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 #include <linux/vs_base.h>
  49 #include <linux/vs_context.h>
  50 #include <linux/vs_network.h>
  51 #include <linux/vs_limit.h>
  52 #include <linux/vs_memory.h>
  53
  54 #include <asm/pgtable.h>
  55 #include <asm/pgalloc.h>
  56 #include <asm/uaccess.h>
  57 #include <asm/mmu_context.h>
  58 #include <asm/cacheflush.h>
  59 #include <asm/tlbflush.h>
  60
  61 /*
  62  * Protected counters by write_lock_irq(&tasklist_lock)
  63  */
  64 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  65 int nr_threads;                 /* The idle threads do not count.. */
  66
  67 int max_threads;                /* tunable limit on nr_threads */
  68
  69 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  70
  71 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  72
  73 int nr_processes(void)
  74 {
  75         int cpu;
  76         int total = 0;
  77
  78         for_each_online_cpu(cpu)
  79                 total += per_cpu(process_counts, cpu);
  80
  81         return total;
  82 }
  83
  84 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  85 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  86 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  87 static kmem_cache_t *task_struct_cachep;
  88 #endif
  89
  90 /* SLAB cache for signal_struct structures (tsk->signal) */
  91 static kmem_cache_t *signal_cachep;
  92
  93 /* SLAB cache for sighand_struct structures (tsk->sighand) */
  94 kmem_cache_t *sighand_cachep;
  95
  96 /* SLAB cache for files_struct structures (tsk->files) */
  97 kmem_cache_t *files_cachep;
  98
  99 /* SLAB cache for fs_struct structures (tsk->fs) */
 100 kmem_cache_t *fs_cachep;
 101
 102 /* SLAB cache for vm_area_struct structures */
 103 kmem_cache_t *vm_area_cachep;
 104
 105 /* SLAB cache for mm_struct structures (tsk->mm) */
 106 static kmem_cache_t *mm_cachep;
 107
 108 void free_task(struct task_struct *tsk)
 109 {
 110         free_thread_info(tsk->thread_info);
 111         rt_mutex_debug_task_free(tsk);
 112         clr_vx_info(&tsk->vx_info);
 113         clr_nx_info(&tsk->nx_info);
 114         free_task_struct(tsk);
 115 }
 116 EXPORT_SYMBOL(free_task);
 117
 118 void __put_task_struct(struct task_struct *tsk)
 119 {
 120         WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
 121         WARN_ON(atomic_read(&tsk->usage));
 122         WARN_ON(tsk == current);
 123
 124         security_task_free(tsk);
 125         free_uid(tsk->user);
 126         put_group_info(tsk->group_info);
 127         delayacct_tsk_free(tsk);
 128
 129         if (!profile_handoff_task(tsk))
 130                 free_task(tsk);
 131 }
 132
 133 void __init fork_init(unsigned long mempages)
 134 {
 135 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 136 #ifndef ARCH_MIN_TASKALIGN
 137 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 138 #endif
 139         /* create a slab on which task_structs can be allocated */
 140         task_struct_cachep =
 141                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 142                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
 143 #endif
 144
 145         /*
 146          * The default maximum number of threads is set to a safe
 147          * value: the thread structures can take up at most half
 148          * of memory.
 149          */
 150         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 151
 152         /*
 153          * we need to allow at least 20 threads to boot a system
 154          */
 155         if(max_threads < 20)
 156                 max_threads = 20;
 157
 158         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 159         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 160         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 161                 init_task.signal->rlim[RLIMIT_NPROC];
 162 }
 163
 164 static struct task_struct *dup_task_struct(struct task_struct *orig)
 165 {
 166         struct task_struct *tsk;
 167         struct thread_info *ti;
 168
 169         prepare_to_copy(orig);
 170
 171         tsk = alloc_task_struct();
 172         if (!tsk)
 173                 return NULL;
 174
 175         ti = alloc_thread_info(tsk);
 176         if (!ti) {
 177                 free_task_struct(tsk);
 178                 return NULL;
 179         }
 180
 181         *tsk = *orig;
 182         tsk->thread_info = ti;
 183         setup_thread_stack(tsk, orig);
 184
 185         /* One for us, one for whoever does the "release_task()" (usually parent) */
 186         atomic_set(&tsk->usage,2);
 187         atomic_set(&tsk->fs_excl, 0);
 188         tsk->btrace_seq = 0;
 189         tsk->splice_pipe = NULL;
 190         return tsk;
 191 }
 192
 193 #ifdef CONFIG_MMU
 194 static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 195 {
 196         struct vm_area_struct *mpnt, *tmp, **pprev;
 197         struct rb_node **rb_link, *rb_parent;
 198         int retval;
 199         unsigned long charge;
 200         struct mempolicy *pol;
 201
 202         down_write(&oldmm->mmap_sem);
 203         flush_cache_mm(oldmm);
 204         /*
 205          * Not linked in yet - no deadlock potential:
 206          */
 207         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 208
 209         mm->locked_vm = 0;
 210         mm->mmap = NULL;
 211         mm->mmap_cache = NULL;
 212         mm->free_area_cache = oldmm->mmap_base;
 213         mm->cached_hole_size = ~0UL;
 214         mm->map_count = 0;
 215         __set_mm_counter(mm, file_rss, 0);
 216         __set_mm_counter(mm, anon_rss, 0);
 217         cpus_clear(mm->cpu_vm_mask);
 218         mm->mm_rb = RB_ROOT;
 219         rb_link = &mm->mm_rb.rb_node;
 220         rb_parent = NULL;
 221         pprev = &mm->mmap;
 222
 223         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 224                 struct file *file;
 225
 226                 if (mpnt->vm_flags & VM_DONTCOPY) {
 227                         long pages = vma_pages(mpnt);
 228                         vx_vmpages_sub(mm, pages);
 229                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 230                                                                 -pages);
 231                         continue;
 232                 }
 233                 charge = 0;
 234                 if (mpnt->vm_flags & VM_ACCOUNT) {
 235                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 236                         if (security_vm_enough_memory(len))
 237                                 goto fail_nomem;
 238                         charge = len;
 239                 }
 240                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 241                 if (!tmp)
 242                         goto fail_nomem;
 243                 *tmp = *mpnt;
 244                 pol = mpol_copy(vma_policy(mpnt));
 245                 retval = PTR_ERR(pol);
 246                 if (IS_ERR(pol))
 247                         goto fail_nomem_policy;
 248                 vma_set_policy(tmp, pol);
 249                 tmp->vm_flags &= ~VM_LOCKED;
 250                 tmp->vm_mm = mm;
 251                 tmp->vm_next = NULL;
 252                 anon_vma_link(tmp);
 253                 file = tmp->vm_file;
 254                 if (file) {
 255                         struct inode *inode = file->f_dentry->d_inode;
 256                         get_file(file);
 257                         if (tmp->vm_flags & VM_DENYWRITE)
 258                                 atomic_dec(&inode->i_writecount);
 259
 260                         /* insert tmp into the share list, just after mpnt */
 261                         spin_lock(&file->f_mapping->i_mmap_lock);
 262                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 263                         flush_dcache_mmap_lock(file->f_mapping);
 264                         vma_prio_tree_add(tmp, mpnt);
 265                         flush_dcache_mmap_unlock(file->f_mapping);
 266                         spin_unlock(&file->f_mapping->i_mmap_lock);
 267                 }
 268
 269                 /*
 270                  * Link in the new vma and copy the page table entries.
 271                  */
 272                 *pprev = tmp;
 273                 pprev = &tmp->vm_next;
 274
 275                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 276                 rb_link = &tmp->vm_rb.rb_right;
 277                 rb_parent = &tmp->vm_rb;
 278
 279                 mm->map_count++;
 280                 retval = copy_page_range(mm, oldmm, mpnt);
 281
 282                 if (tmp->vm_ops && tmp->vm_ops->open)
 283                         tmp->vm_ops->open(tmp);
 284
 285                 if (retval)
 286                         goto out;
 287         }
 288 #ifdef arch_dup_mmap
 289         arch_dup_mmap(mm, oldmm);
 290 #endif
 291         retval = 0;
 292 out:
 293         up_write(&mm->mmap_sem);
 294         flush_tlb_mm(oldmm);
 295         up_write(&oldmm->mmap_sem);
 296         return retval;
 297 fail_nomem_policy:
 298         kmem_cache_free(vm_area_cachep, tmp);
 299 fail_nomem:
 300         retval = -ENOMEM;
 301         vm_unacct_memory(charge);
 302         goto out;
 303 }
 304
 305 static inline int mm_alloc_pgd(struct mm_struct * mm)
 306 {
 307         mm->pgd = pgd_alloc(mm);
 308         if (unlikely(!mm->pgd))
 309                 return -ENOMEM;
 310         return 0;
 311 }
 312
 313 static inline void mm_free_pgd(struct mm_struct * mm)
 314 {
 315         pgd_free(mm->pgd);
 316 }
 317 #else
 318 #define dup_mmap(mm, oldmm)     (0)
 319 #define mm_alloc_pgd(mm)        (0)
 320 #define mm_free_pgd(mm)
 321 #endif /* CONFIG_MMU */
 322
 323  __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 324
 325 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
 326 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 327
 328 #include <linux/init_task.h>
 329
 330 static struct mm_struct * mm_init(struct mm_struct * mm)
 331 {
 332         atomic_set(&mm->mm_users, 1);
 333         atomic_set(&mm->mm_count, 1);
 334         init_rwsem(&mm->mmap_sem);
 335         INIT_LIST_HEAD(&mm->mmlist);
 336         mm->core_waiters = 0;
 337         mm->nr_ptes = 0;
 338         spin_lock_init(&mm->page_table_lock);
 339         rwlock_init(&mm->ioctx_list_lock);
 340         mm->ioctx_list = NULL;
 341         mm->free_area_cache = TASK_UNMAPPED_BASE;
 342         mm->cached_hole_size = ~0UL;
 343
 344         if (likely(!mm_alloc_pgd(mm))) {
 345                 mm->def_flags = 0;
 346                 set_vx_info(&mm->mm_vx_info, current->vx_info);
 347                 return mm;
 348         }
 349         free_mm(mm);
 350         return NULL;
 351 }
 352
 353 /*
 354  * Allocate and initialize an mm_struct.
 355  */
 356 struct mm_struct * mm_alloc(void)
 357 {
 358         struct mm_struct * mm;
 359
 360         mm = allocate_mm();
 361         if (mm) {
 362                 memset(mm, 0, sizeof(*mm));
 363                 mm = mm_init(mm);
 364         }
 365         return mm;
 366 }
 367
 368 /*
 369  * Called when the last reference to the mm
 370  * is dropped: either by a lazy thread or by
 371  * mmput. Free the page directory and the mm.
 372  */
 373 void fastcall __mmdrop(struct mm_struct *mm)
 374 {
 375         BUG_ON(mm == &init_mm);
 376         mm_free_pgd(mm);
 377         destroy_context(mm);
 378         clr_vx_info(&mm->mm_vx_info);
 379         free_mm(mm);
 380 }
 381
 382 /*
 383  * Decrement the use count and release all resources for an mm.
 384  */
 385 void mmput(struct mm_struct *mm)
 386 {
 387         might_sleep();
 388
 389         if (atomic_dec_and_test(&mm->mm_users)) {
 390                 exit_aio(mm);
 391                 exit_mmap(mm);
 392                 if (!list_empty(&mm->mmlist)) {
 393                         spin_lock(&mmlist_lock);
 394                         list_del(&mm->mmlist);
 395                         spin_unlock(&mmlist_lock);
 396                 }
 397                 put_swap_token(mm);
 398                 mmdrop(mm);
 399         }
 400 }
 401 EXPORT_SYMBOL_GPL(mmput);
 402
 403 /**
 404  * get_task_mm - acquire a reference to the task's mm
 405  *
 406  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
 407  * this kernel workthread has transiently adopted a user mm with use_mm,
 408  * to do its AIO) is not set and if so returns a reference to it, after
 409  * bumping up the use count.  User must release the mm via mmput()
 410  * after use.  Typically used by /proc and ptrace.
 411  */
 412 struct mm_struct *get_task_mm(struct task_struct *task)
 413 {
 414         struct mm_struct *mm;
 415
 416         task_lock(task);
 417         mm = task->mm;
 418         if (mm) {
 419                 if (task->flags & PF_BORROWED_MM)
 420                         mm = NULL;
 421                 else
 422                         atomic_inc(&mm->mm_users);
 423         }
 424         task_unlock(task);
 425         return mm;
 426 }
 427 EXPORT_SYMBOL_GPL(get_task_mm);
 428
 429 /* Please note the differences between mmput and mm_release.
 430  * mmput is called whenever we stop holding onto a mm_struct,
 431  * error success whatever.
 432  *
 433  * mm_release is called after a mm_struct has been removed
 434  * from the current process.
 435  *
 436  * This difference is important for error handling, when we
 437  * only half set up a mm_struct for a new process and need to restore
 438  * the old one.  Because we mmput the new mm_struct before
 439  * restoring the old one. . .
 440  * Eric Biederman 10 January 1998
 441  */
 442 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 443 {
 444         struct completion *vfork_done = tsk->vfork_done;
 445
 446         /* Get rid of any cached register state */
 447         deactivate_mm(tsk, mm);
 448
 449         /* notify parent sleeping on vfork() */
 450         if (vfork_done) {
 451                 tsk->vfork_done = NULL;
 452                 complete(vfork_done);
 453         }
 454         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
 455                 u32 __user * tidptr = tsk->clear_child_tid;
 456                 tsk->clear_child_tid = NULL;
 457
 458                 /*
 459                  * We don't check the error code - if userspace has
 460                  * not set up a proper pointer then tough luck.
 461                  */
 462                 put_user(0, tidptr);
 463                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 464         }
 465 }
 466
 467 /*
 468  * Allocate a new mm structure and copy contents from the
 469  * mm structure of the passed in task structure.
 470  */
 471 static struct mm_struct *dup_mm(struct task_struct *tsk)
 472 {
 473         struct mm_struct *mm, *oldmm = current->mm;
 474         int err;
 475
 476         if (!oldmm)
 477                 return NULL;
 478
 479         mm = allocate_mm();
 480         if (!mm)
 481                 goto fail_nomem;
 482
 483         memcpy(mm, oldmm, sizeof(*mm));
 484         mm->mm_vx_info = NULL;
 485
 486         if (!mm_init(mm))
 487                 goto fail_nomem;
 488
 489         if (init_new_context(tsk, mm))
 490                 goto fail_nocontext;
 491
 492         err = dup_mmap(mm, oldmm);
 493         if (err)
 494                 goto free_pt;
 495
 496         mm->hiwater_rss = get_mm_rss(mm);
 497         mm->hiwater_vm = mm->total_vm;
 498
 499         return mm;
 500
 501 free_pt:
 502         mmput(mm);
 503
 504 fail_nomem:
 505         return NULL;
 506
 507 fail_nocontext:
 508         /*
 509          * If init_new_context() failed, we cannot use mmput() to free the mm
 510          * because it calls destroy_context()
 511          */
 512         clr_vx_info(&mm->mm_vx_info);
 513         mm_free_pgd(mm);
 514         free_mm(mm);
 515         return NULL;
 516 }
 517
 518 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 519 {
 520         struct mm_struct * mm, *oldmm;
 521         int retval;
 522
 523         tsk->min_flt = tsk->maj_flt = 0;
 524         tsk->nvcsw = tsk->nivcsw = 0;
 525
 526         tsk->mm = NULL;
 527         tsk->active_mm = NULL;
 528
 529         /*
 530          * Are we cloning a kernel thread?
 531          *
 532          * We need to steal a active VM for that..
 533          */
 534         oldmm = current->mm;
 535         if (!oldmm)
 536                 return 0;
 537
 538         if (clone_flags & CLONE_VM) {
 539                 atomic_inc(&oldmm->mm_users);
 540                 mm = oldmm;
 541                 goto good_mm;
 542         }
 543
 544         retval = -ENOMEM;
 545         mm = dup_mm(tsk);
 546         if (!mm)
 547                 goto fail_nomem;
 548
 549 good_mm:
 550         tsk->mm = mm;
 551         tsk->active_mm = mm;
 552         return 0;
 553
 554 fail_nomem:
 555         return retval;
 556 }
 557
 558 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 559 {
 560         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 561         /* We don't need to lock fs - think why ;-) */
 562         if (fs) {
 563                 atomic_set(&fs->count, 1);
 564                 rwlock_init(&fs->lock);
 565                 fs->umask = old->umask;
 566                 read_lock(&old->lock);
 567                 fs->rootmnt = mntget(old->rootmnt);
 568                 fs->root = dget(old->root);
 569                 fs->pwdmnt = mntget(old->pwdmnt);
 570                 fs->pwd = dget(old->pwd);
 571                 if (old->altroot) {
 572                         fs->altrootmnt = mntget(old->altrootmnt);
 573                         fs->altroot = dget(old->altroot);
 574                 } else {
 575                         fs->altrootmnt = NULL;
 576                         fs->altroot = NULL;
 577                 }
 578                 read_unlock(&old->lock);
 579         }
 580         return fs;
 581 }
 582
 583 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 584 {
 585         return __copy_fs_struct(old);
 586 }
 587
 588 EXPORT_SYMBOL_GPL(copy_fs_struct);
 589
 590 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 591 {
 592         if (clone_flags & CLONE_FS) {
 593                 atomic_inc(&current->fs->count);
 594                 return 0;
 595         }
 596         tsk->fs = __copy_fs_struct(current->fs);
 597         if (!tsk->fs)
 598                 return -ENOMEM;
 599         return 0;
 600 }
 601
 602 static int count_open_files(struct fdtable *fdt)
 603 {
 604         int size = fdt->max_fdset;
 605         int i;
 606
 607         /* Find the last open fd */
 608         for (i = size/(8*sizeof(long)); i > 0; ) {
 609                 if (fdt->open_fds->fds_bits[--i])
 610                         break;
 611         }
 612         i = (i+1) * 8 * sizeof(long);
 613         return i;
 614 }
 615
 616 static struct files_struct *alloc_files(void)
 617 {
 618         struct files_struct *newf;
 619         struct fdtable *fdt;
 620
 621         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
 622         if (!newf)
 623                 goto out;
 624
 625         atomic_set(&newf->count, 1);
 626
 627         spin_lock_init(&newf->file_lock);
 628         newf->next_fd = 0;
 629         fdt = &newf->fdtab;
 630         fdt->max_fds = NR_OPEN_DEFAULT;
 631         fdt->max_fdset = EMBEDDED_FD_SET_SIZE;
 632         fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
 633         fdt->open_fds = (fd_set *)&newf->open_fds_init;
 634         fdt->fd = &newf->fd_array[0];
 635         INIT_RCU_HEAD(&fdt->rcu);
 636         fdt->free_files = NULL;
 637         fdt->next = NULL;
 638         rcu_assign_pointer(newf->fdt, fdt);
 639 out:
 640         return newf;
 641 }
 642
 643 /*
 644  * Allocate a new files structure and copy contents from the
 645  * passed in files structure.
 646  * errorp will be valid only when the returned files_struct is NULL.
 647  */
 648 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
 649 {
 650         struct files_struct *newf;
 651         struct file **old_fds, **new_fds;
 652         int open_files, size, i, expand;
 653         struct fdtable *old_fdt, *new_fdt;
 654
 655         *errorp = -ENOMEM;
 656         newf = alloc_files();
 657         if (!newf)
 658                 goto out;
 659
 660         spin_lock(&oldf->file_lock);
 661         old_fdt = files_fdtable(oldf);
 662         new_fdt = files_fdtable(newf);
 663         size = old_fdt->max_fdset;
 664         open_files = count_open_files(old_fdt);
 665         expand = 0;
 666
 667         /*
 668          * Check whether we need to allocate a larger fd array or fd set.
 669          * Note: we're not a clone task, so the open count won't  change.
 670          */
 671         if (open_files > new_fdt->max_fdset) {
 672                 new_fdt->max_fdset = 0;
 673                 expand = 1;
 674         }
 675         if (open_files > new_fdt->max_fds) {
 676                 new_fdt->max_fds = 0;
 677                 expand = 1;
 678         }
 679
 680         /* if the old fdset gets grown now, we'll only copy up to "size" fds */
 681         if (expand) {
 682                 spin_unlock(&oldf->file_lock);
 683                 spin_lock(&newf->file_lock);
 684                 *errorp = expand_files(newf, open_files-1);
 685                 spin_unlock(&newf->file_lock);
 686                 if (*errorp < 0)
 687                         goto out_release;
 688                 new_fdt = files_fdtable(newf);
 689                 /*
 690                  * Reacquire the oldf lock and a pointer to its fd table
 691                  * who knows it may have a new bigger fd table. We need
 692                  * the latest pointer.
 693                  */
 694                 spin_lock(&oldf->file_lock);
 695                 old_fdt = files_fdtable(oldf);
 696         }
 697
 698         old_fds = old_fdt->fd;
 699         new_fds = new_fdt->fd;
 700
 701         memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
 702         memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
 703
 704         for (i = open_files; i != 0; i--) {
 705                 struct file *f = *old_fds++;
 706                 if (f) {
 707                         get_file(f);
 708                         /* FIXME: sum it first for check and performance */
 709                         vx_openfd_inc(open_files - i);
 710                 } else {
 711                         /*
 712                          * The fd may be claimed in the fd bitmap but not yet
 713                          * instantiated in the files array if a sibling thread
 714                          * is partway through open().  So make sure that this
 715                          * fd is available to the new process.
 716                          */
 717                         FD_CLR(open_files - i, new_fdt->open_fds);
 718                 }
 719                 rcu_assign_pointer(*new_fds++, f);
 720         }
 721         spin_unlock(&oldf->file_lock);
 722
 723         /* compute the remainder to be cleared */
 724         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
 725
 726         /* This is long word aligned thus could use a optimized version */
 727         memset(new_fds, 0, size);
 728
 729         if (new_fdt->max_fdset > open_files) {
 730                 int left = (new_fdt->max_fdset-open_files)/8;
 731                 int start = open_files / (8 * sizeof(unsigned long));
 732
 733                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
 734                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
 735         }
 736
 737 out:
 738         return newf;
 739
 740 out_release:
 741         free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
 742         free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
 743         free_fd_array(new_fdt->fd, new_fdt->max_fds);
 744         kmem_cache_free(files_cachep, newf);
 745         return NULL;
 746 }
 747
 748 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 749 {
 750         struct files_struct *oldf, *newf;
 751         int error = 0;
 752
 753         /*
 754          * A background process may not have any files ...
 755          */
 756         oldf = current->files;
 757         if (!oldf)
 758                 goto out;
 759
 760         if (clone_flags & CLONE_FILES) {
 761                 atomic_inc(&oldf->count);
 762                 goto out;
 763         }
 764
 765         /*
 766          * Note: we may be using current for both targets (See exec.c)
 767          * This works because we cache current->files (old) as oldf. Don't
 768          * break this.
 769          */
 770         tsk->files = NULL;
 771         newf = dup_fd(oldf, &error);
 772         if (!newf)
 773                 goto out;
 774
 775         tsk->files = newf;
 776         error = 0;
 777 out:
 778         return error;
 779 }
 780
 781 /*
 782  *      Helper to unshare the files of the current task.
 783  *      We don't want to expose copy_files internals to
 784  *      the exec layer of the kernel.
 785  */
 786
 787 int unshare_files(void)
 788 {
 789         struct files_struct *files  = current->files;
 790         int rc;
 791
 792         BUG_ON(!files);
 793
 794         /* This can race but the race causes us to copy when we don't
 795            need to and drop the copy */
 796         if(atomic_read(&files->count) == 1)
 797         {
 798                 atomic_inc(&files->count);
 799                 return 0;
 800         }
 801         rc = copy_files(0, current);
 802         if(rc)
 803                 current->files = files;
 804         return rc;
 805 }
 806
 807 EXPORT_SYMBOL(unshare_files);
 808
 809 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 810 {
 811         struct sighand_struct *sig;
 812
 813         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 814                 atomic_inc(&current->sighand->count);
 815                 return 0;
 816         }
 817         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 818         rcu_assign_pointer(tsk->sighand, sig);
 819         if (!sig)
 820                 return -ENOMEM;
 821         atomic_set(&sig->count, 1);
 822         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 823         return 0;
 824 }
 825
 826 void __cleanup_sighand(struct sighand_struct *sighand)
 827 {
 828         if (atomic_dec_and_test(&sighand->count))
 829                 kmem_cache_free(sighand_cachep, sighand);
 830 }
 831
 832 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 833 {
 834         struct signal_struct *sig;
 835         int ret;
 836
 837         if (clone_flags & CLONE_THREAD) {
 838                 atomic_inc(&current->signal->count);
 839                 atomic_inc(&current->signal->live);
 840                 taskstats_tgid_alloc(current->signal);
 841                 return 0;
 842         }
 843         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 844         tsk->signal = sig;
 845         if (!sig)
 846                 return -ENOMEM;
 847
 848         ret = copy_thread_group_keys(tsk);
 849         if (ret < 0) {
 850                 kmem_cache_free(signal_cachep, sig);
 851                 return ret;
 852         }
 853
 854         atomic_set(&sig->count, 1);
 855         atomic_set(&sig->live, 1);
 856         init_waitqueue_head(&sig->wait_chldexit);
 857         sig->flags = 0;
 858         sig->group_exit_code = 0;
 859         sig->group_exit_task = NULL;
 860         sig->group_stop_count = 0;
 861         sig->curr_target = NULL;
 862         init_sigpending(&sig->shared_pending);
 863         INIT_LIST_HEAD(&sig->posix_timers);
 864
 865         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_REL);
 866         sig->it_real_incr.tv64 = 0;
 867         sig->real_timer.function = it_real_fn;
 868         sig->tsk = tsk;
 869
 870         sig->it_virt_expires = cputime_zero;
 871         sig->it_virt_incr = cputime_zero;
 872         sig->it_prof_expires = cputime_zero;
 873         sig->it_prof_incr = cputime_zero;
 874
 875         sig->leader = 0;        /* session leadership doesn't inherit */
 876         sig->tty_old_pgrp = 0;
 877
 878         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 879         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 880         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 881         sig->sched_time = 0;
 882         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 883         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 884         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 885         taskstats_tgid_init(sig);
 886
 887         task_lock(current->group_leader);
 888         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 889         task_unlock(current->group_leader);
 890
 891         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 892                 /*
 893                  * New sole thread in the process gets an expiry time
 894                  * of the whole CPU time limit.
 895                  */
 896                 tsk->it_prof_expires =
 897                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 898         }
 899         acct_init_pacct(&sig->pacct);
 900
 901         return 0;
 902 }
 903
 904 void __cleanup_signal(struct signal_struct *sig)
 905 {
 906         exit_thread_group_keys(sig);
 907         taskstats_tgid_free(sig);
 908         kmem_cache_free(signal_cachep, sig);
 909 }
 910
 911 static inline void cleanup_signal(struct task_struct *tsk)
 912 {
 913         struct signal_struct *sig = tsk->signal;
 914
 915         atomic_dec(&sig->live);
 916
 917         if (atomic_dec_and_test(&sig->count))
 918                 __cleanup_signal(sig);
 919 }
 920
 921 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 922 {
 923         unsigned long new_flags = p->flags;
 924
 925         new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
 926         new_flags |= PF_FORKNOEXEC;
 927         if (!(clone_flags & CLONE_PTRACE))
 928                 p->ptrace = 0;
 929         p->flags = new_flags;
 930 }
 931
 932 asmlinkage long sys_set_tid_address(int __user *tidptr)
 933 {
 934         current->clear_child_tid = tidptr;
 935
 936         return current->pid;
 937 }
 938
 939 static inline void rt_mutex_init_task(struct task_struct *p)
 940 {
 941 #ifdef CONFIG_RT_MUTEXES
 942         spin_lock_init(&p->pi_lock);
 943         plist_head_init(&p->pi_waiters, &p->pi_lock);
 944         p->pi_blocked_on = NULL;
 945 #endif
 946 }
 947
 948 /*
 949  * This creates a new process as a copy of the old one,
 950  * but does not actually start it yet.
 951  *
 952  * It copies the registers, and all the appropriate
 953  * parts of the process environment (as per the clone
 954  * flags). The actual kick-off is left to the caller.
 955  */
 956 static struct task_struct *copy_process(unsigned long clone_flags,
 957                                         unsigned long stack_start,
 958                                         struct pt_regs *regs,
 959                                         unsigned long stack_size,
 960                                         int __user *parent_tidptr,
 961                                         int __user *child_tidptr,
 962                                         int pid)
 963 {
 964         int retval;
 965         struct task_struct *p = NULL;
 966         struct vx_info *vxi;
 967         struct nx_info *nxi;
 968
 969         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 970                 return ERR_PTR(-EINVAL);
 971
 972         /*
 973          * Thread groups must share signals as well, and detached threads
 974          * can only be started up within the thread group.
 975          */
 976         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 977                 return ERR_PTR(-EINVAL);
 978
 979         /*
 980          * Shared signal handlers imply shared VM. By way of the above,
 981          * thread groups also imply shared VM. Blocking this case allows
 982          * for various simplifications in other code.
 983          */
 984         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 985                 return ERR_PTR(-EINVAL);
 986
 987         retval = security_task_create(clone_flags);
 988         if (retval)
 989                 goto fork_out;
 990
 991         retval = -ENOMEM;
 992         p = dup_task_struct(current);
 993         if (!p)
 994                 goto fork_out;
 995
 996         rt_mutex_init_task(p);
 997
 998         p->tux_info = NULL;
 999
1000 #ifdef CONFIG_TRACE_IRQFLAGS
1001         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1002         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1003 #endif
1004         init_vx_info(&p->vx_info, current->vx_info);
1005         init_nx_info(&p->nx_info, current->nx_info);
1006
1007         /* check vserver memory */
1008         if (p->mm && !(clone_flags & CLONE_VM)) {
1009                 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
1010                         vx_pages_add(p->vx_info, RLIMIT_AS, p->mm->total_vm);
1011                 else
1012                         goto bad_fork_free;
1013         }
1014         if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
1015                 if (!vx_rsspages_avail(p->mm, get_mm_counter(p->mm, file_rss)))
1016                         goto bad_fork_cleanup_vm;
1017         }
1018
1019         retval = -EAGAIN;
1020         if (!vx_nproc_avail(1))
1021                 goto bad_fork_cleanup_vm;
1022
1023         if (atomic_read(&p->user->processes) >=
1024                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1025                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1026                                 p->user != &root_user)
1027                         goto bad_fork_cleanup_vm;
1028         }
1029
1030         atomic_inc(&p->user->__count);
1031         atomic_inc(&p->user->processes);
1032         get_group_info(p->group_info);
1033
1034         /*
1035          * If multiple threads are within copy_process(), then this check
1036          * triggers too late. This doesn't hurt, the check is only there
1037          * to stop root fork bombs.
1038          */
1039         if (nr_threads >= max_threads)
1040                 goto bad_fork_cleanup_count;
1041
1042         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1043                 goto bad_fork_cleanup_count;
1044
1045         if (p->binfmt && !try_module_get(p->binfmt->module))
1046                 goto bad_fork_cleanup_put_domain;
1047
1048         p->did_exec = 0;
1049         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1050         copy_flags(clone_flags, p);
1051         p->pid = pid;
1052         retval = -EFAULT;
1053         if (clone_flags & CLONE_PARENT_SETTID)
1054                 if (put_user(p->pid, parent_tidptr))
1055                         goto bad_fork_cleanup_delays_binfmt;
1056
1057         INIT_LIST_HEAD(&p->children);
1058         INIT_LIST_HEAD(&p->sibling);
1059         p->vfork_done = NULL;
1060         spin_lock_init(&p->alloc_lock);
1061
1062         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1063         init_sigpending(&p->pending);
1064
1065         p->utime = cputime_zero;
1066         p->stime = cputime_zero;
1067         p->sched_time = 0;
1068         p->rchar = 0;           /* I/O counter: bytes read */
1069         p->wchar = 0;           /* I/O counter: bytes written */
1070         p->syscr = 0;           /* I/O counter: read syscalls */
1071         p->syscw = 0;           /* I/O counter: write syscalls */
1072         acct_clear_integrals(p);
1073
1074         p->it_virt_expires = cputime_zero;
1075         p->it_prof_expires = cputime_zero;
1076         p->it_sched_expires = 0;
1077         INIT_LIST_HEAD(&p->cpu_timers[0]);
1078         INIT_LIST_HEAD(&p->cpu_timers[1]);
1079         INIT_LIST_HEAD(&p->cpu_timers[2]);
1080
1081         p->lock_depth = -1;             /* -1 = no lock */
1082         do_posix_clock_monotonic_gettime(&p->start_time);
1083         p->security = NULL;
1084         p->io_context = NULL;
1085         p->io_wait = NULL;
1086         p->audit_context = NULL;
1087         cpuset_fork(p);
1088 #ifdef CONFIG_NUMA
1089         p->mempolicy = mpol_copy(p->mempolicy);
1090         if (IS_ERR(p->mempolicy)) {
1091                 retval = PTR_ERR(p->mempolicy);
1092                 p->mempolicy = NULL;
1093                 goto bad_fork_cleanup_cpuset;
1094         }
1095         mpol_fix_fork_child_flag(p);
1096 #endif
1097 #ifdef CONFIG_TRACE_IRQFLAGS
1098         p->irq_events = 0;
1099         p->hardirqs_enabled = 0;
1100         p->hardirq_enable_ip = 0;
1101         p->hardirq_enable_event = 0;
1102         p->hardirq_disable_ip = _THIS_IP_;
1103         p->hardirq_disable_event = 0;
1104         p->softirqs_enabled = 1;
1105         p->softirq_enable_ip = _THIS_IP_;
1106         p->softirq_enable_event = 0;
1107         p->softirq_disable_ip = 0;
1108         p->softirq_disable_event = 0;
1109         p->hardirq_context = 0;
1110         p->softirq_context = 0;
1111 #endif
1112 #ifdef CONFIG_LOCKDEP
1113         p->lockdep_depth = 0; /* no locks held yet */
1114         p->curr_chain_key = 0;
1115         p->lockdep_recursion = 0;
1116 #endif
1117
1118 #ifdef CONFIG_DEBUG_MUTEXES
1119         p->blocked_on = NULL; /* not blocked yet */
1120 #endif
1121
1122         p->tgid = p->pid;
1123         if (clone_flags & CLONE_THREAD)
1124                 p->tgid = current->tgid;
1125
1126         if ((retval = security_task_alloc(p)))
1127                 goto bad_fork_cleanup_policy;
1128         if ((retval = audit_alloc(p)))
1129                 goto bad_fork_cleanup_security;
1130         /* copy all the process information */
1131         if ((retval = copy_semundo(clone_flags, p)))
1132                 goto bad_fork_cleanup_audit;
1133         if ((retval = copy_files(clone_flags, p)))
1134                 goto bad_fork_cleanup_semundo;
1135         if ((retval = copy_fs(clone_flags, p)))
1136                 goto bad_fork_cleanup_files;
1137         if ((retval = copy_sighand(clone_flags, p)))
1138                 goto bad_fork_cleanup_fs;
1139         if ((retval = copy_signal(clone_flags, p)))
1140                 goto bad_fork_cleanup_sighand;
1141         if ((retval = copy_mm(clone_flags, p)))
1142                 goto bad_fork_cleanup_signal;
1143         if ((retval = copy_keys(clone_flags, p)))
1144                 goto bad_fork_cleanup_mm;
1145         if ((retval = copy_namespace(clone_flags, p)))
1146                 goto bad_fork_cleanup_keys;
1147         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1148         if (retval)
1149                 goto bad_fork_cleanup_namespace;
1150
1151         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1152         /*
1153          * Clear TID on mm_release()?
1154          */
1155         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1156         p->robust_list = NULL;
1157 #ifdef CONFIG_COMPAT
1158         p->compat_robust_list = NULL;
1159 #endif
1160         INIT_LIST_HEAD(&p->pi_state_list);
1161         p->pi_state_cache = NULL;
1162
1163         /*
1164          * sigaltstack should be cleared when sharing the same VM
1165          */
1166         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1167                 p->sas_ss_sp = p->sas_ss_size = 0;
1168
1169         /*
1170          * Syscall tracing should be turned off in the child regardless
1171          * of CLONE_PTRACE.
1172          */
1173         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1174 #ifdef TIF_SYSCALL_EMU
1175         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1176 #endif
1177
1178         /* Our parent execution domain becomes current domain
1179            These must match for thread signalling to apply */
1180
1181         p->parent_exec_id = p->self_exec_id;
1182
1183         /* ok, now we should be set up.. */
1184         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1185         p->pdeath_signal = 0;
1186         p->exit_state = 0;
1187
1188         /*
1189          * Ok, make it visible to the rest of the system.
1190          * We dont wake it up yet.
1191          */
1192         p->group_leader = p;
1193         INIT_LIST_HEAD(&p->thread_group);
1194         INIT_LIST_HEAD(&p->ptrace_children);
1195         INIT_LIST_HEAD(&p->ptrace_list);
1196
1197         /* Perform scheduler related setup. Assign this task to a CPU. */
1198         sched_fork(p, clone_flags);
1199
1200         /* Need tasklist lock for parent etc handling! */
1201         write_lock_irq(&tasklist_lock);
1202
1203         /*
1204          * The task hasn't been attached yet, so its cpus_allowed mask will
1205          * not be changed, nor will its assigned CPU.
1206          *
1207          * The cpus_allowed mask of the parent may have changed after it was
1208          * copied first time - so re-copy it here, then check the child's CPU
1209          * to ensure it is on a valid CPU (and if not, just force it back to
1210          * parent's CPU). This avoids alot of nasty races.
1211          */
1212         p->cpus_allowed = current->cpus_allowed;
1213         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1214                         !cpu_online(task_cpu(p))))
1215                 set_task_cpu(p, smp_processor_id());
1216
1217         /* CLONE_PARENT re-uses the old parent */
1218         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1219                 p->real_parent = current->real_parent;
1220         else
1221                 p->real_parent = current;
1222         p->parent = p->real_parent;
1223
1224         spin_lock(&current->sighand->siglock);
1225
1226         /*
1227          * Process group and session signals need to be delivered to just the
1228          * parent before the fork or both the parent and the child after the
1229          * fork. Restart if a signal comes in before we add the new process to
1230          * it's process group.
1231          * A fatal signal pending means that current will exit, so the new
1232          * thread can't slip out of an OOM kill (or normal SIGKILL).
1233          */
1234         recalc_sigpending();
1235         if (signal_pending(current)) {
1236                 spin_unlock(&current->sighand->siglock);
1237                 write_unlock_irq(&tasklist_lock);
1238                 retval = -ERESTARTNOINTR;
1239                 goto bad_fork_cleanup_namespace;
1240         }
1241
1242         if (clone_flags & CLONE_THREAD) {
1243                 p->group_leader = current->group_leader;
1244                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1245
1246                 if (!cputime_eq(current->signal->it_virt_expires,
1247                                 cputime_zero) ||
1248                     !cputime_eq(current->signal->it_prof_expires,
1249                                 cputime_zero) ||
1250                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1251                     !list_empty(&current->signal->cpu_timers[0]) ||
1252                     !list_empty(&current->signal->cpu_timers[1]) ||
1253                     !list_empty(&current->signal->cpu_timers[2])) {
1254                         /*
1255                          * Have child wake up on its first tick to check
1256                          * for process CPU timers.
1257                          */
1258                         p->it_prof_expires = jiffies_to_cputime(1);
1259                 }
1260         }
1261
1262         /*
1263          * inherit ioprio
1264          */
1265         p->ioprio = current->ioprio;
1266
1267         if (likely(p->pid)) {
1268                 add_parent(p);
1269                 if (unlikely(p->ptrace & PT_PTRACED))
1270                         __ptrace_link(p, current->parent);
1271
1272                 if (thread_group_leader(p)) {
1273                         p->signal->tty = current->signal->tty;
1274                         p->signal->pgrp = process_group(current);
1275                         p->signal->session = current->signal->session;
1276                         attach_pid(p, PIDTYPE_PGID, process_group(p));
1277                         attach_pid(p, PIDTYPE_SID, p->signal->session);
1278
1279                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1280                         __get_cpu_var(process_counts)++;
1281                 }
1282                 attach_pid(p, PIDTYPE_PID, p->pid);
1283                 nr_threads++;
1284         }
1285
1286         total_forks++;
1287         spin_unlock(&current->sighand->siglock);
1288
1289         /* p is copy of current */
1290         vxi = p->vx_info;
1291         if (vxi) {
1292                 claim_vx_info(vxi, p);
1293                 atomic_inc(&vxi->cvirt.nr_threads);
1294                 atomic_inc(&vxi->cvirt.total_forks);
1295                 vx_nproc_inc(p);
1296         }
1297         nxi = p->nx_info;
1298         if (nxi)
1299                 claim_nx_info(nxi, p);
1300         write_unlock_irq(&tasklist_lock);
1301         proc_fork_connector(p);
1302         return p;
1303
1304 bad_fork_cleanup_namespace:
1305         exit_namespace(p);
1306 bad_fork_cleanup_keys:
1307         exit_keys(p);
1308 bad_fork_cleanup_mm:
1309         if (p->mm)
1310                 mmput(p->mm);
1311 bad_fork_cleanup_signal:
1312         cleanup_signal(p);
1313 bad_fork_cleanup_sighand:
1314         __cleanup_sighand(p->sighand);
1315 bad_fork_cleanup_fs:
1316         exit_fs(p); /* blocking */
1317 bad_fork_cleanup_files:
1318         exit_files(p); /* blocking */
1319 bad_fork_cleanup_semundo:
1320         exit_sem(p);
1321 bad_fork_cleanup_audit:
1322         audit_free(p);
1323 bad_fork_cleanup_security:
1324         security_task_free(p);
1325 bad_fork_cleanup_policy:
1326 #ifdef CONFIG_NUMA
1327         mpol_free(p->mempolicy);
1328 bad_fork_cleanup_cpuset:
1329 #endif
1330         cpuset_exit(p);
1331 bad_fork_cleanup_delays_binfmt:
1332         delayacct_tsk_free(p);
1333         if (p->binfmt)
1334                 module_put(p->binfmt->module);
1335 bad_fork_cleanup_put_domain:
1336         module_put(task_thread_info(p)->exec_domain->module);
1337 bad_fork_cleanup_count:
1338         put_group_info(p->group_info);
1339         atomic_dec(&p->user->processes);
1340         free_uid(p->user);
1341 bad_fork_cleanup_vm:
1342         if (p->mm && !(clone_flags & CLONE_VM))
1343                 vx_pages_sub(p->vx_info, RLIMIT_AS, p->mm->total_vm);
1344 bad_fork_free:
1345         free_task(p);
1346 fork_out:
1347         return ERR_PTR(retval);
1348 }
1349
1350 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1351 {
1352         memset(regs, 0, sizeof(struct pt_regs));
1353         return regs;
1354 }
1355
1356 struct task_struct * __devinit fork_idle(int cpu)
1357 {
1358         struct task_struct *task;
1359         struct pt_regs regs;
1360
1361         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1362         if (!task)
1363                 return ERR_PTR(-ENOMEM);
1364         init_idle(task, cpu);
1365
1366         return task;
1367 }
1368
1369 static inline int fork_traceflag (unsigned clone_flags)
1370 {
1371         if (clone_flags & CLONE_UNTRACED)
1372                 return 0;
1373         else if (clone_flags & CLONE_VFORK) {
1374                 if (current->ptrace & PT_TRACE_VFORK)
1375                         return PTRACE_EVENT_VFORK;
1376         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1377                 if (current->ptrace & PT_TRACE_CLONE)
1378                         return PTRACE_EVENT_CLONE;
1379         } else if (current->ptrace & PT_TRACE_FORK)
1380                 return PTRACE_EVENT_FORK;
1381
1382         return 0;
1383 }
1384
1385 /*
1386  *  Ok, this is the main fork-routine.
1387  *
1388  * It copies the process, and if successful kick-starts
1389  * it and waits for it to finish using the VM if required.
1390  */
1391 long do_fork(unsigned long clone_flags,
1392               unsigned long stack_start,
1393               struct pt_regs *regs,
1394               unsigned long stack_size,
1395               int __user *parent_tidptr,
1396               int __user *child_tidptr)
1397 {
1398         struct task_struct *p;
1399         int trace = 0;
1400         struct pid *pid = alloc_pid();
1401         long nr;
1402
1403         if (!pid)
1404                 return -EAGAIN;
1405         nr = pid->nr;
1406         if (unlikely(current->ptrace)) {
1407                 trace = fork_traceflag (clone_flags);
1408                 if (trace)
1409                         clone_flags |= CLONE_PTRACE;
1410         }
1411
1412         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, nr);
1413         /*
1414          * Do this prior waking up the new thread - the thread pointer
1415          * might get invalid after that point, if the thread exits quickly.
1416          */
1417         if (!IS_ERR(p)) {
1418                 struct completion vfork;
1419
1420                 if (clone_flags & CLONE_VFORK) {
1421                         p->vfork_done = &vfork;
1422                         init_completion(&vfork);
1423                 }
1424
1425                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1426                         /*
1427                          * We'll start up with an immediate SIGSTOP.
1428                          */
1429                         sigaddset(&p->pending.signal, SIGSTOP);
1430                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1431                 }
1432
1433                 if (!(clone_flags & CLONE_STOPPED))
1434                         wake_up_new_task(p, clone_flags);
1435                 else
1436                         p->state = TASK_STOPPED;
1437
1438                 if (unlikely (trace)) {
1439                         current->ptrace_message = nr;
1440                         ptrace_notify ((trace << 8) | SIGTRAP);
1441                 }
1442
1443                 if (clone_flags & CLONE_VFORK) {
1444                         wait_for_completion(&vfork);
1445                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1446                                 current->ptrace_message = nr;
1447                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1448                         }
1449                 }
1450         } else {
1451                 free_pid(pid);
1452                 nr = PTR_ERR(p);
1453         }
1454         return nr;
1455 }
1456
1457 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1458 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1459 #endif
1460
1461 static void sighand_ctor(void *data, kmem_cache_t *cachep, unsigned long flags)
1462 {
1463         struct sighand_struct *sighand = data;
1464
1465         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
1466                                         SLAB_CTOR_CONSTRUCTOR)
1467                 spin_lock_init(&sighand->siglock);
1468 }
1469
1470 void __init proc_caches_init(void)
1471 {
1472         sighand_cachep = kmem_cache_create("sighand_cache",
1473                         sizeof(struct sighand_struct), 0,
1474                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1475                         sighand_ctor, NULL);
1476         signal_cachep = kmem_cache_create("signal_cache",
1477                         sizeof(struct signal_struct), 0,
1478                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1479         files_cachep = kmem_cache_create("files_cache",
1480                         sizeof(struct files_struct), 0,
1481                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1482         fs_cachep = kmem_cache_create("fs_cache",
1483                         sizeof(struct fs_struct), 0,
1484                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1485         vm_area_cachep = kmem_cache_create("vm_area_struct",
1486                         sizeof(struct vm_area_struct), 0,
1487                         SLAB_PANIC, NULL, NULL);
1488         mm_cachep = kmem_cache_create("mm_struct",
1489                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1490                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1491 }
1492
1493
1494 /*
1495  * Check constraints on flags passed to the unshare system call and
1496  * force unsharing of additional process context as appropriate.
1497  */
1498 static inline void check_unshare_flags(unsigned long *flags_ptr)
1499 {
1500         /*
1501          * If unsharing a thread from a thread group, must also
1502          * unshare vm.
1503          */
1504         if (*flags_ptr & CLONE_THREAD)
1505                 *flags_ptr |= CLONE_VM;
1506
1507         /*
1508          * If unsharing vm, must also unshare signal handlers.
1509          */
1510         if (*flags_ptr & CLONE_VM)
1511                 *flags_ptr |= CLONE_SIGHAND;
1512
1513         /*
1514          * If unsharing signal handlers and the task was created
1515          * using CLONE_THREAD, then must unshare the thread
1516          */
1517         if ((*flags_ptr & CLONE_SIGHAND) &&
1518             (atomic_read(&current->signal->count) > 1))
1519                 *flags_ptr |= CLONE_THREAD;
1520
1521         /*
1522          * If unsharing namespace, must also unshare filesystem information.
1523          */
1524         if (*flags_ptr & CLONE_NEWNS)
1525                 *flags_ptr |= CLONE_FS;
1526 }
1527
1528 /*
1529  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1530  */
1531 static int unshare_thread(unsigned long unshare_flags)
1532 {
1533         if (unshare_flags & CLONE_THREAD)
1534                 return -EINVAL;
1535
1536         return 0;
1537 }
1538
1539 /*
1540  * Unshare the filesystem structure if it is being shared
1541  */
1542 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1543 {
1544         struct fs_struct *fs = current->fs;
1545
1546         if ((unshare_flags & CLONE_FS) &&
1547             (fs && atomic_read(&fs->count) > 1)) {
1548                 *new_fsp = __copy_fs_struct(current->fs);
1549                 if (!*new_fsp)
1550                         return -ENOMEM;
1551         }
1552
1553         return 0;
1554 }
1555
1556 /*
1557  * Unshare the namespace structure if it is being shared
1558  */
1559 static int unshare_namespace(unsigned long unshare_flags, struct namespace **new_nsp, struct fs_struct *new_fs)
1560 {
1561         struct namespace *ns = current->namespace;
1562
1563         if ((unshare_flags & CLONE_NEWNS) &&
1564             (ns && atomic_read(&ns->count) > 1)) {
1565                 if (!capable(CAP_SYS_ADMIN))
1566                         return -EPERM;
1567
1568                 *new_nsp = dup_namespace(current, new_fs ? new_fs : current->fs);
1569                 if (!*new_nsp)
1570                         return -ENOMEM;
1571         }
1572
1573         return 0;
1574 }
1575
1576 /*
1577  * Unsharing of sighand for tasks created with CLONE_SIGHAND is not
1578  * supported yet
1579  */
1580 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1581 {
1582         struct sighand_struct *sigh = current->sighand;
1583
1584         if ((unshare_flags & CLONE_SIGHAND) &&
1585             (sigh && atomic_read(&sigh->count) > 1))
1586                 return -EINVAL;
1587         else
1588                 return 0;
1589 }
1590
1591 /*
1592  * Unshare vm if it is being shared
1593  */
1594 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1595 {
1596         struct mm_struct *mm = current->mm;
1597
1598         if ((unshare_flags & CLONE_VM) &&
1599             (mm && atomic_read(&mm->mm_users) > 1)) {
1600                 return -EINVAL;
1601         }
1602
1603         return 0;
1604 }
1605
1606 /*
1607  * Unshare file descriptor table if it is being shared
1608  */
1609 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1610 {
1611         struct files_struct *fd = current->files;
1612         int error = 0;
1613
1614         if ((unshare_flags & CLONE_FILES) &&
1615             (fd && atomic_read(&fd->count) > 1)) {
1616                 *new_fdp = dup_fd(fd, &error);
1617                 if (!*new_fdp)
1618                         return error;
1619         }
1620
1621         return 0;
1622 }
1623
1624 /*
1625  * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1626  * supported yet
1627  */
1628 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1629 {
1630         if (unshare_flags & CLONE_SYSVSEM)
1631                 return -EINVAL;
1632
1633         return 0;
1634 }
1635
1636 /*
1637  * unshare allows a process to 'unshare' part of the process
1638  * context which was originally shared using clone.  copy_*
1639  * functions used by do_fork() cannot be used here directly
1640  * because they modify an inactive task_struct that is being
1641  * constructed. Here we are modifying the current, active,
1642  * task_struct.
1643  */
1644 asmlinkage long sys_unshare(unsigned long unshare_flags)
1645 {
1646         int err = 0;
1647         struct fs_struct *fs, *new_fs = NULL;
1648         struct namespace *ns, *new_ns = NULL;
1649         struct sighand_struct *sigh, *new_sigh = NULL;
1650         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1651         struct files_struct *fd, *new_fd = NULL;
1652         struct sem_undo_list *new_ulist = NULL;
1653
1654         check_unshare_flags(&unshare_flags);
1655
1656         /* Return -EINVAL for all unsupported flags */
1657         err = -EINVAL;
1658         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1659                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM))
1660                 goto bad_unshare_out;
1661
1662         if ((err = unshare_thread(unshare_flags)))
1663                 goto bad_unshare_out;
1664         if ((err = unshare_fs(unshare_flags, &new_fs)))
1665                 goto bad_unshare_cleanup_thread;
1666         if ((err = unshare_namespace(unshare_flags, &new_ns, new_fs)))
1667                 goto bad_unshare_cleanup_fs;
1668         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1669                 goto bad_unshare_cleanup_ns;
1670         if ((err = unshare_vm(unshare_flags, &new_mm)))
1671                 goto bad_unshare_cleanup_sigh;
1672         if ((err = unshare_fd(unshare_flags, &new_fd)))
1673                 goto bad_unshare_cleanup_vm;
1674         if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1675                 goto bad_unshare_cleanup_fd;
1676
1677         if (new_fs || new_ns || new_sigh || new_mm || new_fd || new_ulist) {
1678
1679                 task_lock(current);
1680
1681                 if (new_fs) {
1682                         fs = current->fs;
1683                         current->fs = new_fs;
1684                         new_fs = fs;
1685                 }
1686
1687                 if (new_ns) {
1688                         ns = current->namespace;
1689                         current->namespace = new_ns;
1690                         new_ns = ns;
1691                 }
1692
1693                 if (new_sigh) {
1694                         sigh = current->sighand;
1695                         rcu_assign_pointer(current->sighand, new_sigh);
1696                         new_sigh = sigh;
1697                 }
1698
1699                 if (new_mm) {
1700                         mm = current->mm;
1701                         active_mm = current->active_mm;
1702                         current->mm = new_mm;
1703                         current->active_mm = new_mm;
1704                         activate_mm(active_mm, new_mm);
1705                         new_mm = mm;
1706                 }
1707
1708                 if (new_fd) {
1709                         fd = current->files;
1710                         current->files = new_fd;
1711                         new_fd = fd;
1712                 }
1713
1714                 task_unlock(current);
1715         }
1716
1717 bad_unshare_cleanup_fd:
1718         if (new_fd)
1719                 put_files_struct(new_fd);
1720
1721 bad_unshare_cleanup_vm:
1722         if (new_mm)
1723                 mmput(new_mm);
1724
1725 bad_unshare_cleanup_sigh:
1726         if (new_sigh)
1727                 if (atomic_dec_and_test(&new_sigh->count))
1728                         kmem_cache_free(sighand_cachep, new_sigh);
1729
1730 bad_unshare_cleanup_ns:
1731         if (new_ns)
1732                 put_namespace(new_ns);
1733
1734 bad_unshare_cleanup_fs:
1735         if (new_fs)
1736                 put_fs_struct(new_fs);
1737
1738 bad_unshare_cleanup_thread:
1739 bad_unshare_out:
1740         return err;
1741 }