kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/mm.h>
  32 #include <linux/vmacache.h>
  33 #include <linux/nsproxy.h>
  34 #include <linux/capability.h>
  35 #include <linux/cpu.h>
  36 #include <linux/cgroup.h>
  37 #include <linux/security.h>
  38 #include <linux/hugetlb.h>
  39 #include <linux/seccomp.h>
  40 #include <linux/swap.h>
  41 #include <linux/syscalls.h>
  42 #include <linux/jiffies.h>
  43 #include <linux/futex.h>
  44 #include <linux/compat.h>
  45 #include <linux/kthread.h>
  46 #include <linux/task_io_accounting_ops.h>
  47 #include <linux/rcupdate.h>
  48 #include <linux/ptrace.h>
  49 #include <linux/mount.h>
  50 #include <linux/audit.h>
  51 #include <linux/memcontrol.h>
  52 #include <linux/ftrace.h>
  53 #include <linux/proc_fs.h>
  54 #include <linux/profile.h>
  55 #include <linux/rmap.h>
  56 #include <linux/ksm.h>
  57 #include <linux/acct.h>
  58 #include <linux/tsacct_kern.h>
  59 #include <linux/cn_proc.h>
  60 #include <linux/freezer.h>
  61 #include <linux/delayacct.h>
  62 #include <linux/taskstats_kern.h>
  63 #include <linux/random.h>
  64 #include <linux/tty.h>
  65 #include <linux/blkdev.h>
  66 #include <linux/fs_struct.h>
  67 #include <linux/magic.h>
  68 #include <linux/perf_event.h>
  69 #include <linux/posix-timers.h>
  70 #include <linux/user-return-notifier.h>
  71 #include <linux/oom.h>
  72 #include <linux/khugepaged.h>
  73 #include <linux/signalfd.h>
  74 #include <linux/uprobes.h>
  75 #include <linux/aio.h>
  76 #include <linux/compiler.h>
  77
  78 #include <asm/pgtable.h>
  79 #include <asm/pgalloc.h>
  80 #include <asm/uaccess.h>
  81 #include <asm/mmu_context.h>
  82 #include <asm/cacheflush.h>
  83 #include <asm/tlbflush.h>
  84
  85 #include <trace/events/sched.h>
  86
  87 #define CREATE_TRACE_POINTS
  88 #include <trace/events/task.h>
  89
  90 /*
  91  * Protected counters by write_lock_irq(&tasklist_lock)
  92  */
  93 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  94 int nr_threads;                 /* The idle threads do not count.. */
  95
  96 int max_threads;                /* tunable limit on nr_threads */
  97
  98 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  99
 100 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 101
 102 #ifdef CONFIG_PROVE_RCU
 103 int lockdep_tasklist_lock_is_held(void)
 104 {
 105         return lockdep_is_held(&tasklist_lock);
 106 }
 107 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 108 #endif /* #ifdef CONFIG_PROVE_RCU */
 109
 110 int nr_processes(void)
 111 {
 112         int cpu;
 113         int total = 0;
 114
 115         for_each_possible_cpu(cpu)
 116                 total += per_cpu(process_counts, cpu);
 117
 118         return total;
 119 }
 120
 121 void __weak arch_release_task_struct(struct task_struct *tsk)
 122 {
 123 }
 124
 125 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 126 static struct kmem_cache *task_struct_cachep;
 127
 128 static inline struct task_struct *alloc_task_struct_node(int node)
 129 {
 130         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 131 }
 132
 133 static inline void free_task_struct(struct task_struct *tsk)
 134 {
 135         kmem_cache_free(task_struct_cachep, tsk);
 136 }
 137 #endif
 138
 139 void __weak arch_release_thread_info(struct thread_info *ti)
 140 {
 141 }
 142
 143 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
 144
 145 /*
 146  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 147  * kmemcache based allocator.
 148  */
 149 # if THREAD_SIZE >= PAGE_SIZE
 150 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 151                                                   int node)
 152 {
 153         struct page *page = alloc_kmem_pages_node(node, THREADINFO_GFP,
 154                                                   THREAD_SIZE_ORDER);
 155
 156         return page ? page_address(page) : NULL;
 157 }
 158
 159 static inline void free_thread_info(struct thread_info *ti)
 160 {
 161         free_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 162 }
 163 # else
 164 static struct kmem_cache *thread_info_cache;
 165
 166 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 167                                                   int node)
 168 {
 169         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
 170 }
 171
 172 static void free_thread_info(struct thread_info *ti)
 173 {
 174         kmem_cache_free(thread_info_cache, ti);
 175 }
 176
 177 void thread_info_cache_init(void)
 178 {
 179         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
 180                                               THREAD_SIZE, 0, NULL);
 181         BUG_ON(thread_info_cache == NULL);
 182 }
 183 # endif
 184 #endif
 185
 186 /* SLAB cache for signal_struct structures (tsk->signal) */
 187 static struct kmem_cache *signal_cachep;
 188
 189 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 190 struct kmem_cache *sighand_cachep;
 191
 192 /* SLAB cache for files_struct structures (tsk->files) */
 193 struct kmem_cache *files_cachep;
 194
 195 /* SLAB cache for fs_struct structures (tsk->fs) */
 196 struct kmem_cache *fs_cachep;
 197
 198 /* SLAB cache for vm_area_struct structures */
 199 struct kmem_cache *vm_area_cachep;
 200
 201 /* SLAB cache for mm_struct structures (tsk->mm) */
 202 static struct kmem_cache *mm_cachep;
 203
 204 static void account_kernel_stack(struct thread_info *ti, int account)
 205 {
 206         struct zone *zone = page_zone(virt_to_page(ti));
 207
 208         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 209 }
 210
 211 void free_task(struct task_struct *tsk)
 212 {
 213         account_kernel_stack(tsk->stack, -1);
 214         arch_release_thread_info(tsk->stack);
 215         free_thread_info(tsk->stack);
 216         rt_mutex_debug_task_free(tsk);
 217         ftrace_graph_exit_task(tsk);
 218         put_seccomp_filter(tsk);
 219         arch_release_task_struct(tsk);
 220         free_task_struct(tsk);
 221 }
 222 EXPORT_SYMBOL(free_task);
 223
 224 static inline void free_signal_struct(struct signal_struct *sig)
 225 {
 226         taskstats_tgid_free(sig);
 227         sched_autogroup_exit(sig);
 228         kmem_cache_free(signal_cachep, sig);
 229 }
 230
 231 static inline void put_signal_struct(struct signal_struct *sig)
 232 {
 233         if (atomic_dec_and_test(&sig->sigcnt))
 234                 free_signal_struct(sig);
 235 }
 236
 237 void __put_task_struct(struct task_struct *tsk)
 238 {
 239         WARN_ON(!tsk->exit_state);
 240         WARN_ON(atomic_read(&tsk->usage));
 241         WARN_ON(tsk == current);
 242
 243         task_numa_free(tsk);
 244         security_task_free(tsk);
 245         exit_creds(tsk);
 246         delayacct_tsk_free(tsk);
 247         put_signal_struct(tsk->signal);
 248
 249         if (!profile_handoff_task(tsk))
 250                 free_task(tsk);
 251 }
 252 EXPORT_SYMBOL_GPL(__put_task_struct);
 253
 254 void __init __weak arch_task_cache_init(void) { }
 255
 256 void __init fork_init(unsigned long mempages)
 257 {
 258 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 259 #ifndef ARCH_MIN_TASKALIGN
 260 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 261 #endif
 262         /* create a slab on which task_structs can be allocated */
 263         task_struct_cachep =
 264                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 265                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 266 #endif
 267
 268         /* do the arch specific task caches init */
 269         arch_task_cache_init();
 270
 271         /*
 272          * The default maximum number of threads is set to a safe
 273          * value: the thread structures can take up at most half
 274          * of memory.
 275          */
 276         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 277
 278         /*
 279          * we need to allow at least 20 threads to boot a system
 280          */
 281         if (max_threads < 20)
 282                 max_threads = 20;
 283
 284         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 285         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 286         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 287                 init_task.signal->rlim[RLIMIT_NPROC];
 288 }
 289
 290 int __weak arch_dup_task_struct(struct task_struct *dst,
 291                                                struct task_struct *src)
 292 {
 293         *dst = *src;
 294         return 0;
 295 }
 296
 297 void set_task_stack_end_magic(struct task_struct *tsk)
 298 {
 299         unsigned long *stackend;
 300
 301         stackend = end_of_stack(tsk);
 302         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 303 }
 304
 305 static struct task_struct *dup_task_struct(struct task_struct *orig)
 306 {
 307         struct task_struct *tsk;
 308         struct thread_info *ti;
 309         int node = tsk_fork_get_node(orig);
 310         int err;
 311
 312         tsk = alloc_task_struct_node(node);
 313         if (!tsk)
 314                 return NULL;
 315
 316         ti = alloc_thread_info_node(tsk, node);
 317         if (!ti)
 318                 goto free_tsk;
 319
 320         err = arch_dup_task_struct(tsk, orig);
 321         if (err)
 322                 goto free_ti;
 323
 324         tsk->stack = ti;
 325 #ifdef CONFIG_SECCOMP
 326         /*
 327          * We must handle setting up seccomp filters once we're under
 328          * the sighand lock in case orig has changed between now and
 329          * then. Until then, filter must be NULL to avoid messing up
 330          * the usage counts on the error path calling free_task.
 331          */
 332         tsk->seccomp.filter = NULL;
 333 #endif
 334
 335         setup_thread_stack(tsk, orig);
 336         clear_user_return_notifier(tsk);
 337         clear_tsk_need_resched(tsk);
 338         set_task_stack_end_magic(tsk);
 339
 340 #ifdef CONFIG_CC_STACKPROTECTOR
 341         tsk->stack_canary = get_random_int();
 342 #endif
 343
 344         /*
 345          * One for us, one for whoever does the "release_task()" (usually
 346          * parent)
 347          */
 348         atomic_set(&tsk->usage, 2);
 349 #ifdef CONFIG_BLK_DEV_IO_TRACE
 350         tsk->btrace_seq = 0;
 351 #endif
 352         tsk->splice_pipe = NULL;
 353         tsk->task_frag.page = NULL;
 354
 355         account_kernel_stack(ti, 1);
 356
 357         return tsk;
 358
 359 free_ti:
 360         free_thread_info(ti);
 361 free_tsk:
 362         free_task_struct(tsk);
 363         return NULL;
 364 }
 365
 366 #ifdef CONFIG_MMU
 367 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 368 {
 369         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 370         struct rb_node **rb_link, *rb_parent;
 371         int retval;
 372         unsigned long charge;
 373
 374         uprobe_start_dup_mmap();
 375         down_write(&oldmm->mmap_sem);
 376         flush_cache_dup_mm(oldmm);
 377         uprobe_dup_mmap(oldmm, mm);
 378         /*
 379          * Not linked in yet - no deadlock potential:
 380          */
 381         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 382
 383         mm->total_vm = oldmm->total_vm;
 384         mm->shared_vm = oldmm->shared_vm;
 385         mm->exec_vm = oldmm->exec_vm;
 386         mm->stack_vm = oldmm->stack_vm;
 387
 388         rb_link = &mm->mm_rb.rb_node;
 389         rb_parent = NULL;
 390         pprev = &mm->mmap;
 391         retval = ksm_fork(mm, oldmm);
 392         if (retval)
 393                 goto out;
 394         retval = khugepaged_fork(mm, oldmm);
 395         if (retval)
 396                 goto out;
 397
 398         prev = NULL;
 399         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 400                 struct file *file;
 401
 402                 if (mpnt->vm_flags & VM_DONTCOPY) {
 403                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 404                                                         -vma_pages(mpnt));
 405                         continue;
 406                 }
 407                 charge = 0;
 408                 if (mpnt->vm_flags & VM_ACCOUNT) {
 409                         unsigned long len = vma_pages(mpnt);
 410
 411                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 412                                 goto fail_nomem;
 413                         charge = len;
 414                 }
 415                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 416                 if (!tmp)
 417                         goto fail_nomem;
 418                 *tmp = *mpnt;
 419                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 420                 retval = vma_dup_policy(mpnt, tmp);
 421                 if (retval)
 422                         goto fail_nomem_policy;
 423                 tmp->vm_mm = mm;
 424                 if (anon_vma_fork(tmp, mpnt))
 425                         goto fail_nomem_anon_vma_fork;
 426                 tmp->vm_flags &= ~VM_LOCKED;
 427                 tmp->vm_next = tmp->vm_prev = NULL;
 428                 file = tmp->vm_file;
 429                 if (file) {
 430                         struct inode *inode = file_inode(file);
 431                         struct address_space *mapping = file->f_mapping;
 432
 433                         get_file(file);
 434                         if (tmp->vm_flags & VM_DENYWRITE)
 435                                 atomic_dec(&inode->i_writecount);
 436                         i_mmap_lock_write(mapping);
 437                         if (tmp->vm_flags & VM_SHARED)
 438                                 atomic_inc(&mapping->i_mmap_writable);
 439                         flush_dcache_mmap_lock(mapping);
 440                         /* insert tmp into the share list, just after mpnt */
 441                         vma_interval_tree_insert_after(tmp, mpnt,
 442                                         &mapping->i_mmap);
 443                         flush_dcache_mmap_unlock(mapping);
 444                         i_mmap_unlock_write(mapping);
 445                 }
 446
 447                 /*
 448                  * Clear hugetlb-related page reserves for children. This only
 449                  * affects MAP_PRIVATE mappings. Faults generated by the child
 450                  * are not guaranteed to succeed, even if read-only
 451                  */
 452                 if (is_vm_hugetlb_page(tmp))
 453                         reset_vma_resv_huge_pages(tmp);
 454
 455                 /*
 456                  * Link in the new vma and copy the page table entries.
 457                  */
 458                 *pprev = tmp;
 459                 pprev = &tmp->vm_next;
 460                 tmp->vm_prev = prev;
 461                 prev = tmp;
 462
 463                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 464                 rb_link = &tmp->vm_rb.rb_right;
 465                 rb_parent = &tmp->vm_rb;
 466
 467                 mm->map_count++;
 468                 retval = copy_page_range(mm, oldmm, mpnt);
 469
 470                 if (tmp->vm_ops && tmp->vm_ops->open)
 471                         tmp->vm_ops->open(tmp);
 472
 473                 if (retval)
 474                         goto out;
 475         }
 476         /* a new mm has just been created */
 477         arch_dup_mmap(oldmm, mm);
 478         retval = 0;
 479 out:
 480         up_write(&mm->mmap_sem);
 481         flush_tlb_mm(oldmm);
 482         up_write(&oldmm->mmap_sem);
 483         uprobe_end_dup_mmap();
 484         return retval;
 485 fail_nomem_anon_vma_fork:
 486         mpol_put(vma_policy(tmp));
 487 fail_nomem_policy:
 488         kmem_cache_free(vm_area_cachep, tmp);
 489 fail_nomem:
 490         retval = -ENOMEM;
 491         vm_unacct_memory(charge);
 492         goto out;
 493 }
 494
 495 static inline int mm_alloc_pgd(struct mm_struct *mm)
 496 {
 497         mm->pgd = pgd_alloc(mm);
 498         if (unlikely(!mm->pgd))
 499                 return -ENOMEM;
 500         return 0;
 501 }
 502
 503 static inline void mm_free_pgd(struct mm_struct *mm)
 504 {
 505         pgd_free(mm, mm->pgd);
 506 }
 507 #else
 508 #define dup_mmap(mm, oldmm)     (0)
 509 #define mm_alloc_pgd(mm)        (0)
 510 #define mm_free_pgd(mm)
 511 #endif /* CONFIG_MMU */
 512
 513 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 514
 515 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 516 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 517
 518 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 519
 520 static int __init coredump_filter_setup(char *s)
 521 {
 522         default_dump_filter =
 523                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 524                 MMF_DUMP_FILTER_MASK;
 525         return 1;
 526 }
 527
 528 __setup("coredump_filter=", coredump_filter_setup);
 529
 530 #include <linux/init_task.h>
 531
 532 static void mm_init_aio(struct mm_struct *mm)
 533 {
 534 #ifdef CONFIG_AIO
 535         spin_lock_init(&mm->ioctx_lock);
 536         mm->ioctx_table = NULL;
 537 #endif
 538 }
 539
 540 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 541 {
 542 #ifdef CONFIG_MEMCG
 543         mm->owner = p;
 544 #endif
 545 }
 546
 547 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 548 {
 549         mm->mmap = NULL;
 550         mm->mm_rb = RB_ROOT;
 551         mm->vmacache_seqnum = 0;
 552         atomic_set(&mm->mm_users, 1);
 553         atomic_set(&mm->mm_count, 1);
 554         init_rwsem(&mm->mmap_sem);
 555         INIT_LIST_HEAD(&mm->mmlist);
 556         mm->core_state = NULL;
 557         atomic_long_set(&mm->nr_ptes, 0);
 558 #ifndef __PAGETABLE_PMD_FOLDED
 559         atomic_long_set(&mm->nr_pmds, 0);
 560 #endif
 561         mm->map_count = 0;
 562         mm->locked_vm = 0;
 563         mm->pinned_vm = 0;
 564         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 565         spin_lock_init(&mm->page_table_lock);
 566         mm_init_cpumask(mm);
 567         mm_init_aio(mm);
 568         mm_init_owner(mm, p);
 569         mmu_notifier_mm_init(mm);
 570         clear_tlb_flush_pending(mm);
 571 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 572         mm->pmd_huge_pte = NULL;
 573 #endif
 574
 575         if (current->mm) {
 576                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 577                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 578         } else {
 579                 mm->flags = default_dump_filter;
 580                 mm->def_flags = 0;
 581         }
 582
 583         if (mm_alloc_pgd(mm))
 584                 goto fail_nopgd;
 585
 586         if (init_new_context(p, mm))
 587                 goto fail_nocontext;
 588
 589         return mm;
 590
 591 fail_nocontext:
 592         mm_free_pgd(mm);
 593 fail_nopgd:
 594         free_mm(mm);
 595         return NULL;
 596 }
 597
 598 static void check_mm(struct mm_struct *mm)
 599 {
 600         int i;
 601
 602         for (i = 0; i < NR_MM_COUNTERS; i++) {
 603                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 604
 605                 if (unlikely(x))
 606                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 607                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 608         }
 609
 610         if (atomic_long_read(&mm->nr_ptes))
 611                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 612                                 atomic_long_read(&mm->nr_ptes));
 613         if (mm_nr_pmds(mm))
 614                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 615                                 mm_nr_pmds(mm));
 616
 617 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 618         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 619 #endif
 620 }
 621
 622 /*
 623  * Allocate and initialize an mm_struct.
 624  */
 625 struct mm_struct *mm_alloc(void)
 626 {
 627         struct mm_struct *mm;
 628
 629         mm = allocate_mm();
 630         if (!mm)
 631                 return NULL;
 632
 633         memset(mm, 0, sizeof(*mm));
 634         return mm_init(mm, current);
 635 }
 636
 637 /*
 638  * Called when the last reference to the mm
 639  * is dropped: either by a lazy thread or by
 640  * mmput. Free the page directory and the mm.
 641  */
 642 void __mmdrop(struct mm_struct *mm)
 643 {
 644         BUG_ON(mm == &init_mm);
 645         mm_free_pgd(mm);
 646         destroy_context(mm);
 647         mmu_notifier_mm_destroy(mm);
 648         check_mm(mm);
 649         free_mm(mm);
 650 }
 651 EXPORT_SYMBOL_GPL(__mmdrop);
 652
 653 /*
 654  * Decrement the use count and release all resources for an mm.
 655  */
 656 void mmput(struct mm_struct *mm)
 657 {
 658         might_sleep();
 659
 660         if (atomic_dec_and_test(&mm->mm_users)) {
 661                 uprobe_clear_state(mm);
 662                 exit_aio(mm);
 663                 ksm_exit(mm);
 664                 khugepaged_exit(mm); /* must run before exit_mmap */
 665                 exit_mmap(mm);
 666                 set_mm_exe_file(mm, NULL);
 667                 if (!list_empty(&mm->mmlist)) {
 668                         spin_lock(&mmlist_lock);
 669                         list_del(&mm->mmlist);
 670                         spin_unlock(&mmlist_lock);
 671                 }
 672                 if (mm->binfmt)
 673                         module_put(mm->binfmt->module);
 674                 mmdrop(mm);
 675         }
 676 }
 677 EXPORT_SYMBOL_GPL(mmput);
 678
 679 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 680 {
 681         if (new_exe_file)
 682                 get_file(new_exe_file);
 683         if (mm->exe_file)
 684                 fput(mm->exe_file);
 685         mm->exe_file = new_exe_file;
 686 }
 687
 688 struct file *get_mm_exe_file(struct mm_struct *mm)
 689 {
 690         struct file *exe_file;
 691
 692         /* We need mmap_sem to protect against races with removal of exe_file */
 693         down_read(&mm->mmap_sem);
 694         exe_file = mm->exe_file;
 695         if (exe_file)
 696                 get_file(exe_file);
 697         up_read(&mm->mmap_sem);
 698         return exe_file;
 699 }
 700
 701 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 702 {
 703         /* It's safe to write the exe_file pointer without exe_file_lock because
 704          * this is called during fork when the task is not yet in /proc */
 705         newmm->exe_file = get_mm_exe_file(oldmm);
 706 }
 707
 708 /**
 709  * get_task_mm - acquire a reference to the task's mm
 710  *
 711  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 712  * this kernel workthread has transiently adopted a user mm with use_mm,
 713  * to do its AIO) is not set and if so returns a reference to it, after
 714  * bumping up the use count.  User must release the mm via mmput()
 715  * after use.  Typically used by /proc and ptrace.
 716  */
 717 struct mm_struct *get_task_mm(struct task_struct *task)
 718 {
 719         struct mm_struct *mm;
 720
 721         task_lock(task);
 722         mm = task->mm;
 723         if (mm) {
 724                 if (task->flags & PF_KTHREAD)
 725                         mm = NULL;
 726                 else
 727                         atomic_inc(&mm->mm_users);
 728         }
 729         task_unlock(task);
 730         return mm;
 731 }
 732 EXPORT_SYMBOL_GPL(get_task_mm);
 733
 734 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 735 {
 736         struct mm_struct *mm;
 737         int err;
 738
 739         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 740         if (err)
 741                 return ERR_PTR(err);
 742
 743         mm = get_task_mm(task);
 744         if (mm && mm != current->mm &&
 745                         !ptrace_may_access(task, mode)) {
 746                 mmput(mm);
 747                 mm = ERR_PTR(-EACCES);
 748         }
 749         mutex_unlock(&task->signal->cred_guard_mutex);
 750
 751         return mm;
 752 }
 753
 754 static void complete_vfork_done(struct task_struct *tsk)
 755 {
 756         struct completion *vfork;
 757
 758         task_lock(tsk);
 759         vfork = tsk->vfork_done;
 760         if (likely(vfork)) {
 761                 tsk->vfork_done = NULL;
 762                 complete(vfork);
 763         }
 764         task_unlock(tsk);
 765 }
 766
 767 static int wait_for_vfork_done(struct task_struct *child,
 768                                 struct completion *vfork)
 769 {
 770         int killed;
 771
 772         freezer_do_not_count();
 773         killed = wait_for_completion_killable(vfork);
 774         freezer_count();
 775
 776         if (killed) {
 777                 task_lock(child);
 778                 child->vfork_done = NULL;
 779                 task_unlock(child);
 780         }
 781
 782         put_task_struct(child);
 783         return killed;
 784 }
 785
 786 /* Please note the differences between mmput and mm_release.
 787  * mmput is called whenever we stop holding onto a mm_struct,
 788  * error success whatever.
 789  *
 790  * mm_release is called after a mm_struct has been removed
 791  * from the current process.
 792  *
 793  * This difference is important for error handling, when we
 794  * only half set up a mm_struct for a new process and need to restore
 795  * the old one.  Because we mmput the new mm_struct before
 796  * restoring the old one. . .
 797  * Eric Biederman 10 January 1998
 798  */
 799 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 800 {
 801         /* Get rid of any futexes when releasing the mm */
 802 #ifdef CONFIG_FUTEX
 803         if (unlikely(tsk->robust_list)) {
 804                 exit_robust_list(tsk);
 805                 tsk->robust_list = NULL;
 806         }
 807 #ifdef CONFIG_COMPAT
 808         if (unlikely(tsk->compat_robust_list)) {
 809                 compat_exit_robust_list(tsk);
 810                 tsk->compat_robust_list = NULL;
 811         }
 812 #endif
 813         if (unlikely(!list_empty(&tsk->pi_state_list)))
 814                 exit_pi_state_list(tsk);
 815 #endif
 816
 817         uprobe_free_utask(tsk);
 818
 819         /* Get rid of any cached register state */
 820         deactivate_mm(tsk, mm);
 821
 822         /*
 823          * If we're exiting normally, clear a user-space tid field if
 824          * requested.  We leave this alone when dying by signal, to leave
 825          * the value intact in a core dump, and to save the unnecessary
 826          * trouble, say, a killed vfork parent shouldn't touch this mm.
 827          * Userland only wants this done for a sys_exit.
 828          */
 829         if (tsk->clear_child_tid) {
 830                 if (!(tsk->flags & PF_SIGNALED) &&
 831                     atomic_read(&mm->mm_users) > 1) {
 832                         /*
 833                          * We don't check the error code - if userspace has
 834                          * not set up a proper pointer then tough luck.
 835                          */
 836                         put_user(0, tsk->clear_child_tid);
 837                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 838                                         1, NULL, NULL, 0);
 839                 }
 840                 tsk->clear_child_tid = NULL;
 841         }
 842
 843         /*
 844          * All done, finally we can wake up parent and return this mm to him.
 845          * Also kthread_stop() uses this completion for synchronization.
 846          */
 847         if (tsk->vfork_done)
 848                 complete_vfork_done(tsk);
 849 }
 850
 851 /*
 852  * Allocate a new mm structure and copy contents from the
 853  * mm structure of the passed in task structure.
 854  */
 855 static struct mm_struct *dup_mm(struct task_struct *tsk)
 856 {
 857         struct mm_struct *mm, *oldmm = current->mm;
 858         int err;
 859
 860         mm = allocate_mm();
 861         if (!mm)
 862                 goto fail_nomem;
 863
 864         memcpy(mm, oldmm, sizeof(*mm));
 865
 866         if (!mm_init(mm, tsk))
 867                 goto fail_nomem;
 868
 869         dup_mm_exe_file(oldmm, mm);
 870
 871         err = dup_mmap(mm, oldmm);
 872         if (err)
 873                 goto free_pt;
 874
 875         mm->hiwater_rss = get_mm_rss(mm);
 876         mm->hiwater_vm = mm->total_vm;
 877
 878         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 879                 goto free_pt;
 880
 881         return mm;
 882
 883 free_pt:
 884         /* don't put binfmt in mmput, we haven't got module yet */
 885         mm->binfmt = NULL;
 886         mmput(mm);
 887
 888 fail_nomem:
 889         return NULL;
 890 }
 891
 892 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 893 {
 894         struct mm_struct *mm, *oldmm;
 895         int retval;
 896
 897         tsk->min_flt = tsk->maj_flt = 0;
 898         tsk->nvcsw = tsk->nivcsw = 0;
 899 #ifdef CONFIG_DETECT_HUNG_TASK
 900         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 901 #endif
 902
 903         tsk->mm = NULL;
 904         tsk->active_mm = NULL;
 905
 906         /*
 907          * Are we cloning a kernel thread?
 908          *
 909          * We need to steal a active VM for that..
 910          */
 911         oldmm = current->mm;
 912         if (!oldmm)
 913                 return 0;
 914
 915         /* initialize the new vmacache entries */
 916         vmacache_flush(tsk);
 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                 /*
1033                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1034                  * without an RCU grace period, see __lock_task_sighand().
1035                  */
1036                 kmem_cache_free(sighand_cachep, sighand);
1037         }
1038 }
1039
1040 /*
1041  * Initialize POSIX timer handling for a thread group.
1042  */
1043 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1044 {
1045         unsigned long cpu_limit;
1046
1047         /* Thread group counters. */
1048         thread_group_cputime_init(sig);
1049
1050         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1051         if (cpu_limit != RLIM_INFINITY) {
1052                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1053                 sig->cputimer.running = 1;
1054         }
1055
1056         /* The timer lists. */
1057         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1058         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1059         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1060 }
1061
1062 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1063 {
1064         struct signal_struct *sig;
1065
1066         if (clone_flags & CLONE_THREAD)
1067                 return 0;
1068
1069         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1070         tsk->signal = sig;
1071         if (!sig)
1072                 return -ENOMEM;
1073
1074         sig->nr_threads = 1;
1075         atomic_set(&sig->live, 1);
1076         atomic_set(&sig->sigcnt, 1);
1077
1078         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1079         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1080         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1081
1082         init_waitqueue_head(&sig->wait_chldexit);
1083         sig->curr_target = tsk;
1084         init_sigpending(&sig->shared_pending);
1085         INIT_LIST_HEAD(&sig->posix_timers);
1086         seqlock_init(&sig->stats_lock);
1087
1088         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1089         sig->real_timer.function = it_real_fn;
1090
1091         task_lock(current->group_leader);
1092         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1093         task_unlock(current->group_leader);
1094
1095         posix_cpu_timers_init_group(sig);
1096
1097         tty_audit_fork(sig);
1098         sched_autogroup_fork(sig);
1099
1100 #ifdef CONFIG_CGROUPS
1101         init_rwsem(&sig->group_rwsem);
1102 #endif
1103
1104         sig->oom_score_adj = current->signal->oom_score_adj;
1105         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1106
1107         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1108                                    current->signal->is_child_subreaper;
1109
1110         mutex_init(&sig->cred_guard_mutex);
1111
1112         return 0;
1113 }
1114
1115 static void copy_seccomp(struct task_struct *p)
1116 {
1117 #ifdef CONFIG_SECCOMP
1118         /*
1119          * Must be called with sighand->lock held, which is common to
1120          * all threads in the group. Holding cred_guard_mutex is not
1121          * needed because this new task is not yet running and cannot
1122          * be racing exec.
1123          */
1124         assert_spin_locked(&current->sighand->siglock);
1125
1126         /* Ref-count the new filter user, and assign it. */
1127         get_seccomp_filter(current);
1128         p->seccomp = current->seccomp;
1129
1130         /*
1131          * Explicitly enable no_new_privs here in case it got set
1132          * between the task_struct being duplicated and holding the
1133          * sighand lock. The seccomp state and nnp must be in sync.
1134          */
1135         if (task_no_new_privs(current))
1136                 task_set_no_new_privs(p);
1137
1138         /*
1139          * If the parent gained a seccomp mode after copying thread
1140          * flags and between before we held the sighand lock, we have
1141          * to manually enable the seccomp thread flag here.
1142          */
1143         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1144                 set_tsk_thread_flag(p, TIF_SECCOMP);
1145 #endif
1146 }
1147
1148 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1149 {
1150         current->clear_child_tid = tidptr;
1151
1152         return task_pid_vnr(current);
1153 }
1154
1155 static void rt_mutex_init_task(struct task_struct *p)
1156 {
1157         raw_spin_lock_init(&p->pi_lock);
1158 #ifdef CONFIG_RT_MUTEXES
1159         p->pi_waiters = RB_ROOT;
1160         p->pi_waiters_leftmost = NULL;
1161         p->pi_blocked_on = NULL;
1162 #endif
1163 }
1164
1165 /*
1166  * Initialize POSIX timer handling for a single task.
1167  */
1168 static void posix_cpu_timers_init(struct task_struct *tsk)
1169 {
1170         tsk->cputime_expires.prof_exp = 0;
1171         tsk->cputime_expires.virt_exp = 0;
1172         tsk->cputime_expires.sched_exp = 0;
1173         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1174         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1175         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1176 }
1177
1178 static inline void
1179 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1180 {
1181          task->pids[type].pid = pid;
1182 }
1183
1184 /*
1185  * This creates a new process as a copy of the old one,
1186  * but does not actually start it yet.
1187  *
1188  * It copies the registers, and all the appropriate
1189  * parts of the process environment (as per the clone
1190  * flags). The actual kick-off is left to the caller.
1191  */
1192 static struct task_struct *copy_process(unsigned long clone_flags,
1193                                         unsigned long stack_start,
1194                                         unsigned long stack_size,
1195                                         int __user *child_tidptr,
1196                                         struct pid *pid,
1197                                         int trace)
1198 {
1199         int retval;
1200         struct task_struct *p;
1201
1202         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1203                 return ERR_PTR(-EINVAL);
1204
1205         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1206                 return ERR_PTR(-EINVAL);
1207
1208         /*
1209          * Thread groups must share signals as well, and detached threads
1210          * can only be started up within the thread group.
1211          */
1212         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1213                 return ERR_PTR(-EINVAL);
1214
1215         /*
1216          * Shared signal handlers imply shared VM. By way of the above,
1217          * thread groups also imply shared VM. Blocking this case allows
1218          * for various simplifications in other code.
1219          */
1220         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1221                 return ERR_PTR(-EINVAL);
1222
1223         /*
1224          * Siblings of global init remain as zombies on exit since they are
1225          * not reaped by their parent (swapper). To solve this and to avoid
1226          * multi-rooted process trees, prevent global and container-inits
1227          * from creating siblings.
1228          */
1229         if ((clone_flags & CLONE_PARENT) &&
1230                                 current->signal->flags & SIGNAL_UNKILLABLE)
1231                 return ERR_PTR(-EINVAL);
1232
1233         /*
1234          * If the new process will be in a different pid or user namespace
1235          * do not allow it to share a thread group or signal handlers or
1236          * parent with the forking task.
1237          */
1238         if (clone_flags & CLONE_SIGHAND) {
1239                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1240                     (task_active_pid_ns(current) !=
1241                                 current->nsproxy->pid_ns_for_children))
1242                         return ERR_PTR(-EINVAL);
1243         }
1244
1245         retval = security_task_create(clone_flags);
1246         if (retval)
1247                 goto fork_out;
1248
1249         retval = -ENOMEM;
1250         p = dup_task_struct(current);
1251         if (!p)
1252                 goto fork_out;
1253
1254         ftrace_graph_init_task(p);
1255
1256         rt_mutex_init_task(p);
1257
1258 #ifdef CONFIG_PROVE_LOCKING
1259         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1260         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1261 #endif
1262         retval = -EAGAIN;
1263         if (atomic_read(&p->real_cred->user->processes) >=
1264                         task_rlimit(p, RLIMIT_NPROC)) {
1265                 if (p->real_cred->user != INIT_USER &&
1266                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1267                         goto bad_fork_free;
1268         }
1269         current->flags &= ~PF_NPROC_EXCEEDED;
1270
1271         retval = copy_creds(p, clone_flags);
1272         if (retval < 0)
1273                 goto bad_fork_free;
1274
1275         /*
1276          * If multiple threads are within copy_process(), then this check
1277          * triggers too late. This doesn't hurt, the check is only there
1278          * to stop root fork bombs.
1279          */
1280         retval = -EAGAIN;
1281         if (nr_threads >= max_threads)
1282                 goto bad_fork_cleanup_count;
1283
1284         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1285                 goto bad_fork_cleanup_count;
1286
1287         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1288         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1289         p->flags |= PF_FORKNOEXEC;
1290         INIT_LIST_HEAD(&p->children);
1291         INIT_LIST_HEAD(&p->sibling);
1292         rcu_copy_process(p);
1293         p->vfork_done = NULL;
1294         spin_lock_init(&p->alloc_lock);
1295
1296         init_sigpending(&p->pending);
1297
1298         p->utime = p->stime = p->gtime = 0;
1299         p->utimescaled = p->stimescaled = 0;
1300 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1301         p->prev_cputime.utime = p->prev_cputime.stime = 0;
1302 #endif
1303 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1304         seqlock_init(&p->vtime_seqlock);
1305         p->vtime_snap = 0;
1306         p->vtime_snap_whence = VTIME_SLEEPING;
1307 #endif
1308
1309 #if defined(SPLIT_RSS_COUNTING)
1310         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1311 #endif
1312
1313         p->default_timer_slack_ns = current->timer_slack_ns;
1314
1315         task_io_accounting_init(&p->ioac);
1316         acct_clear_integrals(p);
1317
1318         posix_cpu_timers_init(p);
1319
1320         p->start_time = ktime_get_ns();
1321         p->real_start_time = ktime_get_boot_ns();
1322         p->io_context = NULL;
1323         p->audit_context = NULL;
1324         if (clone_flags & CLONE_THREAD)
1325                 threadgroup_change_begin(current);
1326         cgroup_fork(p);
1327 #ifdef CONFIG_NUMA
1328         p->mempolicy = mpol_dup(p->mempolicy);
1329         if (IS_ERR(p->mempolicy)) {
1330                 retval = PTR_ERR(p->mempolicy);
1331                 p->mempolicy = NULL;
1332                 goto bad_fork_cleanup_threadgroup_lock;
1333         }
1334 #endif
1335 #ifdef CONFIG_CPUSETS
1336         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1337         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1338         seqcount_init(&p->mems_allowed_seq);
1339 #endif
1340 #ifdef CONFIG_TRACE_IRQFLAGS
1341         p->irq_events = 0;
1342         p->hardirqs_enabled = 0;
1343         p->hardirq_enable_ip = 0;
1344         p->hardirq_enable_event = 0;
1345         p->hardirq_disable_ip = _THIS_IP_;
1346         p->hardirq_disable_event = 0;
1347         p->softirqs_enabled = 1;
1348         p->softirq_enable_ip = _THIS_IP_;
1349         p->softirq_enable_event = 0;
1350         p->softirq_disable_ip = 0;
1351         p->softirq_disable_event = 0;
1352         p->hardirq_context = 0;
1353         p->softirq_context = 0;
1354 #endif
1355 #ifdef CONFIG_LOCKDEP
1356         p->lockdep_depth = 0; /* no locks held yet */
1357         p->curr_chain_key = 0;
1358         p->lockdep_recursion = 0;
1359 #endif
1360
1361 #ifdef CONFIG_DEBUG_MUTEXES
1362         p->blocked_on = NULL; /* not blocked yet */
1363 #endif
1364 #ifdef CONFIG_BCACHE
1365         p->sequential_io        = 0;
1366         p->sequential_io_avg    = 0;
1367 #endif
1368
1369         /* Perform scheduler related setup. Assign this task to a CPU. */
1370         retval = sched_fork(clone_flags, p);
1371         if (retval)
1372                 goto bad_fork_cleanup_policy;
1373
1374         retval = perf_event_init_task(p);
1375         if (retval)
1376                 goto bad_fork_cleanup_policy;
1377         retval = audit_alloc(p);
1378         if (retval)
1379                 goto bad_fork_cleanup_perf;
1380         /* copy all the process information */
1381         shm_init_task(p);
1382         retval = copy_semundo(clone_flags, p);
1383         if (retval)
1384                 goto bad_fork_cleanup_audit;
1385         retval = copy_files(clone_flags, p);
1386         if (retval)
1387                 goto bad_fork_cleanup_semundo;
1388         retval = copy_fs(clone_flags, p);
1389         if (retval)
1390                 goto bad_fork_cleanup_files;
1391         retval = copy_sighand(clone_flags, p);
1392         if (retval)
1393                 goto bad_fork_cleanup_fs;
1394         retval = copy_signal(clone_flags, p);
1395         if (retval)
1396                 goto bad_fork_cleanup_sighand;
1397         retval = copy_mm(clone_flags, p);
1398         if (retval)
1399                 goto bad_fork_cleanup_signal;
1400         retval = copy_namespaces(clone_flags, p);
1401         if (retval)
1402                 goto bad_fork_cleanup_mm;
1403         retval = copy_io(clone_flags, p);
1404         if (retval)
1405                 goto bad_fork_cleanup_namespaces;
1406         retval = copy_thread(clone_flags, stack_start, stack_size, p);
1407         if (retval)
1408                 goto bad_fork_cleanup_io;
1409
1410         if (pid != &init_struct_pid) {
1411                 retval = -ENOMEM;
1412                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1413                 if (!pid)
1414                         goto bad_fork_cleanup_io;
1415         }
1416
1417         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1418         /*
1419          * Clear TID on mm_release()?
1420          */
1421         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1422 #ifdef CONFIG_BLOCK
1423         p->plug = NULL;
1424 #endif
1425 #ifdef CONFIG_FUTEX
1426         p->robust_list = NULL;
1427 #ifdef CONFIG_COMPAT
1428         p->compat_robust_list = NULL;
1429 #endif
1430         INIT_LIST_HEAD(&p->pi_state_list);
1431         p->pi_state_cache = NULL;
1432 #endif
1433         /*
1434          * sigaltstack should be cleared when sharing the same VM
1435          */
1436         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1437                 p->sas_ss_sp = p->sas_ss_size = 0;
1438
1439         /*
1440          * Syscall tracing and stepping should be turned off in the
1441          * child regardless of CLONE_PTRACE.
1442          */
1443         user_disable_single_step(p);
1444         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1445 #ifdef TIF_SYSCALL_EMU
1446         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1447 #endif
1448         clear_all_latency_tracing(p);
1449
1450         /* ok, now we should be set up.. */
1451         p->pid = pid_nr(pid);
1452         if (clone_flags & CLONE_THREAD) {
1453                 p->exit_signal = -1;
1454                 p->group_leader = current->group_leader;
1455                 p->tgid = current->tgid;
1456         } else {
1457                 if (clone_flags & CLONE_PARENT)
1458                         p->exit_signal = current->group_leader->exit_signal;
1459                 else
1460                         p->exit_signal = (clone_flags & CSIGNAL);
1461                 p->group_leader = p;
1462                 p->tgid = p->pid;
1463         }
1464
1465         p->nr_dirtied = 0;
1466         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1467         p->dirty_paused_when = 0;
1468
1469         p->pdeath_signal = 0;
1470         INIT_LIST_HEAD(&p->thread_group);
1471         p->task_works = NULL;
1472
1473         /*
1474          * Make it visible to the rest of the system, but dont wake it up yet.
1475          * Need tasklist lock for parent etc handling!
1476          */
1477         write_lock_irq(&tasklist_lock);
1478
1479         /* CLONE_PARENT re-uses the old parent */
1480         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1481                 p->real_parent = current->real_parent;
1482                 p->parent_exec_id = current->parent_exec_id;
1483         } else {
1484                 p->real_parent = current;
1485                 p->parent_exec_id = current->self_exec_id;
1486         }
1487
1488         spin_lock(&current->sighand->siglock);
1489
1490         /*
1491          * Copy seccomp details explicitly here, in case they were changed
1492          * before holding sighand lock.
1493          */
1494         copy_seccomp(p);
1495
1496         /*
1497          * Process group and session signals need to be delivered to just the
1498          * parent before the fork or both the parent and the child after the
1499          * fork. Restart if a signal comes in before we add the new process to
1500          * it's process group.
1501          * A fatal signal pending means that current will exit, so the new
1502          * thread can't slip out of an OOM kill (or normal SIGKILL).
1503         */
1504         recalc_sigpending();
1505         if (signal_pending(current)) {
1506                 spin_unlock(&current->sighand->siglock);
1507                 write_unlock_irq(&tasklist_lock);
1508                 retval = -ERESTARTNOINTR;
1509                 goto bad_fork_free_pid;
1510         }
1511
1512         if (likely(p->pid)) {
1513                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1514
1515                 init_task_pid(p, PIDTYPE_PID, pid);
1516                 if (thread_group_leader(p)) {
1517                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1518                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1519
1520                         if (is_child_reaper(pid)) {
1521                                 ns_of_pid(pid)->child_reaper = p;
1522                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1523                         }
1524
1525                         p->signal->leader_pid = pid;
1526                         p->signal->tty = tty_kref_get(current->signal->tty);
1527                         list_add_tail(&p->sibling, &p->real_parent->children);
1528                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1529                         attach_pid(p, PIDTYPE_PGID);
1530                         attach_pid(p, PIDTYPE_SID);
1531                         __this_cpu_inc(process_counts);
1532                 } else {
1533                         current->signal->nr_threads++;
1534                         atomic_inc(&current->signal->live);
1535                         atomic_inc(&current->signal->sigcnt);
1536                         list_add_tail_rcu(&p->thread_group,
1537                                           &p->group_leader->thread_group);
1538                         list_add_tail_rcu(&p->thread_node,
1539                                           &p->signal->thread_head);
1540                 }
1541                 attach_pid(p, PIDTYPE_PID);
1542                 nr_threads++;
1543         }
1544
1545         total_forks++;
1546         spin_unlock(&current->sighand->siglock);
1547         syscall_tracepoint_update(p);
1548         write_unlock_irq(&tasklist_lock);
1549
1550         proc_fork_connector(p);
1551         cgroup_post_fork(p);
1552         if (clone_flags & CLONE_THREAD)
1553                 threadgroup_change_end(current);
1554         perf_event_fork(p);
1555
1556         trace_task_newtask(p, clone_flags);
1557         uprobe_copy_process(p, clone_flags);
1558
1559         return p;
1560
1561 bad_fork_free_pid:
1562         if (pid != &init_struct_pid)
1563                 free_pid(pid);
1564 bad_fork_cleanup_io:
1565         if (p->io_context)
1566                 exit_io_context(p);
1567 bad_fork_cleanup_namespaces:
1568         exit_task_namespaces(p);
1569 bad_fork_cleanup_mm:
1570         if (p->mm)
1571                 mmput(p->mm);
1572 bad_fork_cleanup_signal:
1573         if (!(clone_flags & CLONE_THREAD))
1574                 free_signal_struct(p->signal);
1575 bad_fork_cleanup_sighand:
1576         __cleanup_sighand(p->sighand);
1577 bad_fork_cleanup_fs:
1578         exit_fs(p); /* blocking */
1579 bad_fork_cleanup_files:
1580         exit_files(p); /* blocking */
1581 bad_fork_cleanup_semundo:
1582         exit_sem(p);
1583 bad_fork_cleanup_audit:
1584         audit_free(p);
1585 bad_fork_cleanup_perf:
1586         perf_event_free_task(p);
1587 bad_fork_cleanup_policy:
1588 #ifdef CONFIG_NUMA
1589         mpol_put(p->mempolicy);
1590 bad_fork_cleanup_threadgroup_lock:
1591 #endif
1592         if (clone_flags & CLONE_THREAD)
1593                 threadgroup_change_end(current);
1594         delayacct_tsk_free(p);
1595         module_put(task_thread_info(p)->exec_domain->module);
1596 bad_fork_cleanup_count:
1597         atomic_dec(&p->cred->user->processes);
1598         exit_creds(p);
1599 bad_fork_free:
1600         free_task(p);
1601 fork_out:
1602         return ERR_PTR(retval);
1603 }
1604
1605 static inline void init_idle_pids(struct pid_link *links)
1606 {
1607         enum pid_type type;
1608
1609         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1610                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1611                 links[type].pid = &init_struct_pid;
1612         }
1613 }
1614
1615 struct task_struct *fork_idle(int cpu)
1616 {
1617         struct task_struct *task;
1618         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1619         if (!IS_ERR(task)) {
1620                 init_idle_pids(task->pids);
1621                 init_idle(task, cpu);
1622         }
1623
1624         return task;
1625 }
1626
1627 /*
1628  *  Ok, this is the main fork-routine.
1629  *
1630  * It copies the process, and if successful kick-starts
1631  * it and waits for it to finish using the VM if required.
1632  */
1633 long do_fork(unsigned long clone_flags,
1634               unsigned long stack_start,
1635               unsigned long stack_size,
1636               int __user *parent_tidptr,
1637               int __user *child_tidptr)
1638 {
1639         struct task_struct *p;
1640         int trace = 0;
1641         long nr;
1642
1643         /*
1644          * Determine whether and which event to report to ptracer.  When
1645          * called from kernel_thread or CLONE_UNTRACED is explicitly
1646          * requested, no event is reported; otherwise, report if the event
1647          * for the type of forking is enabled.
1648          */
1649         if (!(clone_flags & CLONE_UNTRACED)) {
1650                 if (clone_flags & CLONE_VFORK)
1651                         trace = PTRACE_EVENT_VFORK;
1652                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1653                         trace = PTRACE_EVENT_CLONE;
1654                 else
1655                         trace = PTRACE_EVENT_FORK;
1656
1657                 if (likely(!ptrace_event_enabled(current, trace)))
1658                         trace = 0;
1659         }
1660
1661         p = copy_process(clone_flags, stack_start, stack_size,
1662                          child_tidptr, NULL, trace);
1663         /*
1664          * Do this prior waking up the new thread - the thread pointer
1665          * might get invalid after that point, if the thread exits quickly.
1666          */
1667         if (!IS_ERR(p)) {
1668                 struct completion vfork;
1669                 struct pid *pid;
1670
1671                 trace_sched_process_fork(current, p);
1672
1673                 pid = get_task_pid(p, PIDTYPE_PID);
1674                 nr = pid_vnr(pid);
1675
1676                 if (clone_flags & CLONE_PARENT_SETTID)
1677                         put_user(nr, parent_tidptr);
1678
1679                 if (clone_flags & CLONE_VFORK) {
1680                         p->vfork_done = &vfork;
1681                         init_completion(&vfork);
1682                         get_task_struct(p);
1683                 }
1684
1685                 wake_up_new_task(p);
1686
1687                 /* forking complete and child started to run, tell ptracer */
1688                 if (unlikely(trace))
1689                         ptrace_event_pid(trace, pid);
1690
1691                 if (clone_flags & CLONE_VFORK) {
1692                         if (!wait_for_vfork_done(p, &vfork))
1693                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1694                 }
1695
1696                 put_pid(pid);
1697         } else {
1698                 nr = PTR_ERR(p);
1699         }
1700         return nr;
1701 }
1702
1703 /*
1704  * Create a kernel thread.
1705  */
1706 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1707 {
1708         return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1709                 (unsigned long)arg, NULL, NULL);
1710 }
1711
1712 #ifdef __ARCH_WANT_SYS_FORK
1713 SYSCALL_DEFINE0(fork)
1714 {
1715 #ifdef CONFIG_MMU
1716         return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1717 #else
1718         /* can not support in nommu mode */
1719         return -EINVAL;
1720 #endif
1721 }
1722 #endif
1723
1724 #ifdef __ARCH_WANT_SYS_VFORK
1725 SYSCALL_DEFINE0(vfork)
1726 {
1727         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1728                         0, NULL, NULL);
1729 }
1730 #endif
1731
1732 #ifdef __ARCH_WANT_SYS_CLONE
1733 #ifdef CONFIG_CLONE_BACKWARDS
1734 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1735                  int __user *, parent_tidptr,
1736                  int, tls_val,
1737                  int __user *, child_tidptr)
1738 #elif defined(CONFIG_CLONE_BACKWARDS2)
1739 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1740                  int __user *, parent_tidptr,
1741                  int __user *, child_tidptr,
1742                  int, tls_val)
1743 #elif defined(CONFIG_CLONE_BACKWARDS3)
1744 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1745                 int, stack_size,
1746                 int __user *, parent_tidptr,
1747                 int __user *, child_tidptr,
1748                 int, tls_val)
1749 #else
1750 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1751                  int __user *, parent_tidptr,
1752                  int __user *, child_tidptr,
1753                  int, tls_val)
1754 #endif
1755 {
1756         return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1757 }
1758 #endif
1759
1760 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1761 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1762 #endif
1763
1764 static void sighand_ctor(void *data)
1765 {
1766         struct sighand_struct *sighand = data;
1767
1768         spin_lock_init(&sighand->siglock);
1769         init_waitqueue_head(&sighand->signalfd_wqh);
1770 }
1771
1772 void __init proc_caches_init(void)
1773 {
1774         sighand_cachep = kmem_cache_create("sighand_cache",
1775                         sizeof(struct sighand_struct), 0,
1776                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1777                         SLAB_NOTRACK, sighand_ctor);
1778         signal_cachep = kmem_cache_create("signal_cache",
1779                         sizeof(struct signal_struct), 0,
1780                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1781         files_cachep = kmem_cache_create("files_cache",
1782                         sizeof(struct files_struct), 0,
1783                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1784         fs_cachep = kmem_cache_create("fs_cache",
1785                         sizeof(struct fs_struct), 0,
1786                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1787         /*
1788          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1789          * whole struct cpumask for the OFFSTACK case. We could change
1790          * this to *only* allocate as much of it as required by the
1791          * maximum number of CPU's we can ever have.  The cpumask_allocation
1792          * is at the end of the structure, exactly for that reason.
1793          */
1794         mm_cachep = kmem_cache_create("mm_struct",
1795                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1796                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1797         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1798         mmap_init();
1799         nsproxy_cache_init();
1800 }
1801
1802 /*
1803  * Check constraints on flags passed to the unshare system call.
1804  */
1805 static int check_unshare_flags(unsigned long unshare_flags)
1806 {
1807         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1808                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1809                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1810                                 CLONE_NEWUSER|CLONE_NEWPID))
1811                 return -EINVAL;
1812         /*
1813          * Not implemented, but pretend it works if there is nothing to
1814          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1815          * needs to unshare vm.
1816          */
1817         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1818                 /* FIXME: get_task_mm() increments ->mm_users */
1819                 if (atomic_read(&current->mm->mm_users) > 1)
1820                         return -EINVAL;
1821         }
1822
1823         return 0;
1824 }
1825
1826 /*
1827  * Unshare the filesystem structure if it is being shared
1828  */
1829 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1830 {
1831         struct fs_struct *fs = current->fs;
1832
1833         if (!(unshare_flags & CLONE_FS) || !fs)
1834                 return 0;
1835
1836         /* don't need lock here; in the worst case we'll do useless copy */
1837         if (fs->users == 1)
1838                 return 0;
1839
1840         *new_fsp = copy_fs_struct(fs);
1841         if (!*new_fsp)
1842                 return -ENOMEM;
1843
1844         return 0;
1845 }
1846
1847 /*
1848  * Unshare file descriptor table if it is being shared
1849  */
1850 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1851 {
1852         struct files_struct *fd = current->files;
1853         int error = 0;
1854
1855         if ((unshare_flags & CLONE_FILES) &&
1856             (fd && atomic_read(&fd->count) > 1)) {
1857                 *new_fdp = dup_fd(fd, &error);
1858                 if (!*new_fdp)
1859                         return error;
1860         }
1861
1862         return 0;
1863 }
1864
1865 /*
1866  * unshare allows a process to 'unshare' part of the process
1867  * context which was originally shared using clone.  copy_*
1868  * functions used by do_fork() cannot be used here directly
1869  * because they modify an inactive task_struct that is being
1870  * constructed. Here we are modifying the current, active,
1871  * task_struct.
1872  */
1873 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1874 {
1875         struct fs_struct *fs, *new_fs = NULL;
1876         struct files_struct *fd, *new_fd = NULL;
1877         struct cred *new_cred = NULL;
1878         struct nsproxy *new_nsproxy = NULL;
1879         int do_sysvsem = 0;
1880         int err;
1881
1882         /*
1883          * If unsharing a user namespace must also unshare the thread.
1884          */
1885         if (unshare_flags & CLONE_NEWUSER)
1886                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1887         /*
1888          * If unsharing a thread from a thread group, must also unshare vm.
1889          */
1890         if (unshare_flags & CLONE_THREAD)
1891                 unshare_flags |= CLONE_VM;
1892         /*
1893          * If unsharing vm, must also unshare signal handlers.
1894          */
1895         if (unshare_flags & CLONE_VM)
1896                 unshare_flags |= CLONE_SIGHAND;
1897         /*
1898          * If unsharing namespace, must also unshare filesystem information.
1899          */
1900         if (unshare_flags & CLONE_NEWNS)
1901                 unshare_flags |= CLONE_FS;
1902
1903         err = check_unshare_flags(unshare_flags);
1904         if (err)
1905                 goto bad_unshare_out;
1906         /*
1907          * CLONE_NEWIPC must also detach from the undolist: after switching
1908          * to a new ipc namespace, the semaphore arrays from the old
1909          * namespace are unreachable.
1910          */
1911         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1912                 do_sysvsem = 1;
1913         err = unshare_fs(unshare_flags, &new_fs);
1914         if (err)
1915                 goto bad_unshare_out;
1916         err = unshare_fd(unshare_flags, &new_fd);
1917         if (err)
1918                 goto bad_unshare_cleanup_fs;
1919         err = unshare_userns(unshare_flags, &new_cred);
1920         if (err)
1921                 goto bad_unshare_cleanup_fd;
1922         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1923                                          new_cred, new_fs);
1924         if (err)
1925                 goto bad_unshare_cleanup_cred;
1926
1927         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1928                 if (do_sysvsem) {
1929                         /*
1930                          * CLONE_SYSVSEM is equivalent to sys_exit().
1931                          */
1932                         exit_sem(current);
1933                 }
1934                 if (unshare_flags & CLONE_NEWIPC) {
1935                         /* Orphan segments in old ns (see sem above). */
1936                         exit_shm(current);
1937                         shm_init_task(current);
1938                 }
1939
1940                 if (new_nsproxy)
1941                         switch_task_namespaces(current, new_nsproxy);
1942
1943                 task_lock(current);
1944
1945                 if (new_fs) {
1946                         fs = current->fs;
1947                         spin_lock(&fs->lock);
1948                         current->fs = new_fs;
1949                         if (--fs->users)
1950                                 new_fs = NULL;
1951                         else
1952                                 new_fs = fs;
1953                         spin_unlock(&fs->lock);
1954                 }
1955
1956                 if (new_fd) {
1957                         fd = current->files;
1958                         current->files = new_fd;
1959                         new_fd = fd;
1960                 }
1961
1962                 task_unlock(current);
1963
1964                 if (new_cred) {
1965                         /* Install the new user namespace */
1966                         commit_creds(new_cred);
1967                         new_cred = NULL;
1968                 }
1969         }
1970
1971 bad_unshare_cleanup_cred:
1972         if (new_cred)
1973                 put_cred(new_cred);
1974 bad_unshare_cleanup_fd:
1975         if (new_fd)
1976                 put_files_struct(new_fd);
1977
1978 bad_unshare_cleanup_fs:
1979         if (new_fs)
1980                 free_fs_struct(new_fs);
1981
1982 bad_unshare_out:
1983         return err;
1984 }
1985
1986 /*
1987  *      Helper to unshare the files of the current task.
1988  *      We don't want to expose copy_files internals to
1989  *      the exec layer of the kernel.
1990  */
1991
1992 int unshare_files(struct files_struct **displaced)
1993 {
1994         struct task_struct *task = current;
1995         struct files_struct *copy = NULL;
1996         int error;
1997
1998         error = unshare_fd(CLONE_FILES, &copy);
1999         if (error || !copy) {
2000                 *displaced = NULL;
2001                 return error;
2002         }
2003         *displaced = task->files;
2004         task_lock(task);
2005         task->files = copy;
2006         task_unlock(task);
2007         return 0;
2008 }