime = ktime_get_boottime_ns();
2385
2386         /*
2387          * Make it visible to the rest of the system, but dont wake it up yet.
2388          * Need tasklist lock for parent etc handling!
2389          */
2390         write_lock_irq(&tasklist_lock);
2391
2392         /* CLONE_PARENT re-uses the old parent */
2393         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2394                 p->real_parent = current->real_parent;
2395                 p->parent_exec_id = current->parent_exec_id;
2396                 if (clone_flags & CLONE_THREAD)
2397                         p->exit_signal = -1;
2398                 else
2399                         p->exit_signal = current->group_leader->exit_signal;
2400         } else {
2401                 p->real_parent = current;
2402                 p->parent_exec_id = current->self_exec_id;
2403                 p->exit_signal = args->exit_signal;
2404         }
2405
2406         klp_copy_process(p);
2407
2408         sched_core_fork(p);
2409
2410         spin_lock(&current->sighand->siglock);
2411
2412         /*
2413          * Copy seccomp details explicitly here, in case they were changed
2414          * before holding sighand lock.
2415          */
2416         copy_seccomp(p);
2417
2418         rv_task_fork(p);
2419
2420         rseq_fork(p, clone_flags);
2421
2422         /* Don't start children in a dying pid namespace */
2423         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2424                 retval = -ENOMEM;
2425                 goto bad_fork_cancel_cgroup;
2426         }
2427
2428         /* Let kill terminate clone/fork in the middle */
2429         if (fatal_signal_pending(current)) {
2430                 retval = -EINTR;
2431                 goto bad_fork_cancel_cgroup;
2432         }
2433
2434         init_task_pid_links(p);
2435         if (likely(p->pid)) {
2436                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2437
2438                 init_task_pid(p, PIDTYPE_PID, pid);
2439                 if (thread_group_leader(p)) {
2440                         init_task_pid(p, PIDTYPE_TGID, pid);
2441                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2442                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2443
2444                         if (is_child_reaper(pid)) {
2445                                 ns_of_pid(pid)->child_reaper = p;
2446                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2447                         }
2448                         p->signal->shared_pending.signal = delayed.signal;
2449                         p->signal->tty = tty_kref_get(current->signal->tty);
2450                         /*
2451                          * Inherit has_child_subreaper flag under the same
2452                          * tasklist_lock with adding child to the process tree
2453                          * for propagate_has_child_subreaper optimization.
2454                          */
2455                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2456                                                          p->real_parent->signal->is_child_subreaper;
2457                         list_add_tail(&p->sibling, &p->real_parent->children);
2458                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2459                         attach_pid(p, PIDTYPE_TGID);
2460                         attach_pid(p, PIDTYPE_PGID);
2461                         attach_pid(p, PIDTYPE_SID);
2462                         __this_cpu_inc(process_counts);
2463                 } else {
2464                         current->signal->nr_threads++;
2465                         atomic_inc(&current->signal->live);
2466                         refcount_inc(&current->signal->sigcnt);
2467                         task_join_group_stop(p);
2468                         list_add_tail_rcu(&p->thread_group,
2469                                           &p->group_leader->thread_group);
2470                         list_add_tail_rcu(&p->thread_node,
2471                                           &p->signal->thread_head);
2472                 }
2473                 attach_pid(p, PIDTYPE_PID);
2474                 nr_threads++;
2475         }
2476         total_forks++;
2477         hlist_del_init(&delayed.node);
2478         spin_unlock(&current->sighand->siglock);
2479         syscall_tracepoint_update(p);
2480         write_unlock_irq(&tasklist_lock);
2481
2482         if (pidfile)
2483                 fd_install(pidfd, pidfile);
2484
2485         proc_fork_connector(p);
2486         sched_post_fork(p);
2487         cgroup_post_fork(p, args);
2488         perf_event_fork(p);
2489
2490         trace_task_newtask(p, clone_flags);
2491         uprobe_copy_process(p, clone_flags);
2492
2493         copy_oom_score_adj(clone_flags, p);
2494
2495         return p;
2496
2497 bad_fork_cancel_cgroup:
2498         sched_core_free(p);
2499         spin_unlock(&current->sighand->siglock);
2500         write_unlock_irq(&tasklist_lock);
2501         cgroup_cancel_fork(p, args);
2502 bad_fork_put_pidfd:
2503         if (clone_flags & CLONE_PIDFD) {
2504                 fput(pidfile);
2505                 put_unused_fd(pidfd);
2506         }
2507 bad_fork_free_pid:
2508         if (pid != &init_struct_pid)
2509                 free_pid(pid);
2510 bad_fork_cleanup_thread:
2511         exit_thread(p);
2512 bad_fork_cleanup_io:
2513         if (p->io_context)
2514                 exit_io_context(p);
2515 bad_fork_cleanup_namespaces:
2516         exit_task_namespaces(p);
2517 bad_fork_cleanup_mm:
2518         if (p->mm) {
2519                 mm_clear_owner(p->mm, p);
2520                 mmput(p->mm);
2521         }
2522 bad_fork_cleanup_signal:
2523         if (!(clone_flags & CLONE_THREAD))
2524                 free_signal_struct(p->signal);
2525 bad_fork_cleanup_sighand:
2526         __cleanup_sighand(p->sighand);
2527 bad_fork_cleanup_fs:
2528         exit_fs(p); /* blocking */
2529 bad_fork_cleanup_files:
2530         exit_files(p); /* blocking */
2531 bad_fork_cleanup_semundo:
2532         exit_sem(p);
2533 bad_fork_cleanup_security:
2534         security_task_free(p);
2535 bad_fork_cleanup_audit:
2536         audit_free(p);
2537 bad_fork_cleanup_perf:
2538         perf_event_free_task(p);
2539 bad_fork_cleanup_policy:
2540         lockdep_free_task(p);
2541 #ifdef CONFIG_NUMA
2542         mpol_put(p->mempolicy);
2543 #endif
2544 bad_fork_cleanup_delayacct:
2545         delayacct_tsk_free(p);
2546 bad_fork_cleanup_count:
2547         dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2548         exit_creds(p);
2549 bad_fork_free:
2550         WRITE_ONCE(p->__state, TASK_DEAD);
2551         exit_task_stack_account(p);
2552         put_task_stack(p);
2553         delayed_free_task(p);
2554 fork_out:
2555         spin_lock_irq(&current->sighand->siglock);
2556         hlist_del_init(&delayed.node);
2557         spin_unlock_irq(&current->sighand->siglock);
2558         return ERR_PTR(retval);
2559 }
2560
2561 static inline void init_idle_pids(struct task_struct *idle)
2562 {
2563         enum pid_type type;
2564
2565         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2566                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2567                 init_task_pid(idle, type, &init_struct_pid);
2568         }
2569 }
2570
2571 static int idle_dummy(void *dummy)
2572 {
2573         /* This function is never called */
2574         return 0;
2575 }
2576
2577 struct task_struct * __init fork_idle(int cpu)
2578 {
2579         struct task_struct *task;
2580         struct kernel_clone_args args = {
2581                 .flags          = CLONE_VM,
2582                 .fn             = &idle_dummy,
2583                 .fn_arg         = NULL,
2584                 .kthread        = 1,
2585                 .idle           = 1,
2586         };
2587
2588         task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2589         if (!IS_ERR(task)) {
2590                 init_idle_pids(task);
2591                 init_idle(task, cpu);
2592         }
2593
2594         return task;
2595 }
2596
2597 struct mm_struct *copy_init_mm(void)
2598 {
2599         return dup_mm(NULL, &init_mm);
2600 }
2601
2602 /*
2603  * This is like kernel_clone(), but shaved down and tailored to just
2604  * creating io_uring workers. It returns a created task, or an error pointer.
2605  * The returned task is inactive, and the caller must fire it up through
2606  * wake_up_new_task(p). All signals are blocked in the created task.
2607  */
2608 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2609 {
2610         unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2611                                 CLONE_IO;
2612         struct kernel_clone_args args = {
2613                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2614                                     CLONE_UNTRACED) & ~CSIGNAL),
2615                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2616                 .fn             = fn,
2617                 .fn_arg         = arg,
2618                 .io_thread      = 1,
2619         };
2620
2621         return copy_process(NULL, 0, node, &args);
2622 }
2623
2624 /*
2625  *  Ok, this is the main fork-routine.
2626  *
2627  * It copies the process, and if successful kick-starts
2628  * it and waits for it to finish using the VM if required.
2629  *
2630  * args->exit_signal is expected to be checked for sanity by the caller.
2631  */
2632 pid_t kernel_clone(struct kernel_clone_args *args)
2633 {
2634         u64 clone_flags = args->flags;
2635         struct completion vfork;
2636         struct pid *pid;
2637         struct task_struct *p;
2638         int trace = 0;
2639         pid_t nr;
2640
2641         /*
2642          * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2643          * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2644          * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2645          * field in struct clone_args and it still doesn't make sense to have
2646          * them both point at the same memory location. Performing this check
2647          * here has the advantage that we don't need to have a separate helper
2648          * to check for legacy clone().
2649          */
2650         if ((args->flags & CLONE_PIDFD) &&
2651             (args->flags & CLONE_PARENT_SETTID) &&
2652             (args->pidfd == args->parent_tid))
2653                 return -EINVAL;
2654
2655         /*
2656          * Determine whether and which event to report to ptracer.  When
2657          * called from kernel_thread or CLONE_UNTRACED is explicitly
2658          * requested, no event is reported; otherwise, report if the event
2659          * for the type of forking is enabled.
2660          */
2661         if (!(clone_flags & CLONE_UNTRACED)) {
2662                 if (clone_flags & CLONE_VFORK)
2663                         trace = PTRACE_EVENT_VFORK;
2664                 else if (args->exit_signal != SIGCHLD)
2665                         trace = PTRACE_EVENT_CLONE;
2666                 else
2667                         trace = PTRACE_EVENT_FORK;
2668
2669                 if (likely(!ptrace_event_enabled(current, trace)))
2670                         trace = 0;
2671         }
2672
2673         p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2674         add_latent_entropy();
2675
2676         if (IS_ERR(p))
2677                 return PTR_ERR(p);
2678
2679         /*
2680          * Do this prior waking up the new thread - the thread pointer
2681          * might get invalid after that point, if the thread exits quickly.
2682          */
2683         trace_sched_process_fork(current, p);
2684
2685         pid = get_task_pid(p, PIDTYPE_PID);
2686         nr = pid_vnr(pid);
2687
2688         if (clone_flags & CLONE_PARENT_SETTID)
2689                 put_user(nr, args->parent_tid);
2690
2691         if (clone_flags & CLONE_VFORK) {
2692                 p->vfork_done = &vfork;
2693                 init_completion(&vfork);
2694                 get_task_struct(p);
2695         }
2696
2697         if (IS_ENABLED(CONFIG_LRU_GEN) && !(clone_flags & CLONE_VM)) {
2698                 /* lock the task to synchronize with memcg migration */
2699                 task_lock(p);
2700                 lru_gen_add_mm(p->mm);
2701                 task_unlock(p);
2702         }
2703
2704         wake_up_new_task(p);
2705
2706         /* forking complete and child started to run, tell ptracer */
2707         if (unlikely(trace))
2708                 ptrace_event_pid(trace, pid);
2709
2710         if (clone_flags & CLONE_VFORK) {
2711                 if (!wait_for_vfork_done(p, &vfork))
2712                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2713         }
2714
2715         put_pid(pid);
2716         return nr;
2717 }
2718
2719 /*
2720  * Create a kernel thread.
2721  */
2722 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2723 {
2724         struct kernel_clone_args args = {
2725                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2726                                     CLONE_UNTRACED) & ~CSIGNAL),
2727                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2728                 .fn             = fn,
2729                 .fn_arg         = arg,
2730                 .kthread        = 1,
2731         };
2732
2733         return kernel_clone(&args);
2734 }
2735
2736 /*
2737  * Create a user mode thread.
2738  */
2739 pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2740 {
2741         struct kernel_clone_args args = {
2742                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2743                                     CLONE_UNTRACED) & ~CSIGNAL),
2744                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2745                 .fn             = fn,
2746                 .fn_arg         = arg,
2747         };
2748
2749         return kernel_clone(&args);
2750 }
2751
2752 #ifdef __ARCH_WANT_SYS_FORK
2753 SYSCALL_DEFINE0(fork)
2754 {
2755 #ifdef CONFIG_MMU
2756         struct kernel_clone_args args = {
2757                 .exit_signal = SIGCHLD,
2758         };
2759
2760         return kernel_clone(&args);
2761 #else
2762         /* can not support in nommu mode */
2763         return -EINVAL;
2764 #endif
2765 }
2766 #endif
2767
2768 #ifdef __ARCH_WANT_SYS_VFORK
2769 SYSCALL_DEFINE0(vfork)
2770 {
2771         struct kernel_clone_args args = {
2772                 .flags          = CLONE_VFORK | CLONE_VM,
2773                 .exit_signal    = SIGCHLD,
2774         };
2775
2776         return kernel_clone(&args);
2777 }
2778 #endif
2779
2780 #ifdef __ARCH_WANT_SYS_CLONE
2781 #ifdef CONFIG_CLONE_BACKWARDS
2782 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2783                  int __user *, parent_tidptr,
2784                  unsigned long, tls,
2785                  int __user *, child_tidptr)
2786 #elif defined(CONFIG_CLONE_BACKWARDS2)
2787 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2788                  int __user *, parent_tidptr,
2789                  int __user *, child_tidptr,
2790                  unsigned long, tls)
2791 #elif defined(CONFIG_CLONE_BACKWARDS3)
2792 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2793                 int, stack_size,
2794                 int __user *, parent_tidptr,
2795                 int __user *, child_tidptr,
2796                 unsigned long, tls)
2797 #else
2798 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2799                  int __user *, parent_tidptr,
2800                  int __user *, child_tidptr,
2801                  unsigned long, tls)
2802 #endif
2803 {
2804         struct kernel_clone_args args = {
2805                 .flags          = (lower_32_bits(clone_flags) & ~CSIGNAL),
2806                 .pidfd          = parent_tidptr,
2807                 .child_tid      = child_tidptr,
2808                 .parent_tid     = parent_tidptr,
2809                 .exit_signal    = (lower_32_bits(clone_flags) & CSIGNAL),
2810                 .stack          = newsp,
2811                 .tls            = tls,
2812         };
2813
2814         return kernel_clone(&args);
2815 }
2816 #endif
2817
2818 #ifdef __ARCH_WANT_SYS_CLONE3
2819
2820 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2821                                               struct clone_args __user *uargs,
2822                                               size_t usize)
2823 {
2824         int err;
2825         struct clone_args args;
2826         pid_t *kset_tid = kargs->set_tid;
2827
2828         BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2829                      CLONE_ARGS_SIZE_VER0);
2830         BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2831                      CLONE_ARGS_SIZE_VER1);
2832         BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2833                      CLONE_ARGS_SIZE_VER2);
2834         BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2835
2836         if (unlikely(usize > PAGE_SIZE))
2837                 return -E2BIG;
2838         if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2839                 return -EINVAL;
2840
2841         err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2842         if (err)
2843                 return err;
2844
2845         if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2846                 return -EINVAL;
2847
2848         if (unlikely(!args.set_tid && args.set_tid_size > 0))
2849                 return -EINVAL;
2850
2851         if (unlikely(args.set_tid && args.set_tid_size == 0))
2852                 return -EINVAL;
2853
2854         /*
2855          * Verify that higher 32bits of exit_signal are unset and that
2856          * it is a valid signal
2857          */
2858         if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2859                      !valid_signal(args.exit_signal)))
2860                 return -EINVAL;
2861
2862         if ((args.flags & CLONE_INTO_CGROUP) &&
2863             (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2864                 return -EINVAL;
2865
2866         *kargs = (struct kernel_clone_args){
2867                 .flags          = args.flags,
2868                 .pidfd          = u64_to_user_ptr(args.pidfd),
2869                 .child_tid      = u64_to_user_ptr(args.child_tid),
2870                 .parent_tid     = u64_to_user_ptr(args.parent_tid),
2871                 .exit_signal    = args.exit_signal,
2872                 .stack          = args.stack,
2873                 .stack_size     = args.stack_size,
2874                 .tls            = args.tls,
2875                 .set_tid_size   = args.set_tid_size,
2876                 .cgroup         = args.cgroup,
2877         };
2878
2879         if (args.set_tid &&
2880                 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2881                         (kargs->set_tid_size * sizeof(pid_t))))
2882                 return -EFAULT;
2883
2884         kargs->set_tid = kset_tid;
2885
2886         return 0;
2887 }
2888
2889 /**
2890  * clone3_stack_valid - check and prepare stack
2891  * @kargs: kernel clone args
2892  *
2893  * Verify that the stack arguments userspace gave us are sane.
2894  * In addition, set the stack direction for userspace since it's easy for us to
2895  * determine.
2896  */
2897 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2898 {
2899         if (kargs->stack == 0) {
2900                 if (kargs->stack_size > 0)
2901                         return false;
2902         } else {
2903                 if (kargs->stack_size == 0)
2904                         return false;
2905
2906                 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2907                         return false;
2908
2909 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2910                 kargs->stack += kargs->stack_size;
2911 #endif
2912         }
2913
2914         return true;
2915 }
2916
2917 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2918 {
2919         /* Verify that no unknown flags are passed along. */
2920         if (kargs->flags &
2921             ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2922                 return false;
2923
2924         /*
2925          * - make the CLONE_DETACHED bit reusable for clone3
2926          * - make the CSIGNAL bits reusable for clone3
2927          */
2928         if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2929                 return false;
2930
2931         if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2932             (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2933                 return false;
2934
2935         if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2936             kargs->exit_signal)
2937                 return false;
2938
2939         if (!clone3_stack_valid(kargs))
2940                 return false;
2941
2942         return true;
2943 }
2944
2945 /**
2946  * clone3 - create a new process with specific properties
2947  * @uargs: argument structure
2948  * @size:  size of @uargs
2949  *
2950  * clone3() is the extensible successor to clone()/clone2().
2951  * It takes a struct as argument that is versioned by its size.
2952  *
2953  * Return: On success, a positive PID for the child process.
2954  *         On error, a negative errno number.
2955  */
2956 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2957 {
2958         int err;
2959
2960         struct kernel_clone_args kargs;
2961         pid_t set_tid[MAX_PID_NS_LEVEL];
2962
2963         kargs.set_tid = set_tid;
2964
2965         err = copy_clone_args_from_user(&kargs, uargs, size);
2966         if (err)
2967                 return err;
2968
2969         if (!clone3_args_valid(&kargs))
2970                 return -EINVAL;
2971
2972         return kernel_clone(&kargs);
2973 }
2974 #endif
2975
2976 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2977 {
2978         struct task_struct *leader, *parent, *child;
2979         int res;
2980
2981         read_lock(&tasklist_lock);
2982         leader = top = top->group_leader;
2983 down:
2984         for_each_thread(leader, parent) {
2985                 list_for_each_entry(child, &parent->children, sibling) {
2986                         res = visitor(child, data);
2987                         if (res) {
2988                                 if (res < 0)
2989                                         goto out;
2990                                 leader = child;
2991                                 goto down;
2992                         }
2993 up:
2994                         ;
2995                 }
2996         }
2997
2998         if (leader != top) {
2999                 child = leader;
3000                 parent = child->real_parent;
3001                 leader = parent->group_leader;
3002                 goto up;
3003         }
3004 out:
3005         read_unlock(&tasklist_lock);
3006 }
3007
3008 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
3009 #define ARCH_MIN_MMSTRUCT_ALIGN 0
3010 #endif
3011
3012 static void sighand_ctor(void *data)
3013 {
3014         struct sighand_struct *sighand = data;
3015
3016         spin_lock_init(&sighand->siglock);
3017         init_waitqueue_head(&sighand->signalfd_wqh);
3018 }
3019
3020 void __init proc_caches_init(void)
3021 {
3022         unsigned int mm_size;
3023
3024         sighand_cachep = kmem_cache_create("sighand_cache",
3025                         sizeof(struct sighand_struct), 0,
3026                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3027                         SLAB_ACCOUNT, sighand_ctor);
3028         signal_cachep = kmem_cache_create("signal_cache",
3029                         sizeof(struct signal_struct), 0,
3030                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3031                         NULL);
3032         files_cachep = kmem_cache_create("files_cache",
3033                         sizeof(struct files_struct), 0,
3034                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3035                         NULL);
3036         fs_cachep = kmem_cache_create("fs_cache",
3037                         sizeof(struct fs_struct), 0,
3038                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3039                         NULL);
3040
3041         /*
3042          * The mm_cpumask is located at the end of mm_struct, and is
3043          * dynamically sized based on the maximum CPU number this system
3044          * can have, taking hotplug into account (nr_cpu_ids).
3045          */
3046         mm_size = sizeof(struct mm_struct) + cpumask_size();
3047
3048         mm_cachep = kmem_cache_create_usercopy("mm_struct",
3049                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3050                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3051                         offsetof(struct mm_struct, saved_auxv),
3052                         sizeof_field(struct mm_struct, saved_auxv),
3053                         NULL);
3054         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3055         mmap_init();
3056         nsproxy_cache_init();
3057 }
3058
3059 /*
3060  * Check constraints on flags passed to the unshare system call.
3061  */
3062 static int check_unshare_flags(unsigned long unshare_flags)
3063 {
3064         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3065                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3066                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3067                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3068                                 CLONE_NEWTIME))
3069                 return -EINVAL;
3070         /*
3071          * Not implemented, but pretend it works if there is nothing
3072          * to unshare.  Note that unsharing the address space or the
3073          * signal handlers also need to unshare the signal queues (aka
3074          * CLONE_THREAD).
3075          */
3076         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3077                 if (!thread_group_empty(current))
3078                         return -EINVAL;
3079         }
3080         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3081                 if (refcount_read(&current->sighand->count) > 1)
3082                         return -EINVAL;
3083         }
3084         if (unshare_flags & CLONE_VM) {
3085                 if (!current_is_single_threaded())
3086                         return -EINVAL;
3087         }
3088
3089         return 0;
3090 }
3091
3092 /*
3093  * Unshare the filesystem structure if it is being shared
3094  */
3095 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3096 {
3097         struct fs_struct *fs = current->fs;
3098
3099         if (!(unshare_flags & CLONE_FS) || !fs)
3100                 return 0;
3101
3102         /* don't need lock here; in the worst case we'll do useless copy */
3103         if (fs->users == 1)
3104                 return 0;
3105
3106         *new_fsp = copy_fs_struct(fs);
3107         if (!*new_fsp)
3108                 return -ENOMEM;
3109
3110         return 0;
3111 }
3112
3113 /*
3114  * Unshare file descriptor table if it is being shared
3115  */
3116 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3117                struct files_struct **new_fdp)
3118 {
3119         struct files_struct *fd = current->files;
3120         int error = 0;
3121
3122         if ((unshare_flags & CLONE_FILES) &&
3123             (fd && atomic_read(&fd->count) > 1)) {
3124                 *new_fdp = dup_fd(fd, max_fds, &error);
3125                 if (!*new_fdp)
3126                         return error;
3127         }
3128
3129         return 0;
3130 }
3131
3132 /*
3133  * unshare allows a process to 'unshare' part of the process
3134  * context which was originally shared using clone.  copy_*
3135  * functions used by kernel_clone() cannot be used here directly
3136  * because they modify an inactive task_struct that is being
3137  * constructed. Here we are modifying the current, active,
3138  * task_struct.
3139  */
3140 int ksys_unshare(unsigned long unshare_flags)
3141 {
3142         struct fs_struct *fs, *new_fs = NULL;
3143         struct files_struct *new_fd = NULL;
3144         struct cred *new_cred = NULL;
3145         struct nsproxy *new_nsproxy = NULL;
3146         int do_sysvsem = 0;
3147         int err;
3148
3149         /*
3150          * If unsharing a user namespace must also unshare the thread group
3151          * and unshare the filesystem root and working directories.
3152          */
3153         if (unshare_flags & CLONE_NEWUSER)
3154                 unshare_flags |= CLONE_THREAD | CLONE_FS;
3155         /*
3156          * If unsharing vm, must also unshare signal handlers.
3157          */
3158         if (unshare_flags & CLONE_VM)
3159                 unshare_flags |= CLONE_SIGHAND;
3160         /*
3161          * If unsharing a signal handlers, must also unshare the signal queues.
3162          */
3163         if (unshare_flags & CLONE_SIGHAND)
3164                 unshare_flags |= CLONE_THREAD;
3165         /*
3166          * If unsharing namespace, must also unshare filesystem information.
3167          */
3168         if (unshare_flags & CLONE_NEWNS)
3169                 unshare_flags |= CLONE_FS;
3170
3171         err = check_unshare_flags(unshare_flags);
3172         if (err)
3173                 goto bad_unshare_out;
3174         /*
3175          * CLONE_NEWIPC must also detach from the undolist: after switching
3176          * to a new ipc namespace, the semaphore arrays from the old
3177          * namespace are unreachable.
3178          */
3179         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3180                 do_sysvsem = 1;
3181         err = unshare_fs(unshare_flags, &new_fs);
3182         if (err)
3183                 goto bad_unshare_out;
3184         err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3185         if (err)
3186                 goto bad_unshare_cleanup_fs;
3187         err = unshare_userns(unshare_flags, &new_cred);
3188         if (err)
3189                 goto bad_unshare_cleanup_fd;
3190         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3191                                          new_cred, new_fs);
3192         if (err)
3193                 goto bad_unshare_cleanup_cred;
3194
3195         if (new_cred) {
3196                 err = set_cred_ucounts(new_cred);
3197                 if (err)
3198                         goto bad_unshare_cleanup_cred;
3199         }
3200
3201         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3202                 if (do_sysvsem) {
3203                         /*
3204                          * CLONE_SYSVSEM is equivalent to sys_exit().
3205                          */
3206                         exit_sem(current);
3207                 }
3208                 if (unshare_flags & CLONE_NEWIPC) {
3209                         /* Orphan segments in old ns (see sem above). */
3210                         exit_shm(current);
3211                         shm_init_task(current);
3212                 }
3213
3214                 if (new_nsproxy)
3215                         switch_task_namespaces(current, new_nsproxy);
3216
3217                 task_lock(current);
3218
3219                 if (new_fs) {
3220                         fs = current->fs;
3221                         spin_lock(&fs->lock);
3222                         current->fs = new_fs;
3223                         if (--fs->users)
3224                                 new_fs = NULL;
3225                         else
3226                                 new_fs = fs;
3227                         spin_unlock(&fs->lock);
3228                 }
3229
3230                 if (new_fd)
3231                         swap(current->files, new_fd);
3232
3233                 task_unlock(current);
3234
3235                 if (new_cred) {
3236                         /* Install the new user namespace */
3237                         commit_creds(new_cred);
3238                         new_cred = NULL;
3239                 }
3240         }
3241
3242         perf_event_namespaces(current);
3243
3244 bad_unshare_cleanup_cred:
3245         if (new_cred)
3246                 put_cred(new_cred);
3247 bad_unshare_cleanup_fd:
3248         if (new_fd)
3249                 put_files_struct(new_fd);
3250
3251 bad_unshare_cleanup_fs:
3252         if (new_fs)
3253                 free_fs_struct(new_fs);
3254
3255 bad_unshare_out:
3256         return err;
3257 }
3258
3259 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3260 {
3261         return ksys_unshare(unshare_flags);
3262 }
3263
3264 /*
3265  *      Helper to unshare the files of the current task.
3266  *      We don't want to expose copy_files internals to
3267  *      the exec layer of the kernel.
3268  */
3269
3270 int unshare_files(void)
3271 {
3272         struct task_struct *task = current;
3273         struct files_struct *old, *copy = NULL;
3274         int error;
3275
3276         error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, &copy);
3277         if (error || !copy)
3278                 return error;
3279
3280         old = task->files;
3281         task_lock(task);
3282         task->files = copy;
3283         task_unlock(task);
3284         put_files_struct(old);
3285         return 0;
3286 }
3287
3288 int sysctl_max_threads(struct ctl_table *table, int write,
3289                        void *buffer, size_t *lenp, loff_t *ppos)
3290 {
3291         struct ctl_table t;
3292         int ret;
3293         int threads = max_threads;
3294         int min = 1;
3295         int max = MAX_THREADS;
3296
3297         t = *table;
3298         t.data = &threads;
3299         t.extra1 = &min;
3300         t.extra2 = &max;
3301
3302         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3303         if (ret || !write)
3304                 return ret;
3305
3306         max_threads = threads;
3307
3308         return 0;
3309 }