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/nsproxy.h>
  32 #include <linux/capability.h>
  33 #include <linux/cpu.h>
  34 #include <linux/cgroup.h>
  35 #include <linux/security.h>
  36 #include <linux/hugetlb.h>
  37 #include <linux/seccomp.h>
  38 #include <linux/swap.h>
  39 #include <linux/syscalls.h>
  40 #include <linux/jiffies.h>
  41 #include <linux/futex.h>
  42 #include <linux/compat.h>
  43 #include <linux/kthread.h>
  44 #include <linux/task_io_accounting_ops.h>
  45 #include <linux/rcupdate.h>
  46 #include <linux/ptrace.h>
  47 #include <linux/mount.h>
  48 #include <linux/audit.h>
  49 #include <linux/memcontrol.h>
  50 #include <linux/ftrace.h>
  51 #include <linux/proc_fs.h>
  52 #include <linux/profile.h>
  53 #include <linux/rmap.h>
  54 #include <linux/ksm.h>
  55 #include <linux/acct.h>
  56 #include <linux/tsacct_kern.h>
  57 #include <linux/cn_proc.h>
  58 #include <linux/freezer.h>
  59 #include <linux/delayacct.h>
  60 #include <linux/taskstats_kern.h>
  61 #include <linux/random.h>
  62 #include <linux/tty.h>
  63 #include <linux/blkdev.h>
  64 #include <linux/fs_struct.h>
  65 #include <linux/magic.h>
  66 #include <linux/perf_event.h>
  67 #include <linux/posix-timers.h>
  68 #include <linux/user-return-notifier.h>
  69 #include <linux/oom.h>
  70 #include <linux/khugepaged.h>
  71 #include <linux/signalfd.h>
  72 #include <linux/uprobes.h>
  73 #include <linux/aio.h>
  74
  75 #include <asm/pgtable.h>
  76 #include <asm/pgalloc.h>
  77 #include <asm/uaccess.h>
  78 #include <asm/mmu_context.h>
  79 #include <asm/cacheflush.h>
  80 #include <asm/tlbflush.h>
  81
  82 #include <trace/events/sched.h>
  83
  84 #define CREATE_TRACE_POINTS
  85 #include <trace/events/task.h>
  86
  87 /*
  88  * Protected counters by write_lock_irq(&tasklist_lock)
  89  */
  90 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  91 int nr_threads;                 /* The idle threads do not count.. */
  92
  93 int max_threads;                /* tunable limit on nr_threads */
  94
  95 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  96
  97 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  98
  99 #ifdef CONFIG_PROVE_RCU
 100 int lockdep_tasklist_lock_is_held(void)
 101 {
 102         return lockdep_is_held(&tasklist_lock);
 103 }
 104 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 105 #endif /* #ifdef CONFIG_PROVE_RCU */
 106
 107 int nr_processes(void)
 108 {
 109         int cpu;
 110         int total = 0;
 111
 112         for_each_possible_cpu(cpu)
 113                 total += per_cpu(process_counts, cpu);
 114
 115         return total;
 116 }
 117
 118 void __weak arch_release_task_struct(struct task_struct *tsk)
 119 {
 120 }
 121
 122 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 123 static struct kmem_cache *task_struct_cachep;
 124
 125 static inline struct task_struct *alloc_task_struct_node(int node)
 126 {
 127         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 128 }
 129
 130 static inline void free_task_struct(struct task_struct *tsk)
 131 {
 132         kmem_cache_free(task_struct_cachep, tsk);
 133 }
 134 #endif
 135
 136 void __weak arch_release_thread_info(struct thread_info *ti)
 137 {
 138 }
 139
 140 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
 141
 142 /*
 143  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 144  * kmemcache based allocator.
 145  */
 146 # if THREAD_SIZE >= PAGE_SIZE
 147 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 148                                                   int node)
 149 {
 150         struct page *page = alloc_pages_node(node, THREADINFO_GFP_ACCOUNTED,
 151                                              THREAD_SIZE_ORDER);
 152
 153         return page ? page_address(page) : NULL;
 154 }
 155
 156 static inline void free_thread_info(struct thread_info *ti)
 157 {
 158         free_memcg_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 159 }
 160 # else
 161 static struct kmem_cache *thread_info_cache;
 162
 163 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 164                                                   int node)
 165 {
 166         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
 167 }
 168
 169 static void free_thread_info(struct thread_info *ti)
 170 {
 171         kmem_cache_free(thread_info_cache, ti);
 172 }
 173
 174 void thread_info_cache_init(void)
 175 {
 176         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
 177                                               THREAD_SIZE, 0, NULL);
 178         BUG_ON(thread_info_cache == NULL);
 179 }
 180 # endif
 181 #endif
 182
 183 /* SLAB cache for signal_struct structures (tsk->signal) */
 184 static struct kmem_cache *signal_cachep;
 185
 186 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 187 struct kmem_cache *sighand_cachep;
 188
 189 /* SLAB cache for files_struct structures (tsk->files) */
 190 struct kmem_cache *files_cachep;
 191
 192 /* SLAB cache for fs_struct structures (tsk->fs) */
 193 struct kmem_cache *fs_cachep;
 194
 195 /* SLAB cache for vm_area_struct structures */
 196 struct kmem_cache *vm_area_cachep;
 197
 198 /* SLAB cache for mm_struct structures (tsk->mm) */
 199 static struct kmem_cache *mm_cachep;
 200
 201 /* Notifier list called when a task struct is freed */
 202 static ATOMIC_NOTIFIER_HEAD(task_free_notifier);
 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 int task_free_register(struct notifier_block *n)
 238 {
 239         return atomic_notifier_chain_register(&task_free_notifier, n);
 240 }
 241 EXPORT_SYMBOL(task_free_register);
 242
 243 int task_free_unregister(struct notifier_block *n)
 244 {
 245         return atomic_notifier_chain_unregister(&task_free_notifier, n);
 246 }
 247 EXPORT_SYMBOL(task_free_unregister);
 248
 249 void __put_task_struct(struct task_struct *tsk)
 250 {
 251         WARN_ON(!tsk->exit_state);
 252         WARN_ON(atomic_read(&tsk->usage));
 253         WARN_ON(tsk == current);
 254
 255         security_task_free(tsk);
 256         exit_creds(tsk);
 257         delayacct_tsk_free(tsk);
 258         put_signal_struct(tsk->signal);
 259
 260         atomic_notifier_call_chain(&task_free_notifier, 0, tsk);
 261         if (!profile_handoff_task(tsk))
 262                 free_task(tsk);
 263 }
 264 EXPORT_SYMBOL_GPL(__put_task_struct);
 265
 266 void __init __weak arch_task_cache_init(void) { }
 267
 268 void __init fork_init(unsigned long mempages)
 269 {
 270 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 271 #ifndef ARCH_MIN_TASKALIGN
 272 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 273 #endif
 274         /* create a slab on which task_structs can be allocated */
 275         task_struct_cachep =
 276                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 277                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 278 #endif
 279
 280         /* do the arch specific task caches init */
 281         arch_task_cache_init();
 282
 283         /*
 284          * The default maximum number of threads is set to a safe
 285          * value: the thread structures can take up at most half
 286          * of memory.
 287          */
 288         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 289
 290         /*
 291          * we need to allow at least 20 threads to boot a system
 292          */
 293         if (max_threads < 20)
 294                 max_threads = 20;
 295
 296         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 297         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 298         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 299                 init_task.signal->rlim[RLIMIT_NPROC];
 300 }
 301
 302 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 303                                                struct task_struct *src)
 304 {
 305         *dst = *src;
 306         return 0;
 307 }
 308
 309 static struct task_struct *dup_task_struct(struct task_struct *orig)
 310 {
 311         struct task_struct *tsk;
 312         struct thread_info *ti;
 313         unsigned long *stackend;
 314         int node = tsk_fork_get_node(orig);
 315         int err;
 316
 317         tsk = alloc_task_struct_node(node);
 318         if (!tsk)
 319                 return NULL;
 320
 321         ti = alloc_thread_info_node(tsk, node);
 322         if (!ti)
 323                 goto free_tsk;
 324
 325         err = arch_dup_task_struct(tsk, orig);
 326         if (err)
 327                 goto free_ti;
 328
 329         tsk->stack = ti;
 330 #ifdef CONFIG_SECCOMP
 331         /*
 332          * We must handle setting up seccomp filters once we're under
 333          * the sighand lock in case orig has changed between now and
 334          * then. Until then, filter must be NULL to avoid messing up
 335          * the usage counts on the error path calling free_task.
 336          */
 337         tsk->seccomp.filter = NULL;
 338 #endif
 339
 340         setup_thread_stack(tsk, orig);
 341         clear_user_return_notifier(tsk);
 342         clear_tsk_need_resched(tsk);
 343         stackend = end_of_stack(tsk);
 344         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 345
 346 #ifdef CONFIG_CC_STACKPROTECTOR
 347         tsk->stack_canary = get_random_int();
 348 #endif
 349
 350         /*
 351          * One for us, one for whoever does the "release_task()" (usually
 352          * parent)
 353          */
 354         atomic_set(&tsk->usage, 2);
 355 #ifdef CONFIG_BLK_DEV_IO_TRACE
 356         tsk->btrace_seq = 0;
 357 #endif
 358         tsk->splice_pipe = NULL;
 359         tsk->task_frag.page = NULL;
 360
 361         account_kernel_stack(ti, 1);
 362
 363         return tsk;
 364
 365 free_ti:
 366         free_thread_info(ti);
 367 free_tsk:
 368         free_task_struct(tsk);
 369         return NULL;
 370 }
 371
 372 #ifdef CONFIG_MMU
 373 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 374 {
 375         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 376         struct rb_node **rb_link, *rb_parent;
 377         int retval;
 378         unsigned long charge;
 379         struct mempolicy *pol;
 380
 381         uprobe_start_dup_mmap();
 382         down_write(&oldmm->mmap_sem);
 383         flush_cache_dup_mm(oldmm);
 384         uprobe_dup_mmap(oldmm, mm);
 385         /*
 386          * Not linked in yet - no deadlock potential:
 387          */
 388         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 389
 390         mm->locked_vm = 0;
 391         mm->mmap = NULL;
 392         mm->mmap_cache = NULL;
 393         mm->free_area_cache = oldmm->mmap_base;
 394         mm->cached_hole_size = ~0UL;
 395         mm->map_count = 0;
 396         cpumask_clear(mm_cpumask(mm));
 397         mm->mm_rb = RB_ROOT;
 398         rb_link = &mm->mm_rb.rb_node;
 399         rb_parent = NULL;
 400         pprev = &mm->mmap;
 401         retval = ksm_fork(mm, oldmm);
 402         if (retval)
 403                 goto out;
 404         retval = khugepaged_fork(mm, oldmm);
 405         if (retval)
 406                 goto out;
 407
 408         prev = NULL;
 409         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 410                 struct file *file;
 411
 412                 if (mpnt->vm_flags & VM_DONTCOPY) {
 413                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 414                                                         -vma_pages(mpnt));
 415                         continue;
 416                 }
 417                 charge = 0;
 418                 if (mpnt->vm_flags & VM_ACCOUNT) {
 419                         unsigned long len = vma_pages(mpnt);
 420
 421                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 422                                 goto fail_nomem;
 423                         charge = len;
 424                 }
 425                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 426                 if (!tmp)
 427                         goto fail_nomem;
 428                 *tmp = *mpnt;
 429                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 430                 pol = mpol_dup(vma_policy(mpnt));
 431                 retval = PTR_ERR(pol);
 432                 if (IS_ERR(pol))
 433                         goto fail_nomem_policy;
 434                 vma_set_policy(tmp, pol);
 435                 tmp->vm_mm = mm;
 436                 if (anon_vma_fork(tmp, mpnt))
 437                         goto fail_nomem_anon_vma_fork;
 438                 tmp->vm_flags &= ~VM_LOCKED;
 439                 tmp->vm_next = tmp->vm_prev = NULL;
 440                 file = tmp->vm_file;
 441                 if (file) {
 442                         struct inode *inode = file_inode(file);
 443                         struct address_space *mapping = file->f_mapping;
 444
 445                         get_file(file);
 446                         if (tmp->vm_flags & VM_DENYWRITE)
 447                                 atomic_dec(&inode->i_writecount);
 448                         mutex_lock(&mapping->i_mmap_mutex);
 449                         if (tmp->vm_flags & VM_SHARED)
 450                                 mapping->i_mmap_writable++;
 451                         flush_dcache_mmap_lock(mapping);
 452                         /* insert tmp into the share list, just after mpnt */
 453                         if (unlikely(tmp->vm_flags & VM_NONLINEAR))
 454                                 vma_nonlinear_insert(tmp,
 455                                                 &mapping->i_mmap_nonlinear);
 456                         else
 457                                 vma_interval_tree_insert_after(tmp, mpnt,
 458                                                         &mapping->i_mmap);
 459                         flush_dcache_mmap_unlock(mapping);
 460                         mutex_unlock(&mapping->i_mmap_mutex);
 461                 }
 462
 463                 /*
 464                  * Clear hugetlb-related page reserves for children. This only
 465                  * affects MAP_PRIVATE mappings. Faults generated by the child
 466                  * are not guaranteed to succeed, even if read-only
 467                  */
 468                 if (is_vm_hugetlb_page(tmp))
 469                         reset_vma_resv_huge_pages(tmp);
 470
 471                 /*
 472                  * Link in the new vma and copy the page table entries.
 473                  */
 474                 *pprev = tmp;
 475                 pprev = &tmp->vm_next;
 476                 tmp->vm_prev = prev;
 477                 prev = tmp;
 478
 479                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 480                 rb_link = &tmp->vm_rb.rb_right;
 481                 rb_parent = &tmp->vm_rb;
 482
 483                 mm->map_count++;
 484                 retval = copy_page_range(mm, oldmm, mpnt);
 485
 486                 if (tmp->vm_ops && tmp->vm_ops->open)
 487                         tmp->vm_ops->open(tmp);
 488
 489                 if (retval)
 490                         goto out;
 491         }
 492         /* a new mm has just been created */
 493         arch_dup_mmap(oldmm, mm);
 494         retval = 0;
 495 out:
 496         up_write(&mm->mmap_sem);
 497         flush_tlb_mm(oldmm);
 498         up_write(&oldmm->mmap_sem);
 499         uprobe_end_dup_mmap();
 500         return retval;
 501 fail_nomem_anon_vma_fork:
 502         mpol_put(pol);
 503 fail_nomem_policy:
 504         kmem_cache_free(vm_area_cachep, tmp);
 505 fail_nomem:
 506         retval = -ENOMEM;
 507         vm_unacct_memory(charge);
 508         goto out;
 509 }
 510
 511 static inline int mm_alloc_pgd(struct mm_struct *mm)
 512 {
 513         mm->pgd = pgd_alloc(mm);
 514         if (unlikely(!mm->pgd))
 515                 return -ENOMEM;
 516         return 0;
 517 }
 518
 519 static inline void mm_free_pgd(struct mm_struct *mm)
 520 {
 521         pgd_free(mm, mm->pgd);
 522 }
 523 #else
 524 #define dup_mmap(mm, oldmm)     (0)
 525 #define mm_alloc_pgd(mm)        (0)
 526 #define mm_free_pgd(mm)
 527 #endif /* CONFIG_MMU */
 528
 529 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 530
 531 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 532 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 533
 534 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 535
 536 static int __init coredump_filter_setup(char *s)
 537 {
 538         default_dump_filter =
 539                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 540                 MMF_DUMP_FILTER_MASK;
 541         return 1;
 542 }
 543
 544 __setup("coredump_filter=", coredump_filter_setup);
 545
 546 #include <linux/init_task.h>
 547
 548 static void mm_init_aio(struct mm_struct *mm)
 549 {
 550 #ifdef CONFIG_AIO
 551         spin_lock_init(&mm->ioctx_lock);
 552         INIT_HLIST_HEAD(&mm->ioctx_list);
 553 #endif
 554 }
 555
 556 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 557 {
 558         atomic_set(&mm->mm_users, 1);
 559         atomic_set(&mm->mm_count, 1);
 560         init_rwsem(&mm->mmap_sem);
 561         INIT_LIST_HEAD(&mm->mmlist);
 562         mm->flags = (current->mm) ?
 563                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 564         mm->core_state = NULL;
 565         mm->nr_ptes = 0;
 566         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 567         spin_lock_init(&mm->page_table_lock);
 568         mm->free_area_cache = TASK_UNMAPPED_BASE;
 569         mm->cached_hole_size = ~0UL;
 570         mm_init_aio(mm);
 571         mm_init_owner(mm, p);
 572         clear_tlb_flush_pending(mm);
 573
 574         if (likely(!mm_alloc_pgd(mm))) {
 575                 mm->def_flags = 0;
 576                 mmu_notifier_mm_init(mm);
 577                 return mm;
 578         }
 579
 580         free_mm(mm);
 581         return NULL;
 582 }
 583
 584 static void check_mm(struct mm_struct *mm)
 585 {
 586         int i;
 587
 588         for (i = 0; i < NR_MM_COUNTERS; i++) {
 589                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 590
 591                 if (unlikely(x))
 592                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 593                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 594         }
 595
 596 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 597         VM_BUG_ON(mm->pmd_huge_pte);
 598 #endif
 599 }
 600
 601 /*
 602  * Allocate and initialize an mm_struct.
 603  */
 604 struct mm_struct *mm_alloc(void)
 605 {
 606         struct mm_struct *mm;
 607
 608         mm = allocate_mm();
 609         if (!mm)
 610                 return NULL;
 611
 612         memset(mm, 0, sizeof(*mm));
 613         mm_init_cpumask(mm);
 614         return mm_init(mm, current);
 615 }
 616
 617 /*
 618  * Called when the last reference to the mm
 619  * is dropped: either by a lazy thread or by
 620  * mmput. Free the page directory and the mm.
 621  */
 622 void __mmdrop(struct mm_struct *mm)
 623 {
 624         BUG_ON(mm == &init_mm);
 625         mm_free_pgd(mm);
 626         destroy_context(mm);
 627         mmu_notifier_mm_destroy(mm);
 628         check_mm(mm);
 629         free_mm(mm);
 630 }
 631 EXPORT_SYMBOL_GPL(__mmdrop);
 632
 633 /*
 634  * Decrement the use count and release all resources for an mm.
 635  */
 636 void mmput(struct mm_struct *mm)
 637 {
 638         might_sleep();
 639
 640         if (atomic_dec_and_test(&mm->mm_users)) {
 641                 uprobe_clear_state(mm);
 642                 exit_aio(mm);
 643                 ksm_exit(mm);
 644                 khugepaged_exit(mm); /* must run before exit_mmap */
 645                 exit_mmap(mm);
 646                 set_mm_exe_file(mm, NULL);
 647                 if (!list_empty(&mm->mmlist)) {
 648                         spin_lock(&mmlist_lock);
 649                         list_del(&mm->mmlist);
 650                         spin_unlock(&mmlist_lock);
 651                 }
 652                 if (mm->binfmt)
 653                         module_put(mm->binfmt->module);
 654                 mmdrop(mm);
 655         }
 656 }
 657 EXPORT_SYMBOL_GPL(mmput);
 658
 659 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 660 {
 661         if (new_exe_file)
 662                 get_file(new_exe_file);
 663         if (mm->exe_file)
 664                 fput(mm->exe_file);
 665         mm->exe_file = new_exe_file;
 666 }
 667
 668 struct file *get_mm_exe_file(struct mm_struct *mm)
 669 {
 670         struct file *exe_file;
 671
 672         /* We need mmap_sem to protect against races with removal of exe_file */
 673         down_read(&mm->mmap_sem);
 674         exe_file = mm->exe_file;
 675         if (exe_file)
 676                 get_file(exe_file);
 677         up_read(&mm->mmap_sem);
 678         return exe_file;
 679 }
 680
 681 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 682 {
 683         /* It's safe to write the exe_file pointer without exe_file_lock because
 684          * this is called during fork when the task is not yet in /proc */
 685         newmm->exe_file = get_mm_exe_file(oldmm);
 686 }
 687
 688 /**
 689  * get_task_mm - acquire a reference to the task's mm
 690  *
 691  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 692  * this kernel workthread has transiently adopted a user mm with use_mm,
 693  * to do its AIO) is not set and if so returns a reference to it, after
 694  * bumping up the use count.  User must release the mm via mmput()
 695  * after use.  Typically used by /proc and ptrace.
 696  */
 697 struct mm_struct *get_task_mm(struct task_struct *task)
 698 {
 699         struct mm_struct *mm;
 700
 701         task_lock(task);
 702         mm = task->mm;
 703         if (mm) {
 704                 if (task->flags & PF_KTHREAD)
 705                         mm = NULL;
 706                 else
 707                         atomic_inc(&mm->mm_users);
 708         }
 709         task_unlock(task);
 710         return mm;
 711 }
 712 EXPORT_SYMBOL_GPL(get_task_mm);
 713
 714 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 715 {
 716         struct mm_struct *mm;
 717         int err;
 718
 719         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 720         if (err)
 721                 return ERR_PTR(err);
 722
 723         mm = get_task_mm(task);
 724         if (mm && mm != current->mm &&
 725                         !ptrace_may_access(task, mode) &&
 726                         !capable(CAP_SYS_RESOURCE)) {
 727                 mmput(mm);
 728                 mm = ERR_PTR(-EACCES);
 729         }
 730         mutex_unlock(&task->signal->cred_guard_mutex);
 731
 732         return mm;
 733 }
 734
 735 static void complete_vfork_done(struct task_struct *tsk)
 736 {
 737         struct completion *vfork;
 738
 739         task_lock(tsk);
 740         vfork = tsk->vfork_done;
 741         if (likely(vfork)) {
 742                 tsk->vfork_done = NULL;
 743                 complete(vfork);
 744         }
 745         task_unlock(tsk);
 746 }
 747
 748 static int wait_for_vfork_done(struct task_struct *child,
 749                                 struct completion *vfork)
 750 {
 751         int killed;
 752
 753         freezer_do_not_count();
 754         killed = wait_for_completion_killable(vfork);
 755         freezer_count();
 756
 757         if (killed) {
 758                 task_lock(child);
 759                 child->vfork_done = NULL;
 760                 task_unlock(child);
 761         }
 762
 763         put_task_struct(child);
 764         return killed;
 765 }
 766
 767 /* Please note the differences between mmput and mm_release.
 768  * mmput is called whenever we stop holding onto a mm_struct,
 769  * error success whatever.
 770  *
 771  * mm_release is called after a mm_struct has been removed
 772  * from the current process.
 773  *
 774  * This difference is important for error handling, when we
 775  * only half set up a mm_struct for a new process and need to restore
 776  * the old one.  Because we mmput the new mm_struct before
 777  * restoring the old one. . .
 778  * Eric Biederman 10 January 1998
 779  */
 780 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 781 {
 782         /* Get rid of any futexes when releasing the mm */
 783 #ifdef CONFIG_FUTEX
 784         if (unlikely(tsk->robust_list)) {
 785                 exit_robust_list(tsk);
 786                 tsk->robust_list = NULL;
 787         }
 788 #ifdef CONFIG_COMPAT
 789         if (unlikely(tsk->compat_robust_list)) {
 790                 compat_exit_robust_list(tsk);
 791                 tsk->compat_robust_list = NULL;
 792         }
 793 #endif
 794         if (unlikely(!list_empty(&tsk->pi_state_list)))
 795                 exit_pi_state_list(tsk);
 796 #endif
 797
 798         uprobe_free_utask(tsk);
 799
 800         /* Get rid of any cached register state */
 801         deactivate_mm(tsk, mm);
 802
 803         /*
 804          * If we're exiting normally, clear a user-space tid field if
 805          * requested.  We leave this alone when dying by signal, to leave
 806          * the value intact in a core dump, and to save the unnecessary
 807          * trouble, say, a killed vfork parent shouldn't touch this mm.
 808          * Userland only wants this done for a sys_exit.
 809          */
 810         if (tsk->clear_child_tid) {
 811                 if (!(tsk->flags & PF_SIGNALED) &&
 812                     atomic_read(&mm->mm_users) > 1) {
 813                         /*
 814                          * We don't check the error code - if userspace has
 815                          * not set up a proper pointer then tough luck.
 816                          */
 817                         put_user(0, tsk->clear_child_tid);
 818                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 819                                         1, NULL, NULL, 0);
 820                 }
 821                 tsk->clear_child_tid = NULL;
 822         }
 823
 824         /*
 825          * All done, finally we can wake up parent and return this mm to him.
 826          * Also kthread_stop() uses this completion for synchronization.
 827          */
 828         if (tsk->vfork_done)
 829                 complete_vfork_done(tsk);
 830 }
 831
 832 /*
 833  * Allocate a new mm structure and copy contents from the
 834  * mm structure of the passed in task structure.
 835  */
 836 struct mm_struct *dup_mm(struct task_struct *tsk)
 837 {
 838         struct mm_struct *mm, *oldmm = current->mm;
 839         int err;
 840
 841         if (!oldmm)
 842                 return NULL;
 843
 844         mm = allocate_mm();
 845         if (!mm)
 846                 goto fail_nomem;
 847
 848         memcpy(mm, oldmm, sizeof(*mm));
 849         mm_init_cpumask(mm);
 850
 851 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 852         mm->pmd_huge_pte = NULL;
 853 #endif
 854 #ifdef CONFIG_NUMA_BALANCING
 855         mm->first_nid = NUMA_PTE_SCAN_INIT;
 856 #endif
 857         if (!mm_init(mm, tsk))
 858                 goto fail_nomem;
 859
 860         if (init_new_context(tsk, mm))
 861                 goto fail_nocontext;
 862
 863         dup_mm_exe_file(oldmm, mm);
 864
 865         err = dup_mmap(mm, oldmm);
 866         if (err)
 867                 goto free_pt;
 868
 869         mm->hiwater_rss = get_mm_rss(mm);
 870         mm->hiwater_vm = mm->total_vm;
 871
 872         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 873                 goto free_pt;
 874
 875         return mm;
 876
 877 free_pt:
 878         /* don't put binfmt in mmput, we haven't got module yet */
 879         mm->binfmt = NULL;
 880         mmput(mm);
 881
 882 fail_nomem:
 883         return NULL;
 884
 885 fail_nocontext:
 886         /*
 887          * If init_new_context() failed, we cannot use mmput() to free the mm
 888          * because it calls destroy_context()
 889          */
 890         mm_free_pgd(mm);
 891         free_mm(mm);
 892         return NULL;
 893 }
 894
 895 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 896 {
 897         struct mm_struct *mm, *oldmm;
 898         int retval;
 899
 900         tsk->min_flt = tsk->maj_flt = 0;
 901         tsk->nvcsw = tsk->nivcsw = 0;
 902 #ifdef CONFIG_DETECT_HUNG_TASK
 903         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 904 #endif
 905
 906         tsk->mm = NULL;
 907         tsk->active_mm = NULL;
 908
 909         /*
 910          * Are we cloning a kernel thread?
 911          *
 912          * We need to steal a active VM for that..
 913          */
 914         oldmm = current->mm;
 915         if (!oldmm)
 916                 return 0;
 917
 918         if (clone_flags & CLONE_VM) {
 919                 atomic_inc(&oldmm->mm_users);
 920                 mm = oldmm;
 921                 goto good_mm;
 922         }
 923
 924         retval = -ENOMEM;
 925         mm = dup_mm(tsk);
 926         if (!mm)
 927                 goto fail_nomem;
 928
 929 good_mm:
 930         tsk->mm = mm;
 931         tsk->active_mm = mm;
 932         return 0;
 933
 934 fail_nomem:
 935         return retval;
 936 }
 937
 938 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 939 {
 940         struct fs_struct *fs = current->fs;
 941         if (clone_flags & CLONE_FS) {
 942                 /* tsk->fs is already what we want */
 943                 spin_lock(&fs->lock);
 944                 if (fs->in_exec) {
 945                         spin_unlock(&fs->lock);
 946                         return -EAGAIN;
 947                 }
 948                 fs->users++;
 949                 spin_unlock(&fs->lock);
 950                 return 0;
 951         }
 952         tsk->fs = copy_fs_struct(fs);
 953         if (!tsk->fs)
 954                 return -ENOMEM;
 955         return 0;
 956 }
 957
 958 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 959 {
 960         struct files_struct *oldf, *newf;
 961         int error = 0;
 962
 963         /*
 964          * A background process may not have any files ...
 965          */
 966         oldf = current->files;
 967         if (!oldf)
 968                 goto out;
 969
 970         if (clone_flags & CLONE_FILES) {
 971                 atomic_inc(&oldf->count);
 972                 goto out;
 973         }
 974
 975         newf = dup_fd(oldf, &error);
 976         if (!newf)
 977                 goto out;
 978
 979         tsk->files = newf;
 980         error = 0;
 981 out:
 982         return error;
 983 }
 984
 985 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 986 {
 987 #ifdef CONFIG_BLOCK
 988         struct io_context *ioc = current->io_context;
 989         struct io_context *new_ioc;
 990
 991         if (!ioc)
 992                 return 0;
 993         /*
 994          * Share io context with parent, if CLONE_IO is set
 995          */
 996         if (clone_flags & CLONE_IO) {
 997                 ioc_task_link(ioc);
 998                 tsk->io_context = ioc;
 999         } else if (ioprio_valid(ioc->ioprio)) {
1000                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1001                 if (unlikely(!new_ioc))
1002                         return -ENOMEM;
1003
1004                 new_ioc->ioprio = ioc->ioprio;
1005                 put_io_context(new_ioc);
1006         }
1007 #endif
1008         return 0;
1009 }
1010
1011 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1012 {
1013         struct sighand_struct *sig;
1014
1015         if (clone_flags & CLONE_SIGHAND) {
1016                 atomic_inc(&current->sighand->count);
1017                 return 0;
1018         }
1019         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1020         rcu_assign_pointer(tsk->sighand, sig);
1021         if (!sig)
1022                 return -ENOMEM;
1023         atomic_set(&sig->count, 1);
1024         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1025         return 0;
1026 }
1027
1028 void __cleanup_sighand(struct sighand_struct *sighand)
1029 {
1030         if (atomic_dec_and_test(&sighand->count)) {
1031                 signalfd_cleanup(sighand);
1032                 kmem_cache_free(sighand_cachep, sighand);
1033         }
1034 }
1035
1036
1037 /*
1038  * Initialize POSIX timer handling for a thread group.
1039  */
1040 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1041 {
1042         unsigned long cpu_limit;
1043
1044         /* Thread group counters. */
1045         thread_group_cputime_init(sig);
1046
1047         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1048         if (cpu_limit != RLIM_INFINITY) {
1049                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1050                 sig->cputimer.running = 1;
1051         }
1052
1053         /* The timer lists. */
1054         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1055         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1056         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1057 }
1058
1059 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1060 {
1061         struct signal_struct *sig;
1062
1063         if (clone_flags & CLONE_THREAD)
1064                 return 0;
1065
1066         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1067         tsk->signal = sig;
1068         if (!sig)
1069                 return -ENOMEM;
1070
1071         sig->nr_threads = 1;
1072         atomic_set(&sig->live, 1);
1073         atomic_set(&sig->sigcnt, 1);
1074
1075         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1076         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1077         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1078
1079         init_waitqueue_head(&sig->wait_chldexit);
1080         sig->curr_target = tsk;
1081         init_sigpending(&sig->shared_pending);
1082         INIT_LIST_HEAD(&sig->posix_timers);
1083
1084         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1085         sig->real_timer.function = it_real_fn;
1086
1087         task_lock(current->group_leader);
1088         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1089         task_unlock(current->group_leader);
1090
1091         posix_cpu_timers_init_group(sig);
1092
1093         tty_audit_fork(sig);
1094         sched_autogroup_fork(sig);
1095
1096 #ifdef CONFIG_CGROUPS
1097         init_rwsem(&sig->group_rwsem);
1098 #endif
1099
1100         sig->oom_score_adj = current->signal->oom_score_adj;
1101         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1102
1103         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1104                                    current->signal->is_child_subreaper;
1105
1106         mutex_init(&sig->cred_guard_mutex);
1107
1108         return 0;
1109 }
1110
1111 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1112 {
1113         unsigned long new_flags = p->flags;
1114
1115         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1116         new_flags |= PF_FORKNOEXEC;
1117         p->flags = new_flags;
1118 }
1119
1120 static void copy_seccomp(struct task_struct *p)
1121 {
1122 #ifdef CONFIG_SECCOMP
1123         /*
1124          * Must be called with sighand->lock held, which is common to
1125          * all threads in the group. Holding cred_guard_mutex is not
1126          * needed because this new task is not yet running and cannot
1127          * be racing exec.
1128          */
1129         assert_spin_locked(&current->sighand->siglock);
1130
1131         /* Ref-count the new filter user, and assign it. */
1132         get_seccomp_filter(current);
1133         p->seccomp = current->seccomp;
1134
1135         /*
1136          * Explicitly enable no_new_privs here in case it got set
1137          * between the task_struct being duplicated and holding the
1138          * sighand lock. The seccomp state and nnp must be in sync.
1139          */
1140         if (task_no_new_privs(current))
1141                 task_set_no_new_privs(p);
1142
1143         /*
1144          * If the parent gained a seccomp mode after copying thread
1145          * flags and between before we held the sighand lock, we have
1146          * to manually enable the seccomp thread flag here.
1147          */
1148         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1149                 set_tsk_thread_flag(p, TIF_SECCOMP);
1150 #endif
1151 }
1152
1153 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1154 {
1155         current->clear_child_tid = tidptr;
1156
1157         return task_pid_vnr(current);
1158 }
1159
1160 static void rt_mutex_init_task(struct task_struct *p)
1161 {
1162         raw_spin_lock_init(&p->pi_lock);
1163 #ifdef CONFIG_RT_MUTEXES
1164         plist_head_init(&p->pi_waiters);
1165         p->pi_blocked_on = NULL;
1166 #endif
1167 }
1168
1169 #ifdef CONFIG_MM_OWNER
1170 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1171 {
1172         mm->owner = p;
1173 }
1174 #endif /* CONFIG_MM_OWNER */
1175
1176 /*
1177  * Initialize POSIX timer handling for a single task.
1178  */
1179 static void posix_cpu_timers_init(struct task_struct *tsk)
1180 {
1181         tsk->cputime_expires.prof_exp = 0;
1182         tsk->cputime_expires.virt_exp = 0;
1183         tsk->cputime_expires.sched_exp = 0;
1184         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1185         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1186         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1187 }
1188
1189 /*
1190  * This creates a new process as a copy of the old one,
1191  * but does not actually start it yet.
1192  *
1193  * It copies the registers, and all the appropriate
1194  * parts of the process environment (as per the clone
1195  * flags). The actual kick-off is left to the caller.
1196  */
1197 static struct task_struct *copy_process(unsigned long clone_flags,
1198                                         unsigned long stack_start,
1199                                         unsigned long stack_size,
1200                                         int __user *child_tidptr,
1201                                         struct pid *pid,
1202                                         int trace)
1203 {
1204         int retval;
1205         struct task_struct *p;
1206
1207         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1208                 return ERR_PTR(-EINVAL);
1209
1210         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1211                 return ERR_PTR(-EINVAL);
1212
1213         /*
1214          * Thread groups must share signals as well, and detached threads
1215          * can only be started up within the thread group.
1216          */
1217         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1218                 return ERR_PTR(-EINVAL);
1219
1220         /*
1221          * Shared signal handlers imply shared VM. By way of the above,
1222          * thread groups also imply shared VM. Blocking this case allows
1223          * for various simplifications in other code.
1224          */
1225         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1226                 return ERR_PTR(-EINVAL);
1227
1228         /*
1229          * Siblings of global init remain as zombies on exit since they are
1230          * not reaped by their parent (swapper). To solve this and to avoid
1231          * multi-rooted process trees, prevent global and container-inits
1232          * from creating siblings.
1233          */
1234         if ((clone_flags & CLONE_PARENT) &&
1235                                 current->signal->flags & SIGNAL_UNKILLABLE)
1236                 return ERR_PTR(-EINVAL);
1237
1238         /*
1239          * If the new process will be in a different pid namespace don't
1240          * allow it to share a thread group or signal handlers with the
1241          * forking task.
1242          */
1243         if ((clone_flags & (CLONE_SIGHAND | CLONE_NEWPID)) &&
1244             (task_active_pid_ns(current) != current->nsproxy->pid_ns))
1245                 return ERR_PTR(-EINVAL);
1246
1247         retval = security_task_create(clone_flags);
1248         if (retval)
1249                 goto fork_out;
1250
1251         retval = -ENOMEM;
1252         p = dup_task_struct(current);
1253         if (!p)
1254                 goto fork_out;
1255
1256         ftrace_graph_init_task(p);
1257
1258         rt_mutex_init_task(p);
1259
1260 #ifdef CONFIG_PROVE_LOCKING
1261         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1262         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1263 #endif
1264         retval = -EAGAIN;
1265         if (atomic_read(&p->real_cred->user->processes) >=
1266                         task_rlimit(p, RLIMIT_NPROC)) {
1267                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1268                     p->real_cred->user != INIT_USER)
1269                         goto bad_fork_free;
1270         }
1271         current->flags &= ~PF_NPROC_EXCEEDED;
1272
1273         retval = copy_creds(p, clone_flags);
1274         if (retval < 0)
1275                 goto bad_fork_free;
1276
1277         /*
1278          * If multiple threads are within copy_process(), then this check
1279          * triggers too late. This doesn't hurt, the check is only there
1280          * to stop root fork bombs.
1281          */
1282         retval = -EAGAIN;
1283         if (nr_threads >= max_threads)
1284                 goto bad_fork_cleanup_count;
1285
1286         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1287                 goto bad_fork_cleanup_count;
1288
1289         p->did_exec = 0;
1290         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1291         copy_flags(clone_flags, p);
1292         INIT_LIST_HEAD(&p->children);
1293         INIT_LIST_HEAD(&p->sibling);
1294         rcu_copy_process(p);
1295         p->vfork_done = NULL;
1296         spin_lock_init(&p->alloc_lock);
1297
1298         init_sigpending(&p->pending);
1299
1300         p->utime = p->stime = p->gtime = 0;
1301         p->utimescaled = p->stimescaled = 0;
1302         p->cpu_power = 0;
1303 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1304         p->prev_cputime.utime = p->prev_cputime.stime = 0;
1305 #endif
1306 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1307         seqlock_init(&p->vtime_seqlock);
1308         p->vtime_snap = 0;
1309         p->vtime_snap_whence = VTIME_SLEEPING;
1310 #endif
1311
1312 #if defined(SPLIT_RSS_COUNTING)
1313         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1314 #endif
1315
1316         p->default_timer_slack_ns = current->timer_slack_ns;
1317
1318         task_io_accounting_init(&p->ioac);
1319         acct_clear_integrals(p);
1320
1321         posix_cpu_timers_init(p);
1322
1323         do_posix_clock_monotonic_gettime(&p->start_time);
1324         p->real_start_time = p->start_time;
1325         monotonic_to_bootbased(&p->real_start_time);
1326         p->io_context = NULL;
1327         p->audit_context = NULL;
1328         if (clone_flags & CLONE_THREAD)
1329                 threadgroup_change_begin(current);
1330         cgroup_fork(p);
1331 #ifdef CONFIG_NUMA
1332         p->mempolicy = mpol_dup(p->mempolicy);
1333         if (IS_ERR(p->mempolicy)) {
1334                 retval = PTR_ERR(p->mempolicy);
1335                 p->mempolicy = NULL;
1336                 goto bad_fork_cleanup_cgroup;
1337         }
1338         mpol_fix_fork_child_flag(p);
1339 #endif
1340 #ifdef CONFIG_CPUSETS
1341         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1342         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1343         seqcount_init(&p->mems_allowed_seq);
1344 #endif
1345 #ifdef CONFIG_TRACE_IRQFLAGS
1346         p->irq_events = 0;
1347         p->hardirqs_enabled = 0;
1348         p->hardirq_enable_ip = 0;
1349         p->hardirq_enable_event = 0;
1350         p->hardirq_disable_ip = _THIS_IP_;
1351         p->hardirq_disable_event = 0;
1352         p->softirqs_enabled = 1;
1353         p->softirq_enable_ip = _THIS_IP_;
1354         p->softirq_enable_event = 0;
1355         p->softirq_disable_ip = 0;
1356         p->softirq_disable_event = 0;
1357         p->hardirq_context = 0;
1358         p->softirq_context = 0;
1359 #endif
1360 #ifdef CONFIG_LOCKDEP
1361         p->lockdep_depth = 0; /* no locks held yet */
1362         p->curr_chain_key = 0;
1363         p->lockdep_recursion = 0;
1364 #endif
1365
1366 #ifdef CONFIG_DEBUG_MUTEXES
1367         p->blocked_on = NULL; /* not blocked yet */
1368 #endif
1369 #ifdef CONFIG_MEMCG
1370         p->memcg_batch.do_batch = 0;
1371         p->memcg_batch.memcg = NULL;
1372 #endif
1373 #ifdef CONFIG_BCACHE
1374         p->sequential_io        = 0;
1375         p->sequential_io_avg    = 0;
1376 #endif
1377
1378         /* Perform scheduler related setup. Assign this task to a CPU. */
1379         sched_fork(p);
1380
1381         retval = perf_event_init_task(p);
1382         if (retval)
1383                 goto bad_fork_cleanup_policy;
1384         retval = audit_alloc(p);
1385         if (retval)
1386                 goto bad_fork_cleanup_perf;
1387         /* copy all the process information */
1388         retval = copy_semundo(clone_flags, p);
1389         if (retval)
1390                 goto bad_fork_cleanup_audit;
1391         retval = copy_files(clone_flags, p);
1392         if (retval)
1393                 goto bad_fork_cleanup_semundo;
1394         retval = copy_fs(clone_flags, p);
1395         if (retval)
1396                 goto bad_fork_cleanup_files;
1397         retval = copy_sighand(clone_flags, p);
1398         if (retval)
1399                 goto bad_fork_cleanup_fs;
1400         retval = copy_signal(clone_flags, p);
1401         if (retval)
1402                 goto bad_fork_cleanup_sighand;
1403         retval = copy_mm(clone_flags, p);
1404         if (retval)
1405                 goto bad_fork_cleanup_signal;
1406         retval = copy_namespaces(clone_flags, p);
1407         if (retval)
1408                 goto bad_fork_cleanup_mm;
1409         retval = copy_io(clone_flags, p);
1410         if (retval)
1411                 goto bad_fork_cleanup_namespaces;
1412         retval = copy_thread(clone_flags, stack_start, stack_size, p);
1413         if (retval)
1414                 goto bad_fork_cleanup_io;
1415
1416         if (pid != &init_struct_pid) {
1417                 retval = -ENOMEM;
1418                 pid = alloc_pid(p->nsproxy->pid_ns);
1419                 if (!pid)
1420                         goto bad_fork_cleanup_io;
1421         }
1422
1423         p->pid = pid_nr(pid);
1424         p->tgid = p->pid;
1425         if (clone_flags & CLONE_THREAD)
1426                 p->tgid = current->tgid;
1427
1428         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1429         /*
1430          * Clear TID on mm_release()?
1431          */
1432         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1433 #ifdef CONFIG_BLOCK
1434         p->plug = NULL;
1435 #endif
1436 #ifdef CONFIG_FUTEX
1437         p->robust_list = NULL;
1438 #ifdef CONFIG_COMPAT
1439         p->compat_robust_list = NULL;
1440 #endif
1441         INIT_LIST_HEAD(&p->pi_state_list);
1442         p->pi_state_cache = NULL;
1443 #endif
1444         uprobe_copy_process(p);
1445         /*
1446          * sigaltstack should be cleared when sharing the same VM
1447          */
1448         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1449                 p->sas_ss_sp = p->sas_ss_size = 0;
1450
1451         /*
1452          * Syscall tracing and stepping should be turned off in the
1453          * child regardless of CLONE_PTRACE.
1454          */
1455         user_disable_single_step(p);
1456         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1457 #ifdef TIF_SYSCALL_EMU
1458         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1459 #endif
1460         clear_all_latency_tracing(p);
1461
1462         /* ok, now we should be set up.. */
1463         if (clone_flags & CLONE_THREAD)
1464                 p->exit_signal = -1;
1465         else if (clone_flags & CLONE_PARENT)
1466                 p->exit_signal = current->group_leader->exit_signal;
1467         else
1468                 p->exit_signal = (clone_flags & CSIGNAL);
1469
1470         p->pdeath_signal = 0;
1471         p->exit_state = 0;
1472
1473         p->nr_dirtied = 0;
1474         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1475         p->dirty_paused_when = 0;
1476
1477         /*
1478          * Ok, make it visible to the rest of the system.
1479          * We dont wake it up yet.
1480          */
1481         p->group_leader = p;
1482         INIT_LIST_HEAD(&p->thread_group);
1483         p->task_works = NULL;
1484
1485         /* Need tasklist lock for parent etc handling! */
1486         write_lock_irq(&tasklist_lock);
1487
1488         /* CLONE_PARENT re-uses the old parent */
1489         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1490                 p->real_parent = current->real_parent;
1491                 p->parent_exec_id = current->parent_exec_id;
1492         } else {
1493                 p->real_parent = current;
1494                 p->parent_exec_id = current->self_exec_id;
1495         }
1496
1497         spin_lock(&current->sighand->siglock);
1498
1499         /*
1500          * Copy seccomp details explicitly here, in case they were changed
1501          * before holding sighand lock.
1502          */
1503         copy_seccomp(p);
1504
1505         /*
1506          * Process group and session signals need to be delivered to just the
1507          * parent before the fork or both the parent and the child after the
1508          * fork. Restart if a signal comes in before we add the new process to
1509          * it's process group.
1510          * A fatal signal pending means that current will exit, so the new
1511          * thread can't slip out of an OOM kill (or normal SIGKILL).
1512         */
1513         recalc_sigpending();
1514         if (signal_pending(current)) {
1515                 spin_unlock(&current->sighand->siglock);
1516                 write_unlock_irq(&tasklist_lock);
1517                 retval = -ERESTARTNOINTR;
1518                 goto bad_fork_free_pid;
1519         }
1520
1521         if (likely(p->pid)) {
1522                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1523
1524                 if (thread_group_leader(p)) {
1525                         if (is_child_reaper(pid)) {
1526                                 ns_of_pid(pid)->child_reaper = p;
1527                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1528                         }
1529
1530                         p->signal->leader_pid = pid;
1531                         p->signal->tty = tty_kref_get(current->signal->tty);
1532                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1533                         attach_pid(p, PIDTYPE_SID, task_session(current));
1534                         list_add_tail(&p->sibling, &p->real_parent->children);
1535                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1536                         __this_cpu_inc(process_counts);
1537                 } else {
1538                         current->signal->nr_threads++;
1539                         atomic_inc(&current->signal->live);
1540                         atomic_inc(&current->signal->sigcnt);
1541                         p->group_leader = current->group_leader;
1542                         list_add_tail_rcu(&p->thread_group,
1543                                           &p->group_leader->thread_group);
1544                         list_add_tail_rcu(&p->thread_node,
1545                                           &p->signal->thread_head);
1546                 }
1547                 attach_pid(p, PIDTYPE_PID, pid);
1548                 nr_threads++;
1549         }
1550
1551         total_forks++;
1552         spin_unlock(&current->sighand->siglock);
1553         syscall_tracepoint_update(p);
1554         write_unlock_irq(&tasklist_lock);
1555
1556         proc_fork_connector(p);
1557         cgroup_post_fork(p);
1558         if (clone_flags & CLONE_THREAD)
1559                 threadgroup_change_end(current);
1560         perf_event_fork(p);
1561
1562         trace_task_newtask(p, clone_flags);
1563
1564         return p;
1565
1566 bad_fork_free_pid:
1567         if (pid != &init_struct_pid)
1568                 free_pid(pid);
1569 bad_fork_cleanup_io:
1570         if (p->io_context)
1571                 exit_io_context(p);
1572 bad_fork_cleanup_namespaces:
1573         exit_task_namespaces(p);
1574 bad_fork_cleanup_mm:
1575         if (p->mm)
1576                 mmput(p->mm);
1577 bad_fork_cleanup_signal:
1578         if (!(clone_flags & CLONE_THREAD))
1579                 free_signal_struct(p->signal);
1580 bad_fork_cleanup_sighand:
1581         __cleanup_sighand(p->sighand);
1582 bad_fork_cleanup_fs:
1583         exit_fs(p); /* blocking */
1584 bad_fork_cleanup_files:
1585         exit_files(p); /* blocking */
1586 bad_fork_cleanup_semundo:
1587         exit_sem(p);
1588 bad_fork_cleanup_audit:
1589         audit_free(p);
1590 bad_fork_cleanup_perf:
1591         perf_event_free_task(p);
1592 bad_fork_cleanup_policy:
1593 #ifdef CONFIG_NUMA
1594         mpol_put(p->mempolicy);
1595 bad_fork_cleanup_cgroup:
1596 #endif
1597         if (clone_flags & CLONE_THREAD)
1598                 threadgroup_change_end(current);
1599         cgroup_exit(p, 0);
1600         delayacct_tsk_free(p);
1601         module_put(task_thread_info(p)->exec_domain->module);
1602 bad_fork_cleanup_count:
1603         atomic_dec(&p->cred->user->processes);
1604         exit_creds(p);
1605 bad_fork_free:
1606         free_task(p);
1607 fork_out:
1608         return ERR_PTR(retval);
1609 }
1610
1611 static inline void init_idle_pids(struct pid_link *links)
1612 {
1613         enum pid_type type;
1614
1615         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1616                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1617                 links[type].pid = &init_struct_pid;
1618         }
1619 }
1620
1621 struct task_struct * __cpuinit fork_idle(int cpu)
1622 {
1623         struct task_struct *task;
1624         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1625         if (!IS_ERR(task)) {
1626                 init_idle_pids(task->pids);
1627                 init_idle(task, cpu);
1628         }
1629
1630         return task;
1631 }
1632
1633 /*
1634  *  Ok, this is the main fork-routine.
1635  *
1636  * It copies the process, and if successful kick-starts
1637  * it and waits for it to finish using the VM if required.
1638  */
1639 long do_fork(unsigned long clone_flags,
1640               unsigned long stack_start,
1641               unsigned long stack_size,
1642               int __user *parent_tidptr,
1643               int __user *child_tidptr)
1644 {
1645         struct task_struct *p;
1646         int trace = 0;
1647         long nr;
1648
1649         /*
1650          * Do some preliminary argument and permissions checking before we
1651          * actually start allocating stuff
1652          */
1653         if (clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) {
1654                 if (clone_flags & (CLONE_THREAD|CLONE_PARENT))
1655                         return -EINVAL;
1656         }
1657
1658         /*
1659          * Determine whether and which event to report to ptracer.  When
1660          * called from kernel_thread or CLONE_UNTRACED is explicitly
1661          * requested, no event is reported; otherwise, report if the event
1662          * for the type of forking is enabled.
1663          */
1664         if (!(clone_flags & CLONE_UNTRACED)) {
1665                 if (clone_flags & CLONE_VFORK)
1666                         trace = PTRACE_EVENT_VFORK;
1667                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1668                         trace = PTRACE_EVENT_CLONE;
1669                 else
1670                         trace = PTRACE_EVENT_FORK;
1671
1672                 if (likely(!ptrace_event_enabled(current, trace)))
1673                         trace = 0;
1674         }
1675
1676         p = copy_process(clone_flags, stack_start, stack_size,
1677                          child_tidptr, NULL, trace);
1678         /*
1679          * Do this prior waking up the new thread - the thread pointer
1680          * might get invalid after that point, if the thread exits quickly.
1681          */
1682         if (!IS_ERR(p)) {
1683                 struct completion vfork;
1684                 struct pid *pid;
1685
1686                 trace_sched_process_fork(current, p);
1687
1688                 pid = get_task_pid(p, PIDTYPE_PID);
1689                 nr = pid_vnr(pid);
1690
1691                 if (clone_flags & CLONE_PARENT_SETTID)
1692                         put_user(nr, parent_tidptr);
1693
1694                 if (clone_flags & CLONE_VFORK) {
1695                         p->vfork_done = &vfork;
1696                         init_completion(&vfork);
1697                         get_task_struct(p);
1698                 }
1699
1700                 wake_up_new_task(p);
1701
1702                 /* forking complete and child started to run, tell ptracer */
1703                 if (unlikely(trace))
1704                         ptrace_event_pid(trace, pid);
1705
1706                 if (clone_flags & CLONE_VFORK) {
1707                         if (!wait_for_vfork_done(p, &vfork))
1708                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1709                 }
1710
1711                 put_pid(pid);
1712         } else {
1713                 nr = PTR_ERR(p);
1714         }
1715         return nr;
1716 }
1717
1718 /*
1719  * Create a kernel thread.
1720  */
1721 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1722 {
1723         return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1724                 (unsigned long)arg, NULL, NULL);
1725 }
1726
1727 #ifdef __ARCH_WANT_SYS_FORK
1728 SYSCALL_DEFINE0(fork)
1729 {
1730 #ifdef CONFIG_MMU
1731         return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1732 #else
1733         /* can not support in nommu mode */
1734         return(-EINVAL);
1735 #endif
1736 }
1737 #endif
1738
1739 #ifdef __ARCH_WANT_SYS_VFORK
1740 SYSCALL_DEFINE0(vfork)
1741 {
1742         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1743                         0, NULL, NULL);
1744 }
1745 #endif
1746
1747 #ifdef __ARCH_WANT_SYS_CLONE
1748 #ifdef CONFIG_CLONE_BACKWARDS
1749 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1750                  int __user *, parent_tidptr,
1751                  int, tls_val,
1752                  int __user *, child_tidptr)
1753 #elif defined(CONFIG_CLONE_BACKWARDS2)
1754 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1755                  int __user *, parent_tidptr,
1756                  int __user *, child_tidptr,
1757                  int, tls_val)
1758 #elif defined(CONFIG_CLONE_BACKWARDS3)
1759 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1760                 int, stack_size,
1761                 int __user *, parent_tidptr,
1762                 int __user *, child_tidptr,
1763                 int, tls_val)
1764 #else
1765 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1766                  int __user *, parent_tidptr,
1767                  int __user *, child_tidptr,
1768                  int, tls_val)
1769 #endif
1770 {
1771         return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1772 }
1773 #endif
1774
1775 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1776 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1777 #endif
1778
1779 static void sighand_ctor(void *data)
1780 {
1781         struct sighand_struct *sighand = data;
1782
1783         spin_lock_init(&sighand->siglock);
1784         init_waitqueue_head(&sighand->signalfd_wqh);
1785 }
1786
1787 void __init proc_caches_init(void)
1788 {
1789         sighand_cachep = kmem_cache_create("sighand_cache",
1790                         sizeof(struct sighand_struct), 0,
1791                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1792                         SLAB_NOTRACK, sighand_ctor);
1793         signal_cachep = kmem_cache_create("signal_cache",
1794                         sizeof(struct signal_struct), 0,
1795                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1796         files_cachep = kmem_cache_create("files_cache",
1797                         sizeof(struct files_struct), 0,
1798                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1799         fs_cachep = kmem_cache_create("fs_cache",
1800                         sizeof(struct fs_struct), 0,
1801                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1802         /*
1803          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1804          * whole struct cpumask for the OFFSTACK case. We could change
1805          * this to *only* allocate as much of it as required by the
1806          * maximum number of CPU's we can ever have.  The cpumask_allocation
1807          * is at the end of the structure, exactly for that reason.
1808          */
1809         mm_cachep = kmem_cache_create("mm_struct",
1810                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1811                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1812         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1813         mmap_init();
1814         nsproxy_cache_init();
1815 }
1816
1817 /*
1818  * Check constraints on flags passed to the unshare system call.
1819  */
1820 static int check_unshare_flags(unsigned long unshare_flags)
1821 {
1822         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1823                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1824                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1825                                 CLONE_NEWUSER|CLONE_NEWPID))
1826                 return -EINVAL;
1827         /*
1828          * Not implemented, but pretend it works if there is nothing
1829          * to unshare.  Note that unsharing the address space or the
1830          * signal handlers also need to unshare the signal queues (aka
1831          * CLONE_THREAD).
1832          */
1833         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1834                 if (!thread_group_empty(current))
1835                         return -EINVAL;
1836         }
1837         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
1838                 if (atomic_read(&current->sighand->count) > 1)
1839                         return -EINVAL;
1840         }
1841         if (unshare_flags & CLONE_VM) {
1842                 if (!current_is_single_threaded())
1843                         return -EINVAL;
1844         }
1845
1846         return 0;
1847 }
1848
1849 /*
1850  * Unshare the filesystem structure if it is being shared
1851  */
1852 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1853 {
1854         struct fs_struct *fs = current->fs;
1855
1856         if (!(unshare_flags & CLONE_FS) || !fs)
1857                 return 0;
1858
1859         /* don't need lock here; in the worst case we'll do useless copy */
1860         if (fs->users == 1)
1861                 return 0;
1862
1863         *new_fsp = copy_fs_struct(fs);
1864         if (!*new_fsp)
1865                 return -ENOMEM;
1866
1867         return 0;
1868 }
1869
1870 /*
1871  * Unshare file descriptor table if it is being shared
1872  */
1873 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1874 {
1875         struct files_struct *fd = current->files;
1876         int error = 0;
1877
1878         if ((unshare_flags & CLONE_FILES) &&
1879             (fd && atomic_read(&fd->count) > 1)) {
1880                 *new_fdp = dup_fd(fd, &error);
1881                 if (!*new_fdp)
1882                         return error;
1883         }
1884
1885         return 0;
1886 }
1887
1888 /*
1889  * unshare allows a process to 'unshare' part of the process
1890  * context which was originally shared using clone.  copy_*
1891  * functions used by do_fork() cannot be used here directly
1892  * because they modify an inactive task_struct that is being
1893  * constructed. Here we are modifying the current, active,
1894  * task_struct.
1895  */
1896 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1897 {
1898         struct fs_struct *fs, *new_fs = NULL;
1899         struct files_struct *fd, *new_fd = NULL;
1900         struct cred *new_cred = NULL;
1901         struct nsproxy *new_nsproxy = NULL;
1902         int do_sysvsem = 0;
1903         int err;
1904
1905         /*
1906          * If unsharing a user namespace must also unshare the thread.
1907          */
1908         if (unshare_flags & CLONE_NEWUSER)
1909                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1910         /*
1911          * If unsharing a pid namespace must also unshare the thread.
1912          */
1913         if (unshare_flags & CLONE_NEWPID)
1914                 unshare_flags |= CLONE_THREAD;
1915         /*
1916          * If unsharing vm, must also unshare signal handlers.
1917          */
1918         if (unshare_flags & CLONE_VM)
1919                 unshare_flags |= CLONE_SIGHAND;
1920         /*
1921          * If unsharing a signal handlers, must also unshare the signal queues.
1922          */
1923         if (unshare_flags & CLONE_SIGHAND)
1924                 unshare_flags |= CLONE_THREAD;
1925         /*
1926          * If unsharing namespace, must also unshare filesystem information.
1927          */
1928         if (unshare_flags & CLONE_NEWNS)
1929                 unshare_flags |= CLONE_FS;
1930
1931         err = check_unshare_flags(unshare_flags);
1932         if (err)
1933                 goto bad_unshare_out;
1934         /*
1935          * CLONE_NEWIPC must also detach from the undolist: after switching
1936          * to a new ipc namespace, the semaphore arrays from the old
1937          * namespace are unreachable.
1938          */
1939         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1940                 do_sysvsem = 1;
1941         err = unshare_fs(unshare_flags, &new_fs);
1942         if (err)
1943                 goto bad_unshare_out;
1944         err = unshare_fd(unshare_flags, &new_fd);
1945         if (err)
1946                 goto bad_unshare_cleanup_fs;
1947         err = unshare_userns(unshare_flags, &new_cred);
1948         if (err)
1949                 goto bad_unshare_cleanup_fd;
1950         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1951                                          new_cred, new_fs);
1952         if (err)
1953                 goto bad_unshare_cleanup_cred;
1954
1955         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1956                 if (do_sysvsem) {
1957                         /*
1958                          * CLONE_SYSVSEM is equivalent to sys_exit().
1959                          */
1960                         exit_sem(current);
1961                 }
1962
1963                 if (new_nsproxy)
1964                         switch_task_namespaces(current, new_nsproxy);
1965
1966                 task_lock(current);
1967
1968                 if (new_fs) {
1969                         fs = current->fs;
1970                         spin_lock(&fs->lock);
1971                         current->fs = new_fs;
1972                         if (--fs->users)
1973                                 new_fs = NULL;
1974                         else
1975                                 new_fs = fs;
1976                         spin_unlock(&fs->lock);
1977                 }
1978
1979                 if (new_fd) {
1980                         fd = current->files;
1981                         current->files = new_fd;
1982                         new_fd = fd;
1983                 }
1984
1985                 task_unlock(current);
1986
1987                 if (new_cred) {
1988                         /* Install the new user namespace */
1989                         commit_creds(new_cred);
1990                         new_cred = NULL;
1991                 }
1992         }
1993
1994 bad_unshare_cleanup_cred:
1995         if (new_cred)
1996                 put_cred(new_cred);
1997 bad_unshare_cleanup_fd:
1998         if (new_fd)
1999                 put_files_struct(new_fd);
2000
2001 bad_unshare_cleanup_fs:
2002         if (new_fs)
2003                 free_fs_struct(new_fs);
2004
2005 bad_unshare_out:
2006         return err;
2007 }
2008
2009 /*
2010  *      Helper to unshare the files of the current task.
2011  *      We don't want to expose copy_files internals to
2012  *      the exec layer of the kernel.
2013  */
2014
2015 int unshare_files(struct files_struct **displaced)
2016 {
2017         struct task_struct *task = current;
2018         struct files_struct *copy = NULL;
2019         int error;
2020
2021         error = unshare_fd(CLONE_FILES, &copy);
2022         if (error || !copy) {
2023                 *displaced = NULL;
2024                 return error;
2025         }
2026         *displaced = task->files;
2027         task_lock(task);
2028         task->files = copy;
2029         task_unlock(task);
2030         return 0;
2031 }