kernel/fork.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/kernel/fork.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  */
   7
   8 /*
   9  *  'fork.c' contains the help-routines for the 'fork' system call
  10  * (see also entry.S and others).
  11  * Fork is rather simple, once you get the hang of it, but the memory
  12  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  13  */
  14
  15 #include <linux/anon_inodes.h>
  16 #include <linux/slab.h>
  17 #include <linux/sched/autogroup.h>
  18 #include <linux/sched/mm.h>
  19 #include <linux/sched/coredump.h>
  20 #include <linux/sched/user.h>
  21 #include <linux/sched/numa_balancing.h>
  22 #include <linux/sched/stat.h>
  23 #include <linux/sched/task.h>
  24 #include <linux/sched/task_stack.h>
  25 #include <linux/sched/cputime.h>
  26 #include <linux/seq_file.h>
  27 #include <linux/rtmutex.h>
  28 #include <linux/init.h>
  29 #include <linux/unistd.h>
  30 #include <linux/module.h>
  31 #include <linux/vmalloc.h>
  32 #include <linux/completion.h>
  33 #include <linux/personality.h>
  34 #include <linux/mempolicy.h>
  35 #include <linux/sem.h>
  36 #include <linux/file.h>
  37 #include <linux/fdtable.h>
  38 #include <linux/iocontext.h>
  39 #include <linux/key.h>
  40 #include <linux/kmsan.h>
  41 #include <linux/binfmts.h>
  42 #include <linux/mman.h>
  43 #include <linux/mmu_notifier.h>
  44 #include <linux/fs.h>
  45 #include <linux/mm.h>
  46 #include <linux/mm_inline.h>
  47 #include <linux/nsproxy.h>
  48 #include <linux/capability.h>
  49 #include <linux/cpu.h>
  50 #include <linux/cgroup.h>
  51 #include <linux/security.h>
  52 #include <linux/hugetlb.h>
  53 #include <linux/seccomp.h>
  54 #include <linux/swap.h>
  55 #include <linux/syscalls.h>
  56 #include <linux/jiffies.h>
  57 #include <linux/futex.h>
  58 #include <linux/compat.h>
  59 #include <linux/kthread.h>
  60 #include <linux/task_io_accounting_ops.h>
  61 #include <linux/rcupdate.h>
  62 #include <linux/ptrace.h>
  63 #include <linux/mount.h>
  64 #include <linux/audit.h>
  65 #include <linux/memcontrol.h>
  66 #include <linux/ftrace.h>
  67 #include <linux/proc_fs.h>
  68 #include <linux/profile.h>
  69 #include <linux/rmap.h>
  70 #include <linux/ksm.h>
  71 #include <linux/acct.h>
  72 #include <linux/userfaultfd_k.h>
  73 #include <linux/tsacct_kern.h>
  74 #include <linux/cn_proc.h>
  75 #include <linux/freezer.h>
  76 #include <linux/delayacct.h>
  77 #include <linux/taskstats_kern.h>
  78 #include <linux/tty.h>
  79 #include <linux/fs_struct.h>
  80 #include <linux/magic.h>
  81 #include <linux/perf_event.h>
  82 #include <linux/posix-timers.h>
  83 #include <linux/user-return-notifier.h>
  84 #include <linux/oom.h>
  85 #include <linux/khugepaged.h>
  86 #include <linux/signalfd.h>
  87 #include <linux/uprobes.h>
  88 #include <linux/aio.h>
  89 #include <linux/compiler.h>
  90 #include <linux/sysctl.h>
  91 #include <linux/kcov.h>
  92 #include <linux/livepatch.h>
  93 #include <linux/thread_info.h>
  94 #include <linux/stackleak.h>
  95 #include <linux/kasan.h>
  96 #include <linux/scs.h>
  97 #include <linux/io_uring.h>
  98 #include <linux/bpf.h>
  99 #include <linux/stackprotector.h>
 100
 101 #include <asm/pgalloc.h>
 102 #include <linux/uaccess.h>
 103 #include <asm/mmu_context.h>
 104 #include <asm/cacheflush.h>
 105 #include <asm/tlbflush.h>
 106
 107 #include <trace/events/sched.h>
 108
 109 #define CREATE_TRACE_POINTS
 110 #include <trace/events/task.h>
 111
 112 /*
 113  * Minimum number of threads to boot the kernel
 114  */
 115 #define MIN_THREADS 20
 116
 117 /*
 118  * Maximum number of threads
 119  */
 120 #define MAX_THREADS FUTEX_TID_MASK
 121
 122 /*
 123  * Protected counters by write_lock_irq(&tasklist_lock)
 124  */
 125 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 126 int nr_threads;                 /* The idle threads do not count.. */
 127
 128 static int max_threads;         /* tunable limit on nr_threads */
 129
 130 #define NAMED_ARRAY_INDEX(x)    [x] = __stringify(x)
 131
 132 static const char * const resident_page_types[] = {
 133         NAMED_ARRAY_INDEX(MM_FILEPAGES),
 134         NAMED_ARRAY_INDEX(MM_ANONPAGES),
 135         NAMED_ARRAY_INDEX(MM_SWAPENTS),
 136         NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
 137 };
 138
 139 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 140
 141 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 142
 143 #ifdef CONFIG_PROVE_RCU
 144 int lockdep_tasklist_lock_is_held(void)
 145 {
 146         return lockdep_is_held(&tasklist_lock);
 147 }
 148 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 149 #endif /* #ifdef CONFIG_PROVE_RCU */
 150
 151 int nr_processes(void)
 152 {
 153         int cpu;
 154         int total = 0;
 155
 156         for_each_possible_cpu(cpu)
 157                 total += per_cpu(process_counts, cpu);
 158
 159         return total;
 160 }
 161
 162 void __weak arch_release_task_struct(struct task_struct *tsk)
 163 {
 164 }
 165
 166 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 167 static struct kmem_cache *task_struct_cachep;
 168
 169 static inline struct task_struct *alloc_task_struct_node(int node)
 170 {
 171         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 172 }
 173
 174 static inline void free_task_struct(struct task_struct *tsk)
 175 {
 176         kmem_cache_free(task_struct_cachep, tsk);
 177 }
 178 #endif
 179
 180 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 181
 182 /*
 183  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 184  * kmemcache based allocator.
 185  */
 186 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 187
 188 #  ifdef CONFIG_VMAP_STACK
 189 /*
 190  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 191  * flush.  Try to minimize the number of calls by caching stacks.
 192  */
 193 #define NR_CACHED_STACKS 2
 194 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 195
 196 struct vm_stack {
 197         struct rcu_head rcu;
 198         struct vm_struct *stack_vm_area;
 199 };
 200
 201 static bool try_release_thread_stack_to_cache(struct vm_struct *vm)
 202 {
 203         unsigned int i;
 204
 205         for (i = 0; i < NR_CACHED_STACKS; i++) {
 206                 if (this_cpu_cmpxchg(cached_stacks[i], NULL, vm) != NULL)
 207                         continue;
 208                 return true;
 209         }
 210         return false;
 211 }
 212
 213 static void thread_stack_free_rcu(struct rcu_head *rh)
 214 {
 215         struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
 216
 217         if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
 218                 return;
 219
 220         vfree(vm_stack);
 221 }
 222
 223 static void thread_stack_delayed_free(struct task_struct *tsk)
 224 {
 225         struct vm_stack *vm_stack = tsk->stack;
 226
 227         vm_stack->stack_vm_area = tsk->stack_vm_area;
 228         call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
 229 }
 230
 231 static int free_vm_stack_cache(unsigned int cpu)
 232 {
 233         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 234         int i;
 235
 236         for (i = 0; i < NR_CACHED_STACKS; i++) {
 237                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 238
 239                 if (!vm_stack)
 240                         continue;
 241
 242                 vfree(vm_stack->addr);
 243                 cached_vm_stacks[i] = NULL;
 244         }
 245
 246         return 0;
 247 }
 248
 249 static int memcg_charge_kernel_stack(struct vm_struct *vm)
 250 {
 251         int i;
 252         int ret;
 253
 254         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 255         BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 256
 257         for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 258                 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0);
 259                 if (ret)
 260                         goto err;
 261         }
 262         return 0;
 263 err:
 264         /*
 265          * If memcg_kmem_charge_page() fails, page's memory cgroup pointer is
 266          * NULL, and memcg_kmem_uncharge_page() in free_thread_stack() will
 267          * ignore this page.
 268          */
 269         for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 270                 memcg_kmem_uncharge_page(vm->pages[i], 0);
 271         return ret;
 272 }
 273
 274 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 275 {
 276         struct vm_struct *vm;
 277         void *stack;
 278         int i;
 279
 280         for (i = 0; i < NR_CACHED_STACKS; i++) {
 281                 struct vm_struct *s;
 282
 283                 s = this_cpu_xchg(cached_stacks[i], NULL);
 284
 285                 if (!s)
 286                         continue;
 287
 288                 /* Reset stack metadata. */
 289                 kasan_unpoison_range(s->addr, THREAD_SIZE);
 290
 291                 stack = kasan_reset_tag(s->addr);
 292
 293                 /* Clear stale pointers from reused stack. */
 294                 memset(stack, 0, THREAD_SIZE);
 295
 296                 if (memcg_charge_kernel_stack(s)) {
 297                         vfree(s->addr);
 298                         return -ENOMEM;
 299                 }
 300
 301                 tsk->stack_vm_area = s;
 302                 tsk->stack = stack;
 303                 return 0;
 304         }
 305
 306         /*
 307          * Allocated stacks are cached and later reused by new threads,
 308          * so memcg accounting is performed manually on assigning/releasing
 309          * stacks to tasks. Drop __GFP_ACCOUNT.
 310          */
 311         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 312                                      VMALLOC_START, VMALLOC_END,
 313                                      THREADINFO_GFP & ~__GFP_ACCOUNT,
 314                                      PAGE_KERNEL,
 315                                      0, node, __builtin_return_address(0));
 316         if (!stack)
 317                 return -ENOMEM;
 318
 319         vm = find_vm_area(stack);
 320         if (memcg_charge_kernel_stack(vm)) {
 321                 vfree(stack);
 322                 return -ENOMEM;
 323         }
 324         /*
 325          * We can't call find_vm_area() in interrupt context, and
 326          * free_thread_stack() can be called in interrupt context,
 327          * so cache the vm_struct.
 328          */
 329         tsk->stack_vm_area = vm;
 330         stack = kasan_reset_tag(stack);
 331         tsk->stack = stack;
 332         return 0;
 333 }
 334
 335 static void free_thread_stack(struct task_struct *tsk)
 336 {
 337         if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
 338                 thread_stack_delayed_free(tsk);
 339
 340         tsk->stack = NULL;
 341         tsk->stack_vm_area = NULL;
 342 }
 343
 344 #  else /* !CONFIG_VMAP_STACK */
 345
 346 static void thread_stack_free_rcu(struct rcu_head *rh)
 347 {
 348         __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
 349 }
 350
 351 static void thread_stack_delayed_free(struct task_struct *tsk)
 352 {
 353         struct rcu_head *rh = tsk->stack;
 354
 355         call_rcu(rh, thread_stack_free_rcu);
 356 }
 357
 358 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 359 {
 360         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 361                                              THREAD_SIZE_ORDER);
 362
 363         if (likely(page)) {
 364                 tsk->stack = kasan_reset_tag(page_address(page));
 365                 return 0;
 366         }
 367         return -ENOMEM;
 368 }
 369
 370 static void free_thread_stack(struct task_struct *tsk)
 371 {
 372         thread_stack_delayed_free(tsk);
 373         tsk->stack = NULL;
 374 }
 375
 376 #  endif /* CONFIG_VMAP_STACK */
 377 # else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
 378
 379 static struct kmem_cache *thread_stack_cache;
 380
 381 static void thread_stack_free_rcu(struct rcu_head *rh)
 382 {
 383         kmem_cache_free(thread_stack_cache, rh);
 384 }
 385
 386 static void thread_stack_delayed_free(struct task_struct *tsk)
 387 {
 388         struct rcu_head *rh = tsk->stack;
 389
 390         call_rcu(rh, thread_stack_free_rcu);
 391 }
 392
 393 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 394 {
 395         unsigned long *stack;
 396         stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 397         stack = kasan_reset_tag(stack);
 398         tsk->stack = stack;
 399         return stack ? 0 : -ENOMEM;
 400 }
 401
 402 static void free_thread_stack(struct task_struct *tsk)
 403 {
 404         thread_stack_delayed_free(tsk);
 405         tsk->stack = NULL;
 406 }
 407
 408 void thread_stack_cache_init(void)
 409 {
 410         thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 411                                         THREAD_SIZE, THREAD_SIZE, 0, 0,
 412                                         THREAD_SIZE, NULL);
 413         BUG_ON(thread_stack_cache == NULL);
 414 }
 415
 416 # endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
 417 #else /* CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
 418
 419 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 420 {
 421         unsigned long *stack;
 422
 423         stack = arch_alloc_thread_stack_node(tsk, node);
 424         tsk->stack = stack;
 425         return stack ? 0 : -ENOMEM;
 426 }
 427
 428 static void free_thread_stack(struct task_struct *tsk)
 429 {
 430         arch_free_thread_stack(tsk);
 431         tsk->stack = NULL;
 432 }
 433
 434 #endif /* !CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
 435
 436 /* SLAB cache for signal_struct structures (tsk->signal) */
 437 static struct kmem_cache *signal_cachep;
 438
 439 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 440 struct kmem_cache *sighand_cachep;
 441
 442 /* SLAB cache for files_struct structures (tsk->files) */
 443 struct kmem_cache *files_cachep;
 444
 445 /* SLAB cache for fs_struct structures (tsk->fs) */
 446 struct kmem_cache *fs_cachep;
 447
 448 /* SLAB cache for vm_area_struct structures */
 449 static struct kmem_cache *vm_area_cachep;
 450
 451 /* SLAB cache for mm_struct structures (tsk->mm) */
 452 static struct kmem_cache *mm_cachep;
 453
 454 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 455 {
 456         struct vm_area_struct *vma;
 457
 458         vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 459         if (vma)
 460                 vma_init(vma, mm);
 461         return vma;
 462 }
 463
 464 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 465 {
 466         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 467
 468         if (new) {
 469                 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
 470                 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
 471                 /*
 472                  * orig->shared.rb may be modified concurrently, but the clone
 473                  * will be reinitialized.
 474                  */
 475                 *new = data_race(*orig);
 476                 INIT_LIST_HEAD(&new->anon_vma_chain);
 477                 dup_anon_vma_name(orig, new);
 478         }
 479         return new;
 480 }
 481
 482 void vm_area_free(struct vm_area_struct *vma)
 483 {
 484         free_anon_vma_name(vma);
 485         kmem_cache_free(vm_area_cachep, vma);
 486 }
 487
 488 static void account_kernel_stack(struct task_struct *tsk, int account)
 489 {
 490         if (IS_ENABLED(CONFIG_VMAP_STACK)) {
 491                 struct vm_struct *vm = task_stack_vm_area(tsk);
 492                 int i;
 493
 494                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 495                         mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
 496                                               account * (PAGE_SIZE / 1024));
 497         } else {
 498                 void *stack = task_stack_page(tsk);
 499
 500                 /* All stack pages are in the same node. */
 501                 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
 502                                       account * (THREAD_SIZE / 1024));
 503         }
 504 }
 505
 506 void exit_task_stack_account(struct task_struct *tsk)
 507 {
 508         account_kernel_stack(tsk, -1);
 509
 510         if (IS_ENABLED(CONFIG_VMAP_STACK)) {
 511                 struct vm_struct *vm;
 512                 int i;
 513
 514                 vm = task_stack_vm_area(tsk);
 515                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 516                         memcg_kmem_uncharge_page(vm->pages[i], 0);
 517         }
 518 }
 519
 520 static void release_task_stack(struct task_struct *tsk)
 521 {
 522         if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
 523                 return;  /* Better to leak the stack than to free prematurely */
 524
 525         free_thread_stack(tsk);
 526 }
 527
 528 #ifdef CONFIG_THREAD_INFO_IN_TASK
 529 void put_task_stack(struct task_struct *tsk)
 530 {
 531         if (refcount_dec_and_test(&tsk->stack_refcount))
 532                 release_task_stack(tsk);
 533 }
 534 #endif
 535
 536 void free_task(struct task_struct *tsk)
 537 {
 538 #ifdef CONFIG_SECCOMP
 539         WARN_ON_ONCE(tsk->seccomp.filter);
 540 #endif
 541         release_user_cpus_ptr(tsk);
 542         scs_release(tsk);
 543
 544 #ifndef CONFIG_THREAD_INFO_IN_TASK
 545         /*
 546          * The task is finally done with both the stack and thread_info,
 547          * so free both.
 548          */
 549         release_task_stack(tsk);
 550 #else
 551         /*
 552          * If the task had a separate stack allocation, it should be gone
 553          * by now.
 554          */
 555         WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
 556 #endif
 557         rt_mutex_debug_task_free(tsk);
 558         ftrace_graph_exit_task(tsk);
 559         arch_release_task_struct(tsk);
 560         if (tsk->flags & PF_KTHREAD)
 561                 free_kthread_struct(tsk);
 562         free_task_struct(tsk);
 563 }
 564 EXPORT_SYMBOL(free_task);
 565
 566 static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
 567 {
 568         struct file *exe_file;
 569
 570         exe_file = get_mm_exe_file(oldmm);
 571         RCU_INIT_POINTER(mm->exe_file, exe_file);
 572         /*
 573          * We depend on the oldmm having properly denied write access to the
 574          * exe_file already.
 575          */
 576         if (exe_file && deny_write_access(exe_file))
 577                 pr_warn_once("deny_write_access() failed in %s\n", __func__);
 578 }
 579
 580 #ifdef CONFIG_MMU
 581 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 582                                         struct mm_struct *oldmm)
 583 {
 584         struct vm_area_struct *mpnt, *tmp;
 585         int retval;
 586         unsigned long charge = 0;
 587         LIST_HEAD(uf);
 588         MA_STATE(old_mas, &oldmm->mm_mt, 0, 0);
 589         MA_STATE(mas, &mm->mm_mt, 0, 0);
 590
 591         uprobe_start_dup_mmap();
 592         if (mmap_write_lock_killable(oldmm)) {
 593                 retval = -EINTR;
 594                 goto fail_uprobe_end;
 595         }
 596         flush_cache_dup_mm(oldmm);
 597         uprobe_dup_mmap(oldmm, mm);
 598         /*
 599          * Not linked in yet - no deadlock potential:
 600          */
 601         mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
 602
 603         /* No ordering required: file already has been exposed. */
 604         dup_mm_exe_file(mm, oldmm);
 605
 606         mm->total_vm = oldmm->total_vm;
 607         mm->data_vm = oldmm->data_vm;
 608         mm->exec_vm = oldmm->exec_vm;
 609         mm->stack_vm = oldmm->stack_vm;
 610
 611         retval = ksm_fork(mm, oldmm);
 612         if (retval)
 613                 goto out;
 614         khugepaged_fork(mm, oldmm);
 615
 616         retval = mas_expected_entries(&mas, oldmm->map_count);
 617         if (retval)
 618                 goto out;
 619
 620         mas_for_each(&old_mas, mpnt, ULONG_MAX) {
 621                 struct file *file;
 622
 623                 if (mpnt->vm_flags & VM_DONTCOPY) {
 624                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 625                         continue;
 626                 }
 627                 charge = 0;
 628                 /*
 629                  * Don't duplicate many vmas if we've been oom-killed (for
 630                  * example)
 631                  */
 632                 if (fatal_signal_pending(current)) {
 633                         retval = -EINTR;
 634                         goto loop_out;
 635                 }
 636                 if (mpnt->vm_flags & VM_ACCOUNT) {
 637                         unsigned long len = vma_pages(mpnt);
 638
 639                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 640                                 goto fail_nomem;
 641                         charge = len;
 642                 }
 643                 tmp = vm_area_dup(mpnt);
 644                 if (!tmp)
 645                         goto fail_nomem;
 646                 retval = vma_dup_policy(mpnt, tmp);
 647                 if (retval)
 648                         goto fail_nomem_policy;
 649                 tmp->vm_mm = mm;
 650                 retval = dup_userfaultfd(tmp, &uf);
 651                 if (retval)
 652                         goto fail_nomem_anon_vma_fork;
 653                 if (tmp->vm_flags & VM_WIPEONFORK) {
 654                         /*
 655                          * VM_WIPEONFORK gets a clean slate in the child.
 656                          * Don't prepare anon_vma until fault since we don't
 657                          * copy page for current vma.
 658                          */
 659                         tmp->anon_vma = NULL;
 660                 } else if (anon_vma_fork(tmp, mpnt))
 661                         goto fail_nomem_anon_vma_fork;
 662                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 663                 file = tmp->vm_file;
 664                 if (file) {
 665                         struct address_space *mapping = file->f_mapping;
 666
 667                         get_file(file);
 668                         i_mmap_lock_write(mapping);
 669                         if (tmp->vm_flags & VM_SHARED)
 670                                 mapping_allow_writable(mapping);
 671                         flush_dcache_mmap_lock(mapping);
 672                         /* insert tmp into the share list, just after mpnt */
 673                         vma_interval_tree_insert_after(tmp, mpnt,
 674                                         &mapping->i_mmap);
 675                         flush_dcache_mmap_unlock(mapping);
 676                         i_mmap_unlock_write(mapping);
 677                 }
 678
 679                 /*
 680                  * Copy/update hugetlb private vma information.
 681                  */
 682                 if (is_vm_hugetlb_page(tmp))
 683                         hugetlb_dup_vma_private(tmp);
 684
 685                 /* Link the vma into the MT */
 686                 mas.index = tmp->vm_start;
 687                 mas.last = tmp->vm_end - 1;
 688                 mas_store(&mas, tmp);
 689                 if (mas_is_err(&mas))
 690                         goto fail_nomem_mas_store;
 691
 692                 mm->map_count++;
 693                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 694                         retval = copy_page_range(tmp, mpnt);
 695
 696                 if (tmp->vm_ops && tmp->vm_ops->open)
 697                         tmp->vm_ops->open(tmp);
 698
 699                 if (retval)
 700                         goto loop_out;
 701         }
 702         /* a new mm has just been created */
 703         retval = arch_dup_mmap(oldmm, mm);
 704 loop_out:
 705         mas_destroy(&mas);
 706 out:
 707         mmap_write_unlock(mm);
 708         flush_tlb_mm(oldmm);
 709         mmap_write_unlock(oldmm);
 710         dup_userfaultfd_complete(&uf);
 711 fail_uprobe_end:
 712         uprobe_end_dup_mmap();
 713         return retval;
 714
 715 fail_nomem_mas_store:
 716         unlink_anon_vmas(tmp);
 717 fail_nomem_anon_vma_fork:
 718         mpol_put(vma_policy(tmp));
 719 fail_nomem_policy:
 720         vm_area_free(tmp);
 721 fail_nomem:
 722         retval = -ENOMEM;
 723         vm_unacct_memory(charge);
 724         goto loop_out;
 725 }
 726
 727 static inline int mm_alloc_pgd(struct mm_struct *mm)
 728 {
 729         mm->pgd = pgd_alloc(mm);
 730         if (unlikely(!mm->pgd))
 731                 return -ENOMEM;
 732         return 0;
 733 }
 734
 735 static inline void mm_free_pgd(struct mm_struct *mm)
 736 {
 737         pgd_free(mm, mm->pgd);
 738 }
 739 #else
 740 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 741 {
 742         mmap_write_lock(oldmm);
 743         dup_mm_exe_file(mm, oldmm);
 744         mmap_write_unlock(oldmm);
 745         return 0;
 746 }
 747 #define mm_alloc_pgd(mm)        (0)
 748 #define mm_free_pgd(mm)
 749 #endif /* CONFIG_MMU */
 750
 751 static void check_mm(struct mm_struct *mm)
 752 {
 753         int i;
 754
 755         BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
 756                          "Please make sure 'struct resident_page_types[]' is updated as well");
 757
 758         for (i = 0; i < NR_MM_COUNTERS; i++) {
 759                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 760
 761                 if (unlikely(x))
 762                         pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
 763                                  mm, resident_page_types[i], x);
 764         }
 765
 766         if (mm_pgtables_bytes(mm))
 767                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 768                                 mm_pgtables_bytes(mm));
 769
 770 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 771         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 772 #endif
 773 }
 774
 775 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 776 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 777
 778 /*
 779  * Called when the last reference to the mm
 780  * is dropped: either by a lazy thread or by
 781  * mmput. Free the page directory and the mm.
 782  */
 783 void __mmdrop(struct mm_struct *mm)
 784 {
 785         BUG_ON(mm == &init_mm);
 786         WARN_ON_ONCE(mm == current->mm);
 787         WARN_ON_ONCE(mm == current->active_mm);
 788         mm_free_pgd(mm);
 789         destroy_context(mm);
 790         mmu_notifier_subscriptions_destroy(mm);
 791         check_mm(mm);
 792         put_user_ns(mm->user_ns);
 793         mm_pasid_drop(mm);
 794         free_mm(mm);
 795 }
 796 EXPORT_SYMBOL_GPL(__mmdrop);
 797
 798 static void mmdrop_async_fn(struct work_struct *work)
 799 {
 800         struct mm_struct *mm;
 801
 802         mm = container_of(work, struct mm_struct, async_put_work);
 803         __mmdrop(mm);
 804 }
 805
 806 static void mmdrop_async(struct mm_struct *mm)
 807 {
 808         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 809                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 810                 schedule_work(&mm->async_put_work);
 811         }
 812 }
 813
 814 static inline void free_signal_struct(struct signal_struct *sig)
 815 {
 816         taskstats_tgid_free(sig);
 817         sched_autogroup_exit(sig);
 818         /*
 819          * __mmdrop is not safe to call from softirq context on x86 due to
 820          * pgd_dtor so postpone it to the async context
 821          */
 822         if (sig->oom_mm)
 823                 mmdrop_async(sig->oom_mm);
 824         kmem_cache_free(signal_cachep, sig);
 825 }
 826
 827 static inline void put_signal_struct(struct signal_struct *sig)
 828 {
 829         if (refcount_dec_and_test(&sig->sigcnt))
 830                 free_signal_struct(sig);
 831 }
 832
 833 void __put_task_struct(struct task_struct *tsk)
 834 {
 835         WARN_ON(!tsk->exit_state);
 836         WARN_ON(refcount_read(&tsk->usage));
 837         WARN_ON(tsk == current);
 838
 839         io_uring_free(tsk);
 840         cgroup_free(tsk);
 841         task_numa_free(tsk, true);
 842         security_task_free(tsk);
 843         bpf_task_storage_free(tsk);
 844         exit_creds(tsk);
 845         delayacct_tsk_free(tsk);
 846         put_signal_struct(tsk->signal);
 847         sched_core_free(tsk);
 848         free_task(tsk);
 849 }
 850 EXPORT_SYMBOL_GPL(__put_task_struct);
 851
 852 void __init __weak arch_task_cache_init(void) { }
 853
 854 /*
 855  * set_max_threads
 856  */
 857 static void set_max_threads(unsigned int max_threads_suggested)
 858 {
 859         u64 threads;
 860         unsigned long nr_pages = totalram_pages();
 861
 862         /*
 863          * The number of threads shall be limited such that the thread
 864          * structures may only consume a small part of the available memory.
 865          */
 866         if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
 867                 threads = MAX_THREADS;
 868         else
 869                 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
 870                                     (u64) THREAD_SIZE * 8UL);
 871
 872         if (threads > max_threads_suggested)
 873                 threads = max_threads_suggested;
 874
 875         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 876 }
 877
 878 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 879 /* Initialized by the architecture: */
 880 int arch_task_struct_size __read_mostly;
 881 #endif
 882
 883 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 884 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 885 {
 886         /* Fetch thread_struct whitelist for the architecture. */
 887         arch_thread_struct_whitelist(offset, size);
 888
 889         /*
 890          * Handle zero-sized whitelist or empty thread_struct, otherwise
 891          * adjust offset to position of thread_struct in task_struct.
 892          */
 893         if (unlikely(*size == 0))
 894                 *offset = 0;
 895         else
 896                 *offset += offsetof(struct task_struct, thread);
 897 }
 898 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
 899
 900 void __init fork_init(void)
 901 {
 902         int i;
 903 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 904 #ifndef ARCH_MIN_TASKALIGN
 905 #define ARCH_MIN_TASKALIGN      0
 906 #endif
 907         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 908         unsigned long useroffset, usersize;
 909
 910         /* create a slab on which task_structs can be allocated */
 911         task_struct_whitelist(&useroffset, &usersize);
 912         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 913                         arch_task_struct_size, align,
 914                         SLAB_PANIC|SLAB_ACCOUNT,
 915                         useroffset, usersize, NULL);
 916 #endif
 917
 918         /* do the arch specific task caches init */
 919         arch_task_cache_init();
 920
 921         set_max_threads(MAX_THREADS);
 922
 923         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 924         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 925         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 926                 init_task.signal->rlim[RLIMIT_NPROC];
 927
 928         for (i = 0; i < UCOUNT_COUNTS; i++)
 929                 init_user_ns.ucount_max[i] = max_threads/2;
 930
 931         set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC,      RLIM_INFINITY);
 932         set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE,   RLIM_INFINITY);
 933         set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
 934         set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK,    RLIM_INFINITY);
 935
 936 #ifdef CONFIG_VMAP_STACK
 937         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 938                           NULL, free_vm_stack_cache);
 939 #endif
 940
 941         scs_init();
 942
 943         lockdep_init_task(&init_task);
 944         uprobes_init();
 945 }
 946
 947 int __weak arch_dup_task_struct(struct task_struct *dst,
 948                                                struct task_struct *src)
 949 {
 950         *dst = *src;
 951         return 0;
 952 }
 953
 954 void set_task_stack_end_magic(struct task_struct *tsk)
 955 {
 956         unsigned long *stackend;
 957
 958         stackend = end_of_stack(tsk);
 959         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 960 }
 961
 962 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 963 {
 964         struct task_struct *tsk;
 965         int err;
 966
 967         if (node == NUMA_NO_NODE)
 968                 node = tsk_fork_get_node(orig);
 969         tsk = alloc_task_struct_node(node);
 970         if (!tsk)
 971                 return NULL;
 972
 973         err = arch_dup_task_struct(tsk, orig);
 974         if (err)
 975                 goto free_tsk;
 976
 977         err = alloc_thread_stack_node(tsk, node);
 978         if (err)
 979                 goto free_tsk;
 980
 981 #ifdef CONFIG_THREAD_INFO_IN_TASK
 982         refcount_set(&tsk->stack_refcount, 1);
 983 #endif
 984         account_kernel_stack(tsk, 1);
 985
 986         err = scs_prepare(tsk, node);
 987         if (err)
 988                 goto free_stack;
 989
 990 #ifdef CONFIG_SECCOMP
 991         /*
 992          * We must handle setting up seccomp filters once we're under
 993          * the sighand lock in case orig has changed between now and
 994          * then. Until then, filter must be NULL to avoid messing up
 995          * the usage counts on the error path calling free_task.
 996          */
 997         tsk->seccomp.filter = NULL;
 998 #endif
 999
1000         setup_thread_stack(tsk, orig);
1001         clear_user_return_notifier(tsk);
1002         clear_tsk_need_resched(tsk);
1003         set_task_stack_end_magic(tsk);
1004         clear_syscall_work_syscall_user_dispatch(tsk);
1005
1006 #ifdef CONFIG_STACKPROTECTOR
1007         tsk->stack_canary = get_random_canary();
1008 #endif
1009         if (orig->cpus_ptr == &orig->cpus_mask)
1010                 tsk->cpus_ptr = &tsk->cpus_mask;
1011         dup_user_cpus_ptr(tsk, orig, node);
1012
1013         /*
1014          * One for the user space visible state that goes away when reaped.
1015          * One for the scheduler.
1016          */
1017         refcount_set(&tsk->rcu_users, 2);
1018         /* One for the rcu users */
1019         refcount_set(&tsk->usage, 1);
1020 #ifdef CONFIG_BLK_DEV_IO_TRACE
1021         tsk->btrace_seq = 0;
1022 #endif
1023         tsk->splice_pipe = NULL;
1024         tsk->task_frag.page = NULL;
1025         tsk->wake_q.next = NULL;
1026         tsk->worker_private = NULL;
1027
1028         kcov_task_init(tsk);
1029         kmsan_task_create(tsk);
1030         kmap_local_fork(tsk);
1031
1032 #ifdef CONFIG_FAULT_INJECTION
1033         tsk->fail_nth = 0;
1034 #endif
1035
1036 #ifdef CONFIG_BLK_CGROUP
1037         tsk->throttle_queue = NULL;
1038         tsk->use_memdelay = 0;
1039 #endif
1040
1041 #ifdef CONFIG_IOMMU_SVA
1042         tsk->pasid_activated = 0;
1043 #endif
1044
1045 #ifdef CONFIG_MEMCG
1046         tsk->active_memcg = NULL;
1047 #endif
1048
1049 #ifdef CONFIG_CPU_SUP_INTEL
1050         tsk->reported_split_lock = 0;
1051 #endif
1052
1053         return tsk;
1054
1055 free_stack:
1056         exit_task_stack_account(tsk);
1057         free_thread_stack(tsk);
1058 free_tsk:
1059         free_task_struct(tsk);
1060         return NULL;
1061 }
1062
1063 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1064
1065 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1066
1067 static int __init coredump_filter_setup(char *s)
1068 {
1069         default_dump_filter =
1070                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1071                 MMF_DUMP_FILTER_MASK;
1072         return 1;
1073 }
1074
1075 __setup("coredump_filter=", coredump_filter_setup);
1076
1077 #include <linux/init_task.h>
1078
1079 static void mm_init_aio(struct mm_struct *mm)
1080 {
1081 #ifdef CONFIG_AIO
1082         spin_lock_init(&mm->ioctx_lock);
1083         mm->ioctx_table = NULL;
1084 #endif
1085 }
1086
1087 static __always_inline void mm_clear_owner(struct mm_struct *mm,
1088                                            struct task_struct *p)
1089 {
1090 #ifdef CONFIG_MEMCG
1091         if (mm->owner == p)
1092                 WRITE_ONCE(mm->owner, NULL);
1093 #endif
1094 }
1095
1096 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1097 {
1098 #ifdef CONFIG_MEMCG
1099         mm->owner = p;
1100 #endif
1101 }
1102
1103 static void mm_init_uprobes_state(struct mm_struct *mm)
1104 {
1105 #ifdef CONFIG_UPROBES
1106         mm->uprobes_state.xol_area = NULL;
1107 #endif
1108 }
1109
1110 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1111         struct user_namespace *user_ns)
1112 {
1113         mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1114         mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1115         atomic_set(&mm->mm_users, 1);
1116         atomic_set(&mm->mm_count, 1);
1117         seqcount_init(&mm->write_protect_seq);
1118         mmap_init_lock(mm);
1119         INIT_LIST_HEAD(&mm->mmlist);
1120         mm_pgtables_bytes_init(mm);
1121         mm->map_count = 0;
1122         mm->locked_vm = 0;
1123         atomic64_set(&mm->pinned_vm, 0);
1124         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1125         spin_lock_init(&mm->page_table_lock);
1126         spin_lock_init(&mm->arg_lock);
1127         mm_init_cpumask(mm);
1128         mm_init_aio(mm);
1129         mm_init_owner(mm, p);
1130         mm_pasid_init(mm);
1131         RCU_INIT_POINTER(mm->exe_file, NULL);
1132         mmu_notifier_subscriptions_init(mm);
1133         init_tlb_flush_pending(mm);
1134 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1135         mm->pmd_huge_pte = NULL;
1136 #endif
1137         mm_init_uprobes_state(mm);
1138         hugetlb_count_init(mm);
1139
1140         if (current->mm) {
1141                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1142                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1143         } else {
1144                 mm->flags = default_dump_filter;
1145                 mm->def_flags = 0;
1146         }
1147
1148         if (mm_alloc_pgd(mm))
1149                 goto fail_nopgd;
1150
1151         if (init_new_context(p, mm))
1152                 goto fail_nocontext;
1153
1154         mm->user_ns = get_user_ns(user_ns);
1155         lru_gen_init_mm(mm);
1156         return mm;
1157
1158 fail_nocontext:
1159         mm_free_pgd(mm);
1160 fail_nopgd:
1161         free_mm(mm);
1162         return NULL;
1163 }
1164
1165 /*
1166  * Allocate and initialize an mm_struct.
1167  */
1168 struct mm_struct *mm_alloc(void)
1169 {
1170         struct mm_struct *mm;
1171
1172         mm = allocate_mm();
1173         if (!mm)
1174                 return NULL;
1175
1176         memset(mm, 0, sizeof(*mm));
1177         return mm_init(mm, current, current_user_ns());
1178 }
1179
1180 static inline void __mmput(struct mm_struct *mm)
1181 {
1182         VM_BUG_ON(atomic_read(&mm->mm_users));
1183
1184         uprobe_clear_state(mm);
1185         exit_aio(mm);
1186         ksm_exit(mm);
1187         khugepaged_exit(mm); /* must run before exit_mmap */
1188         exit_mmap(mm);
1189         mm_put_huge_zero_page(mm);
1190         set_mm_exe_file(mm, NULL);
1191         if (!list_empty(&mm->mmlist)) {
1192                 spin_lock(&mmlist_lock);
1193                 list_del(&mm->mmlist);
1194                 spin_unlock(&mmlist_lock);
1195         }
1196         if (mm->binfmt)
1197                 module_put(mm->binfmt->module);
1198         lru_gen_del_mm(mm);
1199         mmdrop(mm);
1200 }
1201
1202 /*
1203  * Decrement the use count and release all resources for an mm.
1204  */
1205 void mmput(struct mm_struct *mm)
1206 {
1207         might_sleep();
1208
1209         if (atomic_dec_and_test(&mm->mm_users))
1210                 __mmput(mm);
1211 }
1212 EXPORT_SYMBOL_GPL(mmput);
1213
1214 #ifdef CONFIG_MMU
1215 static void mmput_async_fn(struct work_struct *work)
1216 {
1217         struct mm_struct *mm = container_of(work, struct mm_struct,
1218                                             async_put_work);
1219
1220         __mmput(mm);
1221 }
1222
1223 void mmput_async(struct mm_struct *mm)
1224 {
1225         if (atomic_dec_and_test(&mm->mm_users)) {
1226                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1227                 schedule_work(&mm->async_put_work);
1228         }
1229 }
1230 EXPORT_SYMBOL_GPL(mmput_async);
1231 #endif
1232
1233 /**
1234  * set_mm_exe_file - change a reference to the mm's executable file
1235  *
1236  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1237  *
1238  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1239  * invocations: in mmput() nobody alive left, in execve task is single
1240  * threaded.
1241  *
1242  * Can only fail if new_exe_file != NULL.
1243  */
1244 int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1245 {
1246         struct file *old_exe_file;
1247
1248         /*
1249          * It is safe to dereference the exe_file without RCU as
1250          * this function is only called if nobody else can access
1251          * this mm -- see comment above for justification.
1252          */
1253         old_exe_file = rcu_dereference_raw(mm->exe_file);
1254
1255         if (new_exe_file) {
1256                 /*
1257                  * We expect the caller (i.e., sys_execve) to already denied
1258                  * write access, so this is unlikely to fail.
1259                  */
1260                 if (unlikely(deny_write_access(new_exe_file)))
1261                         return -EACCES;
1262                 get_file(new_exe_file);
1263         }
1264         rcu_assign_pointer(mm->exe_file, new_exe_file);
1265         if (old_exe_file) {
1266                 allow_write_access(old_exe_file);
1267                 fput(old_exe_file);
1268         }
1269         return 0;
1270 }
1271
1272 /**
1273  * replace_mm_exe_file - replace a reference to the mm's executable file
1274  *
1275  * This changes mm's executable file (shown as symlink /proc/[pid]/exe),
1276  * dealing with concurrent invocation and without grabbing the mmap lock in
1277  * write mode.
1278  *
1279  * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1280  */
1281 int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1282 {
1283         struct vm_area_struct *vma;
1284         struct file *old_exe_file;
1285         int ret = 0;
1286
1287         /* Forbid mm->exe_file change if old file still mapped. */
1288         old_exe_file = get_mm_exe_file(mm);
1289         if (old_exe_file) {
1290                 VMA_ITERATOR(vmi, mm, 0);
1291                 mmap_read_lock(mm);
1292                 for_each_vma(vmi, vma) {
1293                         if (!vma->vm_file)
1294                                 continue;
1295                         if (path_equal(&vma->vm_file->f_path,
1296                                        &old_exe_file->f_path)) {
1297                                 ret = -EBUSY;
1298                                 break;
1299                         }
1300                 }
1301                 mmap_read_unlock(mm);
1302                 fput(old_exe_file);
1303                 if (ret)
1304                         return ret;
1305         }
1306
1307         /* set the new file, lockless */
1308         ret = deny_write_access(new_exe_file);
1309         if (ret)
1310                 return -EACCES;
1311         get_file(new_exe_file);
1312
1313         old_exe_file = xchg(&mm->exe_file, new_exe_file);
1314         if (old_exe_file) {
1315                 /*
1316                  * Don't race with dup_mmap() getting the file and disallowing
1317                  * write access while someone might open the file writable.
1318                  */
1319                 mmap_read_lock(mm);
1320                 allow_write_access(old_exe_file);
1321                 fput(old_exe_file);
1322                 mmap_read_unlock(mm);
1323         }
1324         return 0;
1325 }
1326
1327 /**
1328  * get_mm_exe_file - acquire a reference to the mm's executable file
1329  *
1330  * Returns %NULL if mm has no associated executable file.
1331  * User must release file via fput().
1332  */
1333 struct file *get_mm_exe_file(struct mm_struct *mm)
1334 {
1335         struct file *exe_file;
1336
1337         rcu_read_lock();
1338         exe_file = rcu_dereference(mm->exe_file);
1339         if (exe_file && !get_file_rcu(exe_file))
1340                 exe_file = NULL;
1341         rcu_read_unlock();
1342         return exe_file;
1343 }
1344
1345 /**
1346  * get_task_exe_file - acquire a reference to the task's executable file
1347  *
1348  * Returns %NULL if task's mm (if any) has no associated executable file or
1349  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1350  * User must release file via fput().
1351  */
1352 struct file *get_task_exe_file(struct task_struct *task)
1353 {
1354         struct file *exe_file = NULL;
1355         struct mm_struct *mm;
1356
1357         task_lock(task);
1358         mm = task->mm;
1359         if (mm) {
1360                 if (!(task->flags & PF_KTHREAD))
1361                         exe_file = get_mm_exe_file(mm);
1362         }
1363         task_unlock(task);
1364         return exe_file;
1365 }
1366
1367 /**
1368  * get_task_mm - acquire a reference to the task's mm
1369  *
1370  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1371  * this kernel workthread has transiently adopted a user mm with use_mm,
1372  * to do its AIO) is not set and if so returns a reference to it, after
1373  * bumping up the use count.  User must release the mm via mmput()
1374  * after use.  Typically used by /proc and ptrace.
1375  */
1376 struct mm_struct *get_task_mm(struct task_struct *task)
1377 {
1378         struct mm_struct *mm;
1379
1380         task_lock(task);
1381         mm = task->mm;
1382         if (mm) {
1383                 if (task->flags & PF_KTHREAD)
1384                         mm = NULL;
1385                 else
1386                         mmget(mm);
1387         }
1388         task_unlock(task);
1389         return mm;
1390 }
1391 EXPORT_SYMBOL_GPL(get_task_mm);
1392
1393 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1394 {
1395         struct mm_struct *mm;
1396         int err;
1397
1398         err =  down_read_killable(&task->signal->exec_update_lock);
1399         if (err)
1400                 return ERR_PTR(err);
1401
1402         mm = get_task_mm(task);
1403         if (mm && mm != current->mm &&
1404                         !ptrace_may_access(task, mode)) {
1405                 mmput(mm);
1406                 mm = ERR_PTR(-EACCES);
1407         }
1408         up_read(&task->signal->exec_update_lock);
1409
1410         return mm;
1411 }
1412
1413 static void complete_vfork_done(struct task_struct *tsk)
1414 {
1415         struct completion *vfork;
1416
1417         task_lock(tsk);
1418         vfork = tsk->vfork_done;
1419         if (likely(vfork)) {
1420                 tsk->vfork_done = NULL;
1421                 complete(vfork);
1422         }
1423         task_unlock(tsk);
1424 }
1425
1426 static int wait_for_vfork_done(struct task_struct *child,
1427                                 struct completion *vfork)
1428 {
1429         unsigned int state = TASK_UNINTERRUPTIBLE|TASK_KILLABLE|TASK_FREEZABLE;
1430         int killed;
1431
1432         cgroup_enter_frozen();
1433         killed = wait_for_completion_state(vfork, state);
1434         cgroup_leave_frozen(false);
1435
1436         if (killed) {
1437                 task_lock(child);
1438                 child->vfork_done = NULL;
1439                 task_unlock(child);
1440         }
1441
1442         put_task_struct(child);
1443         return killed;
1444 }
1445
1446 /* Please note the differences between mmput and mm_release.
1447  * mmput is called whenever we stop holding onto a mm_struct,
1448  * error success whatever.
1449  *
1450  * mm_release is called after a mm_struct has been removed
1451  * from the current process.
1452  *
1453  * This difference is important for error handling, when we
1454  * only half set up a mm_struct for a new process and need to restore
1455  * the old one.  Because we mmput the new mm_struct before
1456  * restoring the old one. . .
1457  * Eric Biederman 10 January 1998
1458  */
1459 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1460 {
1461         uprobe_free_utask(tsk);
1462
1463         /* Get rid of any cached register state */
1464         deactivate_mm(tsk, mm);
1465
1466         /*
1467          * Signal userspace if we're not exiting with a core dump
1468          * because we want to leave the value intact for debugging
1469          * purposes.
1470          */
1471         if (tsk->clear_child_tid) {
1472                 if (atomic_read(&mm->mm_users) > 1) {
1473                         /*
1474                          * We don't check the error code - if userspace has
1475                          * not set up a proper pointer then tough luck.
1476                          */
1477                         put_user(0, tsk->clear_child_tid);
1478                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1479                                         1, NULL, NULL, 0, 0);
1480                 }
1481                 tsk->clear_child_tid = NULL;
1482         }
1483
1484         /*
1485          * All done, finally we can wake up parent and return this mm to him.
1486          * Also kthread_stop() uses this completion for synchronization.
1487          */
1488         if (tsk->vfork_done)
1489                 complete_vfork_done(tsk);
1490 }
1491
1492 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1493 {
1494         futex_exit_release(tsk);
1495         mm_release(tsk, mm);
1496 }
1497
1498 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1499 {
1500         futex_exec_release(tsk);
1501         mm_release(tsk, mm);
1502 }
1503
1504 /**
1505  * dup_mm() - duplicates an existing mm structure
1506  * @tsk: the task_struct with which the new mm will be associated.
1507  * @oldmm: the mm to duplicate.
1508  *
1509  * Allocates a new mm structure and duplicates the provided @oldmm structure
1510  * content into it.
1511  *
1512  * Return: the duplicated mm or NULL on failure.
1513  */
1514 static struct mm_struct *dup_mm(struct task_struct *tsk,
1515                                 struct mm_struct *oldmm)
1516 {
1517         struct mm_struct *mm;
1518         int err;
1519
1520         mm = allocate_mm();
1521         if (!mm)
1522                 goto fail_nomem;
1523
1524         memcpy(mm, oldmm, sizeof(*mm));
1525
1526         if (!mm_init(mm, tsk, mm->user_ns))
1527                 goto fail_nomem;
1528
1529         err = dup_mmap(mm, oldmm);
1530         if (err)
1531                 goto free_pt;
1532
1533         mm->hiwater_rss = get_mm_rss(mm);
1534         mm->hiwater_vm = mm->total_vm;
1535
1536         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1537                 goto free_pt;
1538
1539         return mm;
1540
1541 free_pt:
1542         /* don't put binfmt in mmput, we haven't got module yet */
1543         mm->binfmt = NULL;
1544         mm_init_owner(mm, NULL);
1545         mmput(mm);
1546
1547 fail_nomem:
1548         return NULL;
1549 }
1550
1551 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1552 {
1553         struct mm_struct *mm, *oldmm;
1554
1555         tsk->min_flt = tsk->maj_flt = 0;
1556         tsk->nvcsw = tsk->nivcsw = 0;
1557 #ifdef CONFIG_DETECT_HUNG_TASK
1558         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1559         tsk->last_switch_time = 0;
1560 #endif
1561
1562         tsk->mm = NULL;
1563         tsk->active_mm = NULL;
1564
1565         /*
1566          * Are we cloning a kernel thread?
1567          *
1568          * We need to steal a active VM for that..
1569          */
1570         oldmm = current->mm;
1571         if (!oldmm)
1572                 return 0;
1573
1574         if (clone_flags & CLONE_VM) {
1575                 mmget(oldmm);
1576                 mm = oldmm;
1577         } else {
1578                 mm = dup_mm(tsk, current->mm);
1579                 if (!mm)
1580                         return -ENOMEM;
1581         }
1582
1583         tsk->mm = mm;
1584         tsk->active_mm = mm;
1585         return 0;
1586 }
1587
1588 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1589 {
1590         struct fs_struct *fs = current->fs;
1591         if (clone_flags & CLONE_FS) {
1592                 /* tsk->fs is already what we want */
1593                 spin_lock(&fs->lock);
1594                 if (fs->in_exec) {
1595                         spin_unlock(&fs->lock);
1596                         return -EAGAIN;
1597                 }
1598                 fs->users++;
1599                 spin_unlock(&fs->lock);
1600                 return 0;
1601         }
1602         tsk->fs = copy_fs_struct(fs);
1603         if (!tsk->fs)
1604                 return -ENOMEM;
1605         return 0;
1606 }
1607
1608 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1609 {
1610         struct files_struct *oldf, *newf;
1611         int error = 0;
1612
1613         /*
1614          * A background process may not have any files ...
1615          */
1616         oldf = current->files;
1617         if (!oldf)
1618                 goto out;
1619
1620         if (clone_flags & CLONE_FILES) {
1621                 atomic_inc(&oldf->count);
1622                 goto out;
1623         }
1624
1625         newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1626         if (!newf)
1627                 goto out;
1628
1629         tsk->files = newf;
1630         error = 0;
1631 out:
1632         return error;
1633 }
1634
1635 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1636 {
1637         struct sighand_struct *sig;
1638
1639         if (clone_flags & CLONE_SIGHAND) {
1640                 refcount_inc(&current->sighand->count);
1641                 return 0;
1642         }
1643         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1644         RCU_INIT_POINTER(tsk->sighand, sig);
1645         if (!sig)
1646                 return -ENOMEM;
1647
1648         refcount_set(&sig->count, 1);
1649         spin_lock_irq(&current->sighand->siglock);
1650         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1651         spin_unlock_irq(&current->sighand->siglock);
1652
1653         /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1654         if (clone_flags & CLONE_CLEAR_SIGHAND)
1655                 flush_signal_handlers(tsk, 0);
1656
1657         return 0;
1658 }
1659
1660 void __cleanup_sighand(struct sighand_struct *sighand)
1661 {
1662         if (refcount_dec_and_test(&sighand->count)) {
1663                 signalfd_cleanup(sighand);
1664                 /*
1665                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1666                  * without an RCU grace period, see __lock_task_sighand().
1667                  */
1668                 kmem_cache_free(sighand_cachep, sighand);
1669         }
1670 }
1671
1672 /*
1673  * Initialize POSIX timer handling for a thread group.
1674  */
1675 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1676 {
1677         struct posix_cputimers *pct = &sig->posix_cputimers;
1678         unsigned long cpu_limit;
1679
1680         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1681         posix_cputimers_group_init(pct, cpu_limit);
1682 }
1683
1684 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1685 {
1686         struct signal_struct *sig;
1687
1688         if (clone_flags & CLONE_THREAD)
1689                 return 0;
1690
1691         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1692         tsk->signal = sig;
1693         if (!sig)
1694                 return -ENOMEM;
1695
1696         sig->nr_threads = 1;
1697         sig->quick_threads = 1;
1698         atomic_set(&sig->live, 1);
1699         refcount_set(&sig->sigcnt, 1);
1700
1701         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1702         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1703         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1704
1705         init_waitqueue_head(&sig->wait_chldexit);
1706         sig->curr_target = tsk;
1707         init_sigpending(&sig->shared_pending);
1708         INIT_HLIST_HEAD(&sig->multiprocess);
1709         seqlock_init(&sig->stats_lock);
1710         prev_cputime_init(&sig->prev_cputime);
1711
1712 #ifdef CONFIG_POSIX_TIMERS
1713         INIT_LIST_HEAD(&sig->posix_timers);
1714         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1715         sig->real_timer.function = it_real_fn;
1716 #endif
1717
1718         task_lock(current->group_leader);
1719         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1720         task_unlock(current->group_leader);
1721
1722         posix_cpu_timers_init_group(sig);
1723
1724         tty_audit_fork(sig);
1725         sched_autogroup_fork(sig);
1726
1727         sig->oom_score_adj = current->signal->oom_score_adj;
1728         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1729
1730         mutex_init(&sig->cred_guard_mutex);
1731         init_rwsem(&sig->exec_update_lock);
1732
1733         return 0;
1734 }
1735
1736 static void copy_seccomp(struct task_struct *p)
1737 {
1738 #ifdef CONFIG_SECCOMP
1739         /*
1740          * Must be called with sighand->lock held, which is common to
1741          * all threads in the group. Holding cred_guard_mutex is not
1742          * needed because this new task is not yet running and cannot
1743          * be racing exec.
1744          */
1745         assert_spin_locked(&current->sighand->siglock);
1746
1747         /* Ref-count the new filter user, and assign it. */
1748         get_seccomp_filter(current);
1749         p->seccomp = current->seccomp;
1750
1751         /*
1752          * Explicitly enable no_new_privs here in case it got set
1753          * between the task_struct being duplicated and holding the
1754          * sighand lock. The seccomp state and nnp must be in sync.
1755          */
1756         if (task_no_new_privs(current))
1757                 task_set_no_new_privs(p);
1758
1759         /*
1760          * If the parent gained a seccomp mode after copying thread
1761          * flags and between before we held the sighand lock, we have
1762          * to manually enable the seccomp thread flag here.
1763          */
1764         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1765                 set_task_syscall_work(p, SECCOMP);
1766 #endif
1767 }
1768
1769 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1770 {
1771         current->clear_child_tid = tidptr;
1772
1773         return task_pid_vnr(current);
1774 }
1775
1776 static void rt_mutex_init_task(struct task_struct *p)
1777 {
1778         raw_spin_lock_init(&p->pi_lock);
1779 #ifdef CONFIG_RT_MUTEXES
1780         p->pi_waiters = RB_ROOT_CACHED;
1781         p->pi_top_task = NULL;
1782         p->pi_blocked_on = NULL;
1783 #endif
1784 }
1785
1786 static inline void init_task_pid_links(struct task_struct *task)
1787 {
1788         enum pid_type type;
1789
1790         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1791                 INIT_HLIST_NODE(&task->pid_links[type]);
1792 }
1793
1794 static inline void
1795 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1796 {
1797         if (type == PIDTYPE_PID)
1798                 task->thread_pid = pid;
1799         else
1800                 task->signal->pids[type] = pid;
1801 }
1802
1803 static inline void rcu_copy_process(struct task_struct *p)
1804 {
1805 #ifdef CONFIG_PREEMPT_RCU
1806         p->rcu_read_lock_nesting = 0;
1807         p->rcu_read_unlock_special.s = 0;
1808         p->rcu_blocked_node = NULL;
1809         INIT_LIST_HEAD(&p->rcu_node_entry);
1810 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1811 #ifdef CONFIG_TASKS_RCU
1812         p->rcu_tasks_holdout = false;
1813         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1814         p->rcu_tasks_idle_cpu = -1;
1815 #endif /* #ifdef CONFIG_TASKS_RCU */
1816 #ifdef CONFIG_TASKS_TRACE_RCU
1817         p->trc_reader_nesting = 0;
1818         p->trc_reader_special.s = 0;
1819         INIT_LIST_HEAD(&p->trc_holdout_list);
1820         INIT_LIST_HEAD(&p->trc_blkd_node);
1821 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1822 }
1823
1824 struct pid *pidfd_pid(const struct file *file)
1825 {
1826         if (file->f_op == &pidfd_fops)
1827                 return file->private_data;
1828
1829         return ERR_PTR(-EBADF);
1830 }
1831
1832 static int pidfd_release(struct inode *inode, struct file *file)
1833 {
1834         struct pid *pid = file->private_data;
1835
1836         file->private_data = NULL;
1837         put_pid(pid);
1838         return 0;
1839 }
1840
1841 #ifdef CONFIG_PROC_FS
1842 /**
1843  * pidfd_show_fdinfo - print information about a pidfd
1844  * @m: proc fdinfo file
1845  * @f: file referencing a pidfd
1846  *
1847  * Pid:
1848  * This function will print the pid that a given pidfd refers to in the
1849  * pid namespace of the procfs instance.
1850  * If the pid namespace of the process is not a descendant of the pid
1851  * namespace of the procfs instance 0 will be shown as its pid. This is
1852  * similar to calling getppid() on a process whose parent is outside of
1853  * its pid namespace.
1854  *
1855  * NSpid:
1856  * If pid namespaces are supported then this function will also print
1857  * the pid of a given pidfd refers to for all descendant pid namespaces
1858  * starting from the current pid namespace of the instance, i.e. the
1859  * Pid field and the first entry in the NSpid field will be identical.
1860  * If the pid namespace of the process is not a descendant of the pid
1861  * namespace of the procfs instance 0 will be shown as its first NSpid
1862  * entry and no others will be shown.
1863  * Note that this differs from the Pid and NSpid fields in
1864  * /proc/<pid>/status where Pid and NSpid are always shown relative to
1865  * the  pid namespace of the procfs instance. The difference becomes
1866  * obvious when sending around a pidfd between pid namespaces from a
1867  * different branch of the tree, i.e. where no ancestral relation is
1868  * present between the pid namespaces:
1869  * - create two new pid namespaces ns1 and ns2 in the initial pid
1870  *   namespace (also take care to create new mount namespaces in the
1871  *   new pid namespace and mount procfs)
1872  * - create a process with a pidfd in ns1
1873  * - send pidfd from ns1 to ns2
1874  * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1875  *   have exactly one entry, which is 0
1876  */
1877 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1878 {
1879         struct pid *pid = f->private_data;
1880         struct pid_namespace *ns;
1881         pid_t nr = -1;
1882
1883         if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1884                 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1885                 nr = pid_nr_ns(pid, ns);
1886         }
1887
1888         seq_put_decimal_ll(m, "Pid:\t", nr);
1889
1890 #ifdef CONFIG_PID_NS
1891         seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1892         if (nr > 0) {
1893                 int i;
1894
1895                 /* If nr is non-zero it means that 'pid' is valid and that
1896                  * ns, i.e. the pid namespace associated with the procfs
1897                  * instance, is in the pid namespace hierarchy of pid.
1898                  * Start at one below the already printed level.
1899                  */
1900                 for (i = ns->level + 1; i <= pid->level; i++)
1901                         seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1902         }
1903 #endif
1904         seq_putc(m, '\n');
1905 }
1906 #endif
1907
1908 /*
1909  * Poll support for process exit notification.
1910  */
1911 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1912 {
1913         struct pid *pid = file->private_data;
1914         __poll_t poll_flags = 0;
1915
1916         poll_wait(file, &pid->wait_pidfd, pts);
1917
1918         /*
1919          * Inform pollers only when the whole thread group exits.
1920          * If the thread group leader exits before all other threads in the
1921          * group, then poll(2) should block, similar to the wait(2) family.
1922          */
1923         if (thread_group_exited(pid))
1924                 poll_flags = EPOLLIN | EPOLLRDNORM;
1925
1926         return poll_flags;
1927 }
1928
1929 const struct file_operations pidfd_fops = {
1930         .release = pidfd_release,
1931         .poll = pidfd_poll,
1932 #ifdef CONFIG_PROC_FS
1933         .show_fdinfo = pidfd_show_fdinfo,
1934 #endif
1935 };
1936
1937 static void __delayed_free_task(struct rcu_head *rhp)
1938 {
1939         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1940
1941         free_task(tsk);
1942 }
1943
1944 static __always_inline void delayed_free_task(struct task_struct *tsk)
1945 {
1946         if (IS_ENABLED(CONFIG_MEMCG))
1947                 call_rcu(&tsk->rcu, __delayed_free_task);
1948         else
1949                 free_task(tsk);
1950 }
1951
1952 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1953 {
1954         /* Skip if kernel thread */
1955         if (!tsk->mm)
1956                 return;
1957
1958         /* Skip if spawning a thread or using vfork */
1959         if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1960                 return;
1961
1962         /* We need to synchronize with __set_oom_adj */
1963         mutex_lock(&oom_adj_mutex);
1964         set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1965         /* Update the values in case they were changed after copy_signal */
1966         tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1967         tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1968         mutex_unlock(&oom_adj_mutex);
1969 }
1970
1971 #ifdef CONFIG_RV
1972 static void rv_task_fork(struct task_struct *p)
1973 {
1974         int i;
1975
1976         for (i = 0; i < RV_PER_TASK_MONITORS; i++)
1977                 p->rv[i].da_mon.monitoring = false;
1978 }
1979 #else
1980 #define rv_task_fork(p) do {} while (0)
1981 #endif
1982
1983 /*
1984  * This creates a new process as a copy of the old one,
1985  * but does not actually start it yet.
1986  *
1987  * It copies the registers, and all the appropriate
1988  * parts of the process environment (as per the clone
1989  * flags). The actual kick-off is left to the caller.
1990  */
1991 static __latent_entropy struct task_struct *copy_process(
1992                                         struct pid *pid,
1993                                         int trace,
1994                                         int node,
1995                                         struct kernel_clone_args *args)
1996 {
1997         int pidfd = -1, retval;
1998         struct task_struct *p;
1999         struct multiprocess_signals delayed;
2000         struct file *pidfile = NULL;
2001         const u64 clone_flags = args->flags;
2002         struct nsproxy *nsp = current->nsproxy;
2003
2004         /*
2005          * Don't allow sharing the root directory with processes in a different
2006          * namespace
2007          */
2008         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2009                 return ERR_PTR(-EINVAL);
2010
2011         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2012                 return ERR_PTR(-EINVAL);
2013
2014         /*
2015          * Thread groups must share signals as well, and detached threads
2016          * can only be started up within the thread group.
2017          */
2018         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2019                 return ERR_PTR(-EINVAL);
2020
2021         /*
2022          * Shared signal handlers imply shared VM. By way of the above,
2023          * thread groups also imply shared VM. Blocking this case allows
2024          * for various simplifications in other code.
2025          */
2026         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2027                 return ERR_PTR(-EINVAL);
2028
2029         /*
2030          * Siblings of global init remain as zombies on exit since they are
2031          * not reaped by their parent (swapper). To solve this and to avoid
2032          * multi-rooted process trees, prevent global and container-inits
2033          * from creating siblings.
2034          */
2035         if ((clone_flags & CLONE_PARENT) &&
2036                                 current->signal->flags & SIGNAL_UNKILLABLE)
2037                 return ERR_PTR(-EINVAL);
2038
2039         /*
2040          * If the new process will be in a different pid or user namespace
2041          * do not allow it to share a thread group with the forking task.
2042          */
2043         if (clone_flags & CLONE_THREAD) {
2044                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2045                     (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2046                         return ERR_PTR(-EINVAL);
2047         }
2048
2049         if (clone_flags & CLONE_PIDFD) {
2050                 /*
2051                  * - CLONE_DETACHED is blocked so that we can potentially
2052                  *   reuse it later for CLONE_PIDFD.
2053                  * - CLONE_THREAD is blocked until someone really needs it.
2054                  */
2055                 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
2056                         return ERR_PTR(-EINVAL);
2057         }
2058
2059         /*
2060          * Force any signals received before this point to be delivered
2061          * before the fork happens.  Collect up signals sent to multiple
2062          * processes that happen during the fork and delay them so that
2063          * they appear to happen after the fork.
2064          */
2065         sigemptyset(&delayed.signal);
2066         INIT_HLIST_NODE(&delayed.node);
2067
2068         spin_lock_irq(&current->sighand->siglock);
2069         if (!(clone_flags & CLONE_THREAD))
2070                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
2071         recalc_sigpending();
2072         spin_unlock_irq(&current->sighand->siglock);
2073         retval = -ERESTARTNOINTR;
2074         if (task_sigpending(current))
2075                 goto fork_out;
2076
2077         retval = -ENOMEM;
2078         p = dup_task_struct(current, node);
2079         if (!p)
2080                 goto fork_out;
2081         p->flags &= ~PF_KTHREAD;
2082         if (args->kthread)
2083                 p->flags |= PF_KTHREAD;
2084         if (args->io_thread) {
2085                 /*
2086                  * Mark us an IO worker, and block any signal that isn't
2087                  * fatal or STOP
2088                  */
2089                 p->flags |= PF_IO_WORKER;
2090                 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2091         }
2092
2093         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2094         /*
2095          * Clear TID on mm_release()?
2096          */
2097         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2098
2099         ftrace_graph_init_task(p);
2100
2101         rt_mutex_init_task(p);
2102
2103         lockdep_assert_irqs_enabled();
2104 #ifdef CONFIG_PROVE_LOCKING
2105         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2106 #endif
2107         retval = copy_creds(p, clone_flags);
2108         if (retval < 0)
2109                 goto bad_fork_free;
2110
2111         retval = -EAGAIN;
2112         if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2113                 if (p->real_cred->user != INIT_USER &&
2114                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2115                         goto bad_fork_cleanup_count;
2116         }
2117         current->flags &= ~PF_NPROC_EXCEEDED;
2118
2119         /*
2120          * If multiple threads are within copy_process(), then this check
2121          * triggers too late. This doesn't hurt, the check is only there
2122          * to stop root fork bombs.
2123          */
2124         retval = -EAGAIN;
2125         if (data_race(nr_threads >= max_threads))
2126                 goto bad_fork_cleanup_count;
2127
2128         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
2129         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2130         p->flags |= PF_FORKNOEXEC;
2131         INIT_LIST_HEAD(&p->children);
2132         INIT_LIST_HEAD(&p->sibling);
2133         rcu_copy_process(p);
2134         p->vfork_done = NULL;
2135         spin_lock_init(&p->alloc_lock);
2136
2137         init_sigpending(&p->pending);
2138
2139         p->utime = p->stime = p->gtime = 0;
2140 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2141         p->utimescaled = p->stimescaled = 0;
2142 #endif
2143         prev_cputime_init(&p->prev_cputime);
2144
2145 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2146         seqcount_init(&p->vtime.seqcount);
2147         p->vtime.starttime = 0;
2148         p->vtime.state = VTIME_INACTIVE;
2149 #endif
2150
2151 #ifdef CONFIG_IO_URING
2152         p->io_uring = NULL;
2153 #endif
2154
2155 #if defined(SPLIT_RSS_COUNTING)
2156         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2157 #endif
2158
2159         p->default_timer_slack_ns = current->timer_slack_ns;
2160
2161 #ifdef CONFIG_PSI
2162         p->psi_flags = 0;
2163 #endif
2164
2165         task_io_accounting_init(&p->ioac);
2166         acct_clear_integrals(p);
2167
2168         posix_cputimers_init(&p->posix_cputimers);
2169
2170         p->io_context = NULL;
2171         audit_set_context(p, NULL);
2172         cgroup_fork(p);
2173         if (args->kthread) {
2174                 if (!set_kthread_struct(p))
2175                         goto bad_fork_cleanup_delayacct;
2176         }
2177 #ifdef CONFIG_NUMA
2178         p->mempolicy = mpol_dup(p->mempolicy);
2179         if (IS_ERR(p->mempolicy)) {
2180                 retval = PTR_ERR(p->mempolicy);
2181                 p->mempolicy = NULL;
2182                 goto bad_fork_cleanup_delayacct;
2183         }
2184 #endif
2185 #ifdef CONFIG_CPUSETS
2186         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2187         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2188         seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2189 #endif
2190 #ifdef CONFIG_TRACE_IRQFLAGS
2191         memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2192         p->irqtrace.hardirq_disable_ip  = _THIS_IP_;
2193         p->irqtrace.softirq_enable_ip   = _THIS_IP_;
2194         p->softirqs_enabled             = 1;
2195         p->softirq_context              = 0;
2196 #endif
2197
2198         p->pagefault_disabled = 0;
2199
2200 #ifdef CONFIG_LOCKDEP
2201         lockdep_init_task(p);
2202 #endif
2203
2204 #ifdef CONFIG_DEBUG_MUTEXES
2205         p->blocked_on = NULL; /* not blocked yet */
2206 #endif
2207 #ifdef CONFIG_BCACHE
2208         p->sequential_io        = 0;
2209         p->sequential_io_avg    = 0;
2210 #endif
2211 #ifdef CONFIG_BPF_SYSCALL
2212         RCU_INIT_POINTER(p->bpf_storage, NULL);
2213         p->bpf_ctx = NULL;
2214 #endif
2215
2216         /* Perform scheduler related setup. Assign this task to a CPU. */
2217         retval = sched_fork(clone_flags, p);
2218         if (retval)
2219                 goto bad_fork_cleanup_policy;
2220
2221         retval = perf_event_init_task(p, clone_flags);
2222         if (retval)
2223                 goto bad_fork_cleanup_policy;
2224         retval = audit_alloc(p);
2225         if (retval)
2226                 goto bad_fork_cleanup_perf;
2227         /* copy all the process information */
2228         shm_init_task(p);
2229         retval = security_task_alloc(p, clone_flags);
2230         if (retval)
2231                 goto bad_fork_cleanup_audit;
2232         retval = copy_semundo(clone_flags, p);
2233         if (retval)
2234                 goto bad_fork_cleanup_security;
2235         retval = copy_files(clone_flags, p);
2236         if (retval)
2237                 goto bad_fork_cleanup_semundo;
2238         retval = copy_fs(clone_flags, p);
2239         if (retval)
2240                 goto bad_fork_cleanup_files;
2241         retval = copy_sighand(clone_flags, p);
2242         if (retval)
2243                 goto bad_fork_cleanup_fs;
2244         retval = copy_signal(clone_flags, p);
2245         if (retval)
2246                 goto bad_fork_cleanup_sighand;
2247         retval = copy_mm(clone_flags, p);
2248         if (retval)
2249                 goto bad_fork_cleanup_signal;
2250         retval = copy_namespaces(clone_flags, p);
2251         if (retval)
2252                 goto bad_fork_cleanup_mm;
2253         retval = copy_io(clone_flags, p);
2254         if (retval)
2255                 goto bad_fork_cleanup_namespaces;
2256         retval = copy_thread(p, args);
2257         if (retval)
2258                 goto bad_fork_cleanup_io;
2259
2260         stackleak_task_init(p);
2261
2262         if (pid != &init_struct_pid) {
2263                 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2264                                 args->set_tid_size);
2265                 if (IS_ERR(pid)) {
2266                         retval = PTR_ERR(pid);
2267                         goto bad_fork_cleanup_thread;
2268                 }
2269         }
2270
2271         /*
2272          * This has to happen after we've potentially unshared the file
2273          * descriptor table (so that the pidfd doesn't leak into the child
2274          * if the fd table isn't shared).
2275          */
2276         if (clone_flags & CLONE_PIDFD) {
2277                 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2278                 if (retval < 0)
2279                         goto bad_fork_free_pid;
2280
2281                 pidfd = retval;
2282
2283                 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2284                                               O_RDWR | O_CLOEXEC);
2285                 if (IS_ERR(pidfile)) {
2286                         put_unused_fd(pidfd);
2287                         retval = PTR_ERR(pidfile);
2288                         goto bad_fork_free_pid;
2289                 }
2290                 get_pid(pid);   /* held by pidfile now */
2291
2292                 retval = put_user(pidfd, args->pidfd);
2293                 if (retval)
2294                         goto bad_fork_put_pidfd;
2295         }
2296
2297 #ifdef CONFIG_BLOCK
2298         p->plug = NULL;
2299 #endif
2300         futex_init_task(p);
2301
2302         /*
2303          * sigaltstack should be cleared when sharing the same VM
2304          */
2305         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2306                 sas_ss_reset(p);
2307
2308         /*
2309          * Syscall tracing and stepping should be turned off in the
2310          * child regardless of CLONE_PTRACE.
2311          */
2312         user_disable_single_step(p);
2313         clear_task_syscall_work(p, SYSCALL_TRACE);
2314 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2315         clear_task_syscall_work(p, SYSCALL_EMU);
2316 #endif
2317         clear_tsk_latency_tracing(p);
2318
2319         /* ok, now we should be set up.. */
2320         p->pid = pid_nr(pid);
2321         if (clone_flags & CLONE_THREAD) {
2322                 p->group_leader = current->group_leader;
2323                 p->tgid = current->tgid;
2324         } else {
2325                 p->group_leader = p;
2326                 p->tgid = p->pid;
2327         }
2328
2329         p->nr_dirtied = 0;
2330         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2331         p->dirty_paused_when = 0;
2332
2333         p->pdeath_signal = 0;
2334         INIT_LIST_HEAD(&p->thread_group);
2335         p->task_works = NULL;
2336         clear_posix_cputimers_work(p);
2337
2338 #ifdef CONFIG_KRETPROBES
2339         p->kretprobe_instances.first = NULL;
2340 #endif
2341 #ifdef CONFIG_RETHOOK
2342         p->rethooks.first = NULL;
2343 #endif
2344
2345         /*
2346          * Ensure that the cgroup subsystem policies allow the new process to be
2347          * forked. It should be noted that the new process's css_set can be changed
2348          * between here and cgroup_post_fork() if an organisation operation is in
2349          * progress.
2350          */
2351         retval = cgroup_can_fork(p, args);
2352         if (retval)
2353                 goto bad_fork_put_pidfd;
2354
2355         /*
2356          * Now that the cgroups are pinned, re-clone the parent cgroup and put
2357          * the new task on the correct runqueue. All this *before* the task
2358          * becomes visible.
2359          *
2360          * This isn't part of ->can_fork() because while the re-cloning is
2361          * cgroup specific, it unconditionally needs to place the task on a
2362          * runqueue.
2363          */
2364         sched_cgroup_fork(p, args);
2365
2366         /*
2367          * From this point on we must avoid any synchronous user-space
2368          * communication until we take the tasklist-lock. In particular, we do
2369          * not want user-space to be able to predict the process start-time by
2370          * stalling fork(2) after we recorded the start_time but before it is
2371          * visible to the system.
2372          */
2373
2374         p->start_time = ktime_get_ns();
2375         p->start_boottime = ktime_get_boottime_ns();
2376
2377         /*
2378          * Make it visible to the rest of the system, but dont wake it up yet.
2379          * Need tasklist lock for parent etc handling!
2380          */
2381         write_lock_irq(&tasklist_lock);
2382
2383         /* CLONE_PARENT re-uses the old parent */
2384         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2385                 p->real_parent = current->real_parent;
2386                 p->parent_exec_id = current->parent_exec_id;
2387                 if (clone_flags & CLONE_THREAD)
2388                         p->exit_signal = -1;
2389                 else
2390                         p->exit_signal = current->group_leader->exit_signal;
2391         } else {
2392                 p->real_parent = current;
2393                 p->parent_exec_id = current->self_exec_id;
2394                 p->exit_signal = args->exit_signal;
2395         }
2396
2397         klp_copy_process(p);
2398
2399         sched_core_fork(p);
2400
2401         spin_lock(&current->sighand->siglock);
2402
2403         rv_task_fork(p);
2404
2405         rseq_fork(p, clone_flags);
2406
2407         /* Don't start children in a dying pid namespace */
2408         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2409                 retval = -ENOMEM;
2410                 goto bad_fork_cancel_cgroup;
2411         }
2412
2413         /* Let kill terminate clone/fork in the middle */
2414         if (fatal_signal_pending(current)) {
2415                 retval = -EINTR;
2416                 goto bad_fork_cancel_cgroup;
2417         }
2418
2419         /* No more failure paths after this point. */
2420
2421         /*
2422          * Copy seccomp details explicitly here, in case they were changed
2423          * before holding sighand lock.
2424          */
2425         copy_seccomp(p);
2426
2427         init_task_pid_links(p);
2428         if (likely(p->pid)) {
2429                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2430
2431                 init_task_pid(p, PIDTYPE_PID, pid);
2432                 if (thread_group_leader(p)) {
2433                         init_task_pid(p, PIDTYPE_TGID, pid);
2434                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2435                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2436
2437                         if (is_child_reaper(pid)) {
2438                                 ns_of_pid(pid)->child_reaper = p;
2439                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2440                         }
2441                         p->signal->shared_pending.signal = delayed.signal;
2442                         p->signal->tty = tty_kref_get(current->signal->tty);
2443                         /*
2444                          * Inherit has_child_subreaper flag under the same
2445                          * tasklist_lock with adding child to the process tree
2446                          * for propagate_has_child_subreaper optimization.
2447                          */
2448                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2449                                                          p->real_parent->signal->is_child_subreaper;
2450                         list_add_tail(&p->sibling, &p->real_parent->children);
2451                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2452                         attach_pid(p, PIDTYPE_TGID);
2453                         attach_pid(p, PIDTYPE_PGID);
2454                         attach_pid(p, PIDTYPE_SID);
2455                         __this_cpu_inc(process_counts);
2456                 } else {
2457                         current->signal->nr_threads++;
2458                         current->signal->quick_threads++;
2459                         atomic_inc(&current->signal->live);
2460                         refcount_inc(&current->signal->sigcnt);
2461                         task_join_group_stop(p);
2462                         list_add_tail_rcu(&p->thread_group,
2463                                           &p->group_leader->thread_group);
2464                         list_add_tail_rcu(&p->thread_node,
2465                                           &p->signal->thread_head);
2466                 }
2467                 attach_pid(p, PIDTYPE_PID);
2468                 nr_threads++;
2469         }
2470         total_forks++;
2471         hlist_del_init(&delayed.node);
2472         spin_unlock(&current->sighand->siglock);
2473         syscall_tracepoint_update(p);
2474         write_unlock_irq(&tasklist_lock);
2475
2476         if (pidfile)
2477                 fd_install(pidfd, pidfile);
2478
2479         proc_fork_connector(p);
2480         sched_post_fork(p);
2481         cgroup_post_fork(p, args);
2482         perf_event_fork(p);
2483
2484         trace_task_newtask(p, clone_flags);
2485         uprobe_copy_process(p, clone_flags);
2486
2487         copy_oom_score_adj(clone_flags, p);
2488
2489         return p;
2490
2491 bad_fork_cancel_cgroup:
2492         sched_core_free(p);
2493         spin_unlock(&current->sighand->siglock);
2494         write_unlock_irq(&tasklist_lock);
2495         cgroup_cancel_fork(p, args);
2496 bad_fork_put_pidfd:
2497         if (clone_flags & CLONE_PIDFD) {
2498                 fput(pidfile);
2499                 put_unused_fd(pidfd);
2500         }
2501 bad_fork_free_pid:
2502         if (pid != &init_struct_pid)
2503                 free_pid(pid);
2504 bad_fork_cleanup_thread:
2505         exit_thread(p);
2506 bad_fork_cleanup_io:
2507         if (p->io_context)
2508                 exit_io_context(p);
2509 bad_fork_cleanup_namespaces:
2510         exit_task_namespaces(p);
2511 bad_fork_cleanup_mm:
2512         if (p->mm) {
2513                 mm_clear_owner(p->mm, p);
2514                 mmput(p->mm);
2515         }
2516 bad_fork_cleanup_signal:
2517         if (!(clone_flags & CLONE_THREAD))
2518                 free_signal_struct(p->signal);
2519 bad_fork_cleanup_sighand:
2520         __cleanup_sighand(p->sighand);
2521 bad_fork_cleanup_fs:
2522         exit_fs(p); /* blocking */
2523 bad_fork_cleanup_files:
2524         exit_files(p); /* blocking */
2525 bad_fork_cleanup_semundo:
2526         exit_sem(p);
2527 bad_fork_cleanup_security:
2528         security_task_free(p);
2529 bad_fork_cleanup_audit:
2530         audit_free(p);
2531 bad_fork_cleanup_perf:
2532         perf_event_free_task(p);
2533 bad_fork_cleanup_policy:
2534         lockdep_free_task(p);
2535 #ifdef CONFIG_NUMA
2536         mpol_put(p->mempolicy);
2537 #endif
2538 bad_fork_cleanup_delayacct:
2539         delayacct_tsk_free(p);
2540 bad_fork_cleanup_count:
2541         dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2542         exit_creds(p);
2543 bad_fork_free:
2544         WRITE_ONCE(p->__state, TASK_DEAD);
2545         exit_task_stack_account(p);
2546         put_task_stack(p);
2547         delayed_free_task(p);
2548 fork_out:
2549         spin_lock_irq(&current->sighand->siglock);
2550         hlist_del_init(&delayed.node);
2551         spin_unlock_irq(&current->sighand->siglock);
2552         return ERR_PTR(retval);
2553 }
2554
2555 static inline void init_idle_pids(struct task_struct *idle)
2556 {
2557         enum pid_type type;
2558
2559         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2560                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2561                 init_task_pid(idle, type, &init_struct_pid);
2562         }
2563 }
2564
2565 static int idle_dummy(void *dummy)
2566 {
2567         /* This function is never called */
2568         return 0;
2569 }
2570
2571 struct task_struct * __init fork_idle(int cpu)
2572 {
2573         struct task_struct *task;
2574         struct kernel_clone_args args = {
2575                 .flags          = CLONE_VM,
2576                 .fn             = &idle_dummy,
2577                 .fn_arg         = NULL,
2578                 .kthread        = 1,
2579                 .idle           = 1,
2580         };
2581
2582         task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2583         if (!IS_ERR(task)) {
2584                 init_idle_pids(task);
2585                 init_idle(task, cpu);
2586         }
2587
2588         return task;
2589 }
2590
2591 struct mm_struct *copy_init_mm(void)
2592 {
2593         return dup_mm(NULL, &init_mm);
2594 }
2595
2596 /*
2597  * This is like kernel_clone(), but shaved down and tailored to just
2598  * creating io_uring workers. It returns a created task, or an error pointer.
2599  * The returned task is inactive, and the caller must fire it up through
2600  * wake_up_new_task(p). All signals are blocked in the created task.
2601  */
2602 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2603 {
2604         unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2605                                 CLONE_IO;
2606         struct kernel_clone_args args = {
2607                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2608                                     CLONE_UNTRACED) & ~CSIGNAL),
2609                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2610                 .fn             = fn,
2611                 .fn_arg         = arg,
2612                 .io_thread      = 1,
2613         };
2614
2615         return copy_process(NULL, 0, node, &args);
2616 }
2617
2618 /*
2619  *  Ok, this is the main fork-routine.
2620  *
2621  * It copies the process, and if successful kick-starts
2622  * it and waits for it to finish using the VM if required.
2623  *
2624  * args->exit_signal is expected to be checked for sanity by the caller.
2625  */
2626 pid_t kernel_clone(struct kernel_clone_args *args)
2627 {
2628         u64 clone_flags = args->flags;
2629         struct completion vfork;
2630         struct pid *pid;
2631         struct task_struct *p;
2632         int trace = 0;
2633         pid_t nr;
2634
2635         /*
2636          * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2637          * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2638          * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2639          * field in struct clone_args and it still doesn't make sense to have
2640          * them both point at the same memory location. Performing this check
2641          * here has the advantage that we don't need to have a separate helper
2642          * to check for legacy clone().
2643          */
2644         if ((args->flags & CLONE_PIDFD) &&
2645             (args->flags & CLONE_PARENT_SETTID) &&
2646             (args->pidfd == args->parent_tid))
2647                 return -EINVAL;
2648
2649         /*
2650          * Determine whether and which event to report to ptracer.  When
2651          * called from kernel_thread or CLONE_UNTRACED is explicitly
2652          * requested, no event is reported; otherwise, report if the event
2653          * for the type of forking is enabled.
2654          */
2655         if (!(clone_flags & CLONE_UNTRACED)) {
2656                 if (clone_flags & CLONE_VFORK)
2657                         trace = PTRACE_EVENT_VFORK;
2658                 else if (args->exit_signal != SIGCHLD)
2659                         trace = PTRACE_EVENT_CLONE;
2660                 else
2661                         trace = PTRACE_EVENT_FORK;
2662
2663                 if (likely(!ptrace_event_enabled(current, trace)))
2664                         trace = 0;
2665         }
2666
2667         p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2668         add_latent_entropy();
2669
2670         if (IS_ERR(p))
2671                 return PTR_ERR(p);
2672
2673         /*
2674          * Do this prior waking up the new thread - the thread pointer
2675          * might get invalid after that point, if the thread exits quickly.
2676          */
2677         trace_sched_process_fork(current, p);
2678
2679         pid = get_task_pid(p, PIDTYPE_PID);
2680         nr = pid_vnr(pid);
2681
2682         if (clone_flags & CLONE_PARENT_SETTID)
2683                 put_user(nr, args->parent_tid);
2684
2685         if (clone_flags & CLONE_VFORK) {
2686                 p->vfork_done = &vfork;
2687                 init_completion(&vfork);
2688                 get_task_struct(p);
2689         }
2690
2691         if (IS_ENABLED(CONFIG_LRU_GEN) && !(clone_flags & CLONE_VM)) {
2692                 /* lock the task to synchronize with memcg migration */
2693                 task_lock(p);
2694                 lru_gen_add_mm(p->mm);
2695                 task_unlock(p);
2696         }
2697
2698         wake_up_new_task(p);
2699
2700         /* forking complete and child started to run, tell ptracer */
2701         if (unlikely(trace))
2702                 ptrace_event_pid(trace, pid);
2703
2704         if (clone_flags & CLONE_VFORK) {
2705                 if (!wait_for_vfork_done(p, &vfork))
2706                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2707         }
2708
2709         put_pid(pid);
2710         return nr;
2711 }
2712
2713 /*
2714  * Create a kernel thread.
2715  */
2716 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2717 {
2718         struct kernel_clone_args args = {
2719                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2720                                     CLONE_UNTRACED) & ~CSIGNAL),
2721                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2722                 .fn             = fn,
2723                 .fn_arg         = arg,
2724                 .kthread        = 1,
2725         };
2726
2727         return kernel_clone(&args);
2728 }
2729
2730 /*
2731  * Create a user mode thread.
2732  */
2733 pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2734 {
2735         struct kernel_clone_args args = {
2736                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2737                                     CLONE_UNTRACED) & ~CSIGNAL),
2738                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2739                 .fn             = fn,
2740                 .fn_arg         = arg,
2741         };
2742
2743         return kernel_clone(&args);
2744 }
2745
2746 #ifdef __ARCH_WANT_SYS_FORK
2747 SYSCALL_DEFINE0(fork)
2748 {
2749 #ifdef CONFIG_MMU
2750         struct kernel_clone_args args = {
2751                 .exit_signal = SIGCHLD,
2752         };
2753
2754         return kernel_clone(&args);
2755 #else
2756         /* can not support in nommu mode */
2757         return -EINVAL;
2758 #endif
2759 }
2760 #endif
2761
2762 #ifdef __ARCH_WANT_SYS_VFORK
2763 SYSCALL_DEFINE0(vfork)
2764 {
2765         struct kernel_clone_args args = {
2766                 .flags          = CLONE_VFORK | CLONE_VM,
2767                 .exit_signal    = SIGCHLD,
2768         };
2769
2770         return kernel_clone(&args);
2771 }
2772 #endif
2773
2774 #ifdef __ARCH_WANT_SYS_CLONE
2775 #ifdef CONFIG_CLONE_BACKWARDS
2776 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2777                  int __user *, parent_tidptr,
2778                  unsigned long, tls,
2779                  int __user *, child_tidptr)
2780 #elif defined(CONFIG_CLONE_BACKWARDS2)
2781 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2782                  int __user *, parent_tidptr,
2783                  int __user *, child_tidptr,
2784                  unsigned long, tls)
2785 #elif defined(CONFIG_CLONE_BACKWARDS3)
2786 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2787                 int, stack_size,
2788                 int __user *, parent_tidptr,
2789                 int __user *, child_tidptr,
2790                 unsigned long, tls)
2791 #else
2792 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2793                  int __user *, parent_tidptr,
2794                  int __user *, child_tidptr,
2795                  unsigned long, tls)
2796 #endif
2797 {
2798         struct kernel_clone_args args = {
2799                 .flags          = (lower_32_bits(clone_flags) & ~CSIGNAL),
2800                 .pidfd          = parent_tidptr,
2801                 .child_tid      = child_tidptr,
2802                 .parent_tid     = parent_tidptr,
2803                 .exit_signal    = (lower_32_bits(clone_flags) & CSIGNAL),
2804                 .stack          = newsp,
2805                 .tls            = tls,
2806         };
2807
2808         return kernel_clone(&args);
2809 }
2810 #endif
2811
2812 #ifdef __ARCH_WANT_SYS_CLONE3
2813
2814 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2815                                               struct clone_args __user *uargs,
2816                                               size_t usize)
2817 {
2818         int err;
2819         struct clone_args args;
2820         pid_t *kset_tid = kargs->set_tid;
2821
2822         BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2823                      CLONE_ARGS_SIZE_VER0);
2824         BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2825                      CLONE_ARGS_SIZE_VER1);
2826         BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2827                      CLONE_ARGS_SIZE_VER2);
2828         BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2829
2830         if (unlikely(usize > PAGE_SIZE))
2831                 return -E2BIG;
2832         if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2833                 return -EINVAL;
2834
2835         err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2836         if (err)
2837                 return err;
2838
2839         if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2840                 return -EINVAL;
2841
2842         if (unlikely(!args.set_tid && args.set_tid_size > 0))
2843                 return -EINVAL;
2844
2845         if (unlikely(args.set_tid && args.set_tid_size == 0))
2846                 return -EINVAL;
2847
2848         /*
2849          * Verify that higher 32bits of exit_signal are unset and that
2850          * it is a valid signal
2851          */
2852         if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2853                      !valid_signal(args.exit_signal)))
2854                 return -EINVAL;
2855
2856         if ((args.flags & CLONE_INTO_CGROUP) &&
2857             (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2858                 return -EINVAL;
2859
2860         *kargs = (struct kernel_clone_args){
2861                 .flags          = args.flags,
2862                 .pidfd          = u64_to_user_ptr(args.pidfd),
2863                 .child_tid      = u64_to_user_ptr(args.child_tid),
2864                 .parent_tid     = u64_to_user_ptr(args.parent_tid),
2865                 .exit_signal    = args.exit_signal,
2866                 .stack          = args.stack,
2867                 .stack_size     = args.stack_size,
2868                 .tls            = args.tls,
2869                 .set_tid_size   = args.set_tid_size,
2870                 .cgroup         = args.cgroup,
2871         };
2872
2873         if (args.set_tid &&
2874                 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2875                         (kargs->set_tid_size * sizeof(pid_t))))
2876                 return -EFAULT;
2877
2878         kargs->set_tid = kset_tid;
2879
2880         return 0;
2881 }
2882
2883 /**
2884  * clone3_stack_valid - check and prepare stack
2885  * @kargs: kernel clone args
2886  *
2887  * Verify that the stack arguments userspace gave us are sane.
2888  * In addition, set the stack direction for userspace since it's easy for us to
2889  * determine.
2890  */
2891 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2892 {
2893         if (kargs->stack == 0) {
2894                 if (kargs->stack_size > 0)
2895                         return false;
2896         } else {
2897                 if (kargs->stack_size == 0)
2898                         return false;
2899
2900                 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2901                         return false;
2902
2903 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2904                 kargs->stack += kargs->stack_size;
2905 #endif
2906         }
2907
2908         return true;
2909 }
2910
2911 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2912 {
2913         /* Verify that no unknown flags are passed along. */
2914         if (kargs->flags &
2915             ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2916                 return false;
2917
2918         /*
2919          * - make the CLONE_DETACHED bit reusable for clone3
2920          * - make the CSIGNAL bits reusable for clone3
2921          */
2922         if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2923                 return false;
2924
2925         if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2926             (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2927                 return false;
2928
2929         if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2930             kargs->exit_signal)
2931                 return false;
2932
2933         if (!clone3_stack_valid(kargs))
2934                 return false;
2935
2936         return true;
2937 }
2938
2939 /**
2940  * clone3 - create a new process with specific properties
2941  * @uargs: argument structure
2942  * @size:  size of @uargs
2943  *
2944  * clone3() is the extensible successor to clone()/clone2().
2945  * It takes a struct as argument that is versioned by its size.
2946  *
2947  * Return: On success, a positive PID for the child process.
2948  *         On error, a negative errno number.
2949  */
2950 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2951 {
2952         int err;
2953
2954         struct kernel_clone_args kargs;
2955         pid_t set_tid[MAX_PID_NS_LEVEL];
2956
2957         kargs.set_tid = set_tid;
2958
2959         err = copy_clone_args_from_user(&kargs, uargs, size);
2960         if (err)
2961                 return err;
2962
2963         if (!clone3_args_valid(&kargs))
2964                 return -EINVAL;
2965
2966         return kernel_clone(&kargs);
2967 }
2968 #endif
2969
2970 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2971 {
2972         struct task_struct *leader, *parent, *child;
2973         int res;
2974
2975         read_lock(&tasklist_lock);
2976         leader = top = top->group_leader;
2977 down:
2978         for_each_thread(leader, parent) {
2979                 list_for_each_entry(child, &parent->children, sibling) {
2980                         res = visitor(child, data);
2981                         if (res) {
2982                                 if (res < 0)
2983                                         goto out;
2984                                 leader = child;
2985                                 goto down;
2986                         }
2987 up:
2988                         ;
2989                 }
2990         }
2991
2992         if (leader != top) {
2993                 child = leader;
2994                 parent = child->real_parent;
2995                 leader = parent->group_leader;
2996                 goto up;
2997         }
2998 out:
2999         read_unlock(&tasklist_lock);
3000 }
3001
3002 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
3003 #define ARCH_MIN_MMSTRUCT_ALIGN 0
3004 #endif
3005
3006 static void sighand_ctor(void *data)
3007 {
3008         struct sighand_struct *sighand = data;
3009
3010         spin_lock_init(&sighand->siglock);
3011         init_waitqueue_head(&sighand->signalfd_wqh);
3012 }
3013
3014 void __init proc_caches_init(void)
3015 {
3016         unsigned int mm_size;
3017
3018         sighand_cachep = kmem_cache_create("sighand_cache",
3019                         sizeof(struct sighand_struct), 0,
3020                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3021                         SLAB_ACCOUNT, sighand_ctor);
3022         signal_cachep = kmem_cache_create("signal_cache",
3023                         sizeof(struct signal_struct), 0,
3024                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3025                         NULL);
3026         files_cachep = kmem_cache_create("files_cache",
3027                         sizeof(struct files_struct), 0,
3028                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3029                         NULL);
3030         fs_cachep = kmem_cache_create("fs_cache",
3031                         sizeof(struct fs_struct), 0,
3032                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3033                         NULL);
3034
3035         /*
3036          * The mm_cpumask is located at the end of mm_struct, and is
3037          * dynamically sized based on the maximum CPU number this system
3038          * can have, taking hotplug into account (nr_cpu_ids).
3039          */
3040         mm_size = sizeof(struct mm_struct) + cpumask_size();
3041
3042         mm_cachep = kmem_cache_create_usercopy("mm_struct",
3043                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3044                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3045                         offsetof(struct mm_struct, saved_auxv),
3046                         sizeof_field(struct mm_struct, saved_auxv),
3047                         NULL);
3048         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3049         mmap_init();
3050         nsproxy_cache_init();
3051 }
3052
3053 /*
3054  * Check constraints on flags passed to the unshare system call.
3055  */
3056 static int check_unshare_flags(unsigned long unshare_flags)
3057 {
3058         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3059                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3060                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3061                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3062                                 CLONE_NEWTIME))
3063                 return -EINVAL;
3064         /*
3065          * Not implemented, but pretend it works if there is nothing
3066          * to unshare.  Note that unsharing the address space or the
3067          * signal handlers also need to unshare the signal queues (aka
3068          * CLONE_THREAD).
3069          */
3070         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3071                 if (!thread_group_empty(current))
3072                         return -EINVAL;
3073         }
3074         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3075                 if (refcount_read(&current->sighand->count) > 1)
3076                         return -EINVAL;
3077         }
3078         if (unshare_flags & CLONE_VM) {
3079                 if (!current_is_single_threaded())
3080                         return -EINVAL;
3081         }
3082
3083         return 0;
3084 }
3085
3086 /*
3087  * Unshare the filesystem structure if it is being shared
3088  */
3089 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3090 {
3091         struct fs_struct *fs = current->fs;
3092
3093         if (!(unshare_flags & CLONE_FS) || !fs)
3094                 return 0;
3095
3096         /* don't need lock here; in the worst case we'll do useless copy */
3097         if (fs->users == 1)
3098                 return 0;
3099
3100         *new_fsp = copy_fs_struct(fs);
3101         if (!*new_fsp)
3102                 return -ENOMEM;
3103
3104         return 0;
3105 }
3106
3107 /*
3108  * Unshare file descriptor table if it is being shared
3109  */
3110 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3111                struct files_struct **new_fdp)
3112 {
3113         struct files_struct *fd = current->files;
3114         int error = 0;
3115
3116         if ((unshare_flags & CLONE_FILES) &&
3117             (fd && atomic_read(&fd->count) > 1)) {
3118                 *new_fdp = dup_fd(fd, max_fds, &error);
3119                 if (!*new_fdp)
3120                         return error;
3121         }
3122
3123         return 0;
3124 }
3125
3126 /*
3127  * unshare allows a process to 'unshare' part of the process
3128  * context which was originally shared using clone.  copy_*
3129  * functions used by kernel_clone() cannot be used here directly
3130  * because they modify an inactive task_struct that is being
3131  * constructed. Here we are modifying the current, active,
3132  * task_struct.
3133  */
3134 int ksys_unshare(unsigned long unshare_flags)
3135 {
3136         struct fs_struct *fs, *new_fs = NULL;
3137         struct files_struct *new_fd = NULL;
3138         struct cred *new_cred = NULL;
3139         struct nsproxy *new_nsproxy = NULL;
3140         int do_sysvsem = 0;
3141         int err;
3142
3143         /*
3144          * If unsharing a user namespace must also unshare the thread group
3145          * and unshare the filesystem root and working directories.
3146          */
3147         if (unshare_flags & CLONE_NEWUSER)
3148                 unshare_flags |= CLONE_THREAD | CLONE_FS;
3149         /*
3150          * If unsharing vm, must also unshare signal handlers.
3151          */
3152         if (unshare_flags & CLONE_VM)
3153                 unshare_flags |= CLONE_SIGHAND;
3154         /*
3155          * If unsharing a signal handlers, must also unshare the signal queues.
3156          */
3157         if (unshare_flags & CLONE_SIGHAND)
3158                 unshare_flags |= CLONE_THREAD;
3159         /*
3160          * If unsharing namespace, must also unshare filesystem information.
3161          */
3162         if (unshare_flags & CLONE_NEWNS)
3163                 unshare_flags |= CLONE_FS;
3164
3165         err = check_unshare_flags(unshare_flags);
3166         if (err)
3167                 goto bad_unshare_out;
3168         /*
3169          * CLONE_NEWIPC must also detach from the undolist: after switching
3170          * to a new ipc namespace, the semaphore arrays from the old
3171          * namespace are unreachable.
3172          */
3173         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3174                 do_sysvsem = 1;
3175         err = unshare_fs(unshare_flags, &new_fs);
3176         if (err)
3177                 goto bad_unshare_out;
3178         err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3179         if (err)
3180                 goto bad_unshare_cleanup_fs;
3181         err = unshare_userns(unshare_flags, &new_cred);
3182         if (err)
3183                 goto bad_unshare_cleanup_fd;
3184         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3185                                          new_cred, new_fs);
3186         if (err)
3187                 goto bad_unshare_cleanup_cred;
3188
3189         if (new_cred) {
3190                 err = set_cred_ucounts(new_cred);
3191                 if (err)
3192                         goto bad_unshare_cleanup_cred;
3193         }
3194
3195         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3196                 if (do_sysvsem) {
3197                         /*
3198                          * CLONE_SYSVSEM is equivalent to sys_exit().
3199                          */
3200                         exit_sem(current);
3201                 }
3202                 if (unshare_flags & CLONE_NEWIPC) {
3203                         /* Orphan segments in old ns (see sem above). */
3204                         exit_shm(current);
3205                         shm_init_task(current);
3206                 }
3207
3208                 if (new_nsproxy)
3209                         switch_task_namespaces(current, new_nsproxy);
3210
3211                 task_lock(current);
3212
3213                 if (new_fs) {
3214                         fs = current->fs;
3215                         spin_lock(&fs->lock);
3216                         current->fs = new_fs;
3217                         if (--fs->users)
3218                                 new_fs = NULL;
3219                         else
3220                                 new_fs = fs;
3221                         spin_unlock(&fs->lock);
3222                 }
3223
3224                 if (new_fd)
3225                         swap(current->files, new_fd);
3226
3227                 task_unlock(current);
3228
3229                 if (new_cred) {
3230                         /* Install the new user namespace */
3231                         commit_creds(new_cred);
3232                         new_cred = NULL;
3233                 }
3234         }
3235
3236         perf_event_namespaces(current);
3237
3238 bad_unshare_cleanup_cred:
3239         if (new_cred)
3240                 put_cred(new_cred);
3241 bad_unshare_cleanup_fd:
3242         if (new_fd)
3243                 put_files_struct(new_fd);
3244
3245 bad_unshare_cleanup_fs:
3246         if (new_fs)
3247                 free_fs_struct(new_fs);
3248
3249 bad_unshare_out:
3250         return err;
3251 }
3252
3253 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3254 {
3255         return ksys_unshare(unshare_flags);
3256 }
3257
3258 /*
3259  *      Helper to unshare the files of the current task.
3260  *      We don't want to expose copy_files internals to
3261  *      the exec layer of the kernel.
3262  */
3263
3264 int unshare_files(void)
3265 {
3266         struct task_struct *task = current;
3267         struct files_struct *old, *copy = NULL;
3268         int error;
3269
3270         error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, &copy);
3271         if (error || !copy)
3272                 return error;
3273
3274         old = task->files;
3275         task_lock(task);
3276         task->files = copy;
3277         task_unlock(task);
3278         put_files_struct(old);
3279         return 0;
3280 }
3281
3282 int sysctl_max_threads(struct ctl_table *table, int write,
3283                        void *buffer, size_t *lenp, loff_t *ppos)
3284 {
3285         struct ctl_table t;
3286         int ret;
3287         int threads = max_threads;
3288         int min = 1;
3289         int max = MAX_THREADS;
3290
3291         t = *table;
3292         t.data = &threads;
3293         t.extra1 = &min;
3294         t.extra2 = &max;
3295
3296         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3297         if (ret || !write)
3298                 return ret;
3299
3300         max_threads = threads;
3301
3302         return 0;
3303 }