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/smp_lock.h>
  18 #include <linux/module.h>
  19 #include <linux/vmalloc.h>
  20 #include <linux/completion.h>
  21 #include <linux/namespace.h>
  22 #include <linux/personality.h>
  23 #include <linux/mempolicy.h>
  24 #include <linux/sem.h>
  25 #include <linux/file.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/fs.h>
  30 #include <linux/capability.h>
  31 #include <linux/cpu.h>
  32 #include <linux/cpuset.h>
  33 #include <linux/security.h>
  34 #include <linux/swap.h>
  35 #include <linux/syscalls.h>
  36 #include <linux/jiffies.h>
  37 #include <linux/futex.h>
  38 #include <linux/rcupdate.h>
  39 #include <linux/ptrace.h>
  40 #include <linux/mount.h>
  41 #include <linux/audit.h>
  42 #include <linux/profile.h>
  43 #include <linux/rmap.h>
  44 #include <linux/acct.h>
  45 #include <linux/cn_proc.h>
  46 #include <linux/delayacct.h>
  47 #include <linux/taskstats_kern.h>
  48
  49 #include <asm/pgtable.h>
  50 #include <asm/pgalloc.h>
  51 #include <asm/uaccess.h>
  52 #include <asm/mmu_context.h>
  53 #include <asm/cacheflush.h>
  54 #include <asm/tlbflush.h>
  55
  56 /*
  57  * Protected counters by write_lock_irq(&tasklist_lock)
  58  */
  59 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  60 int nr_threads;                 /* The idle threads do not count.. */
  61
  62 int max_threads;                /* tunable limit on nr_threads */
  63
  64 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  65
  66 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  67
  68 int nr_processes(void)
  69 {
  70         int cpu;
  71         int total = 0;
  72
  73         for_each_online_cpu(cpu)
  74                 total += per_cpu(process_counts, cpu);
  75
  76         return total;
  77 }
  78
  79 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  80 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  81 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  82 static kmem_cache_t *task_struct_cachep;
  83 #endif
  84
  85 /* SLAB cache for signal_struct structures (tsk->signal) */
  86 static kmem_cache_t *signal_cachep;
  87
  88 /* SLAB cache for sighand_struct structures (tsk->sighand) */
  89 kmem_cache_t *sighand_cachep;
  90
  91 /* SLAB cache for files_struct structures (tsk->files) */
  92 kmem_cache_t *files_cachep;
  93
  94 /* SLAB cache for fs_struct structures (tsk->fs) */
  95 kmem_cache_t *fs_cachep;
  96
  97 /* SLAB cache for vm_area_struct structures */
  98 kmem_cache_t *vm_area_cachep;
  99
 100 /* SLAB cache for mm_struct structures (tsk->mm) */
 101 static kmem_cache_t *mm_cachep;
 102
 103 void free_task(struct task_struct *tsk)
 104 {
 105         free_thread_info(tsk->thread_info);
 106         rt_mutex_debug_task_free(tsk);
 107         free_task_struct(tsk);
 108 }
 109 EXPORT_SYMBOL(free_task);
 110
 111 void __put_task_struct(struct task_struct *tsk)
 112 {
 113         WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
 114         WARN_ON(atomic_read(&tsk->usage));
 115         WARN_ON(tsk == current);
 116
 117         security_task_free(tsk);
 118         free_uid(tsk->user);
 119         put_group_info(tsk->group_info);
 120
 121         if (!profile_handoff_task(tsk))
 122                 free_task(tsk);
 123 }
 124
 125 void __init fork_init(unsigned long mempages)
 126 {
 127 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 128 #ifndef ARCH_MIN_TASKALIGN
 129 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 130 #endif
 131         /* create a slab on which task_structs can be allocated */
 132         task_struct_cachep =
 133                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 134                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
 135 #endif
 136
 137         /*
 138          * The default maximum number of threads is set to a safe
 139          * value: the thread structures can take up at most half
 140          * of memory.
 141          */
 142         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 143
 144         /*
 145          * we need to allow at least 20 threads to boot a system
 146          */
 147         if(max_threads < 20)
 148                 max_threads = 20;
 149
 150         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 151         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 152         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 153                 init_task.signal->rlim[RLIMIT_NPROC];
 154 }
 155
 156 static struct task_struct *dup_task_struct(struct task_struct *orig)
 157 {
 158         struct task_struct *tsk;
 159         struct thread_info *ti;
 160
 161         prepare_to_copy(orig);
 162
 163         tsk = alloc_task_struct();
 164         if (!tsk)
 165                 return NULL;
 166
 167         ti = alloc_thread_info(tsk);
 168         if (!ti) {
 169                 free_task_struct(tsk);
 170                 return NULL;
 171         }
 172
 173         *tsk = *orig;
 174         tsk->thread_info = ti;
 175         setup_thread_stack(tsk, orig);
 176
 177         /* One for us, one for whoever does the "release_task()" (usually parent) */
 178         atomic_set(&tsk->usage,2);
 179         atomic_set(&tsk->fs_excl, 0);
 180         tsk->btrace_seq = 0;
 181         tsk->splice_pipe = NULL;
 182         return tsk;
 183 }
 184
 185 #ifdef CONFIG_MMU
 186 static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 187 {
 188         struct vm_area_struct *mpnt, *tmp, **pprev;
 189         struct rb_node **rb_link, *rb_parent;
 190         int retval;
 191         unsigned long charge;
 192         struct mempolicy *pol;
 193
 194         down_write(&oldmm->mmap_sem);
 195         flush_cache_mm(oldmm);
 196         /*
 197          * Not linked in yet - no deadlock potential:
 198          */
 199         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 200
 201         mm->locked_vm = 0;
 202         mm->mmap = NULL;
 203         mm->mmap_cache = NULL;
 204         mm->free_area_cache = oldmm->mmap_base;
 205         mm->cached_hole_size = ~0UL;
 206         mm->map_count = 0;
 207         cpus_clear(mm->cpu_vm_mask);
 208         mm->mm_rb = RB_ROOT;
 209         rb_link = &mm->mm_rb.rb_node;
 210         rb_parent = NULL;
 211         pprev = &mm->mmap;
 212
 213         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 214                 struct file *file;
 215
 216                 if (mpnt->vm_flags & VM_DONTCOPY) {
 217                         long pages = vma_pages(mpnt);
 218                         mm->total_vm -= pages;
 219                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 220                                                                 -pages);
 221                         continue;
 222                 }
 223                 charge = 0;
 224                 if (mpnt->vm_flags & VM_ACCOUNT) {
 225                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 226                         if (security_vm_enough_memory(len))
 227                                 goto fail_nomem;
 228                         charge = len;
 229                 }
 230                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 231                 if (!tmp)
 232                         goto fail_nomem;
 233                 *tmp = *mpnt;
 234                 pol = mpol_copy(vma_policy(mpnt));
 235                 retval = PTR_ERR(pol);
 236                 if (IS_ERR(pol))
 237                         goto fail_nomem_policy;
 238                 vma_set_policy(tmp, pol);
 239                 tmp->vm_flags &= ~VM_LOCKED;
 240                 tmp->vm_mm = mm;
 241                 tmp->vm_next = NULL;
 242                 anon_vma_link(tmp);
 243                 file = tmp->vm_file;
 244                 if (file) {
 245                         struct inode *inode = file->f_dentry->d_inode;
 246                         get_file(file);
 247                         if (tmp->vm_flags & VM_DENYWRITE)
 248                                 atomic_dec(&inode->i_writecount);
 249
 250                         /* insert tmp into the share list, just after mpnt */
 251                         spin_lock(&file->f_mapping->i_mmap_lock);
 252                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 253                         flush_dcache_mmap_lock(file->f_mapping);
 254                         vma_prio_tree_add(tmp, mpnt);
 255                         flush_dcache_mmap_unlock(file->f_mapping);
 256                         spin_unlock(&file->f_mapping->i_mmap_lock);
 257                 }
 258
 259                 /*
 260                  * Link in the new vma and copy the page table entries.
 261                  */
 262                 *pprev = tmp;
 263                 pprev = &tmp->vm_next;
 264
 265                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 266                 rb_link = &tmp->vm_rb.rb_right;
 267                 rb_parent = &tmp->vm_rb;
 268
 269                 mm->map_count++;
 270                 retval = copy_page_range(mm, oldmm, mpnt);
 271
 272                 if (tmp->vm_ops && tmp->vm_ops->open)
 273                         tmp->vm_ops->open(tmp);
 274
 275                 if (retval)
 276                         goto out;
 277         }
 278         retval = 0;
 279 out:
 280         up_write(&mm->mmap_sem);
 281         flush_tlb_mm(oldmm);
 282         up_write(&oldmm->mmap_sem);
 283         return retval;
 284 fail_nomem_policy:
 285         kmem_cache_free(vm_area_cachep, tmp);
 286 fail_nomem:
 287         retval = -ENOMEM;
 288         vm_unacct_memory(charge);
 289         goto out;
 290 }
 291
 292 static inline int mm_alloc_pgd(struct mm_struct * mm)
 293 {
 294         mm->pgd = pgd_alloc(mm);
 295         if (unlikely(!mm->pgd))
 296                 return -ENOMEM;
 297         return 0;
 298 }
 299
 300 static inline void mm_free_pgd(struct mm_struct * mm)
 301 {
 302         pgd_free(mm->pgd);
 303 }
 304 #else
 305 #define dup_mmap(mm, oldmm)     (0)
 306 #define mm_alloc_pgd(mm)        (0)
 307 #define mm_free_pgd(mm)
 308 #endif /* CONFIG_MMU */
 309
 310  __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 311
 312 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
 313 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 314
 315 #include <linux/init_task.h>
 316
 317 static struct mm_struct * mm_init(struct mm_struct * mm)
 318 {
 319         atomic_set(&mm->mm_users, 1);
 320         atomic_set(&mm->mm_count, 1);
 321         init_rwsem(&mm->mmap_sem);
 322         INIT_LIST_HEAD(&mm->mmlist);
 323         mm->core_waiters = 0;
 324         mm->nr_ptes = 0;
 325         set_mm_counter(mm, file_rss, 0);
 326         set_mm_counter(mm, anon_rss, 0);
 327         spin_lock_init(&mm->page_table_lock);
 328         rwlock_init(&mm->ioctx_list_lock);
 329         mm->ioctx_list = NULL;
 330         mm->free_area_cache = TASK_UNMAPPED_BASE;
 331         mm->cached_hole_size = ~0UL;
 332
 333         if (likely(!mm_alloc_pgd(mm))) {
 334                 mm->def_flags = 0;
 335                 return mm;
 336         }
 337         free_mm(mm);
 338         return NULL;
 339 }
 340
 341 /*
 342  * Allocate and initialize an mm_struct.
 343  */
 344 struct mm_struct * mm_alloc(void)
 345 {
 346         struct mm_struct * mm;
 347
 348         mm = allocate_mm();
 349         if (mm) {
 350                 memset(mm, 0, sizeof(*mm));
 351                 mm = mm_init(mm);
 352         }
 353         return mm;
 354 }
 355
 356 /*
 357  * Called when the last reference to the mm
 358  * is dropped: either by a lazy thread or by
 359  * mmput. Free the page directory and the mm.
 360  */
 361 void fastcall __mmdrop(struct mm_struct *mm)
 362 {
 363         BUG_ON(mm == &init_mm);
 364         mm_free_pgd(mm);
 365         destroy_context(mm);
 366         free_mm(mm);
 367 }
 368
 369 /*
 370  * Decrement the use count and release all resources for an mm.
 371  */
 372 void mmput(struct mm_struct *mm)
 373 {
 374         might_sleep();
 375
 376         if (atomic_dec_and_test(&mm->mm_users)) {
 377                 exit_aio(mm);
 378                 exit_mmap(mm);
 379                 if (!list_empty(&mm->mmlist)) {
 380                         spin_lock(&mmlist_lock);
 381                         list_del(&mm->mmlist);
 382                         spin_unlock(&mmlist_lock);
 383                 }
 384                 put_swap_token(mm);
 385                 mmdrop(mm);
 386         }
 387 }
 388 EXPORT_SYMBOL_GPL(mmput);
 389
 390 /**
 391  * get_task_mm - acquire a reference to the task's mm
 392  *
 393  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
 394  * this kernel workthread has transiently adopted a user mm with use_mm,
 395  * to do its AIO) is not set and if so returns a reference to it, after
 396  * bumping up the use count.  User must release the mm via mmput()
 397  * after use.  Typically used by /proc and ptrace.
 398  */
 399 struct mm_struct *get_task_mm(struct task_struct *task)
 400 {
 401         struct mm_struct *mm;
 402
 403         task_lock(task);
 404         mm = task->mm;
 405         if (mm) {
 406                 if (task->flags & PF_BORROWED_MM)
 407                         mm = NULL;
 408                 else
 409                         atomic_inc(&mm->mm_users);
 410         }
 411         task_unlock(task);
 412         return mm;
 413 }
 414 EXPORT_SYMBOL_GPL(get_task_mm);
 415
 416 /* Please note the differences between mmput and mm_release.
 417  * mmput is called whenever we stop holding onto a mm_struct,
 418  * error success whatever.
 419  *
 420  * mm_release is called after a mm_struct has been removed
 421  * from the current process.
 422  *
 423  * This difference is important for error handling, when we
 424  * only half set up a mm_struct for a new process and need to restore
 425  * the old one.  Because we mmput the new mm_struct before
 426  * restoring the old one. . .
 427  * Eric Biederman 10 January 1998
 428  */
 429 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 430 {
 431         struct completion *vfork_done = tsk->vfork_done;
 432
 433         /* Get rid of any cached register state */
 434         deactivate_mm(tsk, mm);
 435
 436         /* notify parent sleeping on vfork() */
 437         if (vfork_done) {
 438                 tsk->vfork_done = NULL;
 439                 complete(vfork_done);
 440         }
 441         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
 442                 u32 __user * tidptr = tsk->clear_child_tid;
 443                 tsk->clear_child_tid = NULL;
 444
 445                 /*
 446                  * We don't check the error code - if userspace has
 447                  * not set up a proper pointer then tough luck.
 448                  */
 449                 put_user(0, tidptr);
 450                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 451         }
 452 }
 453
 454 /*
 455  * Allocate a new mm structure and copy contents from the
 456  * mm structure of the passed in task structure.
 457  */
 458 static struct mm_struct *dup_mm(struct task_struct *tsk)
 459 {
 460         struct mm_struct *mm, *oldmm = current->mm;
 461         int err;
 462
 463         if (!oldmm)
 464                 return NULL;
 465
 466         mm = allocate_mm();
 467         if (!mm)
 468                 goto fail_nomem;
 469
 470         memcpy(mm, oldmm, sizeof(*mm));
 471
 472         if (!mm_init(mm))
 473                 goto fail_nomem;
 474
 475         if (init_new_context(tsk, mm))
 476                 goto fail_nocontext;
 477
 478         err = dup_mmap(mm, oldmm);
 479         if (err)
 480                 goto free_pt;
 481
 482         mm->hiwater_rss = get_mm_rss(mm);
 483         mm->hiwater_vm = mm->total_vm;
 484
 485         return mm;
 486
 487 free_pt:
 488         mmput(mm);
 489
 490 fail_nomem:
 491         return NULL;
 492
 493 fail_nocontext:
 494         /*
 495          * If init_new_context() failed, we cannot use mmput() to free the mm
 496          * because it calls destroy_context()
 497          */
 498         mm_free_pgd(mm);
 499         free_mm(mm);
 500         return NULL;
 501 }
 502
 503 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 504 {
 505         struct mm_struct * mm, *oldmm;
 506         int retval;
 507
 508         tsk->min_flt = tsk->maj_flt = 0;
 509         tsk->nvcsw = tsk->nivcsw = 0;
 510
 511         tsk->mm = NULL;
 512         tsk->active_mm = NULL;
 513
 514         /*
 515          * Are we cloning a kernel thread?
 516          *
 517          * We need to steal a active VM for that..
 518          */
 519         oldmm = current->mm;
 520         if (!oldmm)
 521                 return 0;
 522
 523         if (clone_flags & CLONE_VM) {
 524                 atomic_inc(&oldmm->mm_users);
 525                 mm = oldmm;
 526                 goto good_mm;
 527         }
 528
 529         retval = -ENOMEM;
 530         mm = dup_mm(tsk);
 531         if (!mm)
 532                 goto fail_nomem;
 533
 534 good_mm:
 535         tsk->mm = mm;
 536         tsk->active_mm = mm;
 537         return 0;
 538
 539 fail_nomem:
 540         return retval;
 541 }
 542
 543 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 544 {
 545         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 546         /* We don't need to lock fs - think why ;-) */
 547         if (fs) {
 548                 atomic_set(&fs->count, 1);
 549                 rwlock_init(&fs->lock);
 550                 fs->umask = old->umask;
 551                 read_lock(&old->lock);
 552                 fs->rootmnt = mntget(old->rootmnt);
 553                 fs->root = dget(old->root);
 554                 fs->pwdmnt = mntget(old->pwdmnt);
 555                 fs->pwd = dget(old->pwd);
 556                 if (old->altroot) {
 557                         fs->altrootmnt = mntget(old->altrootmnt);
 558                         fs->altroot = dget(old->altroot);
 559                 } else {
 560                         fs->altrootmnt = NULL;
 561                         fs->altroot = NULL;
 562                 }
 563                 read_unlock(&old->lock);
 564         }
 565         return fs;
 566 }
 567
 568 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 569 {
 570         return __copy_fs_struct(old);
 571 }
 572
 573 EXPORT_SYMBOL_GPL(copy_fs_struct);
 574
 575 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 576 {
 577         if (clone_flags & CLONE_FS) {
 578                 atomic_inc(&current->fs->count);
 579                 return 0;
 580         }
 581         tsk->fs = __copy_fs_struct(current->fs);
 582         if (!tsk->fs)
 583                 return -ENOMEM;
 584         return 0;
 585 }
 586
 587 static int count_open_files(struct fdtable *fdt)
 588 {
 589         int size = fdt->max_fdset;
 590         int i;
 591
 592         /* Find the last open fd */
 593         for (i = size/(8*sizeof(long)); i > 0; ) {
 594                 if (fdt->open_fds->fds_bits[--i])
 595                         break;
 596         }
 597         i = (i+1) * 8 * sizeof(long);
 598         return i;
 599 }
 600
 601 static struct files_struct *alloc_files(void)
 602 {
 603         struct files_struct *newf;
 604         struct fdtable *fdt;
 605
 606         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
 607         if (!newf)
 608                 goto out;
 609
 610         atomic_set(&newf->count, 1);
 611
 612         spin_lock_init(&newf->file_lock);
 613         newf->next_fd = 0;
 614         fdt = &newf->fdtab;
 615         fdt->max_fds = NR_OPEN_DEFAULT;
 616         fdt->max_fdset = EMBEDDED_FD_SET_SIZE;
 617         fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
 618         fdt->open_fds = (fd_set *)&newf->open_fds_init;
 619         fdt->fd = &newf->fd_array[0];
 620         INIT_RCU_HEAD(&fdt->rcu);
 621         fdt->free_files = NULL;
 622         fdt->next = NULL;
 623         rcu_assign_pointer(newf->fdt, fdt);
 624 out:
 625         return newf;
 626 }
 627
 628 /*
 629  * Allocate a new files structure and copy contents from the
 630  * passed in files structure.
 631  * errorp will be valid only when the returned files_struct is NULL.
 632  */
 633 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
 634 {
 635         struct files_struct *newf;
 636         struct file **old_fds, **new_fds;
 637         int open_files, size, i, expand;
 638         struct fdtable *old_fdt, *new_fdt;
 639
 640         *errorp = -ENOMEM;
 641         newf = alloc_files();
 642         if (!newf)
 643                 goto out;
 644
 645         spin_lock(&oldf->file_lock);
 646         old_fdt = files_fdtable(oldf);
 647         new_fdt = files_fdtable(newf);
 648         size = old_fdt->max_fdset;
 649         open_files = count_open_files(old_fdt);
 650         expand = 0;
 651
 652         /*
 653          * Check whether we need to allocate a larger fd array or fd set.
 654          * Note: we're not a clone task, so the open count won't  change.
 655          */
 656         if (open_files > new_fdt->max_fdset) {
 657                 new_fdt->max_fdset = 0;
 658                 expand = 1;
 659         }
 660         if (open_files > new_fdt->max_fds) {
 661                 new_fdt->max_fds = 0;
 662                 expand = 1;
 663         }
 664
 665         /* if the old fdset gets grown now, we'll only copy up to "size" fds */
 666         if (expand) {
 667                 spin_unlock(&oldf->file_lock);
 668                 spin_lock(&newf->file_lock);
 669                 *errorp = expand_files(newf, open_files-1);
 670                 spin_unlock(&newf->file_lock);
 671                 if (*errorp < 0)
 672                         goto out_release;
 673                 new_fdt = files_fdtable(newf);
 674                 /*
 675                  * Reacquire the oldf lock and a pointer to its fd table
 676                  * who knows it may have a new bigger fd table. We need
 677                  * the latest pointer.
 678                  */
 679                 spin_lock(&oldf->file_lock);
 680                 old_fdt = files_fdtable(oldf);
 681         }
 682
 683         old_fds = old_fdt->fd;
 684         new_fds = new_fdt->fd;
 685
 686         memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
 687         memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
 688
 689         for (i = open_files; i != 0; i--) {
 690                 struct file *f = *old_fds++;
 691                 if (f) {
 692                         get_file(f);
 693                 } else {
 694                         /*
 695                          * The fd may be claimed in the fd bitmap but not yet
 696                          * instantiated in the files array if a sibling thread
 697                          * is partway through open().  So make sure that this
 698                          * fd is available to the new process.
 699                          */
 700                         FD_CLR(open_files - i, new_fdt->open_fds);
 701                 }
 702                 rcu_assign_pointer(*new_fds++, f);
 703         }
 704         spin_unlock(&oldf->file_lock);
 705
 706         /* compute the remainder to be cleared */
 707         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
 708
 709         /* This is long word aligned thus could use a optimized version */
 710         memset(new_fds, 0, size);
 711
 712         if (new_fdt->max_fdset > open_files) {
 713                 int left = (new_fdt->max_fdset-open_files)/8;
 714                 int start = open_files / (8 * sizeof(unsigned long));
 715
 716                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
 717                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
 718         }
 719
 720 out:
 721         return newf;
 722
 723 out_release:
 724         free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
 725         free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
 726         free_fd_array(new_fdt->fd, new_fdt->max_fds);
 727         kmem_cache_free(files_cachep, newf);
 728         return NULL;
 729 }
 730
 731 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 732 {
 733         struct files_struct *oldf, *newf;
 734         int error = 0;
 735
 736         /*
 737          * A background process may not have any files ...
 738          */
 739         oldf = current->files;
 740         if (!oldf)
 741                 goto out;
 742
 743         if (clone_flags & CLONE_FILES) {
 744                 atomic_inc(&oldf->count);
 745                 goto out;
 746         }
 747
 748         /*
 749          * Note: we may be using current for both targets (See exec.c)
 750          * This works because we cache current->files (old) as oldf. Don't
 751          * break this.
 752          */
 753         tsk->files = NULL;
 754         newf = dup_fd(oldf, &error);
 755         if (!newf)
 756                 goto out;
 757
 758         tsk->files = newf;
 759         error = 0;
 760 out:
 761         return error;
 762 }
 763
 764 /*
 765  *      Helper to unshare the files of the current task.
 766  *      We don't want to expose copy_files internals to
 767  *      the exec layer of the kernel.
 768  */
 769
 770 int unshare_files(void)
 771 {
 772         struct files_struct *files  = current->files;
 773         int rc;
 774
 775         BUG_ON(!files);
 776
 777         /* This can race but the race causes us to copy when we don't
 778            need to and drop the copy */
 779         if(atomic_read(&files->count) == 1)
 780         {
 781                 atomic_inc(&files->count);
 782                 return 0;
 783         }
 784         rc = copy_files(0, current);
 785         if(rc)
 786                 current->files = files;
 787         return rc;
 788 }
 789
 790 EXPORT_SYMBOL(unshare_files);
 791
 792 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 793 {
 794         struct sighand_struct *sig;
 795
 796         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 797                 atomic_inc(&current->sighand->count);
 798                 return 0;
 799         }
 800         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 801         rcu_assign_pointer(tsk->sighand, sig);
 802         if (!sig)
 803                 return -ENOMEM;
 804         atomic_set(&sig->count, 1);
 805         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 806         return 0;
 807 }
 808
 809 void __cleanup_sighand(struct sighand_struct *sighand)
 810 {
 811         if (atomic_dec_and_test(&sighand->count))
 812                 kmem_cache_free(sighand_cachep, sighand);
 813 }
 814
 815 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 816 {
 817         struct signal_struct *sig;
 818         int ret;
 819
 820         if (clone_flags & CLONE_THREAD) {
 821                 atomic_inc(&current->signal->count);
 822                 atomic_inc(&current->signal->live);
 823                 taskstats_tgid_alloc(current->signal);
 824                 return 0;
 825         }
 826         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 827         tsk->signal = sig;
 828         if (!sig)
 829                 return -ENOMEM;
 830
 831         ret = copy_thread_group_keys(tsk);
 832         if (ret < 0) {
 833                 kmem_cache_free(signal_cachep, sig);
 834                 return ret;
 835         }
 836
 837         atomic_set(&sig->count, 1);
 838         atomic_set(&sig->live, 1);
 839         init_waitqueue_head(&sig->wait_chldexit);
 840         sig->flags = 0;
 841         sig->group_exit_code = 0;
 842         sig->group_exit_task = NULL;
 843         sig->group_stop_count = 0;
 844         sig->curr_target = NULL;
 845         init_sigpending(&sig->shared_pending);
 846         INIT_LIST_HEAD(&sig->posix_timers);
 847
 848         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_REL);
 849         sig->it_real_incr.tv64 = 0;
 850         sig->real_timer.function = it_real_fn;
 851         sig->tsk = tsk;
 852
 853         sig->it_virt_expires = cputime_zero;
 854         sig->it_virt_incr = cputime_zero;
 855         sig->it_prof_expires = cputime_zero;
 856         sig->it_prof_incr = cputime_zero;
 857
 858         sig->leader = 0;        /* session leadership doesn't inherit */
 859         sig->tty_old_pgrp = 0;
 860
 861         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 862         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 863         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 864         sig->sched_time = 0;
 865         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 866         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 867         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 868         taskstats_tgid_init(sig);
 869
 870         task_lock(current->group_leader);
 871         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 872         task_unlock(current->group_leader);
 873
 874         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 875                 /*
 876                  * New sole thread in the process gets an expiry time
 877                  * of the whole CPU time limit.
 878                  */
 879                 tsk->it_prof_expires =
 880                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 881         }
 882         acct_init_pacct(&sig->pacct);
 883
 884         return 0;
 885 }
 886
 887 void __cleanup_signal(struct signal_struct *sig)
 888 {
 889         exit_thread_group_keys(sig);
 890         taskstats_tgid_free(sig);
 891         kmem_cache_free(signal_cachep, sig);
 892 }
 893
 894 static inline void cleanup_signal(struct task_struct *tsk)
 895 {
 896         struct signal_struct *sig = tsk->signal;
 897
 898         atomic_dec(&sig->live);
 899
 900         if (atomic_dec_and_test(&sig->count))
 901                 __cleanup_signal(sig);
 902 }
 903
 904 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 905 {
 906         unsigned long new_flags = p->flags;
 907
 908         new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
 909         new_flags |= PF_FORKNOEXEC;
 910         if (!(clone_flags & CLONE_PTRACE))
 911                 p->ptrace = 0;
 912         p->flags = new_flags;
 913 }
 914
 915 asmlinkage long sys_set_tid_address(int __user *tidptr)
 916 {
 917         current->clear_child_tid = tidptr;
 918
 919         return current->pid;
 920 }
 921
 922 static inline void rt_mutex_init_task(struct task_struct *p)
 923 {
 924 #ifdef CONFIG_RT_MUTEXES
 925         spin_lock_init(&p->pi_lock);
 926         plist_head_init(&p->pi_waiters, &p->pi_lock);
 927         p->pi_blocked_on = NULL;
 928 #endif
 929 }
 930
 931 /*
 932  * This creates a new process as a copy of the old one,
 933  * but does not actually start it yet.
 934  *
 935  * It copies the registers, and all the appropriate
 936  * parts of the process environment (as per the clone
 937  * flags). The actual kick-off is left to the caller.
 938  */
 939 static struct task_struct *copy_process(unsigned long clone_flags,
 940                                         unsigned long stack_start,
 941                                         struct pt_regs *regs,
 942                                         unsigned long stack_size,
 943                                         int __user *parent_tidptr,
 944                                         int __user *child_tidptr,
 945                                         int pid)
 946 {
 947         int retval;
 948         struct task_struct *p = NULL;
 949
 950         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 951                 return ERR_PTR(-EINVAL);
 952
 953         /*
 954          * Thread groups must share signals as well, and detached threads
 955          * can only be started up within the thread group.
 956          */
 957         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 958                 return ERR_PTR(-EINVAL);
 959
 960         /*
 961          * Shared signal handlers imply shared VM. By way of the above,
 962          * thread groups also imply shared VM. Blocking this case allows
 963          * for various simplifications in other code.
 964          */
 965         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 966                 return ERR_PTR(-EINVAL);
 967
 968         retval = security_task_create(clone_flags);
 969         if (retval)
 970                 goto fork_out;
 971
 972         retval = -ENOMEM;
 973         p = dup_task_struct(current);
 974         if (!p)
 975                 goto fork_out;
 976
 977 #ifdef CONFIG_TRACE_IRQFLAGS
 978         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
 979         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
 980 #endif
 981         retval = -EAGAIN;
 982         if (atomic_read(&p->user->processes) >=
 983                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
 984                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 985                                 p->user != &root_user)
 986                         goto bad_fork_free;
 987         }
 988
 989         atomic_inc(&p->user->__count);
 990         atomic_inc(&p->user->processes);
 991         get_group_info(p->group_info);
 992
 993         /*
 994          * If multiple threads are within copy_process(), then this check
 995          * triggers too late. This doesn't hurt, the check is only there
 996          * to stop root fork bombs.
 997          */
 998         if (nr_threads >= max_threads)
 999                 goto bad_fork_cleanup_count;
1000
1001         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1002                 goto bad_fork_cleanup_count;
1003
1004         if (p->binfmt && !try_module_get(p->binfmt->module))
1005                 goto bad_fork_cleanup_put_domain;
1006
1007         p->did_exec = 0;
1008         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1009         copy_flags(clone_flags, p);
1010         p->pid = pid;
1011         retval = -EFAULT;
1012         if (clone_flags & CLONE_PARENT_SETTID)
1013                 if (put_user(p->pid, parent_tidptr))
1014                         goto bad_fork_cleanup;
1015
1016         INIT_LIST_HEAD(&p->children);
1017         INIT_LIST_HEAD(&p->sibling);
1018         p->vfork_done = NULL;
1019         spin_lock_init(&p->alloc_lock);
1020
1021         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1022         init_sigpending(&p->pending);
1023
1024         p->utime = cputime_zero;
1025         p->stime = cputime_zero;
1026         p->sched_time = 0;
1027         p->rchar = 0;           /* I/O counter: bytes read */
1028         p->wchar = 0;           /* I/O counter: bytes written */
1029         p->syscr = 0;           /* I/O counter: read syscalls */
1030         p->syscw = 0;           /* I/O counter: write syscalls */
1031         acct_clear_integrals(p);
1032
1033         p->it_virt_expires = cputime_zero;
1034         p->it_prof_expires = cputime_zero;
1035         p->it_sched_expires = 0;
1036         INIT_LIST_HEAD(&p->cpu_timers[0]);
1037         INIT_LIST_HEAD(&p->cpu_timers[1]);
1038         INIT_LIST_HEAD(&p->cpu_timers[2]);
1039
1040         p->lock_depth = -1;             /* -1 = no lock */
1041         do_posix_clock_monotonic_gettime(&p->start_time);
1042         p->security = NULL;
1043         p->io_context = NULL;
1044         p->io_wait = NULL;
1045         p->audit_context = NULL;
1046         cpuset_fork(p);
1047 #ifdef CONFIG_NUMA
1048         p->mempolicy = mpol_copy(p->mempolicy);
1049         if (IS_ERR(p->mempolicy)) {
1050                 retval = PTR_ERR(p->mempolicy);
1051                 p->mempolicy = NULL;
1052                 goto bad_fork_cleanup_cpuset;
1053         }
1054         mpol_fix_fork_child_flag(p);
1055 #endif
1056 #ifdef CONFIG_TRACE_IRQFLAGS
1057         p->irq_events = 0;
1058         p->hardirqs_enabled = 0;
1059         p->hardirq_enable_ip = 0;
1060         p->hardirq_enable_event = 0;
1061         p->hardirq_disable_ip = _THIS_IP_;
1062         p->hardirq_disable_event = 0;
1063         p->softirqs_enabled = 1;
1064         p->softirq_enable_ip = _THIS_IP_;
1065         p->softirq_enable_event = 0;
1066         p->softirq_disable_ip = 0;
1067         p->softirq_disable_event = 0;
1068         p->hardirq_context = 0;
1069         p->softirq_context = 0;
1070 #endif
1071 #ifdef CONFIG_LOCKDEP
1072         p->lockdep_depth = 0; /* no locks held yet */
1073         p->curr_chain_key = 0;
1074         p->lockdep_recursion = 0;
1075 #endif
1076
1077         rt_mutex_init_task(p);
1078
1079 #ifdef CONFIG_DEBUG_MUTEXES
1080         p->blocked_on = NULL; /* not blocked yet */
1081 #endif
1082
1083         p->tgid = p->pid;
1084         if (clone_flags & CLONE_THREAD)
1085                 p->tgid = current->tgid;
1086
1087         if ((retval = security_task_alloc(p)))
1088                 goto bad_fork_cleanup_policy;
1089         if ((retval = audit_alloc(p)))
1090                 goto bad_fork_cleanup_security;
1091         /* copy all the process information */
1092         if ((retval = copy_semundo(clone_flags, p)))
1093                 goto bad_fork_cleanup_audit;
1094         if ((retval = copy_files(clone_flags, p)))
1095                 goto bad_fork_cleanup_semundo;
1096         if ((retval = copy_fs(clone_flags, p)))
1097                 goto bad_fork_cleanup_files;
1098         if ((retval = copy_sighand(clone_flags, p)))
1099                 goto bad_fork_cleanup_fs;
1100         if ((retval = copy_signal(clone_flags, p)))
1101                 goto bad_fork_cleanup_sighand;
1102         if ((retval = copy_mm(clone_flags, p)))
1103                 goto bad_fork_cleanup_signal;
1104         if ((retval = copy_keys(clone_flags, p)))
1105                 goto bad_fork_cleanup_mm;
1106         if ((retval = copy_namespace(clone_flags, p)))
1107                 goto bad_fork_cleanup_keys;
1108         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1109         if (retval)
1110                 goto bad_fork_cleanup_namespace;
1111
1112         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1113         /*
1114          * Clear TID on mm_release()?
1115          */
1116         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1117         p->robust_list = NULL;
1118 #ifdef CONFIG_COMPAT
1119         p->compat_robust_list = NULL;
1120 #endif
1121         INIT_LIST_HEAD(&p->pi_state_list);
1122         p->pi_state_cache = NULL;
1123
1124         /*
1125          * sigaltstack should be cleared when sharing the same VM
1126          */
1127         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1128                 p->sas_ss_sp = p->sas_ss_size = 0;
1129
1130         /*
1131          * Syscall tracing should be turned off in the child regardless
1132          * of CLONE_PTRACE.
1133          */
1134         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1135 #ifdef TIF_SYSCALL_EMU
1136         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1137 #endif
1138
1139         /* Our parent execution domain becomes current domain
1140            These must match for thread signalling to apply */
1141
1142         p->parent_exec_id = p->self_exec_id;
1143
1144         /* ok, now we should be set up.. */
1145         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1146         p->pdeath_signal = 0;
1147         p->exit_state = 0;
1148
1149         /*
1150          * Ok, make it visible to the rest of the system.
1151          * We dont wake it up yet.
1152          */
1153         p->group_leader = p;
1154         INIT_LIST_HEAD(&p->thread_group);
1155         INIT_LIST_HEAD(&p->ptrace_children);
1156         INIT_LIST_HEAD(&p->ptrace_list);
1157
1158         /* Perform scheduler related setup. Assign this task to a CPU. */
1159         sched_fork(p, clone_flags);
1160
1161         /* Need tasklist lock for parent etc handling! */
1162         write_lock_irq(&tasklist_lock);
1163
1164         /*
1165          * The task hasn't been attached yet, so its cpus_allowed mask will
1166          * not be changed, nor will its assigned CPU.
1167          *
1168          * The cpus_allowed mask of the parent may have changed after it was
1169          * copied first time - so re-copy it here, then check the child's CPU
1170          * to ensure it is on a valid CPU (and if not, just force it back to
1171          * parent's CPU). This avoids alot of nasty races.
1172          */
1173         p->cpus_allowed = current->cpus_allowed;
1174         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1175                         !cpu_online(task_cpu(p))))
1176                 set_task_cpu(p, smp_processor_id());
1177
1178         /* CLONE_PARENT re-uses the old parent */
1179         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1180                 p->real_parent = current->real_parent;
1181         else
1182                 p->real_parent = current;
1183         p->parent = p->real_parent;
1184
1185         spin_lock(&current->sighand->siglock);
1186
1187         /*
1188          * Process group and session signals need to be delivered to just the
1189          * parent before the fork or both the parent and the child after the
1190          * fork. Restart if a signal comes in before we add the new process to
1191          * it's process group.
1192          * A fatal signal pending means that current will exit, so the new
1193          * thread can't slip out of an OOM kill (or normal SIGKILL).
1194          */
1195         recalc_sigpending();
1196         if (signal_pending(current)) {
1197                 spin_unlock(&current->sighand->siglock);
1198                 write_unlock_irq(&tasklist_lock);
1199                 retval = -ERESTARTNOINTR;
1200                 goto bad_fork_cleanup_namespace;
1201         }
1202
1203         if (clone_flags & CLONE_THREAD) {
1204                 p->group_leader = current->group_leader;
1205                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1206
1207                 if (!cputime_eq(current->signal->it_virt_expires,
1208                                 cputime_zero) ||
1209                     !cputime_eq(current->signal->it_prof_expires,
1210                                 cputime_zero) ||
1211                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1212                     !list_empty(&current->signal->cpu_timers[0]) ||
1213                     !list_empty(&current->signal->cpu_timers[1]) ||
1214                     !list_empty(&current->signal->cpu_timers[2])) {
1215                         /*
1216                          * Have child wake up on its first tick to check
1217                          * for process CPU timers.
1218                          */
1219                         p->it_prof_expires = jiffies_to_cputime(1);
1220                 }
1221         }
1222
1223         /*
1224          * inherit ioprio
1225          */
1226         p->ioprio = current->ioprio;
1227
1228         if (likely(p->pid)) {
1229                 add_parent(p);
1230                 if (unlikely(p->ptrace & PT_PTRACED))
1231                         __ptrace_link(p, current->parent);
1232
1233                 if (thread_group_leader(p)) {
1234                         p->signal->tty = current->signal->tty;
1235                         p->signal->pgrp = process_group(current);
1236                         p->signal->session = current->signal->session;
1237                         attach_pid(p, PIDTYPE_PGID, process_group(p));
1238                         attach_pid(p, PIDTYPE_SID, p->signal->session);
1239
1240                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1241                         __get_cpu_var(process_counts)++;
1242                 }
1243                 attach_pid(p, PIDTYPE_PID, p->pid);
1244                 nr_threads++;
1245         }
1246
1247         total_forks++;
1248         spin_unlock(&current->sighand->siglock);
1249         write_unlock_irq(&tasklist_lock);
1250         proc_fork_connector(p);
1251         return p;
1252
1253 bad_fork_cleanup_namespace:
1254         exit_namespace(p);
1255 bad_fork_cleanup_keys:
1256         exit_keys(p);
1257 bad_fork_cleanup_mm:
1258         if (p->mm)
1259                 mmput(p->mm);
1260 bad_fork_cleanup_signal:
1261         cleanup_signal(p);
1262 bad_fork_cleanup_sighand:
1263         __cleanup_sighand(p->sighand);
1264 bad_fork_cleanup_fs:
1265         exit_fs(p); /* blocking */
1266 bad_fork_cleanup_files:
1267         exit_files(p); /* blocking */
1268 bad_fork_cleanup_semundo:
1269         exit_sem(p);
1270 bad_fork_cleanup_audit:
1271         audit_free(p);
1272 bad_fork_cleanup_security:
1273         security_task_free(p);
1274 bad_fork_cleanup_policy:
1275 #ifdef CONFIG_NUMA
1276         mpol_free(p->mempolicy);
1277 bad_fork_cleanup_cpuset:
1278 #endif
1279         cpuset_exit(p);
1280 bad_fork_cleanup:
1281         if (p->binfmt)
1282                 module_put(p->binfmt->module);
1283 bad_fork_cleanup_put_domain:
1284         module_put(task_thread_info(p)->exec_domain->module);
1285 bad_fork_cleanup_count:
1286         put_group_info(p->group_info);
1287         atomic_dec(&p->user->processes);
1288         free_uid(p->user);
1289 bad_fork_free:
1290         free_task(p);
1291 fork_out:
1292         return ERR_PTR(retval);
1293 }
1294
1295 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1296 {
1297         memset(regs, 0, sizeof(struct pt_regs));
1298         return regs;
1299 }
1300
1301 struct task_struct * __devinit fork_idle(int cpu)
1302 {
1303         struct task_struct *task;
1304         struct pt_regs regs;
1305
1306         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1307         if (!task)
1308                 return ERR_PTR(-ENOMEM);
1309         init_idle(task, cpu);
1310
1311         return task;
1312 }
1313
1314 static inline int fork_traceflag (unsigned clone_flags)
1315 {
1316         if (clone_flags & CLONE_UNTRACED)
1317                 return 0;
1318         else if (clone_flags & CLONE_VFORK) {
1319                 if (current->ptrace & PT_TRACE_VFORK)
1320                         return PTRACE_EVENT_VFORK;
1321         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1322                 if (current->ptrace & PT_TRACE_CLONE)
1323                         return PTRACE_EVENT_CLONE;
1324         } else if (current->ptrace & PT_TRACE_FORK)
1325                 return PTRACE_EVENT_FORK;
1326
1327         return 0;
1328 }
1329
1330 /*
1331  *  Ok, this is the main fork-routine.
1332  *
1333  * It copies the process, and if successful kick-starts
1334  * it and waits for it to finish using the VM if required.
1335  */
1336 long do_fork(unsigned long clone_flags,
1337               unsigned long stack_start,
1338               struct pt_regs *regs,
1339               unsigned long stack_size,
1340               int __user *parent_tidptr,
1341               int __user *child_tidptr)
1342 {
1343         struct task_struct *p;
1344         int trace = 0;
1345         struct pid *pid = alloc_pid();
1346         long nr;
1347
1348         if (!pid)
1349                 return -EAGAIN;
1350         nr = pid->nr;
1351         if (unlikely(current->ptrace)) {
1352                 trace = fork_traceflag (clone_flags);
1353                 if (trace)
1354                         clone_flags |= CLONE_PTRACE;
1355         }
1356
1357         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, nr);
1358         /*
1359          * Do this prior waking up the new thread - the thread pointer
1360          * might get invalid after that point, if the thread exits quickly.
1361          */
1362         if (!IS_ERR(p)) {
1363                 struct completion vfork;
1364
1365                 if (clone_flags & CLONE_VFORK) {
1366                         p->vfork_done = &vfork;
1367                         init_completion(&vfork);
1368                 }
1369
1370                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1371                         /*
1372                          * We'll start up with an immediate SIGSTOP.
1373                          */
1374                         sigaddset(&p->pending.signal, SIGSTOP);
1375                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1376                 }
1377
1378                 if (!(clone_flags & CLONE_STOPPED))
1379                         wake_up_new_task(p, clone_flags);
1380                 else
1381                         p->state = TASK_STOPPED;
1382
1383                 if (unlikely (trace)) {
1384                         current->ptrace_message = nr;
1385                         ptrace_notify ((trace << 8) | SIGTRAP);
1386                 }
1387
1388                 if (clone_flags & CLONE_VFORK) {
1389                         wait_for_completion(&vfork);
1390                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1391                                 current->ptrace_message = nr;
1392                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1393                         }
1394                 }
1395         } else {
1396                 free_pid(pid);
1397                 nr = PTR_ERR(p);
1398         }
1399         return nr;
1400 }
1401
1402 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1403 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1404 #endif
1405
1406 static void sighand_ctor(void *data, kmem_cache_t *cachep, unsigned long flags)
1407 {
1408         struct sighand_struct *sighand = data;
1409
1410         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
1411                                         SLAB_CTOR_CONSTRUCTOR)
1412                 spin_lock_init(&sighand->siglock);
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, NULL);
1421         signal_cachep = kmem_cache_create("signal_cache",
1422                         sizeof(struct signal_struct), 0,
1423                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1424         files_cachep = kmem_cache_create("files_cache",
1425                         sizeof(struct files_struct), 0,
1426                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1427         fs_cachep = kmem_cache_create("fs_cache",
1428                         sizeof(struct fs_struct), 0,
1429                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1430         vm_area_cachep = kmem_cache_create("vm_area_struct",
1431                         sizeof(struct vm_area_struct), 0,
1432                         SLAB_PANIC, NULL, 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, NULL);
1436 }
1437
1438
1439 /*
1440  * Check constraints on flags passed to the unshare system call and
1441  * force unsharing of additional process context as appropriate.
1442  */
1443 static inline void check_unshare_flags(unsigned long *flags_ptr)
1444 {
1445         /*
1446          * If unsharing a thread from a thread group, must also
1447          * unshare vm.
1448          */
1449         if (*flags_ptr & CLONE_THREAD)
1450                 *flags_ptr |= CLONE_VM;
1451
1452         /*
1453          * If unsharing vm, must also unshare signal handlers.
1454          */
1455         if (*flags_ptr & CLONE_VM)
1456                 *flags_ptr |= CLONE_SIGHAND;
1457
1458         /*
1459          * If unsharing signal handlers and the task was created
1460          * using CLONE_THREAD, then must unshare the thread
1461          */
1462         if ((*flags_ptr & CLONE_SIGHAND) &&
1463             (atomic_read(&current->signal->count) > 1))
1464                 *flags_ptr |= CLONE_THREAD;
1465
1466         /*
1467          * If unsharing namespace, must also unshare filesystem information.
1468          */
1469         if (*flags_ptr & CLONE_NEWNS)
1470                 *flags_ptr |= CLONE_FS;
1471 }
1472
1473 /*
1474  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1475  */
1476 static int unshare_thread(unsigned long unshare_flags)
1477 {
1478         if (unshare_flags & CLONE_THREAD)
1479                 return -EINVAL;
1480
1481         return 0;
1482 }
1483
1484 /*
1485  * Unshare the filesystem structure if it is being shared
1486  */
1487 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1488 {
1489         struct fs_struct *fs = current->fs;
1490
1491         if ((unshare_flags & CLONE_FS) &&
1492             (fs && atomic_read(&fs->count) > 1)) {
1493                 *new_fsp = __copy_fs_struct(current->fs);
1494                 if (!*new_fsp)
1495                         return -ENOMEM;
1496         }
1497
1498         return 0;
1499 }
1500
1501 /*
1502  * Unshare the namespace structure if it is being shared
1503  */
1504 static int unshare_namespace(unsigned long unshare_flags, struct namespace **new_nsp, struct fs_struct *new_fs)
1505 {
1506         struct namespace *ns = current->namespace;
1507
1508         if ((unshare_flags & CLONE_NEWNS) &&
1509             (ns && atomic_read(&ns->count) > 1)) {
1510                 if (!capable(CAP_SYS_ADMIN))
1511                         return -EPERM;
1512
1513                 *new_nsp = dup_namespace(current, new_fs ? new_fs : current->fs);
1514                 if (!*new_nsp)
1515                         return -ENOMEM;
1516         }
1517
1518         return 0;
1519 }
1520
1521 /*
1522  * Unsharing of sighand for tasks created with CLONE_SIGHAND is not
1523  * supported yet
1524  */
1525 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1526 {
1527         struct sighand_struct *sigh = current->sighand;
1528
1529         if ((unshare_flags & CLONE_SIGHAND) &&
1530             (sigh && atomic_read(&sigh->count) > 1))
1531                 return -EINVAL;
1532         else
1533                 return 0;
1534 }
1535
1536 /*
1537  * Unshare vm if it is being shared
1538  */
1539 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1540 {
1541         struct mm_struct *mm = current->mm;
1542
1543         if ((unshare_flags & CLONE_VM) &&
1544             (mm && atomic_read(&mm->mm_users) > 1)) {
1545                 return -EINVAL;
1546         }
1547
1548         return 0;
1549 }
1550
1551 /*
1552  * Unshare file descriptor table if it is being shared
1553  */
1554 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1555 {
1556         struct files_struct *fd = current->files;
1557         int error = 0;
1558
1559         if ((unshare_flags & CLONE_FILES) &&
1560             (fd && atomic_read(&fd->count) > 1)) {
1561                 *new_fdp = dup_fd(fd, &error);
1562                 if (!*new_fdp)
1563                         return error;
1564         }
1565
1566         return 0;
1567 }
1568
1569 /*
1570  * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1571  * supported yet
1572  */
1573 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1574 {
1575         if (unshare_flags & CLONE_SYSVSEM)
1576                 return -EINVAL;
1577
1578         return 0;
1579 }
1580
1581 /*
1582  * unshare allows a process to 'unshare' part of the process
1583  * context which was originally shared using clone.  copy_*
1584  * functions used by do_fork() cannot be used here directly
1585  * because they modify an inactive task_struct that is being
1586  * constructed. Here we are modifying the current, active,
1587  * task_struct.
1588  */
1589 asmlinkage long sys_unshare(unsigned long unshare_flags)
1590 {
1591         int err = 0;
1592         struct fs_struct *fs, *new_fs = NULL;
1593         struct namespace *ns, *new_ns = NULL;
1594         struct sighand_struct *sigh, *new_sigh = NULL;
1595         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1596         struct files_struct *fd, *new_fd = NULL;
1597         struct sem_undo_list *new_ulist = NULL;
1598
1599         check_unshare_flags(&unshare_flags);
1600
1601         /* Return -EINVAL for all unsupported flags */
1602         err = -EINVAL;
1603         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1604                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM))
1605                 goto bad_unshare_out;
1606
1607         if ((err = unshare_thread(unshare_flags)))
1608                 goto bad_unshare_out;
1609         if ((err = unshare_fs(unshare_flags, &new_fs)))
1610                 goto bad_unshare_cleanup_thread;
1611         if ((err = unshare_namespace(unshare_flags, &new_ns, new_fs)))
1612                 goto bad_unshare_cleanup_fs;
1613         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1614                 goto bad_unshare_cleanup_ns;
1615         if ((err = unshare_vm(unshare_flags, &new_mm)))
1616                 goto bad_unshare_cleanup_sigh;
1617         if ((err = unshare_fd(unshare_flags, &new_fd)))
1618                 goto bad_unshare_cleanup_vm;
1619         if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1620                 goto bad_unshare_cleanup_fd;
1621
1622         if (new_fs || new_ns || new_sigh || new_mm || new_fd || new_ulist) {
1623
1624                 task_lock(current);
1625
1626                 if (new_fs) {
1627                         fs = current->fs;
1628                         current->fs = new_fs;
1629                         new_fs = fs;
1630                 }
1631
1632                 if (new_ns) {
1633                         ns = current->namespace;
1634                         current->namespace = new_ns;
1635                         new_ns = ns;
1636                 }
1637
1638                 if (new_sigh) {
1639                         sigh = current->sighand;
1640                         rcu_assign_pointer(current->sighand, new_sigh);
1641                         new_sigh = sigh;
1642                 }
1643
1644                 if (new_mm) {
1645                         mm = current->mm;
1646                         active_mm = current->active_mm;
1647                         current->mm = new_mm;
1648                         current->active_mm = new_mm;
1649                         activate_mm(active_mm, new_mm);
1650                         new_mm = mm;
1651                 }
1652
1653                 if (new_fd) {
1654                         fd = current->files;
1655                         current->files = new_fd;
1656                         new_fd = fd;
1657                 }
1658
1659                 task_unlock(current);
1660         }
1661
1662 bad_unshare_cleanup_fd:
1663         if (new_fd)
1664                 put_files_struct(new_fd);
1665
1666 bad_unshare_cleanup_vm:
1667         if (new_mm)
1668                 mmput(new_mm);
1669
1670 bad_unshare_cleanup_sigh:
1671         if (new_sigh)
1672                 if (atomic_dec_and_test(&new_sigh->count))
1673                         kmem_cache_free(sighand_cachep, new_sigh);
1674
1675 bad_unshare_cleanup_ns:
1676         if (new_ns)
1677                 put_namespace(new_ns);
1678
1679 bad_unshare_cleanup_fs:
1680         if (new_fs)
1681                 put_fs_struct(new_fs);
1682
1683 bad_unshare_cleanup_thread:
1684 bad_unshare_out:
1685         return err;
1686 }