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/mm.h>
  32 #include <linux/vmacache.h>
  33 #include <linux/nsproxy.h>
  34 #include <linux/capability.h>
  35 #include <linux/cpu.h>
  36 #include <linux/cgroup.h>
  37 #include <linux/security.h>
  38 #include <linux/hugetlb.h>
  39 #include <linux/seccomp.h>
  40 #include <linux/swap.h>
  41 #include <linux/syscalls.h>
  42 #include <linux/jiffies.h>
  43 #include <linux/futex.h>
  44 #include <linux/compat.h>
  45 #include <linux/kthread.h>
  46 #include <linux/task_io_accounting_ops.h>
  47 #include <linux/rcupdate.h>
  48 #include <linux/ptrace.h>
  49 #include <linux/mount.h>
  50 #include <linux/audit.h>
  51 #include <linux/memcontrol.h>
  52 #include <linux/ftrace.h>
  53 #include <linux/proc_fs.h>
  54 #include <linux/profile.h>
  55 #include <linux/rmap.h>
  56 #include <linux/ksm.h>
  57 #include <linux/acct.h>
  58 #include <linux/tsacct_kern.h>
  59 #include <linux/cn_proc.h>
  60 #include <linux/freezer.h>
  61 #include <linux/delayacct.h>
  62 #include <linux/taskstats_kern.h>
  63 #include <linux/random.h>
  64 #include <linux/tty.h>
  65 #include <linux/blkdev.h>
  66 #include <linux/fs_struct.h>
  67 #include <linux/magic.h>
  68 #include <linux/perf_event.h>
  69 #include <linux/posix-timers.h>
  70 #include <linux/user-return-notifier.h>
  71 #include <linux/oom.h>
  72 #include <linux/khugepaged.h>
  73 #include <linux/signalfd.h>
  74 #include <linux/uprobes.h>
  75 #include <linux/aio.h>
  76 #include <linux/compiler.h>
  77
  78 #include <asm/pgtable.h>
  79 #include <asm/pgalloc.h>
  80 #include <asm/uaccess.h>
  81 #include <asm/mmu_context.h>
  82 #include <asm/cacheflush.h>
  83 #include <asm/tlbflush.h>
  84
  85 #include <trace/events/sched.h>
  86
  87 #define CREATE_TRACE_POINTS
  88 #include <trace/events/task.h>
  89
  90 /*
  91  * Minimum number of threads to boot the kernel
  92  */
  93 #define MIN_THREADS 20
  94
  95 /*
  96  * Maximum number of threads
  97  */
  98 #define MAX_THREADS FUTEX_TID_MASK
  99
 100 /*
 101  * Protected counters by write_lock_irq(&tasklist_lock)
 102  */
 103 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 104 int nr_threads;                 /* The idle threads do not count.. */
 105
 106 int max_threads;                /* tunable limit on nr_threads */
 107
 108 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 109
 110 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 111
 112 #ifdef CONFIG_PROVE_RCU
 113 int lockdep_tasklist_lock_is_held(void)
 114 {
 115         return lockdep_is_held(&tasklist_lock);
 116 }
 117 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 118 #endif /* #ifdef CONFIG_PROVE_RCU */
 119
 120 int nr_processes(void)
 121 {
 122         int cpu;
 123         int total = 0;
 124
 125         for_each_possible_cpu(cpu)
 126                 total += per_cpu(process_counts, cpu);
 127
 128         return total;
 129 }
 130
 131 void __weak arch_release_task_struct(struct task_struct *tsk)
 132 {
 133 }
 134
 135 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 136 static struct kmem_cache *task_struct_cachep;
 137
 138 static inline struct task_struct *alloc_task_struct_node(int node)
 139 {
 140         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 141 }
 142
 143 static inline void free_task_struct(struct task_struct *tsk)
 144 {
 145         kmem_cache_free(task_struct_cachep, tsk);
 146 }
 147 #endif
 148
 149 void __weak arch_release_thread_info(struct thread_info *ti)
 150 {
 151 }
 152
 153 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
 154
 155 /*
 156  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 157  * kmemcache based allocator.
 158  */
 159 # if THREAD_SIZE >= PAGE_SIZE
 160 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 161                                                   int node)
 162 {
 163         struct page *page = alloc_kmem_pages_node(node, THREADINFO_GFP,
 164                                                   THREAD_SIZE_ORDER);
 165
 166         return page ? page_address(page) : NULL;
 167 }
 168
 169 static inline void free_thread_info(struct thread_info *ti)
 170 {
 171         free_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 172 }
 173 # else
 174 static struct kmem_cache *thread_info_cache;
 175
 176 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 177                                                   int node)
 178 {
 179         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
 180 }
 181
 182 static void free_thread_info(struct thread_info *ti)
 183 {
 184         kmem_cache_free(thread_info_cache, ti);
 185 }
 186
 187 void thread_info_cache_init(void)
 188 {
 189         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
 190                                               THREAD_SIZE, 0, NULL);
 191         BUG_ON(thread_info_cache == NULL);
 192 }
 193 # endif
 194 #endif
 195
 196 /* SLAB cache for signal_struct structures (tsk->signal) */
 197 static struct kmem_cache *signal_cachep;
 198
 199 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 200 struct kmem_cache *sighand_cachep;
 201
 202 /* SLAB cache for files_struct structures (tsk->files) */
 203 struct kmem_cache *files_cachep;
 204
 205 /* SLAB cache for fs_struct structures (tsk->fs) */
 206 struct kmem_cache *fs_cachep;
 207
 208 /* SLAB cache for vm_area_struct structures */
 209 struct kmem_cache *vm_area_cachep;
 210
 211 /* SLAB cache for mm_struct structures (tsk->mm) */
 212 static struct kmem_cache *mm_cachep;
 213
 214 static void account_kernel_stack(struct thread_info *ti, int account)
 215 {
 216         struct zone *zone = page_zone(virt_to_page(ti));
 217
 218         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 219 }
 220
 221 void free_task(struct task_struct *tsk)
 222 {
 223         account_kernel_stack(tsk->stack, -1);
 224         arch_release_thread_info(tsk->stack);
 225         free_thread_info(tsk->stack);
 226         rt_mutex_debug_task_free(tsk);
 227         ftrace_graph_exit_task(tsk);
 228         put_seccomp_filter(tsk);
 229         arch_release_task_struct(tsk);
 230         free_task_struct(tsk);
 231 }
 232 EXPORT_SYMBOL(free_task);
 233
 234 static inline void free_signal_struct(struct signal_struct *sig)
 235 {
 236         taskstats_tgid_free(sig);
 237         sched_autogroup_exit(sig);
 238         kmem_cache_free(signal_cachep, sig);
 239 }
 240
 241 static inline void put_signal_struct(struct signal_struct *sig)
 242 {
 243         if (atomic_dec_and_test(&sig->sigcnt))
 244                 free_signal_struct(sig);
 245 }
 246
 247 void __put_task_struct(struct task_struct *tsk)
 248 {
 249         WARN_ON(!tsk->exit_state);
 250         WARN_ON(atomic_read(&tsk->usage));
 251         WARN_ON(tsk == current);
 252
 253         task_numa_free(tsk);
 254         security_task_free(tsk);
 255         exit_creds(tsk);
 256         delayacct_tsk_free(tsk);
 257         put_signal_struct(tsk->signal);
 258
 259         if (!profile_handoff_task(tsk))
 260                 free_task(tsk);
 261 }
 262 EXPORT_SYMBOL_GPL(__put_task_struct);
 263
 264 void __init __weak arch_task_cache_init(void) { }
 265
 266 /*
 267  * set_max_threads
 268  */
 269 static void set_max_threads(void)
 270 {
 271         u64 threads;
 272
 273         /*
 274          * The number of threads shall be limited such that the thread
 275          * structures may only consume a small part of the available memory.
 276          */
 277         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 278                 threads = MAX_THREADS;
 279         else
 280                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 281                                     (u64) THREAD_SIZE * 8UL);
 282
 283         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 284 }
 285
 286 void __init fork_init(void)
 287 {
 288 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 289 #ifndef ARCH_MIN_TASKALIGN
 290 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 291 #endif
 292         /* create a slab on which task_structs can be allocated */
 293         task_struct_cachep =
 294                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 295                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 296 #endif
 297
 298         /* do the arch specific task caches init */
 299         arch_task_cache_init();
 300
 301         set_max_threads();
 302
 303         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 304         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 305         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 306                 init_task.signal->rlim[RLIMIT_NPROC];
 307 }
 308
 309 int __weak arch_dup_task_struct(struct task_struct *dst,
 310                                                struct task_struct *src)
 311 {
 312         *dst = *src;
 313         return 0;
 314 }
 315
 316 void set_task_stack_end_magic(struct task_struct *tsk)
 317 {
 318         unsigned long *stackend;
 319
 320         stackend = end_of_stack(tsk);
 321         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 322 }
 323
 324 static struct task_struct *dup_task_struct(struct task_struct *orig)
 325 {
 326         struct task_struct *tsk;
 327         struct thread_info *ti;
 328         int node = tsk_fork_get_node(orig);
 329         int err;
 330
 331         tsk = alloc_task_struct_node(node);
 332         if (!tsk)
 333                 return NULL;
 334
 335         ti = alloc_thread_info_node(tsk, node);
 336         if (!ti)
 337                 goto free_tsk;
 338
 339         err = arch_dup_task_struct(tsk, orig);
 340         if (err)
 341                 goto free_ti;
 342
 343         tsk->stack = ti;
 344 #ifdef CONFIG_SECCOMP
 345         /*
 346          * We must handle setting up seccomp filters once we're under
 347          * the sighand lock in case orig has changed between now and
 348          * then. Until then, filter must be NULL to avoid messing up
 349          * the usage counts on the error path calling free_task.
 350          */
 351         tsk->seccomp.filter = NULL;
 352 #endif
 353
 354         setup_thread_stack(tsk, orig);
 355         clear_user_return_notifier(tsk);
 356         clear_tsk_need_resched(tsk);
 357         set_task_stack_end_magic(tsk);
 358
 359 #ifdef CONFIG_CC_STACKPROTECTOR
 360         tsk->stack_canary = get_random_int();
 361 #endif
 362
 363         /*
 364          * One for us, one for whoever does the "release_task()" (usually
 365          * parent)
 366          */
 367         atomic_set(&tsk->usage, 2);
 368 #ifdef CONFIG_BLK_DEV_IO_TRACE
 369         tsk->btrace_seq = 0;
 370 #endif
 371         tsk->splice_pipe = NULL;
 372         tsk->task_frag.page = NULL;
 373
 374         account_kernel_stack(ti, 1);
 375
 376         return tsk;
 377
 378 free_ti:
 379         free_thread_info(ti);
 380 free_tsk:
 381         free_task_struct(tsk);
 382         return NULL;
 383 }
 384
 385 #ifdef CONFIG_MMU
 386 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 387 {
 388         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 389         struct rb_node **rb_link, *rb_parent;
 390         int retval;
 391         unsigned long charge;
 392
 393         uprobe_start_dup_mmap();
 394         down_write(&oldmm->mmap_sem);
 395         flush_cache_dup_mm(oldmm);
 396         uprobe_dup_mmap(oldmm, mm);
 397         /*
 398          * Not linked in yet - no deadlock potential:
 399          */
 400         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 401
 402         mm->total_vm = oldmm->total_vm;
 403         mm->shared_vm = oldmm->shared_vm;
 404         mm->exec_vm = oldmm->exec_vm;
 405         mm->stack_vm = oldmm->stack_vm;
 406
 407         rb_link = &mm->mm_rb.rb_node;
 408         rb_parent = NULL;
 409         pprev = &mm->mmap;
 410         retval = ksm_fork(mm, oldmm);
 411         if (retval)
 412                 goto out;
 413         retval = khugepaged_fork(mm, oldmm);
 414         if (retval)
 415                 goto out;
 416
 417         prev = NULL;
 418         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 419                 struct file *file;
 420
 421                 if (mpnt->vm_flags & VM_DONTCOPY) {
 422                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 423                                                         -vma_pages(mpnt));
 424                         continue;
 425                 }
 426                 charge = 0;
 427                 if (mpnt->vm_flags & VM_ACCOUNT) {
 428                         unsigned long len = vma_pages(mpnt);
 429
 430                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 431                                 goto fail_nomem;
 432                         charge = len;
 433                 }
 434                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 435                 if (!tmp)
 436                         goto fail_nomem;
 437                 *tmp = *mpnt;
 438                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 439                 retval = vma_dup_policy(mpnt, tmp);
 440                 if (retval)
 441                         goto fail_nomem_policy;
 442                 tmp->vm_mm = mm;
 443                 if (anon_vma_fork(tmp, mpnt))
 444                         goto fail_nomem_anon_vma_fork;
 445                 tmp->vm_flags &= ~VM_LOCKED;
 446                 tmp->vm_next = tmp->vm_prev = NULL;
 447                 file = tmp->vm_file;
 448                 if (file) {
 449                         struct inode *inode = file_inode(file);
 450                         struct address_space *mapping = file->f_mapping;
 451
 452                         get_file(file);
 453                         if (tmp->vm_flags & VM_DENYWRITE)
 454                                 atomic_dec(&inode->i_writecount);
 455                         i_mmap_lock_write(mapping);
 456                         if (tmp->vm_flags & VM_SHARED)
 457                                 atomic_inc(&mapping->i_mmap_writable);
 458                         flush_dcache_mmap_lock(mapping);
 459                         /* insert tmp into the share list, just after mpnt */
 460                         vma_interval_tree_insert_after(tmp, mpnt,
 461                                         &mapping->i_mmap);
 462                         flush_dcache_mmap_unlock(mapping);
 463                         i_mmap_unlock_write(mapping);
 464                 }
 465
 466                 /*
 467                  * Clear hugetlb-related page reserves for children. This only
 468                  * affects MAP_PRIVATE mappings. Faults generated by the child
 469                  * are not guaranteed to succeed, even if read-only
 470                  */
 471                 if (is_vm_hugetlb_page(tmp))
 472                         reset_vma_resv_huge_pages(tmp);
 473
 474                 /*
 475                  * Link in the new vma and copy the page table entries.
 476                  */
 477                 *pprev = tmp;
 478                 pprev = &tmp->vm_next;
 479                 tmp->vm_prev = prev;
 480                 prev = tmp;
 481
 482                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 483                 rb_link = &tmp->vm_rb.rb_right;
 484                 rb_parent = &tmp->vm_rb;
 485
 486                 mm->map_count++;
 487                 retval = copy_page_range(mm, oldmm, mpnt);
 488
 489                 if (tmp->vm_ops && tmp->vm_ops->open)
 490                         tmp->vm_ops->open(tmp);
 491
 492                 if (retval)
 493                         goto out;
 494         }
 495         /* a new mm has just been created */
 496         arch_dup_mmap(oldmm, mm);
 497         retval = 0;
 498 out:
 499         up_write(&mm->mmap_sem);
 500         flush_tlb_mm(oldmm);
 501         up_write(&oldmm->mmap_sem);
 502         uprobe_end_dup_mmap();
 503         return retval;
 504 fail_nomem_anon_vma_fork:
 505         mpol_put(vma_policy(tmp));
 506 fail_nomem_policy:
 507         kmem_cache_free(vm_area_cachep, tmp);
 508 fail_nomem:
 509         retval = -ENOMEM;
 510         vm_unacct_memory(charge);
 511         goto out;
 512 }
 513
 514 static inline int mm_alloc_pgd(struct mm_struct *mm)
 515 {
 516         mm->pgd = pgd_alloc(mm);
 517         if (unlikely(!mm->pgd))
 518                 return -ENOMEM;
 519         return 0;
 520 }
 521
 522 static inline void mm_free_pgd(struct mm_struct *mm)
 523 {
 524         pgd_free(mm, mm->pgd);
 525 }
 526 #else
 527 #define dup_mmap(mm, oldmm)     (0)
 528 #define mm_alloc_pgd(mm)        (0)
 529 #define mm_free_pgd(mm)
 530 #endif /* CONFIG_MMU */
 531
 532 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 533
 534 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 535 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 536
 537 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 538
 539 static int __init coredump_filter_setup(char *s)
 540 {
 541         default_dump_filter =
 542                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 543                 MMF_DUMP_FILTER_MASK;
 544         return 1;
 545 }
 546
 547 __setup("coredump_filter=", coredump_filter_setup);
 548
 549 #include <linux/init_task.h>
 550
 551 static void mm_init_aio(struct mm_struct *mm)
 552 {
 553 #ifdef CONFIG_AIO
 554         spin_lock_init(&mm->ioctx_lock);
 555         mm->ioctx_table = NULL;
 556 #endif
 557 }
 558
 559 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 560 {
 561 #ifdef CONFIG_MEMCG
 562         mm->owner = p;
 563 #endif
 564 }
 565
 566 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 567 {
 568         mm->mmap = NULL;
 569         mm->mm_rb = RB_ROOT;
 570         mm->vmacache_seqnum = 0;
 571         atomic_set(&mm->mm_users, 1);
 572         atomic_set(&mm->mm_count, 1);
 573         init_rwsem(&mm->mmap_sem);
 574         INIT_LIST_HEAD(&mm->mmlist);
 575         mm->core_state = NULL;
 576         atomic_long_set(&mm->nr_ptes, 0);
 577         mm_nr_pmds_init(mm);
 578         mm->map_count = 0;
 579         mm->locked_vm = 0;
 580         mm->pinned_vm = 0;
 581         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 582         spin_lock_init(&mm->page_table_lock);
 583         mm_init_cpumask(mm);
 584         mm_init_aio(mm);
 585         mm_init_owner(mm, p);
 586         mmu_notifier_mm_init(mm);
 587         clear_tlb_flush_pending(mm);
 588 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 589         mm->pmd_huge_pte = NULL;
 590 #endif
 591
 592         if (current->mm) {
 593                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 594                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 595         } else {
 596                 mm->flags = default_dump_filter;
 597                 mm->def_flags = 0;
 598         }
 599
 600         if (mm_alloc_pgd(mm))
 601                 goto fail_nopgd;
 602
 603         if (init_new_context(p, mm))
 604                 goto fail_nocontext;
 605
 606         return mm;
 607
 608 fail_nocontext:
 609         mm_free_pgd(mm);
 610 fail_nopgd:
 611         free_mm(mm);
 612         return NULL;
 613 }
 614
 615 static void check_mm(struct mm_struct *mm)
 616 {
 617         int i;
 618
 619         for (i = 0; i < NR_MM_COUNTERS; i++) {
 620                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 621
 622                 if (unlikely(x))
 623                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 624                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 625         }
 626
 627         if (atomic_long_read(&mm->nr_ptes))
 628                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 629                                 atomic_long_read(&mm->nr_ptes));
 630         if (mm_nr_pmds(mm))
 631                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 632                                 mm_nr_pmds(mm));
 633
 634 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 635         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 636 #endif
 637 }
 638
 639 /*
 640  * Allocate and initialize an mm_struct.
 641  */
 642 struct mm_struct *mm_alloc(void)
 643 {
 644         struct mm_struct *mm;
 645
 646         mm = allocate_mm();
 647         if (!mm)
 648                 return NULL;
 649
 650         memset(mm, 0, sizeof(*mm));
 651         return mm_init(mm, current);
 652 }
 653
 654 /*
 655  * Called when the last reference to the mm
 656  * is dropped: either by a lazy thread or by
 657  * mmput. Free the page directory and the mm.
 658  */
 659 void __mmdrop(struct mm_struct *mm)
 660 {
 661         BUG_ON(mm == &init_mm);
 662         mm_free_pgd(mm);
 663         destroy_context(mm);
 664         mmu_notifier_mm_destroy(mm);
 665         check_mm(mm);
 666         free_mm(mm);
 667 }
 668 EXPORT_SYMBOL_GPL(__mmdrop);
 669
 670 /*
 671  * Decrement the use count and release all resources for an mm.
 672  */
 673 void mmput(struct mm_struct *mm)
 674 {
 675         might_sleep();
 676
 677         if (atomic_dec_and_test(&mm->mm_users)) {
 678                 uprobe_clear_state(mm);
 679                 exit_aio(mm);
 680                 ksm_exit(mm);
 681                 khugepaged_exit(mm); /* must run before exit_mmap */
 682                 exit_mmap(mm);
 683                 set_mm_exe_file(mm, NULL);
 684                 if (!list_empty(&mm->mmlist)) {
 685                         spin_lock(&mmlist_lock);
 686                         list_del(&mm->mmlist);
 687                         spin_unlock(&mmlist_lock);
 688                 }
 689                 if (mm->binfmt)
 690                         module_put(mm->binfmt->module);
 691                 mmdrop(mm);
 692         }
 693 }
 694 EXPORT_SYMBOL_GPL(mmput);
 695
 696 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 697 {
 698         if (new_exe_file)
 699                 get_file(new_exe_file);
 700         if (mm->exe_file)
 701                 fput(mm->exe_file);
 702         mm->exe_file = new_exe_file;
 703 }
 704
 705 struct file *get_mm_exe_file(struct mm_struct *mm)
 706 {
 707         struct file *exe_file;
 708
 709         /* We need mmap_sem to protect against races with removal of exe_file */
 710         down_read(&mm->mmap_sem);
 711         exe_file = mm->exe_file;
 712         if (exe_file)
 713                 get_file(exe_file);
 714         up_read(&mm->mmap_sem);
 715         return exe_file;
 716 }
 717
 718 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 719 {
 720         /* It's safe to write the exe_file pointer without exe_file_lock because
 721          * this is called during fork when the task is not yet in /proc */
 722         newmm->exe_file = get_mm_exe_file(oldmm);
 723 }
 724
 725 /**
 726  * get_task_mm - acquire a reference to the task's mm
 727  *
 728  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 729  * this kernel workthread has transiently adopted a user mm with use_mm,
 730  * to do its AIO) is not set and if so returns a reference to it, after
 731  * bumping up the use count.  User must release the mm via mmput()
 732  * after use.  Typically used by /proc and ptrace.
 733  */
 734 struct mm_struct *get_task_mm(struct task_struct *task)
 735 {
 736         struct mm_struct *mm;
 737
 738         task_lock(task);
 739         mm = task->mm;
 740         if (mm) {
 741                 if (task->flags & PF_KTHREAD)
 742                         mm = NULL;
 743                 else
 744                         atomic_inc(&mm->mm_users);
 745         }
 746         task_unlock(task);
 747         return mm;
 748 }
 749 EXPORT_SYMBOL_GPL(get_task_mm);
 750
 751 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 752 {
 753         struct mm_struct *mm;
 754         int err;
 755
 756         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 757         if (err)
 758                 return ERR_PTR(err);
 759
 760         mm = get_task_mm(task);
 761         if (mm && mm != current->mm &&
 762                         !ptrace_may_access(task, mode)) {
 763                 mmput(mm);
 764                 mm = ERR_PTR(-EACCES);
 765         }
 766         mutex_unlock(&task->signal->cred_guard_mutex);
 767
 768         return mm;
 769 }
 770
 771 static void complete_vfork_done(struct task_struct *tsk)
 772 {
 773         struct completion *vfork;
 774
 775         task_lock(tsk);
 776         vfork = tsk->vfork_done;
 777         if (likely(vfork)) {
 778                 tsk->vfork_done = NULL;
 779                 complete(vfork);
 780         }
 781         task_unlock(tsk);
 782 }
 783
 784 static int wait_for_vfork_done(struct task_struct *child,
 785                                 struct completion *vfork)
 786 {
 787         int killed;
 788
 789         freezer_do_not_count();
 790         killed = wait_for_completion_killable(vfork);
 791         freezer_count();
 792
 793         if (killed) {
 794                 task_lock(child);
 795                 child->vfork_done = NULL;
 796                 task_unlock(child);
 797         }
 798
 799         put_task_struct(child);
 800         return killed;
 801 }
 802
 803 /* Please note the differences between mmput and mm_release.
 804  * mmput is called whenever we stop holding onto a mm_struct,
 805  * error success whatever.
 806  *
 807  * mm_release is called after a mm_struct has been removed
 808  * from the current process.
 809  *
 810  * This difference is important for error handling, when we
 811  * only half set up a mm_struct for a new process and need to restore
 812  * the old one.  Because we mmput the new mm_struct before
 813  * restoring the old one. . .
 814  * Eric Biederman 10 January 1998
 815  */
 816 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 817 {
 818         /* Get rid of any futexes when releasing the mm */
 819 #ifdef CONFIG_FUTEX
 820         if (unlikely(tsk->robust_list)) {
 821                 exit_robust_list(tsk);
 822                 tsk->robust_list = NULL;
 823         }
 824 #ifdef CONFIG_COMPAT
 825         if (unlikely(tsk->compat_robust_list)) {
 826                 compat_exit_robust_list(tsk);
 827                 tsk->compat_robust_list = NULL;
 828         }
 829 #endif
 830         if (unlikely(!list_empty(&tsk->pi_state_list)))
 831                 exit_pi_state_list(tsk);
 832 #endif
 833
 834         uprobe_free_utask(tsk);
 835
 836         /* Get rid of any cached register state */
 837         deactivate_mm(tsk, mm);
 838
 839         /*
 840          * If we're exiting normally, clear a user-space tid field if
 841          * requested.  We leave this alone when dying by signal, to leave
 842          * the value intact in a core dump, and to save the unnecessary
 843          * trouble, say, a killed vfork parent shouldn't touch this mm.
 844          * Userland only wants this done for a sys_exit.
 845          */
 846         if (tsk->clear_child_tid) {
 847                 if (!(tsk->flags & PF_SIGNALED) &&
 848                     atomic_read(&mm->mm_users) > 1) {
 849                         /*
 850                          * We don't check the error code - if userspace has
 851                          * not set up a proper pointer then tough luck.
 852                          */
 853                         put_user(0, tsk->clear_child_tid);
 854                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 855                                         1, NULL, NULL, 0);
 856                 }
 857                 tsk->clear_child_tid = NULL;
 858         }
 859
 860         /*
 861          * All done, finally we can wake up parent and return this mm to him.
 862          * Also kthread_stop() uses this completion for synchronization.
 863          */
 864         if (tsk->vfork_done)
 865                 complete_vfork_done(tsk);
 866 }
 867
 868 /*
 869  * Allocate a new mm structure and copy contents from the
 870  * mm structure of the passed in task structure.
 871  */
 872 static struct mm_struct *dup_mm(struct task_struct *tsk)
 873 {
 874         struct mm_struct *mm, *oldmm = current->mm;
 875         int err;
 876
 877         mm = allocate_mm();
 878         if (!mm)
 879                 goto fail_nomem;
 880
 881         memcpy(mm, oldmm, sizeof(*mm));
 882
 883         if (!mm_init(mm, tsk))
 884                 goto fail_nomem;
 885
 886         dup_mm_exe_file(oldmm, mm);
 887
 888         err = dup_mmap(mm, oldmm);
 889         if (err)
 890                 goto free_pt;
 891
 892         mm->hiwater_rss = get_mm_rss(mm);
 893         mm->hiwater_vm = mm->total_vm;
 894
 895         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 896                 goto free_pt;
 897
 898         return mm;
 899
 900 free_pt:
 901         /* don't put binfmt in mmput, we haven't got module yet */
 902         mm->binfmt = NULL;
 903         mmput(mm);
 904
 905 fail_nomem:
 906         return NULL;
 907 }
 908
 909 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 910 {
 911         struct mm_struct *mm, *oldmm;
 912         int retval;
 913
 914         tsk->min_flt = tsk->maj_flt = 0;
 915         tsk->nvcsw = tsk->nivcsw = 0;
 916 #ifdef CONFIG_DETECT_HUNG_TASK
 917         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 918 #endif
 919
 920         tsk->mm = NULL;
 921         tsk->active_mm = NULL;
 922
 923         /*
 924          * Are we cloning a kernel thread?
 925          *
 926          * We need to steal a active VM for that..
 927          */
 928         oldmm = current->mm;
 929         if (!oldmm)
 930                 return 0;
 931
 932         /* initialize the new vmacache entries */
 933         vmacache_flush(tsk);
 934
 935         if (clone_flags & CLONE_VM) {
 936                 atomic_inc(&oldmm->mm_users);
 937                 mm = oldmm;
 938                 goto good_mm;
 939         }
 940
 941         retval = -ENOMEM;
 942         mm = dup_mm(tsk);
 943         if (!mm)
 944                 goto fail_nomem;
 945
 946 good_mm:
 947         tsk->mm = mm;
 948         tsk->active_mm = mm;
 949         return 0;
 950
 951 fail_nomem:
 952         return retval;
 953 }
 954
 955 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 956 {
 957         struct fs_struct *fs = current->fs;
 958         if (clone_flags & CLONE_FS) {
 959                 /* tsk->fs is already what we want */
 960                 spin_lock(&fs->lock);
 961                 if (fs->in_exec) {
 962                         spin_unlock(&fs->lock);
 963                         return -EAGAIN;
 964                 }
 965                 fs->users++;
 966                 spin_unlock(&fs->lock);
 967                 return 0;
 968         }
 969         tsk->fs = copy_fs_struct(fs);
 970         if (!tsk->fs)
 971                 return -ENOMEM;
 972         return 0;
 973 }
 974
 975 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 976 {
 977         struct files_struct *oldf, *newf;
 978         int error = 0;
 979
 980         /*
 981          * A background process may not have any files ...
 982          */
 983         oldf = current->files;
 984         if (!oldf)
 985                 goto out;
 986
 987         if (clone_flags & CLONE_FILES) {
 988                 atomic_inc(&oldf->count);
 989                 goto out;
 990         }
 991
 992         newf = dup_fd(oldf, &error);
 993         if (!newf)
 994                 goto out;
 995
 996         tsk->files = newf;
 997         error = 0;
 998 out:
 999         return error;
1000 }
1001
1002 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1003 {
1004 #ifdef CONFIG_BLOCK
1005         struct io_context *ioc = current->io_context;
1006         struct io_context *new_ioc;
1007
1008         if (!ioc)
1009                 return 0;
1010         /*
1011          * Share io context with parent, if CLONE_IO is set
1012          */
1013         if (clone_flags & CLONE_IO) {
1014                 ioc_task_link(ioc);
1015                 tsk->io_context = ioc;
1016         } else if (ioprio_valid(ioc->ioprio)) {
1017                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1018                 if (unlikely(!new_ioc))
1019                         return -ENOMEM;
1020
1021                 new_ioc->ioprio = ioc->ioprio;
1022                 put_io_context(new_ioc);
1023         }
1024 #endif
1025         return 0;
1026 }
1027
1028 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1029 {
1030         struct sighand_struct *sig;
1031
1032         if (clone_flags & CLONE_SIGHAND) {
1033                 atomic_inc(&current->sighand->count);
1034                 return 0;
1035         }
1036         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1037         rcu_assign_pointer(tsk->sighand, sig);
1038         if (!sig)
1039                 return -ENOMEM;
1040         atomic_set(&sig->count, 1);
1041         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1042         return 0;
1043 }
1044
1045 void __cleanup_sighand(struct sighand_struct *sighand)
1046 {
1047         if (atomic_dec_and_test(&sighand->count)) {
1048                 signalfd_cleanup(sighand);
1049                 /*
1050                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1051                  * without an RCU grace period, see __lock_task_sighand().
1052                  */
1053                 kmem_cache_free(sighand_cachep, sighand);
1054         }
1055 }
1056
1057 /*
1058  * Initialize POSIX timer handling for a thread group.
1059  */
1060 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1061 {
1062         unsigned long cpu_limit;
1063
1064         /* Thread group counters. */
1065         thread_group_cputime_init(sig);
1066
1067         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1068         if (cpu_limit != RLIM_INFINITY) {
1069                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1070                 sig->cputimer.running = 1;
1071         }
1072
1073         /* The timer lists. */
1074         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1075         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1076         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1077 }
1078
1079 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1080 {
1081         struct signal_struct *sig;
1082
1083         if (clone_flags & CLONE_THREAD)
1084                 return 0;
1085
1086         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1087         tsk->signal = sig;
1088         if (!sig)
1089                 return -ENOMEM;
1090
1091         sig->nr_threads = 1;
1092         atomic_set(&sig->live, 1);
1093         atomic_set(&sig->sigcnt, 1);
1094
1095         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1096         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1097         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1098
1099         init_waitqueue_head(&sig->wait_chldexit);
1100         sig->curr_target = tsk;
1101         init_sigpending(&sig->shared_pending);
1102         INIT_LIST_HEAD(&sig->posix_timers);
1103         seqlock_init(&sig->stats_lock);
1104
1105         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1106         sig->real_timer.function = it_real_fn;
1107
1108         task_lock(current->group_leader);
1109         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1110         task_unlock(current->group_leader);
1111
1112         posix_cpu_timers_init_group(sig);
1113
1114         tty_audit_fork(sig);
1115         sched_autogroup_fork(sig);
1116
1117 #ifdef CONFIG_CGROUPS
1118         init_rwsem(&sig->group_rwsem);
1119 #endif
1120
1121         sig->oom_score_adj = current->signal->oom_score_adj;
1122         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1123
1124         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1125                                    current->signal->is_child_subreaper;
1126
1127         mutex_init(&sig->cred_guard_mutex);
1128
1129         return 0;
1130 }
1131
1132 static void copy_seccomp(struct task_struct *p)
1133 {
1134 #ifdef CONFIG_SECCOMP
1135         /*
1136          * Must be called with sighand->lock held, which is common to
1137          * all threads in the group. Holding cred_guard_mutex is not
1138          * needed because this new task is not yet running and cannot
1139          * be racing exec.
1140          */
1141         assert_spin_locked(&current->sighand->siglock);
1142
1143         /* Ref-count the new filter user, and assign it. */
1144         get_seccomp_filter(current);
1145         p->seccomp = current->seccomp;
1146
1147         /*
1148          * Explicitly enable no_new_privs here in case it got set
1149          * between the task_struct being duplicated and holding the
1150          * sighand lock. The seccomp state and nnp must be in sync.
1151          */
1152         if (task_no_new_privs(current))
1153                 task_set_no_new_privs(p);
1154
1155         /*
1156          * If the parent gained a seccomp mode after copying thread
1157          * flags and between before we held the sighand lock, we have
1158          * to manually enable the seccomp thread flag here.
1159          */
1160         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1161                 set_tsk_thread_flag(p, TIF_SECCOMP);
1162 #endif
1163 }
1164
1165 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1166 {
1167         current->clear_child_tid = tidptr;
1168
1169         return task_pid_vnr(current);
1170 }
1171
1172 static void rt_mutex_init_task(struct task_struct *p)
1173 {
1174         raw_spin_lock_init(&p->pi_lock);
1175 #ifdef CONFIG_RT_MUTEXES
1176         p->pi_waiters = RB_ROOT;
1177         p->pi_waiters_leftmost = NULL;
1178         p->pi_blocked_on = NULL;
1179 #endif
1180 }
1181
1182 /*
1183  * Initialize POSIX timer handling for a single task.
1184  */
1185 static void posix_cpu_timers_init(struct task_struct *tsk)
1186 {
1187         tsk->cputime_expires.prof_exp = 0;
1188         tsk->cputime_expires.virt_exp = 0;
1189         tsk->cputime_expires.sched_exp = 0;
1190         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1191         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1192         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1193 }
1194
1195 static inline void
1196 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1197 {
1198          task->pids[type].pid = pid;
1199 }
1200
1201 /*
1202  * This creates a new process as a copy of the old one,
1203  * but does not actually start it yet.
1204  *
1205  * It copies the registers, and all the appropriate
1206  * parts of the process environment (as per the clone
1207  * flags). The actual kick-off is left to the caller.
1208  */
1209 static struct task_struct *copy_process(unsigned long clone_flags,
1210                                         unsigned long stack_start,
1211                                         unsigned long stack_size,
1212                                         int __user *child_tidptr,
1213                                         struct pid *pid,
1214                                         int trace)
1215 {
1216         int retval;
1217         struct task_struct *p;
1218
1219         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1220                 return ERR_PTR(-EINVAL);
1221
1222         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1223                 return ERR_PTR(-EINVAL);
1224
1225         /*
1226          * Thread groups must share signals as well, and detached threads
1227          * can only be started up within the thread group.
1228          */
1229         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1230                 return ERR_PTR(-EINVAL);
1231
1232         /*
1233          * Shared signal handlers imply shared VM. By way of the above,
1234          * thread groups also imply shared VM. Blocking this case allows
1235          * for various simplifications in other code.
1236          */
1237         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1238                 return ERR_PTR(-EINVAL);
1239
1240         /*
1241          * Siblings of global init remain as zombies on exit since they are
1242          * not reaped by their parent (swapper). To solve this and to avoid
1243          * multi-rooted process trees, prevent global and container-inits
1244          * from creating siblings.
1245          */
1246         if ((clone_flags & CLONE_PARENT) &&
1247                                 current->signal->flags & SIGNAL_UNKILLABLE)
1248                 return ERR_PTR(-EINVAL);
1249
1250         /*
1251          * If the new process will be in a different pid or user namespace
1252          * do not allow it to share a thread group or signal handlers or
1253          * parent with the forking task.
1254          */
1255         if (clone_flags & CLONE_SIGHAND) {
1256                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1257                     (task_active_pid_ns(current) !=
1258                                 current->nsproxy->pid_ns_for_children))
1259                         return ERR_PTR(-EINVAL);
1260         }
1261
1262         retval = security_task_create(clone_flags);
1263         if (retval)
1264                 goto fork_out;
1265
1266         retval = -ENOMEM;
1267         p = dup_task_struct(current);
1268         if (!p)
1269                 goto fork_out;
1270
1271         ftrace_graph_init_task(p);
1272
1273         rt_mutex_init_task(p);
1274
1275 #ifdef CONFIG_PROVE_LOCKING
1276         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1277         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1278 #endif
1279         retval = -EAGAIN;
1280         if (atomic_read(&p->real_cred->user->processes) >=
1281                         task_rlimit(p, RLIMIT_NPROC)) {
1282                 if (p->real_cred->user != INIT_USER &&
1283                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1284                         goto bad_fork_free;
1285         }
1286         current->flags &= ~PF_NPROC_EXCEEDED;
1287
1288         retval = copy_creds(p, clone_flags);
1289         if (retval < 0)
1290                 goto bad_fork_free;
1291
1292         /*
1293          * If multiple threads are within copy_process(), then this check
1294          * triggers too late. This doesn't hurt, the check is only there
1295          * to stop root fork bombs.
1296          */
1297         retval = -EAGAIN;
1298         if (nr_threads >= max_threads)
1299                 goto bad_fork_cleanup_count;
1300
1301         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1302         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1303         p->flags |= PF_FORKNOEXEC;
1304         INIT_LIST_HEAD(&p->children);
1305         INIT_LIST_HEAD(&p->sibling);
1306         rcu_copy_process(p);
1307         p->vfork_done = NULL;
1308         spin_lock_init(&p->alloc_lock);
1309
1310         init_sigpending(&p->pending);
1311
1312         p->utime = p->stime = p->gtime = 0;
1313         p->utimescaled = p->stimescaled = 0;
1314 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1315         p->prev_cputime.utime = p->prev_cputime.stime = 0;
1316 #endif
1317 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1318         seqlock_init(&p->vtime_seqlock);
1319         p->vtime_snap = 0;
1320         p->vtime_snap_whence = VTIME_SLEEPING;
1321 #endif
1322
1323 #if defined(SPLIT_RSS_COUNTING)
1324         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1325 #endif
1326
1327         p->default_timer_slack_ns = current->timer_slack_ns;
1328
1329         task_io_accounting_init(&p->ioac);
1330         acct_clear_integrals(p);
1331
1332         posix_cpu_timers_init(p);
1333
1334         p->start_time = ktime_get_ns();
1335         p->real_start_time = ktime_get_boot_ns();
1336         p->io_context = NULL;
1337         p->audit_context = NULL;
1338         if (clone_flags & CLONE_THREAD)
1339                 threadgroup_change_begin(current);
1340         cgroup_fork(p);
1341 #ifdef CONFIG_NUMA
1342         p->mempolicy = mpol_dup(p->mempolicy);
1343         if (IS_ERR(p->mempolicy)) {
1344                 retval = PTR_ERR(p->mempolicy);
1345                 p->mempolicy = NULL;
1346                 goto bad_fork_cleanup_threadgroup_lock;
1347         }
1348 #endif
1349 #ifdef CONFIG_CPUSETS
1350         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1351         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1352         seqcount_init(&p->mems_allowed_seq);
1353 #endif
1354 #ifdef CONFIG_TRACE_IRQFLAGS
1355         p->irq_events = 0;
1356         p->hardirqs_enabled = 0;
1357         p->hardirq_enable_ip = 0;
1358         p->hardirq_enable_event = 0;
1359         p->hardirq_disable_ip = _THIS_IP_;
1360         p->hardirq_disable_event = 0;
1361         p->softirqs_enabled = 1;
1362         p->softirq_enable_ip = _THIS_IP_;
1363         p->softirq_enable_event = 0;
1364         p->softirq_disable_ip = 0;
1365         p->softirq_disable_event = 0;
1366         p->hardirq_context = 0;
1367         p->softirq_context = 0;
1368 #endif
1369 #ifdef CONFIG_LOCKDEP
1370         p->lockdep_depth = 0; /* no locks held yet */
1371         p->curr_chain_key = 0;
1372         p->lockdep_recursion = 0;
1373 #endif
1374
1375 #ifdef CONFIG_DEBUG_MUTEXES
1376         p->blocked_on = NULL; /* not blocked yet */
1377 #endif
1378 #ifdef CONFIG_BCACHE
1379         p->sequential_io        = 0;
1380         p->sequential_io_avg    = 0;
1381 #endif
1382
1383         /* Perform scheduler related setup. Assign this task to a CPU. */
1384         retval = sched_fork(clone_flags, p);
1385         if (retval)
1386                 goto bad_fork_cleanup_policy;
1387
1388         retval = perf_event_init_task(p);
1389         if (retval)
1390                 goto bad_fork_cleanup_policy;
1391         retval = audit_alloc(p);
1392         if (retval)
1393                 goto bad_fork_cleanup_perf;
1394         /* copy all the process information */
1395         shm_init_task(p);
1396         retval = copy_semundo(clone_flags, p);
1397         if (retval)
1398                 goto bad_fork_cleanup_audit;
1399         retval = copy_files(clone_flags, p);
1400         if (retval)
1401                 goto bad_fork_cleanup_semundo;
1402         retval = copy_fs(clone_flags, p);
1403         if (retval)
1404                 goto bad_fork_cleanup_files;
1405         retval = copy_sighand(clone_flags, p);
1406         if (retval)
1407                 goto bad_fork_cleanup_fs;
1408         retval = copy_signal(clone_flags, p);
1409         if (retval)
1410                 goto bad_fork_cleanup_sighand;
1411         retval = copy_mm(clone_flags, p);
1412         if (retval)
1413                 goto bad_fork_cleanup_signal;
1414         retval = copy_namespaces(clone_flags, p);
1415         if (retval)
1416                 goto bad_fork_cleanup_mm;
1417         retval = copy_io(clone_flags, p);
1418         if (retval)
1419                 goto bad_fork_cleanup_namespaces;
1420         retval = copy_thread(clone_flags, stack_start, stack_size, p);
1421         if (retval)
1422                 goto bad_fork_cleanup_io;
1423
1424         if (pid != &init_struct_pid) {
1425                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1426                 if (IS_ERR(pid)) {
1427                         retval = PTR_ERR(pid);
1428                         goto bad_fork_cleanup_io;
1429                 }
1430         }
1431
1432         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1433         /*
1434          * Clear TID on mm_release()?
1435          */
1436         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1437 #ifdef CONFIG_BLOCK
1438         p->plug = NULL;
1439 #endif
1440 #ifdef CONFIG_FUTEX
1441         p->robust_list = NULL;
1442 #ifdef CONFIG_COMPAT
1443         p->compat_robust_list = NULL;
1444 #endif
1445         INIT_LIST_HEAD(&p->pi_state_list);
1446         p->pi_state_cache = NULL;
1447 #endif
1448         /*
1449          * sigaltstack should be cleared when sharing the same VM
1450          */
1451         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1452                 p->sas_ss_sp = p->sas_ss_size = 0;
1453
1454         /*
1455          * Syscall tracing and stepping should be turned off in the
1456          * child regardless of CLONE_PTRACE.
1457          */
1458         user_disable_single_step(p);
1459         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1460 #ifdef TIF_SYSCALL_EMU
1461         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1462 #endif
1463         clear_all_latency_tracing(p);
1464
1465         /* ok, now we should be set up.. */
1466         p->pid = pid_nr(pid);
1467         if (clone_flags & CLONE_THREAD) {
1468                 p->exit_signal = -1;
1469                 p->group_leader = current->group_leader;
1470                 p->tgid = current->tgid;
1471         } else {
1472                 if (clone_flags & CLONE_PARENT)
1473                         p->exit_signal = current->group_leader->exit_signal;
1474                 else
1475                         p->exit_signal = (clone_flags & CSIGNAL);
1476                 p->group_leader = p;
1477                 p->tgid = p->pid;
1478         }
1479
1480         p->nr_dirtied = 0;
1481         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1482         p->dirty_paused_when = 0;
1483
1484         p->pdeath_signal = 0;
1485         INIT_LIST_HEAD(&p->thread_group);
1486         p->task_works = NULL;
1487
1488         /*
1489          * Make it visible to the rest of the system, but dont wake it up yet.
1490          * Need tasklist lock for parent etc handling!
1491          */
1492         write_lock_irq(&tasklist_lock);
1493
1494         /* CLONE_PARENT re-uses the old parent */
1495         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1496                 p->real_parent = current->real_parent;
1497                 p->parent_exec_id = current->parent_exec_id;
1498         } else {
1499                 p->real_parent = current;
1500                 p->parent_exec_id = current->self_exec_id;
1501         }
1502
1503         spin_lock(&current->sighand->siglock);
1504
1505         /*
1506          * Copy seccomp details explicitly here, in case they were changed
1507          * before holding sighand lock.
1508          */
1509         copy_seccomp(p);
1510
1511         /*
1512          * Process group and session signals need to be delivered to just the
1513          * parent before the fork or both the parent and the child after the
1514          * fork. Restart if a signal comes in before we add the new process to
1515          * it's process group.
1516          * A fatal signal pending means that current will exit, so the new
1517          * thread can't slip out of an OOM kill (or normal SIGKILL).
1518         */
1519         recalc_sigpending();
1520         if (signal_pending(current)) {
1521                 spin_unlock(&current->sighand->siglock);
1522                 write_unlock_irq(&tasklist_lock);
1523                 retval = -ERESTARTNOINTR;
1524                 goto bad_fork_free_pid;
1525         }
1526
1527         if (likely(p->pid)) {
1528                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1529
1530                 init_task_pid(p, PIDTYPE_PID, pid);
1531                 if (thread_group_leader(p)) {
1532                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1533                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1534
1535                         if (is_child_reaper(pid)) {
1536                                 ns_of_pid(pid)->child_reaper = p;
1537                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1538                         }
1539
1540                         p->signal->leader_pid = pid;
1541                         p->signal->tty = tty_kref_get(current->signal->tty);
1542                         list_add_tail(&p->sibling, &p->real_parent->children);
1543                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1544                         attach_pid(p, PIDTYPE_PGID);
1545                         attach_pid(p, PIDTYPE_SID);
1546                         __this_cpu_inc(process_counts);
1547                 } else {
1548                         current->signal->nr_threads++;
1549                         atomic_inc(&current->signal->live);
1550                         atomic_inc(&current->signal->sigcnt);
1551                         list_add_tail_rcu(&p->thread_group,
1552                                           &p->group_leader->thread_group);
1553                         list_add_tail_rcu(&p->thread_node,
1554                                           &p->signal->thread_head);
1555                 }
1556                 attach_pid(p, PIDTYPE_PID);
1557                 nr_threads++;
1558         }
1559
1560         total_forks++;
1561         spin_unlock(&current->sighand->siglock);
1562         syscall_tracepoint_update(p);
1563         write_unlock_irq(&tasklist_lock);
1564
1565         proc_fork_connector(p);
1566         cgroup_post_fork(p);
1567         if (clone_flags & CLONE_THREAD)
1568                 threadgroup_change_end(current);
1569         perf_event_fork(p);
1570
1571         trace_task_newtask(p, clone_flags);
1572         uprobe_copy_process(p, clone_flags);
1573
1574         return p;
1575
1576 bad_fork_free_pid:
1577         if (pid != &init_struct_pid)
1578                 free_pid(pid);
1579 bad_fork_cleanup_io:
1580         if (p->io_context)
1581                 exit_io_context(p);
1582 bad_fork_cleanup_namespaces:
1583         exit_task_namespaces(p);
1584 bad_fork_cleanup_mm:
1585         if (p->mm)
1586                 mmput(p->mm);
1587 bad_fork_cleanup_signal:
1588         if (!(clone_flags & CLONE_THREAD))
1589                 free_signal_struct(p->signal);
1590 bad_fork_cleanup_sighand:
1591         __cleanup_sighand(p->sighand);
1592 bad_fork_cleanup_fs:
1593         exit_fs(p); /* blocking */
1594 bad_fork_cleanup_files:
1595         exit_files(p); /* blocking */
1596 bad_fork_cleanup_semundo:
1597         exit_sem(p);
1598 bad_fork_cleanup_audit:
1599         audit_free(p);
1600 bad_fork_cleanup_perf:
1601         perf_event_free_task(p);
1602 bad_fork_cleanup_policy:
1603 #ifdef CONFIG_NUMA
1604         mpol_put(p->mempolicy);
1605 bad_fork_cleanup_threadgroup_lock:
1606 #endif
1607         if (clone_flags & CLONE_THREAD)
1608                 threadgroup_change_end(current);
1609         delayacct_tsk_free(p);
1610 bad_fork_cleanup_count:
1611         atomic_dec(&p->cred->user->processes);
1612         exit_creds(p);
1613 bad_fork_free:
1614         free_task(p);
1615 fork_out:
1616         return ERR_PTR(retval);
1617 }
1618
1619 static inline void init_idle_pids(struct pid_link *links)
1620 {
1621         enum pid_type type;
1622
1623         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1624                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1625                 links[type].pid = &init_struct_pid;
1626         }
1627 }
1628
1629 struct task_struct *fork_idle(int cpu)
1630 {
1631         struct task_struct *task;
1632         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1633         if (!IS_ERR(task)) {
1634                 init_idle_pids(task->pids);
1635                 init_idle(task, cpu);
1636         }
1637
1638         return task;
1639 }
1640
1641 /*
1642  *  Ok, this is the main fork-routine.
1643  *
1644  * It copies the process, and if successful kick-starts
1645  * it and waits for it to finish using the VM if required.
1646  */
1647 long do_fork(unsigned long clone_flags,
1648               unsigned long stack_start,
1649               unsigned long stack_size,
1650               int __user *parent_tidptr,
1651               int __user *child_tidptr)
1652 {
1653         struct task_struct *p;
1654         int trace = 0;
1655         long nr;
1656
1657         /*
1658          * Determine whether and which event to report to ptracer.  When
1659          * called from kernel_thread or CLONE_UNTRACED is explicitly
1660          * requested, no event is reported; otherwise, report if the event
1661          * for the type of forking is enabled.
1662          */
1663         if (!(clone_flags & CLONE_UNTRACED)) {
1664                 if (clone_flags & CLONE_VFORK)
1665                         trace = PTRACE_EVENT_VFORK;
1666                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1667                         trace = PTRACE_EVENT_CLONE;
1668                 else
1669                         trace = PTRACE_EVENT_FORK;
1670
1671                 if (likely(!ptrace_event_enabled(current, trace)))
1672                         trace = 0;
1673         }
1674
1675         p = copy_process(clone_flags, stack_start, stack_size,
1676                          child_tidptr, NULL, trace);
1677         /*
1678          * Do this prior waking up the new thread - the thread pointer
1679          * might get invalid after that point, if the thread exits quickly.
1680          */
1681         if (!IS_ERR(p)) {
1682                 struct completion vfork;
1683                 struct pid *pid;
1684
1685                 trace_sched_process_fork(current, p);
1686
1687                 pid = get_task_pid(p, PIDTYPE_PID);
1688                 nr = pid_vnr(pid);
1689
1690                 if (clone_flags & CLONE_PARENT_SETTID)
1691                         put_user(nr, parent_tidptr);
1692
1693                 if (clone_flags & CLONE_VFORK) {
1694                         p->vfork_done = &vfork;
1695                         init_completion(&vfork);
1696                         get_task_struct(p);
1697                 }
1698
1699                 wake_up_new_task(p);
1700
1701                 /* forking complete and child started to run, tell ptracer */
1702                 if (unlikely(trace))
1703                         ptrace_event_pid(trace, pid);
1704
1705                 if (clone_flags & CLONE_VFORK) {
1706                         if (!wait_for_vfork_done(p, &vfork))
1707                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1708                 }
1709
1710                 put_pid(pid);
1711         } else {
1712                 nr = PTR_ERR(p);
1713         }
1714         return nr;
1715 }
1716
1717 /*
1718  * Create a kernel thread.
1719  */
1720 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1721 {
1722         return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1723                 (unsigned long)arg, NULL, NULL);
1724 }
1725
1726 #ifdef __ARCH_WANT_SYS_FORK
1727 SYSCALL_DEFINE0(fork)
1728 {
1729 #ifdef CONFIG_MMU
1730         return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1731 #else
1732         /* can not support in nommu mode */
1733         return -EINVAL;
1734 #endif
1735 }
1736 #endif
1737
1738 #ifdef __ARCH_WANT_SYS_VFORK
1739 SYSCALL_DEFINE0(vfork)
1740 {
1741         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1742                         0, NULL, NULL);
1743 }
1744 #endif
1745
1746 #ifdef __ARCH_WANT_SYS_CLONE
1747 #ifdef CONFIG_CLONE_BACKWARDS
1748 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1749                  int __user *, parent_tidptr,
1750                  int, tls_val,
1751                  int __user *, child_tidptr)
1752 #elif defined(CONFIG_CLONE_BACKWARDS2)
1753 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1754                  int __user *, parent_tidptr,
1755                  int __user *, child_tidptr,
1756                  int, tls_val)
1757 #elif defined(CONFIG_CLONE_BACKWARDS3)
1758 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1759                 int, stack_size,
1760                 int __user *, parent_tidptr,
1761                 int __user *, child_tidptr,
1762                 int, tls_val)
1763 #else
1764 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1765                  int __user *, parent_tidptr,
1766                  int __user *, child_tidptr,
1767                  int, tls_val)
1768 #endif
1769 {
1770         return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1771 }
1772 #endif
1773
1774 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1775 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1776 #endif
1777
1778 static void sighand_ctor(void *data)
1779 {
1780         struct sighand_struct *sighand = data;
1781
1782         spin_lock_init(&sighand->siglock);
1783         init_waitqueue_head(&sighand->signalfd_wqh);
1784 }
1785
1786 void __init proc_caches_init(void)
1787 {
1788         sighand_cachep = kmem_cache_create("sighand_cache",
1789                         sizeof(struct sighand_struct), 0,
1790                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1791                         SLAB_NOTRACK, sighand_ctor);
1792         signal_cachep = kmem_cache_create("signal_cache",
1793                         sizeof(struct signal_struct), 0,
1794                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1795         files_cachep = kmem_cache_create("files_cache",
1796                         sizeof(struct files_struct), 0,
1797                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1798         fs_cachep = kmem_cache_create("fs_cache",
1799                         sizeof(struct fs_struct), 0,
1800                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1801         /*
1802          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1803          * whole struct cpumask for the OFFSTACK case. We could change
1804          * this to *only* allocate as much of it as required by the
1805          * maximum number of CPU's we can ever have.  The cpumask_allocation
1806          * is at the end of the structure, exactly for that reason.
1807          */
1808         mm_cachep = kmem_cache_create("mm_struct",
1809                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1810                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1811         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1812         mmap_init();
1813         nsproxy_cache_init();
1814 }
1815
1816 /*
1817  * Check constraints on flags passed to the unshare system call.
1818  */
1819 static int check_unshare_flags(unsigned long unshare_flags)
1820 {
1821         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1822                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1823                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1824                                 CLONE_NEWUSER|CLONE_NEWPID))
1825                 return -EINVAL;
1826         /*
1827          * Not implemented, but pretend it works if there is nothing to
1828          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1829          * needs to unshare vm.
1830          */
1831         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1832                 /* FIXME: get_task_mm() increments ->mm_users */
1833                 if (atomic_read(&current->mm->mm_users) > 1)
1834                         return -EINVAL;
1835         }
1836
1837         return 0;
1838 }
1839
1840 /*
1841  * Unshare the filesystem structure if it is being shared
1842  */
1843 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1844 {
1845         struct fs_struct *fs = current->fs;
1846
1847         if (!(unshare_flags & CLONE_FS) || !fs)
1848                 return 0;
1849
1850         /* don't need lock here; in the worst case we'll do useless copy */
1851         if (fs->users == 1)
1852                 return 0;
1853
1854         *new_fsp = copy_fs_struct(fs);
1855         if (!*new_fsp)
1856                 return -ENOMEM;
1857
1858         return 0;
1859 }
1860
1861 /*
1862  * Unshare file descriptor table if it is being shared
1863  */
1864 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1865 {
1866         struct files_struct *fd = current->files;
1867         int error = 0;
1868
1869         if ((unshare_flags & CLONE_FILES) &&
1870             (fd && atomic_read(&fd->count) > 1)) {
1871                 *new_fdp = dup_fd(fd, &error);
1872                 if (!*new_fdp)
1873                         return error;
1874         }
1875
1876         return 0;
1877 }
1878
1879 /*
1880  * unshare allows a process to 'unshare' part of the process
1881  * context which was originally shared using clone.  copy_*
1882  * functions used by do_fork() cannot be used here directly
1883  * because they modify an inactive task_struct that is being
1884  * constructed. Here we are modifying the current, active,
1885  * task_struct.
1886  */
1887 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1888 {
1889         struct fs_struct *fs, *new_fs = NULL;
1890         struct files_struct *fd, *new_fd = NULL;
1891         struct cred *new_cred = NULL;
1892         struct nsproxy *new_nsproxy = NULL;
1893         int do_sysvsem = 0;
1894         int err;
1895
1896         /*
1897          * If unsharing a user namespace must also unshare the thread.
1898          */
1899         if (unshare_flags & CLONE_NEWUSER)
1900                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1901         /*
1902          * If unsharing a thread from a thread group, must also unshare vm.
1903          */
1904         if (unshare_flags & CLONE_THREAD)
1905                 unshare_flags |= CLONE_VM;
1906         /*
1907          * If unsharing vm, must also unshare signal handlers.
1908          */
1909         if (unshare_flags & CLONE_VM)
1910                 unshare_flags |= CLONE_SIGHAND;
1911         /*
1912          * If unsharing namespace, must also unshare filesystem information.
1913          */
1914         if (unshare_flags & CLONE_NEWNS)
1915                 unshare_flags |= CLONE_FS;
1916
1917         err = check_unshare_flags(unshare_flags);
1918         if (err)
1919                 goto bad_unshare_out;
1920         /*
1921          * CLONE_NEWIPC must also detach from the undolist: after switching
1922          * to a new ipc namespace, the semaphore arrays from the old
1923          * namespace are unreachable.
1924          */
1925         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1926                 do_sysvsem = 1;
1927         err = unshare_fs(unshare_flags, &new_fs);
1928         if (err)
1929                 goto bad_unshare_out;
1930         err = unshare_fd(unshare_flags, &new_fd);
1931         if (err)
1932                 goto bad_unshare_cleanup_fs;
1933         err = unshare_userns(unshare_flags, &new_cred);
1934         if (err)
1935                 goto bad_unshare_cleanup_fd;
1936         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1937                                          new_cred, new_fs);
1938         if (err)
1939                 goto bad_unshare_cleanup_cred;
1940
1941         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1942                 if (do_sysvsem) {
1943                         /*
1944                          * CLONE_SYSVSEM is equivalent to sys_exit().
1945                          */
1946                         exit_sem(current);
1947                 }
1948                 if (unshare_flags & CLONE_NEWIPC) {
1949                         /* Orphan segments in old ns (see sem above). */
1950                         exit_shm(current);
1951                         shm_init_task(current);
1952                 }
1953
1954                 if (new_nsproxy)
1955                         switch_task_namespaces(current, new_nsproxy);
1956
1957                 task_lock(current);
1958
1959                 if (new_fs) {
1960                         fs = current->fs;
1961                         spin_lock(&fs->lock);
1962                         current->fs = new_fs;
1963                         if (--fs->users)
1964                                 new_fs = NULL;
1965                         else
1966                                 new_fs = fs;
1967                         spin_unlock(&fs->lock);
1968                 }
1969
1970                 if (new_fd) {
1971                         fd = current->files;
1972                         current->files = new_fd;
1973                         new_fd = fd;
1974                 }
1975
1976                 task_unlock(current);
1977
1978                 if (new_cred) {
1979                         /* Install the new user namespace */
1980                         commit_creds(new_cred);
1981                         new_cred = NULL;
1982                 }
1983         }
1984
1985 bad_unshare_cleanup_cred:
1986         if (new_cred)
1987                 put_cred(new_cred);
1988 bad_unshare_cleanup_fd:
1989         if (new_fd)
1990                 put_files_struct(new_fd);
1991
1992 bad_unshare_cleanup_fs:
1993         if (new_fs)
1994                 free_fs_struct(new_fs);
1995
1996 bad_unshare_out:
1997         return err;
1998 }
1999
2000 /*
2001  *      Helper to unshare the files of the current task.
2002  *      We don't want to expose copy_files internals to
2003  *      the exec layer of the kernel.
2004  */
2005
2006 int unshare_files(struct files_struct **displaced)
2007 {
2008         struct task_struct *task = current;
2009         struct files_struct *copy = NULL;
2010         int error;
2011
2012         error = unshare_fd(CLONE_FILES, &copy);
2013         if (error || !copy) {
2014                 *displaced = NULL;
2015                 return error;
2016         }
2017         *displaced = task->files;
2018         task_lock(task);
2019         task->files = copy;
2020         task_unlock(task);
2021         return 0;
2022 }