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