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