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