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/tty.h>
  79 #include <linux/fs_struct.h>
  80 #include <linux/magic.h>
  81 #include <linux/perf_event.h>
  82 #include <linux/posix-timers.h>
  83 #include <linux/user-return-notifier.h>
  84 #include <linux/oom.h>
  85 #include <linux/khugepaged.h>
  86 #include <linux/signalfd.h>
  87 #include <linux/uprobes.h>
  88 #include <linux/aio.h>
  89 #include <linux/compiler.h>
  90 #include <linux/sysctl.h>
  91 #include <linux/kcov.h>
  92 #include <linux/livepatch.h>
  93 #include <linux/thread_info.h>
  94 #include <linux/stackleak.h>
  95 #include <linux/kasan.h>
  96 #include <linux/scs.h>
  97 #include <linux/io_uring.h>
  98 #include <linux/bpf.h>
  99 #include <linux/stackprotector.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         free_task_struct(tsk);
 563 }
 564 EXPORT_SYMBOL(free_task);
 565
 566 static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
 567 {
 568         struct file *exe_file;
 569
 570         exe_file = get_mm_exe_file(oldmm);
 571         RCU_INIT_POINTER(mm->exe_file, exe_file);
 572         /*
 573          * We depend on the oldmm having properly denied write access to the
 574          * exe_file already.
 575          */
 576         if (exe_file && deny_write_access(exe_file))
 577                 pr_warn_once("deny_write_access() failed in %s\n", __func__);
 578 }
 579
 580 #ifdef CONFIG_MMU
 581 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 582                                         struct mm_struct *oldmm)
 583 {
 584         struct vm_area_struct *mpnt, *tmp;
 585         int retval;
 586         unsigned long charge = 0;
 587         LIST_HEAD(uf);
 588         MA_STATE(old_mas, &oldmm->mm_mt, 0, 0);
 589         MA_STATE(mas, &mm->mm_mt, 0, 0);
 590
 591         uprobe_start_dup_mmap();
 592         if (mmap_write_lock_killable(oldmm)) {
 593                 retval = -EINTR;
 594                 goto fail_uprobe_end;
 595         }
 596         flush_cache_dup_mm(oldmm);
 597         uprobe_dup_mmap(oldmm, mm);
 598         /*
 599          * Not linked in yet - no deadlock potential:
 600          */
 601         mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
 602
 603         /* No ordering required: file already has been exposed. */
 604         dup_mm_exe_file(mm, oldmm);
 605
 606         mm->total_vm = oldmm->total_vm;
 607         mm->data_vm = oldmm->data_vm;
 608         mm->exec_vm = oldmm->exec_vm;
 609         mm->stack_vm = oldmm->stack_vm;
 610
 611         retval = ksm_fork(mm, oldmm);
 612         if (retval)
 613                 goto out;
 614         khugepaged_fork(mm, oldmm);
 615
 616         retval = mas_expected_entries(&mas, oldmm->map_count);
 617         if (retval)
 618                 goto out;
 619
 620         mas_for_each(&old_mas, mpnt, ULONG_MAX) {
 621                 struct file *file;
 622
 623                 if (mpnt->vm_flags & VM_DONTCOPY) {
 624                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 625                         continue;
 626                 }
 627                 charge = 0;
 628                 /*
 629                  * Don't duplicate many vmas if we've been oom-killed (for
 630                  * example)
 631                  */
 632                 if (fatal_signal_pending(current)) {
 633                         retval = -EINTR;
 634                         goto loop_out;
 635                 }
 636                 if (mpnt->vm_flags & VM_ACCOUNT) {
 637                         unsigned long len = vma_pages(mpnt);
 638
 639                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 640                                 goto fail_nomem;
 641                         charge = len;
 642                 }
 643                 tmp = vm_area_dup(mpnt);
 644                 if (!tmp)
 645                         goto fail_nomem;
 646                 retval = vma_dup_policy(mpnt, tmp);
 647                 if (retval)
 648                         goto fail_nomem_policy;
 649                 tmp->vm_mm = mm;
 650                 retval = dup_userfaultfd(tmp, &uf);
 651                 if (retval)
 652                         goto fail_nomem_anon_vma_fork;
 653                 if (tmp->vm_flags & VM_WIPEONFORK) {
 654                         /*
 655                          * VM_WIPEONFORK gets a clean slate in the child.
 656                          * Don't prepare anon_vma until fault since we don't
 657                          * copy page for current vma.
 658                          */
 659                         tmp->anon_vma = NULL;
 660                 } else if (anon_vma_fork(tmp, mpnt))
 661                         goto fail_nomem_anon_vma_fork;
 662                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 663                 file = tmp->vm_file;
 664                 if (file) {
 665                         struct address_space *mapping = file->f_mapping;
 666
 667                         get_file(file);
 668                         i_mmap_lock_write(mapping);
 669                         if (tmp->vm_flags & VM_SHARED)
 670                                 mapping_allow_writable(mapping);
 671                         flush_dcache_mmap_lock(mapping);
 672                         /* insert tmp into the share list, just after mpnt */
 673                         vma_interval_tree_insert_after(tmp, mpnt,
 674                                         &mapping->i_mmap);
 675                         flush_dcache_mmap_unlock(mapping);
 676                         i_mmap_unlock_write(mapping);
 677                 }
 678
 679                 /*
 680                  * Copy/update hugetlb private vma information.
 681                  */
 682                 if (is_vm_hugetlb_page(tmp))
 683                         hugetlb_dup_vma_private(tmp);
 684
 685                 /* Link the vma into the MT */
 686                 mas.index = tmp->vm_start;
 687                 mas.last = tmp->vm_end - 1;
 688                 mas_store(&mas, tmp);
 689                 if (mas_is_err(&mas))
 690                         goto fail_nomem_mas_store;
 691
 692                 mm->map_count++;
 693                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 694                         retval = copy_page_range(tmp, mpnt);
 695
 696                 if (tmp->vm_ops && tmp->vm_ops->open)
 697                         tmp->vm_ops->open(tmp);
 698
 699                 if (retval)
 700                         goto loop_out;
 701         }
 702         /* a new mm has just been created */
 703         retval = arch_dup_mmap(oldmm, mm);
 704 loop_out:
 705         mas_destroy(&mas);
 706 out:
 707         mmap_write_unlock(mm);
 708         flush_tlb_mm(oldmm);
 709         mmap_write_unlock(oldmm);
 710         dup_userfaultfd_complete(&uf);
 711 fail_uprobe_end:
 712         uprobe_end_dup_mmap();
 713         return retval;
 714
 715 fail_nomem_mas_store:
 716         unlink_anon_vmas(tmp);
 717 fail_nomem_anon_vma_fork:
 718         mpol_put(vma_policy(tmp));
 719 fail_nomem_policy:
 720         vm_area_free(tmp);
 721 fail_nomem:
 722         retval = -ENOMEM;
 723         vm_unacct_memory(charge);
 724         goto loop_out;
 725 }
 726
 727 static inline int mm_alloc_pgd(struct mm_struct *mm)
 728 {
 729         mm->pgd = pgd_alloc(mm);
 730         if (unlikely(!mm->pgd))
 731                 return -ENOMEM;
 732         return 0;
 733 }
 734
 735 static inline void mm_free_pgd(struct mm_struct *mm)
 736 {
 737         pgd_free(mm, mm->pgd);
 738 }
 739 #else
 740 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 741 {
 742         mmap_write_lock(oldmm);
 743         dup_mm_exe_file(mm, oldmm);
 744         mmap_write_unlock(oldmm);
 745         return 0;
 746 }
 747 #define mm_alloc_pgd(mm)        (0)
 748 #define mm_free_pgd(mm)
 749 #endif /* CONFIG_MMU */
 750
 751 static void check_mm(struct mm_struct *mm)
 752 {
 753         int i;
 754
 755         BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
 756                          "Please make sure 'struct resident_page_types[]' is updated as well");
 757
 758         for (i = 0; i < NR_MM_COUNTERS; i++) {
 759                 long x = percpu_counter_sum(&mm->rss_stat[i]);
 760
 761                 if (likely(!x))
 762                         continue;
 763
 764                 /* Making sure this is not due to race with CPU offlining. */
 765                 x = percpu_counter_sum_all(&mm->rss_stat[i]);
 766                 if (unlikely(x))
 767                         pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
 768                                  mm, resident_page_types[i], x);
 769         }
 770
 771         if (mm_pgtables_bytes(mm))
 772                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 773                                 mm_pgtables_bytes(mm));
 774
 775 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 776         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 777 #endif
 778 }
 779
 780 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 781 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 782
 783 /*
 784  * Called when the last reference to the mm
 785  * is dropped: either by a lazy thread or by
 786  * mmput. Free the page directory and the mm.
 787  */
 788 void __mmdrop(struct mm_struct *mm)
 789 {
 790         int i;
 791
 792         BUG_ON(mm == &init_mm);
 793         WARN_ON_ONCE(mm == current->mm);
 794         WARN_ON_ONCE(mm == current->active_mm);
 795         mm_free_pgd(mm);
 796         destroy_context(mm);
 797         mmu_notifier_subscriptions_destroy(mm);
 798         check_mm(mm);
 799         put_user_ns(mm->user_ns);
 800         mm_pasid_drop(mm);
 801
 802         for (i = 0; i < NR_MM_COUNTERS; i++)
 803                 percpu_counter_destroy(&mm->rss_stat[i]);
 804         free_mm(mm);
 805 }
 806 EXPORT_SYMBOL_GPL(__mmdrop);
 807
 808 static void mmdrop_async_fn(struct work_struct *work)
 809 {
 810         struct mm_struct *mm;
 811
 812         mm = container_of(work, struct mm_struct, async_put_work);
 813         __mmdrop(mm);
 814 }
 815
 816 static void mmdrop_async(struct mm_struct *mm)
 817 {
 818         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 819                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 820                 schedule_work(&mm->async_put_work);
 821         }
 822 }
 823
 824 static inline void free_signal_struct(struct signal_struct *sig)
 825 {
 826         taskstats_tgid_free(sig);
 827         sched_autogroup_exit(sig);
 828         /*
 829          * __mmdrop is not safe to call from softirq context on x86 due to
 830          * pgd_dtor so postpone it to the async context
 831          */
 832         if (sig->oom_mm)
 833                 mmdrop_async(sig->oom_mm);
 834         kmem_cache_free(signal_cachep, sig);
 835 }
 836
 837 static inline void put_signal_struct(struct signal_struct *sig)
 838 {
 839         if (refcount_dec_and_test(&sig->sigcnt))
 840                 free_signal_struct(sig);
 841 }
 842
 843 void __put_task_struct(struct task_struct *tsk)
 844 {
 845         WARN_ON(!tsk->exit_state);
 846         WARN_ON(refcount_read(&tsk->usage));
 847         WARN_ON(tsk == current);
 848
 849         io_uring_free(tsk);
 850         cgroup_free(tsk);
 851         task_numa_free(tsk, true);
 852         security_task_free(tsk);
 853         bpf_task_storage_free(tsk);
 854         exit_creds(tsk);
 855         delayacct_tsk_free(tsk);
 856         put_signal_struct(tsk->signal);
 857         sched_core_free(tsk);
 858         free_task(tsk);
 859 }
 860 EXPORT_SYMBOL_GPL(__put_task_struct);
 861
 862 void __init __weak arch_task_cache_init(void) { }
 863
 864 /*
 865  * set_max_threads
 866  */
 867 static void set_max_threads(unsigned int max_threads_suggested)
 868 {
 869         u64 threads;
 870         unsigned long nr_pages = totalram_pages();
 871
 872         /*
 873          * The number of threads shall be limited such that the thread
 874          * structures may only consume a small part of the available memory.
 875          */
 876         if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
 877                 threads = MAX_THREADS;
 878         else
 879                 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
 880                                     (u64) THREAD_SIZE * 8UL);
 881
 882         if (threads > max_threads_suggested)
 883                 threads = max_threads_suggested;
 884
 885         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 886 }
 887
 888 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 889 /* Initialized by the architecture: */
 890 int arch_task_struct_size __read_mostly;
 891 #endif
 892
 893 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 894 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 895 {
 896         /* Fetch thread_struct whitelist for the architecture. */
 897         arch_thread_struct_whitelist(offset, size);
 898
 899         /*
 900          * Handle zero-sized whitelist or empty thread_struct, otherwise
 901          * adjust offset to position of thread_struct in task_struct.
 902          */
 903         if (unlikely(*size == 0))
 904                 *offset = 0;
 905         else
 906                 *offset += offsetof(struct task_struct, thread);
 907 }
 908 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
 909
 910 void __init fork_init(void)
 911 {
 912         int i;
 913 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 914 #ifndef ARCH_MIN_TASKALIGN
 915 #define ARCH_MIN_TASKALIGN      0
 916 #endif
 917         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 918         unsigned long useroffset, usersize;
 919
 920         /* create a slab on which task_structs can be allocated */
 921         task_struct_whitelist(&useroffset, &usersize);
 922         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 923                         arch_task_struct_size, align,
 924                         SLAB_PANIC|SLAB_ACCOUNT,
 925                         useroffset, usersize, NULL);
 926 #endif
 927
 928         /* do the arch specific task caches init */
 929         arch_task_cache_init();
 930
 931         set_max_threads(MAX_THREADS);
 932
 933         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 934         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 935         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 936                 init_task.signal->rlim[RLIMIT_NPROC];
 937
 938         for (i = 0; i < UCOUNT_COUNTS; i++)
 939                 init_user_ns.ucount_max[i] = max_threads/2;
 940
 941         set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC,      RLIM_INFINITY);
 942         set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE,   RLIM_INFINITY);
 943         set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
 944         set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK,    RLIM_INFINITY);
 945
 946 #ifdef CONFIG_VMAP_STACK
 947         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 948                           NULL, free_vm_stack_cache);
 949 #endif
 950
 951         scs_init();
 952
 953         lockdep_init_task(&init_task);
 954         uprobes_init();
 955 }
 956
 957 int __weak arch_dup_task_struct(struct task_struct *dst,
 958                                                struct task_struct *src)
 959 {
 960         *dst = *src;
 961         return 0;
 962 }
 963
 964 void set_task_stack_end_magic(struct task_struct *tsk)
 965 {
 966         unsigned long *stackend;
 967
 968         stackend = end_of_stack(tsk);
 969         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 970 }
 971
 972 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 973 {
 974         struct task_struct *tsk;
 975         int err;
 976
 977         if (node == NUMA_NO_NODE)
 978                 node = tsk_fork_get_node(orig);
 979         tsk = alloc_task_struct_node(node);
 980         if (!tsk)
 981                 return NULL;
 982
 983         err = arch_dup_task_struct(tsk, orig);
 984         if (err)
 985                 goto free_tsk;
 986
 987         err = alloc_thread_stack_node(tsk, node);
 988         if (err)
 989                 goto free_tsk;
 990
 991 #ifdef CONFIG_THREAD_INFO_IN_TASK
 992         refcount_set(&tsk->stack_refcount, 1);
 993 #endif
 994         account_kernel_stack(tsk, 1);
 995
 996         err = scs_prepare(tsk, node);
 997         if (err)
 998                 goto free_stack;
 999
1000 #ifdef CONFIG_SECCOMP
1001         /*
1002          * We must handle setting up seccomp filters once we're under
1003          * the sighand lock in case orig has changed between now and
1004          * then. Until then, filter must be NULL to avoid messing up
1005          * the usage counts on the error path calling free_task.
1006          */
1007         tsk->seccomp.filter = NULL;
1008 #endif
1009
1010         setup_thread_stack(tsk, orig);
1011         clear_user_return_notifier(tsk);
1012         clear_tsk_need_resched(tsk);
1013         set_task_stack_end_magic(tsk);
1014         clear_syscall_work_syscall_user_dispatch(tsk);
1015
1016 #ifdef CONFIG_STACKPROTECTOR
1017         tsk->stack_canary = get_random_canary();
1018 #endif
1019         if (orig->cpus_ptr == &orig->cpus_mask)
1020                 tsk->cpus_ptr = &tsk->cpus_mask;
1021         dup_user_cpus_ptr(tsk, orig, node);
1022
1023         /*
1024          * One for the user space visible state that goes away when reaped.
1025          * One for the scheduler.
1026          */
1027         refcount_set(&tsk->rcu_users, 2);
1028         /* One for the rcu users */
1029         refcount_set(&tsk->usage, 1);
1030 #ifdef CONFIG_BLK_DEV_IO_TRACE
1031         tsk->btrace_seq = 0;
1032 #endif
1033         tsk->splice_pipe = NULL;
1034         tsk->task_frag.page = NULL;
1035         tsk->wake_q.next = NULL;
1036         tsk->worker_private = NULL;
1037
1038         kcov_task_init(tsk);
1039         kmsan_task_create(tsk);
1040         kmap_local_fork(tsk);
1041
1042 #ifdef CONFIG_FAULT_INJECTION
1043         tsk->fail_nth = 0;
1044 #endif
1045
1046 #ifdef CONFIG_BLK_CGROUP
1047         tsk->throttle_disk = NULL;
1048         tsk->use_memdelay = 0;
1049 #endif
1050
1051 #ifdef CONFIG_IOMMU_SVA
1052         tsk->pasid_activated = 0;
1053 #endif
1054
1055 #ifdef CONFIG_MEMCG
1056         tsk->active_memcg = NULL;
1057 #endif
1058
1059 #ifdef CONFIG_CPU_SUP_INTEL
1060         tsk->reported_split_lock = 0;
1061 #endif
1062
1063 #ifdef CONFIG_SCHED_MM_CID
1064         tsk->mm_cid = -1;
1065         tsk->mm_cid_active = 0;
1066 #endif
1067         return tsk;
1068
1069 free_stack:
1070         exit_task_stack_account(tsk);
1071         free_thread_stack(tsk);
1072 free_tsk:
1073         free_task_struct(tsk);
1074         return NULL;
1075 }
1076
1077 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1078
1079 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1080
1081 static int __init coredump_filter_setup(char *s)
1082 {
1083         default_dump_filter =
1084                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1085                 MMF_DUMP_FILTER_MASK;
1086         return 1;
1087 }
1088
1089 __setup("coredump_filter=", coredump_filter_setup);
1090
1091 #include <linux/init_task.h>
1092
1093 static void mm_init_aio(struct mm_struct *mm)
1094 {
1095 #ifdef CONFIG_AIO
1096         spin_lock_init(&mm->ioctx_lock);
1097         mm->ioctx_table = NULL;
1098 #endif
1099 }
1100
1101 static __always_inline void mm_clear_owner(struct mm_struct *mm,
1102                                            struct task_struct *p)
1103 {
1104 #ifdef CONFIG_MEMCG
1105         if (mm->owner == p)
1106                 WRITE_ONCE(mm->owner, NULL);
1107 #endif
1108 }
1109
1110 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1111 {
1112 #ifdef CONFIG_MEMCG
1113         mm->owner = p;
1114 #endif
1115 }
1116
1117 static void mm_init_uprobes_state(struct mm_struct *mm)
1118 {
1119 #ifdef CONFIG_UPROBES
1120         mm->uprobes_state.xol_area = NULL;
1121 #endif
1122 }
1123
1124 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1125         struct user_namespace *user_ns)
1126 {
1127         int i;
1128
1129         mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1130         mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1131         atomic_set(&mm->mm_users, 1);
1132         atomic_set(&mm->mm_count, 1);
1133         seqcount_init(&mm->write_protect_seq);
1134         mmap_init_lock(mm);
1135         INIT_LIST_HEAD(&mm->mmlist);
1136         mm_pgtables_bytes_init(mm);
1137         mm->map_count = 0;
1138         mm->locked_vm = 0;
1139         atomic64_set(&mm->pinned_vm, 0);
1140         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1141         spin_lock_init(&mm->page_table_lock);
1142         spin_lock_init(&mm->arg_lock);
1143         mm_init_cpumask(mm);
1144         mm_init_aio(mm);
1145         mm_init_owner(mm, p);
1146         mm_pasid_init(mm);
1147         RCU_INIT_POINTER(mm->exe_file, NULL);
1148         mmu_notifier_subscriptions_init(mm);
1149         init_tlb_flush_pending(mm);
1150 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1151         mm->pmd_huge_pte = NULL;
1152 #endif
1153         mm_init_uprobes_state(mm);
1154         hugetlb_count_init(mm);
1155
1156         if (current->mm) {
1157                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1158                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1159         } else {
1160                 mm->flags = default_dump_filter;
1161                 mm->def_flags = 0;
1162         }
1163
1164         if (mm_alloc_pgd(mm))
1165                 goto fail_nopgd;
1166
1167         if (init_new_context(p, mm))
1168                 goto fail_nocontext;
1169
1170         for (i = 0; i < NR_MM_COUNTERS; i++)
1171                 if (percpu_counter_init(&mm->rss_stat[i], 0, GFP_KERNEL_ACCOUNT))
1172                         goto fail_pcpu;
1173
1174         mm->user_ns = get_user_ns(user_ns);
1175         lru_gen_init_mm(mm);
1176         mm_init_cid(mm);
1177         return mm;
1178
1179 fail_pcpu:
1180         while (i > 0)
1181                 percpu_counter_destroy(&mm->rss_stat[--i]);
1182 fail_nocontext:
1183         mm_free_pgd(mm);
1184 fail_nopgd:
1185         free_mm(mm);
1186         return NULL;
1187 }
1188
1189 /*
1190  * Allocate and initialize an mm_struct.
1191  */
1192 struct mm_struct *mm_alloc(void)
1193 {
1194         struct mm_struct *mm;
1195
1196         mm = allocate_mm();
1197         if (!mm)
1198                 return NULL;
1199
1200         memset(mm, 0, sizeof(*mm));
1201         return mm_init(mm, current, current_user_ns());
1202 }
1203
1204 static inline void __mmput(struct mm_struct *mm)
1205 {
1206         VM_BUG_ON(atomic_read(&mm->mm_users));
1207
1208         uprobe_clear_state(mm);
1209         exit_aio(mm);
1210         ksm_exit(mm);
1211         khugepaged_exit(mm); /* must run before exit_mmap */
1212         exit_mmap(mm);
1213         mm_put_huge_zero_page(mm);
1214         set_mm_exe_file(mm, NULL);
1215         if (!list_empty(&mm->mmlist)) {
1216                 spin_lock(&mmlist_lock);
1217                 list_del(&mm->mmlist);
1218                 spin_unlock(&mmlist_lock);
1219         }
1220         if (mm->binfmt)
1221                 module_put(mm->binfmt->module);
1222         lru_gen_del_mm(mm);
1223         mmdrop(mm);
1224 }
1225
1226 /*
1227  * Decrement the use count and release all resources for an mm.
1228  */
1229 void mmput(struct mm_struct *mm)
1230 {
1231         might_sleep();
1232
1233         if (atomic_dec_and_test(&mm->mm_users))
1234                 __mmput(mm);
1235 }
1236 EXPORT_SYMBOL_GPL(mmput);
1237
1238 #ifdef CONFIG_MMU
1239 static void mmput_async_fn(struct work_struct *work)
1240 {
1241         struct mm_struct *mm = container_of(work, struct mm_struct,
1242                                             async_put_work);
1243
1244         __mmput(mm);
1245 }
1246
1247 void mmput_async(struct mm_struct *mm)
1248 {
1249         if (atomic_dec_and_test(&mm->mm_users)) {
1250                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1251                 schedule_work(&mm->async_put_work);
1252         }
1253 }
1254 EXPORT_SYMBOL_GPL(mmput_async);
1255 #endif
1256
1257 /**
1258  * set_mm_exe_file - change a reference to the mm's executable file
1259  *
1260  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1261  *
1262  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1263  * invocations: in mmput() nobody alive left, in execve task is single
1264  * threaded.
1265  *
1266  * Can only fail if new_exe_file != NULL.
1267  */
1268 int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1269 {
1270         struct file *old_exe_file;
1271
1272         /*
1273          * It is safe to dereference the exe_file without RCU as
1274          * this function is only called if nobody else can access
1275          * this mm -- see comment above for justification.
1276          */
1277         old_exe_file = rcu_dereference_raw(mm->exe_file);
1278
1279         if (new_exe_file) {
1280                 /*
1281                  * We expect the caller (i.e., sys_execve) to already denied
1282                  * write access, so this is unlikely to fail.
1283                  */
1284                 if (unlikely(deny_write_access(new_exe_file)))
1285                         return -EACCES;
1286                 get_file(new_exe_file);
1287         }
1288         rcu_assign_pointer(mm->exe_file, new_exe_file);
1289         if (old_exe_file) {
1290                 allow_write_access(old_exe_file);
1291                 fput(old_exe_file);
1292         }
1293         return 0;
1294 }
1295
1296 /**
1297  * replace_mm_exe_file - replace a reference to the mm's executable file
1298  *
1299  * This changes mm's executable file (shown as symlink /proc/[pid]/exe),
1300  * dealing with concurrent invocation and without grabbing the mmap lock in
1301  * write mode.
1302  *
1303  * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1304  */
1305 int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1306 {
1307         struct vm_area_struct *vma;
1308         struct file *old_exe_file;
1309         int ret = 0;
1310
1311         /* Forbid mm->exe_file change if old file still mapped. */
1312         old_exe_file = get_mm_exe_file(mm);
1313         if (old_exe_file) {
1314                 VMA_ITERATOR(vmi, mm, 0);
1315                 mmap_read_lock(mm);
1316                 for_each_vma(vmi, vma) {
1317                         if (!vma->vm_file)
1318                                 continue;
1319                         if (path_equal(&vma->vm_file->f_path,
1320                                        &old_exe_file->f_path)) {
1321                                 ret = -EBUSY;
1322                                 break;
1323                         }
1324                 }
1325                 mmap_read_unlock(mm);
1326                 fput(old_exe_file);
1327                 if (ret)
1328                         return ret;
1329         }
1330
1331         /* set the new file, lockless */
1332         ret = deny_write_access(new_exe_file);
1333         if (ret)
1334                 return -EACCES;
1335         get_file(new_exe_file);
1336
1337         old_exe_file = xchg(&mm->exe_file, new_exe_file);
1338         if (old_exe_file) {
1339                 /*
1340                  * Don't race with dup_mmap() getting the file and disallowing
1341                  * write access while someone might open the file writable.
1342                  */
1343                 mmap_read_lock(mm);
1344                 allow_write_access(old_exe_file);
1345                 fput(old_exe_file);
1346                 mmap_read_unlock(mm);
1347         }
1348         return 0;
1349 }
1350
1351 /**
1352  * get_mm_exe_file - acquire a reference to the mm's executable file
1353  *
1354  * Returns %NULL if mm has no associated executable file.
1355  * User must release file via fput().
1356  */
1357 struct file *get_mm_exe_file(struct mm_struct *mm)
1358 {
1359         struct file *exe_file;
1360
1361         rcu_read_lock();
1362         exe_file = rcu_dereference(mm->exe_file);
1363         if (exe_file && !get_file_rcu(exe_file))
1364                 exe_file = NULL;
1365         rcu_read_unlock();
1366         return exe_file;
1367 }
1368
1369 /**
1370  * get_task_exe_file - acquire a reference to the task's executable file
1371  *
1372  * Returns %NULL if task's mm (if any) has no associated executable file or
1373  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1374  * User must release file via fput().
1375  */
1376 struct file *get_task_exe_file(struct task_struct *task)
1377 {
1378         struct file *exe_file = NULL;
1379         struct mm_struct *mm;
1380
1381         task_lock(task);
1382         mm = task->mm;
1383         if (mm) {
1384                 if (!(task->flags & PF_KTHREAD))
1385                         exe_file = get_mm_exe_file(mm);
1386         }
1387         task_unlock(task);
1388         return exe_file;
1389 }
1390
1391 /**
1392  * get_task_mm - acquire a reference to the task's mm
1393  *
1394  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1395  * this kernel workthread has transiently adopted a user mm with use_mm,
1396  * to do its AIO) is not set and if so returns a reference to it, after
1397  * bumping up the use count.  User must release the mm via mmput()
1398  * after use.  Typically used by /proc and ptrace.
1399  */
1400 struct mm_struct *get_task_mm(struct task_struct *task)
1401 {
1402         struct mm_struct *mm;
1403
1404         task_lock(task);
1405         mm = task->mm;
1406         if (mm) {
1407                 if (task->flags & PF_KTHREAD)
1408                         mm = NULL;
1409                 else
1410                         mmget(mm);
1411         }
1412         task_unlock(task);
1413         return mm;
1414 }
1415 EXPORT_SYMBOL_GPL(get_task_mm);
1416
1417 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1418 {
1419         struct mm_struct *mm;
1420         int err;
1421
1422         err =  down_read_killable(&task->signal->exec_update_lock);
1423         if (err)
1424                 return ERR_PTR(err);
1425
1426         mm = get_task_mm(task);
1427         if (mm && mm != current->mm &&
1428                         !ptrace_may_access(task, mode)) {
1429                 mmput(mm);
1430                 mm = ERR_PTR(-EACCES);
1431         }
1432         up_read(&task->signal->exec_update_lock);
1433
1434         return mm;
1435 }
1436
1437 static void complete_vfork_done(struct task_struct *tsk)
1438 {
1439         struct completion *vfork;
1440
1441         task_lock(tsk);
1442         vfork = tsk->vfork_done;
1443         if (likely(vfork)) {
1444                 tsk->vfork_done = NULL;
1445                 complete(vfork);
1446         }
1447         task_unlock(tsk);
1448 }
1449
1450 static int wait_for_vfork_done(struct task_struct *child,
1451                                 struct completion *vfork)
1452 {
1453         unsigned int state = TASK_UNINTERRUPTIBLE|TASK_KILLABLE|TASK_FREEZABLE;
1454         int killed;
1455
1456         cgroup_enter_frozen();
1457         killed = wait_for_completion_state(vfork, state);
1458         cgroup_leave_frozen(false);
1459
1460         if (killed) {
1461                 task_lock(child);
1462                 child->vfork_done = NULL;
1463                 task_unlock(child);
1464         }
1465
1466         put_task_struct(child);
1467         return killed;
1468 }
1469
1470 /* Please note the differences between mmput and mm_release.
1471  * mmput is called whenever we stop holding onto a mm_struct,
1472  * error success whatever.
1473  *
1474  * mm_release is called after a mm_struct has been removed
1475  * from the current process.
1476  *
1477  * This difference is important for error handling, when we
1478  * only half set up a mm_struct for a new process and need to restore
1479  * the old one.  Because we mmput the new mm_struct before
1480  * restoring the old one. . .
1481  * Eric Biederman 10 January 1998
1482  */
1483 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1484 {
1485         uprobe_free_utask(tsk);
1486
1487         /* Get rid of any cached register state */
1488         deactivate_mm(tsk, mm);
1489
1490         /*
1491          * Signal userspace if we're not exiting with a core dump
1492          * because we want to leave the value intact for debugging
1493          * purposes.
1494          */
1495         if (tsk->clear_child_tid) {
1496                 if (atomic_read(&mm->mm_users) > 1) {
1497                         /*
1498                          * We don't check the error code - if userspace has
1499                          * not set up a proper pointer then tough luck.
1500                          */
1501                         put_user(0, tsk->clear_child_tid);
1502                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1503                                         1, NULL, NULL, 0, 0);
1504                 }
1505                 tsk->clear_child_tid = NULL;
1506         }
1507
1508         /*
1509          * All done, finally we can wake up parent and return this mm to him.
1510          * Also kthread_stop() uses this completion for synchronization.
1511          */
1512         if (tsk->vfork_done)
1513                 complete_vfork_done(tsk);
1514 }
1515
1516 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1517 {
1518         futex_exit_release(tsk);
1519         mm_release(tsk, mm);
1520 }
1521
1522 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1523 {
1524         futex_exec_release(tsk);
1525         mm_release(tsk, mm);
1526 }
1527
1528 /**
1529  * dup_mm() - duplicates an existing mm structure
1530  * @tsk: the task_struct with which the new mm will be associated.
1531  * @oldmm: the mm to duplicate.
1532  *
1533  * Allocates a new mm structure and duplicates the provided @oldmm structure
1534  * content into it.
1535  *
1536  * Return: the duplicated mm or NULL on failure.
1537  */
1538 static struct mm_struct *dup_mm(struct task_struct *tsk,
1539                                 struct mm_struct *oldmm)
1540 {
1541         struct mm_struct *mm;
1542         int err;
1543
1544         mm = allocate_mm();
1545         if (!mm)
1546                 goto fail_nomem;
1547
1548         memcpy(mm, oldmm, sizeof(*mm));
1549
1550         if (!mm_init(mm, tsk, mm->user_ns))
1551                 goto fail_nomem;
1552
1553         err = dup_mmap(mm, oldmm);
1554         if (err)
1555                 goto free_pt;
1556
1557         mm->hiwater_rss = get_mm_rss(mm);
1558         mm->hiwater_vm = mm->total_vm;
1559
1560         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1561                 goto free_pt;
1562
1563         return mm;
1564
1565 free_pt:
1566         /* don't put binfmt in mmput, we haven't got module yet */
1567         mm->binfmt = NULL;
1568         mm_init_owner(mm, NULL);
1569         mmput(mm);
1570
1571 fail_nomem:
1572         return NULL;
1573 }
1574
1575 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1576 {
1577         struct mm_struct *mm, *oldmm;
1578
1579         tsk->min_flt = tsk->maj_flt = 0;
1580         tsk->nvcsw = tsk->nivcsw = 0;
1581 #ifdef CONFIG_DETECT_HUNG_TASK
1582         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1583         tsk->last_switch_time = 0;
1584 #endif
1585
1586         tsk->mm = NULL;
1587         tsk->active_mm = NULL;
1588
1589         /*
1590          * Are we cloning a kernel thread?
1591          *
1592          * We need to steal a active VM for that..
1593          */
1594         oldmm = current->mm;
1595         if (!oldmm)
1596                 return 0;
1597
1598         if (clone_flags & CLONE_VM) {
1599                 mmget(oldmm);
1600                 mm = oldmm;
1601         } else {
1602                 mm = dup_mm(tsk, current->mm);
1603                 if (!mm)
1604                         return -ENOMEM;
1605         }
1606
1607         tsk->mm = mm;
1608         tsk->active_mm = mm;
1609         sched_mm_cid_fork(tsk);
1610         return 0;
1611 }
1612
1613 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1614 {
1615         struct fs_struct *fs = current->fs;
1616         if (clone_flags & CLONE_FS) {
1617                 /* tsk->fs is already what we want */
1618                 spin_lock(&fs->lock);
1619                 if (fs->in_exec) {
1620                         spin_unlock(&fs->lock);
1621                         return -EAGAIN;
1622                 }
1623                 fs->users++;
1624                 spin_unlock(&fs->lock);
1625                 return 0;
1626         }
1627         tsk->fs = copy_fs_struct(fs);
1628         if (!tsk->fs)
1629                 return -ENOMEM;
1630         return 0;
1631 }
1632
1633 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1634 {
1635         struct files_struct *oldf, *newf;
1636         int error = 0;
1637
1638         /*
1639          * A background process may not have any files ...
1640          */
1641         oldf = current->files;
1642         if (!oldf)
1643                 goto out;
1644
1645         if (clone_flags & CLONE_FILES) {
1646                 atomic_inc(&oldf->count);
1647                 goto out;
1648         }
1649
1650         newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1651         if (!newf)
1652                 goto out;
1653
1654         tsk->files = newf;
1655         error = 0;
1656 out:
1657         return error;
1658 }
1659
1660 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1661 {
1662         struct sighand_struct *sig;
1663
1664         if (clone_flags & CLONE_SIGHAND) {
1665                 refcount_inc(&current->sighand->count);
1666                 return 0;
1667         }
1668         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1669         RCU_INIT_POINTER(tsk->sighand, sig);
1670         if (!sig)
1671                 return -ENOMEM;
1672
1673         refcount_set(&sig->count, 1);
1674         spin_lock_irq(&current->sighand->siglock);
1675         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1676         spin_unlock_irq(&current->sighand->siglock);
1677
1678         /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1679         if (clone_flags & CLONE_CLEAR_SIGHAND)
1680                 flush_signal_handlers(tsk, 0);
1681
1682         return 0;
1683 }
1684
1685 void __cleanup_sighand(struct sighand_struct *sighand)
1686 {
1687         if (refcount_dec_and_test(&sighand->count)) {
1688                 signalfd_cleanup(sighand);
1689                 /*
1690                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1691                  * without an RCU grace period, see __lock_task_sighand().
1692                  */
1693                 kmem_cache_free(sighand_cachep, sighand);
1694         }
1695 }
1696
1697 /*
1698  * Initialize POSIX timer handling for a thread group.
1699  */
1700 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1701 {
1702         struct posix_cputimers *pct = &sig->posix_cputimers;
1703         unsigned long cpu_limit;
1704
1705         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1706         posix_cputimers_group_init(pct, cpu_limit);
1707 }
1708
1709 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1710 {
1711         struct signal_struct *sig;
1712
1713         if (clone_flags & CLONE_THREAD)
1714                 return 0;
1715
1716         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1717         tsk->signal = sig;
1718         if (!sig)
1719                 return -ENOMEM;
1720
1721         sig->nr_threads = 1;
1722         sig->quick_threads = 1;
1723         atomic_set(&sig->live, 1);
1724         refcount_set(&sig->sigcnt, 1);
1725
1726         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1727         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1728         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1729
1730         init_waitqueue_head(&sig->wait_chldexit);
1731         sig->curr_target = tsk;
1732         init_sigpending(&sig->shared_pending);
1733         INIT_HLIST_HEAD(&sig->multiprocess);
1734         seqlock_init(&sig->stats_lock);
1735         prev_cputime_init(&sig->prev_cputime);
1736
1737 #ifdef CONFIG_POSIX_TIMERS
1738         INIT_LIST_HEAD(&sig->posix_timers);
1739         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1740         sig->real_timer.function = it_real_fn;
1741 #endif
1742
1743         task_lock(current->group_leader);
1744         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1745         task_unlock(current->group_leader);
1746
1747         posix_cpu_timers_init_group(sig);
1748
1749         tty_audit_fork(sig);
1750         sched_autogroup_fork(sig);
1751
1752         sig->oom_score_adj = current->signal->oom_score_adj;
1753         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1754
1755         mutex_init(&sig->cred_guard_mutex);
1756         init_rwsem(&sig->exec_update_lock);
1757
1758         return 0;
1759 }
1760
1761 static void copy_seccomp(struct task_struct *p)
1762 {
1763 #ifdef CONFIG_SECCOMP
1764         /*
1765          * Must be called with sighand->lock held, which is common to
1766          * all threads in the group. Holding cred_guard_mutex is not
1767          * needed because this new task is not yet running and cannot
1768          * be racing exec.
1769          */
1770         assert_spin_locked(&current->sighand->siglock);
1771
1772         /* Ref-count the new filter user, and assign it. */
1773         get_seccomp_filter(current);
1774         p->seccomp = current->seccomp;
1775
1776         /*
1777          * Explicitly enable no_new_privs here in case it got set
1778          * between the task_struct being duplicated and holding the
1779          * sighand lock. The seccomp state and nnp must be in sync.
1780          */
1781         if (task_no_new_privs(current))
1782                 task_set_no_new_privs(p);
1783
1784         /*
1785          * If the parent gained a seccomp mode after copying thread
1786          * flags and between before we held the sighand lock, we have
1787          * to manually enable the seccomp thread flag here.
1788          */
1789         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1790                 set_task_syscall_work(p, SECCOMP);
1791 #endif
1792 }
1793
1794 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1795 {
1796         current->clear_child_tid = tidptr;
1797
1798         return task_pid_vnr(current);
1799 }
1800
1801 static void rt_mutex_init_task(struct task_struct *p)
1802 {
1803         raw_spin_lock_init(&p->pi_lock);
1804 #ifdef CONFIG_RT_MUTEXES
1805         p->pi_waiters = RB_ROOT_CACHED;
1806         p->pi_top_task = NULL;
1807         p->pi_blocked_on = NULL;
1808 #endif
1809 }
1810
1811 static inline void init_task_pid_links(struct task_struct *task)
1812 {
1813         enum pid_type type;
1814
1815         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1816                 INIT_HLIST_NODE(&task->pid_links[type]);
1817 }
1818
1819 static inline void
1820 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1821 {
1822         if (type == PIDTYPE_PID)
1823                 task->thread_pid = pid;
1824         else
1825                 task->signal->pids[type] = pid;
1826 }
1827
1828 static inline void rcu_copy_process(struct task_struct *p)
1829 {
1830 #ifdef CONFIG_PREEMPT_RCU
1831         p->rcu_read_lock_nesting = 0;
1832         p->rcu_read_unlock_special.s = 0;
1833         p->rcu_blocked_node = NULL;
1834         INIT_LIST_HEAD(&p->rcu_node_entry);
1835 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1836 #ifdef CONFIG_TASKS_RCU
1837         p->rcu_tasks_holdout = false;
1838         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1839         p->rcu_tasks_idle_cpu = -1;
1840 #endif /* #ifdef CONFIG_TASKS_RCU */
1841 #ifdef CONFIG_TASKS_TRACE_RCU
1842         p->trc_reader_nesting = 0;
1843         p->trc_reader_special.s = 0;
1844         INIT_LIST_HEAD(&p->trc_holdout_list);
1845         INIT_LIST_HEAD(&p->trc_blkd_node);
1846 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1847 }
1848
1849 struct pid *pidfd_pid(const struct file *file)
1850 {
1851         if (file->f_op == &pidfd_fops)
1852                 return file->private_data;
1853
1854         return ERR_PTR(-EBADF);
1855 }
1856
1857 static int pidfd_release(struct inode *inode, struct file *file)
1858 {
1859         struct pid *pid = file->private_data;
1860
1861         file->private_data = NULL;
1862         put_pid(pid);
1863         return 0;
1864 }
1865
1866 #ifdef CONFIG_PROC_FS
1867 /**
1868  * pidfd_show_fdinfo - print information about a pidfd
1869  * @m: proc fdinfo file
1870  * @f: file referencing a pidfd
1871  *
1872  * Pid:
1873  * This function will print the pid that a given pidfd refers to in the
1874  * pid namespace of the procfs instance.
1875  * If the pid namespace of the process is not a descendant of the pid
1876  * namespace of the procfs instance 0 will be shown as its pid. This is
1877  * similar to calling getppid() on a process whose parent is outside of
1878  * its pid namespace.
1879  *
1880  * NSpid:
1881  * If pid namespaces are supported then this function will also print
1882  * the pid of a given pidfd refers to for all descendant pid namespaces
1883  * starting from the current pid namespace of the instance, i.e. the
1884  * Pid field and the first entry in the NSpid field will be identical.
1885  * If the pid namespace of the process is not a descendant of the pid
1886  * namespace of the procfs instance 0 will be shown as its first NSpid
1887  * entry and no others will be shown.
1888  * Note that this differs from the Pid and NSpid fields in
1889  * /proc/<pid>/status where Pid and NSpid are always shown relative to
1890  * the  pid namespace of the procfs instance. The difference becomes
1891  * obvious when sending around a pidfd between pid namespaces from a
1892  * different branch of the tree, i.e. where no ancestral relation is
1893  * present between the pid namespaces:
1894  * - create two new pid namespaces ns1 and ns2 in the initial pid
1895  *   namespace (also take care to create new mount namespaces in the
1896  *   new pid namespace and mount procfs)
1897  * - create a process with a pidfd in ns1
1898  * - send pidfd from ns1 to ns2
1899  * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1900  *   have exactly one entry, which is 0
1901  */
1902 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1903 {
1904         struct pid *pid = f->private_data;
1905         struct pid_namespace *ns;
1906         pid_t nr = -1;
1907
1908         if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1909                 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1910                 nr = pid_nr_ns(pid, ns);
1911         }
1912
1913         seq_put_decimal_ll(m, "Pid:\t", nr);
1914
1915 #ifdef CONFIG_PID_NS
1916         seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1917         if (nr > 0) {
1918                 int i;
1919
1920                 /* If nr is non-zero it means that 'pid' is valid and that
1921                  * ns, i.e. the pid namespace associated with the procfs
1922                  * instance, is in the pid namespace hierarchy of pid.
1923                  * Start at one below the already printed level.
1924                  */
1925                 for (i = ns->level + 1; i <= pid->level; i++)
1926                         seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1927         }
1928 #endif
1929         seq_putc(m, '\n');
1930 }
1931 #endif
1932
1933 /*
1934  * Poll support for process exit notification.
1935  */
1936 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1937 {
1938         struct pid *pid = file->private_data;
1939         __poll_t poll_flags = 0;
1940
1941         poll_wait(file, &pid->wait_pidfd, pts);
1942
1943         /*
1944          * Inform pollers only when the whole thread group exits.
1945          * If the thread group leader exits before all other threads in the
1946          * group, then poll(2) should block, similar to the wait(2) family.
1947          */
1948         if (thread_group_exited(pid))
1949                 poll_flags = EPOLLIN | EPOLLRDNORM;
1950
1951         return poll_flags;
1952 }
1953
1954 const struct file_operations pidfd_fops = {
1955         .release = pidfd_release,
1956         .poll = pidfd_poll,
1957 #ifdef CONFIG_PROC_FS
1958         .show_fdinfo = pidfd_show_fdinfo,
1959 #endif
1960 };
1961
1962 static void __delayed_free_task(struct rcu_head *rhp)
1963 {
1964         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1965
1966         free_task(tsk);
1967 }
1968
1969 static __always_inline void delayed_free_task(struct task_struct *tsk)
1970 {
1971         if (IS_ENABLED(CONFIG_MEMCG))
1972                 call_rcu(&tsk->rcu, __delayed_free_task);
1973         else
1974                 free_task(tsk);
1975 }
1976
1977 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1978 {
1979         /* Skip if kernel thread */
1980         if (!tsk->mm)
1981                 return;
1982
1983         /* Skip if spawning a thread or using vfork */
1984         if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1985                 return;
1986
1987         /* We need to synchronize with __set_oom_adj */
1988         mutex_lock(&oom_adj_mutex);
1989         set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1990         /* Update the values in case they were changed after copy_signal */
1991         tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1992         tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1993         mutex_unlock(&oom_adj_mutex);
1994 }
1995
1996 #ifdef CONFIG_RV
1997 static void rv_task_fork(struct task_struct *p)
1998 {
1999         int i;
2000
2001         for (i = 0; i < RV_PER_TASK_MONITORS; i++)
2002                 p->rv[i].da_mon.monitoring = false;
2003 }
2004 #else
2005 #define rv_task_fork(p) do {} while (0)
2006 #endif
2007
2008 /*
2009  * This creates a new process as a copy of the old one,
2010  * but does not actually start it yet.
2011  *
2012  * It copies the registers, and all the appropriate
2013  * parts of the process environment (as per the clone
2014  * flags). The actual kick-off is left to the caller.
2015  */
2016 static __latent_entropy struct task_struct *copy_process(
2017                                         struct pid *pid,
2018                                         int trace,
2019                                         int node,
2020                                         struct kernel_clone_args *args)
2021 {
2022         int pidfd = -1, retval;
2023         struct task_struct *p;
2024         struct multiprocess_signals delayed;
2025         struct file *pidfile = NULL;
2026         const u64 clone_flags = args->flags;
2027         struct nsproxy *nsp = current->nsproxy;
2028
2029         /*
2030          * Don't allow sharing the root directory with processes in a different
2031          * namespace
2032          */
2033         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2034                 return ERR_PTR(-EINVAL);
2035
2036         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2037                 return ERR_PTR(-EINVAL);
2038
2039         /*
2040          * Thread groups must share signals as well, and detached threads
2041          * can only be started up within the thread group.
2042          */
2043         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2044                 return ERR_PTR(-EINVAL);
2045
2046         /*
2047          * Shared signal handlers imply shared VM. By way of the above,
2048          * thread groups also imply shared VM. Blocking this case allows
2049          * for various simplifications in other code.
2050          */
2051         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2052                 return ERR_PTR(-EINVAL);
2053
2054         /*
2055          * Siblings of global init remain as zombies on exit since they are
2056          * not reaped by their parent (swapper). To solve this and to avoid
2057          * multi-rooted process trees, prevent global and container-inits
2058          * from creating siblings.
2059          */
2060         if ((clone_flags & CLONE_PARENT) &&
2061                                 current->signal->flags & SIGNAL_UNKILLABLE)
2062                 return ERR_PTR(-EINVAL);
2063
2064         /*
2065          * If the new process will be in a different pid or user namespace
2066          * do not allow it to share a thread group with the forking task.
2067          */
2068         if (clone_flags & CLONE_THREAD) {
2069                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2070                     (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2071                         return ERR_PTR(-EINVAL);
2072         }
2073
2074         if (clone_flags & CLONE_PIDFD) {
2075                 /*
2076                  * - CLONE_DETACHED is blocked so that we can potentially
2077                  *   reuse it later for CLONE_PIDFD.
2078                  * - CLONE_THREAD is blocked until someone really needs it.
2079                  */
2080                 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
2081                         return ERR_PTR(-EINVAL);
2082         }
2083
2084         /*
2085          * Force any signals received before this point to be delivered
2086          * before the fork happens.  Collect up signals sent to multiple
2087          * processes that happen during the fork and delay them so that
2088          * they appear to happen after the fork.
2089          */
2090         sigemptyset(&delayed.signal);
2091         INIT_HLIST_NODE(&delayed.node);
2092
2093         spin_lock_irq(&current->sighand->siglock);
2094         if (!(clone_flags & CLONE_THREAD))
2095                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
2096         recalc_sigpending();
2097         spin_unlock_irq(&current->sighand->siglock);
2098         retval = -ERESTARTNOINTR;
2099         if (task_sigpending(current))
2100                 goto fork_out;
2101
2102         retval = -ENOMEM;
2103         p = dup_task_struct(current, node);
2104         if (!p)
2105                 goto fork_out;
2106         p->flags &= ~PF_KTHREAD;
2107         if (args->kthread)
2108                 p->flags |= PF_KTHREAD;
2109         if (args->io_thread) {
2110                 /*
2111                  * Mark us an IO worker, and block any signal that isn't
2112                  * fatal or STOP
2113                  */
2114                 p->flags |= PF_IO_WORKER;
2115                 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2116         }
2117
2118         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2119         /*
2120          * Clear TID on mm_release()?
2121          */
2122         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2123
2124         ftrace_graph_init_task(p);
2125
2126         rt_mutex_init_task(p);
2127
2128         lockdep_assert_irqs_enabled();
2129 #ifdef CONFIG_PROVE_LOCKING
2130         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2131 #endif
2132         retval = copy_creds(p, clone_flags);
2133         if (retval < 0)
2134                 goto bad_fork_free;
2135
2136         retval = -EAGAIN;
2137         if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2138                 if (p->real_cred->user != INIT_USER &&
2139                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2140                         goto bad_fork_cleanup_count;
2141         }
2142         current->flags &= ~PF_NPROC_EXCEEDED;
2143
2144         /*
2145          * If multiple threads are within copy_process(), then this check
2146          * triggers too late. This doesn't hurt, the check is only there
2147          * to stop root fork bombs.
2148          */
2149         retval = -EAGAIN;
2150         if (data_race(nr_threads >= max_threads))
2151                 goto bad_fork_cleanup_count;
2152
2153         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
2154         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2155         p->flags |= PF_FORKNOEXEC;
2156         INIT_LIST_HEAD(&p->children);
2157         INIT_LIST_HEAD(&p->sibling);
2158         rcu_copy_process(p);
2159         p->vfork_done = NULL;
2160         spin_lock_init(&p->alloc_lock);
2161
2162         init_sigpending(&p->pending);
2163
2164         p->utime = p->stime = p->gtime = 0;
2165 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2166         p->utimescaled = p->stimescaled = 0;
2167 #endif
2168         prev_cputime_init(&p->prev_cputime);
2169
2170 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2171         seqcount_init(&p->vtime.seqcount);
2172         p->vtime.starttime = 0;
2173         p->vtime.state = VTIME_INACTIVE;
2174 #endif
2175
2176 #ifdef CONFIG_IO_URING
2177         p->io_uring = NULL;
2178 #endif
2179
2180 #if defined(SPLIT_RSS_COUNTING)
2181         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2182 #endif
2183
2184         p->default_timer_slack_ns = current->timer_slack_ns;
2185
2186 #ifdef CONFIG_PSI
2187         p->psi_flags = 0;
2188 #endif
2189
2190         task_io_accounting_init(&p->ioac);
2191         acct_clear_integrals(p);
2192
2193         posix_cputimers_init(&p->posix_cputimers);
2194
2195         p->io_context = NULL;
2196         audit_set_context(p, NULL);
2197         cgroup_fork(p);
2198         if (args->kthread) {
2199                 if (!set_kthread_struct(p))
2200                         goto bad_fork_cleanup_delayacct;
2201         }
2202 #ifdef CONFIG_NUMA
2203         p->mempolicy = mpol_dup(p->mempolicy);
2204         if (IS_ERR(p->mempolicy)) {
2205                 retval = PTR_ERR(p->mempolicy);
2206                 p->mempolicy = NULL;
2207                 goto bad_fork_cleanup_delayacct;
2208         }
2209 #endif
2210 #ifdef CONFIG_CPUSETS
2211         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2212         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2213         seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2214 #endif
2215 #ifdef CONFIG_TRACE_IRQFLAGS
2216         memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2217         p->irqtrace.hardirq_disable_ip  = _THIS_IP_;
2218         p->irqtrace.softirq_enable_ip   = _THIS_IP_;
2219         p->softirqs_enabled             = 1;
2220         p->softirq_context              = 0;
2221 #endif
2222
2223         p->pagefault_disabled = 0;
2224
2225 #ifdef CONFIG_LOCKDEP
2226         lockdep_init_task(p);
2227 #endif
2228
2229 #ifdef CONFIG_DEBUG_MUTEXES
2230         p->blocked_on = NULL; /* not blocked yet */
2231 #endif
2232 #ifdef CONFIG_BCACHE
2233         p->sequential_io        = 0;
2234         p->sequential_io_avg    = 0;
2235 #endif
2236 #ifdef CONFIG_BPF_SYSCALL
2237         RCU_INIT_POINTER(p->bpf_storage, NULL);
2238         p->bpf_ctx = NULL;
2239 #endif
2240
2241         /* Perform scheduler related setup. Assign this task to a CPU. */
2242         retval = sched_fork(clone_flags, p);
2243         if (retval)
2244                 goto bad_fork_cleanup_policy;
2245
2246         retval = perf_event_init_task(p, clone_flags);
2247         if (retval)
2248                 goto bad_fork_cleanup_policy;
2249         retval = audit_alloc(p);
2250         if (retval)
2251                 goto bad_fork_cleanup_perf;
2252         /* copy all the process information */
2253         shm_init_task(p);
2254         retval = security_task_alloc(p, clone_flags);
2255         if (retval)
2256                 goto bad_fork_cleanup_audit;
2257         retval = copy_semundo(clone_flags, p);
2258         if (retval)
2259                 goto bad_fork_cleanup_security;
2260         retval = copy_files(clone_flags, p);
2261         if (retval)
2262                 goto bad_fork_cleanup_semundo;
2263         retval = copy_fs(clone_flags, p);
2264         if (retval)
2265                 goto bad_fork_cleanup_files;
2266         retval = copy_sighand(clone_flags, p);
2267         if (retval)
2268                 goto bad_fork_cleanup_fs;
2269         retval = copy_signal(clone_flags, p);
2270         if (retval)
2271                 goto bad_fork_cleanup_sighand;
2272         retval = copy_mm(clone_flags, p);
2273         if (retval)
2274                 goto bad_fork_cleanup_signal;
2275         retval = copy_namespaces(clone_flags, p);
2276         if (retval)
2277                 goto bad_fork_cleanup_mm;
2278         retval = copy_io(clone_flags, p);
2279         if (retval)
2280                 goto bad_fork_cleanup_namespaces;
2281         retval = copy_thread(p, args);
2282         if (retval)
2283                 goto bad_fork_cleanup_io;
2284
2285         stackleak_task_init(p);
2286
2287         if (pid != &init_struct_pid) {
2288                 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2289                                 args->set_tid_size);
2290                 if (IS_ERR(pid)) {
2291                         retval = PTR_ERR(pid);
2292                         goto bad_fork_cleanup_thread;
2293                 }
2294         }
2295
2296         /*
2297          * This has to happen after we've potentially unshared the file
2298          * descriptor table (so that the pidfd doesn't leak into the child
2299          * if the fd table isn't shared).
2300          */
2301         if (clone_flags & CLONE_PIDFD) {
2302                 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2303                 if (retval < 0)
2304                         goto bad_fork_free_pid;
2305
2306                 pidfd = retval;
2307
2308                 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2309                                               O_RDWR | O_CLOEXEC);
2310                 if (IS_ERR(pidfile)) {
2311                         put_unused_fd(pidfd);
2312                         retval = PTR_ERR(pidfile);
2313                         goto bad_fork_free_pid;
2314                 }
2315                 get_pid(pid);   /* held by pidfile now */
2316
2317                 retval = put_user(pidfd, args->pidfd);
2318                 if (retval)
2319                         goto bad_fork_put_pidfd;
2320         }
2321
2322 #ifdef CONFIG_BLOCK
2323         p->plug = NULL;
2324 #endif
2325         futex_init_task(p);
2326
2327         /*
2328          * sigaltstack should be cleared when sharing the same VM
2329          */
2330         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2331                 sas_ss_reset(p);
2332
2333         /*
2334          * Syscall tracing and stepping should be turned off in the
2335          * child regardless of CLONE_PTRACE.
2336          */
2337         user_disable_single_step(p);
2338         clear_task_syscall_work(p, SYSCALL_TRACE);
2339 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2340         clear_task_syscall_work(p, SYSCALL_EMU);
2341 #endif
2342         clear_tsk_latency_tracing(p);
2343
2344         /* ok, now we should be set up.. */
2345         p->pid = pid_nr(pid);
2346         if (clone_flags & CLONE_THREAD) {
2347                 p->group_leader = current->group_leader;
2348                 p->tgid = current->tgid;
2349         } else {
2350                 p->group_leader = p;
2351                 p->tgid = p->pid;
2352         }
2353
2354         p->nr_dirtied = 0;
2355         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2356         p->dirty_paused_when = 0;
2357
2358         p->pdeath_signal = 0;
2359         INIT_LIST_HEAD(&p->thread_group);
2360         p->task_works = NULL;
2361         clear_posix_cputimers_work(p);
2362
2363 #ifdef CONFIG_KRETPROBES
2364         p->kretprobe_instances.first = NULL;
2365 #endif
2366 #ifdef CONFIG_RETHOOK
2367         p->rethooks.first = NULL;
2368 #endif
2369
2370         /*
2371          * Ensure that the cgroup subsystem policies allow the new process to be
2372          * forked. It should be noted that the new process's css_set can be changed
2373          * between here and cgroup_post_fork() if an organisation operation is in
2374          * progress.
2375          */
2376         retval = cgroup_can_fork(p, args);
2377         if (retval)
2378                 goto bad_fork_put_pidfd;
2379
2380         /*
2381          * Now that the cgroups are pinned, re-clone the parent cgroup and put
2382          * the new task on the correct runqueue. All this *before* the task
2383          * becomes visible.
2384          *
2385          * This isn't part of ->can_fork() because while the re-cloning is
2386          * cgroup specific, it unconditionally needs to place the task on a
2387          * runqueue.
2388          */
2389         sched_cgroup_fork(p, args);
2390
2391         /*
2392          * From this point on we must avoid any synchronous user-space
2393          * communication until we take the tasklist-lock. In particular, we do
2394          * not want user-space to be able to predict the process start-time by
2395          * stalling fork(2) after we recorded the start_time but before it is
2396          * visible to the system.
2397          */
2398
2399         p->start_time = ktime_get_ns();
2400         p->start_boottime = ktime_get_boottime_ns();
2401
2402         /*
2403          * Make it visible to the rest of the system, but dont wake it up yet.
2404          * Need tasklist lock for parent etc handling!
2405          */
2406         write_lock_irq(&tasklist_lock);
2407
2408         /* CLONE_PARENT re-uses the old parent */
2409         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2410                 p->real_parent = current->real_parent;
2411                 p->parent_exec_id = current->parent_exec_id;
2412                 if (clone_flags & CLONE_THREAD)
2413                         p->exit_signal = -1;
2414                 else
2415                         p->exit_signal = current->group_leader->exit_signal;
2416         } else {
2417                 p->real_parent = current;
2418                 p->parent_exec_id = current->self_exec_id;
2419                 p->exit_signal = args->exit_signal;
2420         }
2421
2422         klp_copy_process(p);
2423
2424         sched_core_fork(p);
2425
2426         spin_lock(&current->sighand->siglock);
2427
2428         rv_task_fork(p);
2429
2430         rseq_fork(p, clone_flags);
2431
2432         /* Don't start children in a dying pid namespace */
2433         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2434                 retval = -ENOMEM;
2435                 goto bad_fork_cancel_cgroup;
2436         }
2437
2438         /* Let kill terminate clone/fork in the middle */
2439         if (fatal_signal_pending(current)) {
2440                 retval = -EINTR;
2441                 goto bad_fork_cancel_cgroup;
2442         }
2443
2444         /* No more failure paths after this point. */
2445
2446         /*
2447          * Copy seccomp details explicitly here, in case they were changed
2448          * before holding sighand lock.
2449          */
2450         copy_seccomp(p);
2451
2452         init_task_pid_links(p);
2453         if (likely(p->pid)) {
2454                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2455
2456                 init_task_pid(p, PIDTYPE_PID, pid);
2457                 if (thread_group_leader(p)) {
2458                         init_task_pid(p, PIDTYPE_TGID, pid);
2459                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2460                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2461
2462                         if (is_child_reaper(pid)) {
2463                                 ns_of_pid(pid)->child_reaper = p;
2464                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2465                         }
2466                         p->signal->shared_pending.signal = delayed.signal;
2467                         p->signal->tty = tty_kref_get(current->signal->tty);
2468                         /*
2469                          * Inherit has_child_subreaper flag under the same
2470                          * tasklist_lock with adding child to the process tree
2471                          * for propagate_has_child_subreaper optimization.
2472                          */
2473                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2474                                                          p->real_parent->signal->is_child_subreaper;
2475                         list_add_tail(&p->sibling, &p->real_parent->children);
2476                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2477                         attach_pid(p, PIDTYPE_TGID);
2478                         attach_pid(p, PIDTYPE_PGID);
2479                         attach_pid(p, PIDTYPE_SID);
2480                         __this_cpu_inc(process_counts);
2481                 } else {
2482                         current->signal->nr_threads++;
2483                         current->signal->quick_threads++;
2484                         atomic_inc(&current->signal->live);
2485                         refcount_inc(&current->signal->sigcnt);
2486                         task_join_group_stop(p);
2487                         list_add_tail_rcu(&p->thread_group,
2488                                           &p->group_leader->thread_group);
2489                         list_add_tail_rcu(&p->thread_node,
2490                                           &p->signal->thread_head);
2491                 }
2492                 attach_pid(p, PIDTYPE_PID);
2493                 nr_threads++;
2494         }
2495         total_forks++;
2496         hlist_del_init(&delayed.node);
2497         spin_unlock(&current->sighand->siglock);
2498         syscall_tracepoint_update(p);
2499         write_unlock_irq(&tasklist_lock);
2500
2501         if (pidfile)
2502                 fd_install(pidfd, pidfile);
2503
2504         proc_fork_connector(p);
2505         sched_post_fork(p);
2506         cgroup_post_fork(p, args);
2507         perf_event_fork(p);
2508
2509         trace_task_newtask(p, clone_flags);
2510         uprobe_copy_process(p, clone_flags);
2511
2512         copy_oom_score_adj(clone_flags, p);
2513
2514         return p;
2515
2516 bad_fork_cancel_cgroup:
2517         sched_core_free(p);
2518         spin_unlock(&current->sighand->siglock);
2519         write_unlock_irq(&tasklist_lock);
2520         cgroup_cancel_fork(p, args);
2521 bad_fork_put_pidfd:
2522         if (clone_flags & CLONE_PIDFD) {
2523                 fput(pidfile);
2524                 put_unused_fd(pidfd);
2525         }
2526 bad_fork_free_pid:
2527         if (pid != &init_struct_pid)
2528                 free_pid(pid);
2529 bad_fork_cleanup_thread:
2530         exit_thread(p);
2531 bad_fork_cleanup_io:
2532         if (p->io_context)
2533                 exit_io_context(p);
2534 bad_fork_cleanup_namespaces:
2535         exit_task_namespaces(p);
2536 bad_fork_cleanup_mm:
2537         if (p->mm) {
2538                 mm_clear_owner(p->mm, p);
2539                 mmput(p->mm);
2540         }
2541 bad_fork_cleanup_signal:
2542         if (!(clone_flags & CLONE_THREAD))
2543                 free_signal_struct(p->signal);
2544 bad_fork_cleanup_sighand:
2545         __cleanup_sighand(p->sighand);
2546 bad_fork_cleanup_fs:
2547         exit_fs(p); /* blocking */
2548 bad_fork_cleanup_files:
2549         exit_files(p); /* blocking */
2550 bad_fork_cleanup_semundo:
2551         exit_sem(p);
2552 bad_fork_cleanup_security:
2553         security_task_free(p);
2554 bad_fork_cleanup_audit:
2555         audit_free(p);
2556 bad_fork_cleanup_perf:
2557         perf_event_free_task(p);
2558 bad_fork_cleanup_policy:
2559         lockdep_free_task(p);
2560 #ifdef CONFIG_NUMA
2561         mpol_put(p->mempolicy);
2562 #endif
2563 bad_fork_cleanup_delayacct:
2564         delayacct_tsk_free(p);
2565 bad_fork_cleanup_count:
2566         dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2567         exit_creds(p);
2568 bad_fork_free:
2569         WRITE_ONCE(p->__state, TASK_DEAD);
2570         exit_task_stack_account(p);
2571         put_task_stack(p);
2572         delayed_free_task(p);
2573 fork_out:
2574         spin_lock_irq(&current->sighand->siglock);
2575         hlist_del_init(&delayed.node);
2576         spin_unlock_irq(&current->sighand->siglock);
2577         return ERR_PTR(retval);
2578 }
2579
2580 static inline void init_idle_pids(struct task_struct *idle)
2581 {
2582         enum pid_type type;
2583
2584         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2585                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2586                 init_task_pid(idle, type, &init_struct_pid);
2587         }
2588 }
2589
2590 static int idle_dummy(void *dummy)
2591 {
2592         /* This function is never called */
2593         return 0;
2594 }
2595
2596 struct task_struct * __init fork_idle(int cpu)
2597 {
2598         struct task_struct *task;
2599         struct kernel_clone_args args = {
2600                 .flags          = CLONE_VM,
2601                 .fn             = &idle_dummy,
2602                 .fn_arg         = NULL,
2603                 .kthread        = 1,
2604                 .idle           = 1,
2605         };
2606
2607         task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2608         if (!IS_ERR(task)) {
2609                 init_idle_pids(task);
2610                 init_idle(task, cpu);
2611         }
2612
2613         return task;
2614 }
2615
2616 /*
2617  * This is like kernel_clone(), but shaved down and tailored to just
2618  * creating io_uring workers. It returns a created task, or an error pointer.
2619  * The returned task is inactive, and the caller must fire it up through
2620  * wake_up_new_task(p). All signals are blocked in the created task.
2621  */
2622 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2623 {
2624         unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2625                                 CLONE_IO;
2626         struct kernel_clone_args args = {
2627                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2628                                     CLONE_UNTRACED) & ~CSIGNAL),
2629                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2630                 .fn             = fn,
2631                 .fn_arg         = arg,
2632                 .io_thread      = 1,
2633         };
2634
2635         return copy_process(NULL, 0, node, &args);
2636 }
2637
2638 /*
2639  *  Ok, this is the main fork-routine.
2640  *
2641  * It copies the process, and if successful kick-starts
2642  * it and waits for it to finish using the VM if required.
2643  *
2644  * args->exit_signal is expected to be checked for sanity by the caller.
2645  */
2646 pid_t kernel_clone(struct kernel_clone_args *args)
2647 {
2648         u64 clone_flags = args->flags;
2649         struct completion vfork;
2650         struct pid *pid;
2651         struct task_struct *p;
2652         int trace = 0;
2653         pid_t nr;
2654
2655         /*
2656          * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2657          * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2658          * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2659          * field in struct clone_args and it still doesn't make sense to have
2660          * them both point at the same memory location. Performing this check
2661          * here has the advantage that we don't need to have a separate helper
2662          * to check for legacy clone().
2663          */
2664         if ((args->flags & CLONE_PIDFD) &&
2665             (args->flags & CLONE_PARENT_SETTID) &&
2666             (args->pidfd == args->parent_tid))
2667                 return -EINVAL;
2668
2669         /*
2670          * Determine whether and which event to report to ptracer.  When
2671          * called from kernel_thread or CLONE_UNTRACED is explicitly
2672          * requested, no event is reported; otherwise, report if the event
2673          * for the type of forking is enabled.
2674          */
2675         if (!(clone_flags & CLONE_UNTRACED)) {
2676                 if (clone_flags & CLONE_VFORK)
2677                         trace = PTRACE_EVENT_VFORK;
2678                 else if (args->exit_signal != SIGCHLD)
2679                         trace = PTRACE_EVENT_CLONE;
2680                 else
2681                         trace = PTRACE_EVENT_FORK;
2682
2683                 if (likely(!ptrace_event_enabled(current, trace)))
2684                         trace = 0;
2685         }
2686
2687         p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2688         add_latent_entropy();
2689
2690         if (IS_ERR(p))
2691                 return PTR_ERR(p);
2692
2693         /*
2694          * Do this prior waking up the new thread - the thread pointer
2695          * might get invalid after that point, if the thread exits quickly.
2696          */
2697         trace_sched_process_fork(current, p);
2698
2699         pid = get_task_pid(p, PIDTYPE_PID);
2700         nr = pid_vnr(pid);
2701
2702         if (clone_flags & CLONE_PARENT_SETTID)
2703                 put_user(nr, args->parent_tid);
2704
2705         if (clone_flags & CLONE_VFORK) {
2706                 p->vfork_done = &vfork;
2707                 init_completion(&vfork);
2708                 get_task_struct(p);
2709         }
2710
2711         if (IS_ENABLED(CONFIG_LRU_GEN) && !(clone_flags & CLONE_VM)) {
2712                 /* lock the task to synchronize with memcg migration */
2713                 task_lock(p);
2714                 lru_gen_add_mm(p->mm);
2715                 task_unlock(p);
2716         }
2717
2718         wake_up_new_task(p);
2719
2720         /* forking complete and child started to run, tell ptracer */
2721         if (unlikely(trace))
2722                 ptrace_event_pid(trace, pid);
2723
2724         if (clone_flags & CLONE_VFORK) {
2725                 if (!wait_for_vfork_done(p, &vfork))
2726                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2727         }
2728
2729         put_pid(pid);
2730         return nr;
2731 }
2732
2733 /*
2734  * Create a kernel thread.
2735  */
2736 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2737 {
2738         struct kernel_clone_args args = {
2739                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2740                                     CLONE_UNTRACED) & ~CSIGNAL),
2741                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2742                 .fn             = fn,
2743                 .fn_arg         = arg,
2744                 .kthread        = 1,
2745         };
2746
2747         return kernel_clone(&args);
2748 }
2749
2750 /*
2751  * Create a user mode thread.
2752  */
2753 pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2754 {
2755         struct kernel_clone_args args = {
2756                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2757                                     CLONE_UNTRACED) & ~CSIGNAL),
2758                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2759                 .fn             = fn,
2760                 .fn_arg         = arg,
2761         };
2762
2763         return kernel_clone(&args);
2764 }
2765
2766 #ifdef __ARCH_WANT_SYS_FORK
2767 SYSCALL_DEFINE0(fork)
2768 {
2769 #ifdef CONFIG_MMU
2770         struct kernel_clone_args args = {
2771                 .exit_signal = SIGCHLD,
2772         };
2773
2774         return kernel_clone(&args);
2775 #else
2776         /* can not support in nommu mode */
2777         return -EINVAL;
2778 #endif
2779 }
2780 #endif
2781
2782 #ifdef __ARCH_WANT_SYS_VFORK
2783 SYSCALL_DEFINE0(vfork)
2784 {
2785         struct kernel_clone_args args = {
2786                 .flags          = CLONE_VFORK | CLONE_VM,
2787                 .exit_signal    = SIGCHLD,
2788         };
2789
2790         return kernel_clone(&args);
2791 }
2792 #endif
2793
2794 #ifdef __ARCH_WANT_SYS_CLONE
2795 #ifdef CONFIG_CLONE_BACKWARDS
2796 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2797                  int __user *, parent_tidptr,
2798                  unsigned long, tls,
2799                  int __user *, child_tidptr)
2800 #elif defined(CONFIG_CLONE_BACKWARDS2)
2801 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2802                  int __user *, parent_tidptr,
2803                  int __user *, child_tidptr,
2804                  unsigned long, tls)
2805 #elif defined(CONFIG_CLONE_BACKWARDS3)
2806 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2807                 int, stack_size,
2808                 int __user *, parent_tidptr,
2809                 int __user *, child_tidptr,
2810                 unsigned long, tls)
2811 #else
2812 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2813                  int __user *, parent_tidptr,
2814                  int __user *, child_tidptr,
2815                  unsigned long, tls)
2816 #endif
2817 {
2818         struct kernel_clone_args args = {
2819                 .flags          = (lower_32_bits(clone_flags) & ~CSIGNAL),
2820                 .pidfd          = parent_tidptr,
2821                 .child_tid      = child_tidptr,
2822                 .parent_tid     = parent_tidptr,
2823                 .exit_signal    = (lower_32_bits(clone_flags) & CSIGNAL),
2824                 .stack          = newsp,
2825                 .tls            = tls,
2826         };
2827
2828         return kernel_clone(&args);
2829 }
2830 #endif
2831
2832 #ifdef __ARCH_WANT_SYS_CLONE3
2833
2834 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2835                                               struct clone_args __user *uargs,
2836                                               size_t usize)
2837 {
2838         int err;
2839         struct clone_args args;
2840         pid_t *kset_tid = kargs->set_tid;
2841
2842         BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2843                      CLONE_ARGS_SIZE_VER0);
2844         BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2845                      CLONE_ARGS_SIZE_VER1);
2846         BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2847                      CLONE_ARGS_SIZE_VER2);
2848         BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2849
2850         if (unlikely(usize > PAGE_SIZE))
2851                 return -E2BIG;
2852         if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2853                 return -EINVAL;
2854
2855         err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2856         if (err)
2857                 return err;
2858
2859         if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2860                 return -EINVAL;
2861
2862         if (unlikely(!args.set_tid && args.set_tid_size > 0))
2863                 return -EINVAL;
2864
2865         if (unlikely(args.set_tid && args.set_tid_size == 0))
2866                 return -EINVAL;
2867
2868         /*
2869          * Verify that higher 32bits of exit_signal are unset and that
2870          * it is a valid signal
2871          */
2872         if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2873                      !valid_signal(args.exit_signal)))
2874                 return -EINVAL;
2875
2876         if ((args.flags & CLONE_INTO_CGROUP) &&
2877             (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2878                 return -EINVAL;
2879
2880         *kargs = (struct kernel_clone_args){
2881                 .flags          = args.flags,
2882                 .pidfd          = u64_to_user_ptr(args.pidfd),
2883                 .child_tid      = u64_to_user_ptr(args.child_tid),
2884                 .parent_tid     = u64_to_user_ptr(args.parent_tid),
2885                 .exit_signal    = args.exit_signal,
2886                 .stack          = args.stack,
2887                 .stack_size     = args.stack_size,
2888                 .tls            = args.tls,
2889                 .set_tid_size   = args.set_tid_size,
2890                 .cgroup         = args.cgroup,
2891         };
2892
2893         if (args.set_tid &&
2894                 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2895                         (kargs->set_tid_size * sizeof(pid_t))))
2896                 return -EFAULT;
2897
2898         kargs->set_tid = kset_tid;
2899
2900         return 0;
2901 }
2902
2903 /**
2904  * clone3_stack_valid - check and prepare stack
2905  * @kargs: kernel clone args
2906  *
2907  * Verify that the stack arguments userspace gave us are sane.
2908  * In addition, set the stack direction for userspace since it's easy for us to
2909  * determine.
2910  */
2911 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2912 {
2913         if (kargs->stack == 0) {
2914                 if (kargs->stack_size > 0)
2915                         return false;
2916         } else {
2917                 if (kargs->stack_size == 0)
2918                         return false;
2919
2920                 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2921                         return false;
2922
2923 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2924                 kargs->stack += kargs->stack_size;
2925 #endif
2926         }
2927
2928         return true;
2929 }
2930
2931 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2932 {
2933         /* Verify that no unknown flags are passed along. */
2934         if (kargs->flags &
2935             ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2936                 return false;
2937
2938         /*
2939          * - make the CLONE_DETACHED bit reusable for clone3
2940          * - make the CSIGNAL bits reusable for clone3
2941          */
2942         if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2943                 return false;
2944
2945         if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2946             (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2947                 return false;
2948
2949         if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2950             kargs->exit_signal)
2951                 return false;
2952
2953         if (!clone3_stack_valid(kargs))
2954                 return false;
2955
2956         return true;
2957 }
2958
2959 /**
2960  * clone3 - create a new process with specific properties
2961  * @uargs: argument structure
2962  * @size:  size of @uargs
2963  *
2964  * clone3() is the extensible successor to clone()/clone2().
2965  * It takes a struct as argument that is versioned by its size.
2966  *
2967  * Return: On success, a positive PID for the child process.
2968  *         On error, a negative errno number.
2969  */
2970 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2971 {
2972         int err;
2973
2974         struct kernel_clone_args kargs;
2975         pid_t set_tid[MAX_PID_NS_LEVEL];
2976
2977         kargs.set_tid = set_tid;
2978
2979         err = copy_clone_args_from_user(&kargs, uargs, size);
2980         if (err)
2981                 return err;
2982
2983         if (!clone3_args_valid(&kargs))
2984                 return -EINVAL;
2985
2986         return kernel_clone(&kargs);
2987 }
2988 #endif
2989
2990 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2991 {
2992         struct task_struct *leader, *parent, *child;
2993         int res;
2994
2995         read_lock(&tasklist_lock);
2996         leader = top = top->group_leader;
2997 down:
2998         for_each_thread(leader, parent) {
2999                 list_for_each_entry(child, &parent->children, sibling) {
3000                         res = visitor(child, data);
3001                         if (res) {
3002                                 if (res < 0)
3003                                         goto out;
3004                                 leader = child;
3005                                 goto down;
3006                         }
3007 up:
3008                         ;
3009                 }
3010         }
3011
3012         if (leader != top) {
3013                 child = leader;
3014                 parent = child->real_parent;
3015                 leader = parent->group_leader;
3016                 goto up;
3017         }
3018 out:
3019         read_unlock(&tasklist_lock);
3020 }
3021
3022 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
3023 #define ARCH_MIN_MMSTRUCT_ALIGN 0
3024 #endif
3025
3026 static void sighand_ctor(void *data)
3027 {
3028         struct sighand_struct *sighand = data;
3029
3030         spin_lock_init(&sighand->siglock);
3031         init_waitqueue_head(&sighand->signalfd_wqh);
3032 }
3033
3034 void __init mm_cache_init(void)
3035 {
3036         unsigned int mm_size;
3037
3038         /*
3039          * The mm_cpumask is located at the end of mm_struct, and is
3040          * dynamically sized based on the maximum CPU number this system
3041          * can have, taking hotplug into account (nr_cpu_ids).
3042          */
3043         mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
3044
3045         mm_cachep = kmem_cache_create_usercopy("mm_struct",
3046                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3047                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3048                         offsetof(struct mm_struct, saved_auxv),
3049                         sizeof_field(struct mm_struct, saved_auxv),
3050                         NULL);
3051 }
3052
3053 void __init proc_caches_init(void)
3054 {
3055         sighand_cachep = kmem_cache_create("sighand_cache",
3056                         sizeof(struct sighand_struct), 0,
3057                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3058                         SLAB_ACCOUNT, sighand_ctor);
3059         signal_cachep = kmem_cache_create("signal_cache",
3060                         sizeof(struct signal_struct), 0,
3061                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3062                         NULL);
3063         files_cachep = kmem_cache_create("files_cache",
3064                         sizeof(struct files_struct), 0,
3065                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3066                         NULL);
3067         fs_cachep = kmem_cache_create("fs_cache",
3068                         sizeof(struct fs_struct), 0,
3069                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3070                         NULL);
3071
3072         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3073         mmap_init();
3074         nsproxy_cache_init();
3075 }
3076
3077 /*
3078  * Check constraints on flags passed to the unshare system call.
3079  */
3080 static int check_unshare_flags(unsigned long unshare_flags)
3081 {
3082         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3083                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3084                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3085                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3086                                 CLONE_NEWTIME))
3087                 return -EINVAL;
3088         /*
3089          * Not implemented, but pretend it works if there is nothing
3090          * to unshare.  Note that unsharing the address space or the
3091          * signal handlers also need to unshare the signal queues (aka
3092          * CLONE_THREAD).
3093          */
3094         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3095                 if (!thread_group_empty(current))
3096                         return -EINVAL;
3097         }
3098         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3099                 if (refcount_read(&current->sighand->count) > 1)
3100                         return -EINVAL;
3101         }
3102         if (unshare_flags & CLONE_VM) {
3103                 if (!current_is_single_threaded())
3104                         return -EINVAL;
3105         }
3106
3107         return 0;
3108 }
3109
3110 /*
3111  * Unshare the filesystem structure if it is being shared
3112  */
3113 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3114 {
3115         struct fs_struct *fs = current->fs;
3116
3117         if (!(unshare_flags & CLONE_FS) || !fs)
3118                 return 0;
3119
3120         /* don't need lock here; in the worst case we'll do useless copy */
3121         if (fs->users == 1)
3122                 return 0;
3123
3124         *new_fsp = copy_fs_struct(fs);
3125         if (!*new_fsp)
3126                 return -ENOMEM;
3127
3128         return 0;
3129 }
3130
3131 /*
3132  * Unshare file descriptor table if it is being shared
3133  */
3134 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3135                struct files_struct **new_fdp)
3136 {
3137         struct files_struct *fd = current->files;
3138         int error = 0;
3139
3140         if ((unshare_flags & CLONE_FILES) &&
3141             (fd && atomic_read(&fd->count) > 1)) {
3142                 *new_fdp = dup_fd(fd, max_fds, &error);
3143                 if (!*new_fdp)
3144                         return error;
3145         }
3146
3147         return 0;
3148 }
3149
3150 /*
3151  * unshare allows a process to 'unshare' part of the process
3152  * context which was originally shared using clone.  copy_*
3153  * functions used by kernel_clone() cannot be used here directly
3154  * because they modify an inactive task_struct that is being
3155  * constructed. Here we are modifying the current, active,
3156  * task_struct.
3157  */
3158 int ksys_unshare(unsigned long unshare_flags)
3159 {
3160         struct fs_struct *fs, *new_fs = NULL;
3161         struct files_struct *new_fd = NULL;
3162         struct cred *new_cred = NULL;
3163         struct nsproxy *new_nsproxy = NULL;
3164         int do_sysvsem = 0;
3165         int err;
3166
3167         /*
3168          * If unsharing a user namespace must also unshare the thread group
3169          * and unshare the filesystem root and working directories.
3170          */
3171         if (unshare_flags & CLONE_NEWUSER)
3172                 unshare_flags |= CLONE_THREAD | CLONE_FS;
3173         /*
3174          * If unsharing vm, must also unshare signal handlers.
3175          */
3176         if (unshare_flags & CLONE_VM)
3177                 unshare_flags |= CLONE_SIGHAND;
3178         /*
3179          * If unsharing a signal handlers, must also unshare the signal queues.
3180          */
3181         if (unshare_flags & CLONE_SIGHAND)
3182                 unshare_flags |= CLONE_THREAD;
3183         /*
3184          * If unsharing namespace, must also unshare filesystem information.
3185          */
3186         if (unshare_flags & CLONE_NEWNS)
3187                 unshare_flags |= CLONE_FS;
3188
3189         err = check_unshare_flags(unshare_flags);
3190         if (err)
3191                 goto bad_unshare_out;
3192         /*
3193          * CLONE_NEWIPC must also detach from the undolist: after switching
3194          * to a new ipc namespace, the semaphore arrays from the old
3195          * namespace are unreachable.
3196          */
3197         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3198                 do_sysvsem = 1;
3199         err = unshare_fs(unshare_flags, &new_fs);
3200         if (err)
3201                 goto bad_unshare_out;
3202         err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3203         if (err)
3204                 goto bad_unshare_cleanup_fs;
3205         err = unshare_userns(unshare_flags, &new_cred);
3206         if (err)
3207                 goto bad_unshare_cleanup_fd;
3208         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3209                                          new_cred, new_fs);
3210         if (err)
3211                 goto bad_unshare_cleanup_cred;
3212
3213         if (new_cred) {
3214                 err = set_cred_ucounts(new_cred);
3215                 if (err)
3216                         goto bad_unshare_cleanup_cred;
3217         }
3218
3219         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3220                 if (do_sysvsem) {
3221                         /*
3222                          * CLONE_SYSVSEM is equivalent to sys_exit().
3223                          */
3224                         exit_sem(current);
3225                 }
3226                 if (unshare_flags & CLONE_NEWIPC) {
3227                         /* Orphan segments in old ns (see sem above). */
3228                         exit_shm(current);
3229                         shm_init_task(current);
3230                 }
3231
3232                 if (new_nsproxy)
3233                         switch_task_namespaces(current, new_nsproxy);
3234
3235                 task_lock(current);
3236
3237                 if (new_fs) {
3238                         fs = current->fs;
3239                         spin_lock(&fs->lock);
3240                         current->fs = new_fs;
3241                         if (--fs->users)
3242                                 new_fs = NULL;
3243                         else
3244                                 new_fs = fs;
3245                         spin_unlock(&fs->lock);
3246                 }
3247
3248                 if (new_fd)
3249                         swap(current->files, new_fd);
3250
3251                 task_unlock(current);
3252
3253                 if (new_cred) {
3254                         /* Install the new user namespace */
3255                         commit_creds(new_cred);
3256                         new_cred = NULL;
3257                 }
3258         }
3259
3260         perf_event_namespaces(current);
3261
3262 bad_unshare_cleanup_cred:
3263         if (new_cred)
3264                 put_cred(new_cred);
3265 bad_unshare_cleanup_fd:
3266         if (new_fd)
3267                 put_files_struct(new_fd);
3268
3269 bad_unshare_cleanup_fs:
3270         if (new_fs)
3271                 free_fs_struct(new_fs);
3272
3273 bad_unshare_out:
3274         return err;
3275 }
3276
3277 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3278 {
3279         return ksys_unshare(unshare_flags);
3280 }
3281
3282 /*
3283  *      Helper to unshare the files of the current task.
3284  *      We don't want to expose copy_files internals to
3285  *      the exec layer of the kernel.
3286  */
3287
3288 int unshare_files(void)
3289 {
3290         struct task_struct *task = current;
3291         struct files_struct *old, *copy = NULL;
3292         int error;
3293
3294         error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, &copy);
3295         if (error || !copy)
3296                 return error;
3297
3298         old = task->files;
3299         task_lock(task);
3300         task->files = copy;
3301         task_unlock(task);
3302         put_files_struct(old);
3303         return 0;
3304 }
3305
3306 int sysctl_max_threads(struct ctl_table *table, int write,
3307                        void *buffer, size_t *lenp, loff_t *ppos)
3308 {
3309         struct ctl_table t;
3310         int ret;
3311         int threads = max_threads;
3312         int min = 1;
3313         int max = MAX_THREADS;
3314
3315         t = *table;
3316         t.data = &threads;
3317         t.extra1 = &min;
3318         t.extra2 = &max;
3319
3320         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3321         if (ret || !write)
3322                 return ret;
3323
3324         max_threads = threads;
3325
3326         return 0;
3327 }