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