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 #include <linux/sysctl.h>
  78
  79 #include <asm/pgtable.h>
  80 #include <asm/pgalloc.h>
  81 #include <asm/uaccess.h>
  82 #include <asm/mmu_context.h>
  83 #include <asm/cacheflush.h>
  84 #include <asm/tlbflush.h>
  85
  86 #include <trace/events/sched.h>
  87
  88 #define CREATE_TRACE_POINTS
  89 #include <trace/events/task.h>
  90
  91 /*
  92  * Minimum number of threads to boot the kernel
  93  */
  94 #define MIN_THREADS 20
  95
  96 /*
  97  * Maximum number of threads
  98  */
  99 #define MAX_THREADS FUTEX_TID_MASK
 100
 101 /*
 102  * Protected counters by write_lock_irq(&tasklist_lock)
 103  */
 104 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 105 int nr_threads;                 /* The idle threads do not count.. */
 106
 107 int max_threads;                /* tunable limit on nr_threads */
 108
 109 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 110
 111 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 112
 113 #ifdef CONFIG_PROVE_RCU
 114 int lockdep_tasklist_lock_is_held(void)
 115 {
 116         return lockdep_is_held(&tasklist_lock);
 117 }
 118 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 119 #endif /* #ifdef CONFIG_PROVE_RCU */
 120
 121 int nr_processes(void)
 122 {
 123         int cpu;
 124         int total = 0;
 125
 126         for_each_possible_cpu(cpu)
 127                 total += per_cpu(process_counts, cpu);
 128
 129         return total;
 130 }
 131
 132 void __weak arch_release_task_struct(struct task_struct *tsk)
 133 {
 134 }
 135
 136 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 137 static struct kmem_cache *task_struct_cachep;
 138
 139 static inline struct task_struct *alloc_task_struct_node(int node)
 140 {
 141         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 142 }
 143
 144 static inline void free_task_struct(struct task_struct *tsk)
 145 {
 146         kmem_cache_free(task_struct_cachep, tsk);
 147 }
 148 #endif
 149
 150 void __weak arch_release_thread_info(struct thread_info *ti)
 151 {
 152 }
 153
 154 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
 155
 156 /*
 157  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 158  * kmemcache based allocator.
 159  */
 160 # if THREAD_SIZE >= PAGE_SIZE
 161 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 162                                                   int node)
 163 {
 164         struct page *page = alloc_kmem_pages_node(node, THREADINFO_GFP,
 165                                                   THREAD_SIZE_ORDER);
 166
 167         return page ? page_address(page) : NULL;
 168 }
 169
 170 static inline void free_thread_info(struct thread_info *ti)
 171 {
 172         free_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 173 }
 174 # else
 175 static struct kmem_cache *thread_info_cache;
 176
 177 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 178                                                   int node)
 179 {
 180         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
 181 }
 182
 183 static void free_thread_info(struct thread_info *ti)
 184 {
 185         kmem_cache_free(thread_info_cache, ti);
 186 }
 187
 188 void thread_info_cache_init(void)
 189 {
 190         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
 191                                               THREAD_SIZE, 0, NULL);
 192         BUG_ON(thread_info_cache == NULL);
 193 }
 194 # endif
 195 #endif
 196
 197 /* SLAB cache for signal_struct structures (tsk->signal) */
 198 static struct kmem_cache *signal_cachep;
 199
 200 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 201 struct kmem_cache *sighand_cachep;
 202
 203 /* SLAB cache for files_struct structures (tsk->files) */
 204 struct kmem_cache *files_cachep;
 205
 206 /* SLAB cache for fs_struct structures (tsk->fs) */
 207 struct kmem_cache *fs_cachep;
 208
 209 /* SLAB cache for vm_area_struct structures */
 210 struct kmem_cache *vm_area_cachep;
 211
 212 /* SLAB cache for mm_struct structures (tsk->mm) */
 213 static struct kmem_cache *mm_cachep;
 214
 215 static void account_kernel_stack(struct thread_info *ti, int account)
 216 {
 217         struct zone *zone = page_zone(virt_to_page(ti));
 218
 219         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 220 }
 221
 222 void free_task(struct task_struct *tsk)
 223 {
 224         account_kernel_stack(tsk->stack, -1);
 225         arch_release_thread_info(tsk->stack);
 226         free_thread_info(tsk->stack);
 227         rt_mutex_debug_task_free(tsk);
 228         ftrace_graph_exit_task(tsk);
 229         put_seccomp_filter(tsk);
 230         arch_release_task_struct(tsk);
 231         free_task_struct(tsk);
 232 }
 233 EXPORT_SYMBOL(free_task);
 234
 235 static inline void free_signal_struct(struct signal_struct *sig)
 236 {
 237         taskstats_tgid_free(sig);
 238         sched_autogroup_exit(sig);
 239         kmem_cache_free(signal_cachep, sig);
 240 }
 241
 242 static inline void put_signal_struct(struct signal_struct *sig)
 243 {
 244         if (atomic_dec_and_test(&sig->sigcnt))
 245                 free_signal_struct(sig);
 246 }
 247
 248 void __put_task_struct(struct task_struct *tsk)
 249 {
 250         WARN_ON(!tsk->exit_state);
 251         WARN_ON(atomic_read(&tsk->usage));
 252         WARN_ON(tsk == current);
 253
 254         task_numa_free(tsk);
 255         security_task_free(tsk);
 256         exit_creds(tsk);
 257         delayacct_tsk_free(tsk);
 258         put_signal_struct(tsk->signal);
 259
 260         if (!profile_handoff_task(tsk))
 261                 free_task(tsk);
 262 }
 263 EXPORT_SYMBOL_GPL(__put_task_struct);
 264
 265 void __init __weak arch_task_cache_init(void) { }
 266
 267 /*
 268  * set_max_threads
 269  */
 270 static void set_max_threads(unsigned int max_threads_suggested)
 271 {
 272         u64 threads;
 273
 274         /*
 275          * The number of threads shall be limited such that the thread
 276          * structures may only consume a small part of the available memory.
 277          */
 278         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 279                 threads = MAX_THREADS;
 280         else
 281                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 282                                     (u64) THREAD_SIZE * 8UL);
 283
 284         if (threads > max_threads_suggested)
 285                 threads = max_threads_suggested;
 286
 287         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 288 }
 289
 290 void __init fork_init(void)
 291 {
 292 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 293 #ifndef ARCH_MIN_TASKALIGN
 294 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 295 #endif
 296         /* create a slab on which task_structs can be allocated */
 297         task_struct_cachep =
 298                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 299                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 300 #endif
 301
 302         /* do the arch specific task caches init */
 303         arch_task_cache_init();
 304
 305         set_max_threads(MAX_THREADS);
 306
 307         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 308         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 309         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 310                 init_task.signal->rlim[RLIMIT_NPROC];
 311 }
 312
 313 int __weak arch_dup_task_struct(struct task_struct *dst,
 314                                                struct task_struct *src)
 315 {
 316         *dst = *src;
 317         return 0;
 318 }
 319
 320 void set_task_stack_end_magic(struct task_struct *tsk)
 321 {
 322         unsigned long *stackend;
 323
 324         stackend = end_of_stack(tsk);
 325         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 326 }
 327
 328 static struct task_struct *dup_task_struct(struct task_struct *orig)
 329 {
 330         struct task_struct *tsk;
 331         struct thread_info *ti;
 332         int node = tsk_fork_get_node(orig);
 333         int err;
 334
 335         tsk = alloc_task_struct_node(node);
 336         if (!tsk)
 337                 return NULL;
 338
 339         ti = alloc_thread_info_node(tsk, node);
 340         if (!ti)
 341                 goto free_tsk;
 342
 343         err = arch_dup_task_struct(tsk, orig);
 344         if (err)
 345                 goto free_ti;
 346
 347         tsk->stack = ti;
 348 #ifdef CONFIG_SECCOMP
 349         /*
 350          * We must handle setting up seccomp filters once we're under
 351          * the sighand lock in case orig has changed between now and
 352          * then. Until then, filter must be NULL to avoid messing up
 353          * the usage counts on the error path calling free_task.
 354          */
 355         tsk->seccomp.filter = NULL;
 356 #endif
 357
 358         setup_thread_stack(tsk, orig);
 359         clear_user_return_notifier(tsk);
 360         clear_tsk_need_resched(tsk);
 361         set_task_stack_end_magic(tsk);
 362
 363 #ifdef CONFIG_CC_STACKPROTECTOR
 364         tsk->stack_canary = get_random_int();
 365 #endif
 366
 367         /*
 368          * One for us, one for whoever does the "release_task()" (usually
 369          * parent)
 370          */
 371         atomic_set(&tsk->usage, 2);
 372 #ifdef CONFIG_BLK_DEV_IO_TRACE
 373         tsk->btrace_seq = 0;
 374 #endif
 375         tsk->splice_pipe = NULL;
 376         tsk->task_frag.page = NULL;
 377
 378         account_kernel_stack(ti, 1);
 379
 380         return tsk;
 381
 382 free_ti:
 383         free_thread_info(ti);
 384 free_tsk:
 385         free_task_struct(tsk);
 386         return NULL;
 387 }
 388
 389 #ifdef CONFIG_MMU
 390 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 391 {
 392         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 393         struct rb_node **rb_link, *rb_parent;
 394         int retval;
 395         unsigned long charge;
 396
 397         uprobe_start_dup_mmap();
 398         down_write(&oldmm->mmap_sem);
 399         flush_cache_dup_mm(oldmm);
 400         uprobe_dup_mmap(oldmm, mm);
 401         /*
 402          * Not linked in yet - no deadlock potential:
 403          */
 404         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 405
 406         mm->total_vm = oldmm->total_vm;
 407         mm->shared_vm = oldmm->shared_vm;
 408         mm->exec_vm = oldmm->exec_vm;
 409         mm->stack_vm = oldmm->stack_vm;
 410
 411         rb_link = &mm->mm_rb.rb_node;
 412         rb_parent = NULL;
 413         pprev = &mm->mmap;
 414         retval = ksm_fork(mm, oldmm);
 415         if (retval)
 416                 goto out;
 417         retval = khugepaged_fork(mm, oldmm);
 418         if (retval)
 419                 goto out;
 420
 421         prev = NULL;
 422         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 423                 struct file *file;
 424
 425                 if (mpnt->vm_flags & VM_DONTCOPY) {
 426                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 427                                                         -vma_pages(mpnt));
 428                         continue;
 429                 }
 430                 charge = 0;
 431                 if (mpnt->vm_flags & VM_ACCOUNT) {
 432                         unsigned long len = vma_pages(mpnt);
 433
 434                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 435                                 goto fail_nomem;
 436                         charge = len;
 437                 }
 438                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 439                 if (!tmp)
 440                         goto fail_nomem;
 441                 *tmp = *mpnt;
 442                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 443                 retval = vma_dup_policy(mpnt, tmp);
 444                 if (retval)
 445                         goto fail_nomem_policy;
 446                 tmp->vm_mm = mm;
 447                 if (anon_vma_fork(tmp, mpnt))
 448                         goto fail_nomem_anon_vma_fork;
 449                 tmp->vm_flags &= ~VM_LOCKED;
 450                 tmp->vm_next = tmp->vm_prev = NULL;
 451                 file = tmp->vm_file;
 452                 if (file) {
 453                         struct inode *inode = file_inode(file);
 454                         struct address_space *mapping = file->f_mapping;
 455
 456                         get_file(file);
 457                         if (tmp->vm_flags & VM_DENYWRITE)
 458                                 atomic_dec(&inode->i_writecount);
 459                         i_mmap_lock_write(mapping);
 460                         if (tmp->vm_flags & VM_SHARED)
 461                                 atomic_inc(&mapping->i_mmap_writable);
 462                         flush_dcache_mmap_lock(mapping);
 463                         /* insert tmp into the share list, just after mpnt */
 464                         vma_interval_tree_insert_after(tmp, mpnt,
 465                                         &mapping->i_mmap);
 466                         flush_dcache_mmap_unlock(mapping);
 467                         i_mmap_unlock_write(mapping);
 468                 }
 469
 470                 /*
 471                  * Clear hugetlb-related page reserves for children. This only
 472                  * affects MAP_PRIVATE mappings. Faults generated by the child
 473                  * are not guaranteed to succeed, even if read-only
 474                  */
 475                 if (is_vm_hugetlb_page(tmp))
 476                         reset_vma_resv_huge_pages(tmp);
 477
 478                 /*
 479                  * Link in the new vma and copy the page table entries.
 480                  */
 481                 *pprev = tmp;
 482                 pprev = &tmp->vm_next;
 483                 tmp->vm_prev = prev;
 484                 prev = tmp;
 485
 486                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 487                 rb_link = &tmp->vm_rb.rb_right;
 488                 rb_parent = &tmp->vm_rb;
 489
 490                 mm->map_count++;
 491                 retval = copy_page_range(mm, oldmm, mpnt);
 492
 493                 if (tmp->vm_ops && tmp->vm_ops->open)
 494                         tmp->vm_ops->open(tmp);
 495
 496                 if (retval)
 497                         goto out;
 498         }
 499         /* a new mm has just been created */
 500         arch_dup_mmap(oldmm, mm);
 501         retval = 0;
 502 out:
 503         up_write(&mm->mmap_sem);
 504         flush_tlb_mm(oldmm);
 505         up_write(&oldmm->mmap_sem);
 506         uprobe_end_dup_mmap();
 507         return retval;
 508 fail_nomem_anon_vma_fork:
 509         mpol_put(vma_policy(tmp));
 510 fail_nomem_policy:
 511         kmem_cache_free(vm_area_cachep, tmp);
 512 fail_nomem:
 513         retval = -ENOMEM;
 514         vm_unacct_memory(charge);
 515         goto out;
 516 }
 517
 518 static inline int mm_alloc_pgd(struct mm_struct *mm)
 519 {
 520         mm->pgd = pgd_alloc(mm);
 521         if (unlikely(!mm->pgd))
 522                 return -ENOMEM;
 523         return 0;
 524 }
 525
 526 static inline void mm_free_pgd(struct mm_struct *mm)
 527 {
 528         pgd_free(mm, mm->pgd);
 529 }
 530 #else
 531 #define dup_mmap(mm, oldmm)     (0)
 532 #define mm_alloc_pgd(mm)        (0)
 533 #define mm_free_pgd(mm)
 534 #endif /* CONFIG_MMU */
 535
 536 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 537
 538 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 539 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 540
 541 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 542
 543 static int __init coredump_filter_setup(char *s)
 544 {
 545         default_dump_filter =
 546                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 547                 MMF_DUMP_FILTER_MASK;
 548         return 1;
 549 }
 550
 551 __setup("coredump_filter=", coredump_filter_setup);
 552
 553 #include <linux/init_task.h>
 554
 555 static void mm_init_aio(struct mm_struct *mm)
 556 {
 557 #ifdef CONFIG_AIO
 558         spin_lock_init(&mm->ioctx_lock);
 559         mm->ioctx_table = NULL;
 560 #endif
 561 }
 562
 563 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 564 {
 565 #ifdef CONFIG_MEMCG
 566         mm->owner = p;
 567 #endif
 568 }
 569
 570 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 571 {
 572         mm->mmap = NULL;
 573         mm->mm_rb = RB_ROOT;
 574         mm->vmacache_seqnum = 0;
 575         atomic_set(&mm->mm_users, 1);
 576         atomic_set(&mm->mm_count, 1);
 577         init_rwsem(&mm->mmap_sem);
 578         INIT_LIST_HEAD(&mm->mmlist);
 579         mm->core_state = NULL;
 580         atomic_long_set(&mm->nr_ptes, 0);
 581         mm_nr_pmds_init(mm);
 582         mm->map_count = 0;
 583         mm->locked_vm = 0;
 584         mm->pinned_vm = 0;
 585         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 586         spin_lock_init(&mm->page_table_lock);
 587         mm_init_cpumask(mm);
 588         mm_init_aio(mm);
 589         mm_init_owner(mm, p);
 590         mmu_notifier_mm_init(mm);
 591         clear_tlb_flush_pending(mm);
 592 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 593         mm->pmd_huge_pte = NULL;
 594 #endif
 595
 596         if (current->mm) {
 597                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 598                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 599         } else {
 600                 mm->flags = default_dump_filter;
 601                 mm->def_flags = 0;
 602         }
 603
 604         if (mm_alloc_pgd(mm))
 605                 goto fail_nopgd;
 606
 607         if (init_new_context(p, mm))
 608                 goto fail_nocontext;
 609
 610         return mm;
 611
 612 fail_nocontext:
 613         mm_free_pgd(mm);
 614 fail_nopgd:
 615         free_mm(mm);
 616         return NULL;
 617 }
 618
 619 static void check_mm(struct mm_struct *mm)
 620 {
 621         int i;
 622
 623         for (i = 0; i < NR_MM_COUNTERS; i++) {
 624                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 625
 626                 if (unlikely(x))
 627                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 628                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 629         }
 630
 631         if (atomic_long_read(&mm->nr_ptes))
 632                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 633                                 atomic_long_read(&mm->nr_ptes));
 634         if (mm_nr_pmds(mm))
 635                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 636                                 mm_nr_pmds(mm));
 637
 638 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 639         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 640 #endif
 641 }
 642
 643 /*
 644  * Allocate and initialize an mm_struct.
 645  */
 646 struct mm_struct *mm_alloc(void)
 647 {
 648         struct mm_struct *mm;
 649
 650         mm = allocate_mm();
 651         if (!mm)
 652                 return NULL;
 653
 654         memset(mm, 0, sizeof(*mm));
 655         return mm_init(mm, current);
 656 }
 657
 658 /*
 659  * Called when the last reference to the mm
 660  * is dropped: either by a lazy thread or by
 661  * mmput. Free the page directory and the mm.
 662  */
 663 void __mmdrop(struct mm_struct *mm)
 664 {
 665         BUG_ON(mm == &init_mm);
 666         mm_free_pgd(mm);
 667         destroy_context(mm);
 668         mmu_notifier_mm_destroy(mm);
 669         check_mm(mm);
 670         free_mm(mm);
 671 }
 672 EXPORT_SYMBOL_GPL(__mmdrop);
 673
 674 /*
 675  * Decrement the use count and release all resources for an mm.
 676  */
 677 void mmput(struct mm_struct *mm)
 678 {
 679         might_sleep();
 680
 681         if (atomic_dec_and_test(&mm->mm_users)) {
 682                 uprobe_clear_state(mm);
 683                 exit_aio(mm);
 684                 ksm_exit(mm);
 685                 khugepaged_exit(mm); /* must run before exit_mmap */
 686                 exit_mmap(mm);
 687                 set_mm_exe_file(mm, NULL);
 688                 if (!list_empty(&mm->mmlist)) {
 689                         spin_lock(&mmlist_lock);
 690                         list_del(&mm->mmlist);
 691                         spin_unlock(&mmlist_lock);
 692                 }
 693                 if (mm->binfmt)
 694                         module_put(mm->binfmt->module);
 695                 mmdrop(mm);
 696         }
 697 }
 698 EXPORT_SYMBOL_GPL(mmput);
 699
 700 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 701 {
 702         if (new_exe_file)
 703                 get_file(new_exe_file);
 704         if (mm->exe_file)
 705                 fput(mm->exe_file);
 706         mm->exe_file = new_exe_file;
 707 }
 708
 709 struct file *get_mm_exe_file(struct mm_struct *mm)
 710 {
 711         struct file *exe_file;
 712
 713         /* We need mmap_sem to protect against races with removal of exe_file */
 714         down_read(&mm->mmap_sem);
 715         exe_file = mm->exe_file;
 716         if (exe_file)
 717                 get_file(exe_file);
 718         up_read(&mm->mmap_sem);
 719         return exe_file;
 720 }
 721
 722 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 723 {
 724         /* It's safe to write the exe_file pointer without exe_file_lock because
 725          * this is called during fork when the task is not yet in /proc */
 726         newmm->exe_file = get_mm_exe_file(oldmm);
 727 }
 728
 729 /**
 730  * get_task_mm - acquire a reference to the task's mm
 731  *
 732  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 733  * this kernel workthread has transiently adopted a user mm with use_mm,
 734  * to do its AIO) is not set and if so returns a reference to it, after
 735  * bumping up the use count.  User must release the mm via mmput()
 736  * after use.  Typically used by /proc and ptrace.
 737  */
 738 struct mm_struct *get_task_mm(struct task_struct *task)
 739 {
 740         struct mm_struct *mm;
 741
 742         task_lock(task);
 743         mm = task->mm;
 744         if (mm) {
 745                 if (task->flags & PF_KTHREAD)
 746                         mm = NULL;
 747                 else
 748                         atomic_inc(&mm->mm_users);
 749         }
 750         task_unlock(task);
 751         return mm;
 752 }
 753 EXPORT_SYMBOL_GPL(get_task_mm);
 754
 755 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 756 {
 757         struct mm_struct *mm;
 758         int err;
 759
 760         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 761         if (err)
 762                 return ERR_PTR(err);
 763
 764         mm = get_task_mm(task);
 765         if (mm && mm != current->mm &&
 766                         !ptrace_may_access(task, mode)) {
 767                 mmput(mm);
 768                 mm = ERR_PTR(-EACCES);
 769         }
 770         mutex_unlock(&task->signal->cred_guard_mutex);
 771
 772         return mm;
 773 }
 774
 775 static void complete_vfork_done(struct task_struct *tsk)
 776 {
 777         struct completion *vfork;
 778
 779         task_lock(tsk);
 780         vfork = tsk->vfork_done;
 781         if (likely(vfork)) {
 782                 tsk->vfork_done = NULL;
 783                 complete(vfork);
 784         }
 785         task_unlock(tsk);
 786 }
 787
 788 static int wait_for_vfork_done(struct task_struct *child,
 789                                 struct completion *vfork)
 790 {
 791         int killed;
 792
 793         freezer_do_not_count();
 794         killed = wait_for_completion_killable(vfork);
 795         freezer_count();
 796
 797         if (killed) {
 798                 task_lock(child);
 799                 child->vfork_done = NULL;
 800                 task_unlock(child);
 801         }
 802
 803         put_task_struct(child);
 804         return killed;
 805 }
 806
 807 /* Please note the differences between mmput and mm_release.
 808  * mmput is called whenever we stop holding onto a mm_struct,
 809  * error success whatever.
 810  *
 811  * mm_release is called after a mm_struct has been removed
 812  * from the current process.
 813  *
 814  * This difference is important for error handling, when we
 815  * only half set up a mm_struct for a new process and need to restore
 816  * the old one.  Because we mmput the new mm_struct before
 817  * restoring the old one. . .
 818  * Eric Biederman 10 January 1998
 819  */
 820 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 821 {
 822         /* Get rid of any futexes when releasing the mm */
 823 #ifdef CONFIG_FUTEX
 824         if (unlikely(tsk->robust_list)) {
 825                 exit_robust_list(tsk);
 826                 tsk->robust_list = NULL;
 827         }
 828 #ifdef CONFIG_COMPAT
 829         if (unlikely(tsk->compat_robust_list)) {
 830                 compat_exit_robust_list(tsk);
 831                 tsk->compat_robust_list = NULL;
 832         }
 833 #endif
 834         if (unlikely(!list_empty(&tsk->pi_state_list)))
 835                 exit_pi_state_list(tsk);
 836 #endif
 837
 838         uprobe_free_utask(tsk);
 839
 840         /* Get rid of any cached register state */
 841         deactivate_mm(tsk, mm);
 842
 843         /*
 844          * If we're exiting normally, clear a user-space tid field if
 845          * requested.  We leave this alone when dying by signal, to leave
 846          * the value intact in a core dump, and to save the unnecessary
 847          * trouble, say, a killed vfork parent shouldn't touch this mm.
 848          * Userland only wants this done for a sys_exit.
 849          */
 850         if (tsk->clear_child_tid) {
 851                 if (!(tsk->flags & PF_SIGNALED) &&
 852                     atomic_read(&mm->mm_users) > 1) {
 853                         /*
 854                          * We don't check the error code - if userspace has
 855                          * not set up a proper pointer then tough luck.
 856                          */
 857                         put_user(0, tsk->clear_child_tid);
 858                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 859                                         1, NULL, NULL, 0);
 860                 }
 861                 tsk->clear_child_tid = NULL;
 862         }
 863
 864         /*
 865          * All done, finally we can wake up parent and return this mm to him.
 866          * Also kthread_stop() uses this completion for synchronization.
 867          */
 868         if (tsk->vfork_done)
 869                 complete_vfork_done(tsk);
 870 }
 871
 872 /*
 873  * Allocate a new mm structure and copy contents from the
 874  * mm structure of the passed in task structure.
 875  */
 876 static struct mm_struct *dup_mm(struct task_struct *tsk)
 877 {
 878         struct mm_struct *mm, *oldmm = current->mm;
 879         int err;
 880
 881         mm = allocate_mm();
 882         if (!mm)
 883                 goto fail_nomem;
 884
 885         memcpy(mm, oldmm, sizeof(*mm));
 886
 887         if (!mm_init(mm, tsk))
 888                 goto fail_nomem;
 889
 890         dup_mm_exe_file(oldmm, mm);
 891
 892         err = dup_mmap(mm, oldmm);
 893         if (err)
 894                 goto free_pt;
 895
 896         mm->hiwater_rss = get_mm_rss(mm);
 897         mm->hiwater_vm = mm->total_vm;
 898
 899         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 900                 goto free_pt;
 901
 902         return mm;
 903
 904 free_pt:
 905         /* don't put binfmt in mmput, we haven't got module yet */
 906         mm->binfmt = NULL;
 907         mmput(mm);
 908
 909 fail_nomem:
 910         return NULL;
 911 }
 912
 913 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 914 {
 915         struct mm_struct *mm, *oldmm;
 916         int retval;
 917
 918         tsk->min_flt = tsk->maj_flt = 0;
 919         tsk->nvcsw = tsk->nivcsw = 0;
 920 #ifdef CONFIG_DETECT_HUNG_TASK
 921         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 922 #endif
 923
 924         tsk->mm = NULL;
 925         tsk->active_mm = NULL;
 926
 927         /*
 928          * Are we cloning a kernel thread?
 929          *
 930          * We need to steal a active VM for that..
 931          */
 932         oldmm = current->mm;
 933         if (!oldmm)
 934                 return 0;
 935
 936         /* initialize the new vmacache entries */
 937         vmacache_flush(tsk);
 938
 939         if (clone_flags & CLONE_VM) {
 940                 atomic_inc(&oldmm->mm_users);
 941                 mm = oldmm;
 942                 goto good_mm;
 943         }
 944
 945         retval = -ENOMEM;
 946         mm = dup_mm(tsk);
 947         if (!mm)
 948                 goto fail_nomem;
 949
 950 good_mm:
 951         tsk->mm = mm;
 952         tsk->active_mm = mm;
 953         return 0;
 954
 955 fail_nomem:
 956         return retval;
 957 }
 958
 959 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 960 {
 961         struct fs_struct *fs = current->fs;
 962         if (clone_flags & CLONE_FS) {
 963                 /* tsk->fs is already what we want */
 964                 spin_lock(&fs->lock);
 965                 if (fs->in_exec) {
 966                         spin_unlock(&fs->lock);
 967                         return -EAGAIN;
 968                 }
 969                 fs->users++;
 970                 spin_unlock(&fs->lock);
 971                 return 0;
 972         }
 973         tsk->fs = copy_fs_struct(fs);
 974         if (!tsk->fs)
 975                 return -ENOMEM;
 976         return 0;
 977 }
 978
 979 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 980 {
 981         struct files_struct *oldf, *newf;
 982         int error = 0;
 983
 984         /*
 985          * A background process may not have any files ...
 986          */
 987         oldf = current->files;
 988         if (!oldf)
 989                 goto out;
 990
 991         if (clone_flags & CLONE_FILES) {
 992                 atomic_inc(&oldf->count);
 993                 goto out;
 994         }
 995
 996         newf = dup_fd(oldf, &error);
 997         if (!newf)
 998                 goto out;
 999
1000         tsk->files = newf;
1001         error = 0;
1002 out:
1003         return error;
1004 }
1005
1006 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1007 {
1008 #ifdef CONFIG_BLOCK
1009         struct io_context *ioc = current->io_context;
1010         struct io_context *new_ioc;
1011
1012         if (!ioc)
1013                 return 0;
1014         /*
1015          * Share io context with parent, if CLONE_IO is set
1016          */
1017         if (clone_flags & CLONE_IO) {
1018                 ioc_task_link(ioc);
1019                 tsk->io_context = ioc;
1020         } else if (ioprio_valid(ioc->ioprio)) {
1021                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1022                 if (unlikely(!new_ioc))
1023                         return -ENOMEM;
1024
1025                 new_ioc->ioprio = ioc->ioprio;
1026                 put_io_context(new_ioc);
1027         }
1028 #endif
1029         return 0;
1030 }
1031
1032 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1033 {
1034         struct sighand_struct *sig;
1035
1036         if (clone_flags & CLONE_SIGHAND) {
1037                 atomic_inc(&current->sighand->count);
1038                 return 0;
1039         }
1040         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1041         rcu_assign_pointer(tsk->sighand, sig);
1042         if (!sig)
1043                 return -ENOMEM;
1044         atomic_set(&sig->count, 1);
1045         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1046         return 0;
1047 }
1048
1049 void __cleanup_sighand(struct sighand_struct *sighand)
1050 {
1051         if (atomic_dec_and_test(&sighand->count)) {
1052                 signalfd_cleanup(sighand);
1053                 /*
1054                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1055                  * without an RCU grace period, see __lock_task_sighand().
1056                  */
1057                 kmem_cache_free(sighand_cachep, sighand);
1058         }
1059 }
1060
1061 /*
1062  * Initialize POSIX timer handling for a thread group.
1063  */
1064 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1065 {
1066         unsigned long cpu_limit;
1067
1068         /* Thread group counters. */
1069         thread_group_cputime_init(sig);
1070
1071         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1072         if (cpu_limit != RLIM_INFINITY) {
1073                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1074                 sig->cputimer.running = 1;
1075         }
1076
1077         /* The timer lists. */
1078         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1079         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1080         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1081 }
1082
1083 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1084 {
1085         struct signal_struct *sig;
1086
1087         if (clone_flags & CLONE_THREAD)
1088                 return 0;
1089
1090         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1091         tsk->signal = sig;
1092         if (!sig)
1093                 return -ENOMEM;
1094
1095         sig->nr_threads = 1;
1096         atomic_set(&sig->live, 1);
1097         atomic_set(&sig->sigcnt, 1);
1098
1099         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1100         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1101         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1102
1103         init_waitqueue_head(&sig->wait_chldexit);
1104         sig->curr_target = tsk;
1105         init_sigpending(&sig->shared_pending);
1106         INIT_LIST_HEAD(&sig->posix_timers);
1107         seqlock_init(&sig->stats_lock);
1108
1109         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1110         sig->real_timer.function = it_real_fn;
1111
1112         task_lock(current->group_leader);
1113         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1114         task_unlock(current->group_leader);
1115
1116         posix_cpu_timers_init_group(sig);
1117
1118         tty_audit_fork(sig);
1119         sched_autogroup_fork(sig);
1120
1121 #ifdef CONFIG_CGROUPS
1122         init_rwsem(&sig->group_rwsem);
1123 #endif
1124
1125         sig->oom_score_adj = current->signal->oom_score_adj;
1126         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1127
1128         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1129                                    current->signal->is_child_subreaper;
1130
1131         mutex_init(&sig->cred_guard_mutex);
1132
1133         return 0;
1134 }
1135
1136 static void copy_seccomp(struct task_struct *p)
1137 {
1138 #ifdef CONFIG_SECCOMP
1139         /*
1140          * Must be called with sighand->lock held, which is common to
1141          * all threads in the group. Holding cred_guard_mutex is not
1142          * needed because this new task is not yet running and cannot
1143          * be racing exec.
1144          */
1145         assert_spin_locked(&current->sighand->siglock);
1146
1147         /* Ref-count the new filter user, and assign it. */
1148         get_seccomp_filter(current);
1149         p->seccomp = current->seccomp;
1150
1151         /*
1152          * Explicitly enable no_new_privs here in case it got set
1153          * between the task_struct being duplicated and holding the
1154          * sighand lock. The seccomp state and nnp must be in sync.
1155          */
1156         if (task_no_new_privs(current))
1157                 task_set_no_new_privs(p);
1158
1159         /*
1160          * If the parent gained a seccomp mode after copying thread
1161          * flags and between before we held the sighand lock, we have
1162          * to manually enable the seccomp thread flag here.
1163          */
1164         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1165                 set_tsk_thread_flag(p, TIF_SECCOMP);
1166 #endif
1167 }
1168
1169 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1170 {
1171         current->clear_child_tid = tidptr;
1172
1173         return task_pid_vnr(current);
1174 }
1175
1176 static void rt_mutex_init_task(struct task_struct *p)
1177 {
1178         raw_spin_lock_init(&p->pi_lock);
1179 #ifdef CONFIG_RT_MUTEXES
1180         p->pi_waiters = RB_ROOT;
1181         p->pi_waiters_leftmost = NULL;
1182         p->pi_blocked_on = NULL;
1183 #endif
1184 }
1185
1186 /*
1187  * Initialize POSIX timer handling for a single task.
1188  */
1189 static void posix_cpu_timers_init(struct task_struct *tsk)
1190 {
1191         tsk->cputime_expires.prof_exp = 0;
1192         tsk->cputime_expires.virt_exp = 0;
1193         tsk->cputime_expires.sched_exp = 0;
1194         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1195         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1196         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1197 }
1198
1199 static inline void
1200 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1201 {
1202          task->pids[type].pid = pid;
1203 }
1204
1205 /*
1206  * This creates a new process as a copy of the old one,
1207  * but does not actually start it yet.
1208  *
1209  * It copies the registers, and all the appropriate
1210  * parts of the process environment (as per the clone
1211  * flags). The actual kick-off is left to the caller.
1212  */
1213 static struct task_struct *copy_process(unsigned long clone_flags,
1214                                         unsigned long stack_start,
1215                                         unsigned long stack_size,
1216                                         int __user *child_tidptr,
1217                                         struct pid *pid,
1218                                         int trace)
1219 {
1220         int retval;
1221         struct task_struct *p;
1222
1223         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1224                 return ERR_PTR(-EINVAL);
1225
1226         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1227                 return ERR_PTR(-EINVAL);
1228
1229         /*
1230          * Thread groups must share signals as well, and detached threads
1231          * can only be started up within the thread group.
1232          */
1233         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1234                 return ERR_PTR(-EINVAL);
1235
1236         /*
1237          * Shared signal handlers imply shared VM. By way of the above,
1238          * thread groups also imply shared VM. Blocking this case allows
1239          * for various simplifications in other code.
1240          */
1241         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1242                 return ERR_PTR(-EINVAL);
1243
1244         /*
1245          * Siblings of global init remain as zombies on exit since they are
1246          * not reaped by their parent (swapper). To solve this and to avoid
1247          * multi-rooted process trees, prevent global and container-inits
1248          * from creating siblings.
1249          */
1250         if ((clone_flags & CLONE_PARENT) &&
1251                                 current->signal->flags & SIGNAL_UNKILLABLE)
1252                 return ERR_PTR(-EINVAL);
1253
1254         /*
1255          * If the new process will be in a different pid or user namespace
1256          * do not allow it to share a thread group or signal handlers or
1257          * parent with the forking task.
1258          */
1259         if (clone_flags & CLONE_SIGHAND) {
1260                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1261                     (task_active_pid_ns(current) !=
1262                                 current->nsproxy->pid_ns_for_children))
1263                         return ERR_PTR(-EINVAL);
1264         }
1265
1266         retval = security_task_create(clone_flags);
1267         if (retval)
1268                 goto fork_out;
1269
1270         retval = -ENOMEM;
1271         p = dup_task_struct(current);
1272         if (!p)
1273                 goto fork_out;
1274
1275         ftrace_graph_init_task(p);
1276
1277         rt_mutex_init_task(p);
1278
1279 #ifdef CONFIG_PROVE_LOCKING
1280         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1281         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1282 #endif
1283         retval = -EAGAIN;
1284         if (atomic_read(&p->real_cred->user->processes) >=
1285                         task_rlimit(p, RLIMIT_NPROC)) {
1286                 if (p->real_cred->user != INIT_USER &&
1287                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1288                         goto bad_fork_free;
1289         }
1290         current->flags &= ~PF_NPROC_EXCEEDED;
1291
1292         retval = copy_creds(p, clone_flags);
1293         if (retval < 0)
1294                 goto bad_fork_free;
1295
1296         /*
1297          * If multiple threads are within copy_process(), then this check
1298          * triggers too late. This doesn't hurt, the check is only there
1299          * to stop root fork bombs.
1300          */
1301         retval = -EAGAIN;
1302         if (nr_threads >= max_threads)
1303                 goto bad_fork_cleanup_count;
1304
1305         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1306         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1307         p->flags |= PF_FORKNOEXEC;
1308         INIT_LIST_HEAD(&p->children);
1309         INIT_LIST_HEAD(&p->sibling);
1310         rcu_copy_process(p);
1311         p->vfork_done = NULL;
1312         spin_lock_init(&p->alloc_lock);
1313
1314         init_sigpending(&p->pending);
1315
1316         p->utime = p->stime = p->gtime = 0;
1317         p->utimescaled = p->stimescaled = 0;
1318 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1319         p->prev_cputime.utime = p->prev_cputime.stime = 0;
1320 #endif
1321 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1322         seqlock_init(&p->vtime_seqlock);
1323         p->vtime_snap = 0;
1324         p->vtime_snap_whence = VTIME_SLEEPING;
1325 #endif
1326
1327 #if defined(SPLIT_RSS_COUNTING)
1328         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1329 #endif
1330
1331         p->default_timer_slack_ns = current->timer_slack_ns;
1332
1333         task_io_accounting_init(&p->ioac);
1334         acct_clear_integrals(p);
1335
1336         posix_cpu_timers_init(p);
1337
1338         p->start_time = ktime_get_ns();
1339         p->real_start_time = ktime_get_boot_ns();
1340         p->io_context = NULL;
1341         p->audit_context = NULL;
1342         if (clone_flags & CLONE_THREAD)
1343                 threadgroup_change_begin(current);
1344         cgroup_fork(p);
1345 #ifdef CONFIG_NUMA
1346         p->mempolicy = mpol_dup(p->mempolicy);
1347         if (IS_ERR(p->mempolicy)) {
1348                 retval = PTR_ERR(p->mempolicy);
1349                 p->mempolicy = NULL;
1350                 goto bad_fork_cleanup_threadgroup_lock;
1351         }
1352 #endif
1353 #ifdef CONFIG_CPUSETS
1354         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1355         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1356         seqcount_init(&p->mems_allowed_seq);
1357 #endif
1358 #ifdef CONFIG_TRACE_IRQFLAGS
1359         p->irq_events = 0;
1360         p->hardirqs_enabled = 0;
1361         p->hardirq_enable_ip = 0;
1362         p->hardirq_enable_event = 0;
1363         p->hardirq_disable_ip = _THIS_IP_;
1364         p->hardirq_disable_event = 0;
1365         p->softirqs_enabled = 1;
1366         p->softirq_enable_ip = _THIS_IP_;
1367         p->softirq_enable_event = 0;
1368         p->softirq_disable_ip = 0;
1369         p->softirq_disable_event = 0;
1370         p->hardirq_context = 0;
1371         p->softirq_context = 0;
1372 #endif
1373 #ifdef CONFIG_LOCKDEP
1374         p->lockdep_depth = 0; /* no locks held yet */
1375         p->curr_chain_key = 0;
1376         p->lockdep_recursion = 0;
1377 #endif
1378
1379 #ifdef CONFIG_DEBUG_MUTEXES
1380         p->blocked_on = NULL; /* not blocked yet */
1381 #endif
1382 #ifdef CONFIG_BCACHE
1383         p->sequential_io        = 0;
1384         p->sequential_io_avg    = 0;
1385 #endif
1386
1387         /* Perform scheduler related setup. Assign this task to a CPU. */
1388         retval = sched_fork(clone_flags, p);
1389         if (retval)
1390                 goto bad_fork_cleanup_policy;
1391
1392         retval = perf_event_init_task(p);
1393         if (retval)
1394                 goto bad_fork_cleanup_policy;
1395         retval = audit_alloc(p);
1396         if (retval)
1397                 goto bad_fork_cleanup_perf;
1398         /* copy all the process information */
1399         shm_init_task(p);
1400         retval = copy_semundo(clone_flags, p);
1401         if (retval)
1402                 goto bad_fork_cleanup_audit;
1403         retval = copy_files(clone_flags, p);
1404         if (retval)
1405                 goto bad_fork_cleanup_semundo;
1406         retval = copy_fs(clone_flags, p);
1407         if (retval)
1408                 goto bad_fork_cleanup_files;
1409         retval = copy_sighand(clone_flags, p);
1410         if (retval)
1411                 goto bad_fork_cleanup_fs;
1412         retval = copy_signal(clone_flags, p);
1413         if (retval)
1414                 goto bad_fork_cleanup_sighand;
1415         retval = copy_mm(clone_flags, p);
1416         if (retval)
1417                 goto bad_fork_cleanup_signal;
1418         retval = copy_namespaces(clone_flags, p);
1419         if (retval)
1420                 goto bad_fork_cleanup_mm;
1421         retval = copy_io(clone_flags, p);
1422         if (retval)
1423                 goto bad_fork_cleanup_namespaces;
1424         retval = copy_thread(clone_flags, stack_start, stack_size, p);
1425         if (retval)
1426                 goto bad_fork_cleanup_io;
1427
1428         if (pid != &init_struct_pid) {
1429                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1430                 if (IS_ERR(pid)) {
1431                         retval = PTR_ERR(pid);
1432                         goto bad_fork_cleanup_io;
1433                 }
1434         }
1435
1436         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1437         /*
1438          * Clear TID on mm_release()?
1439          */
1440         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1441 #ifdef CONFIG_BLOCK
1442         p->plug = NULL;
1443 #endif
1444 #ifdef CONFIG_FUTEX
1445         p->robust_list = NULL;
1446 #ifdef CONFIG_COMPAT
1447         p->compat_robust_list = NULL;
1448 #endif
1449         INIT_LIST_HEAD(&p->pi_state_list);
1450         p->pi_state_cache = NULL;
1451 #endif
1452         /*
1453          * sigaltstack should be cleared when sharing the same VM
1454          */
1455         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1456                 p->sas_ss_sp = p->sas_ss_size = 0;
1457
1458         /*
1459          * Syscall tracing and stepping should be turned off in the
1460          * child regardless of CLONE_PTRACE.
1461          */
1462         user_disable_single_step(p);
1463         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1464 #ifdef TIF_SYSCALL_EMU
1465         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1466 #endif
1467         clear_all_latency_tracing(p);
1468
1469         /* ok, now we should be set up.. */
1470         p->pid = pid_nr(pid);
1471         if (clone_flags & CLONE_THREAD) {
1472                 p->exit_signal = -1;
1473                 p->group_leader = current->group_leader;
1474                 p->tgid = current->tgid;
1475         } else {
1476                 if (clone_flags & CLONE_PARENT)
1477                         p->exit_signal = current->group_leader->exit_signal;
1478                 else
1479                         p->exit_signal = (clone_flags & CSIGNAL);
1480                 p->group_leader = p;
1481                 p->tgid = p->pid;
1482         }
1483
1484         p->nr_dirtied = 0;
1485         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1486         p->dirty_paused_when = 0;
1487
1488         p->pdeath_signal = 0;
1489         INIT_LIST_HEAD(&p->thread_group);
1490         p->task_works = NULL;
1491
1492         /*
1493          * Make it visible to the rest of the system, but dont wake it up yet.
1494          * Need tasklist lock for parent etc handling!
1495          */
1496         write_lock_irq(&tasklist_lock);
1497
1498         /* CLONE_PARENT re-uses the old parent */
1499         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1500                 p->real_parent = current->real_parent;
1501                 p->parent_exec_id = current->parent_exec_id;
1502         } else {
1503                 p->real_parent = current;
1504                 p->parent_exec_id = current->self_exec_id;
1505         }
1506
1507         spin_lock(&current->sighand->siglock);
1508
1509         /*
1510          * Copy seccomp details explicitly here, in case they were changed
1511          * before holding sighand lock.
1512          */
1513         copy_seccomp(p);
1514
1515         /*
1516          * Process group and session signals need to be delivered to just the
1517          * parent before the fork or both the parent and the child after the
1518          * fork. Restart if a signal comes in before we add the new process to
1519          * it's process group.
1520          * A fatal signal pending means that current will exit, so the new
1521          * thread can't slip out of an OOM kill (or normal SIGKILL).
1522         */
1523         recalc_sigpending();
1524         if (signal_pending(current)) {
1525                 spin_unlock(&current->sighand->siglock);
1526                 write_unlock_irq(&tasklist_lock);
1527                 retval = -ERESTARTNOINTR;
1528                 goto bad_fork_free_pid;
1529         }
1530
1531         if (likely(p->pid)) {
1532                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1533
1534                 init_task_pid(p, PIDTYPE_PID, pid);
1535                 if (thread_group_leader(p)) {
1536                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1537                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1538
1539                         if (is_child_reaper(pid)) {
1540                                 ns_of_pid(pid)->child_reaper = p;
1541                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1542                         }
1543
1544                         p->signal->leader_pid = pid;
1545                         p->signal->tty = tty_kref_get(current->signal->tty);
1546                         list_add_tail(&p->sibling, &p->real_parent->children);
1547                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1548                         attach_pid(p, PIDTYPE_PGID);
1549                         attach_pid(p, PIDTYPE_SID);
1550                         __this_cpu_inc(process_counts);
1551                 } else {
1552                         current->signal->nr_threads++;
1553                         atomic_inc(&current->signal->live);
1554                         atomic_inc(&current->signal->sigcnt);
1555                         list_add_tail_rcu(&p->thread_group,
1556                                           &p->group_leader->thread_group);
1557                         list_add_tail_rcu(&p->thread_node,
1558                                           &p->signal->thread_head);
1559                 }
1560                 attach_pid(p, PIDTYPE_PID);
1561                 nr_threads++;
1562         }
1563
1564         total_forks++;
1565         spin_unlock(&current->sighand->siglock);
1566         syscall_tracepoint_update(p);
1567         write_unlock_irq(&tasklist_lock);
1568
1569         proc_fork_connector(p);
1570         cgroup_post_fork(p);
1571         if (clone_flags & CLONE_THREAD)
1572                 threadgroup_change_end(current);
1573         perf_event_fork(p);
1574
1575         trace_task_newtask(p, clone_flags);
1576         uprobe_copy_process(p, clone_flags);
1577
1578         return p;
1579
1580 bad_fork_free_pid:
1581         if (pid != &init_struct_pid)
1582                 free_pid(pid);
1583 bad_fork_cleanup_io:
1584         if (p->io_context)
1585                 exit_io_context(p);
1586 bad_fork_cleanup_namespaces:
1587         exit_task_namespaces(p);
1588 bad_fork_cleanup_mm:
1589         if (p->mm)
1590                 mmput(p->mm);
1591 bad_fork_cleanup_signal:
1592         if (!(clone_flags & CLONE_THREAD))
1593                 free_signal_struct(p->signal);
1594 bad_fork_cleanup_sighand:
1595         __cleanup_sighand(p->sighand);
1596 bad_fork_cleanup_fs:
1597         exit_fs(p); /* blocking */
1598 bad_fork_cleanup_files:
1599         exit_files(p); /* blocking */
1600 bad_fork_cleanup_semundo:
1601         exit_sem(p);
1602 bad_fork_cleanup_audit:
1603         audit_free(p);
1604 bad_fork_cleanup_perf:
1605         perf_event_free_task(p);
1606 bad_fork_cleanup_policy:
1607 #ifdef CONFIG_NUMA
1608         mpol_put(p->mempolicy);
1609 bad_fork_cleanup_threadgroup_lock:
1610 #endif
1611         if (clone_flags & CLONE_THREAD)
1612                 threadgroup_change_end(current);
1613         delayacct_tsk_free(p);
1614 bad_fork_cleanup_count:
1615         atomic_dec(&p->cred->user->processes);
1616         exit_creds(p);
1617 bad_fork_free:
1618         free_task(p);
1619 fork_out:
1620         return ERR_PTR(retval);
1621 }
1622
1623 static inline void init_idle_pids(struct pid_link *links)
1624 {
1625         enum pid_type type;
1626
1627         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1628                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1629                 links[type].pid = &init_struct_pid;
1630         }
1631 }
1632
1633 struct task_struct *fork_idle(int cpu)
1634 {
1635         struct task_struct *task;
1636         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1637         if (!IS_ERR(task)) {
1638                 init_idle_pids(task->pids);
1639                 init_idle(task, cpu);
1640         }
1641
1642         return task;
1643 }
1644
1645 /*
1646  *  Ok, this is the main fork-routine.
1647  *
1648  * It copies the process, and if successful kick-starts
1649  * it and waits for it to finish using the VM if required.
1650  */
1651 long do_fork(unsigned long clone_flags,
1652               unsigned long stack_start,
1653               unsigned long stack_size,
1654               int __user *parent_tidptr,
1655               int __user *child_tidptr)
1656 {
1657         struct task_struct *p;
1658         int trace = 0;
1659         long nr;
1660
1661         /*
1662          * Determine whether and which event to report to ptracer.  When
1663          * called from kernel_thread or CLONE_UNTRACED is explicitly
1664          * requested, no event is reported; otherwise, report if the event
1665          * for the type of forking is enabled.
1666          */
1667         if (!(clone_flags & CLONE_UNTRACED)) {
1668                 if (clone_flags & CLONE_VFORK)
1669                         trace = PTRACE_EVENT_VFORK;
1670                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1671                         trace = PTRACE_EVENT_CLONE;
1672                 else
1673                         trace = PTRACE_EVENT_FORK;
1674
1675                 if (likely(!ptrace_event_enabled(current, trace)))
1676                         trace = 0;
1677         }
1678
1679         p = copy_process(clone_flags, stack_start, stack_size,
1680                          child_tidptr, NULL, trace);
1681         /*
1682          * Do this prior waking up the new thread - the thread pointer
1683          * might get invalid after that point, if the thread exits quickly.
1684          */
1685         if (!IS_ERR(p)) {
1686                 struct completion vfork;
1687                 struct pid *pid;
1688
1689                 trace_sched_process_fork(current, p);
1690
1691                 pid = get_task_pid(p, PIDTYPE_PID);
1692                 nr = pid_vnr(pid);
1693
1694                 if (clone_flags & CLONE_PARENT_SETTID)
1695                         put_user(nr, parent_tidptr);
1696
1697                 if (clone_flags & CLONE_VFORK) {
1698                         p->vfork_done = &vfork;
1699                         init_completion(&vfork);
1700                         get_task_struct(p);
1701                 }
1702
1703                 wake_up_new_task(p);
1704
1705                 /* forking complete and child started to run, tell ptracer */
1706                 if (unlikely(trace))
1707                         ptrace_event_pid(trace, pid);
1708
1709                 if (clone_flags & CLONE_VFORK) {
1710                         if (!wait_for_vfork_done(p, &vfork))
1711                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1712                 }
1713
1714                 put_pid(pid);
1715         } else {
1716                 nr = PTR_ERR(p);
1717         }
1718         return nr;
1719 }
1720
1721 /*
1722  * Create a kernel thread.
1723  */
1724 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1725 {
1726         return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1727                 (unsigned long)arg, NULL, NULL);
1728 }
1729
1730 #ifdef __ARCH_WANT_SYS_FORK
1731 SYSCALL_DEFINE0(fork)
1732 {
1733 #ifdef CONFIG_MMU
1734         return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1735 #else
1736         /* can not support in nommu mode */
1737         return -EINVAL;
1738 #endif
1739 }
1740 #endif
1741
1742 #ifdef __ARCH_WANT_SYS_VFORK
1743 SYSCALL_DEFINE0(vfork)
1744 {
1745         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1746                         0, NULL, NULL);
1747 }
1748 #endif
1749
1750 #ifdef __ARCH_WANT_SYS_CLONE
1751 #ifdef CONFIG_CLONE_BACKWARDS
1752 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1753                  int __user *, parent_tidptr,
1754                  int, tls_val,
1755                  int __user *, child_tidptr)
1756 #elif defined(CONFIG_CLONE_BACKWARDS2)
1757 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1758                  int __user *, parent_tidptr,
1759                  int __user *, child_tidptr,
1760                  int, tls_val)
1761 #elif defined(CONFIG_CLONE_BACKWARDS3)
1762 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1763                 int, stack_size,
1764                 int __user *, parent_tidptr,
1765                 int __user *, child_tidptr,
1766                 int, tls_val)
1767 #else
1768 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1769                  int __user *, parent_tidptr,
1770                  int __user *, child_tidptr,
1771                  int, tls_val)
1772 #endif
1773 {
1774         return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1775 }
1776 #endif
1777
1778 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1779 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1780 #endif
1781
1782 static void sighand_ctor(void *data)
1783 {
1784         struct sighand_struct *sighand = data;
1785
1786         spin_lock_init(&sighand->siglock);
1787         init_waitqueue_head(&sighand->signalfd_wqh);
1788 }
1789
1790 void __init proc_caches_init(void)
1791 {
1792         sighand_cachep = kmem_cache_create("sighand_cache",
1793                         sizeof(struct sighand_struct), 0,
1794                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1795                         SLAB_NOTRACK, sighand_ctor);
1796         signal_cachep = kmem_cache_create("signal_cache",
1797                         sizeof(struct signal_struct), 0,
1798                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1799         files_cachep = kmem_cache_create("files_cache",
1800                         sizeof(struct files_struct), 0,
1801                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1802         fs_cachep = kmem_cache_create("fs_cache",
1803                         sizeof(struct fs_struct), 0,
1804                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1805         /*
1806          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1807          * whole struct cpumask for the OFFSTACK case. We could change
1808          * this to *only* allocate as much of it as required by the
1809          * maximum number of CPU's we can ever have.  The cpumask_allocation
1810          * is at the end of the structure, exactly for that reason.
1811          */
1812         mm_cachep = kmem_cache_create("mm_struct",
1813                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1814                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1815         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1816         mmap_init();
1817         nsproxy_cache_init();
1818 }
1819
1820 /*
1821  * Check constraints on flags passed to the unshare system call.
1822  */
1823 static int check_unshare_flags(unsigned long unshare_flags)
1824 {
1825         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1826                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1827                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1828                                 CLONE_NEWUSER|CLONE_NEWPID))
1829                 return -EINVAL;
1830         /*
1831          * Not implemented, but pretend it works if there is nothing to
1832          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1833          * needs to unshare vm.
1834          */
1835         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1836                 /* FIXME: get_task_mm() increments ->mm_users */
1837                 if (atomic_read(&current->mm->mm_users) > 1)
1838                         return -EINVAL;
1839         }
1840
1841         return 0;
1842 }
1843
1844 /*
1845  * Unshare the filesystem structure if it is being shared
1846  */
1847 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1848 {
1849         struct fs_struct *fs = current->fs;
1850
1851         if (!(unshare_flags & CLONE_FS) || !fs)
1852                 return 0;
1853
1854         /* don't need lock here; in the worst case we'll do useless copy */
1855         if (fs->users == 1)
1856                 return 0;
1857
1858         *new_fsp = copy_fs_struct(fs);
1859         if (!*new_fsp)
1860                 return -ENOMEM;
1861
1862         return 0;
1863 }
1864
1865 /*
1866  * Unshare file descriptor table if it is being shared
1867  */
1868 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1869 {
1870         struct files_struct *fd = current->files;
1871         int error = 0;
1872
1873         if ((unshare_flags & CLONE_FILES) &&
1874             (fd && atomic_read(&fd->count) > 1)) {
1875                 *new_fdp = dup_fd(fd, &error);
1876                 if (!*new_fdp)
1877                         return error;
1878         }
1879
1880         return 0;
1881 }
1882
1883 /*
1884  * unshare allows a process to 'unshare' part of the process
1885  * context which was originally shared using clone.  copy_*
1886  * functions used by do_fork() cannot be used here directly
1887  * because they modify an inactive task_struct that is being
1888  * constructed. Here we are modifying the current, active,
1889  * task_struct.
1890  */
1891 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1892 {
1893         struct fs_struct *fs, *new_fs = NULL;
1894         struct files_struct *fd, *new_fd = NULL;
1895         struct cred *new_cred = NULL;
1896         struct nsproxy *new_nsproxy = NULL;
1897         int do_sysvsem = 0;
1898         int err;
1899
1900         /*
1901          * If unsharing a user namespace must also unshare the thread.
1902          */
1903         if (unshare_flags & CLONE_NEWUSER)
1904                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1905         /*
1906          * If unsharing a thread from a thread group, must also unshare vm.
1907          */
1908         if (unshare_flags & CLONE_THREAD)
1909                 unshare_flags |= CLONE_VM;
1910         /*
1911          * If unsharing vm, must also unshare signal handlers.
1912          */
1913         if (unshare_flags & CLONE_VM)
1914                 unshare_flags |= CLONE_SIGHAND;
1915         /*
1916          * If unsharing namespace, must also unshare filesystem information.
1917          */
1918         if (unshare_flags & CLONE_NEWNS)
1919                 unshare_flags |= CLONE_FS;
1920
1921         err = check_unshare_flags(unshare_flags);
1922         if (err)
1923                 goto bad_unshare_out;
1924         /*
1925          * CLONE_NEWIPC must also detach from the undolist: after switching
1926          * to a new ipc namespace, the semaphore arrays from the old
1927          * namespace are unreachable.
1928          */
1929         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1930                 do_sysvsem = 1;
1931         err = unshare_fs(unshare_flags, &new_fs);
1932         if (err)
1933                 goto bad_unshare_out;
1934         err = unshare_fd(unshare_flags, &new_fd);
1935         if (err)
1936                 goto bad_unshare_cleanup_fs;
1937         err = unshare_userns(unshare_flags, &new_cred);
1938         if (err)
1939                 goto bad_unshare_cleanup_fd;
1940         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1941                                          new_cred, new_fs);
1942         if (err)
1943                 goto bad_unshare_cleanup_cred;
1944
1945         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1946                 if (do_sysvsem) {
1947                         /*
1948                          * CLONE_SYSVSEM is equivalent to sys_exit().
1949                          */
1950                         exit_sem(current);
1951                 }
1952                 if (unshare_flags & CLONE_NEWIPC) {
1953                         /* Orphan segments in old ns (see sem above). */
1954                         exit_shm(current);
1955                         shm_init_task(current);
1956                 }
1957
1958                 if (new_nsproxy)
1959                         switch_task_namespaces(current, new_nsproxy);
1960
1961                 task_lock(current);
1962
1963                 if (new_fs) {
1964                         fs = current->fs;
1965                         spin_lock(&fs->lock);
1966                         current->fs = new_fs;
1967                         if (--fs->users)
1968                                 new_fs = NULL;
1969                         else
1970                                 new_fs = fs;
1971                         spin_unlock(&fs->lock);
1972                 }
1973
1974                 if (new_fd) {
1975                         fd = current->files;
1976                         current->files = new_fd;
1977                         new_fd = fd;
1978                 }
1979
1980                 task_unlock(current);
1981
1982                 if (new_cred) {
1983                         /* Install the new user namespace */
1984                         commit_creds(new_cred);
1985                         new_cred = NULL;
1986                 }
1987         }
1988
1989 bad_unshare_cleanup_cred:
1990         if (new_cred)
1991                 put_cred(new_cred);
1992 bad_unshare_cleanup_fd:
1993         if (new_fd)
1994                 put_files_struct(new_fd);
1995
1996 bad_unshare_cleanup_fs:
1997         if (new_fs)
1998                 free_fs_struct(new_fs);
1999
2000 bad_unshare_out:
2001         return err;
2002 }
2003
2004 /*
2005  *      Helper to unshare the files of the current task.
2006  *      We don't want to expose copy_files internals to
2007  *      the exec layer of the kernel.
2008  */
2009
2010 int unshare_files(struct files_struct **displaced)
2011 {
2012         struct task_struct *task = current;
2013         struct files_struct *copy = NULL;
2014         int error;
2015
2016         error = unshare_fd(CLONE_FILES, &copy);
2017         if (error || !copy) {
2018                 *displaced = NULL;
2019                 return error;
2020         }
2021         *displaced = task->files;
2022         task_lock(task);
2023         task->files = copy;
2024         task_unlock(task);
2025         return 0;
2026 }
2027
2028 int sysctl_max_threads(struct ctl_table *table, int write,
2029                        void __user *buffer, size_t *lenp, loff_t *ppos)
2030 {
2031         struct ctl_table t;
2032         int ret;
2033         int threads = max_threads;
2034         int min = MIN_THREADS;
2035         int max = MAX_THREADS;
2036
2037         t = *table;
2038         t.data = &threads;
2039         t.extra1 = &min;
2040         t.extra2 = &max;
2041
2042         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2043         if (ret || !write)
2044                 return ret;
2045
2046         set_max_threads(threads);
2047
2048         return 0;
2049 }