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