kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/sched/autogroup.h>
  16 #include <linux/sched/mm.h>
  17 #include <linux/sched/coredump.h>
  18 #include <linux/sched/user.h>
  19 #include <linux/sched/numa_balancing.h>
  20 #include <linux/sched/stat.h>
  21 #include <linux/sched/task.h>
  22 #include <linux/sched/task_stack.h>
  23 #include <linux/sched/cputime.h>
  24 #include <linux/rtmutex.h>
  25 #include <linux/init.h>
  26 #include <linux/unistd.h>
  27 #include <linux/module.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/completion.h>
  30 #include <linux/personality.h>
  31 #include <linux/mempolicy.h>
  32 #include <linux/sem.h>
  33 #include <linux/file.h>
  34 #include <linux/fdtable.h>
  35 #include <linux/iocontext.h>
  36 #include <linux/key.h>
  37 #include <linux/binfmts.h>
  38 #include <linux/mman.h>
  39 #include <linux/mmu_notifier.h>
  40 #include <linux/hmm.h>
  41 #include <linux/fs.h>
  42 #include <linux/mm.h>
  43 #include <linux/vmacache.h>
  44 #include <linux/nsproxy.h>
  45 #include <linux/capability.h>
  46 #include <linux/cpu.h>
  47 #include <linux/cgroup.h>
  48 #include <linux/security.h>
  49 #include <linux/hugetlb.h>
  50 #include <linux/seccomp.h>
  51 #include <linux/swap.h>
  52 #include <linux/syscalls.h>
  53 #include <linux/jiffies.h>
  54 #include <linux/futex.h>
  55 #include <linux/compat.h>
  56 #include <linux/kthread.h>
  57 #include <linux/task_io_accounting_ops.h>
  58 #include <linux/rcupdate.h>
  59 #include <linux/ptrace.h>
  60 #include <linux/mount.h>
  61 #include <linux/audit.h>
  62 #include <linux/memcontrol.h>
  63 #include <linux/ftrace.h>
  64 #include <linux/proc_fs.h>
  65 #include <linux/profile.h>
  66 #include <linux/rmap.h>
  67 #include <linux/ksm.h>
  68 #include <linux/acct.h>
  69 #include <linux/userfaultfd_k.h>
  70 #include <linux/tsacct_kern.h>
  71 #include <linux/cn_proc.h>
  72 #include <linux/freezer.h>
  73 #include <linux/delayacct.h>
  74 #include <linux/taskstats_kern.h>
  75 #include <linux/random.h>
  76 #include <linux/tty.h>
  77 #include <linux/blkdev.h>
  78 #include <linux/fs_struct.h>
  79 #include <linux/magic.h>
  80 #include <linux/sched/mm.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
  95 #include <asm/pgtable.h>
  96 #include <asm/pgalloc.h>
  97 #include <linux/uaccess.h>
  98 #include <asm/mmu_context.h>
  99 #include <asm/cacheflush.h>
 100 #include <asm/tlbflush.h>
 101
 102 #include <trace/events/sched.h>
 103
 104 #define CREATE_TRACE_POINTS
 105 #include <trace/events/task.h>
 106
 107 /*
 108  * Minimum number of threads to boot the kernel
 109  */
 110 #define MIN_THREADS 20
 111
 112 /*
 113  * Maximum number of threads
 114  */
 115 #define MAX_THREADS FUTEX_TID_MASK
 116
 117 /*
 118  * Protected counters by write_lock_irq(&tasklist_lock)
 119  */
 120 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 121 int nr_threads;                 /* The idle threads do not count.. */
 122
 123 int max_threads;                /* tunable limit on nr_threads */
 124
 125 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 126
 127 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 128
 129 #ifdef CONFIG_PROVE_RCU
 130 int lockdep_tasklist_lock_is_held(void)
 131 {
 132         return lockdep_is_held(&tasklist_lock);
 133 }
 134 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 135 #endif /* #ifdef CONFIG_PROVE_RCU */
 136
 137 int nr_processes(void)
 138 {
 139         int cpu;
 140         int total = 0;
 141
 142         for_each_possible_cpu(cpu)
 143                 total += per_cpu(process_counts, cpu);
 144
 145         return total;
 146 }
 147
 148 void __weak arch_release_task_struct(struct task_struct *tsk)
 149 {
 150 }
 151
 152 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 153 static struct kmem_cache *task_struct_cachep;
 154
 155 static inline struct task_struct *alloc_task_struct_node(int node)
 156 {
 157         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 158 }
 159
 160 static inline void free_task_struct(struct task_struct *tsk)
 161 {
 162         kmem_cache_free(task_struct_cachep, tsk);
 163 }
 164 #endif
 165
 166 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 167
 168 /*
 169  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 170  * kmemcache based allocator.
 171  */
 172 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 173
 174 #ifdef CONFIG_VMAP_STACK
 175 /*
 176  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 177  * flush.  Try to minimize the number of calls by caching stacks.
 178  */
 179 #define NR_CACHED_STACKS 2
 180 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 181
 182 static int free_vm_stack_cache(unsigned int cpu)
 183 {
 184         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 185         int i;
 186
 187         for (i = 0; i < NR_CACHED_STACKS; i++) {
 188                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 189
 190                 if (!vm_stack)
 191                         continue;
 192
 193                 vfree(vm_stack->addr);
 194                 cached_vm_stacks[i] = NULL;
 195         }
 196
 197         return 0;
 198 }
 199 #endif
 200
 201 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 202 {
 203 #ifdef CONFIG_VMAP_STACK
 204         void *stack;
 205         int i;
 206
 207         for (i = 0; i < NR_CACHED_STACKS; i++) {
 208                 struct vm_struct *s;
 209
 210                 s = this_cpu_xchg(cached_stacks[i], NULL);
 211
 212                 if (!s)
 213                         continue;
 214
 215                 /* Clear stale pointers from reused stack. */
 216                 memset(s->addr, 0, THREAD_SIZE);
 217
 218                 tsk->stack_vm_area = s;
 219                 tsk->stack = s->addr;
 220                 return s->addr;
 221         }
 222
 223         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 224                                      VMALLOC_START, VMALLOC_END,
 225                                      THREADINFO_GFP,
 226                                      PAGE_KERNEL,
 227                                      0, node, __builtin_return_address(0));
 228
 229         /*
 230          * We can't call find_vm_area() in interrupt context, and
 231          * free_thread_stack() can be called in interrupt context,
 232          * so cache the vm_struct.
 233          */
 234         if (stack) {
 235                 tsk->stack_vm_area = find_vm_area(stack);
 236                 tsk->stack = stack;
 237         }
 238         return stack;
 239 #else
 240         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 241                                              THREAD_SIZE_ORDER);
 242
 243         if (likely(page)) {
 244                 tsk->stack = page_address(page);
 245                 return tsk->stack;
 246         }
 247         return NULL;
 248 #endif
 249 }
 250
 251 static inline void free_thread_stack(struct task_struct *tsk)
 252 {
 253 #ifdef CONFIG_VMAP_STACK
 254         if (task_stack_vm_area(tsk)) {
 255                 int i;
 256
 257                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 258                         if (this_cpu_cmpxchg(cached_stacks[i],
 259                                         NULL, tsk->stack_vm_area) != NULL)
 260                                 continue;
 261
 262                         return;
 263                 }
 264
 265                 vfree_atomic(tsk->stack);
 266                 return;
 267         }
 268 #endif
 269
 270         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 271 }
 272 # else
 273 static struct kmem_cache *thread_stack_cache;
 274
 275 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 276                                                   int node)
 277 {
 278         unsigned long *stack;
 279         stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 280         tsk->stack = stack;
 281         return stack;
 282 }
 283
 284 static void free_thread_stack(struct task_struct *tsk)
 285 {
 286         kmem_cache_free(thread_stack_cache, tsk->stack);
 287 }
 288
 289 void thread_stack_cache_init(void)
 290 {
 291         thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 292                                         THREAD_SIZE, THREAD_SIZE, 0, 0,
 293                                         THREAD_SIZE, NULL);
 294         BUG_ON(thread_stack_cache == NULL);
 295 }
 296 # endif
 297 #endif
 298
 299 /* SLAB cache for signal_struct structures (tsk->signal) */
 300 static struct kmem_cache *signal_cachep;
 301
 302 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 303 struct kmem_cache *sighand_cachep;
 304
 305 /* SLAB cache for files_struct structures (tsk->files) */
 306 struct kmem_cache *files_cachep;
 307
 308 /* SLAB cache for fs_struct structures (tsk->fs) */
 309 struct kmem_cache *fs_cachep;
 310
 311 /* SLAB cache for vm_area_struct structures */
 312 static struct kmem_cache *vm_area_cachep;
 313
 314 /* SLAB cache for mm_struct structures (tsk->mm) */
 315 static struct kmem_cache *mm_cachep;
 316
 317 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 318 {
 319         struct vm_area_struct *vma;
 320
 321         vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 322         if (vma)
 323                 vma_init(vma, mm);
 324         return vma;
 325 }
 326
 327 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 328 {
 329         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 330
 331         if (new) {
 332                 *new = *orig;
 333                 INIT_LIST_HEAD(&new->anon_vma_chain);
 334         }
 335         return new;
 336 }
 337
 338 void vm_area_free(struct vm_area_struct *vma)
 339 {
 340         kmem_cache_free(vm_area_cachep, vma);
 341 }
 342
 343 static void account_kernel_stack(struct task_struct *tsk, int account)
 344 {
 345         void *stack = task_stack_page(tsk);
 346         struct vm_struct *vm = task_stack_vm_area(tsk);
 347
 348         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 349
 350         if (vm) {
 351                 int i;
 352
 353                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 354
 355                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 356                         mod_zone_page_state(page_zone(vm->pages[i]),
 357                                             NR_KERNEL_STACK_KB,
 358                                             PAGE_SIZE / 1024 * account);
 359                 }
 360
 361                 /* All stack pages belong to the same memcg. */
 362                 mod_memcg_page_state(vm->pages[0], MEMCG_KERNEL_STACK_KB,
 363                                      account * (THREAD_SIZE / 1024));
 364         } else {
 365                 /*
 366                  * All stack pages are in the same zone and belong to the
 367                  * same memcg.
 368                  */
 369                 struct page *first_page = virt_to_page(stack);
 370
 371                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 372                                     THREAD_SIZE / 1024 * account);
 373
 374                 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
 375                                      account * (THREAD_SIZE / 1024));
 376         }
 377 }
 378
 379 static void release_task_stack(struct task_struct *tsk)
 380 {
 381         if (WARN_ON(tsk->state != TASK_DEAD))
 382                 return;  /* Better to leak the stack than to free prematurely */
 383
 384         account_kernel_stack(tsk, -1);
 385         free_thread_stack(tsk);
 386         tsk->stack = NULL;
 387 #ifdef CONFIG_VMAP_STACK
 388         tsk->stack_vm_area = NULL;
 389 #endif
 390 }
 391
 392 #ifdef CONFIG_THREAD_INFO_IN_TASK
 393 void put_task_stack(struct task_struct *tsk)
 394 {
 395         if (atomic_dec_and_test(&tsk->stack_refcount))
 396                 release_task_stack(tsk);
 397 }
 398 #endif
 399
 400 void free_task(struct task_struct *tsk)
 401 {
 402 #ifndef CONFIG_THREAD_INFO_IN_TASK
 403         /*
 404          * The task is finally done with both the stack and thread_info,
 405          * so free both.
 406          */
 407         release_task_stack(tsk);
 408 #else
 409         /*
 410          * If the task had a separate stack allocation, it should be gone
 411          * by now.
 412          */
 413         WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
 414 #endif
 415         rt_mutex_debug_task_free(tsk);
 416         ftrace_graph_exit_task(tsk);
 417         put_seccomp_filter(tsk);
 418         arch_release_task_struct(tsk);
 419         if (tsk->flags & PF_KTHREAD)
 420                 free_kthread_struct(tsk);
 421         free_task_struct(tsk);
 422 }
 423 EXPORT_SYMBOL(free_task);
 424
 425 #ifdef CONFIG_MMU
 426 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 427                                         struct mm_struct *oldmm)
 428 {
 429         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 430         struct rb_node **rb_link, *rb_parent;
 431         int retval;
 432         unsigned long charge;
 433         LIST_HEAD(uf);
 434
 435         uprobe_start_dup_mmap();
 436         if (down_write_killable(&oldmm->mmap_sem)) {
 437                 retval = -EINTR;
 438                 goto fail_uprobe_end;
 439         }
 440         flush_cache_dup_mm(oldmm);
 441         uprobe_dup_mmap(oldmm, mm);
 442         /*
 443          * Not linked in yet - no deadlock potential:
 444          */
 445         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 446
 447         /* No ordering required: file already has been exposed. */
 448         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 449
 450         mm->total_vm = oldmm->total_vm;
 451         mm->data_vm = oldmm->data_vm;
 452         mm->exec_vm = oldmm->exec_vm;
 453         mm->stack_vm = oldmm->stack_vm;
 454
 455         rb_link = &mm->mm_rb.rb_node;
 456         rb_parent = NULL;
 457         pprev = &mm->mmap;
 458         retval = ksm_fork(mm, oldmm);
 459         if (retval)
 460                 goto out;
 461         retval = khugepaged_fork(mm, oldmm);
 462         if (retval)
 463                 goto out;
 464
 465         prev = NULL;
 466         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 467                 struct file *file;
 468
 469                 if (mpnt->vm_flags & VM_DONTCOPY) {
 470                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 471                         continue;
 472                 }
 473                 charge = 0;
 474                 /*
 475                  * Don't duplicate many vmas if we've been oom-killed (for
 476                  * example)
 477                  */
 478                 if (fatal_signal_pending(current)) {
 479                         retval = -EINTR;
 480                         goto out;
 481                 }
 482                 if (mpnt->vm_flags & VM_ACCOUNT) {
 483                         unsigned long len = vma_pages(mpnt);
 484
 485                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 486                                 goto fail_nomem;
 487                         charge = len;
 488                 }
 489                 tmp = vm_area_dup(mpnt);
 490                 if (!tmp)
 491                         goto fail_nomem;
 492                 retval = vma_dup_policy(mpnt, tmp);
 493                 if (retval)
 494                         goto fail_nomem_policy;
 495                 tmp->vm_mm = mm;
 496                 retval = dup_userfaultfd(tmp, &uf);
 497                 if (retval)
 498                         goto fail_nomem_anon_vma_fork;
 499                 if (tmp->vm_flags & VM_WIPEONFORK) {
 500                         /* VM_WIPEONFORK gets a clean slate in the child. */
 501                         tmp->anon_vma = NULL;
 502                         if (anon_vma_prepare(tmp))
 503                                 goto fail_nomem_anon_vma_fork;
 504                 } else if (anon_vma_fork(tmp, mpnt))
 505                         goto fail_nomem_anon_vma_fork;
 506                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 507                 tmp->vm_next = tmp->vm_prev = NULL;
 508                 file = tmp->vm_file;
 509                 if (file) {
 510                         struct inode *inode = file_inode(file);
 511                         struct address_space *mapping = file->f_mapping;
 512
 513                         get_file(file);
 514                         if (tmp->vm_flags & VM_DENYWRITE)
 515                                 atomic_dec(&inode->i_writecount);
 516                         i_mmap_lock_write(mapping);
 517                         if (tmp->vm_flags & VM_SHARED)
 518                                 atomic_inc(&mapping->i_mmap_writable);
 519                         flush_dcache_mmap_lock(mapping);
 520                         /* insert tmp into the share list, just after mpnt */
 521                         vma_interval_tree_insert_after(tmp, mpnt,
 522                                         &mapping->i_mmap);
 523                         flush_dcache_mmap_unlock(mapping);
 524                         i_mmap_unlock_write(mapping);
 525                 }
 526
 527                 /*
 528                  * Clear hugetlb-related page reserves for children. This only
 529                  * affects MAP_PRIVATE mappings. Faults generated by the child
 530                  * are not guaranteed to succeed, even if read-only
 531                  */
 532                 if (is_vm_hugetlb_page(tmp))
 533                         reset_vma_resv_huge_pages(tmp);
 534
 535                 /*
 536                  * Link in the new vma and copy the page table entries.
 537                  */
 538                 *pprev = tmp;
 539                 pprev = &tmp->vm_next;
 540                 tmp->vm_prev = prev;
 541                 prev = tmp;
 542
 543                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 544                 rb_link = &tmp->vm_rb.rb_right;
 545                 rb_parent = &tmp->vm_rb;
 546
 547                 mm->map_count++;
 548                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 549                         retval = copy_page_range(mm, oldmm, mpnt);
 550
 551                 if (tmp->vm_ops && tmp->vm_ops->open)
 552                         tmp->vm_ops->open(tmp);
 553
 554                 if (retval)
 555                         goto out;
 556         }
 557         /* a new mm has just been created */
 558         retval = arch_dup_mmap(oldmm, mm);
 559 out:
 560         up_write(&mm->mmap_sem);
 561         flush_tlb_mm(oldmm);
 562         up_write(&oldmm->mmap_sem);
 563         dup_userfaultfd_complete(&uf);
 564 fail_uprobe_end:
 565         uprobe_end_dup_mmap();
 566         return retval;
 567 fail_nomem_anon_vma_fork:
 568         mpol_put(vma_policy(tmp));
 569 fail_nomem_policy:
 570         vm_area_free(tmp);
 571 fail_nomem:
 572         retval = -ENOMEM;
 573         vm_unacct_memory(charge);
 574         goto out;
 575 }
 576
 577 static inline int mm_alloc_pgd(struct mm_struct *mm)
 578 {
 579         mm->pgd = pgd_alloc(mm);
 580         if (unlikely(!mm->pgd))
 581                 return -ENOMEM;
 582         return 0;
 583 }
 584
 585 static inline void mm_free_pgd(struct mm_struct *mm)
 586 {
 587         pgd_free(mm, mm->pgd);
 588 }
 589 #else
 590 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 591 {
 592         down_write(&oldmm->mmap_sem);
 593         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 594         up_write(&oldmm->mmap_sem);
 595         return 0;
 596 }
 597 #define mm_alloc_pgd(mm)        (0)
 598 #define mm_free_pgd(mm)
 599 #endif /* CONFIG_MMU */
 600
 601 static void check_mm(struct mm_struct *mm)
 602 {
 603         int i;
 604
 605         for (i = 0; i < NR_MM_COUNTERS; i++) {
 606                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 607
 608                 if (unlikely(x))
 609                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 610                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 611         }
 612
 613         if (mm_pgtables_bytes(mm))
 614                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 615                                 mm_pgtables_bytes(mm));
 616
 617 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 618         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 619 #endif
 620 }
 621
 622 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 623 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 624
 625 /*
 626  * Called when the last reference to the mm
 627  * is dropped: either by a lazy thread or by
 628  * mmput. Free the page directory and the mm.
 629  */
 630 void __mmdrop(struct mm_struct *mm)
 631 {
 632         BUG_ON(mm == &init_mm);
 633         WARN_ON_ONCE(mm == current->mm);
 634         WARN_ON_ONCE(mm == current->active_mm);
 635         mm_free_pgd(mm);
 636         destroy_context(mm);
 637         hmm_mm_destroy(mm);
 638         mmu_notifier_mm_destroy(mm);
 639         check_mm(mm);
 640         put_user_ns(mm->user_ns);
 641         free_mm(mm);
 642 }
 643 EXPORT_SYMBOL_GPL(__mmdrop);
 644
 645 static void mmdrop_async_fn(struct work_struct *work)
 646 {
 647         struct mm_struct *mm;
 648
 649         mm = container_of(work, struct mm_struct, async_put_work);
 650         __mmdrop(mm);
 651 }
 652
 653 static void mmdrop_async(struct mm_struct *mm)
 654 {
 655         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 656                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 657                 schedule_work(&mm->async_put_work);
 658         }
 659 }
 660
 661 static inline void free_signal_struct(struct signal_struct *sig)
 662 {
 663         taskstats_tgid_free(sig);
 664         sched_autogroup_exit(sig);
 665         /*
 666          * __mmdrop is not safe to call from softirq context on x86 due to
 667          * pgd_dtor so postpone it to the async context
 668          */
 669         if (sig->oom_mm)
 670                 mmdrop_async(sig->oom_mm);
 671         kmem_cache_free(signal_cachep, sig);
 672 }
 673
 674 static inline void put_signal_struct(struct signal_struct *sig)
 675 {
 676         if (atomic_dec_and_test(&sig->sigcnt))
 677                 free_signal_struct(sig);
 678 }
 679
 680 void __put_task_struct(struct task_struct *tsk)
 681 {
 682         WARN_ON(!tsk->exit_state);
 683         WARN_ON(atomic_read(&tsk->usage));
 684         WARN_ON(tsk == current);
 685
 686         cgroup_free(tsk);
 687         task_numa_free(tsk, true);
 688         security_task_free(tsk);
 689         exit_creds(tsk);
 690         delayacct_tsk_free(tsk);
 691         put_signal_struct(tsk->signal);
 692
 693         if (!profile_handoff_task(tsk))
 694                 free_task(tsk);
 695 }
 696 EXPORT_SYMBOL_GPL(__put_task_struct);
 697
 698 void __init __weak arch_task_cache_init(void) { }
 699
 700 /*
 701  * set_max_threads
 702  */
 703 static void set_max_threads(unsigned int max_threads_suggested)
 704 {
 705         u64 threads;
 706
 707         /*
 708          * The number of threads shall be limited such that the thread
 709          * structures may only consume a small part of the available memory.
 710          */
 711         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 712                 threads = MAX_THREADS;
 713         else
 714                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 715                                     (u64) THREAD_SIZE * 8UL);
 716
 717         if (threads > max_threads_suggested)
 718                 threads = max_threads_suggested;
 719
 720         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 721 }
 722
 723 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 724 /* Initialized by the architecture: */
 725 int arch_task_struct_size __read_mostly;
 726 #endif
 727
 728 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 729 {
 730         /* Fetch thread_struct whitelist for the architecture. */
 731         arch_thread_struct_whitelist(offset, size);
 732
 733         /*
 734          * Handle zero-sized whitelist or empty thread_struct, otherwise
 735          * adjust offset to position of thread_struct in task_struct.
 736          */
 737         if (unlikely(*size == 0))
 738                 *offset = 0;
 739         else
 740                 *offset += offsetof(struct task_struct, thread);
 741 }
 742
 743 void __init fork_init(void)
 744 {
 745         int i;
 746 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 747 #ifndef ARCH_MIN_TASKALIGN
 748 #define ARCH_MIN_TASKALIGN      0
 749 #endif
 750         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 751         unsigned long useroffset, usersize;
 752
 753         /* create a slab on which task_structs can be allocated */
 754         task_struct_whitelist(&useroffset, &usersize);
 755         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 756                         arch_task_struct_size, align,
 757                         SLAB_PANIC|SLAB_ACCOUNT,
 758                         useroffset, usersize, NULL);
 759 #endif
 760
 761         /* do the arch specific task caches init */
 762         arch_task_cache_init();
 763
 764         set_max_threads(MAX_THREADS);
 765
 766         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 767         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 768         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 769                 init_task.signal->rlim[RLIMIT_NPROC];
 770
 771         for (i = 0; i < UCOUNT_COUNTS; i++) {
 772                 init_user_ns.ucount_max[i] = max_threads/2;
 773         }
 774
 775 #ifdef CONFIG_VMAP_STACK
 776         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 777                           NULL, free_vm_stack_cache);
 778 #endif
 779
 780         lockdep_init_task(&init_task);
 781 }
 782
 783 int __weak arch_dup_task_struct(struct task_struct *dst,
 784                                                struct task_struct *src)
 785 {
 786         *dst = *src;
 787         return 0;
 788 }
 789
 790 void set_task_stack_end_magic(struct task_struct *tsk)
 791 {
 792         unsigned long *stackend;
 793
 794         stackend = end_of_stack(tsk);
 795         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 796 }
 797
 798 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 799 {
 800         struct task_struct *tsk;
 801         unsigned long *stack;
 802         struct vm_struct *stack_vm_area;
 803         int err;
 804
 805         if (node == NUMA_NO_NODE)
 806                 node = tsk_fork_get_node(orig);
 807         tsk = alloc_task_struct_node(node);
 808         if (!tsk)
 809                 return NULL;
 810
 811         stack = alloc_thread_stack_node(tsk, node);
 812         if (!stack)
 813                 goto free_tsk;
 814
 815         stack_vm_area = task_stack_vm_area(tsk);
 816
 817         err = arch_dup_task_struct(tsk, orig);
 818
 819         /*
 820          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 821          * sure they're properly initialized before using any stack-related
 822          * functions again.
 823          */
 824         tsk->stack = stack;
 825 #ifdef CONFIG_VMAP_STACK
 826         tsk->stack_vm_area = stack_vm_area;
 827 #endif
 828 #ifdef CONFIG_THREAD_INFO_IN_TASK
 829         atomic_set(&tsk->stack_refcount, 1);
 830 #endif
 831
 832         if (err)
 833                 goto free_stack;
 834
 835 #ifdef CONFIG_SECCOMP
 836         /*
 837          * We must handle setting up seccomp filters once we're under
 838          * the sighand lock in case orig has changed between now and
 839          * then. Until then, filter must be NULL to avoid messing up
 840          * the usage counts on the error path calling free_task.
 841          */
 842         tsk->seccomp.filter = NULL;
 843 #endif
 844
 845         setup_thread_stack(tsk, orig);
 846         clear_user_return_notifier(tsk);
 847         clear_tsk_need_resched(tsk);
 848         set_task_stack_end_magic(tsk);
 849
 850 #ifdef CONFIG_STACKPROTECTOR
 851         tsk->stack_canary = get_random_canary();
 852 #endif
 853
 854         /*
 855          * One for us, one for whoever does the "release_task()" (usually
 856          * parent)
 857          */
 858         atomic_set(&tsk->usage, 2);
 859 #ifdef CONFIG_BLK_DEV_IO_TRACE
 860         tsk->btrace_seq = 0;
 861 #endif
 862         tsk->splice_pipe = NULL;
 863         tsk->task_frag.page = NULL;
 864         tsk->wake_q.next = NULL;
 865
 866         account_kernel_stack(tsk, 1);
 867
 868         kcov_task_init(tsk);
 869
 870 #ifdef CONFIG_FAULT_INJECTION
 871         tsk->fail_nth = 0;
 872 #endif
 873
 874 #ifdef CONFIG_BLK_CGROUP
 875         tsk->throttle_queue = NULL;
 876         tsk->use_memdelay = 0;
 877 #endif
 878
 879 #ifdef CONFIG_MEMCG
 880         tsk->active_memcg = NULL;
 881 #endif
 882         return tsk;
 883
 884 free_stack:
 885         free_thread_stack(tsk);
 886 free_tsk:
 887         free_task_struct(tsk);
 888         return NULL;
 889 }
 890
 891 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 892
 893 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 894
 895 static int __init coredump_filter_setup(char *s)
 896 {
 897         default_dump_filter =
 898                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 899                 MMF_DUMP_FILTER_MASK;
 900         return 1;
 901 }
 902
 903 __setup("coredump_filter=", coredump_filter_setup);
 904
 905 #include <linux/init_task.h>
 906
 907 static void mm_init_aio(struct mm_struct *mm)
 908 {
 909 #ifdef CONFIG_AIO
 910         spin_lock_init(&mm->ioctx_lock);
 911         mm->ioctx_table = NULL;
 912 #endif
 913 }
 914
 915 static __always_inline void mm_clear_owner(struct mm_struct *mm,
 916                                            struct task_struct *p)
 917 {
 918 #ifdef CONFIG_MEMCG
 919         if (mm->owner == p)
 920                 WRITE_ONCE(mm->owner, NULL);
 921 #endif
 922 }
 923
 924 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 925 {
 926 #ifdef CONFIG_MEMCG
 927         mm->owner = p;
 928 #endif
 929 }
 930
 931 static void mm_init_uprobes_state(struct mm_struct *mm)
 932 {
 933 #ifdef CONFIG_UPROBES
 934         mm->uprobes_state.xol_area = NULL;
 935 #endif
 936 }
 937
 938 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 939         struct user_namespace *user_ns)
 940 {
 941         mm->mmap = NULL;
 942         mm->mm_rb = RB_ROOT;
 943         mm->vmacache_seqnum = 0;
 944         atomic_set(&mm->mm_users, 1);
 945         atomic_set(&mm->mm_count, 1);
 946         init_rwsem(&mm->mmap_sem);
 947         INIT_LIST_HEAD(&mm->mmlist);
 948         mm->core_state = NULL;
 949         mm_pgtables_bytes_init(mm);
 950         mm->map_count = 0;
 951         mm->locked_vm = 0;
 952         mm->pinned_vm = 0;
 953         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 954         spin_lock_init(&mm->page_table_lock);
 955         spin_lock_init(&mm->arg_lock);
 956         mm_init_cpumask(mm);
 957         mm_init_aio(mm);
 958         mm_init_owner(mm, p);
 959         RCU_INIT_POINTER(mm->exe_file, NULL);
 960         mmu_notifier_mm_init(mm);
 961         hmm_mm_init(mm);
 962         init_tlb_flush_pending(mm);
 963 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 964         mm->pmd_huge_pte = NULL;
 965 #endif
 966         mm_init_uprobes_state(mm);
 967         hugetlb_count_init(mm);
 968
 969         if (current->mm) {
 970                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 971                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 972         } else {
 973                 mm->flags = default_dump_filter;
 974                 mm->def_flags = 0;
 975         }
 976
 977         if (mm_alloc_pgd(mm))
 978                 goto fail_nopgd;
 979
 980         if (init_new_context(p, mm))
 981                 goto fail_nocontext;
 982
 983         mm->user_ns = get_user_ns(user_ns);
 984         return mm;
 985
 986 fail_nocontext:
 987         mm_free_pgd(mm);
 988 fail_nopgd:
 989         free_mm(mm);
 990         return NULL;
 991 }
 992
 993 /*
 994  * Allocate and initialize an mm_struct.
 995  */
 996 struct mm_struct *mm_alloc(void)
 997 {
 998         struct mm_struct *mm;
 999
1000         mm = allocate_mm();
1001         if (!mm)
1002                 return NULL;
1003
1004         memset(mm, 0, sizeof(*mm));
1005         return mm_init(mm, current, current_user_ns());
1006 }
1007
1008 static inline void __mmput(struct mm_struct *mm)
1009 {
1010         VM_BUG_ON(atomic_read(&mm->mm_users));
1011
1012         uprobe_clear_state(mm);
1013         exit_aio(mm);
1014         ksm_exit(mm);
1015         khugepaged_exit(mm); /* must run before exit_mmap */
1016         exit_mmap(mm);
1017         mm_put_huge_zero_page(mm);
1018         set_mm_exe_file(mm, NULL);
1019         if (!list_empty(&mm->mmlist)) {
1020                 spin_lock(&mmlist_lock);
1021                 list_del(&mm->mmlist);
1022                 spin_unlock(&mmlist_lock);
1023         }
1024         if (mm->binfmt)
1025                 module_put(mm->binfmt->module);
1026         mmdrop(mm);
1027 }
1028
1029 /*
1030  * Decrement the use count and release all resources for an mm.
1031  */
1032 void mmput(struct mm_struct *mm)
1033 {
1034         might_sleep();
1035
1036         if (atomic_dec_and_test(&mm->mm_users))
1037                 __mmput(mm);
1038 }
1039 EXPORT_SYMBOL_GPL(mmput);
1040
1041 #ifdef CONFIG_MMU
1042 static void mmput_async_fn(struct work_struct *work)
1043 {
1044         struct mm_struct *mm = container_of(work, struct mm_struct,
1045                                             async_put_work);
1046
1047         __mmput(mm);
1048 }
1049
1050 void mmput_async(struct mm_struct *mm)
1051 {
1052         if (atomic_dec_and_test(&mm->mm_users)) {
1053                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1054                 schedule_work(&mm->async_put_work);
1055         }
1056 }
1057 #endif
1058
1059 /**
1060  * set_mm_exe_file - change a reference to the mm's executable file
1061  *
1062  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1063  *
1064  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1065  * invocations: in mmput() nobody alive left, in execve task is single
1066  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1067  * mm->exe_file, but does so without using set_mm_exe_file() in order
1068  * to do avoid the need for any locks.
1069  */
1070 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1071 {
1072         struct file *old_exe_file;
1073
1074         /*
1075          * It is safe to dereference the exe_file without RCU as
1076          * this function is only called if nobody else can access
1077          * this mm -- see comment above for justification.
1078          */
1079         old_exe_file = rcu_dereference_raw(mm->exe_file);
1080
1081         if (new_exe_file)
1082                 get_file(new_exe_file);
1083         rcu_assign_pointer(mm->exe_file, new_exe_file);
1084         if (old_exe_file)
1085                 fput(old_exe_file);
1086 }
1087
1088 /**
1089  * get_mm_exe_file - acquire a reference to the mm's executable file
1090  *
1091  * Returns %NULL if mm has no associated executable file.
1092  * User must release file via fput().
1093  */
1094 struct file *get_mm_exe_file(struct mm_struct *mm)
1095 {
1096         struct file *exe_file;
1097
1098         rcu_read_lock();
1099         exe_file = rcu_dereference(mm->exe_file);
1100         if (exe_file && !get_file_rcu(exe_file))
1101                 exe_file = NULL;
1102         rcu_read_unlock();
1103         return exe_file;
1104 }
1105 EXPORT_SYMBOL(get_mm_exe_file);
1106
1107 /**
1108  * get_task_exe_file - acquire a reference to the task's executable file
1109  *
1110  * Returns %NULL if task's mm (if any) has no associated executable file or
1111  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1112  * User must release file via fput().
1113  */
1114 struct file *get_task_exe_file(struct task_struct *task)
1115 {
1116         struct file *exe_file = NULL;
1117         struct mm_struct *mm;
1118
1119         task_lock(task);
1120         mm = task->mm;
1121         if (mm) {
1122                 if (!(task->flags & PF_KTHREAD))
1123                         exe_file = get_mm_exe_file(mm);
1124         }
1125         task_unlock(task);
1126         return exe_file;
1127 }
1128 EXPORT_SYMBOL(get_task_exe_file);
1129
1130 /**
1131  * get_task_mm - acquire a reference to the task's mm
1132  *
1133  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1134  * this kernel workthread has transiently adopted a user mm with use_mm,
1135  * to do its AIO) is not set and if so returns a reference to it, after
1136  * bumping up the use count.  User must release the mm via mmput()
1137  * after use.  Typically used by /proc and ptrace.
1138  */
1139 struct mm_struct *get_task_mm(struct task_struct *task)
1140 {
1141         struct mm_struct *mm;
1142
1143         task_lock(task);
1144         mm = task->mm;
1145         if (mm) {
1146                 if (task->flags & PF_KTHREAD)
1147                         mm = NULL;
1148                 else
1149                         mmget(mm);
1150         }
1151         task_unlock(task);
1152         return mm;
1153 }
1154 EXPORT_SYMBOL_GPL(get_task_mm);
1155
1156 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1157 {
1158         struct mm_struct *mm;
1159         int err;
1160
1161         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1162         if (err)
1163                 return ERR_PTR(err);
1164
1165         mm = get_task_mm(task);
1166         if (mm && mm != current->mm &&
1167                         !ptrace_may_access(task, mode)) {
1168                 mmput(mm);
1169                 mm = ERR_PTR(-EACCES);
1170         }
1171         mutex_unlock(&task->signal->cred_guard_mutex);
1172
1173         return mm;
1174 }
1175
1176 static void complete_vfork_done(struct task_struct *tsk)
1177 {
1178         struct completion *vfork;
1179
1180         task_lock(tsk);
1181         vfork = tsk->vfork_done;
1182         if (likely(vfork)) {
1183                 tsk->vfork_done = NULL;
1184                 complete(vfork);
1185         }
1186         task_unlock(tsk);
1187 }
1188
1189 static int wait_for_vfork_done(struct task_struct *child,
1190                                 struct completion *vfork)
1191 {
1192         int killed;
1193
1194         freezer_do_not_count();
1195         killed = wait_for_completion_killable(vfork);
1196         freezer_count();
1197
1198         if (killed) {
1199                 task_lock(child);
1200                 child->vfork_done = NULL;
1201                 task_unlock(child);
1202         }
1203
1204         put_task_struct(child);
1205         return killed;
1206 }
1207
1208 /* Please note the differences between mmput and mm_release.
1209  * mmput is called whenever we stop holding onto a mm_struct,
1210  * error success whatever.
1211  *
1212  * mm_release is called after a mm_struct has been removed
1213  * from the current process.
1214  *
1215  * This difference is important for error handling, when we
1216  * only half set up a mm_struct for a new process and need to restore
1217  * the old one.  Because we mmput the new mm_struct before
1218  * restoring the old one. . .
1219  * Eric Biederman 10 January 1998
1220  */
1221 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1222 {
1223         uprobe_free_utask(tsk);
1224
1225         /* Get rid of any cached register state */
1226         deactivate_mm(tsk, mm);
1227
1228         /*
1229          * Signal userspace if we're not exiting with a core dump
1230          * because we want to leave the value intact for debugging
1231          * purposes.
1232          */
1233         if (tsk->clear_child_tid) {
1234                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1235                     atomic_read(&mm->mm_users) > 1) {
1236                         /*
1237                          * We don't check the error code - if userspace has
1238                          * not set up a proper pointer then tough luck.
1239                          */
1240                         put_user(0, tsk->clear_child_tid);
1241                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1242                                         1, NULL, NULL, 0, 0);
1243                 }
1244                 tsk->clear_child_tid = NULL;
1245         }
1246
1247         /*
1248          * All done, finally we can wake up parent and return this mm to him.
1249          * Also kthread_stop() uses this completion for synchronization.
1250          */
1251         if (tsk->vfork_done)
1252                 complete_vfork_done(tsk);
1253 }
1254
1255 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1256 {
1257         futex_exit_release(tsk);
1258         mm_release(tsk, mm);
1259 }
1260
1261 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1262 {
1263         futex_exec_release(tsk);
1264         mm_release(tsk, mm);
1265 }
1266
1267 /*
1268  * Allocate a new mm structure and copy contents from the
1269  * mm structure of the passed in task structure.
1270  */
1271 static struct mm_struct *dup_mm(struct task_struct *tsk)
1272 {
1273         struct mm_struct *mm, *oldmm = current->mm;
1274         int err;
1275
1276         mm = allocate_mm();
1277         if (!mm)
1278                 goto fail_nomem;
1279
1280         memcpy(mm, oldmm, sizeof(*mm));
1281
1282         if (!mm_init(mm, tsk, mm->user_ns))
1283                 goto fail_nomem;
1284
1285         err = dup_mmap(mm, oldmm);
1286         if (err)
1287                 goto free_pt;
1288
1289         mm->hiwater_rss = get_mm_rss(mm);
1290         mm->hiwater_vm = mm->total_vm;
1291
1292         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1293                 goto free_pt;
1294
1295         return mm;
1296
1297 free_pt:
1298         /* don't put binfmt in mmput, we haven't got module yet */
1299         mm->binfmt = NULL;
1300         mm_init_owner(mm, NULL);
1301         mmput(mm);
1302
1303 fail_nomem:
1304         return NULL;
1305 }
1306
1307 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1308 {
1309         struct mm_struct *mm, *oldmm;
1310         int retval;
1311
1312         tsk->min_flt = tsk->maj_flt = 0;
1313         tsk->nvcsw = tsk->nivcsw = 0;
1314 #ifdef CONFIG_DETECT_HUNG_TASK
1315         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1316         tsk->last_switch_time = 0;
1317 #endif
1318
1319         tsk->mm = NULL;
1320         tsk->active_mm = NULL;
1321
1322         /*
1323          * Are we cloning a kernel thread?
1324          *
1325          * We need to steal a active VM for that..
1326          */
1327         oldmm = current->mm;
1328         if (!oldmm)
1329                 return 0;
1330
1331         /* initialize the new vmacache entries */
1332         vmacache_flush(tsk);
1333
1334         if (clone_flags & CLONE_VM) {
1335                 mmget(oldmm);
1336                 mm = oldmm;
1337                 goto good_mm;
1338         }
1339
1340         retval = -ENOMEM;
1341         mm = dup_mm(tsk);
1342         if (!mm)
1343                 goto fail_nomem;
1344
1345 good_mm:
1346         tsk->mm = mm;
1347         tsk->active_mm = mm;
1348         return 0;
1349
1350 fail_nomem:
1351         return retval;
1352 }
1353
1354 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1355 {
1356         struct fs_struct *fs = current->fs;
1357         if (clone_flags & CLONE_FS) {
1358                 /* tsk->fs is already what we want */
1359                 spin_lock(&fs->lock);
1360                 if (fs->in_exec) {
1361                         spin_unlock(&fs->lock);
1362                         return -EAGAIN;
1363                 }
1364                 fs->users++;
1365                 spin_unlock(&fs->lock);
1366                 return 0;
1367         }
1368         tsk->fs = copy_fs_struct(fs);
1369         if (!tsk->fs)
1370                 return -ENOMEM;
1371         return 0;
1372 }
1373
1374 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1375 {
1376         struct files_struct *oldf, *newf;
1377         int error = 0;
1378
1379         /*
1380          * A background process may not have any files ...
1381          */
1382         oldf = current->files;
1383         if (!oldf)
1384                 goto out;
1385
1386         if (clone_flags & CLONE_FILES) {
1387                 atomic_inc(&oldf->count);
1388                 goto out;
1389         }
1390
1391         newf = dup_fd(oldf, &error);
1392         if (!newf)
1393                 goto out;
1394
1395         tsk->files = newf;
1396         error = 0;
1397 out:
1398         return error;
1399 }
1400
1401 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1402 {
1403 #ifdef CONFIG_BLOCK
1404         struct io_context *ioc = current->io_context;
1405         struct io_context *new_ioc;
1406
1407         if (!ioc)
1408                 return 0;
1409         /*
1410          * Share io context with parent, if CLONE_IO is set
1411          */
1412         if (clone_flags & CLONE_IO) {
1413                 ioc_task_link(ioc);
1414                 tsk->io_context = ioc;
1415         } else if (ioprio_valid(ioc->ioprio)) {
1416                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1417                 if (unlikely(!new_ioc))
1418                         return -ENOMEM;
1419
1420                 new_ioc->ioprio = ioc->ioprio;
1421                 put_io_context(new_ioc);
1422         }
1423 #endif
1424         return 0;
1425 }
1426
1427 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1428 {
1429         struct sighand_struct *sig;
1430
1431         if (clone_flags & CLONE_SIGHAND) {
1432                 atomic_inc(&current->sighand->count);
1433                 return 0;
1434         }
1435         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1436         rcu_assign_pointer(tsk->sighand, sig);
1437         if (!sig)
1438                 return -ENOMEM;
1439
1440         atomic_set(&sig->count, 1);
1441         spin_lock_irq(&current->sighand->siglock);
1442         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1443         spin_unlock_irq(&current->sighand->siglock);
1444         return 0;
1445 }
1446
1447 void __cleanup_sighand(struct sighand_struct *sighand)
1448 {
1449         if (atomic_dec_and_test(&sighand->count)) {
1450                 signalfd_cleanup(sighand);
1451                 /*
1452                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1453                  * without an RCU grace period, see __lock_task_sighand().
1454                  */
1455                 kmem_cache_free(sighand_cachep, sighand);
1456         }
1457 }
1458
1459 #ifdef CONFIG_POSIX_TIMERS
1460 /*
1461  * Initialize POSIX timer handling for a thread group.
1462  */
1463 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1464 {
1465         unsigned long cpu_limit;
1466
1467         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1468         if (cpu_limit != RLIM_INFINITY) {
1469                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1470                 sig->cputimer.running = true;
1471         }
1472
1473         /* The timer lists. */
1474         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1475         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1476         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1477 }
1478 #else
1479 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1480 #endif
1481
1482 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1483 {
1484         struct signal_struct *sig;
1485
1486         if (clone_flags & CLONE_THREAD)
1487                 return 0;
1488
1489         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1490         tsk->signal = sig;
1491         if (!sig)
1492                 return -ENOMEM;
1493
1494         sig->nr_threads = 1;
1495         atomic_set(&sig->live, 1);
1496         atomic_set(&sig->sigcnt, 1);
1497
1498         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1499         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1500         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1501
1502         init_waitqueue_head(&sig->wait_chldexit);
1503         sig->curr_target = tsk;
1504         init_sigpending(&sig->shared_pending);
1505         INIT_HLIST_HEAD(&sig->multiprocess);
1506         seqlock_init(&sig->stats_lock);
1507         prev_cputime_init(&sig->prev_cputime);
1508
1509 #ifdef CONFIG_POSIX_TIMERS
1510         INIT_LIST_HEAD(&sig->posix_timers);
1511         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1512         sig->real_timer.function = it_real_fn;
1513 #endif
1514
1515         task_lock(current->group_leader);
1516         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1517         task_unlock(current->group_leader);
1518
1519         posix_cpu_timers_init_group(sig);
1520
1521         tty_audit_fork(sig);
1522         sched_autogroup_fork(sig);
1523
1524         sig->oom_score_adj = current->signal->oom_score_adj;
1525         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1526
1527         mutex_init(&sig->cred_guard_mutex);
1528
1529         return 0;
1530 }
1531
1532 static void copy_seccomp(struct task_struct *p)
1533 {
1534 #ifdef CONFIG_SECCOMP
1535         /*
1536          * Must be called with sighand->lock held, which is common to
1537          * all threads in the group. Holding cred_guard_mutex is not
1538          * needed because this new task is not yet running and cannot
1539          * be racing exec.
1540          */
1541         assert_spin_locked(&current->sighand->siglock);
1542
1543         /* Ref-count the new filter user, and assign it. */
1544         get_seccomp_filter(current);
1545         p->seccomp = current->seccomp;
1546
1547         /*
1548          * Explicitly enable no_new_privs here in case it got set
1549          * between the task_struct being duplicated and holding the
1550          * sighand lock. The seccomp state and nnp must be in sync.
1551          */
1552         if (task_no_new_privs(current))
1553                 task_set_no_new_privs(p);
1554
1555         /*
1556          * If the parent gained a seccomp mode after copying thread
1557          * flags and between before we held the sighand lock, we have
1558          * to manually enable the seccomp thread flag here.
1559          */
1560         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1561                 set_tsk_thread_flag(p, TIF_SECCOMP);
1562 #endif
1563 }
1564
1565 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1566 {
1567         current->clear_child_tid = tidptr;
1568
1569         return task_pid_vnr(current);
1570 }
1571
1572 static void rt_mutex_init_task(struct task_struct *p)
1573 {
1574         raw_spin_lock_init(&p->pi_lock);
1575 #ifdef CONFIG_RT_MUTEXES
1576         p->pi_waiters = RB_ROOT_CACHED;
1577         p->pi_top_task = NULL;
1578         p->pi_blocked_on = NULL;
1579 #endif
1580 }
1581
1582 #ifdef CONFIG_POSIX_TIMERS
1583 /*
1584  * Initialize POSIX timer handling for a single task.
1585  */
1586 static void posix_cpu_timers_init(struct task_struct *tsk)
1587 {
1588         tsk->cputime_expires.prof_exp = 0;
1589         tsk->cputime_expires.virt_exp = 0;
1590         tsk->cputime_expires.sched_exp = 0;
1591         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1592         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1593         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1594 }
1595 #else
1596 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1597 #endif
1598
1599 static inline void init_task_pid_links(struct task_struct *task)
1600 {
1601         enum pid_type type;
1602
1603         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1604                 INIT_HLIST_NODE(&task->pid_links[type]);
1605         }
1606 }
1607
1608 static inline void
1609 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1610 {
1611         if (type == PIDTYPE_PID)
1612                 task->thread_pid = pid;
1613         else
1614                 task->signal->pids[type] = pid;
1615 }
1616
1617 static inline void rcu_copy_process(struct task_struct *p)
1618 {
1619 #ifdef CONFIG_PREEMPT_RCU
1620         p->rcu_read_lock_nesting = 0;
1621         p->rcu_read_unlock_special.s = 0;
1622         p->rcu_blocked_node = NULL;
1623         INIT_LIST_HEAD(&p->rcu_node_entry);
1624 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1625 #ifdef CONFIG_TASKS_RCU
1626         p->rcu_tasks_holdout = false;
1627         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1628         p->rcu_tasks_idle_cpu = -1;
1629 #endif /* #ifdef CONFIG_TASKS_RCU */
1630 }
1631
1632 static void __delayed_free_task(struct rcu_head *rhp)
1633 {
1634         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1635
1636         free_task(tsk);
1637 }
1638
1639 static __always_inline void delayed_free_task(struct task_struct *tsk)
1640 {
1641         if (IS_ENABLED(CONFIG_MEMCG))
1642                 call_rcu(&tsk->rcu, __delayed_free_task);
1643         else
1644                 free_task(tsk);
1645 }
1646
1647 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1648 {
1649         /* Skip if kernel thread */
1650         if (!tsk->mm)
1651                 return;
1652
1653         /* Skip if spawning a thread or using vfork */
1654         if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1655                 return;
1656
1657         /* We need to synchronize with __set_oom_adj */
1658         mutex_lock(&oom_adj_mutex);
1659         set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1660         /* Update the values in case they were changed after copy_signal */
1661         tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1662         tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1663         mutex_unlock(&oom_adj_mutex);
1664 }
1665
1666 /*
1667  * This creates a new process as a copy of the old one,
1668  * but does not actually start it yet.
1669  *
1670  * It copies the registers, and all the appropriate
1671  * parts of the process environment (as per the clone
1672  * flags). The actual kick-off is left to the caller.
1673  */
1674 static __latent_entropy struct task_struct *copy_process(
1675                                         unsigned long clone_flags,
1676                                         unsigned long stack_start,
1677                                         unsigned long stack_size,
1678                                         int __user *child_tidptr,
1679                                         struct pid *pid,
1680                                         int trace,
1681                                         unsigned long tls,
1682                                         int node)
1683 {
1684         int retval;
1685         struct task_struct *p;
1686         struct multiprocess_signals delayed;
1687
1688         /*
1689          * Don't allow sharing the root directory with processes in a different
1690          * namespace
1691          */
1692         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1693                 return ERR_PTR(-EINVAL);
1694
1695         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1696                 return ERR_PTR(-EINVAL);
1697
1698         /*
1699          * Thread groups must share signals as well, and detached threads
1700          * can only be started up within the thread group.
1701          */
1702         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1703                 return ERR_PTR(-EINVAL);
1704
1705         /*
1706          * Shared signal handlers imply shared VM. By way of the above,
1707          * thread groups also imply shared VM. Blocking this case allows
1708          * for various simplifications in other code.
1709          */
1710         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1711                 return ERR_PTR(-EINVAL);
1712
1713         /*
1714          * Siblings of global init remain as zombies on exit since they are
1715          * not reaped by their parent (swapper). To solve this and to avoid
1716          * multi-rooted process trees, prevent global and container-inits
1717          * from creating siblings.
1718          */
1719         if ((clone_flags & CLONE_PARENT) &&
1720                                 current->signal->flags & SIGNAL_UNKILLABLE)
1721                 return ERR_PTR(-EINVAL);
1722
1723         /*
1724          * If the new process will be in a different pid or user namespace
1725          * do not allow it to share a thread group with the forking task.
1726          */
1727         if (clone_flags & CLONE_THREAD) {
1728                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1729                     (task_active_pid_ns(current) !=
1730                                 current->nsproxy->pid_ns_for_children))
1731                         return ERR_PTR(-EINVAL);
1732         }
1733
1734         /*
1735          * Force any signals received before this point to be delivered
1736          * before the fork happens.  Collect up signals sent to multiple
1737          * processes that happen during the fork and delay them so that
1738          * they appear to happen after the fork.
1739          */
1740         sigemptyset(&delayed.signal);
1741         INIT_HLIST_NODE(&delayed.node);
1742
1743         spin_lock_irq(&current->sighand->siglock);
1744         if (!(clone_flags & CLONE_THREAD))
1745                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1746         recalc_sigpending();
1747         spin_unlock_irq(&current->sighand->siglock);
1748         retval = -ERESTARTNOINTR;
1749         if (signal_pending(current))
1750                 goto fork_out;
1751
1752         retval = -ENOMEM;
1753         p = dup_task_struct(current, node);
1754         if (!p)
1755                 goto fork_out;
1756
1757         /*
1758          * This _must_ happen before we call free_task(), i.e. before we jump
1759          * to any of the bad_fork_* labels. This is to avoid freeing
1760          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1761          * kernel threads (PF_KTHREAD).
1762          */
1763         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1764         /*
1765          * Clear TID on mm_release()?
1766          */
1767         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1768
1769         ftrace_graph_init_task(p);
1770
1771         rt_mutex_init_task(p);
1772
1773 #ifdef CONFIG_PROVE_LOCKING
1774         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1775         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1776 #endif
1777         retval = -EAGAIN;
1778         if (atomic_read(&p->real_cred->user->processes) >=
1779                         task_rlimit(p, RLIMIT_NPROC)) {
1780                 if (p->real_cred->user != INIT_USER &&
1781                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1782                         goto bad_fork_free;
1783         }
1784         current->flags &= ~PF_NPROC_EXCEEDED;
1785
1786         retval = copy_creds(p, clone_flags);
1787         if (retval < 0)
1788                 goto bad_fork_free;
1789
1790         /*
1791          * If multiple threads are within copy_process(), then this check
1792          * triggers too late. This doesn't hurt, the check is only there
1793          * to stop root fork bombs.
1794          */
1795         retval = -EAGAIN;
1796         if (nr_threads >= max_threads)
1797                 goto bad_fork_cleanup_count;
1798
1799         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1800         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1801         p->flags |= PF_FORKNOEXEC;
1802         INIT_LIST_HEAD(&p->children);
1803         INIT_LIST_HEAD(&p->sibling);
1804         rcu_copy_process(p);
1805         p->vfork_done = NULL;
1806         spin_lock_init(&p->alloc_lock);
1807
1808         init_sigpending(&p->pending);
1809
1810         p->utime = p->stime = p->gtime = 0;
1811 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1812         p->utimescaled = p->stimescaled = 0;
1813 #endif
1814         prev_cputime_init(&p->prev_cputime);
1815
1816 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1817         seqcount_init(&p->vtime.seqcount);
1818         p->vtime.starttime = 0;
1819         p->vtime.state = VTIME_INACTIVE;
1820 #endif
1821
1822 #if defined(SPLIT_RSS_COUNTING)
1823         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1824 #endif
1825
1826         p->default_timer_slack_ns = current->timer_slack_ns;
1827
1828         task_io_accounting_init(&p->ioac);
1829         acct_clear_integrals(p);
1830
1831         posix_cpu_timers_init(p);
1832
1833         p->io_context = NULL;
1834         audit_set_context(p, NULL);
1835         cgroup_fork(p);
1836 #ifdef CONFIG_NUMA
1837         p->mempolicy = mpol_dup(p->mempolicy);
1838         if (IS_ERR(p->mempolicy)) {
1839                 retval = PTR_ERR(p->mempolicy);
1840                 p->mempolicy = NULL;
1841                 goto bad_fork_cleanup_threadgroup_lock;
1842         }
1843 #endif
1844 #ifdef CONFIG_CPUSETS
1845         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1846         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1847         seqcount_init(&p->mems_allowed_seq);
1848 #endif
1849 #ifdef CONFIG_TRACE_IRQFLAGS
1850         p->irq_events = 0;
1851         p->hardirqs_enabled = 0;
1852         p->hardirq_enable_ip = 0;
1853         p->hardirq_enable_event = 0;
1854         p->hardirq_disable_ip = _THIS_IP_;
1855         p->hardirq_disable_event = 0;
1856         p->softirqs_enabled = 1;
1857         p->softirq_enable_ip = _THIS_IP_;
1858         p->softirq_enable_event = 0;
1859         p->softirq_disable_ip = 0;
1860         p->softirq_disable_event = 0;
1861         p->hardirq_context = 0;
1862         p->softirq_context = 0;
1863 #endif
1864
1865         p->pagefault_disabled = 0;
1866
1867 #ifdef CONFIG_LOCKDEP
1868         p->lockdep_depth = 0; /* no locks held yet */
1869         p->curr_chain_key = 0;
1870         p->lockdep_recursion = 0;
1871         lockdep_init_task(p);
1872 #endif
1873
1874 #ifdef CONFIG_DEBUG_MUTEXES
1875         p->blocked_on = NULL; /* not blocked yet */
1876 #endif
1877 #ifdef CONFIG_BCACHE
1878         p->sequential_io        = 0;
1879         p->sequential_io_avg    = 0;
1880 #endif
1881
1882         /* Perform scheduler related setup. Assign this task to a CPU. */
1883         retval = sched_fork(clone_flags, p);
1884         if (retval)
1885                 goto bad_fork_cleanup_policy;
1886
1887         retval = perf_event_init_task(p);
1888         if (retval)
1889                 goto bad_fork_cleanup_policy;
1890         retval = audit_alloc(p);
1891         if (retval)
1892                 goto bad_fork_cleanup_perf;
1893         /* copy all the process information */
1894         shm_init_task(p);
1895         retval = security_task_alloc(p, clone_flags);
1896         if (retval)
1897                 goto bad_fork_cleanup_audit;
1898         retval = copy_semundo(clone_flags, p);
1899         if (retval)
1900                 goto bad_fork_cleanup_security;
1901         retval = copy_files(clone_flags, p);
1902         if (retval)
1903                 goto bad_fork_cleanup_semundo;
1904         retval = copy_fs(clone_flags, p);
1905         if (retval)
1906                 goto bad_fork_cleanup_files;
1907         retval = copy_sighand(clone_flags, p);
1908         if (retval)
1909                 goto bad_fork_cleanup_fs;
1910         retval = copy_signal(clone_flags, p);
1911         if (retval)
1912                 goto bad_fork_cleanup_sighand;
1913         retval = copy_mm(clone_flags, p);
1914         if (retval)
1915                 goto bad_fork_cleanup_signal;
1916         retval = copy_namespaces(clone_flags, p);
1917         if (retval)
1918                 goto bad_fork_cleanup_mm;
1919         retval = copy_io(clone_flags, p);
1920         if (retval)
1921                 goto bad_fork_cleanup_namespaces;
1922         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1923         if (retval)
1924                 goto bad_fork_cleanup_io;
1925
1926         if (pid != &init_struct_pid) {
1927                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1928                 if (IS_ERR(pid)) {
1929                         retval = PTR_ERR(pid);
1930                         goto bad_fork_cleanup_thread;
1931                 }
1932         }
1933
1934 #ifdef CONFIG_BLOCK
1935         p->plug = NULL;
1936 #endif
1937         futex_init_task(p);
1938
1939         /*
1940          * sigaltstack should be cleared when sharing the same VM
1941          */
1942         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1943                 sas_ss_reset(p);
1944
1945         /*
1946          * Syscall tracing and stepping should be turned off in the
1947          * child regardless of CLONE_PTRACE.
1948          */
1949         user_disable_single_step(p);
1950         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1951 #ifdef TIF_SYSCALL_EMU
1952         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1953 #endif
1954         clear_all_latency_tracing(p);
1955
1956         /* ok, now we should be set up.. */
1957         p->pid = pid_nr(pid);
1958         if (clone_flags & CLONE_THREAD) {
1959                 p->group_leader = current->group_leader;
1960                 p->tgid = current->tgid;
1961         } else {
1962                 p->group_leader = p;
1963                 p->tgid = p->pid;
1964         }
1965
1966         p->nr_dirtied = 0;
1967         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1968         p->dirty_paused_when = 0;
1969
1970         p->pdeath_signal = 0;
1971         INIT_LIST_HEAD(&p->thread_group);
1972         p->task_works = NULL;
1973
1974         cgroup_threadgroup_change_begin(current);
1975         /*
1976          * Ensure that the cgroup subsystem policies allow the new process to be
1977          * forked. It should be noted the the new process's css_set can be changed
1978          * between here and cgroup_post_fork() if an organisation operation is in
1979          * progress.
1980          */
1981         retval = cgroup_can_fork(p);
1982         if (retval)
1983                 goto bad_fork_free_pid;
1984
1985         /*
1986          * From this point on we must avoid any synchronous user-space
1987          * communication until we take the tasklist-lock. In particular, we do
1988          * not want user-space to be able to predict the process start-time by
1989          * stalling fork(2) after we recorded the start_time but before it is
1990          * visible to the system.
1991          */
1992
1993         p->start_time = ktime_get_ns();
1994         p->real_start_time = ktime_get_boot_ns();
1995
1996         /*
1997          * Make it visible to the rest of the system, but dont wake it up yet.
1998          * Need tasklist lock for parent etc handling!
1999          */
2000         write_lock_irq(&tasklist_lock);
2001
2002         /* CLONE_PARENT re-uses the old parent */
2003         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2004                 p->real_parent = current->real_parent;
2005                 p->parent_exec_id = current->parent_exec_id;
2006                 if (clone_flags & CLONE_THREAD)
2007                         p->exit_signal = -1;
2008                 else
2009                         p->exit_signal = current->group_leader->exit_signal;
2010         } else {
2011                 p->real_parent = current;
2012                 p->parent_exec_id = current->self_exec_id;
2013                 p->exit_signal = (clone_flags & CSIGNAL);
2014         }
2015
2016         klp_copy_process(p);
2017
2018         spin_lock(&current->sighand->siglock);
2019
2020         /*
2021          * Copy seccomp details explicitly here, in case they were changed
2022          * before holding sighand lock.
2023          */
2024         copy_seccomp(p);
2025
2026         rseq_fork(p, clone_flags);
2027
2028         /* Don't start children in a dying pid namespace */
2029         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2030                 retval = -ENOMEM;
2031                 goto bad_fork_cancel_cgroup;
2032         }
2033
2034         /* Let kill terminate clone/fork in the middle */
2035         if (fatal_signal_pending(current)) {
2036                 retval = -EINTR;
2037                 goto bad_fork_cancel_cgroup;
2038         }
2039
2040
2041         init_task_pid_links(p);
2042         if (likely(p->pid)) {
2043                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2044
2045                 init_task_pid(p, PIDTYPE_PID, pid);
2046                 if (thread_group_leader(p)) {
2047                         init_task_pid(p, PIDTYPE_TGID, pid);
2048                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2049                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2050
2051                         if (is_child_reaper(pid)) {
2052                                 ns_of_pid(pid)->child_reaper = p;
2053                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2054                         }
2055                         p->signal->shared_pending.signal = delayed.signal;
2056                         p->signal->tty = tty_kref_get(current->signal->tty);
2057                         /*
2058                          * Inherit has_child_subreaper flag under the same
2059                          * tasklist_lock with adding child to the process tree
2060                          * for propagate_has_child_subreaper optimization.
2061                          */
2062                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2063                                                          p->real_parent->signal->is_child_subreaper;
2064                         list_add_tail(&p->sibling, &p->real_parent->children);
2065                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2066                         attach_pid(p, PIDTYPE_TGID);
2067                         attach_pid(p, PIDTYPE_PGID);
2068                         attach_pid(p, PIDTYPE_SID);
2069                         __this_cpu_inc(process_counts);
2070                 } else {
2071                         current->signal->nr_threads++;
2072                         atomic_inc(&current->signal->live);
2073                         atomic_inc(&current->signal->sigcnt);
2074                         task_join_group_stop(p);
2075                         list_add_tail_rcu(&p->thread_group,
2076                                           &p->group_leader->thread_group);
2077                         list_add_tail_rcu(&p->thread_node,
2078                                           &p->signal->thread_head);
2079                 }
2080                 attach_pid(p, PIDTYPE_PID);
2081                 nr_threads++;
2082         }
2083         total_forks++;
2084         hlist_del_init(&delayed.node);
2085         spin_unlock(&current->sighand->siglock);
2086         syscall_tracepoint_update(p);
2087         write_unlock_irq(&tasklist_lock);
2088
2089         proc_fork_connector(p);
2090         cgroup_post_fork(p);
2091         cgroup_threadgroup_change_end(current);
2092         perf_event_fork(p);
2093
2094         trace_task_newtask(p, clone_flags);
2095         uprobe_copy_process(p, clone_flags);
2096
2097         copy_oom_score_adj(clone_flags, p);
2098
2099         return p;
2100
2101 bad_fork_cancel_cgroup:
2102         spin_unlock(&current->sighand->siglock);
2103         write_unlock_irq(&tasklist_lock);
2104         cgroup_cancel_fork(p);
2105 bad_fork_free_pid:
2106         cgroup_threadgroup_change_end(current);
2107         if (pid != &init_struct_pid)
2108                 free_pid(pid);
2109 bad_fork_cleanup_thread:
2110         exit_thread(p);
2111 bad_fork_cleanup_io:
2112         if (p->io_context)
2113                 exit_io_context(p);
2114 bad_fork_cleanup_namespaces:
2115         exit_task_namespaces(p);
2116 bad_fork_cleanup_mm:
2117         if (p->mm) {
2118                 mm_clear_owner(p->mm, p);
2119                 mmput(p->mm);
2120         }
2121 bad_fork_cleanup_signal:
2122         if (!(clone_flags & CLONE_THREAD))
2123                 free_signal_struct(p->signal);
2124 bad_fork_cleanup_sighand:
2125         __cleanup_sighand(p->sighand);
2126 bad_fork_cleanup_fs:
2127         exit_fs(p); /* blocking */
2128 bad_fork_cleanup_files:
2129         exit_files(p); /* blocking */
2130 bad_fork_cleanup_semundo:
2131         exit_sem(p);
2132 bad_fork_cleanup_security:
2133         security_task_free(p);
2134 bad_fork_cleanup_audit:
2135         audit_free(p);
2136 bad_fork_cleanup_perf:
2137         perf_event_free_task(p);
2138 bad_fork_cleanup_policy:
2139         lockdep_free_task(p);
2140 #ifdef CONFIG_NUMA
2141         mpol_put(p->mempolicy);
2142 bad_fork_cleanup_threadgroup_lock:
2143 #endif
2144         delayacct_tsk_free(p);
2145 bad_fork_cleanup_count:
2146         atomic_dec(&p->cred->user->processes);
2147         exit_creds(p);
2148 bad_fork_free:
2149         p->state = TASK_DEAD;
2150         put_task_stack(p);
2151         delayed_free_task(p);
2152 fork_out:
2153         spin_lock_irq(&current->sighand->siglock);
2154         hlist_del_init(&delayed.node);
2155         spin_unlock_irq(&current->sighand->siglock);
2156         return ERR_PTR(retval);
2157 }
2158
2159 static inline void init_idle_pids(struct task_struct *idle)
2160 {
2161         enum pid_type type;
2162
2163         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2164                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2165                 init_task_pid(idle, type, &init_struct_pid);
2166         }
2167 }
2168
2169 struct task_struct *fork_idle(int cpu)
2170 {
2171         struct task_struct *task;
2172         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
2173                             cpu_to_node(cpu));
2174         if (!IS_ERR(task)) {
2175                 init_idle_pids(task);
2176                 init_idle(task, cpu);
2177         }
2178
2179         return task;
2180 }
2181
2182 /*
2183  *  Ok, this is the main fork-routine.
2184  *
2185  * It copies the process, and if successful kick-starts
2186  * it and waits for it to finish using the VM if required.
2187  */
2188 long _do_fork(unsigned long clone_flags,
2189               unsigned long stack_start,
2190               unsigned long stack_size,
2191               int __user *parent_tidptr,
2192               int __user *child_tidptr,
2193               unsigned long tls)
2194 {
2195         struct completion vfork;
2196         struct pid *pid;
2197         struct task_struct *p;
2198         int trace = 0;
2199         long nr;
2200
2201         /*
2202          * Determine whether and which event to report to ptracer.  When
2203          * called from kernel_thread or CLONE_UNTRACED is explicitly
2204          * requested, no event is reported; otherwise, report if the event
2205          * for the type of forking is enabled.
2206          */
2207         if (!(clone_flags & CLONE_UNTRACED)) {
2208                 if (clone_flags & CLONE_VFORK)
2209                         trace = PTRACE_EVENT_VFORK;
2210                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2211                         trace = PTRACE_EVENT_CLONE;
2212                 else
2213                         trace = PTRACE_EVENT_FORK;
2214
2215                 if (likely(!ptrace_event_enabled(current, trace)))
2216                         trace = 0;
2217         }
2218
2219         p = copy_process(clone_flags, stack_start, stack_size,
2220                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2221         add_latent_entropy();
2222
2223         if (IS_ERR(p))
2224                 return PTR_ERR(p);
2225
2226         /*
2227          * Do this prior waking up the new thread - the thread pointer
2228          * might get invalid after that point, if the thread exits quickly.
2229          */
2230         trace_sched_process_fork(current, p);
2231
2232         pid = get_task_pid(p, PIDTYPE_PID);
2233         nr = pid_vnr(pid);
2234
2235         if (clone_flags & CLONE_PARENT_SETTID)
2236                 put_user(nr, parent_tidptr);
2237
2238         if (clone_flags & CLONE_VFORK) {
2239                 p->vfork_done = &vfork;
2240                 init_completion(&vfork);
2241                 get_task_struct(p);
2242         }
2243
2244         wake_up_new_task(p);
2245
2246         /* forking complete and child started to run, tell ptracer */
2247         if (unlikely(trace))
2248                 ptrace_event_pid(trace, pid);
2249
2250         if (clone_flags & CLONE_VFORK) {
2251                 if (!wait_for_vfork_done(p, &vfork))
2252                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2253         }
2254
2255         put_pid(pid);
2256         return nr;
2257 }
2258
2259 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2260 /* For compatibility with architectures that call do_fork directly rather than
2261  * using the syscall entry points below. */
2262 long do_fork(unsigned long clone_flags,
2263               unsigned long stack_start,
2264               unsigned long stack_size,
2265               int __user *parent_tidptr,
2266               int __user *child_tidptr)
2267 {
2268         return _do_fork(clone_flags, stack_start, stack_size,
2269                         parent_tidptr, child_tidptr, 0);
2270 }
2271 #endif
2272
2273 /*
2274  * Create a kernel thread.
2275  */
2276 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2277 {
2278         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2279                 (unsigned long)arg, NULL, NULL, 0);
2280 }
2281
2282 #ifdef __ARCH_WANT_SYS_FORK
2283 SYSCALL_DEFINE0(fork)
2284 {
2285 #ifdef CONFIG_MMU
2286         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2287 #else
2288         /* can not support in nommu mode */
2289         return -EINVAL;
2290 #endif
2291 }
2292 #endif
2293
2294 #ifdef __ARCH_WANT_SYS_VFORK
2295 SYSCALL_DEFINE0(vfork)
2296 {
2297         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2298                         0, NULL, NULL, 0);
2299 }
2300 #endif
2301
2302 #ifdef __ARCH_WANT_SYS_CLONE
2303 #ifdef CONFIG_CLONE_BACKWARDS
2304 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2305                  int __user *, parent_tidptr,
2306                  unsigned long, tls,
2307                  int __user *, child_tidptr)
2308 #elif defined(CONFIG_CLONE_BACKWARDS2)
2309 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2310                  int __user *, parent_tidptr,
2311                  int __user *, child_tidptr,
2312                  unsigned long, tls)
2313 #elif defined(CONFIG_CLONE_BACKWARDS3)
2314 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2315                 int, stack_size,
2316                 int __user *, parent_tidptr,
2317                 int __user *, child_tidptr,
2318                 unsigned long, tls)
2319 #else
2320 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2321                  int __user *, parent_tidptr,
2322                  int __user *, child_tidptr,
2323                  unsigned long, tls)
2324 #endif
2325 {
2326         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2327 }
2328 #endif
2329
2330 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2331 {
2332         struct task_struct *leader, *parent, *child;
2333         int res;
2334
2335         read_lock(&tasklist_lock);
2336         leader = top = top->group_leader;
2337 down:
2338         for_each_thread(leader, parent) {
2339                 list_for_each_entry(child, &parent->children, sibling) {
2340                         res = visitor(child, data);
2341                         if (res) {
2342                                 if (res < 0)
2343                                         goto out;
2344                                 leader = child;
2345                                 goto down;
2346                         }
2347 up:
2348                         ;
2349                 }
2350         }
2351
2352         if (leader != top) {
2353                 child = leader;
2354                 parent = child->real_parent;
2355                 leader = parent->group_leader;
2356                 goto up;
2357         }
2358 out:
2359         read_unlock(&tasklist_lock);
2360 }
2361
2362 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2363 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2364 #endif
2365
2366 static void sighand_ctor(void *data)
2367 {
2368         struct sighand_struct *sighand = data;
2369
2370         spin_lock_init(&sighand->siglock);
2371         init_waitqueue_head(&sighand->signalfd_wqh);
2372 }
2373
2374 void __init proc_caches_init(void)
2375 {
2376         unsigned int mm_size;
2377
2378         sighand_cachep = kmem_cache_create("sighand_cache",
2379                         sizeof(struct sighand_struct), 0,
2380                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2381                         SLAB_ACCOUNT, sighand_ctor);
2382         signal_cachep = kmem_cache_create("signal_cache",
2383                         sizeof(struct signal_struct), 0,
2384                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2385                         NULL);
2386         files_cachep = kmem_cache_create("files_cache",
2387                         sizeof(struct files_struct), 0,
2388                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2389                         NULL);
2390         fs_cachep = kmem_cache_create("fs_cache",
2391                         sizeof(struct fs_struct), 0,
2392                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2393                         NULL);
2394
2395         /*
2396          * The mm_cpumask is located at the end of mm_struct, and is
2397          * dynamically sized based on the maximum CPU number this system
2398          * can have, taking hotplug into account (nr_cpu_ids).
2399          */
2400         mm_size = sizeof(struct mm_struct) + cpumask_size();
2401
2402         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2403                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2404                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2405                         offsetof(struct mm_struct, saved_auxv),
2406                         sizeof_field(struct mm_struct, saved_auxv),
2407                         NULL);
2408         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2409         mmap_init();
2410         nsproxy_cache_init();
2411 }
2412
2413 /*
2414  * Check constraints on flags passed to the unshare system call.
2415  */
2416 static int check_unshare_flags(unsigned long unshare_flags)
2417 {
2418         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2419                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2420                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2421                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2422                 return -EINVAL;
2423         /*
2424          * Not implemented, but pretend it works if there is nothing
2425          * to unshare.  Note that unsharing the address space or the
2426          * signal handlers also need to unshare the signal queues (aka
2427          * CLONE_THREAD).
2428          */
2429         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2430                 if (!thread_group_empty(current))
2431                         return -EINVAL;
2432         }
2433         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2434                 if (atomic_read(&current->sighand->count) > 1)
2435                         return -EINVAL;
2436         }
2437         if (unshare_flags & CLONE_VM) {
2438                 if (!current_is_single_threaded())
2439                         return -EINVAL;
2440         }
2441
2442         return 0;
2443 }
2444
2445 /*
2446  * Unshare the filesystem structure if it is being shared
2447  */
2448 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2449 {
2450         struct fs_struct *fs = current->fs;
2451
2452         if (!(unshare_flags & CLONE_FS) || !fs)
2453                 return 0;
2454
2455         /* don't need lock here; in the worst case we'll do useless copy */
2456         if (fs->users == 1)
2457                 return 0;
2458
2459         *new_fsp = copy_fs_struct(fs);
2460         if (!*new_fsp)
2461                 return -ENOMEM;
2462
2463         return 0;
2464 }
2465
2466 /*
2467  * Unshare file descriptor table if it is being shared
2468  */
2469 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2470 {
2471         struct files_struct *fd = current->files;
2472         int error = 0;
2473
2474         if ((unshare_flags & CLONE_FILES) &&
2475             (fd && atomic_read(&fd->count) > 1)) {
2476                 *new_fdp = dup_fd(fd, &error);
2477                 if (!*new_fdp)
2478                         return error;
2479         }
2480
2481         return 0;
2482 }
2483
2484 /*
2485  * unshare allows a process to 'unshare' part of the process
2486  * context which was originally shared using clone.  copy_*
2487  * functions used by do_fork() cannot be used here directly
2488  * because they modify an inactive task_struct that is being
2489  * constructed. Here we are modifying the current, active,
2490  * task_struct.
2491  */
2492 int ksys_unshare(unsigned long unshare_flags)
2493 {
2494         struct fs_struct *fs, *new_fs = NULL;
2495         struct files_struct *fd, *new_fd = NULL;
2496         struct cred *new_cred = NULL;
2497         struct nsproxy *new_nsproxy = NULL;
2498         int do_sysvsem = 0;
2499         int err;
2500
2501         /*
2502          * If unsharing a user namespace must also unshare the thread group
2503          * and unshare the filesystem root and working directories.
2504          */
2505         if (unshare_flags & CLONE_NEWUSER)
2506                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2507         /*
2508          * If unsharing vm, must also unshare signal handlers.
2509          */
2510         if (unshare_flags & CLONE_VM)
2511                 unshare_flags |= CLONE_SIGHAND;
2512         /*
2513          * If unsharing a signal handlers, must also unshare the signal queues.
2514          */
2515         if (unshare_flags & CLONE_SIGHAND)
2516                 unshare_flags |= CLONE_THREAD;
2517         /*
2518          * If unsharing namespace, must also unshare filesystem information.
2519          */
2520         if (unshare_flags & CLONE_NEWNS)
2521                 unshare_flags |= CLONE_FS;
2522
2523         err = check_unshare_flags(unshare_flags);
2524         if (err)
2525                 goto bad_unshare_out;
2526         /*
2527          * CLONE_NEWIPC must also detach from the undolist: after switching
2528          * to a new ipc namespace, the semaphore arrays from the old
2529          * namespace are unreachable.
2530          */
2531         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2532                 do_sysvsem = 1;
2533         err = unshare_fs(unshare_flags, &new_fs);
2534         if (err)
2535                 goto bad_unshare_out;
2536         err = unshare_fd(unshare_flags, &new_fd);
2537         if (err)
2538                 goto bad_unshare_cleanup_fs;
2539         err = unshare_userns(unshare_flags, &new_cred);
2540         if (err)
2541                 goto bad_unshare_cleanup_fd;
2542         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2543                                          new_cred, new_fs);
2544         if (err)
2545                 goto bad_unshare_cleanup_cred;
2546
2547         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2548                 if (do_sysvsem) {
2549                         /*
2550                          * CLONE_SYSVSEM is equivalent to sys_exit().
2551                          */
2552                         exit_sem(current);
2553                 }
2554                 if (unshare_flags & CLONE_NEWIPC) {
2555                         /* Orphan segments in old ns (see sem above). */
2556                         exit_shm(current);
2557                         shm_init_task(current);
2558                 }
2559
2560                 if (new_nsproxy)
2561                         switch_task_namespaces(current, new_nsproxy);
2562
2563                 task_lock(current);
2564
2565                 if (new_fs) {
2566                         fs = current->fs;
2567                         spin_lock(&fs->lock);
2568                         current->fs = new_fs;
2569                         if (--fs->users)
2570                                 new_fs = NULL;
2571                         else
2572                                 new_fs = fs;
2573                         spin_unlock(&fs->lock);
2574                 }
2575
2576                 if (new_fd) {
2577                         fd = current->files;
2578                         current->files = new_fd;
2579                         new_fd = fd;
2580                 }
2581
2582                 task_unlock(current);
2583
2584                 if (new_cred) {
2585                         /* Install the new user namespace */
2586                         commit_creds(new_cred);
2587                         new_cred = NULL;
2588                 }
2589         }
2590
2591         perf_event_namespaces(current);
2592
2593 bad_unshare_cleanup_cred:
2594         if (new_cred)
2595                 put_cred(new_cred);
2596 bad_unshare_cleanup_fd:
2597         if (new_fd)
2598                 put_files_struct(new_fd);
2599
2600 bad_unshare_cleanup_fs:
2601         if (new_fs)
2602                 free_fs_struct(new_fs);
2603
2604 bad_unshare_out:
2605         return err;
2606 }
2607
2608 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2609 {
2610         return ksys_unshare(unshare_flags);
2611 }
2612
2613 /*
2614  *      Helper to unshare the files of the current task.
2615  *      We don't want to expose copy_files internals to
2616  *      the exec layer of the kernel.
2617  */
2618
2619 int unshare_files(struct files_struct **displaced)
2620 {
2621         struct task_struct *task = current;
2622         struct files_struct *copy = NULL;
2623         int error;
2624
2625         error = unshare_fd(CLONE_FILES, &copy);
2626         if (error || !copy) {
2627                 *displaced = NULL;
2628                 return error;
2629         }
2630         *displaced = task->files;
2631         task_lock(task);
2632         task->files = copy;
2633         task_unlock(task);
2634         return 0;
2635 }
2636
2637 int sysctl_max_threads(struct ctl_table *table, int write,
2638                        void __user *buffer, size_t *lenp, loff_t *ppos)
2639 {
2640         struct ctl_table t;
2641         int ret;
2642         int threads = max_threads;
2643         int min = 1;
2644         int max = MAX_THREADS;
2645
2646         t = *table;
2647         t.data = &threads;
2648         t.extra1 = &min;
2649         t.extra2 = &max;
2650
2651         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2652         if (ret || !write)
2653                 return ret;
2654
2655         max_threads = threads;
2656
2657         return 0;
2658 }