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/config.h>
  15 #include <linux/slab.h>
  16 #include <linux/init.h>
  17 #include <linux/unistd.h>
  18 #include <linux/smp_lock.h>
  19 #include <linux/module.h>
  20 #include <linux/vmalloc.h>
  21 #include <linux/completion.h>
  22 #include <linux/namespace.h>
  23 #include <linux/personality.h>
  24 #include <linux/mempolicy.h>
  25 #include <linux/sem.h>
  26 #include <linux/file.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/fs.h>
  30 #include <linux/cpu.h>
  31 #include <linux/security.h>
  32 #include <linux/syscalls.h>
  33 #include <linux/jiffies.h>
  34 #include <linux/futex.h>
  35 #include <linux/ptrace.h>
  36 #include <linux/mount.h>
  37 #include <linux/audit.h>
  38 #include <linux/rmap.h>
  39 #include <linux/vinline.h>
  40 #include <linux/ninline.h>
  41
  42 #include <asm/pgtable.h>
  43 #include <asm/pgalloc.h>
  44 #include <asm/uaccess.h>
  45 #include <asm/mmu_context.h>
  46 #include <asm/cacheflush.h>
  47 #include <asm/tlbflush.h>
  48
  49 /* The idle threads do not count..
  50  * Protected by write_lock_irq(&tasklist_lock)
  51  */
  52 int nr_threads;
  53
  54 int max_threads;
  55 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  56
  57 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  58
  59 rwlock_t tasklist_lock __cacheline_aligned = RW_LOCK_UNLOCKED;  /* outer */
  60
  61 EXPORT_SYMBOL(tasklist_lock);
  62
  63 int nr_processes(void)
  64 {
  65         int cpu;
  66         int total = 0;
  67
  68         for_each_online_cpu(cpu)
  69                 total += per_cpu(process_counts, cpu);
  70
  71         return total;
  72 }
  73
  74 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  75 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  76 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  77 static kmem_cache_t *task_struct_cachep;
  78 #endif
  79
  80 static void free_task(struct task_struct *tsk)
  81 {
  82         free_thread_info(tsk->thread_info);
  83         clr_vx_info(&tsk->vx_info);
  84         clr_nx_info(&tsk->nx_info);
  85         free_task_struct(tsk);
  86 }
  87
  88 void __put_task_struct(struct task_struct *tsk)
  89 {
  90         WARN_ON(!(tsk->state & (TASK_DEAD | TASK_ZOMBIE)));
  91         WARN_ON(atomic_read(&tsk->usage));
  92         WARN_ON(tsk == current);
  93
  94         if (unlikely(tsk->audit_context))
  95                 audit_free(tsk);
  96         security_task_free(tsk);
  97         free_uid(tsk->user);
  98         put_group_info(tsk->group_info);
  99         free_task(tsk);
 100 }
 101
 102 void fastcall add_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
 103 {
 104         unsigned long flags;
 105
 106         wait->flags &= ~WQ_FLAG_EXCLUSIVE;
 107         spin_lock_irqsave(&q->lock, flags);
 108         __add_wait_queue(q, wait);
 109         spin_unlock_irqrestore(&q->lock, flags);
 110 }
 111
 112 EXPORT_SYMBOL(add_wait_queue);
 113
 114 void fastcall add_wait_queue_exclusive(wait_queue_head_t *q, wait_queue_t * wait)
 115 {
 116         unsigned long flags;
 117
 118         wait->flags |= WQ_FLAG_EXCLUSIVE;
 119         spin_lock_irqsave(&q->lock, flags);
 120         __add_wait_queue_tail(q, wait);
 121         spin_unlock_irqrestore(&q->lock, flags);
 122 }
 123
 124 EXPORT_SYMBOL(add_wait_queue_exclusive);
 125
 126 void fastcall remove_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
 127 {
 128         unsigned long flags;
 129
 130         spin_lock_irqsave(&q->lock, flags);
 131         __remove_wait_queue(q, wait);
 132         spin_unlock_irqrestore(&q->lock, flags);
 133 }
 134
 135 EXPORT_SYMBOL(remove_wait_queue);
 136
 137
 138 /*
 139  * Note: we use "set_current_state()" _after_ the wait-queue add,
 140  * because we need a memory barrier there on SMP, so that any
 141  * wake-function that tests for the wait-queue being active
 142  * will be guaranteed to see waitqueue addition _or_ subsequent
 143  * tests in this thread will see the wakeup having taken place.
 144  *
 145  * The spin_unlock() itself is semi-permeable and only protects
 146  * one way (it only protects stuff inside the critical region and
 147  * stops them from bleeding out - it would still allow subsequent
 148  * loads to move into the the critical region).
 149  */
 150 void fastcall prepare_to_wait(wait_queue_head_t *q, wait_queue_t *wait, int state)
 151 {
 152         unsigned long flags;
 153
 154         wait->flags &= ~WQ_FLAG_EXCLUSIVE;
 155         spin_lock_irqsave(&q->lock, flags);
 156         if (list_empty(&wait->task_list))
 157                 __add_wait_queue(q, wait);
 158         set_current_state(state);
 159         spin_unlock_irqrestore(&q->lock, flags);
 160 }
 161
 162 EXPORT_SYMBOL(prepare_to_wait);
 163
 164 void fastcall
 165 prepare_to_wait_exclusive(wait_queue_head_t *q, wait_queue_t *wait, int state)
 166 {
 167         unsigned long flags;
 168
 169         wait->flags |= WQ_FLAG_EXCLUSIVE;
 170         spin_lock_irqsave(&q->lock, flags);
 171         if (list_empty(&wait->task_list))
 172                 __add_wait_queue_tail(q, wait);
 173         set_current_state(state);
 174         spin_unlock_irqrestore(&q->lock, flags);
 175 }
 176
 177 EXPORT_SYMBOL(prepare_to_wait_exclusive);
 178
 179 void fastcall finish_wait(wait_queue_head_t *q, wait_queue_t *wait)
 180 {
 181         unsigned long flags;
 182
 183         __set_current_state(TASK_RUNNING);
 184         /*
 185          * We can check for list emptiness outside the lock
 186          * IFF:
 187          *  - we use the "careful" check that verifies both
 188          *    the next and prev pointers, so that there cannot
 189          *    be any half-pending updates in progress on other
 190          *    CPU's that we haven't seen yet (and that might
 191          *    still change the stack area.
 192          * and
 193          *  - all other users take the lock (ie we can only
 194          *    have _one_ other CPU that looks at or modifies
 195          *    the list).
 196          */
 197         if (!list_empty_careful(&wait->task_list)) {
 198                 spin_lock_irqsave(&q->lock, flags);
 199                 list_del_init(&wait->task_list);
 200                 spin_unlock_irqrestore(&q->lock, flags);
 201         }
 202 }
 203
 204 EXPORT_SYMBOL(finish_wait);
 205
 206 int autoremove_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
 207 {
 208         int ret = default_wake_function(wait, mode, sync, key);
 209
 210         if (ret)
 211                 list_del_init(&wait->task_list);
 212         return ret;
 213 }
 214
 215 EXPORT_SYMBOL(autoremove_wake_function);
 216
 217 void __init fork_init(unsigned long mempages)
 218 {
 219 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 220 #ifndef ARCH_MIN_TASKALIGN
 221 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 222 #endif
 223         /* create a slab on which task_structs can be allocated */
 224         task_struct_cachep =
 225                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 226                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
 227 #endif
 228
 229         /*
 230          * The default maximum number of threads is set to a safe
 231          * value: the thread structures can take up at most half
 232          * of memory.
 233          */
 234         max_threads = mempages / (THREAD_SIZE/PAGE_SIZE) / 8;
 235         /*
 236          * we need to allow at least 20 threads to boot a system
 237          */
 238         if(max_threads < 20)
 239                 max_threads = 20;
 240
 241         init_task.rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 242         init_task.rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 243 }
 244
 245 static struct task_struct *dup_task_struct(struct task_struct *orig)
 246 {
 247         struct task_struct *tsk;
 248         struct thread_info *ti;
 249
 250         prepare_to_copy(orig);
 251
 252         tsk = alloc_task_struct();
 253         if (!tsk)
 254                 return NULL;
 255
 256         ti = alloc_thread_info(tsk);
 257         if (!ti) {
 258                 free_task_struct(tsk);
 259                 return NULL;
 260         }
 261
 262         *ti = *orig->thread_info;
 263         *tsk = *orig;
 264         tsk->thread_info = ti;
 265         ti->task = tsk;
 266
 267         /* One for us, one for whoever does the "release_task()" (usually parent) */
 268         atomic_set(&tsk->usage,2);
 269         return tsk;
 270 }
 271
 272 #ifdef CONFIG_MMU
 273 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
 274 {
 275         struct vm_area_struct * mpnt, *tmp, **pprev;
 276         struct rb_node **rb_link, *rb_parent;
 277         int retval;
 278         unsigned long charge = 0;
 279         struct mempolicy *pol;
 280
 281         down_write(&oldmm->mmap_sem);
 282         flush_cache_mm(current->mm);
 283         mm->locked_vm = 0;
 284         mm->mmap = NULL;
 285         mm->mmap_cache = NULL;
 286         mm->free_area_cache = TASK_UNMAPPED_BASE;
 287         mm->map_count = 0;
 288         mm->rss = 0;
 289         cpus_clear(mm->cpu_vm_mask);
 290         mm->mm_rb = RB_ROOT;
 291         rb_link = &mm->mm_rb.rb_node;
 292         rb_parent = NULL;
 293         pprev = &mm->mmap;
 294
 295         /*
 296          * Add it to the mmlist after the parent.
 297          * Doing it this way means that we can order the list,
 298          * and fork() won't mess up the ordering significantly.
 299          * Add it first so that swapoff can see any swap entries.
 300          */
 301         spin_lock(&mmlist_lock);
 302         list_add(&mm->mmlist, &current->mm->mmlist);
 303         mmlist_nr++;
 304         spin_unlock(&mmlist_lock);
 305
 306         for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
 307                 struct file *file;
 308
 309                 if(mpnt->vm_flags & VM_DONTCOPY)
 310                         continue;
 311                 if (mpnt->vm_flags & VM_ACCOUNT) {
 312                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 313                         if (security_vm_enough_memory(len))
 314                                 goto fail_nomem;
 315                         charge += len;
 316                 }
 317                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 318                 if (!tmp)
 319                         goto fail_nomem;
 320                 *tmp = *mpnt;
 321                 pol = mpol_copy(vma_policy(mpnt));
 322                 retval = PTR_ERR(pol);
 323                 if (IS_ERR(pol))
 324                         goto fail_nomem_policy;
 325                 vma_set_policy(tmp, pol);
 326                 tmp->vm_flags &= ~VM_LOCKED;
 327                 tmp->vm_mm = mm;
 328                 tmp->vm_next = NULL;
 329                 anon_vma_link(tmp);
 330                 vma_prio_tree_init(tmp);
 331                 file = tmp->vm_file;
 332                 if (file) {
 333                         struct inode *inode = file->f_dentry->d_inode;
 334                         get_file(file);
 335                         if (tmp->vm_flags & VM_DENYWRITE)
 336                                 atomic_dec(&inode->i_writecount);
 337
 338                         /* insert tmp into the share list, just after mpnt */
 339                         spin_lock(&file->f_mapping->i_mmap_lock);
 340                         flush_dcache_mmap_lock(file->f_mapping);
 341                         vma_prio_tree_add(tmp, mpnt);
 342                         flush_dcache_mmap_unlock(file->f_mapping);
 343                         spin_unlock(&file->f_mapping->i_mmap_lock);
 344                 }
 345
 346                 /*
 347                  * Link in the new vma and copy the page table entries:
 348                  * link in first so that swapoff can see swap entries,
 349                  * and try_to_unmap_one's find_vma find the new vma.
 350                  */
 351                 spin_lock(&mm->page_table_lock);
 352                 *pprev = tmp;
 353                 pprev = &tmp->vm_next;
 354
 355                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 356                 rb_link = &tmp->vm_rb.rb_right;
 357                 rb_parent = &tmp->vm_rb;
 358
 359                 mm->map_count++;
 360                 retval = copy_page_range(mm, current->mm, tmp);
 361                 spin_unlock(&mm->page_table_lock);
 362
 363                 if (tmp->vm_ops && tmp->vm_ops->open)
 364                         tmp->vm_ops->open(tmp);
 365
 366                 if (retval)
 367                         goto fail;
 368         }
 369         retval = 0;
 370
 371 out:
 372         flush_tlb_mm(current->mm);
 373         up_write(&oldmm->mmap_sem);
 374         return retval;
 375 fail_nomem_policy:
 376         kmem_cache_free(vm_area_cachep, tmp);
 377 fail_nomem:
 378         retval = -ENOMEM;
 379 fail:
 380         vm_unacct_memory(charge);
 381         goto out;
 382 }
 383 static inline int mm_alloc_pgd(struct mm_struct * mm)
 384 {
 385         mm->pgd = pgd_alloc(mm);
 386         if (unlikely(!mm->pgd))
 387                 return -ENOMEM;
 388         return 0;
 389 }
 390
 391 static inline void mm_free_pgd(struct mm_struct * mm)
 392 {
 393         pgd_free(mm->pgd);
 394 }
 395 #else
 396 #define dup_mmap(mm, oldmm)     (0)
 397 #define mm_alloc_pgd(mm)        (0)
 398 #define mm_free_pgd(mm)
 399 #endif /* CONFIG_MMU */
 400
 401 spinlock_t mmlist_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
 402 int mmlist_nr;
 403
 404 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
 405 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 406
 407 #include <linux/init_task.h>
 408
 409 static struct mm_struct * mm_init(struct mm_struct * mm)
 410 {
 411         atomic_set(&mm->mm_users, 1);
 412         atomic_set(&mm->mm_count, 1);
 413         init_rwsem(&mm->mmap_sem);
 414         mm->core_waiters = 0;
 415         mm->page_table_lock = SPIN_LOCK_UNLOCKED;
 416         mm->ioctx_list_lock = RW_LOCK_UNLOCKED;
 417         mm->ioctx_list = NULL;
 418         mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
 419         mm->free_area_cache = TASK_UNMAPPED_BASE;
 420
 421         if (likely(!mm_alloc_pgd(mm))) {
 422                 mm->def_flags = 0;
 423 #ifdef __HAVE_ARCH_MMAP_TOP
 424                 mm->mmap_top = mmap_top();
 425 #endif
 426                 set_vx_info(&mm->mm_vx_info, current->vx_info);
 427                 return mm;
 428         }
 429         free_mm(mm);
 430         return NULL;
 431 }
 432
 433 /*
 434  * Allocate and initialize an mm_struct.
 435  */
 436 struct mm_struct * mm_alloc(void)
 437 {
 438         struct mm_struct * mm;
 439
 440         mm = allocate_mm();
 441         if (mm) {
 442                 memset(mm, 0, sizeof(*mm));
 443                 mm = mm_init(mm);
 444         }
 445         return mm;
 446 }
 447
 448 /*
 449  * Called when the last reference to the mm
 450  * is dropped: either by a lazy thread or by
 451  * mmput. Free the page directory and the mm.
 452  */
 453 void fastcall __mmdrop(struct mm_struct *mm)
 454 {
 455         BUG_ON(mm == &init_mm);
 456         mm_free_pgd(mm);
 457         destroy_context(mm);
 458         clr_vx_info(&mm->mm_vx_info);
 459         free_mm(mm);
 460 }
 461
 462 /*
 463  * Decrement the use count and release all resources for an mm.
 464  */
 465 void mmput(struct mm_struct *mm)
 466 {
 467         if (atomic_dec_and_lock(&mm->mm_users, &mmlist_lock)) {
 468                 list_del(&mm->mmlist);
 469                 mmlist_nr--;
 470                 spin_unlock(&mmlist_lock);
 471                 exit_aio(mm);
 472                 exit_mmap(mm);
 473                 mmdrop(mm);
 474         }
 475 }
 476
 477 /*
 478  * Checks if the use count of an mm is non-zero and if so
 479  * returns a reference to it after bumping up the use count.
 480  * If the use count is zero, it means this mm is going away,
 481  * so return NULL.
 482  */
 483 struct mm_struct *mmgrab(struct mm_struct *mm)
 484 {
 485         spin_lock(&mmlist_lock);
 486         if (!atomic_read(&mm->mm_users))
 487                 mm = NULL;
 488         else
 489                 atomic_inc(&mm->mm_users);
 490         spin_unlock(&mmlist_lock);
 491         return mm;
 492 }
 493
 494 /* Please note the differences between mmput and mm_release.
 495  * mmput is called whenever we stop holding onto a mm_struct,
 496  * error success whatever.
 497  *
 498  * mm_release is called after a mm_struct has been removed
 499  * from the current process.
 500  *
 501  * This difference is important for error handling, when we
 502  * only half set up a mm_struct for a new process and need to restore
 503  * the old one.  Because we mmput the new mm_struct before
 504  * restoring the old one. . .
 505  * Eric Biederman 10 January 1998
 506  */
 507 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 508 {
 509         struct completion *vfork_done = tsk->vfork_done;
 510
 511         /* Get rid of any cached register state */
 512         deactivate_mm(tsk, mm);
 513
 514         /* notify parent sleeping on vfork() */
 515         if (vfork_done) {
 516                 tsk->vfork_done = NULL;
 517                 complete(vfork_done);
 518         }
 519         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
 520                 u32 __user * tidptr = tsk->clear_child_tid;
 521                 tsk->clear_child_tid = NULL;
 522
 523                 /*
 524                  * We don't check the error code - if userspace has
 525                  * not set up a proper pointer then tough luck.
 526                  */
 527                 put_user(0, tidptr);
 528                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL);
 529         }
 530 }
 531
 532 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 533 {
 534         struct mm_struct * mm, *oldmm;
 535         int retval;
 536
 537         tsk->min_flt = tsk->maj_flt = 0;
 538         tsk->cmin_flt = tsk->cmaj_flt = 0;
 539         tsk->nvcsw = tsk->nivcsw = tsk->cnvcsw = tsk->cnivcsw = 0;
 540
 541         tsk->mm = NULL;
 542         tsk->active_mm = NULL;
 543
 544         /*
 545          * Are we cloning a kernel thread?
 546          *
 547          * We need to steal a active VM for that..
 548          */
 549         oldmm = current->mm;
 550         if (!oldmm)
 551                 return 0;
 552
 553         if (clone_flags & CLONE_VM) {
 554                 atomic_inc(&oldmm->mm_users);
 555                 mm = oldmm;
 556                 /*
 557                  * There are cases where the PTL is held to ensure no
 558                  * new threads start up in user mode using an mm, which
 559                  * allows optimizing out ipis; the tlb_gather_mmu code
 560                  * is an example.
 561                  */
 562                 spin_unlock_wait(&oldmm->page_table_lock);
 563                 goto good_mm;
 564         }
 565
 566         retval = -ENOMEM;
 567         mm = allocate_mm();
 568         if (!mm)
 569                 goto fail_nomem;
 570
 571         /* Copy the current MM stuff.. */
 572         memcpy(mm, oldmm, sizeof(*mm));
 573         mm->mm_vx_info = NULL;
 574         if (!mm_init(mm))
 575                 goto fail_nomem;
 576
 577         if (init_new_context(tsk,mm))
 578                 goto fail_nocontext;
 579
 580         retval = dup_mmap(mm, oldmm);
 581         if (retval)
 582                 goto free_pt;
 583
 584 good_mm:
 585         tsk->mm = mm;
 586         tsk->active_mm = mm;
 587         return 0;
 588
 589 free_pt:
 590         mmput(mm);
 591 fail_nomem:
 592         return retval;
 593
 594 fail_nocontext:
 595         /*
 596          * If init_new_context() failed, we cannot use mmput() to free the mm
 597          * because it calls destroy_context()
 598          */
 599         mm_free_pgd(mm);
 600         free_mm(mm);
 601         return retval;
 602 }
 603
 604 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 605 {
 606         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 607         /* We don't need to lock fs - think why ;-) */
 608         if (fs) {
 609                 atomic_set(&fs->count, 1);
 610                 fs->lock = RW_LOCK_UNLOCKED;
 611                 fs->umask = old->umask;
 612                 read_lock(&old->lock);
 613                 fs->rootmnt = mntget(old->rootmnt);
 614                 fs->root = dget(old->root);
 615                 fs->pwdmnt = mntget(old->pwdmnt);
 616                 fs->pwd = dget(old->pwd);
 617                 if (old->altroot) {
 618                         fs->altrootmnt = mntget(old->altrootmnt);
 619                         fs->altroot = dget(old->altroot);
 620                 } else {
 621                         fs->altrootmnt = NULL;
 622                         fs->altroot = NULL;
 623                 }
 624                 read_unlock(&old->lock);
 625         }
 626         return fs;
 627 }
 628
 629 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 630 {
 631         return __copy_fs_struct(old);
 632 }
 633
 634 EXPORT_SYMBOL_GPL(copy_fs_struct);
 635
 636 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 637 {
 638         if (clone_flags & CLONE_FS) {
 639                 atomic_inc(&current->fs->count);
 640                 return 0;
 641         }
 642         tsk->fs = __copy_fs_struct(current->fs);
 643         if (!tsk->fs)
 644                 return -ENOMEM;
 645         return 0;
 646 }
 647
 648 static int count_open_files(struct files_struct *files, int size)
 649 {
 650         int i;
 651
 652         /* Find the last open fd */
 653         for (i = size/(8*sizeof(long)); i > 0; ) {
 654                 if (files->open_fds->fds_bits[--i])
 655                         break;
 656         }
 657         i = (i+1) * 8 * sizeof(long);
 658         return i;
 659 }
 660
 661 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 662 {
 663         struct files_struct *oldf, *newf;
 664         struct file **old_fds, **new_fds;
 665         int open_files, nfds, size, i, error = 0;
 666
 667         /*
 668          * A background process may not have any files ...
 669          */
 670         oldf = current->files;
 671         if (!oldf)
 672                 goto out;
 673
 674         if (clone_flags & CLONE_FILES) {
 675                 atomic_inc(&oldf->count);
 676                 goto out;
 677         }
 678
 679         /*
 680          * Note: we may be using current for both targets (See exec.c)
 681          * This works because we cache current->files (old) as oldf. Don't
 682          * break this.
 683          */
 684         tsk->files = NULL;
 685         error = -ENOMEM;
 686         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
 687         if (!newf)
 688                 goto out;
 689
 690         atomic_set(&newf->count, 1);
 691
 692         newf->file_lock     = SPIN_LOCK_UNLOCKED;
 693         newf->next_fd       = 0;
 694         newf->max_fds       = NR_OPEN_DEFAULT;
 695         newf->max_fdset     = __FD_SETSIZE;
 696         newf->close_on_exec = &newf->close_on_exec_init;
 697         newf->open_fds      = &newf->open_fds_init;
 698         newf->fd            = &newf->fd_array[0];
 699
 700         /* We don't yet have the oldf readlock, but even if the old
 701            fdset gets grown now, we'll only copy up to "size" fds */
 702         size = oldf->max_fdset;
 703         if (size > __FD_SETSIZE) {
 704                 newf->max_fdset = 0;
 705                 spin_lock(&newf->file_lock);
 706                 error = expand_fdset(newf, size-1);
 707                 spin_unlock(&newf->file_lock);
 708                 if (error)
 709                         goto out_release;
 710         }
 711         spin_lock(&oldf->file_lock);
 712
 713         open_files = count_open_files(oldf, size);
 714
 715         /*
 716          * Check whether we need to allocate a larger fd array.
 717          * Note: we're not a clone task, so the open count won't
 718          * change.
 719          */
 720         nfds = NR_OPEN_DEFAULT;
 721         if (open_files > nfds) {
 722                 spin_unlock(&oldf->file_lock);
 723                 newf->max_fds = 0;
 724                 spin_lock(&newf->file_lock);
 725                 error = expand_fd_array(newf, open_files-1);
 726                 spin_unlock(&newf->file_lock);
 727                 if (error)
 728                         goto out_release;
 729                 nfds = newf->max_fds;
 730                 spin_lock(&oldf->file_lock);
 731         }
 732
 733         old_fds = oldf->fd;
 734         new_fds = newf->fd;
 735
 736         memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
 737         memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
 738
 739         for (i = open_files; i != 0; i--) {
 740                 struct file *f = *old_fds++;
 741                 if (f)
 742                         get_file(f);
 743                 *new_fds++ = f;
 744         }
 745         spin_unlock(&oldf->file_lock);
 746
 747         /* compute the remainder to be cleared */
 748         size = (newf->max_fds - open_files) * sizeof(struct file *);
 749
 750         /* This is long word aligned thus could use a optimized version */
 751         memset(new_fds, 0, size);
 752
 753         if (newf->max_fdset > open_files) {
 754                 int left = (newf->max_fdset-open_files)/8;
 755                 int start = open_files / (8 * sizeof(unsigned long));
 756
 757                 memset(&newf->open_fds->fds_bits[start], 0, left);
 758                 memset(&newf->close_on_exec->fds_bits[start], 0, left);
 759         }
 760
 761         tsk->files = newf;
 762         error = 0;
 763 out:
 764         return error;
 765
 766 out_release:
 767         free_fdset (newf->close_on_exec, newf->max_fdset);
 768         free_fdset (newf->open_fds, newf->max_fdset);
 769         kmem_cache_free(files_cachep, newf);
 770         goto out;
 771 }
 772
 773 /*
 774  *      Helper to unshare the files of the current task.
 775  *      We don't want to expose copy_files internals to
 776  *      the exec layer of the kernel.
 777  */
 778
 779 int unshare_files(void)
 780 {
 781         struct files_struct *files  = current->files;
 782         int rc;
 783
 784         if(!files)
 785                 BUG();
 786
 787         /* This can race but the race causes us to copy when we don't
 788            need to and drop the copy */
 789         if(atomic_read(&files->count) == 1)
 790         {
 791                 atomic_inc(&files->count);
 792                 return 0;
 793         }
 794         rc = copy_files(0, current);
 795         if(rc)
 796                 current->files = files;
 797         return rc;
 798 }
 799
 800 EXPORT_SYMBOL(unshare_files);
 801
 802 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 803 {
 804         struct sighand_struct *sig;
 805
 806         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 807                 atomic_inc(&current->sighand->count);
 808                 return 0;
 809         }
 810         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 811         tsk->sighand = sig;
 812         if (!sig)
 813                 return -ENOMEM;
 814         spin_lock_init(&sig->siglock);
 815         atomic_set(&sig->count, 1);
 816         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 817         return 0;
 818 }
 819
 820 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 821 {
 822         struct signal_struct *sig;
 823
 824         if (clone_flags & CLONE_THREAD) {
 825                 atomic_inc(&current->signal->count);
 826                 return 0;
 827         }
 828         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 829         tsk->signal = sig;
 830         if (!sig)
 831                 return -ENOMEM;
 832         atomic_set(&sig->count, 1);
 833         sig->group_exit = 0;
 834         sig->group_exit_code = 0;
 835         sig->group_exit_task = NULL;
 836         sig->group_stop_count = 0;
 837         sig->curr_target = NULL;
 838         init_sigpending(&sig->shared_pending);
 839         INIT_LIST_HEAD(&sig->posix_timers);
 840
 841         sig->tty = current->signal->tty;
 842         sig->pgrp = process_group(current);
 843         sig->session = current->signal->session;
 844         sig->leader = 0;        /* session leadership doesn't inherit */
 845         sig->tty_old_pgrp = 0;
 846
 847         return 0;
 848 }
 849
 850 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 851 {
 852         unsigned long new_flags = p->flags;
 853
 854         new_flags &= ~PF_SUPERPRIV;
 855         new_flags |= PF_FORKNOEXEC;
 856         if (!(clone_flags & CLONE_PTRACE))
 857                 p->ptrace = 0;
 858         p->flags = new_flags;
 859 }
 860
 861 asmlinkage long sys_set_tid_address(int __user *tidptr)
 862 {
 863         current->clear_child_tid = tidptr;
 864
 865         return current->pid;
 866 }
 867
 868 /*
 869  * This creates a new process as a copy of the old one,
 870  * but does not actually start it yet.
 871  *
 872  * It copies the registers, and all the appropriate
 873  * parts of the process environment (as per the clone
 874  * flags). The actual kick-off is left to the caller.
 875  */
 876 struct task_struct *copy_process(unsigned long clone_flags,
 877                                  unsigned long stack_start,
 878                                  struct pt_regs *regs,
 879                                  unsigned long stack_size,
 880                                  int __user *parent_tidptr,
 881                                  int __user *child_tidptr)
 882 {
 883         int retval;
 884         struct task_struct *p = NULL;
 885         struct vx_info *vxi;
 886         struct nx_info *nxi;
 887
 888         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 889                 return ERR_PTR(-EINVAL);
 890
 891         /*
 892          * Thread groups must share signals as well, and detached threads
 893          * can only be started up within the thread group.
 894          */
 895         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 896                 return ERR_PTR(-EINVAL);
 897
 898         /*
 899          * Shared signal handlers imply shared VM. By way of the above,
 900          * thread groups also imply shared VM. Blocking this case allows
 901          * for various simplifications in other code.
 902          */
 903         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 904                 return ERR_PTR(-EINVAL);
 905
 906         retval = security_task_create(clone_flags);
 907         if (retval)
 908                 goto fork_out;
 909
 910         retval = -ENOMEM;
 911
 912         p = dup_task_struct(current);
 913         if (!p)
 914                 goto fork_out;
 915         p->tux_info = NULL;
 916
 917         vxi = get_vx_info(current->vx_info);
 918         nxi = get_nx_info(current->nx_info);
 919
 920         /* check vserver memory */
 921         if (p->mm && !(clone_flags & CLONE_VM)) {
 922                 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
 923                         vx_pages_add(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
 924                 else
 925                         goto bad_fork_free;
 926         }
 927         if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
 928                 if (!vx_rsspages_avail(p->mm, p->mm->rss))
 929                         goto bad_fork_free;
 930         }
 931
 932         retval = -EAGAIN;
 933         if (vxi && (atomic_read(&vxi->limit.res[RLIMIT_NPROC])
 934                 >= vxi->limit.rlim[RLIMIT_NPROC]))
 935                 goto bad_fork_free;
 936
 937         if (atomic_read(&p->user->processes) >=
 938                         p->rlim[RLIMIT_NPROC].rlim_cur) {
 939                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 940                                 p->user != &root_user)
 941                         goto bad_fork_free;
 942         }
 943
 944         atomic_inc(&p->user->__count);
 945         atomic_inc(&p->user->processes);
 946         get_group_info(p->group_info);
 947
 948         /*
 949          * If multiple threads are within copy_process(), then this check
 950          * triggers too late. This doesn't hurt, the check is only there
 951          * to stop root fork bombs.
 952          */
 953         if (nr_threads >= max_threads)
 954                 goto bad_fork_cleanup_count;
 955
 956         if (!try_module_get(p->thread_info->exec_domain->module))
 957                 goto bad_fork_cleanup_count;
 958
 959         if (p->binfmt && !try_module_get(p->binfmt->module))
 960                 goto bad_fork_cleanup_put_domain;
 961
 962         p->did_exec = 0;
 963         copy_flags(clone_flags, p);
 964         if (clone_flags & CLONE_IDLETASK)
 965                 p->pid = 0;
 966         else {
 967                 p->pid = alloc_pidmap();
 968                 if (p->pid == -1)
 969                         goto bad_fork_cleanup;
 970         }
 971         retval = -EFAULT;
 972         if (clone_flags & CLONE_PARENT_SETTID)
 973                 if (put_user(p->pid, parent_tidptr))
 974                         goto bad_fork_cleanup;
 975
 976         p->proc_dentry = NULL;
 977
 978         INIT_LIST_HEAD(&p->children);
 979         INIT_LIST_HEAD(&p->sibling);
 980         init_waitqueue_head(&p->wait_chldexit);
 981         p->vfork_done = NULL;
 982         spin_lock_init(&p->alloc_lock);
 983         spin_lock_init(&p->proc_lock);
 984
 985         clear_tsk_thread_flag(p, TIF_SIGPENDING);
 986         init_sigpending(&p->pending);
 987
 988         p->it_real_value = p->it_virt_value = p->it_prof_value = 0;
 989         p->it_real_incr = p->it_virt_incr = p->it_prof_incr = 0;
 990         init_timer(&p->real_timer);
 991         p->real_timer.data = (unsigned long) p;
 992
 993         p->utime = p->stime = 0;
 994         p->cutime = p->cstime = 0;
 995         p->lock_depth = -1;             /* -1 = no lock */
 996         p->start_time = get_jiffies_64();
 997         p->security = NULL;
 998         p->io_context = NULL;
 999         p->audit_context = NULL;
1000 #ifdef CONFIG_NUMA
1001         p->mempolicy = mpol_copy(p->mempolicy);
1002         if (IS_ERR(p->mempolicy)) {
1003                 retval = PTR_ERR(p->mempolicy);
1004                 p->mempolicy = NULL;
1005                 goto bad_fork_cleanup;
1006         }
1007 #endif
1008
1009         retval = -ENOMEM;
1010         if ((retval = security_task_alloc(p)))
1011                 goto bad_fork_cleanup_policy;
1012         if ((retval = audit_alloc(p)))
1013                 goto bad_fork_cleanup_security;
1014         /* copy all the process information */
1015         if ((retval = copy_semundo(clone_flags, p)))
1016                 goto bad_fork_cleanup_audit;
1017         if ((retval = copy_files(clone_flags, p)))
1018                 goto bad_fork_cleanup_semundo;
1019         if ((retval = copy_fs(clone_flags, p)))
1020                 goto bad_fork_cleanup_files;
1021         if ((retval = copy_sighand(clone_flags, p)))
1022                 goto bad_fork_cleanup_fs;
1023         if ((retval = copy_signal(clone_flags, p)))
1024                 goto bad_fork_cleanup_sighand;
1025         if ((retval = copy_mm(clone_flags, p)))
1026                 goto bad_fork_cleanup_signal;
1027         if ((retval = copy_namespace(clone_flags, p)))
1028                 goto bad_fork_cleanup_mm;
1029         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1030         if (retval)
1031                 goto bad_fork_cleanup_namespace;
1032
1033         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1034         /*
1035          * Clear TID on mm_release()?
1036          */
1037         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1038
1039         /*
1040          * Syscall tracing should be turned off in the child regardless
1041          * of CLONE_PTRACE.
1042          */
1043         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1044
1045         /* Our parent execution domain becomes current domain
1046            These must match for thread signalling to apply */
1047
1048         p->parent_exec_id = p->self_exec_id;
1049
1050         /* ok, now we should be set up.. */
1051         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1052         p->pdeath_signal = 0;
1053
1054         /* Perform scheduler related setup */
1055         sched_fork(p);
1056
1057         /*
1058          * Ok, make it visible to the rest of the system.
1059          * We dont wake it up yet.
1060          */
1061         p->tgid = p->pid;
1062         p->group_leader = p;
1063         INIT_LIST_HEAD(&p->ptrace_children);
1064         INIT_LIST_HEAD(&p->ptrace_list);
1065
1066         /* Need tasklist lock for parent etc handling! */
1067         write_lock_irq(&tasklist_lock);
1068         /*
1069          * Check for pending SIGKILL! The new thread should not be allowed
1070          * to slip out of an OOM kill. (or normal SIGKILL.)
1071          */
1072         if (sigismember(&current->pending.signal, SIGKILL)) {
1073                 write_unlock_irq(&tasklist_lock);
1074                 retval = -EINTR;
1075                 goto bad_fork_cleanup_namespace;
1076         }
1077
1078         /* CLONE_PARENT re-uses the old parent */
1079         if (clone_flags & CLONE_PARENT)
1080                 p->real_parent = current->real_parent;
1081         else
1082                 p->real_parent = current;
1083         p->parent = p->real_parent;
1084
1085         if (clone_flags & CLONE_THREAD) {
1086                 spin_lock(&current->sighand->siglock);
1087                 /*
1088                  * Important: if an exit-all has been started then
1089                  * do not create this new thread - the whole thread
1090                  * group is supposed to exit anyway.
1091                  */
1092                 if (current->signal->group_exit) {
1093                         spin_unlock(&current->sighand->siglock);
1094                         write_unlock_irq(&tasklist_lock);
1095                         retval = -EAGAIN;
1096                         goto bad_fork_cleanup_namespace;
1097                 }
1098                 p->tgid = current->tgid;
1099                 p->group_leader = current->group_leader;
1100
1101                 if (current->signal->group_stop_count > 0) {
1102                         /*
1103                          * There is an all-stop in progress for the group.
1104                          * We ourselves will stop as soon as we check signals.
1105                          * Make the new thread part of that group stop too.
1106                          */
1107                         current->signal->group_stop_count++;
1108                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1109                 }
1110
1111                 spin_unlock(&current->sighand->siglock);
1112         }
1113
1114         SET_LINKS(p);
1115         if (p->ptrace & PT_PTRACED)
1116                 __ptrace_link(p, current->parent);
1117
1118         attach_pid(p, PIDTYPE_PID, p->pid);
1119         if (thread_group_leader(p)) {
1120                 attach_pid(p, PIDTYPE_TGID, p->tgid);
1121                 attach_pid(p, PIDTYPE_PGID, process_group(p));
1122                 attach_pid(p, PIDTYPE_SID, p->signal->session);
1123                 if (p->pid)
1124                         __get_cpu_var(process_counts)++;
1125         } else
1126                 link_pid(p, p->pids + PIDTYPE_TGID, &p->group_leader->pids[PIDTYPE_TGID].pid);
1127
1128         nr_threads++;
1129         if (vxi) {
1130                 atomic_inc(&vxi->cacct.nr_threads);
1131                 atomic_inc(&vxi->limit.res[RLIMIT_NPROC]);
1132         }
1133         write_unlock_irq(&tasklist_lock);
1134         retval = 0;
1135
1136 fork_out:
1137         if (retval)
1138                 return ERR_PTR(retval);
1139         return p;
1140
1141 bad_fork_cleanup_namespace:
1142         exit_namespace(p);
1143 bad_fork_cleanup_mm:
1144         exit_mm(p);
1145         if (p->active_mm)
1146                 mmdrop(p->active_mm);
1147 bad_fork_cleanup_signal:
1148         exit_signal(p);
1149 bad_fork_cleanup_sighand:
1150         exit_sighand(p);
1151 bad_fork_cleanup_fs:
1152         exit_fs(p); /* blocking */
1153 bad_fork_cleanup_files:
1154         exit_files(p); /* blocking */
1155 bad_fork_cleanup_semundo:
1156         exit_sem(p);
1157 bad_fork_cleanup_audit:
1158         audit_free(p);
1159 bad_fork_cleanup_security:
1160         security_task_free(p);
1161 bad_fork_cleanup_policy:
1162 #ifdef CONFIG_NUMA
1163         mpol_free(p->mempolicy);
1164 #endif
1165 bad_fork_cleanup:
1166         if (p->pid > 0)
1167                 free_pidmap(p->pid);
1168         if (p->binfmt)
1169                 module_put(p->binfmt->module);
1170 bad_fork_cleanup_put_domain:
1171         module_put(p->thread_info->exec_domain->module);
1172 bad_fork_cleanup_count:
1173         put_group_info(p->group_info);
1174         atomic_dec(&p->user->processes);
1175         free_uid(p->user);
1176 bad_fork_free:
1177         free_task(p);
1178         goto fork_out;
1179 }
1180
1181 static inline int fork_traceflag (unsigned clone_flags)
1182 {
1183         if (clone_flags & (CLONE_UNTRACED | CLONE_IDLETASK))
1184                 return 0;
1185         else if (clone_flags & CLONE_VFORK) {
1186                 if (current->ptrace & PT_TRACE_VFORK)
1187                         return PTRACE_EVENT_VFORK;
1188         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1189                 if (current->ptrace & PT_TRACE_CLONE)
1190                         return PTRACE_EVENT_CLONE;
1191         } else if (current->ptrace & PT_TRACE_FORK)
1192                 return PTRACE_EVENT_FORK;
1193
1194         return 0;
1195 }
1196
1197 /*
1198  *  Ok, this is the main fork-routine.
1199  *
1200  * It copies the process, and if successful kick-starts
1201  * it and waits for it to finish using the VM if required.
1202  */
1203 long do_fork(unsigned long clone_flags,
1204               unsigned long stack_start,
1205               struct pt_regs *regs,
1206               unsigned long stack_size,
1207               int __user *parent_tidptr,
1208               int __user *child_tidptr)
1209 {
1210         struct task_struct *p;
1211         int trace = 0;
1212         long pid;
1213
1214         if (unlikely(current->ptrace)) {
1215                 trace = fork_traceflag (clone_flags);
1216                 if (trace)
1217                         clone_flags |= CLONE_PTRACE;
1218         }
1219
1220         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr);
1221         /*
1222          * Do this prior waking up the new thread - the thread pointer
1223          * might get invalid after that point, if the thread exits quickly.
1224          */
1225         pid = IS_ERR(p) ? PTR_ERR(p) : p->pid;
1226
1227         if (!IS_ERR(p)) {
1228                 struct completion vfork;
1229
1230                 if (clone_flags & CLONE_VFORK) {
1231                         p->vfork_done = &vfork;
1232                         init_completion(&vfork);
1233                 }
1234
1235                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1236                         /*
1237                          * We'll start up with an immediate SIGSTOP.
1238                          */
1239                         sigaddset(&p->pending.signal, SIGSTOP);
1240                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1241                 }
1242
1243                 if (!(clone_flags & CLONE_STOPPED)) {
1244                         /*
1245                          * Do the wakeup last. On SMP we treat fork() and
1246                          * CLONE_VM separately, because fork() has already
1247                          * created cache footprint on this CPU (due to
1248                          * copying the pagetables), hence migration would
1249                          * probably be costy. Threads on the other hand
1250                          * have less traction to the current CPU, and if
1251                          * there's an imbalance then the scheduler can
1252                          * migrate this fresh thread now, before it
1253                          * accumulates a larger cache footprint:
1254                          */
1255                         if (clone_flags & CLONE_VM)
1256                                 wake_up_forked_thread(p);
1257                         else
1258                                 wake_up_forked_process(p);
1259                 } else {
1260                         int cpu = get_cpu();
1261
1262                         p->state = TASK_STOPPED;
1263                         if (cpu_is_offline(task_cpu(p)))
1264                                 set_task_cpu(p, cpu);
1265
1266                         put_cpu();
1267                 }
1268                 ++total_forks;
1269
1270                 if (unlikely (trace)) {
1271                         current->ptrace_message = pid;
1272                         ptrace_notify ((trace << 8) | SIGTRAP);
1273                 }
1274
1275                 if (clone_flags & CLONE_VFORK) {
1276                         wait_for_completion(&vfork);
1277                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1278                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1279                 } else
1280                         /*
1281                          * Let the child process run first, to avoid most of the
1282                          * COW overhead when the child exec()s afterwards.
1283                          */
1284                         set_need_resched();
1285         }
1286         return pid;
1287 }
1288
1289 /* SLAB cache for signal_struct structures (tsk->signal) */
1290 kmem_cache_t *signal_cachep;
1291
1292 /* SLAB cache for sighand_struct structures (tsk->sighand) */
1293 kmem_cache_t *sighand_cachep;
1294
1295 /* SLAB cache for files_struct structures (tsk->files) */
1296 kmem_cache_t *files_cachep;
1297
1298 /* SLAB cache for fs_struct structures (tsk->fs) */
1299 kmem_cache_t *fs_cachep;
1300
1301 /* SLAB cache for vm_area_struct structures */
1302 kmem_cache_t *vm_area_cachep;
1303
1304 /* SLAB cache for mm_struct structures (tsk->mm) */
1305 kmem_cache_t *mm_cachep;
1306
1307 void __init proc_caches_init(void)
1308 {
1309         sighand_cachep = kmem_cache_create("sighand_cache",
1310                         sizeof(struct sighand_struct), 0,
1311                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1312         signal_cachep = kmem_cache_create("signal_cache",
1313                         sizeof(struct signal_struct), 0,
1314                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1315         files_cachep = kmem_cache_create("files_cache",
1316                         sizeof(struct files_struct), 0,
1317                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1318         fs_cachep = kmem_cache_create("fs_cache",
1319                         sizeof(struct fs_struct), 0,
1320                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1321         vm_area_cachep = kmem_cache_create("vm_area_struct",
1322                         sizeof(struct vm_area_struct), 0,
1323                         SLAB_PANIC, NULL, NULL);
1324         mm_cachep = kmem_cache_create("mm_struct",
1325                         sizeof(struct mm_struct), 0,
1326                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1327 }