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/vs_network.h>
  40 #include <linux/vs_limit.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         vxdprintk("freeing up task %p\n", tsk);
  84         clr_vx_info(&tsk->vx_info);
  85         clr_nx_info(&tsk->nx_info);
  86         free_task_struct(tsk);
  87 }
  88
  89 void __put_task_struct(struct task_struct *tsk)
  90 {
  91         WARN_ON(!(tsk->state & (TASK_DEAD | TASK_ZOMBIE)));
  92         WARN_ON(atomic_read(&tsk->usage));
  93         WARN_ON(tsk == current);
  94
  95         if (unlikely(tsk->audit_context))
  96                 audit_free(tsk);
  97         security_task_free(tsk);
  98         free_uid(tsk->user);
  99         put_group_info(tsk->group_info);
 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         set_current_state(state);
 160         spin_unlock_irqrestore(&q->lock, flags);
 161 }
 162
 163 EXPORT_SYMBOL(prepare_to_wait);
 164
 165 void fastcall
 166 prepare_to_wait_exclusive(wait_queue_head_t *q, wait_queue_t *wait, int state)
 167 {
 168         unsigned long flags;
 169
 170         wait->flags |= WQ_FLAG_EXCLUSIVE;
 171         spin_lock_irqsave(&q->lock, flags);
 172         if (list_empty(&wait->task_list))
 173                 __add_wait_queue_tail(q, wait);
 174         set_current_state(state);
 175         spin_unlock_irqrestore(&q->lock, flags);
 176 }
 177
 178 EXPORT_SYMBOL(prepare_to_wait_exclusive);
 179
 180 void fastcall finish_wait(wait_queue_head_t *q, wait_queue_t *wait)
 181 {
 182         unsigned long flags;
 183
 184         __set_current_state(TASK_RUNNING);
 185         /*
 186          * We can check for list emptiness outside the lock
 187          * IFF:
 188          *  - we use the "careful" check that verifies both
 189          *    the next and prev pointers, so that there cannot
 190          *    be any half-pending updates in progress on other
 191          *    CPU's that we haven't seen yet (and that might
 192          *    still change the stack area.
 193          * and
 194          *  - all other users take the lock (ie we can only
 195          *    have _one_ other CPU that looks at or modifies
 196          *    the list).
 197          */
 198         if (!list_empty_careful(&wait->task_list)) {
 199                 spin_lock_irqsave(&q->lock, flags);
 200                 list_del_init(&wait->task_list);
 201                 spin_unlock_irqrestore(&q->lock, flags);
 202         }
 203 }
 204
 205 EXPORT_SYMBOL(finish_wait);
 206
 207 int autoremove_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
 208 {
 209         int ret = default_wake_function(wait, mode, sync, key);
 210
 211         if (ret)
 212                 list_del_init(&wait->task_list);
 213         return ret;
 214 }
 215
 216 EXPORT_SYMBOL(autoremove_wake_function);
 217
 218 void __init fork_init(unsigned long mempages)
 219 {
 220 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 221 #ifndef ARCH_MIN_TASKALIGN
 222 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 223 #endif
 224         /* create a slab on which task_structs can be allocated */
 225         task_struct_cachep =
 226                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 227                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
 228 #endif
 229
 230         /*
 231          * The default maximum number of threads is set to a safe
 232          * value: the thread structures can take up at most half
 233          * of memory.
 234          */
 235         max_threads = mempages / (THREAD_SIZE/PAGE_SIZE) / 8;
 236         /*
 237          * we need to allow at least 20 threads to boot a system
 238          */
 239         if(max_threads < 20)
 240                 max_threads = 20;
 241
 242         init_task.rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 243         init_task.rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 244 }
 245
 246 static struct task_struct *dup_task_struct(struct task_struct *orig)
 247 {
 248         struct task_struct *tsk;
 249         struct thread_info *ti;
 250
 251         prepare_to_copy(orig);
 252
 253         tsk = alloc_task_struct();
 254         if (!tsk)
 255                 return NULL;
 256
 257         ti = alloc_thread_info(tsk);
 258         if (!ti) {
 259                 free_task_struct(tsk);
 260                 return NULL;
 261         }
 262
 263         *ti = *orig->thread_info;
 264         *tsk = *orig;
 265         tsk->thread_info = ti;
 266         ti->task = tsk;
 267
 268         /* One for us, one for whoever does the "release_task()" (usually parent) */
 269         atomic_set(&tsk->usage,2);
 270         return tsk;
 271 }
 272
 273 #ifdef CONFIG_MMU
 274 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
 275 {
 276         struct vm_area_struct * mpnt, *tmp, **pprev;
 277         struct rb_node **rb_link, *rb_parent;
 278         int retval;
 279         unsigned long charge;
 280         struct mempolicy *pol;
 281
 282         down_write(&oldmm->mmap_sem);
 283         flush_cache_mm(current->mm);
 284         mm->locked_vm = 0;
 285         mm->mmap = NULL;
 286         mm->mmap_cache = NULL;
 287         mm->free_area_cache = TASK_UNMAPPED_BASE;
 288         mm->map_count = 0;
 289         mm->rss = 0;
 290         cpus_clear(mm->cpu_vm_mask);
 291         mm->mm_rb = RB_ROOT;
 292         rb_link = &mm->mm_rb.rb_node;
 293         rb_parent = NULL;
 294         pprev = &mm->mmap;
 295
 296         /*
 297          * Add it to the mmlist after the parent.
 298          * Doing it this way means that we can order the list,
 299          * and fork() won't mess up the ordering significantly.
 300          * Add it first so that swapoff can see any swap entries.
 301          */
 302         spin_lock(&mmlist_lock);
 303         list_add(&mm->mmlist, &current->mm->mmlist);
 304         mmlist_nr++;
 305         spin_unlock(&mmlist_lock);
 306
 307         for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
 308                 struct file *file;
 309
 310                 if(mpnt->vm_flags & VM_DONTCOPY)
 311                         continue;
 312                 charge = 0;
 313                 if (mpnt->vm_flags & VM_ACCOUNT) {
 314                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 315                         if (security_vm_enough_memory(len))
 316                                 goto fail_nomem;
 317                         charge = len;
 318                 }
 319                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 320                 if (!tmp)
 321                         goto fail_nomem;
 322                 *tmp = *mpnt;
 323                 pol = mpol_copy(vma_policy(mpnt));
 324                 retval = PTR_ERR(pol);
 325                 if (IS_ERR(pol))
 326                         goto fail_nomem_policy;
 327                 vma_set_policy(tmp, pol);
 328                 tmp->vm_flags &= ~VM_LOCKED;
 329                 tmp->vm_mm = mm;
 330                 tmp->vm_next = NULL;
 331                 anon_vma_link(tmp);
 332                 vma_prio_tree_init(tmp);
 333                 file = tmp->vm_file;
 334                 if (file) {
 335                         struct inode *inode = file->f_dentry->d_inode;
 336                         get_file(file);
 337                         if (tmp->vm_flags & VM_DENYWRITE)
 338                                 atomic_dec(&inode->i_writecount);
 339
 340                         /* insert tmp into the share list, just after mpnt */
 341                         spin_lock(&file->f_mapping->i_mmap_lock);
 342                         flush_dcache_mmap_lock(file->f_mapping);
 343                         vma_prio_tree_add(tmp, mpnt);
 344                         flush_dcache_mmap_unlock(file->f_mapping);
 345                         spin_unlock(&file->f_mapping->i_mmap_lock);
 346                 }
 347
 348                 /*
 349                  * Link in the new vma and copy the page table entries:
 350                  * link in first so that swapoff can see swap entries,
 351                  * and try_to_unmap_one's find_vma find the new vma.
 352                  */
 353                 spin_lock(&mm->page_table_lock);
 354                 *pprev = tmp;
 355                 pprev = &tmp->vm_next;
 356
 357                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 358                 rb_link = &tmp->vm_rb.rb_right;
 359                 rb_parent = &tmp->vm_rb;
 360
 361                 mm->map_count++;
 362                 retval = copy_page_range(mm, current->mm, tmp);
 363                 spin_unlock(&mm->page_table_lock);
 364
 365                 if (tmp->vm_ops && tmp->vm_ops->open)
 366                         tmp->vm_ops->open(tmp);
 367
 368                 if (retval)
 369                         goto out;
 370         }
 371         retval = 0;
 372
 373 out:
 374         flush_tlb_mm(current->mm);
 375         up_write(&oldmm->mmap_sem);
 376         return retval;
 377 fail_nomem_policy:
 378         kmem_cache_free(vm_area_cachep, tmp);
 379 fail_nomem:
 380         retval = -ENOMEM;
 381         vm_unacct_memory(charge);
 382         goto out;
 383 }
 384
 385 static inline int mm_alloc_pgd(struct mm_struct * mm)
 386 {
 387         mm->pgd = pgd_alloc(mm);
 388         if (unlikely(!mm->pgd))
 389                 return -ENOMEM;
 390         return 0;
 391 }
 392
 393 static inline void mm_free_pgd(struct mm_struct * mm)
 394 {
 395         pgd_free(mm->pgd);
 396 }
 397 #else
 398 #define dup_mmap(mm, oldmm)     (0)
 399 #define mm_alloc_pgd(mm)        (0)
 400 #define mm_free_pgd(mm)
 401 #endif /* CONFIG_MMU */
 402
 403 spinlock_t mmlist_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
 404 int mmlist_nr;
 405
 406 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
 407 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 408
 409 #include <linux/init_task.h>
 410
 411 static struct mm_struct * mm_init(struct mm_struct * mm)
 412 {
 413         atomic_set(&mm->mm_users, 1);
 414         atomic_set(&mm->mm_count, 1);
 415         init_rwsem(&mm->mmap_sem);
 416         mm->core_waiters = 0;
 417         mm->page_table_lock = SPIN_LOCK_UNLOCKED;
 418         mm->ioctx_list_lock = RW_LOCK_UNLOCKED;
 419         mm->ioctx_list = NULL;
 420         mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
 421         mm->free_area_cache = TASK_UNMAPPED_BASE;
 422
 423         if (likely(!mm_alloc_pgd(mm))) {
 424                 mm->def_flags = 0;
 425                 set_vx_info(&mm->mm_vx_info, current->vx_info);
 426                 return mm;
 427         }
 428         free_mm(mm);
 429         return NULL;
 430 }
 431
 432 /*
 433  * Allocate and initialize an mm_struct.
 434  */
 435 struct mm_struct * mm_alloc(void)
 436 {
 437         struct mm_struct * mm;
 438
 439         mm = allocate_mm();
 440         if (mm) {
 441                 memset(mm, 0, sizeof(*mm));
 442                 mm = mm_init(mm);
 443         }
 444         return mm;
 445 }
 446
 447 /*
 448  * Called when the last reference to the mm
 449  * is dropped: either by a lazy thread or by
 450  * mmput. Free the page directory and the mm.
 451  */
 452 void fastcall __mmdrop(struct mm_struct *mm)
 453 {
 454         BUG_ON(mm == &init_mm);
 455         mm_free_pgd(mm);
 456         destroy_context(mm);
 457         clr_vx_info(&mm->mm_vx_info);
 458         free_mm(mm);
 459 }
 460
 461 /*
 462  * Decrement the use count and release all resources for an mm.
 463  */
 464 void mmput(struct mm_struct *mm)
 465 {
 466         if (atomic_dec_and_lock(&mm->mm_users, &mmlist_lock)) {
 467                 list_del(&mm->mmlist);
 468                 mmlist_nr--;
 469                 spin_unlock(&mmlist_lock);
 470                 exit_aio(mm);
 471                 exit_mmap(mm);
 472                 mmdrop(mm);
 473         }
 474 }
 475
 476 /*
 477  * Checks if the use count of an mm is non-zero and if so
 478  * returns a reference to it after bumping up the use count.
 479  * If the use count is zero, it means this mm is going away,
 480  * so return NULL.
 481  */
 482 struct mm_struct *mmgrab(struct mm_struct *mm)
 483 {
 484         spin_lock(&mmlist_lock);
 485         if (!atomic_read(&mm->mm_users))
 486                 mm = NULL;
 487         else
 488                 atomic_inc(&mm->mm_users);
 489         spin_unlock(&mmlist_lock);
 490         return mm;
 491 }
 492
 493 /* Please note the differences between mmput and mm_release.
 494  * mmput is called whenever we stop holding onto a mm_struct,
 495  * error success whatever.
 496  *
 497  * mm_release is called after a mm_struct has been removed
 498  * from the current process.
 499  *
 500  * This difference is important for error handling, when we
 501  * only half set up a mm_struct for a new process and need to restore
 502  * the old one.  Because we mmput the new mm_struct before
 503  * restoring the old one. . .
 504  * Eric Biederman 10 January 1998
 505  */
 506 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 507 {
 508         struct completion *vfork_done = tsk->vfork_done;
 509
 510         /* Get rid of any cached register state */
 511         deactivate_mm(tsk, mm);
 512
 513         /* notify parent sleeping on vfork() */
 514         if (vfork_done) {
 515                 tsk->vfork_done = NULL;
 516                 complete(vfork_done);
 517         }
 518         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
 519                 u32 __user * tidptr = tsk->clear_child_tid;
 520                 tsk->clear_child_tid = NULL;
 521
 522                 /*
 523                  * We don't check the error code - if userspace has
 524                  * not set up a proper pointer then tough luck.
 525                  */
 526                 put_user(0, tidptr);
 527                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 528         }
 529 }
 530
 531 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 532 {
 533         struct mm_struct * mm, *oldmm;
 534         int retval;
 535
 536         tsk->min_flt = tsk->maj_flt = 0;
 537         tsk->cmin_flt = tsk->cmaj_flt = 0;
 538         tsk->nvcsw = tsk->nivcsw = tsk->cnvcsw = tsk->cnivcsw = 0;
 539
 540         tsk->mm = NULL;
 541         tsk->active_mm = NULL;
 542
 543         /*
 544          * Are we cloning a kernel thread?
 545          *
 546          * We need to steal a active VM for that..
 547          */
 548         oldmm = current->mm;
 549         if (!oldmm)
 550                 return 0;
 551
 552         if (clone_flags & CLONE_VM) {
 553                 atomic_inc(&oldmm->mm_users);
 554                 mm = oldmm;
 555                 /*
 556                  * There are cases where the PTL is held to ensure no
 557                  * new threads start up in user mode using an mm, which
 558                  * allows optimizing out ipis; the tlb_gather_mmu code
 559                  * is an example.
 560                  */
 561                 spin_unlock_wait(&oldmm->page_table_lock);
 562                 goto good_mm;
 563         }
 564
 565         retval = -ENOMEM;
 566         mm = allocate_mm();
 567         if (!mm)
 568                 goto fail_nomem;
 569
 570         /* Copy the current MM stuff.. */
 571         memcpy(mm, oldmm, sizeof(*mm));
 572         mm->mm_vx_info = NULL;
 573         if (!mm_init(mm))
 574                 goto fail_nomem;
 575
 576         if (init_new_context(tsk,mm))
 577                 goto fail_nocontext;
 578
 579         retval = dup_mmap(mm, oldmm);
 580         if (retval)
 581                 goto free_pt;
 582
 583 good_mm:
 584         tsk->mm = mm;
 585         tsk->active_mm = mm;
 586         return 0;
 587
 588 free_pt:
 589         mmput(mm);
 590 fail_nomem:
 591         return retval;
 592
 593 fail_nocontext:
 594         /*
 595          * If init_new_context() failed, we cannot use mmput() to free the mm
 596          * because it calls destroy_context()
 597          */
 598         mm_free_pgd(mm);
 599         free_mm(mm);
 600         return retval;
 601 }
 602
 603 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 604 {
 605         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 606         /* We don't need to lock fs - think why ;-) */
 607         if (fs) {
 608                 atomic_set(&fs->count, 1);
 609                 fs->lock = RW_LOCK_UNLOCKED;
 610                 fs->umask = old->umask;
 611                 read_lock(&old->lock);
 612                 fs->rootmnt = mntget(old->rootmnt);
 613                 fs->root = dget(old->root);
 614                 fs->pwdmnt = mntget(old->pwdmnt);
 615                 fs->pwd = dget(old->pwd);
 616                 if (old->altroot) {
 617                         fs->altrootmnt = mntget(old->altrootmnt);
 618                         fs->altroot = dget(old->altroot);
 619                 } else {
 620                         fs->altrootmnt = NULL;
 621                         fs->altroot = NULL;
 622                 }
 623                 read_unlock(&old->lock);
 624         }
 625         return fs;
 626 }
 627
 628 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 629 {
 630         return __copy_fs_struct(old);
 631 }
 632
 633 EXPORT_SYMBOL_GPL(copy_fs_struct);
 634
 635 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 636 {
 637         if (clone_flags & CLONE_FS) {
 638                 atomic_inc(&current->fs->count);
 639                 return 0;
 640         }
 641         tsk->fs = __copy_fs_struct(current->fs);
 642         if (!tsk->fs)
 643                 return -ENOMEM;
 644         return 0;
 645 }
 646
 647 static int count_open_files(struct files_struct *files, int size)
 648 {
 649         int i;
 650
 651         /* Find the last open fd */
 652         for (i = size/(8*sizeof(long)); i > 0; ) {
 653                 if (files->open_fds->fds_bits[--i])
 654                         break;
 655         }
 656         i = (i+1) * 8 * sizeof(long);
 657         return i;
 658 }
 659
 660 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 661 {
 662         struct files_struct *oldf, *newf;
 663         struct file **old_fds, **new_fds;
 664         int open_files, nfds, size, i, error = 0;
 665
 666         /*
 667          * A background process may not have any files ...
 668          */
 669         oldf = current->files;
 670         if (!oldf)
 671                 goto out;
 672
 673         if (clone_flags & CLONE_FILES) {
 674                 atomic_inc(&oldf->count);
 675                 goto out;
 676         }
 677
 678         /*
 679          * Note: we may be using current for both targets (See exec.c)
 680          * This works because we cache current->files (old) as oldf. Don't
 681          * break this.
 682          */
 683         tsk->files = NULL;
 684         error = -ENOMEM;
 685         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
 686         if (!newf)
 687                 goto out;
 688
 689         atomic_set(&newf->count, 1);
 690
 691         newf->file_lock     = SPIN_LOCK_UNLOCKED;
 692         newf->next_fd       = 0;
 693         newf->max_fds       = NR_OPEN_DEFAULT;
 694         newf->max_fdset     = __FD_SETSIZE;
 695         newf->close_on_exec = &newf->close_on_exec_init;
 696         newf->open_fds      = &newf->open_fds_init;
 697         newf->fd            = &newf->fd_array[0];
 698
 699         /* We don't yet have the oldf readlock, but even if the old
 700            fdset gets grown now, we'll only copy up to "size" fds */
 701         size = oldf->max_fdset;
 702         if (size > __FD_SETSIZE) {
 703                 newf->max_fdset = 0;
 704                 spin_lock(&newf->file_lock);
 705                 error = expand_fdset(newf, size-1);
 706                 spin_unlock(&newf->file_lock);
 707                 if (error)
 708                         goto out_release;
 709         }
 710         spin_lock(&oldf->file_lock);
 711
 712         open_files = count_open_files(oldf, size);
 713
 714         /*
 715          * Check whether we need to allocate a larger fd array.
 716          * Note: we're not a clone task, so the open count won't
 717          * change.
 718          */
 719         nfds = NR_OPEN_DEFAULT;
 720         if (open_files > nfds) {
 721                 spin_unlock(&oldf->file_lock);
 722                 newf->max_fds = 0;
 723                 spin_lock(&newf->file_lock);
 724                 error = expand_fd_array(newf, open_files-1);
 725                 spin_unlock(&newf->file_lock);
 726                 if (error)
 727                         goto out_release;
 728                 nfds = newf->max_fds;
 729                 spin_lock(&oldf->file_lock);
 730         }
 731
 732         old_fds = oldf->fd;
 733         new_fds = newf->fd;
 734
 735         memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
 736         memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
 737
 738         for (i = open_files; i != 0; i--) {
 739                 struct file *f = *old_fds++;
 740                 if (f)
 741                         get_file(f);
 742                 *new_fds++ = f;
 743         }
 744         spin_unlock(&oldf->file_lock);
 745
 746         /* compute the remainder to be cleared */
 747         size = (newf->max_fds - open_files) * sizeof(struct file *);
 748
 749         /* This is long word aligned thus could use a optimized version */
 750         memset(new_fds, 0, size);
 751
 752         if (newf->max_fdset > open_files) {
 753                 int left = (newf->max_fdset-open_files)/8;
 754                 int start = open_files / (8 * sizeof(unsigned long));
 755
 756                 memset(&newf->open_fds->fds_bits[start], 0, left);
 757                 memset(&newf->close_on_exec->fds_bits[start], 0, left);
 758         }
 759
 760         tsk->files = newf;
 761         error = 0;
 762 out:
 763         return error;
 764
 765 out_release:
 766         free_fdset (newf->close_on_exec, newf->max_fdset);
 767         free_fdset (newf->open_fds, newf->max_fdset);
 768         kmem_cache_free(files_cachep, newf);
 769         goto out;
 770 }
 771
 772 /*
 773  *      Helper to unshare the files of the current task.
 774  *      We don't want to expose copy_files internals to
 775  *      the exec layer of the kernel.
 776  */
 777
 778 int unshare_files(void)
 779 {
 780         struct files_struct *files  = current->files;
 781         int rc;
 782
 783         if(!files)
 784                 BUG();
 785
 786         /* This can race but the race causes us to copy when we don't
 787            need to and drop the copy */
 788         if(atomic_read(&files->count) == 1)
 789         {
 790                 atomic_inc(&files->count);
 791                 return 0;
 792         }
 793         rc = copy_files(0, current);
 794         if(rc)
 795                 current->files = files;
 796         return rc;
 797 }
 798
 799 EXPORT_SYMBOL(unshare_files);
 800
 801 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 802 {
 803         struct sighand_struct *sig;
 804
 805         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 806                 atomic_inc(&current->sighand->count);
 807                 return 0;
 808         }
 809         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 810         tsk->sighand = sig;
 811         if (!sig)
 812                 return -ENOMEM;
 813         spin_lock_init(&sig->siglock);
 814         atomic_set(&sig->count, 1);
 815         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 816         return 0;
 817 }
 818
 819 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 820 {
 821         struct signal_struct *sig;
 822
 823         if (clone_flags & CLONE_THREAD) {
 824                 atomic_inc(&current->signal->count);
 825                 return 0;
 826         }
 827         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 828         tsk->signal = sig;
 829         if (!sig)
 830                 return -ENOMEM;
 831         atomic_set(&sig->count, 1);
 832         sig->group_exit = 0;
 833         sig->group_exit_code = 0;
 834         sig->group_exit_task = NULL;
 835         sig->group_stop_count = 0;
 836         sig->curr_target = NULL;
 837         init_sigpending(&sig->shared_pending);
 838         INIT_LIST_HEAD(&sig->posix_timers);
 839
 840         sig->tty = current->signal->tty;
 841         sig->pgrp = process_group(current);
 842         sig->session = current->signal->session;
 843         sig->leader = 0;        /* session leadership doesn't inherit */
 844         sig->tty_old_pgrp = 0;
 845
 846         return 0;
 847 }
 848
 849 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 850 {
 851         unsigned long new_flags = p->flags;
 852
 853         new_flags &= ~PF_SUPERPRIV;
 854         new_flags |= PF_FORKNOEXEC;
 855         if (!(clone_flags & CLONE_PTRACE))
 856                 p->ptrace = 0;
 857         p->flags = new_flags;
 858 }
 859
 860 asmlinkage long sys_set_tid_address(int __user *tidptr)
 861 {
 862         current->clear_child_tid = tidptr;
 863
 864         return current->pid;
 865 }
 866
 867 /*
 868  * This creates a new process as a copy of the old one,
 869  * but does not actually start it yet.
 870  *
 871  * It copies the registers, and all the appropriate
 872  * parts of the process environment (as per the clone
 873  * flags). The actual kick-off is left to the caller.
 874  */
 875 struct task_struct *copy_process(unsigned long clone_flags,
 876                                  unsigned long stack_start,
 877                                  struct pt_regs *regs,
 878                                  unsigned long stack_size,
 879                                  int __user *parent_tidptr,
 880                                  int __user *child_tidptr)
 881 {
 882         int retval;
 883         struct task_struct *p = NULL;
 884         struct vx_info *vxi;
 885
 886         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 887                 return ERR_PTR(-EINVAL);
 888
 889         /*
 890          * Thread groups must share signals as well, and detached threads
 891          * can only be started up within the thread group.
 892          */
 893         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 894                 return ERR_PTR(-EINVAL);
 895
 896         /*
 897          * Shared signal handlers imply shared VM. By way of the above,
 898          * thread groups also imply shared VM. Blocking this case allows
 899          * for various simplifications in other code.
 900          */
 901         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 902                 return ERR_PTR(-EINVAL);
 903
 904         retval = security_task_create(clone_flags);
 905         if (retval)
 906                 goto fork_out;
 907
 908         retval = -ENOMEM;
 909
 910         p = dup_task_struct(current);
 911         if (!p)
 912                 goto fork_out;
 913
 914         p->vx_info = NULL;
 915         set_vx_info(&p->vx_info, current->vx_info);
 916         p->nx_info = NULL;
 917         set_nx_info(&p->nx_info, current->nx_info);
 918
 919         /* check vserver memory */
 920         if (p->mm && !(clone_flags & CLONE_VM)) {
 921                 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
 922                         vx_pages_add(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
 923                 else
 924                         goto bad_fork_free;
 925         }
 926         if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
 927                 if (!vx_rsspages_avail(p->mm, p->mm->rss))
 928                         goto bad_fork_free;
 929         }
 930
 931         retval = -EAGAIN;
 932         if (!vx_nproc_avail(1))
 933                 goto bad_fork_free;
 934
 935         if (atomic_read(&p->user->processes) >=
 936                         p->rlim[RLIMIT_NPROC].rlim_cur) {
 937                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 938                                 p->user != &root_user)
 939                         goto bad_fork_free;
 940         }
 941
 942         atomic_inc(&p->user->__count);
 943         atomic_inc(&p->user->processes);
 944         get_group_info(p->group_info);
 945
 946         /*
 947          * If multiple threads are within copy_process(), then this check
 948          * triggers too late. This doesn't hurt, the check is only there
 949          * to stop root fork bombs.
 950          */
 951         if (nr_threads >= max_threads)
 952                 goto bad_fork_cleanup_count;
 953
 954         if (!try_module_get(p->thread_info->exec_domain->module))
 955                 goto bad_fork_cleanup_count;
 956
 957         if (p->binfmt && !try_module_get(p->binfmt->module))
 958                 goto bad_fork_cleanup_put_domain;
 959
 960         p->did_exec = 0;
 961         copy_flags(clone_flags, p);
 962         if (clone_flags & CLONE_IDLETASK)
 963                 p->pid = 0;
 964         else {
 965                 p->pid = alloc_pidmap();
 966                 if (p->pid == -1)
 967                         goto bad_fork_cleanup;
 968         }
 969         retval = -EFAULT;
 970         if (clone_flags & CLONE_PARENT_SETTID)
 971                 if (put_user(p->pid, parent_tidptr))
 972                         goto bad_fork_cleanup;
 973
 974         p->proc_dentry = NULL;
 975
 976         INIT_LIST_HEAD(&p->children);
 977         INIT_LIST_HEAD(&p->sibling);
 978         init_waitqueue_head(&p->wait_chldexit);
 979         p->vfork_done = NULL;
 980         spin_lock_init(&p->alloc_lock);
 981         spin_lock_init(&p->proc_lock);
 982
 983         clear_tsk_thread_flag(p, TIF_SIGPENDING);
 984         init_sigpending(&p->pending);
 985
 986         p->it_real_value = p->it_virt_value = p->it_prof_value = 0;
 987         p->it_real_incr = p->it_virt_incr = p->it_prof_incr = 0;
 988         init_timer(&p->real_timer);
 989         p->real_timer.data = (unsigned long) p;
 990
 991         p->utime = p->stime = 0;
 992         p->cutime = p->cstime = 0;
 993         p->lock_depth = -1;             /* -1 = no lock */
 994         p->start_time = get_jiffies_64();
 995         p->security = NULL;
 996         p->io_context = NULL;
 997         p->audit_context = NULL;
 998 #ifdef CONFIG_NUMA
 999         p->mempolicy = mpol_copy(p->mempolicy);
1000         if (IS_ERR(p->mempolicy)) {
1001                 retval = PTR_ERR(p->mempolicy);
1002                 p->mempolicy = NULL;
1003                 goto bad_fork_cleanup;
1004         }
1005 #endif
1006
1007         retval = -ENOMEM;
1008         if ((retval = security_task_alloc(p)))
1009                 goto bad_fork_cleanup_policy;
1010         if ((retval = audit_alloc(p)))
1011                 goto bad_fork_cleanup_security;
1012         /* copy all the process information */
1013         if ((retval = copy_semundo(clone_flags, p)))
1014                 goto bad_fork_cleanup_audit;
1015         if ((retval = copy_files(clone_flags, p)))
1016                 goto bad_fork_cleanup_semundo;
1017         if ((retval = copy_fs(clone_flags, p)))
1018                 goto bad_fork_cleanup_files;
1019         if ((retval = copy_sighand(clone_flags, p)))
1020                 goto bad_fork_cleanup_fs;
1021         if ((retval = copy_signal(clone_flags, p)))
1022                 goto bad_fork_cleanup_sighand;
1023         if ((retval = copy_mm(clone_flags, p)))
1024                 goto bad_fork_cleanup_signal;
1025         if ((retval = copy_namespace(clone_flags, p)))
1026                 goto bad_fork_cleanup_mm;
1027         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1028         if (retval)
1029                 goto bad_fork_cleanup_namespace;
1030
1031         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1032         /*
1033          * Clear TID on mm_release()?
1034          */
1035         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1036
1037         /*
1038          * Syscall tracing should be turned off in the child regardless
1039          * of CLONE_PTRACE.
1040          */
1041         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1042
1043         /* Our parent execution domain becomes current domain
1044            These must match for thread signalling to apply */
1045
1046         p->parent_exec_id = p->self_exec_id;
1047
1048         /* ok, now we should be set up.. */
1049         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1050         p->pdeath_signal = 0;
1051
1052         /* Perform scheduler related setup */
1053         sched_fork(p);
1054
1055         /*
1056          * Ok, make it visible to the rest of the system.
1057          * We dont wake it up yet.
1058          */
1059         p->tgid = p->pid;
1060         p->group_leader = p;
1061         INIT_LIST_HEAD(&p->ptrace_children);
1062         INIT_LIST_HEAD(&p->ptrace_list);
1063
1064         /* Need tasklist lock for parent etc handling! */
1065         write_lock_irq(&tasklist_lock);
1066         /*
1067          * Check for pending SIGKILL! The new thread should not be allowed
1068          * to slip out of an OOM kill. (or normal SIGKILL.)
1069          */
1070         if (sigismember(&current->pending.signal, SIGKILL)) {
1071                 write_unlock_irq(&tasklist_lock);
1072                 retval = -EINTR;
1073                 goto bad_fork_cleanup_namespace;
1074         }
1075
1076         /* CLONE_PARENT re-uses the old parent */
1077         if (clone_flags & CLONE_PARENT)
1078                 p->real_parent = current->real_parent;
1079         else
1080                 p->real_parent = current;
1081         p->parent = p->real_parent;
1082
1083         if (clone_flags & CLONE_THREAD) {
1084                 spin_lock(&current->sighand->siglock);
1085                 /*
1086                  * Important: if an exit-all has been started then
1087                  * do not create this new thread - the whole thread
1088                  * group is supposed to exit anyway.
1089                  */
1090                 if (current->signal->group_exit) {
1091                         spin_unlock(&current->sighand->siglock);
1092                         write_unlock_irq(&tasklist_lock);
1093                         retval = -EAGAIN;
1094                         goto bad_fork_cleanup_namespace;
1095                 }
1096                 p->tgid = current->tgid;
1097                 p->group_leader = current->group_leader;
1098
1099                 if (current->signal->group_stop_count > 0) {
1100                         /*
1101                          * There is an all-stop in progress for the group.
1102                          * We ourselves will stop as soon as we check signals.
1103                          * Make the new thread part of that group stop too.
1104                          */
1105                         current->signal->group_stop_count++;
1106                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1107                 }
1108
1109                 spin_unlock(&current->sighand->siglock);
1110         }
1111
1112         SET_LINKS(p);
1113         if (p->ptrace & PT_PTRACED)
1114                 __ptrace_link(p, current->parent);
1115
1116         attach_pid(p, PIDTYPE_PID, p->pid);
1117         if (thread_group_leader(p)) {
1118                 attach_pid(p, PIDTYPE_TGID, p->tgid);
1119                 attach_pid(p, PIDTYPE_PGID, process_group(p));
1120                 attach_pid(p, PIDTYPE_SID, p->signal->session);
1121                 if (p->pid)
1122                         __get_cpu_var(process_counts)++;
1123         } else
1124                 link_pid(p, p->pids + PIDTYPE_TGID, &p->group_leader->pids[PIDTYPE_TGID].pid);
1125
1126         nr_threads++;
1127         vxi = current->vx_info;
1128         if (vxi) {
1129                 atomic_inc(&vxi->cacct.nr_threads);
1130                 // atomic_inc(&vxi->limit.rcur[RLIMIT_NPROC]);
1131         }
1132         vx_nproc_inc();
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 }