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