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