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/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/nsproxy.h>
  32 #include <linux/capability.h>
  33 #include <linux/cpu.h>
  34 #include <linux/cgroup.h>
  35 #include <linux/security.h>
  36 #include <linux/hugetlb.h>
  37 #include <linux/seccomp.h>
  38 #include <linux/swap.h>
  39 #include <linux/syscalls.h>
  40 #include <linux/jiffies.h>
  41 #include <linux/futex.h>
  42 #include <linux/compat.h>
  43 #include <linux/kthread.h>
  44 #include <linux/task_io_accounting_ops.h>
  45 #include <linux/rcupdate.h>
  46 #include <linux/ptrace.h>
  47 #include <linux/mount.h>
  48 #include <linux/audit.h>
  49 #include <linux/memcontrol.h>
  50 #include <linux/ftrace.h>
  51 #include <linux/proc_fs.h>
  52 #include <linux/profile.h>
  53 #include <linux/rmap.h>
  54 #include <linux/ksm.h>
  55 #include <linux/acct.h>
  56 #include <linux/tsacct_kern.h>
  57 #include <linux/cn_proc.h>
  58 #include <linux/freezer.h>
  59 #include <linux/delayacct.h>
  60 #include <linux/taskstats_kern.h>
  61 #include <linux/random.h>
  62 #include <linux/tty.h>
  63 #include <linux/blkdev.h>
  64 #include <linux/fs_struct.h>
  65 #include <linux/magic.h>
  66 #include <linux/perf_event.h>
  67 #include <linux/posix-timers.h>
  68 #include <linux/user-return-notifier.h>
  69 #include <linux/oom.h>
  70 #include <linux/khugepaged.h>
  71 #include <linux/signalfd.h>
  72 #include <linux/uprobes.h>
  73 #include <linux/aio.h>
  74
  75 #include <asm/pgtable.h>
  76 #include <asm/pgalloc.h>
  77 #include <asm/uaccess.h>
  78 #include <asm/mmu_context.h>
  79 #include <asm/cacheflush.h>
  80 #include <asm/tlbflush.h>
  81
  82 #include <trace/events/sched.h>
  83
  84 #define CREATE_TRACE_POINTS
  85 #include <trace/events/task.h>
  86
  87 /*
  88  * Protected counters by write_lock_irq(&tasklist_lock)
  89  */
  90 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  91 int nr_threads;                 /* The idle threads do not count.. */
  92
  93 int max_threads;                /* tunable limit on nr_threads */
  94
  95 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  96
  97 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  98
  99 #ifdef CONFIG_PROVE_RCU
 100 int lockdep_tasklist_lock_is_held(void)
 101 {
 102         return lockdep_is_held(&tasklist_lock);
 103 }
 104 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 105 #endif /* #ifdef CONFIG_PROVE_RCU */
 106
 107 int nr_processes(void)
 108 {
 109         int cpu;
 110         int total = 0;
 111
 112         for_each_possible_cpu(cpu)
 113                 total += per_cpu(process_counts, cpu);
 114
 115         return total;
 116 }
 117
 118 void __weak arch_release_task_struct(struct task_struct *tsk)
 119 {
 120 }
 121
 122 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 123 static struct kmem_cache *task_struct_cachep;
 124
 125 static inline struct task_struct *alloc_task_struct_node(int node)
 126 {
 127         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 128 }
 129
 130 static inline void free_task_struct(struct task_struct *tsk)
 131 {
 132         kmem_cache_free(task_struct_cachep, tsk);
 133 }
 134 #endif
 135
 136 void __weak arch_release_thread_info(struct thread_info *ti)
 137 {
 138 }
 139
 140 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
 141
 142 /*
 143  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 144  * kmemcache based allocator.
 145  */
 146 # if THREAD_SIZE >= PAGE_SIZE
 147 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 148                                                   int node)
 149 {
 150         struct page *page = alloc_pages_node(node, THREADINFO_GFP_ACCOUNTED,
 151                                              THREAD_SIZE_ORDER);
 152
 153         return page ? page_address(page) : NULL;
 154 }
 155
 156 static inline void free_thread_info(struct thread_info *ti)
 157 {
 158         free_memcg_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 159 }
 160 # else
 161 static struct kmem_cache *thread_info_cache;
 162
 163 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 164                                                   int node)
 165 {
 166         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
 167 }
 168
 169 static void free_thread_info(struct thread_info *ti)
 170 {
 171         kmem_cache_free(thread_info_cache, ti);
 172 }
 173
 174 void thread_info_cache_init(void)
 175 {
 176         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
 177                                               THREAD_SIZE, 0, NULL);
 178         BUG_ON(thread_info_cache == NULL);
 179 }
 180 # endif
 181 #endif
 182
 183 /* SLAB cache for signal_struct structures (tsk->signal) */
 184 static struct kmem_cache *signal_cachep;
 185
 186 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 187 struct kmem_cache *sighand_cachep;
 188
 189 /* SLAB cache for files_struct structures (tsk->files) */
 190 struct kmem_cache *files_cachep;
 191
 192 /* SLAB cache for fs_struct structures (tsk->fs) */
 193 struct kmem_cache *fs_cachep;
 194
 195 /* SLAB cache for vm_area_struct structures */
 196 struct kmem_cache *vm_area_cachep;
 197
 198 /* SLAB cache for mm_struct structures (tsk->mm) */
 199 static struct kmem_cache *mm_cachep;
 200
 201 /* Notifier list called when a task struct is freed */
 202 static ATOMIC_NOTIFIER_HEAD(task_free_notifier);
 203
 204 static void account_kernel_stack(struct thread_info *ti, int account)
 205 {
 206         struct zone *zone = page_zone(virt_to_page(ti));
 207
 208         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 209 }
 210
 211 void free_task(struct task_struct *tsk)
 212 {
 213         account_kernel_stack(tsk->stack, -1);
 214         arch_release_thread_info(tsk->stack);
 215         free_thread_info(tsk->stack);
 216         rt_mutex_debug_task_free(tsk);
 217         ftrace_graph_exit_task(tsk);
 218         put_seccomp_filter(tsk);
 219         arch_release_task_struct(tsk);
 220         free_task_struct(tsk);
 221 }
 222 EXPORT_SYMBOL(free_task);
 223
 224 static inline void free_signal_struct(struct signal_struct *sig)
 225 {
 226         taskstats_tgid_free(sig);
 227         sched_autogroup_exit(sig);
 228         kmem_cache_free(signal_cachep, sig);
 229 }
 230
 231 static inline void put_signal_struct(struct signal_struct *sig)
 232 {
 233         if (atomic_dec_and_test(&sig->sigcnt))
 234                 free_signal_struct(sig);
 235 }
 236
 237 int task_free_register(struct notifier_block *n)
 238 {
 239         return atomic_notifier_chain_register(&task_free_notifier, n);
 240 }
 241 EXPORT_SYMBOL(task_free_register);
 242
 243 int task_free_unregister(struct notifier_block *n)
 244 {
 245         return atomic_notifier_chain_unregister(&task_free_notifier, n);
 246 }
 247 EXPORT_SYMBOL(task_free_unregister);
 248
 249 void __put_task_struct(struct task_struct *tsk)
 250 {
 251         WARN_ON(!tsk->exit_state);
 252         WARN_ON(atomic_read(&tsk->usage));
 253         WARN_ON(tsk == current);
 254
 255         security_task_free(tsk);
 256         exit_creds(tsk);
 257         delayacct_tsk_free(tsk);
 258         put_signal_struct(tsk->signal);
 259
 260         atomic_notifier_call_chain(&task_free_notifier, 0, tsk);
 261         if (!profile_handoff_task(tsk))
 262                 free_task(tsk);
 263 }
 264 EXPORT_SYMBOL_GPL(__put_task_struct);
 265
 266 void __init __weak arch_task_cache_init(void) { }
 267
 268 void __init fork_init(unsigned long mempages)
 269 {
 270 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 271 #ifndef ARCH_MIN_TASKALIGN
 272 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 273 #endif
 274         /* create a slab on which task_structs can be allocated */
 275         task_struct_cachep =
 276                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 277                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 278 #endif
 279
 280         /* do the arch specific task caches init */
 281         arch_task_cache_init();
 282
 283         /*
 284          * The default maximum number of threads is set to a safe
 285          * value: the thread structures can take up at most half
 286          * of memory.
 287          */
 288         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 289
 290         /*
 291          * we need to allow at least 20 threads to boot a system
 292          */
 293         if (max_threads < 20)
 294                 max_threads = 20;
 295
 296         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 297         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 298         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 299                 init_task.signal->rlim[RLIMIT_NPROC];
 300 }
 301
 302 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 303                                                struct task_struct *src)
 304 {
 305         *dst = *src;
 306         return 0;
 307 }
 308
 309 static struct task_struct *dup_task_struct(struct task_struct *orig)
 310 {
 311         struct task_struct *tsk;
 312         struct thread_info *ti;
 313         unsigned long *stackend;
 314         int node = tsk_fork_get_node(orig);
 315         int err;
 316
 317         tsk = alloc_task_struct_node(node);
 318         if (!tsk)
 319                 return NULL;
 320
 321         ti = alloc_thread_info_node(tsk, node);
 322         if (!ti)
 323                 goto free_tsk;
 324
 325         err = arch_dup_task_struct(tsk, orig);
 326         if (err)
 327                 goto free_ti;
 328
 329         tsk->stack = ti;
 330 #ifdef CONFIG_SECCOMP
 331         /*
 332          * We must handle setting up seccomp filters once we're under
 333          * the sighand lock in case orig has changed between now and
 334          * then. Until then, filter must be NULL to avoid messing up
 335          * the usage counts on the error path calling free_task.
 336          */
 337         tsk->seccomp.filter = NULL;
 338 #endif
 339
 340         setup_thread_stack(tsk, orig);
 341         clear_user_return_notifier(tsk);
 342         clear_tsk_need_resched(tsk);
 343         stackend = end_of_stack(tsk);
 344         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 345
 346 #ifdef CONFIG_CC_STACKPROTECTOR
 347         tsk->stack_canary = get_random_int();
 348 #endif
 349
 350         /*
 351          * One for us, one for whoever does the "release_task()" (usually
 352          * parent)
 353          */
 354         atomic_set(&tsk->usage, 2);
 355 #ifdef CONFIG_BLK_DEV_IO_TRACE
 356         tsk->btrace_seq = 0;
 357 #endif
 358         tsk->splice_pipe = NULL;
 359         tsk->task_frag.page = NULL;
 360
 361         account_kernel_stack(ti, 1);
 362
 363         return tsk;
 364
 365 free_ti:
 366         free_thread_info(ti);
 367 free_tsk:
 368         free_task_struct(tsk);
 369         return NULL;
 370 }
 371
 372 #ifdef CONFIG_MMU
 373 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 374 {
 375         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 376         struct rb_node **rb_link, *rb_parent;
 377         int retval;
 378         unsigned long charge;
 379         struct mempolicy *pol;
 380
 381         uprobe_start_dup_mmap();
 382         down_write(&oldmm->mmap_sem);
 383         flush_cache_dup_mm(oldmm);
 384         uprobe_dup_mmap(oldmm, mm);
 385         /*
 386          * Not linked in yet - no deadlock potential:
 387          */
 388         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 389
 390         mm->locked_vm = 0;
 391         mm->mmap = NULL;
 392         mm->mmap_cache = NULL;
 393         mm->free_area_cache = oldmm->mmap_base;
 394         mm->cached_hole_size = ~0UL;
 395         mm->map_count = 0;
 396         cpumask_clear(mm_cpumask(mm));
 397         mm->mm_rb = RB_ROOT;
 398         rb_link = &mm->mm_rb.rb_node;
 399         rb_parent = NULL;
 400         pprev = &mm->mmap;
 401         retval = ksm_fork(mm, oldmm);
 402         if (retval)
 403                 goto out;
 404         retval = khugepaged_fork(mm, oldmm);
 405         if (retval)
 406                 goto out;
 407
 408         prev = NULL;
 409         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 410                 struct file *file;
 411
 412                 if (mpnt->vm_flags & VM_DONTCOPY) {
 413                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 414                                                         -vma_pages(mpnt));
 415                         continue;
 416                 }
 417                 charge = 0;
 418                 if (mpnt->vm_flags & VM_ACCOUNT) {
 419                         unsigned long len = vma_pages(mpnt);
 420
 421                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 422                                 goto fail_nomem;
 423                         charge = len;
 424                 }
 425                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 426                 if (!tmp)
 427                         goto fail_nomem;
 428                 *tmp = *mpnt;
 429                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 430                 pol = mpol_dup(vma_policy(mpnt));
 431                 retval = PTR_ERR(pol);
 432                 if (IS_ERR(pol))
 433                         goto fail_nomem_policy;
 434                 vma_set_policy(tmp, pol);
 435                 tmp->vm_mm = mm;
 436                 if (anon_vma_fork(tmp, mpnt))
 437                         goto fail_nomem_anon_vma_fork;
 438                 tmp->vm_flags &= ~VM_LOCKED;
 439                 tmp->vm_next = tmp->vm_prev = NULL;
 440                 file = tmp->vm_file;
 441                 if (file) {
 442                         struct inode *inode = file_inode(file);
 443                         struct address_space *mapping = file->f_mapping;
 444
 445                         get_file(file);
 446                         if (tmp->vm_flags & VM_DENYWRITE)
 447                                 atomic_dec(&inode->i_writecount);
 448                         mutex_lock(&mapping->i_mmap_mutex);
 449                         if (tmp->vm_flags & VM_SHARED)
 450                                 mapping->i_mmap_writable++;
 451                         flush_dcache_mmap_lock(mapping);
 452                         /* insert tmp into the share list, just after mpnt */
 453                         if (unlikely(tmp->vm_flags & VM_NONLINEAR))
 454                                 vma_nonlinear_insert(tmp,
 455                                                 &mapping->i_mmap_nonlinear);
 456                         else
 457                                 vma_interval_tree_insert_after(tmp, mpnt,
 458                                                         &mapping->i_mmap);
 459                         flush_dcache_mmap_unlock(mapping);
 460                         mutex_unlock(&mapping->i_mmap_mutex);
 461                 }
 462
 463                 /*
 464                  * Clear hugetlb-related page reserves for children. This only
 465                  * affects MAP_PRIVATE mappings. Faults generated by the child
 466                  * are not guaranteed to succeed, even if read-only
 467                  */
 468                 if (is_vm_hugetlb_page(tmp))
 469                         reset_vma_resv_huge_pages(tmp);
 470
 471                 /*
 472                  * Link in the new vma and copy the page table entries.
 473                  */
 474                 *pprev = tmp;
 475                 pprev = &tmp->vm_next;
 476                 tmp->vm_prev = prev;
 477                 prev = tmp;
 478
 479                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 480                 rb_link = &tmp->vm_rb.rb_right;
 481                 rb_parent = &tmp->vm_rb;
 482
 483                 mm->map_count++;
 484                 retval = copy_page_range(mm, oldmm, mpnt);
 485
 486                 if (tmp->vm_ops && tmp->vm_ops->open)
 487                         tmp->vm_ops->open(tmp);
 488
 489                 if (retval)
 490                         goto out;
 491         }
 492         /* a new mm has just been created */
 493         arch_dup_mmap(oldmm, mm);
 494         retval = 0;
 495 out:
 496         up_write(&mm->mmap_sem);
 497         flush_tlb_mm(oldmm);
 498         up_write(&oldmm->mmap_sem);
 499         uprobe_end_dup_mmap();
 500         return retval;
 501 fail_nomem_anon_vma_fork:
 502         mpol_put(pol);
 503 fail_nomem_policy:
 504         kmem_cache_free(vm_area_cachep, tmp);
 505 fail_nomem:
 506         retval = -ENOMEM;
 507         vm_unacct_memory(charge);
 508         goto out;
 509 }
 510
 511 static inline int mm_alloc_pgd(struct mm_struct *mm)
 512 {
 513         mm->pgd = pgd_alloc(mm);
 514         if (unlikely(!mm->pgd))
 515                 return -ENOMEM;
 516         return 0;
 517 }
 518
 519 static inline void mm_free_pgd(struct mm_struct *mm)
 520 {
 521         pgd_free(mm, mm->pgd);
 522 }
 523 #else
 524 #define dup_mmap(mm, oldmm)     (0)
 525 #define mm_alloc_pgd(mm)        (0)
 526 #define mm_free_pgd(mm)
 527 #endif /* CONFIG_MMU */
 528
 529 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 530
 531 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 532 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 533
 534 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 535
 536 static int __init coredump_filter_setup(char *s)
 537 {
 538         default_dump_filter =
 539                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 540                 MMF_DUMP_FILTER_MASK;
 541         return 1;
 542 }
 543
 544 __setup("coredump_filter=", coredump_filter_setup);
 545
 546 #include <linux/init_task.h>
 547
 548 static void mm_init_aio(struct mm_struct *mm)
 549 {
 550 #ifdef CONFIG_AIO
 551         spin_lock_init(&mm->ioctx_lock);
 552         INIT_HLIST_HEAD(&mm->ioctx_list);
 553 #endif
 554 }
 555
 556 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 557 {
 558         atomic_set(&mm->mm_users, 1);
 559         atomic_set(&mm->mm_count, 1);
 560         init_rwsem(&mm->mmap_sem);
 561         INIT_LIST_HEAD(&mm->mmlist);
 562         mm->flags = (current->mm) ?
 563                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 564         mm->core_state = NULL;
 565         mm->nr_ptes = 0;
 566         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 567         spin_lock_init(&mm->page_table_lock);
 568         mm->free_area_cache = TASK_UNMAPPED_BASE;
 569         mm->cached_hole_size = ~0UL;
 570         mm_init_aio(mm);
 571         mm_init_owner(mm, p);
 572
 573         if (likely(!mm_alloc_pgd(mm))) {
 574                 mm->def_flags = 0;
 575                 mmu_notifier_mm_init(mm);
 576                 return mm;
 577         }
 578
 579         free_mm(mm);
 580         return NULL;
 581 }
 582
 583 static void check_mm(struct mm_struct *mm)
 584 {
 585         int i;
 586
 587         for (i = 0; i < NR_MM_COUNTERS; i++) {
 588                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 589
 590                 if (unlikely(x))
 591                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 592                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 593         }
 594
 595 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 596         VM_BUG_ON(mm->pmd_huge_pte);
 597 #endif
 598 }
 599
 600 /*
 601  * Allocate and initialize an mm_struct.
 602  */
 603 struct mm_struct *mm_alloc(void)
 604 {
 605         struct mm_struct *mm;
 606
 607         mm = allocate_mm();
 608         if (!mm)
 609                 return NULL;
 610
 611         memset(mm, 0, sizeof(*mm));
 612         mm_init_cpumask(mm);
 613         return mm_init(mm, current);
 614 }
 615
 616 /*
 617  * Called when the last reference to the mm
 618  * is dropped: either by a lazy thread or by
 619  * mmput. Free the page directory and the mm.
 620  */
 621 void __mmdrop(struct mm_struct *mm)
 622 {
 623         BUG_ON(mm == &init_mm);
 624         mm_free_pgd(mm);
 625         destroy_context(mm);
 626         mmu_notifier_mm_destroy(mm);
 627         check_mm(mm);
 628         free_mm(mm);
 629 }
 630 EXPORT_SYMBOL_GPL(__mmdrop);
 631
 632 /*
 633  * Decrement the use count and release all resources for an mm.
 634  */
 635 void mmput(struct mm_struct *mm)
 636 {
 637         might_sleep();
 638
 639         if (atomic_dec_and_test(&mm->mm_users)) {
 640                 uprobe_clear_state(mm);
 641                 exit_aio(mm);
 642                 ksm_exit(mm);
 643                 khugepaged_exit(mm); /* must run before exit_mmap */
 644                 exit_mmap(mm);
 645                 set_mm_exe_file(mm, NULL);
 646                 if (!list_empty(&mm->mmlist)) {
 647                         spin_lock(&mmlist_lock);
 648                         list_del(&mm->mmlist);
 649                         spin_unlock(&mmlist_lock);
 650                 }
 651                 if (mm->binfmt)
 652                         module_put(mm->binfmt->module);
 653                 mmdrop(mm);
 654         }
 655 }
 656 EXPORT_SYMBOL_GPL(mmput);
 657
 658 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 659 {
 660         if (new_exe_file)
 661                 get_file(new_exe_file);
 662         if (mm->exe_file)
 663                 fput(mm->exe_file);
 664         mm->exe_file = new_exe_file;
 665 }
 666
 667 struct file *get_mm_exe_file(struct mm_struct *mm)
 668 {
 669         struct file *exe_file;
 670
 671         /* We need mmap_sem to protect against races with removal of exe_file */
 672         down_read(&mm->mmap_sem);
 673         exe_file = mm->exe_file;
 674         if (exe_file)
 675                 get_file(exe_file);
 676         up_read(&mm->mmap_sem);
 677         return exe_file;
 678 }
 679
 680 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 681 {
 682         /* It's safe to write the exe_file pointer without exe_file_lock because
 683          * this is called during fork when the task is not yet in /proc */
 684         newmm->exe_file = get_mm_exe_file(oldmm);
 685 }
 686
 687 /**
 688  * get_task_mm - acquire a reference to the task's mm
 689  *
 690  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 691  * this kernel workthread has transiently adopted a user mm with use_mm,
 692  * to do its AIO) is not set and if so returns a reference to it, after
 693  * bumping up the use count.  User must release the mm via mmput()
 694  * after use.  Typically used by /proc and ptrace.
 695  */
 696 struct mm_struct *get_task_mm(struct task_struct *task)
 697 {
 698         struct mm_struct *mm;
 699
 700         task_lock(task);
 701         mm = task->mm;
 702         if (mm) {
 703                 if (task->flags & PF_KTHREAD)
 704                         mm = NULL;
 705                 else
 706                         atomic_inc(&mm->mm_users);
 707         }
 708         task_unlock(task);
 709         return mm;
 710 }
 711 EXPORT_SYMBOL_GPL(get_task_mm);
 712
 713 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 714 {
 715         struct mm_struct *mm;
 716         int err;
 717
 718         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 719         if (err)
 720                 return ERR_PTR(err);
 721
 722         mm = get_task_mm(task);
 723         if (mm && mm != current->mm &&
 724                         !ptrace_may_access(task, mode) &&
 725                         !capable(CAP_SYS_RESOURCE)) {
 726                 mmput(mm);
 727                 mm = ERR_PTR(-EACCES);
 728         }
 729         mutex_unlock(&task->signal->cred_guard_mutex);
 730
 731         return mm;
 732 }
 733
 734 static void complete_vfork_done(struct task_struct *tsk)
 735 {
 736         struct completion *vfork;
 737
 738         task_lock(tsk);
 739         vfork = tsk->vfork_done;
 740         if (likely(vfork)) {
 741                 tsk->vfork_done = NULL;
 742                 complete(vfork);
 743         }
 744         task_unlock(tsk);
 745 }
 746
 747 static int wait_for_vfork_done(struct task_struct *child,
 748                                 struct completion *vfork)
 749 {
 750         int killed;
 751
 752         freezer_do_not_count();
 753         killed = wait_for_completion_killable(vfork);
 754         freezer_count();
 755
 756         if (killed) {
 757                 task_lock(child);
 758                 child->vfork_done = NULL;
 759                 task_unlock(child);
 760         }
 761
 762         put_task_struct(child);
 763         return killed;
 764 }
 765
 766 /* Please note the differences between mmput and mm_release.
 767  * mmput is called whenever we stop holding onto a mm_struct,
 768  * error success whatever.
 769  *
 770  * mm_release is called after a mm_struct has been removed
 771  * from the current process.
 772  *
 773  * This difference is important for error handling, when we
 774  * only half set up a mm_struct for a new process and need to restore
 775  * the old one.  Because we mmput the new mm_struct before
 776  * restoring the old one. . .
 777  * Eric Biederman 10 January 1998
 778  */
 779 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 780 {
 781         /* Get rid of any futexes when releasing the mm */
 782 #ifdef CONFIG_FUTEX
 783         if (unlikely(tsk->robust_list)) {
 784                 exit_robust_list(tsk);
 785                 tsk->robust_list = NULL;
 786         }
 787 #ifdef CONFIG_COMPAT
 788         if (unlikely(tsk->compat_robust_list)) {
 789                 compat_exit_robust_list(tsk);
 790                 tsk->compat_robust_list = NULL;
 791         }
 792 #endif
 793         if (unlikely(!list_empty(&tsk->pi_state_list)))
 794                 exit_pi_state_list(tsk);
 795 #endif
 796
 797         uprobe_free_utask(tsk);
 798
 799         /* Get rid of any cached register state */
 800         deactivate_mm(tsk, mm);
 801
 802         /*
 803          * If we're exiting normally, clear a user-space tid field if
 804          * requested.  We leave this alone when dying by signal, to leave
 805          * the value intact in a core dump, and to save the unnecessary
 806          * trouble, say, a killed vfork parent shouldn't touch this mm.
 807          * Userland only wants this done for a sys_exit.
 808          */
 809         if (tsk->clear_child_tid) {
 810                 if (!(tsk->flags & PF_SIGNALED) &&
 811                     atomic_read(&mm->mm_users) > 1) {
 812                         /*
 813                          * We don't check the error code - if userspace has
 814                          * not set up a proper pointer then tough luck.
 815                          */
 816                         put_user(0, tsk->clear_child_tid);
 817                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 818                                         1, NULL, NULL, 0);
 819                 }
 820                 tsk->clear_child_tid = NULL;
 821         }
 822
 823         /*
 824          * All done, finally we can wake up parent and return this mm to him.
 825          * Also kthread_stop() uses this completion for synchronization.
 826          */
 827         if (tsk->vfork_done)
 828                 complete_vfork_done(tsk);
 829 }
 830
 831 /*
 832  * Allocate a new mm structure and copy contents from the
 833  * mm structure of the passed in task structure.
 834  */
 835 struct mm_struct *dup_mm(struct task_struct *tsk)
 836 {
 837         struct mm_struct *mm, *oldmm = current->mm;
 838         int err;
 839
 840         if (!oldmm)
 841                 return NULL;
 842
 843         mm = allocate_mm();
 844         if (!mm)
 845                 goto fail_nomem;
 846
 847         memcpy(mm, oldmm, sizeof(*mm));
 848         mm_init_cpumask(mm);
 849
 850 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 851         mm->pmd_huge_pte = NULL;
 852 #endif
 853 #ifdef CONFIG_NUMA_BALANCING
 854         mm->first_nid = NUMA_PTE_SCAN_INIT;
 855 #endif
 856         if (!mm_init(mm, tsk))
 857                 goto fail_nomem;
 858
 859         if (init_new_context(tsk, mm))
 860                 goto fail_nocontext;
 861
 862         dup_mm_exe_file(oldmm, mm);
 863
 864         err = dup_mmap(mm, oldmm);
 865         if (err)
 866                 goto free_pt;
 867
 868         mm->hiwater_rss = get_mm_rss(mm);
 869         mm->hiwater_vm = mm->total_vm;
 870
 871         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 872                 goto free_pt;
 873
 874         return mm;
 875
 876 free_pt:
 877         /* don't put binfmt in mmput, we haven't got module yet */
 878         mm->binfmt = NULL;
 879         mmput(mm);
 880
 881 fail_nomem:
 882         return NULL;
 883
 884 fail_nocontext:
 885         /*
 886          * If init_new_context() failed, we cannot use mmput() to free the mm
 887          * because it calls destroy_context()
 888          */
 889         mm_free_pgd(mm);
 890         free_mm(mm);
 891         return NULL;
 892 }
 893
 894 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 895 {
 896         struct mm_struct *mm, *oldmm;
 897         int retval;
 898
 899         tsk->min_flt = tsk->maj_flt = 0;
 900         tsk->nvcsw = tsk->nivcsw = 0;
 901 #ifdef CONFIG_DETECT_HUNG_TASK
 902         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 903 #endif
 904
 905         tsk->mm = NULL;
 906         tsk->active_mm = NULL;
 907
 908         /*
 909          * Are we cloning a kernel thread?
 910          *
 911          * We need to steal a active VM for that..
 912          */
 913         oldmm = current->mm;
 914         if (!oldmm)
 915                 return 0;
 916
 917         if (clone_flags & CLONE_VM) {
 918                 atomic_inc(&oldmm->mm_users);
 919                 mm = oldmm;
 920                 goto good_mm;
 921         }
 922
 923         retval = -ENOMEM;
 924         mm = dup_mm(tsk);
 925         if (!mm)
 926                 goto fail_nomem;
 927
 928 good_mm:
 929         tsk->mm = mm;
 930         tsk->active_mm = mm;
 931         return 0;
 932
 933 fail_nomem:
 934         return retval;
 935 }
 936
 937 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 938 {
 939         struct fs_struct *fs = current->fs;
 940         if (clone_flags & CLONE_FS) {
 941                 /* tsk->fs is already what we want */
 942                 spin_lock(&fs->lock);
 943                 if (fs->in_exec) {
 944                         spin_unlock(&fs->lock);
 945                         return -EAGAIN;
 946                 }
 947                 fs->users++;
 948                 spin_unlock(&fs->lock);
 949                 return 0;
 950         }
 951         tsk->fs = copy_fs_struct(fs);
 952         if (!tsk->fs)
 953                 return -ENOMEM;
 954         return 0;
 955 }
 956
 957 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 958 {
 959         struct files_struct *oldf, *newf;
 960         int error = 0;
 961
 962         /*
 963          * A background process may not have any files ...
 964          */
 965         oldf = current->files;
 966         if (!oldf)
 967                 goto out;
 968
 969         if (clone_flags & CLONE_FILES) {
 970                 atomic_inc(&oldf->count);
 971                 goto out;
 972         }
 973
 974         newf = dup_fd(oldf, &error);
 975         if (!newf)
 976                 goto out;
 977
 978         tsk->files = newf;
 979         error = 0;
 980 out:
 981         return error;
 982 }
 983
 984 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 985 {
 986 #ifdef CONFIG_BLOCK
 987         struct io_context *ioc = current->io_context;
 988         struct io_context *new_ioc;
 989
 990         if (!ioc)
 991                 return 0;
 992         /*
 993          * Share io context with parent, if CLONE_IO is set
 994          */
 995         if (clone_flags & CLONE_IO) {
 996                 ioc_task_link(ioc);
 997                 tsk->io_context = ioc;
 998         } else if (ioprio_valid(ioc->ioprio)) {
 999                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1000                 if (unlikely(!new_ioc))
1001                         return -ENOMEM;
1002
1003                 new_ioc->ioprio = ioc->ioprio;
1004                 put_io_context(new_ioc);
1005         }
1006 #endif
1007         return 0;
1008 }
1009
1010 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1011 {
1012         struct sighand_struct *sig;
1013
1014         if (clone_flags & CLONE_SIGHAND) {
1015                 atomic_inc(&current->sighand->count);
1016                 return 0;
1017         }
1018         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1019         rcu_assign_pointer(tsk->sighand, sig);
1020         if (!sig)
1021                 return -ENOMEM;
1022         atomic_set(&sig->count, 1);
1023         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1024         return 0;
1025 }
1026
1027 void __cleanup_sighand(struct sighand_struct *sighand)
1028 {
1029         if (atomic_dec_and_test(&sighand->count)) {
1030                 signalfd_cleanup(sighand);
1031                 kmem_cache_free(sighand_cachep, sighand);
1032         }
1033 }
1034
1035
1036 /*
1037  * Initialize POSIX timer handling for a thread group.
1038  */
1039 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1040 {
1041         unsigned long cpu_limit;
1042
1043         /* Thread group counters. */
1044         thread_group_cputime_init(sig);
1045
1046         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1047         if (cpu_limit != RLIM_INFINITY) {
1048                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1049                 sig->cputimer.running = 1;
1050         }
1051
1052         /* The timer lists. */
1053         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1054         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1055         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1056 }
1057
1058 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1059 {
1060         struct signal_struct *sig;
1061
1062         if (clone_flags & CLONE_THREAD)
1063                 return 0;
1064
1065         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1066         tsk->signal = sig;
1067         if (!sig)
1068                 return -ENOMEM;
1069
1070         sig->nr_threads = 1;
1071         atomic_set(&sig->live, 1);
1072         atomic_set(&sig->sigcnt, 1);
1073
1074         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1075         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1076         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1077
1078         init_waitqueue_head(&sig->wait_chldexit);
1079         sig->curr_target = tsk;
1080         init_sigpending(&sig->shared_pending);
1081         INIT_LIST_HEAD(&sig->posix_timers);
1082
1083         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1084         sig->real_timer.function = it_real_fn;
1085
1086         task_lock(current->group_leader);
1087         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1088         task_unlock(current->group_leader);
1089
1090         posix_cpu_timers_init_group(sig);
1091
1092         tty_audit_fork(sig);
1093         sched_autogroup_fork(sig);
1094
1095 #ifdef CONFIG_CGROUPS
1096         init_rwsem(&sig->group_rwsem);
1097 #endif
1098
1099         sig->oom_score_adj = current->signal->oom_score_adj;
1100         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1101
1102         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1103                                    current->signal->is_child_subreaper;
1104
1105         mutex_init(&sig->cred_guard_mutex);
1106
1107         return 0;
1108 }
1109
1110 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1111 {
1112         unsigned long new_flags = p->flags;
1113
1114         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1115         new_flags |= PF_FORKNOEXEC;
1116         p->flags = new_flags;
1117 }
1118
1119 static void copy_seccomp(struct task_struct *p)
1120 {
1121 #ifdef CONFIG_SECCOMP
1122         /*
1123          * Must be called with sighand->lock held, which is common to
1124          * all threads in the group. Holding cred_guard_mutex is not
1125          * needed because this new task is not yet running and cannot
1126          * be racing exec.
1127          */
1128         assert_spin_locked(&current->sighand->siglock);
1129
1130         /* Ref-count the new filter user, and assign it. */
1131         get_seccomp_filter(current);
1132         p->seccomp = current->seccomp;
1133
1134         /*
1135          * Explicitly enable no_new_privs here in case it got set
1136          * between the task_struct being duplicated and holding the
1137          * sighand lock. The seccomp state and nnp must be in sync.
1138          */
1139         if (task_no_new_privs(current))
1140                 task_set_no_new_privs(p);
1141
1142         /*
1143          * If the parent gained a seccomp mode after copying thread
1144          * flags and between before we held the sighand lock, we have
1145          * to manually enable the seccomp thread flag here.
1146          */
1147         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1148                 set_tsk_thread_flag(p, TIF_SECCOMP);
1149 #endif
1150 }
1151
1152 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1153 {
1154         current->clear_child_tid = tidptr;
1155
1156         return task_pid_vnr(current);
1157 }
1158
1159 static void rt_mutex_init_task(struct task_struct *p)
1160 {
1161         raw_spin_lock_init(&p->pi_lock);
1162 #ifdef CONFIG_RT_MUTEXES
1163         plist_head_init(&p->pi_waiters);
1164         p->pi_blocked_on = NULL;
1165 #endif
1166 }
1167
1168 #ifdef CONFIG_MM_OWNER
1169 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1170 {
1171         mm->owner = p;
1172 }
1173 #endif /* CONFIG_MM_OWNER */
1174
1175 /*
1176  * Initialize POSIX timer handling for a single task.
1177  */
1178 static void posix_cpu_timers_init(struct task_struct *tsk)
1179 {
1180         tsk->cputime_expires.prof_exp = 0;
1181         tsk->cputime_expires.virt_exp = 0;
1182         tsk->cputime_expires.sched_exp = 0;
1183         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1184         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1185         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1186 }
1187
1188 /*
1189  * This creates a new process as a copy of the old one,
1190  * but does not actually start it yet.
1191  *
1192  * It copies the registers, and all the appropriate
1193  * parts of the process environment (as per the clone
1194  * flags). The actual kick-off is left to the caller.
1195  */
1196 static struct task_struct *copy_process(unsigned long clone_flags,
1197                                         unsigned long stack_start,
1198                                         unsigned long stack_size,
1199                                         int __user *child_tidptr,
1200                                         struct pid *pid,
1201                                         int trace)
1202 {
1203         int retval;
1204         struct task_struct *p;
1205
1206         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1207                 return ERR_PTR(-EINVAL);
1208
1209         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1210                 return ERR_PTR(-EINVAL);
1211
1212         /*
1213          * Thread groups must share signals as well, and detached threads
1214          * can only be started up within the thread group.
1215          */
1216         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1217                 return ERR_PTR(-EINVAL);
1218
1219         /*
1220          * Shared signal handlers imply shared VM. By way of the above,
1221          * thread groups also imply shared VM. Blocking this case allows
1222          * for various simplifications in other code.
1223          */
1224         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1225                 return ERR_PTR(-EINVAL);
1226
1227         /*
1228          * Siblings of global init remain as zombies on exit since they are
1229          * not reaped by their parent (swapper). To solve this and to avoid
1230          * multi-rooted process trees, prevent global and container-inits
1231          * from creating siblings.
1232          */
1233         if ((clone_flags & CLONE_PARENT) &&
1234                                 current->signal->flags & SIGNAL_UNKILLABLE)
1235                 return ERR_PTR(-EINVAL);
1236
1237         /*
1238          * If the new process will be in a different pid namespace
1239          * don't allow the creation of threads.
1240          */
1241         if ((clone_flags & (CLONE_VM|CLONE_NEWPID)) &&
1242             (task_active_pid_ns(current) != current->nsproxy->pid_ns))
1243                 return ERR_PTR(-EINVAL);
1244
1245         retval = security_task_create(clone_flags);
1246         if (retval)
1247                 goto fork_out;
1248
1249         retval = -ENOMEM;
1250         p = dup_task_struct(current);
1251         if (!p)
1252                 goto fork_out;
1253
1254         ftrace_graph_init_task(p);
1255
1256         rt_mutex_init_task(p);
1257
1258 #ifdef CONFIG_PROVE_LOCKING
1259         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1260         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1261 #endif
1262         retval = -EAGAIN;
1263         if (atomic_read(&p->real_cred->user->processes) >=
1264                         task_rlimit(p, RLIMIT_NPROC)) {
1265                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1266                     p->real_cred->user != INIT_USER)
1267                         goto bad_fork_free;
1268         }
1269         current->flags &= ~PF_NPROC_EXCEEDED;
1270
1271         retval = copy_creds(p, clone_flags);
1272         if (retval < 0)
1273                 goto bad_fork_free;
1274
1275         /*
1276          * If multiple threads are within copy_process(), then this check
1277          * triggers too late. This doesn't hurt, the check is only there
1278          * to stop root fork bombs.
1279          */
1280         retval = -EAGAIN;
1281         if (nr_threads >= max_threads)
1282                 goto bad_fork_cleanup_count;
1283
1284         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1285                 goto bad_fork_cleanup_count;
1286
1287         p->did_exec = 0;
1288         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1289         copy_flags(clone_flags, p);
1290         INIT_LIST_HEAD(&p->children);
1291         INIT_LIST_HEAD(&p->sibling);
1292         rcu_copy_process(p);
1293         p->vfork_done = NULL;
1294         spin_lock_init(&p->alloc_lock);
1295
1296         init_sigpending(&p->pending);
1297
1298         p->utime = p->stime = p->gtime = 0;
1299         p->utimescaled = p->stimescaled = 0;
1300 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1301         p->prev_cputime.utime = p->prev_cputime.stime = 0;
1302 #endif
1303 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1304         seqlock_init(&p->vtime_seqlock);
1305         p->vtime_snap = 0;
1306         p->vtime_snap_whence = VTIME_SLEEPING;
1307 #endif
1308
1309 #if defined(SPLIT_RSS_COUNTING)
1310         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1311 #endif
1312
1313         p->default_timer_slack_ns = current->timer_slack_ns;
1314
1315         task_io_accounting_init(&p->ioac);
1316         acct_clear_integrals(p);
1317
1318         posix_cpu_timers_init(p);
1319
1320         do_posix_clock_monotonic_gettime(&p->start_time);
1321         p->real_start_time = p->start_time;
1322         monotonic_to_bootbased(&p->real_start_time);
1323         p->io_context = NULL;
1324         p->audit_context = NULL;
1325         if (clone_flags & CLONE_THREAD)
1326                 threadgroup_change_begin(current);
1327         cgroup_fork(p);
1328 #ifdef CONFIG_NUMA
1329         p->mempolicy = mpol_dup(p->mempolicy);
1330         if (IS_ERR(p->mempolicy)) {
1331                 retval = PTR_ERR(p->mempolicy);
1332                 p->mempolicy = NULL;
1333                 goto bad_fork_cleanup_cgroup;
1334         }
1335         mpol_fix_fork_child_flag(p);
1336 #endif
1337 #ifdef CONFIG_CPUSETS
1338         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1339         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1340         seqcount_init(&p->mems_allowed_seq);
1341 #endif
1342 #ifdef CONFIG_TRACE_IRQFLAGS
1343         p->irq_events = 0;
1344         p->hardirqs_enabled = 0;
1345         p->hardirq_enable_ip = 0;
1346         p->hardirq_enable_event = 0;
1347         p->hardirq_disable_ip = _THIS_IP_;
1348         p->hardirq_disable_event = 0;
1349         p->softirqs_enabled = 1;
1350         p->softirq_enable_ip = _THIS_IP_;
1351         p->softirq_enable_event = 0;
1352         p->softirq_disable_ip = 0;
1353         p->softirq_disable_event = 0;
1354         p->hardirq_context = 0;
1355         p->softirq_context = 0;
1356 #endif
1357 #ifdef CONFIG_LOCKDEP
1358         p->lockdep_depth = 0; /* no locks held yet */
1359         p->curr_chain_key = 0;
1360         p->lockdep_recursion = 0;
1361 #endif
1362
1363 #ifdef CONFIG_DEBUG_MUTEXES
1364         p->blocked_on = NULL; /* not blocked yet */
1365 #endif
1366 #ifdef CONFIG_MEMCG
1367         p->memcg_batch.do_batch = 0;
1368         p->memcg_batch.memcg = NULL;
1369 #endif
1370 #ifdef CONFIG_BCACHE
1371         p->sequential_io        = 0;
1372         p->sequential_io_avg    = 0;
1373 #endif
1374
1375         /* Perform scheduler related setup. Assign this task to a CPU. */
1376         sched_fork(p);
1377
1378         retval = perf_event_init_task(p);
1379         if (retval)
1380                 goto bad_fork_cleanup_policy;
1381         retval = audit_alloc(p);
1382         if (retval)
1383                 goto bad_fork_cleanup_policy;
1384         /* copy all the process information */
1385         retval = copy_semundo(clone_flags, p);
1386         if (retval)
1387                 goto bad_fork_cleanup_audit;
1388         retval = copy_files(clone_flags, p);
1389         if (retval)
1390                 goto bad_fork_cleanup_semundo;
1391         retval = copy_fs(clone_flags, p);
1392         if (retval)
1393                 goto bad_fork_cleanup_files;
1394         retval = copy_sighand(clone_flags, p);
1395         if (retval)
1396                 goto bad_fork_cleanup_fs;
1397         retval = copy_signal(clone_flags, p);
1398         if (retval)
1399                 goto bad_fork_cleanup_sighand;
1400         retval = copy_mm(clone_flags, p);
1401         if (retval)
1402                 goto bad_fork_cleanup_signal;
1403         retval = copy_namespaces(clone_flags, p);
1404         if (retval)
1405                 goto bad_fork_cleanup_mm;
1406         retval = copy_io(clone_flags, p);
1407         if (retval)
1408                 goto bad_fork_cleanup_namespaces;
1409         retval = copy_thread(clone_flags, stack_start, stack_size, p);
1410         if (retval)
1411                 goto bad_fork_cleanup_io;
1412
1413         if (pid != &init_struct_pid) {
1414                 retval = -ENOMEM;
1415                 pid = alloc_pid(p->nsproxy->pid_ns);
1416                 if (!pid)
1417                         goto bad_fork_cleanup_io;
1418         }
1419
1420         p->pid = pid_nr(pid);
1421         p->tgid = p->pid;
1422         if (clone_flags & CLONE_THREAD)
1423                 p->tgid = current->tgid;
1424
1425         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1426         /*
1427          * Clear TID on mm_release()?
1428          */
1429         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1430 #ifdef CONFIG_BLOCK
1431         p->plug = NULL;
1432 #endif
1433 #ifdef CONFIG_FUTEX
1434         p->robust_list = NULL;
1435 #ifdef CONFIG_COMPAT
1436         p->compat_robust_list = NULL;
1437 #endif
1438         INIT_LIST_HEAD(&p->pi_state_list);
1439         p->pi_state_cache = NULL;
1440 #endif
1441         uprobe_copy_process(p);
1442         /*
1443          * sigaltstack should be cleared when sharing the same VM
1444          */
1445         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1446                 p->sas_ss_sp = p->sas_ss_size = 0;
1447
1448         /*
1449          * Syscall tracing and stepping should be turned off in the
1450          * child regardless of CLONE_PTRACE.
1451          */
1452         user_disable_single_step(p);
1453         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1454 #ifdef TIF_SYSCALL_EMU
1455         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1456 #endif
1457         clear_all_latency_tracing(p);
1458
1459         /* ok, now we should be set up.. */
1460         if (clone_flags & CLONE_THREAD)
1461                 p->exit_signal = -1;
1462         else if (clone_flags & CLONE_PARENT)
1463                 p->exit_signal = current->group_leader->exit_signal;
1464         else
1465                 p->exit_signal = (clone_flags & CSIGNAL);
1466
1467         p->pdeath_signal = 0;
1468         p->exit_state = 0;
1469
1470         p->nr_dirtied = 0;
1471         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1472         p->dirty_paused_when = 0;
1473
1474         /*
1475          * Ok, make it visible to the rest of the system.
1476          * We dont wake it up yet.
1477          */
1478         p->group_leader = p;
1479         INIT_LIST_HEAD(&p->thread_group);
1480         p->task_works = NULL;
1481
1482         /* Need tasklist lock for parent etc handling! */
1483         write_lock_irq(&tasklist_lock);
1484
1485         /* CLONE_PARENT re-uses the old parent */
1486         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1487                 p->real_parent = current->real_parent;
1488                 p->parent_exec_id = current->parent_exec_id;
1489         } else {
1490                 p->real_parent = current;
1491                 p->parent_exec_id = current->self_exec_id;
1492         }
1493
1494         spin_lock(&current->sighand->siglock);
1495
1496         /*
1497          * Copy seccomp details explicitly here, in case they were changed
1498          * before holding sighand lock.
1499          */
1500         copy_seccomp(p);
1501
1502         /*
1503          * Process group and session signals need to be delivered to just the
1504          * parent before the fork or both the parent and the child after the
1505          * fork. Restart if a signal comes in before we add the new process to
1506          * it's process group.
1507          * A fatal signal pending means that current will exit, so the new
1508          * thread can't slip out of an OOM kill (or normal SIGKILL).
1509         */
1510         recalc_sigpending();
1511         if (signal_pending(current)) {
1512                 spin_unlock(&current->sighand->siglock);
1513                 write_unlock_irq(&tasklist_lock);
1514                 retval = -ERESTARTNOINTR;
1515                 goto bad_fork_free_pid;
1516         }
1517
1518         if (clone_flags & CLONE_THREAD) {
1519                 current->signal->nr_threads++;
1520                 atomic_inc(&current->signal->live);
1521                 atomic_inc(&current->signal->sigcnt);
1522                 p->group_leader = current->group_leader;
1523                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1524         }
1525
1526         if (likely(p->pid)) {
1527                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1528
1529                 if (thread_group_leader(p)) {
1530                         if (is_child_reaper(pid)) {
1531                                 ns_of_pid(pid)->child_reaper = p;
1532                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1533                         }
1534
1535                         p->signal->leader_pid = pid;
1536                         p->signal->tty = tty_kref_get(current->signal->tty);
1537                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1538                         attach_pid(p, PIDTYPE_SID, task_session(current));
1539                         list_add_tail(&p->sibling, &p->real_parent->children);
1540                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1541                         __this_cpu_inc(process_counts);
1542                 } else {
1543                         list_add_tail_rcu(&p->thread_node,
1544                                           &p->signal->thread_head);
1545                 }
1546                 attach_pid(p, PIDTYPE_PID, pid);
1547                 nr_threads++;
1548         }
1549
1550         total_forks++;
1551         spin_unlock(&current->sighand->siglock);
1552         write_unlock_irq(&tasklist_lock);
1553         proc_fork_connector(p);
1554         cgroup_post_fork(p);
1555         if (clone_flags & CLONE_THREAD)
1556                 threadgroup_change_end(current);
1557         perf_event_fork(p);
1558
1559         trace_task_newtask(p, clone_flags);
1560
1561         return p;
1562
1563 bad_fork_free_pid:
1564         if (pid != &init_struct_pid)
1565                 free_pid(pid);
1566 bad_fork_cleanup_io:
1567         if (p->io_context)
1568                 exit_io_context(p);
1569 bad_fork_cleanup_namespaces:
1570         exit_task_namespaces(p);
1571 bad_fork_cleanup_mm:
1572         if (p->mm)
1573                 mmput(p->mm);
1574 bad_fork_cleanup_signal:
1575         if (!(clone_flags & CLONE_THREAD))
1576                 free_signal_struct(p->signal);
1577 bad_fork_cleanup_sighand:
1578         __cleanup_sighand(p->sighand);
1579 bad_fork_cleanup_fs:
1580         exit_fs(p); /* blocking */
1581 bad_fork_cleanup_files:
1582         exit_files(p); /* blocking */
1583 bad_fork_cleanup_semundo:
1584         exit_sem(p);
1585 bad_fork_cleanup_audit:
1586         audit_free(p);
1587 bad_fork_cleanup_policy:
1588         perf_event_free_task(p);
1589 #ifdef CONFIG_NUMA
1590         mpol_put(p->mempolicy);
1591 bad_fork_cleanup_cgroup:
1592 #endif
1593         if (clone_flags & CLONE_THREAD)
1594                 threadgroup_change_end(current);
1595         cgroup_exit(p, 0);
1596         delayacct_tsk_free(p);
1597         module_put(task_thread_info(p)->exec_domain->module);
1598 bad_fork_cleanup_count:
1599         atomic_dec(&p->cred->user->processes);
1600         exit_creds(p);
1601 bad_fork_free:
1602         free_task(p);
1603 fork_out:
1604         return ERR_PTR(retval);
1605 }
1606
1607 static inline void init_idle_pids(struct pid_link *links)
1608 {
1609         enum pid_type type;
1610
1611         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1612                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1613                 links[type].pid = &init_struct_pid;
1614         }
1615 }
1616
1617 struct task_struct * __cpuinit fork_idle(int cpu)
1618 {
1619         struct task_struct *task;
1620         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1621         if (!IS_ERR(task)) {
1622                 init_idle_pids(task->pids);
1623                 init_idle(task, cpu);
1624         }
1625
1626         return task;
1627 }
1628
1629 /*
1630  *  Ok, this is the main fork-routine.
1631  *
1632  * It copies the process, and if successful kick-starts
1633  * it and waits for it to finish using the VM if required.
1634  */
1635 long do_fork(unsigned long clone_flags,
1636               unsigned long stack_start,
1637               unsigned long stack_size,
1638               int __user *parent_tidptr,
1639               int __user *child_tidptr)
1640 {
1641         struct task_struct *p;
1642         int trace = 0;
1643         long nr;
1644
1645         /*
1646          * Do some preliminary argument and permissions checking before we
1647          * actually start allocating stuff
1648          */
1649         if (clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) {
1650                 if (clone_flags & (CLONE_THREAD|CLONE_PARENT))
1651                         return -EINVAL;
1652         }
1653
1654         /*
1655          * Determine whether and which event to report to ptracer.  When
1656          * called from kernel_thread or CLONE_UNTRACED is explicitly
1657          * requested, no event is reported; otherwise, report if the event
1658          * for the type of forking is enabled.
1659          */
1660         if (!(clone_flags & CLONE_UNTRACED)) {
1661                 if (clone_flags & CLONE_VFORK)
1662                         trace = PTRACE_EVENT_VFORK;
1663                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1664                         trace = PTRACE_EVENT_CLONE;
1665                 else
1666                         trace = PTRACE_EVENT_FORK;
1667
1668                 if (likely(!ptrace_event_enabled(current, trace)))
1669                         trace = 0;
1670         }
1671
1672         p = copy_process(clone_flags, stack_start, stack_size,
1673                          child_tidptr, NULL, trace);
1674         /*
1675          * Do this prior waking up the new thread - the thread pointer
1676          * might get invalid after that point, if the thread exits quickly.
1677          */
1678         if (!IS_ERR(p)) {
1679                 struct completion vfork;
1680
1681                 trace_sched_process_fork(current, p);
1682
1683                 nr = task_pid_vnr(p);
1684
1685                 if (clone_flags & CLONE_PARENT_SETTID)
1686                         put_user(nr, parent_tidptr);
1687
1688                 if (clone_flags & CLONE_VFORK) {
1689                         p->vfork_done = &vfork;
1690                         init_completion(&vfork);
1691                         get_task_struct(p);
1692                 }
1693
1694                 wake_up_new_task(p);
1695
1696                 /* forking complete and child started to run, tell ptracer */
1697                 if (unlikely(trace))
1698                         ptrace_event(trace, nr);
1699
1700                 if (clone_flags & CLONE_VFORK) {
1701                         if (!wait_for_vfork_done(p, &vfork))
1702                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1703                 }
1704         } else {
1705                 nr = PTR_ERR(p);
1706         }
1707         return nr;
1708 }
1709
1710 /*
1711  * Create a kernel thread.
1712  */
1713 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1714 {
1715         return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1716                 (unsigned long)arg, NULL, NULL);
1717 }
1718
1719 #ifdef __ARCH_WANT_SYS_FORK
1720 SYSCALL_DEFINE0(fork)
1721 {
1722 #ifdef CONFIG_MMU
1723         return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1724 #else
1725         /* can not support in nommu mode */
1726         return(-EINVAL);
1727 #endif
1728 }
1729 #endif
1730
1731 #ifdef __ARCH_WANT_SYS_VFORK
1732 SYSCALL_DEFINE0(vfork)
1733 {
1734         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1735                         0, NULL, NULL);
1736 }
1737 #endif
1738
1739 #ifdef __ARCH_WANT_SYS_CLONE
1740 #ifdef CONFIG_CLONE_BACKWARDS
1741 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1742                  int __user *, parent_tidptr,
1743                  int, tls_val,
1744                  int __user *, child_tidptr)
1745 #elif defined(CONFIG_CLONE_BACKWARDS2)
1746 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1747                  int __user *, parent_tidptr,
1748                  int __user *, child_tidptr,
1749                  int, tls_val)
1750 #else
1751 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1752                  int __user *, parent_tidptr,
1753                  int __user *, child_tidptr,
1754                  int, tls_val)
1755 #endif
1756 {
1757         return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1758 }
1759 #endif
1760
1761 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1762 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1763 #endif
1764
1765 static void sighand_ctor(void *data)
1766 {
1767         struct sighand_struct *sighand = data;
1768
1769         spin_lock_init(&sighand->siglock);
1770         init_waitqueue_head(&sighand->signalfd_wqh);
1771 }
1772
1773 void __init proc_caches_init(void)
1774 {
1775         sighand_cachep = kmem_cache_create("sighand_cache",
1776                         sizeof(struct sighand_struct), 0,
1777                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1778                         SLAB_NOTRACK, sighand_ctor);
1779         signal_cachep = kmem_cache_create("signal_cache",
1780                         sizeof(struct signal_struct), 0,
1781                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1782         files_cachep = kmem_cache_create("files_cache",
1783                         sizeof(struct files_struct), 0,
1784                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1785         fs_cachep = kmem_cache_create("fs_cache",
1786                         sizeof(struct fs_struct), 0,
1787                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1788         /*
1789          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1790          * whole struct cpumask for the OFFSTACK case. We could change
1791          * this to *only* allocate as much of it as required by the
1792          * maximum number of CPU's we can ever have.  The cpumask_allocation
1793          * is at the end of the structure, exactly for that reason.
1794          */
1795         mm_cachep = kmem_cache_create("mm_struct",
1796                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1797                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1798         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1799         mmap_init();
1800         nsproxy_cache_init();
1801 }
1802
1803 /*
1804  * Check constraints on flags passed to the unshare system call.
1805  */
1806 static int check_unshare_flags(unsigned long unshare_flags)
1807 {
1808         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1809                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1810                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1811                                 CLONE_NEWUSER|CLONE_NEWPID))
1812                 return -EINVAL;
1813         /*
1814          * Not implemented, but pretend it works if there is nothing to
1815          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1816          * needs to unshare vm.
1817          */
1818         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1819                 /* FIXME: get_task_mm() increments ->mm_users */
1820                 if (atomic_read(&current->mm->mm_users) > 1)
1821                         return -EINVAL;
1822         }
1823
1824         return 0;
1825 }
1826
1827 /*
1828  * Unshare the filesystem structure if it is being shared
1829  */
1830 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1831 {
1832         struct fs_struct *fs = current->fs;
1833
1834         if (!(unshare_flags & CLONE_FS) || !fs)
1835                 return 0;
1836
1837         /* don't need lock here; in the worst case we'll do useless copy */
1838         if (fs->users == 1)
1839                 return 0;
1840
1841         *new_fsp = copy_fs_struct(fs);
1842         if (!*new_fsp)
1843                 return -ENOMEM;
1844
1845         return 0;
1846 }
1847
1848 /*
1849  * Unshare file descriptor table if it is being shared
1850  */
1851 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1852 {
1853         struct files_struct *fd = current->files;
1854         int error = 0;
1855
1856         if ((unshare_flags & CLONE_FILES) &&
1857             (fd && atomic_read(&fd->count) > 1)) {
1858                 *new_fdp = dup_fd(fd, &error);
1859                 if (!*new_fdp)
1860                         return error;
1861         }
1862
1863         return 0;
1864 }
1865
1866 /*
1867  * unshare allows a process to 'unshare' part of the process
1868  * context which was originally shared using clone.  copy_*
1869  * functions used by do_fork() cannot be used here directly
1870  * because they modify an inactive task_struct that is being
1871  * constructed. Here we are modifying the current, active,
1872  * task_struct.
1873  */
1874 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1875 {
1876         struct fs_struct *fs, *new_fs = NULL;
1877         struct files_struct *fd, *new_fd = NULL;
1878         struct cred *new_cred = NULL;
1879         struct nsproxy *new_nsproxy = NULL;
1880         int do_sysvsem = 0;
1881         int err;
1882
1883         /*
1884          * If unsharing a user namespace must also unshare the thread.
1885          */
1886         if (unshare_flags & CLONE_NEWUSER)
1887                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1888         /*
1889          * If unsharing a pid namespace must also unshare the thread.
1890          */
1891         if (unshare_flags & CLONE_NEWPID)
1892                 unshare_flags |= CLONE_THREAD;
1893         /*
1894          * If unsharing a thread from a thread group, must also unshare vm.
1895          */
1896         if (unshare_flags & CLONE_THREAD)
1897                 unshare_flags |= CLONE_VM;
1898         /*
1899          * If unsharing vm, must also unshare signal handlers.
1900          */
1901         if (unshare_flags & CLONE_VM)
1902                 unshare_flags |= CLONE_SIGHAND;
1903         /*
1904          * If unsharing namespace, must also unshare filesystem information.
1905          */
1906         if (unshare_flags & CLONE_NEWNS)
1907                 unshare_flags |= CLONE_FS;
1908
1909         err = check_unshare_flags(unshare_flags);
1910         if (err)
1911                 goto bad_unshare_out;
1912         /*
1913          * CLONE_NEWIPC must also detach from the undolist: after switching
1914          * to a new ipc namespace, the semaphore arrays from the old
1915          * namespace are unreachable.
1916          */
1917         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1918                 do_sysvsem = 1;
1919         err = unshare_fs(unshare_flags, &new_fs);
1920         if (err)
1921                 goto bad_unshare_out;
1922         err = unshare_fd(unshare_flags, &new_fd);
1923         if (err)
1924                 goto bad_unshare_cleanup_fs;
1925         err = unshare_userns(unshare_flags, &new_cred);
1926         if (err)
1927                 goto bad_unshare_cleanup_fd;
1928         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1929                                          new_cred, new_fs);
1930         if (err)
1931                 goto bad_unshare_cleanup_cred;
1932
1933         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1934                 if (do_sysvsem) {
1935                         /*
1936                          * CLONE_SYSVSEM is equivalent to sys_exit().
1937                          */
1938                         exit_sem(current);
1939                 }
1940
1941                 if (new_nsproxy)
1942                         switch_task_namespaces(current, new_nsproxy);
1943
1944                 task_lock(current);
1945
1946                 if (new_fs) {
1947                         fs = current->fs;
1948                         spin_lock(&fs->lock);
1949                         current->fs = new_fs;
1950                         if (--fs->users)
1951                                 new_fs = NULL;
1952                         else
1953                                 new_fs = fs;
1954                         spin_unlock(&fs->lock);
1955                 }
1956
1957                 if (new_fd) {
1958                         fd = current->files;
1959                         current->files = new_fd;
1960                         new_fd = fd;
1961                 }
1962
1963                 task_unlock(current);
1964
1965                 if (new_cred) {
1966                         /* Install the new user namespace */
1967                         commit_creds(new_cred);
1968                         new_cred = NULL;
1969                 }
1970         }
1971
1972 bad_unshare_cleanup_cred:
1973         if (new_cred)
1974                 put_cred(new_cred);
1975 bad_unshare_cleanup_fd:
1976         if (new_fd)
1977                 put_files_struct(new_fd);
1978
1979 bad_unshare_cleanup_fs:
1980         if (new_fs)
1981                 free_fs_struct(new_fs);
1982
1983 bad_unshare_out:
1984         return err;
1985 }
1986
1987 /*
1988  *      Helper to unshare the files of the current task.
1989  *      We don't want to expose copy_files internals to
1990  *      the exec layer of the kernel.
1991  */
1992
1993 int unshare_files(struct files_struct **displaced)
1994 {
1995         struct task_struct *task = current;
1996         struct files_struct *copy = NULL;
1997         int error;
1998
1999         error = unshare_fd(CLONE_FILES, &copy);
2000         if (error || !copy) {
2001                 *displaced = NULL;
2002                 return error;
2003         }
2004         *displaced = task->files;
2005         task_lock(task);
2006         task->files = copy;
2007         task_unlock(task);
2008         return 0;
2009 }