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