kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/nsproxy.h>
  32 #include <linux/capability.h>
  33 #include <linux/cpu.h>
  34 #include <linux/cgroup.h>
  35 #include <linux/security.h>
  36 #include <linux/hugetlb.h>
  37 #include <linux/swap.h>
  38 #include <linux/syscalls.h>
  39 #include <linux/jiffies.h>
  40 #include <linux/tracehook.h>
  41 #include <linux/futex.h>
  42 #include <linux/compat.h>
  43 #include <linux/task_io_accounting_ops.h>
  44 #include <linux/rcupdate.h>
  45 #include <linux/ptrace.h>
  46 #include <linux/mount.h>
  47 #include <linux/audit.h>
  48 #include <linux/memcontrol.h>
  49 #include <linux/ftrace.h>
  50 #include <linux/profile.h>
  51 #include <linux/rmap.h>
  52 #include <linux/ksm.h>
  53 #include <linux/acct.h>
  54 #include <linux/tsacct_kern.h>
  55 #include <linux/cn_proc.h>
  56 #include <linux/freezer.h>
  57 #include <linux/delayacct.h>
  58 #include <linux/taskstats_kern.h>
  59 #include <linux/random.h>
  60 #include <linux/tty.h>
  61 #include <linux/proc_fs.h>
  62 #include <linux/blkdev.h>
  63 #include <linux/fs_struct.h>
  64 #include <linux/magic.h>
  65 #include <linux/perf_event.h>
  66 #include <linux/posix-timers.h>
  67
  68 #include <asm/pgtable.h>
  69 #include <asm/pgalloc.h>
  70 #include <asm/uaccess.h>
  71 #include <asm/mmu_context.h>
  72 #include <asm/cacheflush.h>
  73 #include <asm/tlbflush.h>
  74
  75 #include <trace/events/sched.h>
  76
  77 /*
  78  * Protected counters by write_lock_irq(&tasklist_lock)
  79  */
  80 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  81 int nr_threads;                 /* The idle threads do not count.. */
  82
  83 int max_threads;                /* tunable limit on nr_threads */
  84
  85 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  86
  87 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  88
  89 int nr_processes(void)
  90 {
  91         int cpu;
  92         int total = 0;
  93
  94         for_each_possible_cpu(cpu)
  95                 total += per_cpu(process_counts, cpu);
  96
  97         return total;
  98 }
  99
 100 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 101 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
 102 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
 103 static struct kmem_cache *task_struct_cachep;
 104 #endif
 105
 106 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 107 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
 108 {
 109 #ifdef CONFIG_DEBUG_STACK_USAGE
 110         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 111 #else
 112         gfp_t mask = GFP_KERNEL;
 113 #endif
 114         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
 115 }
 116
 117 static inline void free_thread_info(struct thread_info *ti)
 118 {
 119         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 120 }
 121 #endif
 122
 123 /* SLAB cache for signal_struct structures (tsk->signal) */
 124 static struct kmem_cache *signal_cachep;
 125
 126 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 127 struct kmem_cache *sighand_cachep;
 128
 129 /* SLAB cache for files_struct structures (tsk->files) */
 130 struct kmem_cache *files_cachep;
 131
 132 /* SLAB cache for fs_struct structures (tsk->fs) */
 133 struct kmem_cache *fs_cachep;
 134
 135 /* SLAB cache for vm_area_struct structures */
 136 struct kmem_cache *vm_area_cachep;
 137
 138 /* SLAB cache for mm_struct structures (tsk->mm) */
 139 static struct kmem_cache *mm_cachep;
 140
 141 /* Notifier list called when a task struct is freed */
 142 static ATOMIC_NOTIFIER_HEAD(task_free_notifier);
 143
 144 static void account_kernel_stack(struct thread_info *ti, int account)
 145 {
 146         struct zone *zone = page_zone(virt_to_page(ti));
 147
 148         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 149 }
 150
 151 void free_task(struct task_struct *tsk)
 152 {
 153         prop_local_destroy_single(&tsk->dirties);
 154         account_kernel_stack(tsk->stack, -1);
 155         free_thread_info(tsk->stack);
 156         rt_mutex_debug_task_free(tsk);
 157         ftrace_graph_exit_task(tsk);
 158         free_task_struct(tsk);
 159 }
 160 EXPORT_SYMBOL(free_task);
 161
 162 int task_free_register(struct notifier_block *n)
 163 {
 164         return atomic_notifier_chain_register(&task_free_notifier, n);
 165 }
 166 EXPORT_SYMBOL(task_free_register);
 167
 168 int task_free_unregister(struct notifier_block *n)
 169 {
 170         return atomic_notifier_chain_unregister(&task_free_notifier, n);
 171 }
 172 EXPORT_SYMBOL(task_free_unregister);
 173
 174 void __put_task_struct(struct task_struct *tsk)
 175 {
 176         WARN_ON(!tsk->exit_state);
 177         WARN_ON(atomic_read(&tsk->usage));
 178         WARN_ON(tsk == current);
 179
 180         exit_creds(tsk);
 181         delayacct_tsk_free(tsk);
 182
 183         atomic_notifier_call_chain(&task_free_notifier, 0, tsk);
 184         if (!profile_handoff_task(tsk))
 185                 free_task(tsk);
 186 }
 187
 188 /*
 189  * macro override instead of weak attribute alias, to workaround
 190  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 191  */
 192 #ifndef arch_task_cache_init
 193 #define arch_task_cache_init()
 194 #endif
 195
 196 void __init fork_init(unsigned long mempages)
 197 {
 198 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 199 #ifndef ARCH_MIN_TASKALIGN
 200 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 201 #endif
 202         /* create a slab on which task_structs can be allocated */
 203         task_struct_cachep =
 204                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 205                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 206 #endif
 207
 208         /* do the arch specific task caches init */
 209         arch_task_cache_init();
 210
 211         /*
 212          * The default maximum number of threads is set to a safe
 213          * value: the thread structures can take up at most half
 214          * of memory.
 215          */
 216         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 217
 218         /*
 219          * we need to allow at least 20 threads to boot a system
 220          */
 221         if(max_threads < 20)
 222                 max_threads = 20;
 223
 224         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 225         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 226         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 227                 init_task.signal->rlim[RLIMIT_NPROC];
 228 }
 229
 230 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 231                                                struct task_struct *src)
 232 {
 233         *dst = *src;
 234         return 0;
 235 }
 236
 237 static struct task_struct *dup_task_struct(struct task_struct *orig)
 238 {
 239         struct task_struct *tsk;
 240         struct thread_info *ti;
 241         unsigned long *stackend;
 242
 243         int err;
 244
 245         prepare_to_copy(orig);
 246
 247         tsk = alloc_task_struct();
 248         if (!tsk)
 249                 return NULL;
 250
 251         ti = alloc_thread_info(tsk);
 252         if (!ti) {
 253                 free_task_struct(tsk);
 254                 return NULL;
 255         }
 256
 257         err = arch_dup_task_struct(tsk, orig);
 258         if (err)
 259                 goto out;
 260
 261         tsk->stack = ti;
 262
 263         err = prop_local_init_single(&tsk->dirties);
 264         if (err)
 265                 goto out;
 266
 267         setup_thread_stack(tsk, orig);
 268         stackend = end_of_stack(tsk);
 269         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 270
 271 #ifdef CONFIG_CC_STACKPROTECTOR
 272         tsk->stack_canary = get_random_int();
 273 #endif
 274
 275         /* One for us, one for whoever does the "release_task()" (usually parent) */
 276         atomic_set(&tsk->usage,2);
 277         atomic_set(&tsk->fs_excl, 0);
 278 #ifdef CONFIG_BLK_DEV_IO_TRACE
 279         tsk->btrace_seq = 0;
 280 #endif
 281         tsk->splice_pipe = NULL;
 282
 283         account_kernel_stack(ti, 1);
 284
 285         return tsk;
 286
 287 out:
 288         free_thread_info(ti);
 289         free_task_struct(tsk);
 290         return NULL;
 291 }
 292
 293 #ifdef CONFIG_MMU
 294 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 295 {
 296         struct vm_area_struct *mpnt, *tmp, **pprev;
 297         struct rb_node **rb_link, *rb_parent;
 298         int retval;
 299         unsigned long charge;
 300         struct mempolicy *pol;
 301
 302         down_write(&oldmm->mmap_sem);
 303         flush_cache_dup_mm(oldmm);
 304         /*
 305          * Not linked in yet - no deadlock potential:
 306          */
 307         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 308
 309         mm->locked_vm = 0;
 310         mm->mmap = NULL;
 311         mm->mmap_cache = NULL;
 312         mm->free_area_cache = oldmm->mmap_base;
 313         mm->cached_hole_size = ~0UL;
 314         mm->map_count = 0;
 315         cpumask_clear(mm_cpumask(mm));
 316         mm->mm_rb = RB_ROOT;
 317         rb_link = &mm->mm_rb.rb_node;
 318         rb_parent = NULL;
 319         pprev = &mm->mmap;
 320         retval = ksm_fork(mm, oldmm);
 321         if (retval)
 322                 goto out;
 323
 324         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 325                 struct file *file;
 326
 327                 if (mpnt->vm_flags & VM_DONTCOPY) {
 328                         long pages = vma_pages(mpnt);
 329                         mm->total_vm -= pages;
 330                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 331                                                                 -pages);
 332                         continue;
 333                 }
 334                 charge = 0;
 335                 if (mpnt->vm_flags & VM_ACCOUNT) {
 336                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 337                         if (security_vm_enough_memory(len))
 338                                 goto fail_nomem;
 339                         charge = len;
 340                 }
 341                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 342                 if (!tmp)
 343                         goto fail_nomem;
 344                 *tmp = *mpnt;
 345                 pol = mpol_dup(vma_policy(mpnt));
 346                 retval = PTR_ERR(pol);
 347                 if (IS_ERR(pol))
 348                         goto fail_nomem_policy;
 349                 vma_set_policy(tmp, pol);
 350                 tmp->vm_flags &= ~VM_LOCKED;
 351                 tmp->vm_mm = mm;
 352                 tmp->vm_next = NULL;
 353                 anon_vma_link(tmp);
 354                 file = tmp->vm_file;
 355                 if (file) {
 356                         struct inode *inode = file->f_path.dentry->d_inode;
 357                         struct address_space *mapping = file->f_mapping;
 358
 359                         get_file(file);
 360                         if (tmp->vm_flags & VM_DENYWRITE)
 361                                 atomic_dec(&inode->i_writecount);
 362                         spin_lock(&mapping->i_mmap_lock);
 363                         if (tmp->vm_flags & VM_SHARED)
 364                                 mapping->i_mmap_writable++;
 365                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 366                         flush_dcache_mmap_lock(mapping);
 367                         /* insert tmp into the share list, just after mpnt */
 368                         vma_prio_tree_add(tmp, mpnt);
 369                         flush_dcache_mmap_unlock(mapping);
 370                         spin_unlock(&mapping->i_mmap_lock);
 371                 }
 372
 373                 /*
 374                  * Clear hugetlb-related page reserves for children. This only
 375                  * affects MAP_PRIVATE mappings. Faults generated by the child
 376                  * are not guaranteed to succeed, even if read-only
 377                  */
 378                 if (is_vm_hugetlb_page(tmp))
 379                         reset_vma_resv_huge_pages(tmp);
 380
 381                 /*
 382                  * Link in the new vma and copy the page table entries.
 383                  */
 384                 *pprev = tmp;
 385                 pprev = &tmp->vm_next;
 386
 387                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 388                 rb_link = &tmp->vm_rb.rb_right;
 389                 rb_parent = &tmp->vm_rb;
 390
 391                 mm->map_count++;
 392                 retval = copy_page_range(mm, oldmm, mpnt);
 393
 394                 if (tmp->vm_ops && tmp->vm_ops->open)
 395                         tmp->vm_ops->open(tmp);
 396
 397                 if (retval)
 398                         goto out;
 399         }
 400         /* a new mm has just been created */
 401         arch_dup_mmap(oldmm, mm);
 402         retval = 0;
 403 out:
 404         up_write(&mm->mmap_sem);
 405         flush_tlb_mm(oldmm);
 406         up_write(&oldmm->mmap_sem);
 407         return retval;
 408 fail_nomem_policy:
 409         kmem_cache_free(vm_area_cachep, tmp);
 410 fail_nomem:
 411         retval = -ENOMEM;
 412         vm_unacct_memory(charge);
 413         goto out;
 414 }
 415
 416 static inline int mm_alloc_pgd(struct mm_struct * mm)
 417 {
 418         mm->pgd = pgd_alloc(mm);
 419         if (unlikely(!mm->pgd))
 420                 return -ENOMEM;
 421         return 0;
 422 }
 423
 424 static inline void mm_free_pgd(struct mm_struct * mm)
 425 {
 426         pgd_free(mm, mm->pgd);
 427 }
 428 #else
 429 #define dup_mmap(mm, oldmm)     (0)
 430 #define mm_alloc_pgd(mm)        (0)
 431 #define mm_free_pgd(mm)
 432 #endif /* CONFIG_MMU */
 433
 434 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 435
 436 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 437 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 438
 439 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 440
 441 static int __init coredump_filter_setup(char *s)
 442 {
 443         default_dump_filter =
 444                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 445                 MMF_DUMP_FILTER_MASK;
 446         return 1;
 447 }
 448
 449 __setup("coredump_filter=", coredump_filter_setup);
 450
 451 #include <linux/init_task.h>
 452
 453 static void mm_init_aio(struct mm_struct *mm)
 454 {
 455 #ifdef CONFIG_AIO
 456         spin_lock_init(&mm->ioctx_lock);
 457         INIT_HLIST_HEAD(&mm->ioctx_list);
 458 #endif
 459 }
 460
 461 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
 462 {
 463         atomic_set(&mm->mm_users, 1);
 464         atomic_set(&mm->mm_count, 1);
 465         init_rwsem(&mm->mmap_sem);
 466         INIT_LIST_HEAD(&mm->mmlist);
 467         mm->flags = (current->mm) ?
 468                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 469         mm->core_state = NULL;
 470         mm->nr_ptes = 0;
 471         set_mm_counter(mm, file_rss, 0);
 472         set_mm_counter(mm, anon_rss, 0);
 473         spin_lock_init(&mm->page_table_lock);
 474         mm->free_area_cache = TASK_UNMAPPED_BASE;
 475         mm->cached_hole_size = ~0UL;
 476         mm_init_aio(mm);
 477         mm_init_owner(mm, p);
 478
 479         if (likely(!mm_alloc_pgd(mm))) {
 480                 mm->def_flags = 0;
 481                 mmu_notifier_mm_init(mm);
 482                 return mm;
 483         }
 484
 485         free_mm(mm);
 486         return NULL;
 487 }
 488
 489 /*
 490  * Allocate and initialize an mm_struct.
 491  */
 492 struct mm_struct * mm_alloc(void)
 493 {
 494         struct mm_struct * mm;
 495
 496         mm = allocate_mm();
 497         if (mm) {
 498                 memset(mm, 0, sizeof(*mm));
 499                 mm = mm_init(mm, current);
 500         }
 501         return mm;
 502 }
 503
 504 /*
 505  * Called when the last reference to the mm
 506  * is dropped: either by a lazy thread or by
 507  * mmput. Free the page directory and the mm.
 508  */
 509 void __mmdrop(struct mm_struct *mm)
 510 {
 511         BUG_ON(mm == &init_mm);
 512         mm_free_pgd(mm);
 513         destroy_context(mm);
 514         mmu_notifier_mm_destroy(mm);
 515         free_mm(mm);
 516 }
 517 EXPORT_SYMBOL_GPL(__mmdrop);
 518
 519 /*
 520  * Decrement the use count and release all resources for an mm.
 521  */
 522 void mmput(struct mm_struct *mm)
 523 {
 524         might_sleep();
 525
 526         if (atomic_dec_and_test(&mm->mm_users)) {
 527                 exit_aio(mm);
 528                 ksm_exit(mm);
 529                 exit_mmap(mm);
 530                 set_mm_exe_file(mm, NULL);
 531                 if (!list_empty(&mm->mmlist)) {
 532                         spin_lock(&mmlist_lock);
 533                         list_del(&mm->mmlist);
 534                         spin_unlock(&mmlist_lock);
 535                 }
 536                 put_swap_token(mm);
 537                 if (mm->binfmt)
 538                         module_put(mm->binfmt->module);
 539                 mmdrop(mm);
 540         }
 541 }
 542 EXPORT_SYMBOL_GPL(mmput);
 543
 544 /**
 545  * get_task_mm - acquire a reference to the task's mm
 546  *
 547  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 548  * this kernel workthread has transiently adopted a user mm with use_mm,
 549  * to do its AIO) is not set and if so returns a reference to it, after
 550  * bumping up the use count.  User must release the mm via mmput()
 551  * after use.  Typically used by /proc and ptrace.
 552  */
 553 struct mm_struct *get_task_mm(struct task_struct *task)
 554 {
 555         struct mm_struct *mm;
 556
 557         task_lock(task);
 558         mm = task->mm;
 559         if (mm) {
 560                 if (task->flags & PF_KTHREAD)
 561                         mm = NULL;
 562                 else
 563                         atomic_inc(&mm->mm_users);
 564         }
 565         task_unlock(task);
 566         return mm;
 567 }
 568 EXPORT_SYMBOL_GPL(get_task_mm);
 569
 570 /* Please note the differences between mmput and mm_release.
 571  * mmput is called whenever we stop holding onto a mm_struct,
 572  * error success whatever.
 573  *
 574  * mm_release is called after a mm_struct has been removed
 575  * from the current process.
 576  *
 577  * This difference is important for error handling, when we
 578  * only half set up a mm_struct for a new process and need to restore
 579  * the old one.  Because we mmput the new mm_struct before
 580  * restoring the old one. . .
 581  * Eric Biederman 10 January 1998
 582  */
 583 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 584 {
 585         struct completion *vfork_done = tsk->vfork_done;
 586
 587         /* Get rid of any futexes when releasing the mm */
 588 #ifdef CONFIG_FUTEX
 589         if (unlikely(tsk->robust_list)) {
 590                 exit_robust_list(tsk);
 591                 tsk->robust_list = NULL;
 592         }
 593 #ifdef CONFIG_COMPAT
 594         if (unlikely(tsk->compat_robust_list)) {
 595                 compat_exit_robust_list(tsk);
 596                 tsk->compat_robust_list = NULL;
 597         }
 598 #endif
 599         if (unlikely(!list_empty(&tsk->pi_state_list)))
 600                 exit_pi_state_list(tsk);
 601 #endif
 602
 603         /* Get rid of any cached register state */
 604         deactivate_mm(tsk, mm);
 605
 606         /* notify parent sleeping on vfork() */
 607         if (vfork_done) {
 608                 tsk->vfork_done = NULL;
 609                 complete(vfork_done);
 610         }
 611
 612         /*
 613          * If we're exiting normally, clear a user-space tid field if
 614          * requested.  We leave this alone when dying by signal, to leave
 615          * the value intact in a core dump, and to save the unnecessary
 616          * trouble otherwise.  Userland only wants this done for a sys_exit.
 617          */
 618         if (tsk->clear_child_tid) {
 619                 if (!(tsk->flags & PF_SIGNALED) &&
 620                     atomic_read(&mm->mm_users) > 1) {
 621                         /*
 622                          * We don't check the error code - if userspace has
 623                          * not set up a proper pointer then tough luck.
 624                          */
 625                         put_user(0, tsk->clear_child_tid);
 626                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 627                                         1, NULL, NULL, 0);
 628                 }
 629                 tsk->clear_child_tid = NULL;
 630         }
 631 }
 632
 633 /*
 634  * Allocate a new mm structure and copy contents from the
 635  * mm structure of the passed in task structure.
 636  */
 637 struct mm_struct *dup_mm(struct task_struct *tsk)
 638 {
 639         struct mm_struct *mm, *oldmm = current->mm;
 640         int err;
 641
 642         if (!oldmm)
 643                 return NULL;
 644
 645         mm = allocate_mm();
 646         if (!mm)
 647                 goto fail_nomem;
 648
 649         memcpy(mm, oldmm, sizeof(*mm));
 650
 651         /* Initializing for Swap token stuff */
 652         mm->token_priority = 0;
 653         mm->last_interval = 0;
 654
 655         if (!mm_init(mm, tsk))
 656                 goto fail_nomem;
 657
 658         if (init_new_context(tsk, mm))
 659                 goto fail_nocontext;
 660
 661         dup_mm_exe_file(oldmm, mm);
 662
 663         err = dup_mmap(mm, oldmm);
 664         if (err)
 665                 goto free_pt;
 666
 667         mm->hiwater_rss = get_mm_rss(mm);
 668         mm->hiwater_vm = mm->total_vm;
 669
 670         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 671                 goto free_pt;
 672
 673         return mm;
 674
 675 free_pt:
 676         /* don't put binfmt in mmput, we haven't got module yet */
 677         mm->binfmt = NULL;
 678         mmput(mm);
 679
 680 fail_nomem:
 681         return NULL;
 682
 683 fail_nocontext:
 684         /*
 685          * If init_new_context() failed, we cannot use mmput() to free the mm
 686          * because it calls destroy_context()
 687          */
 688         mm_free_pgd(mm);
 689         free_mm(mm);
 690         return NULL;
 691 }
 692
 693 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 694 {
 695         struct mm_struct * mm, *oldmm;
 696         int retval;
 697
 698         tsk->min_flt = tsk->maj_flt = 0;
 699         tsk->nvcsw = tsk->nivcsw = 0;
 700 #ifdef CONFIG_DETECT_HUNG_TASK
 701         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 702 #endif
 703
 704         tsk->mm = NULL;
 705         tsk->active_mm = NULL;
 706
 707         /*
 708          * Are we cloning a kernel thread?
 709          *
 710          * We need to steal a active VM for that..
 711          */
 712         oldmm = current->mm;
 713         if (!oldmm)
 714                 return 0;
 715
 716         if (clone_flags & CLONE_VM) {
 717                 atomic_inc(&oldmm->mm_users);
 718                 mm = oldmm;
 719                 goto good_mm;
 720         }
 721
 722         retval = -ENOMEM;
 723         mm = dup_mm(tsk);
 724         if (!mm)
 725                 goto fail_nomem;
 726
 727 good_mm:
 728         /* Initializing for Swap token stuff */
 729         mm->token_priority = 0;
 730         mm->last_interval = 0;
 731
 732         tsk->mm = mm;
 733         tsk->active_mm = mm;
 734         return 0;
 735
 736 fail_nomem:
 737         return retval;
 738 }
 739
 740 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 741 {
 742         struct fs_struct *fs = current->fs;
 743         if (clone_flags & CLONE_FS) {
 744                 /* tsk->fs is already what we want */
 745                 write_lock(&fs->lock);
 746                 if (fs->in_exec) {
 747                         write_unlock(&fs->lock);
 748                         return -EAGAIN;
 749                 }
 750                 fs->users++;
 751                 write_unlock(&fs->lock);
 752                 return 0;
 753         }
 754         tsk->fs = copy_fs_struct(fs);
 755         if (!tsk->fs)
 756                 return -ENOMEM;
 757         return 0;
 758 }
 759
 760 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 761 {
 762         struct files_struct *oldf, *newf;
 763         int error = 0;
 764
 765         /*
 766          * A background process may not have any files ...
 767          */
 768         oldf = current->files;
 769         if (!oldf)
 770                 goto out;
 771
 772         if (clone_flags & CLONE_FILES) {
 773                 atomic_inc(&oldf->count);
 774                 goto out;
 775         }
 776
 777         newf = dup_fd(oldf, &error);
 778         if (!newf)
 779                 goto out;
 780
 781         tsk->files = newf;
 782         error = 0;
 783 out:
 784         return error;
 785 }
 786
 787 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 788 {
 789 #ifdef CONFIG_BLOCK
 790         struct io_context *ioc = current->io_context;
 791
 792         if (!ioc)
 793                 return 0;
 794         /*
 795          * Share io context with parent, if CLONE_IO is set
 796          */
 797         if (clone_flags & CLONE_IO) {
 798                 tsk->io_context = ioc_task_link(ioc);
 799                 if (unlikely(!tsk->io_context))
 800                         return -ENOMEM;
 801         } else if (ioprio_valid(ioc->ioprio)) {
 802                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 803                 if (unlikely(!tsk->io_context))
 804                         return -ENOMEM;
 805
 806                 tsk->io_context->ioprio = ioc->ioprio;
 807         }
 808 #endif
 809         return 0;
 810 }
 811
 812 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 813 {
 814         struct sighand_struct *sig;
 815
 816         if (clone_flags & CLONE_SIGHAND) {
 817                 atomic_inc(&current->sighand->count);
 818                 return 0;
 819         }
 820         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 821         rcu_assign_pointer(tsk->sighand, sig);
 822         if (!sig)
 823                 return -ENOMEM;
 824         atomic_set(&sig->count, 1);
 825         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 826         return 0;
 827 }
 828
 829 void __cleanup_sighand(struct sighand_struct *sighand)
 830 {
 831         if (atomic_dec_and_test(&sighand->count))
 832                 kmem_cache_free(sighand_cachep, sighand);
 833 }
 834
 835
 836 /*
 837  * Initialize POSIX timer handling for a thread group.
 838  */
 839 static void posix_cpu_timers_init_group(struct signal_struct *sig)
 840 {
 841         /* Thread group counters. */
 842         thread_group_cputime_init(sig);
 843
 844         /* Expiration times and increments. */
 845         sig->it[CPUCLOCK_PROF].expires = cputime_zero;
 846         sig->it[CPUCLOCK_PROF].incr = cputime_zero;
 847         sig->it[CPUCLOCK_VIRT].expires = cputime_zero;
 848         sig->it[CPUCLOCK_VIRT].incr = cputime_zero;
 849
 850         /* Cached expiration times. */
 851         sig->cputime_expires.prof_exp = cputime_zero;
 852         sig->cputime_expires.virt_exp = cputime_zero;
 853         sig->cputime_expires.sched_exp = 0;
 854
 855         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 856                 sig->cputime_expires.prof_exp =
 857                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 858                 sig->cputimer.running = 1;
 859         }
 860
 861         /* The timer lists. */
 862         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 863         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 864         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 865 }
 866
 867 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 868 {
 869         struct signal_struct *sig;
 870
 871         if (clone_flags & CLONE_THREAD)
 872                 return 0;
 873
 874         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 875         tsk->signal = sig;
 876         if (!sig)
 877                 return -ENOMEM;
 878
 879         atomic_set(&sig->count, 1);
 880         atomic_set(&sig->live, 1);
 881         init_waitqueue_head(&sig->wait_chldexit);
 882         sig->flags = 0;
 883         if (clone_flags & CLONE_NEWPID)
 884                 sig->flags |= SIGNAL_UNKILLABLE;
 885         sig->group_exit_code = 0;
 886         sig->group_exit_task = NULL;
 887         sig->group_stop_count = 0;
 888         sig->curr_target = tsk;
 889         init_sigpending(&sig->shared_pending);
 890         INIT_LIST_HEAD(&sig->posix_timers);
 891
 892         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 893         sig->it_real_incr.tv64 = 0;
 894         sig->real_timer.function = it_real_fn;
 895
 896         sig->leader = 0;        /* session leadership doesn't inherit */
 897         sig->tty_old_pgrp = NULL;
 898         sig->tty = NULL;
 899
 900         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 901         sig->gtime = cputime_zero;
 902         sig->cgtime = cputime_zero;
 903         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 904         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 905         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
 906         sig->maxrss = sig->cmaxrss = 0;
 907         task_io_accounting_init(&sig->ioac);
 908         sig->sum_sched_runtime = 0;
 909         taskstats_tgid_init(sig);
 910
 911         task_lock(current->group_leader);
 912         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 913         task_unlock(current->group_leader);
 914
 915         posix_cpu_timers_init_group(sig);
 916
 917         acct_init_pacct(&sig->pacct);
 918
 919         tty_audit_fork(sig);
 920
 921         sig->oom_adj = current->signal->oom_adj;
 922
 923         return 0;
 924 }
 925
 926 void __cleanup_signal(struct signal_struct *sig)
 927 {
 928         thread_group_cputime_free(sig);
 929         tty_kref_put(sig->tty);
 930         kmem_cache_free(signal_cachep, sig);
 931 }
 932
 933 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 934 {
 935         unsigned long new_flags = p->flags;
 936
 937         new_flags &= ~PF_SUPERPRIV;
 938         new_flags |= PF_FORKNOEXEC;
 939         new_flags |= PF_STARTING;
 940         p->flags = new_flags;
 941         clear_freeze_flag(p);
 942 }
 943
 944 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
 945 {
 946         current->clear_child_tid = tidptr;
 947
 948         return task_pid_vnr(current);
 949 }
 950
 951 static void rt_mutex_init_task(struct task_struct *p)
 952 {
 953         spin_lock_init(&p->pi_lock);
 954 #ifdef CONFIG_RT_MUTEXES
 955         plist_head_init(&p->pi_waiters, &p->pi_lock);
 956         p->pi_blocked_on = NULL;
 957 #endif
 958 }
 959
 960 #ifdef CONFIG_MM_OWNER
 961 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 962 {
 963         mm->owner = p;
 964 }
 965 #endif /* CONFIG_MM_OWNER */
 966
 967 /*
 968  * Initialize POSIX timer handling for a single task.
 969  */
 970 static void posix_cpu_timers_init(struct task_struct *tsk)
 971 {
 972         tsk->cputime_expires.prof_exp = cputime_zero;
 973         tsk->cputime_expires.virt_exp = cputime_zero;
 974         tsk->cputime_expires.sched_exp = 0;
 975         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
 976         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
 977         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
 978 }
 979
 980 /*
 981  * This creates a new process as a copy of the old one,
 982  * but does not actually start it yet.
 983  *
 984  * It copies the registers, and all the appropriate
 985  * parts of the process environment (as per the clone
 986  * flags). The actual kick-off is left to the caller.
 987  */
 988 static struct task_struct *copy_process(unsigned long clone_flags,
 989                                         unsigned long stack_start,
 990                                         struct pt_regs *regs,
 991                                         unsigned long stack_size,
 992                                         int __user *child_tidptr,
 993                                         struct pid *pid,
 994                                         int trace)
 995 {
 996         int retval;
 997         struct task_struct *p;
 998         int cgroup_callbacks_done = 0;
 999
1000         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1001                 return ERR_PTR(-EINVAL);
1002
1003         /*
1004          * Thread groups must share signals as well, and detached threads
1005          * can only be started up within the thread group.
1006          */
1007         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1008                 return ERR_PTR(-EINVAL);
1009
1010         /*
1011          * Shared signal handlers imply shared VM. By way of the above,
1012          * thread groups also imply shared VM. Blocking this case allows
1013          * for various simplifications in other code.
1014          */
1015         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1016                 return ERR_PTR(-EINVAL);
1017
1018         /*
1019          * Siblings of global init remain as zombies on exit since they are
1020          * not reaped by their parent (swapper). To solve this and to avoid
1021          * multi-rooted process trees, prevent global and container-inits
1022          * from creating siblings.
1023          */
1024         if ((clone_flags & CLONE_PARENT) &&
1025                                 current->signal->flags & SIGNAL_UNKILLABLE)
1026                 return ERR_PTR(-EINVAL);
1027
1028         retval = security_task_create(clone_flags);
1029         if (retval)
1030                 goto fork_out;
1031
1032         retval = -ENOMEM;
1033         p = dup_task_struct(current);
1034         if (!p)
1035                 goto fork_out;
1036
1037         ftrace_graph_init_task(p);
1038
1039         rt_mutex_init_task(p);
1040
1041 #ifdef CONFIG_PROVE_LOCKING
1042         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1043         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1044 #endif
1045         retval = -EAGAIN;
1046         if (atomic_read(&p->real_cred->user->processes) >=
1047                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1048                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1049                     p->real_cred->user != INIT_USER)
1050                         goto bad_fork_free;
1051         }
1052
1053         retval = copy_creds(p, clone_flags);
1054         if (retval < 0)
1055                 goto bad_fork_free;
1056
1057         /*
1058          * If multiple threads are within copy_process(), then this check
1059          * triggers too late. This doesn't hurt, the check is only there
1060          * to stop root fork bombs.
1061          */
1062         retval = -EAGAIN;
1063         if (nr_threads >= max_threads)
1064                 goto bad_fork_cleanup_count;
1065
1066         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1067                 goto bad_fork_cleanup_count;
1068
1069         p->did_exec = 0;
1070         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1071         copy_flags(clone_flags, p);
1072         INIT_LIST_HEAD(&p->children);
1073         INIT_LIST_HEAD(&p->sibling);
1074         rcu_copy_process(p);
1075         p->vfork_done = NULL;
1076         spin_lock_init(&p->alloc_lock);
1077
1078         init_sigpending(&p->pending);
1079
1080         p->utime = cputime_zero;
1081         p->stime = cputime_zero;
1082         p->gtime = cputime_zero;
1083         p->utimescaled = cputime_zero;
1084         p->stimescaled = cputime_zero;
1085         p->prev_utime = cputime_zero;
1086         p->prev_stime = cputime_zero;
1087
1088         p->default_timer_slack_ns = current->timer_slack_ns;
1089
1090         task_io_accounting_init(&p->ioac);
1091         acct_clear_integrals(p);
1092
1093         posix_cpu_timers_init(p);
1094
1095         p->lock_depth = -1;             /* -1 = no lock */
1096         do_posix_clock_monotonic_gettime(&p->start_time);
1097         p->real_start_time = p->start_time;
1098         monotonic_to_bootbased(&p->real_start_time);
1099         p->io_context = NULL;
1100         p->audit_context = NULL;
1101         cgroup_fork(p);
1102 #ifdef CONFIG_NUMA
1103         p->mempolicy = mpol_dup(p->mempolicy);
1104         if (IS_ERR(p->mempolicy)) {
1105                 retval = PTR_ERR(p->mempolicy);
1106                 p->mempolicy = NULL;
1107                 goto bad_fork_cleanup_cgroup;
1108         }
1109         mpol_fix_fork_child_flag(p);
1110 #endif
1111 #ifdef CONFIG_TRACE_IRQFLAGS
1112         p->irq_events = 0;
1113 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1114         p->hardirqs_enabled = 1;
1115 #else
1116         p->hardirqs_enabled = 0;
1117 #endif
1118         p->hardirq_enable_ip = 0;
1119         p->hardirq_enable_event = 0;
1120         p->hardirq_disable_ip = _THIS_IP_;
1121         p->hardirq_disable_event = 0;
1122         p->softirqs_enabled = 1;
1123         p->softirq_enable_ip = _THIS_IP_;
1124         p->softirq_enable_event = 0;
1125         p->softirq_disable_ip = 0;
1126         p->softirq_disable_event = 0;
1127         p->hardirq_context = 0;
1128         p->softirq_context = 0;
1129 #endif
1130 #ifdef CONFIG_LOCKDEP
1131         p->lockdep_depth = 0; /* no locks held yet */
1132         p->curr_chain_key = 0;
1133         p->lockdep_recursion = 0;
1134 #endif
1135
1136 #ifdef CONFIG_DEBUG_MUTEXES
1137         p->blocked_on = NULL; /* not blocked yet */
1138 #endif
1139
1140         p->bts = NULL;
1141
1142         p->stack_start = stack_start;
1143
1144         /* Perform scheduler related setup. Assign this task to a CPU. */
1145         sched_fork(p, clone_flags);
1146
1147         retval = perf_event_init_task(p);
1148         if (retval)
1149                 goto bad_fork_cleanup_policy;
1150
1151         if ((retval = audit_alloc(p)))
1152                 goto bad_fork_cleanup_policy;
1153         /* copy all the process information */
1154         if ((retval = copy_semundo(clone_flags, p)))
1155                 goto bad_fork_cleanup_audit;
1156         if ((retval = copy_files(clone_flags, p)))
1157                 goto bad_fork_cleanup_semundo;
1158         if ((retval = copy_fs(clone_flags, p)))
1159                 goto bad_fork_cleanup_files;
1160         if ((retval = copy_sighand(clone_flags, p)))
1161                 goto bad_fork_cleanup_fs;
1162         if ((retval = copy_signal(clone_flags, p)))
1163                 goto bad_fork_cleanup_sighand;
1164         if ((retval = copy_mm(clone_flags, p)))
1165                 goto bad_fork_cleanup_signal;
1166         if ((retval = copy_namespaces(clone_flags, p)))
1167                 goto bad_fork_cleanup_mm;
1168         if ((retval = copy_io(clone_flags, p)))
1169                 goto bad_fork_cleanup_namespaces;
1170         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1171         if (retval)
1172                 goto bad_fork_cleanup_io;
1173
1174         if (pid != &init_struct_pid) {
1175                 retval = -ENOMEM;
1176                 pid = alloc_pid(p->nsproxy->pid_ns);
1177                 if (!pid)
1178                         goto bad_fork_cleanup_io;
1179
1180                 if (clone_flags & CLONE_NEWPID) {
1181                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1182                         if (retval < 0)
1183                                 goto bad_fork_free_pid;
1184                 }
1185         }
1186
1187         p->pid = pid_nr(pid);
1188         p->tgid = p->pid;
1189         if (clone_flags & CLONE_THREAD)
1190                 p->tgid = current->tgid;
1191
1192         if (current->nsproxy != p->nsproxy) {
1193                 retval = ns_cgroup_clone(p, pid);
1194                 if (retval)
1195                         goto bad_fork_free_pid;
1196         }
1197
1198         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1199         /*
1200          * Clear TID on mm_release()?
1201          */
1202         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1203 #ifdef CONFIG_FUTEX
1204         p->robust_list = NULL;
1205 #ifdef CONFIG_COMPAT
1206         p->compat_robust_list = NULL;
1207 #endif
1208         INIT_LIST_HEAD(&p->pi_state_list);
1209         p->pi_state_cache = NULL;
1210 #endif
1211         /*
1212          * sigaltstack should be cleared when sharing the same VM
1213          */
1214         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1215                 p->sas_ss_sp = p->sas_ss_size = 0;
1216
1217         /*
1218          * Syscall tracing should be turned off in the child regardless
1219          * of CLONE_PTRACE.
1220          */
1221         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1222 #ifdef TIF_SYSCALL_EMU
1223         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1224 #endif
1225         clear_all_latency_tracing(p);
1226
1227         /* ok, now we should be set up.. */
1228         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1229         p->pdeath_signal = 0;
1230         p->exit_state = 0;
1231
1232         /*
1233          * Ok, make it visible to the rest of the system.
1234          * We dont wake it up yet.
1235          */
1236         p->group_leader = p;
1237         INIT_LIST_HEAD(&p->thread_group);
1238
1239         /* Now that the task is set up, run cgroup callbacks if
1240          * necessary. We need to run them before the task is visible
1241          * on the tasklist. */
1242         cgroup_fork_callbacks(p);
1243         cgroup_callbacks_done = 1;
1244
1245         /* Need tasklist lock for parent etc handling! */
1246         write_lock_irq(&tasklist_lock);
1247
1248         /*
1249          * The task hasn't been attached yet, so its cpus_allowed mask will
1250          * not be changed, nor will its assigned CPU.
1251          *
1252          * The cpus_allowed mask of the parent may have changed after it was
1253          * copied first time - so re-copy it here, then check the child's CPU
1254          * to ensure it is on a valid CPU (and if not, just force it back to
1255          * parent's CPU). This avoids alot of nasty races.
1256          */
1257         p->cpus_allowed = current->cpus_allowed;
1258         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1259         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1260                         !cpu_online(task_cpu(p))))
1261                 set_task_cpu(p, smp_processor_id());
1262
1263         /* CLONE_PARENT re-uses the old parent */
1264         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1265                 p->real_parent = current->real_parent;
1266                 p->parent_exec_id = current->parent_exec_id;
1267         } else {
1268                 p->real_parent = current;
1269                 p->parent_exec_id = current->self_exec_id;
1270         }
1271
1272         spin_lock(&current->sighand->siglock);
1273
1274         /*
1275          * Process group and session signals need to be delivered to just the
1276          * parent before the fork or both the parent and the child after the
1277          * fork. Restart if a signal comes in before we add the new process to
1278          * it's process group.
1279          * A fatal signal pending means that current will exit, so the new
1280          * thread can't slip out of an OOM kill (or normal SIGKILL).
1281          */
1282         recalc_sigpending();
1283         if (signal_pending(current)) {
1284                 spin_unlock(&current->sighand->siglock);
1285                 write_unlock_irq(&tasklist_lock);
1286                 retval = -ERESTARTNOINTR;
1287                 goto bad_fork_free_pid;
1288         }
1289
1290         if (clone_flags & CLONE_THREAD) {
1291                 atomic_inc(&current->signal->count);
1292                 atomic_inc(&current->signal->live);
1293                 p->group_leader = current->group_leader;
1294                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1295         }
1296
1297         if (likely(p->pid)) {
1298                 list_add_tail(&p->sibling, &p->real_parent->children);
1299                 tracehook_finish_clone(p, clone_flags, trace);
1300
1301                 if (thread_group_leader(p)) {
1302                         if (clone_flags & CLONE_NEWPID)
1303                                 p->nsproxy->pid_ns->child_reaper = p;
1304
1305                         p->signal->leader_pid = pid;
1306                         tty_kref_put(p->signal->tty);
1307                         p->signal->tty = tty_kref_get(current->signal->tty);
1308                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1309                         attach_pid(p, PIDTYPE_SID, task_session(current));
1310                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1311                         __get_cpu_var(process_counts)++;
1312                 }
1313                 attach_pid(p, PIDTYPE_PID, pid);
1314                 nr_threads++;
1315         }
1316
1317         total_forks++;
1318         spin_unlock(&current->sighand->siglock);
1319         write_unlock_irq(&tasklist_lock);
1320         proc_fork_connector(p);
1321         cgroup_post_fork(p);
1322         perf_event_fork(p);
1323         return p;
1324
1325 bad_fork_free_pid:
1326         if (pid != &init_struct_pid)
1327                 free_pid(pid);
1328 bad_fork_cleanup_io:
1329         put_io_context(p->io_context);
1330 bad_fork_cleanup_namespaces:
1331         exit_task_namespaces(p);
1332 bad_fork_cleanup_mm:
1333         if (p->mm)
1334                 mmput(p->mm);
1335 bad_fork_cleanup_signal:
1336         if (!(clone_flags & CLONE_THREAD))
1337                 __cleanup_signal(p->signal);
1338 bad_fork_cleanup_sighand:
1339         __cleanup_sighand(p->sighand);
1340 bad_fork_cleanup_fs:
1341         exit_fs(p); /* blocking */
1342 bad_fork_cleanup_files:
1343         exit_files(p); /* blocking */
1344 bad_fork_cleanup_semundo:
1345         exit_sem(p);
1346 bad_fork_cleanup_audit:
1347         audit_free(p);
1348 bad_fork_cleanup_policy:
1349         perf_event_free_task(p);
1350 #ifdef CONFIG_NUMA
1351         mpol_put(p->mempolicy);
1352 bad_fork_cleanup_cgroup:
1353 #endif
1354         cgroup_exit(p, cgroup_callbacks_done);
1355         delayacct_tsk_free(p);
1356         module_put(task_thread_info(p)->exec_domain->module);
1357 bad_fork_cleanup_count:
1358         atomic_dec(&p->cred->user->processes);
1359         exit_creds(p);
1360 bad_fork_free:
1361         free_task(p);
1362 fork_out:
1363         return ERR_PTR(retval);
1364 }
1365
1366 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1367 {
1368         memset(regs, 0, sizeof(struct pt_regs));
1369         return regs;
1370 }
1371
1372 struct task_struct * __cpuinit fork_idle(int cpu)
1373 {
1374         struct task_struct *task;
1375         struct pt_regs regs;
1376
1377         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1378                             &init_struct_pid, 0);
1379         if (!IS_ERR(task))
1380                 init_idle(task, cpu);
1381
1382         return task;
1383 }
1384
1385 /*
1386  *  Ok, this is the main fork-routine.
1387  *
1388  * It copies the process, and if successful kick-starts
1389  * it and waits for it to finish using the VM if required.
1390  */
1391 long do_fork(unsigned long clone_flags,
1392               unsigned long stack_start,
1393               struct pt_regs *regs,
1394               unsigned long stack_size,
1395               int __user *parent_tidptr,
1396               int __user *child_tidptr)
1397 {
1398         struct task_struct *p;
1399         int trace = 0;
1400         long nr;
1401
1402         /*
1403          * Do some preliminary argument and permissions checking before we
1404          * actually start allocating stuff
1405          */
1406         if (clone_flags & CLONE_NEWUSER) {
1407                 if (clone_flags & CLONE_THREAD)
1408                         return -EINVAL;
1409                 /* hopefully this check will go away when userns support is
1410                  * complete
1411                  */
1412                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1413                                 !capable(CAP_SETGID))
1414                         return -EPERM;
1415         }
1416
1417         /*
1418          * We hope to recycle these flags after 2.6.26
1419          */
1420         if (unlikely(clone_flags & CLONE_STOPPED)) {
1421                 static int __read_mostly count = 100;
1422
1423                 if (count > 0 && printk_ratelimit()) {
1424                         char comm[TASK_COMM_LEN];
1425
1426                         count--;
1427                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1428                                         "clone flags 0x%lx\n",
1429                                 get_task_comm(comm, current),
1430                                 clone_flags & CLONE_STOPPED);
1431                 }
1432         }
1433
1434         /*
1435          * When called from kernel_thread, don't do user tracing stuff.
1436          */
1437         if (likely(user_mode(regs)))
1438                 trace = tracehook_prepare_clone(clone_flags);
1439
1440         p = copy_process(clone_flags, stack_start, regs, stack_size,
1441                          child_tidptr, NULL, trace);
1442         /*
1443          * Do this prior waking up the new thread - the thread pointer
1444          * might get invalid after that point, if the thread exits quickly.
1445          */
1446         if (!IS_ERR(p)) {
1447                 struct completion vfork;
1448
1449                 trace_sched_process_fork(current, p);
1450
1451                 nr = task_pid_vnr(p);
1452
1453                 if (clone_flags & CLONE_PARENT_SETTID)
1454                         put_user(nr, parent_tidptr);
1455
1456                 if (clone_flags & CLONE_VFORK) {
1457                         p->vfork_done = &vfork;
1458                         init_completion(&vfork);
1459                 }
1460
1461                 audit_finish_fork(p);
1462                 tracehook_report_clone(regs, clone_flags, nr, p);
1463
1464                 /*
1465                  * We set PF_STARTING at creation in case tracing wants to
1466                  * use this to distinguish a fully live task from one that
1467                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1468                  * clear it and set the child going.
1469                  */
1470                 p->flags &= ~PF_STARTING;
1471
1472                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1473                         /*
1474                          * We'll start up with an immediate SIGSTOP.
1475                          */
1476                         sigaddset(&p->pending.signal, SIGSTOP);
1477                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1478                         __set_task_state(p, TASK_STOPPED);
1479                 } else {
1480                         wake_up_new_task(p, clone_flags);
1481                 }
1482
1483                 tracehook_report_clone_complete(trace, regs,
1484                                                 clone_flags, nr, p);
1485
1486                 if (clone_flags & CLONE_VFORK) {
1487                         freezer_do_not_count();
1488                         wait_for_completion(&vfork);
1489                         freezer_count();
1490                         tracehook_report_vfork_done(p, nr);
1491                 }
1492         } else {
1493                 nr = PTR_ERR(p);
1494         }
1495         return nr;
1496 }
1497
1498 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1499 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1500 #endif
1501
1502 static void sighand_ctor(void *data)
1503 {
1504         struct sighand_struct *sighand = data;
1505
1506         spin_lock_init(&sighand->siglock);
1507         init_waitqueue_head(&sighand->signalfd_wqh);
1508 }
1509
1510 void __init proc_caches_init(void)
1511 {
1512         sighand_cachep = kmem_cache_create("sighand_cache",
1513                         sizeof(struct sighand_struct), 0,
1514                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1515                         SLAB_NOTRACK, sighand_ctor);
1516         signal_cachep = kmem_cache_create("signal_cache",
1517                         sizeof(struct signal_struct), 0,
1518                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1519         files_cachep = kmem_cache_create("files_cache",
1520                         sizeof(struct files_struct), 0,
1521                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1522         fs_cachep = kmem_cache_create("fs_cache",
1523                         sizeof(struct fs_struct), 0,
1524                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1525         mm_cachep = kmem_cache_create("mm_struct",
1526                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1527                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1528         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1529         mmap_init();
1530 }
1531
1532 /*
1533  * Check constraints on flags passed to the unshare system call and
1534  * force unsharing of additional process context as appropriate.
1535  */
1536 static void check_unshare_flags(unsigned long *flags_ptr)
1537 {
1538         /*
1539          * If unsharing a thread from a thread group, must also
1540          * unshare vm.
1541          */
1542         if (*flags_ptr & CLONE_THREAD)
1543                 *flags_ptr |= CLONE_VM;
1544
1545         /*
1546          * If unsharing vm, must also unshare signal handlers.
1547          */
1548         if (*flags_ptr & CLONE_VM)
1549                 *flags_ptr |= CLONE_SIGHAND;
1550
1551         /*
1552          * If unsharing signal handlers and the task was created
1553          * using CLONE_THREAD, then must unshare the thread
1554          */
1555         if ((*flags_ptr & CLONE_SIGHAND) &&
1556             (atomic_read(&current->signal->count) > 1))
1557                 *flags_ptr |= CLONE_THREAD;
1558
1559         /*
1560          * If unsharing namespace, must also unshare filesystem information.
1561          */
1562         if (*flags_ptr & CLONE_NEWNS)
1563                 *flags_ptr |= CLONE_FS;
1564 }
1565
1566 /*
1567  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1568  */
1569 static int unshare_thread(unsigned long unshare_flags)
1570 {
1571         if (unshare_flags & CLONE_THREAD)
1572                 return -EINVAL;
1573
1574         return 0;
1575 }
1576
1577 /*
1578  * Unshare the filesystem structure if it is being shared
1579  */
1580 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1581 {
1582         struct fs_struct *fs = current->fs;
1583
1584         if (!(unshare_flags & CLONE_FS) || !fs)
1585                 return 0;
1586
1587         /* don't need lock here; in the worst case we'll do useless copy */
1588         if (fs->users == 1)
1589                 return 0;
1590
1591         *new_fsp = copy_fs_struct(fs);
1592         if (!*new_fsp)
1593                 return -ENOMEM;
1594
1595         return 0;
1596 }
1597
1598 /*
1599  * Unsharing of sighand is not supported yet
1600  */
1601 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1602 {
1603         struct sighand_struct *sigh = current->sighand;
1604
1605         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1606                 return -EINVAL;
1607         else
1608                 return 0;
1609 }
1610
1611 /*
1612  * Unshare vm if it is being shared
1613  */
1614 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1615 {
1616         struct mm_struct *mm = current->mm;
1617
1618         if ((unshare_flags & CLONE_VM) &&
1619             (mm && atomic_read(&mm->mm_users) > 1)) {
1620                 return -EINVAL;
1621         }
1622
1623         return 0;
1624 }
1625
1626 /*
1627  * Unshare file descriptor table if it is being shared
1628  */
1629 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1630 {
1631         struct files_struct *fd = current->files;
1632         int error = 0;
1633
1634         if ((unshare_flags & CLONE_FILES) &&
1635             (fd && atomic_read(&fd->count) > 1)) {
1636                 *new_fdp = dup_fd(fd, &error);
1637                 if (!*new_fdp)
1638                         return error;
1639         }
1640
1641         return 0;
1642 }
1643
1644 /*
1645  * unshare allows a process to 'unshare' part of the process
1646  * context which was originally shared using clone.  copy_*
1647  * functions used by do_fork() cannot be used here directly
1648  * because they modify an inactive task_struct that is being
1649  * constructed. Here we are modifying the current, active,
1650  * task_struct.
1651  */
1652 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1653 {
1654         int err = 0;
1655         struct fs_struct *fs, *new_fs = NULL;
1656         struct sighand_struct *new_sigh = NULL;
1657         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1658         struct files_struct *fd, *new_fd = NULL;
1659         struct nsproxy *new_nsproxy = NULL;
1660         int do_sysvsem = 0;
1661
1662         check_unshare_flags(&unshare_flags);
1663
1664         /* Return -EINVAL for all unsupported flags */
1665         err = -EINVAL;
1666         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1667                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1668                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1669                 goto bad_unshare_out;
1670
1671         /*
1672          * CLONE_NEWIPC must also detach from the undolist: after switching
1673          * to a new ipc namespace, the semaphore arrays from the old
1674          * namespace are unreachable.
1675          */
1676         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1677                 do_sysvsem = 1;
1678         if ((err = unshare_thread(unshare_flags)))
1679                 goto bad_unshare_out;
1680         if ((err = unshare_fs(unshare_flags, &new_fs)))
1681                 goto bad_unshare_cleanup_thread;
1682         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1683                 goto bad_unshare_cleanup_fs;
1684         if ((err = unshare_vm(unshare_flags, &new_mm)))
1685                 goto bad_unshare_cleanup_sigh;
1686         if ((err = unshare_fd(unshare_flags, &new_fd)))
1687                 goto bad_unshare_cleanup_vm;
1688         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1689                         new_fs)))
1690                 goto bad_unshare_cleanup_fd;
1691
1692         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1693                 if (do_sysvsem) {
1694                         /*
1695                          * CLONE_SYSVSEM is equivalent to sys_exit().
1696                          */
1697                         exit_sem(current);
1698                 }
1699
1700                 if (new_nsproxy) {
1701                         switch_task_namespaces(current, new_nsproxy);
1702                         new_nsproxy = NULL;
1703                 }
1704
1705                 task_lock(current);
1706
1707                 if (new_fs) {
1708                         fs = current->fs;
1709                         write_lock(&fs->lock);
1710                         current->fs = new_fs;
1711                         if (--fs->users)
1712                                 new_fs = NULL;
1713                         else
1714                                 new_fs = fs;
1715                         write_unlock(&fs->lock);
1716                 }
1717
1718                 if (new_mm) {
1719                         mm = current->mm;
1720                         active_mm = current->active_mm;
1721                         current->mm = new_mm;
1722                         current->active_mm = new_mm;
1723                         activate_mm(active_mm, new_mm);
1724                         new_mm = mm;
1725                 }
1726
1727                 if (new_fd) {
1728                         fd = current->files;
1729                         current->files = new_fd;
1730                         new_fd = fd;
1731                 }
1732
1733                 task_unlock(current);
1734         }
1735
1736         if (new_nsproxy)
1737                 put_nsproxy(new_nsproxy);
1738
1739 bad_unshare_cleanup_fd:
1740         if (new_fd)
1741                 put_files_struct(new_fd);
1742
1743 bad_unshare_cleanup_vm:
1744         if (new_mm)
1745                 mmput(new_mm);
1746
1747 bad_unshare_cleanup_sigh:
1748         if (new_sigh)
1749                 if (atomic_dec_and_test(&new_sigh->count))
1750                         kmem_cache_free(sighand_cachep, new_sigh);
1751
1752 bad_unshare_cleanup_fs:
1753         if (new_fs)
1754                 free_fs_struct(new_fs);
1755
1756 bad_unshare_cleanup_thread:
1757 bad_unshare_out:
1758         return err;
1759 }
1760
1761 /*
1762  *      Helper to unshare the files of the current task.
1763  *      We don't want to expose copy_files internals to
1764  *      the exec layer of the kernel.
1765  */
1766
1767 int unshare_files(struct files_struct **displaced)
1768 {
1769         struct task_struct *task = current;
1770         struct files_struct *copy = NULL;
1771         int error;
1772
1773         error = unshare_fd(CLONE_FILES, &copy);
1774         if (error || !copy) {
1775                 *displaced = NULL;
1776                 return error;
1777         }
1778         *displaced = task->files;
1779         task_lock(task);
1780         task->files = copy;
1781         task_unlock(task);
1782         return 0;
1783 }