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