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 #include <linux/ckrm.h>
  42 #include <linux/ckrm_tsk.h>
  43
  44 #include <asm/pgtable.h>
  45 #include <asm/pgalloc.h>
  46 #include <asm/uaccess.h>
  47 #include <asm/mmu_context.h>
  48 #include <asm/cacheflush.h>
  49 #include <asm/tlbflush.h>
  50
  51 /* The idle threads do not count..
  52  * Protected by write_lock_irq(&tasklist_lock)
  53  */
  54 int nr_threads;
  55
  56 int max_threads;
  57 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  58
  59 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  60
  61 rwlock_t tasklist_lock __cacheline_aligned = RW_LOCK_UNLOCKED;  /* outer */
  62
  63 EXPORT_SYMBOL(tasklist_lock);
  64
  65 int nr_processes(void)
  66 {
  67         int cpu;
  68         int total = 0;
  69
  70         for_each_online_cpu(cpu)
  71                 total += per_cpu(process_counts, cpu);
  72
  73         return total;
  74 }
  75
  76 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  77 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  78 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  79 static kmem_cache_t *task_struct_cachep;
  80 #endif
  81
  82 static void free_task(struct task_struct *tsk)
  83 {
  84         free_thread_info(tsk->thread_info);
  85         vxdprintk("freeing up task %p\n", tsk);
  86         clr_vx_info(&tsk->vx_info);
  87         clr_nx_info(&tsk->nx_info);
  88         free_task_struct(tsk);
  89 }
  90
  91 void __put_task_struct(struct task_struct *tsk)
  92 {
  93         WARN_ON(!(tsk->state & (TASK_DEAD | TASK_ZOMBIE)));
  94         WARN_ON(atomic_read(&tsk->usage));
  95         WARN_ON(tsk == current);
  96
  97         if (unlikely(tsk->audit_context))
  98                 audit_free(tsk);
  99         security_task_free(tsk);
 100         free_uid(tsk->user);
 101         put_group_info(tsk->group_info);
 102         free_task(tsk);
 103 }
 104
 105 void fastcall add_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
 106 {
 107         unsigned long flags;
 108
 109         wait->flags &= ~WQ_FLAG_EXCLUSIVE;
 110         spin_lock_irqsave(&q->lock, flags);
 111         __add_wait_queue(q, wait);
 112         spin_unlock_irqrestore(&q->lock, flags);
 113 }
 114
 115 EXPORT_SYMBOL(add_wait_queue);
 116
 117 void fastcall add_wait_queue_exclusive(wait_queue_head_t *q, wait_queue_t * wait)
 118 {
 119         unsigned long flags;
 120
 121         wait->flags |= WQ_FLAG_EXCLUSIVE;
 122         spin_lock_irqsave(&q->lock, flags);
 123         __add_wait_queue_tail(q, wait);
 124         spin_unlock_irqrestore(&q->lock, flags);
 125 }
 126
 127 EXPORT_SYMBOL(add_wait_queue_exclusive);
 128
 129 void fastcall remove_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
 130 {
 131         unsigned long flags;
 132
 133         spin_lock_irqsave(&q->lock, flags);
 134         __remove_wait_queue(q, wait);
 135         spin_unlock_irqrestore(&q->lock, flags);
 136 }
 137
 138 EXPORT_SYMBOL(remove_wait_queue);
 139
 140
 141 /*
 142  * Note: we use "set_current_state()" _after_ the wait-queue add,
 143  * because we need a memory barrier there on SMP, so that any
 144  * wake-function that tests for the wait-queue being active
 145  * will be guaranteed to see waitqueue addition _or_ subsequent
 146  * tests in this thread will see the wakeup having taken place.
 147  *
 148  * The spin_unlock() itself is semi-permeable and only protects
 149  * one way (it only protects stuff inside the critical region and
 150  * stops them from bleeding out - it would still allow subsequent
 151  * loads to move into the the critical region).
 152  */
 153 void fastcall prepare_to_wait(wait_queue_head_t *q, wait_queue_t *wait, int state)
 154 {
 155         unsigned long flags;
 156
 157         wait->flags &= ~WQ_FLAG_EXCLUSIVE;
 158         spin_lock_irqsave(&q->lock, flags);
 159         if (list_empty(&wait->task_list))
 160                 __add_wait_queue(q, wait);
 161         set_current_state(state);
 162         spin_unlock_irqrestore(&q->lock, flags);
 163 }
 164
 165 EXPORT_SYMBOL(prepare_to_wait);
 166
 167 void fastcall
 168 prepare_to_wait_exclusive(wait_queue_head_t *q, wait_queue_t *wait, int state)
 169 {
 170         unsigned long flags;
 171
 172         wait->flags |= WQ_FLAG_EXCLUSIVE;
 173         spin_lock_irqsave(&q->lock, flags);
 174         if (list_empty(&wait->task_list))
 175                 __add_wait_queue_tail(q, wait);
 176         set_current_state(state);
 177         spin_unlock_irqrestore(&q->lock, flags);
 178 }
 179
 180 EXPORT_SYMBOL(prepare_to_wait_exclusive);
 181
 182 void fastcall finish_wait(wait_queue_head_t *q, wait_queue_t *wait)
 183 {
 184         unsigned long flags;
 185
 186         __set_current_state(TASK_RUNNING);
 187         /*
 188          * We can check for list emptiness outside the lock
 189          * IFF:
 190          *  - we use the "careful" check that verifies both
 191          *    the next and prev pointers, so that there cannot
 192          *    be any half-pending updates in progress on other
 193          *    CPU's that we haven't seen yet (and that might
 194          *    still change the stack area.
 195          * and
 196          *  - all other users take the lock (ie we can only
 197          *    have _one_ other CPU that looks at or modifies
 198          *    the list).
 199          */
 200         if (!list_empty_careful(&wait->task_list)) {
 201                 spin_lock_irqsave(&q->lock, flags);
 202                 list_del_init(&wait->task_list);
 203                 spin_unlock_irqrestore(&q->lock, flags);
 204         }
 205 }
 206
 207 EXPORT_SYMBOL(finish_wait);
 208
 209 int autoremove_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
 210 {
 211         int ret = default_wake_function(wait, mode, sync, key);
 212
 213         if (ret)
 214                 list_del_init(&wait->task_list);
 215         return ret;
 216 }
 217
 218 EXPORT_SYMBOL(autoremove_wake_function);
 219
 220 void __init fork_init(unsigned long mempages)
 221 {
 222 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 223 #ifndef ARCH_MIN_TASKALIGN
 224 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 225 #endif
 226         /* create a slab on which task_structs can be allocated */
 227         task_struct_cachep =
 228                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 229                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
 230 #endif
 231
 232         /*
 233          * The default maximum number of threads is set to a safe
 234          * value: the thread structures can take up at most half
 235          * of memory.
 236          */
 237         max_threads = mempages / (THREAD_SIZE/PAGE_SIZE) / 8;
 238         /*
 239          * we need to allow at least 20 threads to boot a system
 240          */
 241         if(max_threads < 20)
 242                 max_threads = 20;
 243
 244         init_task.rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 245         init_task.rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 246 }
 247
 248 static struct task_struct *dup_task_struct(struct task_struct *orig)
 249 {
 250         struct task_struct *tsk;
 251         struct thread_info *ti;
 252
 253         prepare_to_copy(orig);
 254
 255         tsk = alloc_task_struct();
 256         if (!tsk)
 257                 return NULL;
 258
 259         ti = alloc_thread_info(tsk);
 260         if (!ti) {
 261                 free_task_struct(tsk);
 262                 return NULL;
 263         }
 264
 265         *ti = *orig->thread_info;
 266         *tsk = *orig;
 267         tsk->thread_info = ti;
 268         ti->task = tsk;
 269
 270         ckrm_cb_newtask(tsk);
 271         /* One for us, one for whoever does the "release_task()" (usually parent) */
 272         atomic_set(&tsk->usage,2);
 273         return tsk;
 274 }
 275
 276 #ifdef CONFIG_MMU
 277 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
 278 {
 279         struct vm_area_struct * mpnt, *tmp, **pprev;
 280         struct rb_node **rb_link, *rb_parent;
 281         int retval;
 282         unsigned long charge;
 283         struct mempolicy *pol;
 284
 285         down_write(&oldmm->mmap_sem);
 286         flush_cache_mm(current->mm);
 287         mm->locked_vm = 0;
 288         mm->mmap = NULL;
 289         mm->mmap_cache = NULL;
 290         mm->free_area_cache = TASK_UNMAPPED_BASE;
 291         mm->map_count = 0;
 292         mm->rss = 0;
 293         cpus_clear(mm->cpu_vm_mask);
 294         mm->mm_rb = RB_ROOT;
 295         rb_link = &mm->mm_rb.rb_node;
 296         rb_parent = NULL;
 297         pprev = &mm->mmap;
 298
 299         /*
 300          * Add it to the mmlist after the parent.
 301          * Doing it this way means that we can order the list,
 302          * and fork() won't mess up the ordering significantly.
 303          * Add it first so that swapoff can see any swap entries.
 304          */
 305         spin_lock(&mmlist_lock);
 306         list_add(&mm->mmlist, &current->mm->mmlist);
 307         mmlist_nr++;
 308         spin_unlock(&mmlist_lock);
 309
 310         for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
 311                 struct file *file;
 312
 313                 if(mpnt->vm_flags & VM_DONTCOPY)
 314                         continue;
 315                 charge = 0;
 316                 if (mpnt->vm_flags & VM_ACCOUNT) {
 317                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 318                         if (security_vm_enough_memory(len))
 319                                 goto fail_nomem;
 320                         charge = len;
 321                 }
 322                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 323                 if (!tmp)
 324                         goto fail_nomem;
 325                 *tmp = *mpnt;
 326                 pol = mpol_copy(vma_policy(mpnt));
 327                 retval = PTR_ERR(pol);
 328                 if (IS_ERR(pol))
 329                         goto fail_nomem_policy;
 330                 vma_set_policy(tmp, pol);
 331                 tmp->vm_flags &= ~VM_LOCKED;
 332                 tmp->vm_mm = mm;
 333                 tmp->vm_next = NULL;
 334                 anon_vma_link(tmp);
 335                 vma_prio_tree_init(tmp);
 336                 file = tmp->vm_file;
 337                 if (file) {
 338                         struct inode *inode = file->f_dentry->d_inode;
 339                         get_file(file);
 340                         if (tmp->vm_flags & VM_DENYWRITE)
 341                                 atomic_dec(&inode->i_writecount);
 342
 343                         /* insert tmp into the share list, just after mpnt */
 344                         spin_lock(&file->f_mapping->i_mmap_lock);
 345                         flush_dcache_mmap_lock(file->f_mapping);
 346                         vma_prio_tree_add(tmp, mpnt);
 347                         flush_dcache_mmap_unlock(file->f_mapping);
 348                         spin_unlock(&file->f_mapping->i_mmap_lock);
 349                 }
 350
 351                 /*
 352                  * Link in the new vma and copy the page table entries:
 353                  * link in first so that swapoff can see swap entries,
 354                  * and try_to_unmap_one's find_vma find the new vma.
 355                  */
 356                 spin_lock(&mm->page_table_lock);
 357                 *pprev = tmp;
 358                 pprev = &tmp->vm_next;
 359
 360                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 361                 rb_link = &tmp->vm_rb.rb_right;
 362                 rb_parent = &tmp->vm_rb;
 363
 364                 mm->map_count++;
 365                 retval = copy_page_range(mm, current->mm, tmp);
 366                 spin_unlock(&mm->page_table_lock);
 367
 368                 if (tmp->vm_ops && tmp->vm_ops->open)
 369                         tmp->vm_ops->open(tmp);
 370
 371                 if (retval)
 372                         goto out;
 373         }
 374         retval = 0;
 375
 376 out:
 377         flush_tlb_mm(current->mm);
 378         up_write(&oldmm->mmap_sem);
 379         return retval;
 380 fail_nomem_policy:
 381         kmem_cache_free(vm_area_cachep, tmp);
 382 fail_nomem:
 383         retval = -ENOMEM;
 384         vm_unacct_memory(charge);
 385         goto out;
 386 }
 387
 388 static inline int mm_alloc_pgd(struct mm_struct * mm)
 389 {
 390         mm->pgd = pgd_alloc(mm);
 391         if (unlikely(!mm->pgd))
 392                 return -ENOMEM;
 393         return 0;
 394 }
 395
 396 static inline void mm_free_pgd(struct mm_struct * mm)
 397 {
 398         pgd_free(mm->pgd);
 399 }
 400 #else
 401 #define dup_mmap(mm, oldmm)     (0)
 402 #define mm_alloc_pgd(mm)        (0)
 403 #define mm_free_pgd(mm)
 404 #endif /* CONFIG_MMU */
 405
 406 spinlock_t mmlist_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
 407 int mmlist_nr;
 408
 409 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
 410 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 411
 412 #include <linux/init_task.h>
 413
 414 static struct mm_struct * mm_init(struct mm_struct * mm)
 415 {
 416         atomic_set(&mm->mm_users, 1);
 417         atomic_set(&mm->mm_count, 1);
 418         init_rwsem(&mm->mmap_sem);
 419         mm->core_waiters = 0;
 420         mm->page_table_lock = SPIN_LOCK_UNLOCKED;
 421         mm->ioctx_list_lock = RW_LOCK_UNLOCKED;
 422         mm->ioctx_list = NULL;
 423         mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
 424         mm->free_area_cache = TASK_UNMAPPED_BASE;
 425
 426         if (likely(!mm_alloc_pgd(mm))) {
 427                 mm->def_flags = 0;
 428 #ifdef __HAVE_ARCH_MMAP_TOP
 429                 mm->mmap_top = mmap_top();
 430 #endif
 431                 set_vx_info(&mm->mm_vx_info, current->vx_info);
 432                 return mm;
 433         }
 434         free_mm(mm);
 435         return NULL;
 436 }
 437
 438 /*
 439  * Allocate and initialize an mm_struct.
 440  */
 441 struct mm_struct * mm_alloc(void)
 442 {
 443         struct mm_struct * mm;
 444
 445         mm = allocate_mm();
 446         if (mm) {
 447                 memset(mm, 0, sizeof(*mm));
 448                 mm = mm_init(mm);
 449         }
 450         return mm;
 451 }
 452
 453 /*
 454  * Called when the last reference to the mm
 455  * is dropped: either by a lazy thread or by
 456  * mmput. Free the page directory and the mm.
 457  */
 458 void fastcall __mmdrop(struct mm_struct *mm)
 459 {
 460         BUG_ON(mm == &init_mm);
 461         mm_free_pgd(mm);
 462         destroy_context(mm);
 463         clr_vx_info(&mm->mm_vx_info);
 464         free_mm(mm);
 465 }
 466
 467 /*
 468  * Decrement the use count and release all resources for an mm.
 469  */
 470 void mmput(struct mm_struct *mm)
 471 {
 472         if (atomic_dec_and_lock(&mm->mm_users, &mmlist_lock)) {
 473                 list_del(&mm->mmlist);
 474                 mmlist_nr--;
 475                 spin_unlock(&mmlist_lock);
 476                 exit_aio(mm);
 477                 exit_mmap(mm);
 478                 mmdrop(mm);
 479         }
 480 }
 481
 482 /*
 483  * Checks if the use count of an mm is non-zero and if so
 484  * returns a reference to it after bumping up the use count.
 485  * If the use count is zero, it means this mm is going away,
 486  * so return NULL.
 487  */
 488 struct mm_struct *mmgrab(struct mm_struct *mm)
 489 {
 490         spin_lock(&mmlist_lock);
 491         if (!atomic_read(&mm->mm_users))
 492                 mm = NULL;
 493         else
 494                 atomic_inc(&mm->mm_users);
 495         spin_unlock(&mmlist_lock);
 496         return mm;
 497 }
 498
 499 /* Please note the differences between mmput and mm_release.
 500  * mmput is called whenever we stop holding onto a mm_struct,
 501  * error success whatever.
 502  *
 503  * mm_release is called after a mm_struct has been removed
 504  * from the current process.
 505  *
 506  * This difference is important for error handling, when we
 507  * only half set up a mm_struct for a new process and need to restore
 508  * the old one.  Because we mmput the new mm_struct before
 509  * restoring the old one. . .
 510  * Eric Biederman 10 January 1998
 511  */
 512 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 513 {
 514         struct completion *vfork_done = tsk->vfork_done;
 515
 516         /* Get rid of any cached register state */
 517         deactivate_mm(tsk, mm);
 518
 519         /* notify parent sleeping on vfork() */
 520         if (vfork_done) {
 521                 tsk->vfork_done = NULL;
 522                 complete(vfork_done);
 523         }
 524         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
 525                 u32 __user * tidptr = tsk->clear_child_tid;
 526                 tsk->clear_child_tid = NULL;
 527
 528                 /*
 529                  * We don't check the error code - if userspace has
 530                  * not set up a proper pointer then tough luck.
 531                  */
 532                 put_user(0, tidptr);
 533                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 534         }
 535 }
 536
 537 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 538 {
 539         struct mm_struct * mm, *oldmm;
 540         int retval;
 541
 542         tsk->min_flt = tsk->maj_flt = 0;
 543         tsk->cmin_flt = tsk->cmaj_flt = 0;
 544         tsk->nvcsw = tsk->nivcsw = tsk->cnvcsw = tsk->cnivcsw = 0;
 545
 546         tsk->mm = NULL;
 547         tsk->active_mm = NULL;
 548
 549         /*
 550          * Are we cloning a kernel thread?
 551          *
 552          * We need to steal a active VM for that..
 553          */
 554         oldmm = current->mm;
 555         if (!oldmm)
 556                 return 0;
 557
 558         if (clone_flags & CLONE_VM) {
 559                 atomic_inc(&oldmm->mm_users);
 560                 mm = oldmm;
 561                 /*
 562                  * There are cases where the PTL is held to ensure no
 563                  * new threads start up in user mode using an mm, which
 564                  * allows optimizing out ipis; the tlb_gather_mmu code
 565                  * is an example.
 566                  */
 567                 spin_unlock_wait(&oldmm->page_table_lock);
 568                 goto good_mm;
 569         }
 570
 571         retval = -ENOMEM;
 572         mm = allocate_mm();
 573         if (!mm)
 574                 goto fail_nomem;
 575
 576         /* Copy the current MM stuff.. */
 577         memcpy(mm, oldmm, sizeof(*mm));
 578         mm->mm_vx_info = NULL;
 579         if (!mm_init(mm))
 580                 goto fail_nomem;
 581
 582         if (init_new_context(tsk,mm))
 583                 goto fail_nocontext;
 584
 585         retval = dup_mmap(mm, oldmm);
 586         if (retval)
 587                 goto free_pt;
 588
 589 good_mm:
 590         tsk->mm = mm;
 591         tsk->active_mm = mm;
 592         return 0;
 593
 594 free_pt:
 595         mmput(mm);
 596 fail_nomem:
 597         return retval;
 598
 599 fail_nocontext:
 600         /*
 601          * If init_new_context() failed, we cannot use mmput() to free the mm
 602          * because it calls destroy_context()
 603          */
 604         mm_free_pgd(mm);
 605         free_mm(mm);
 606         return retval;
 607 }
 608
 609 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 610 {
 611         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 612         /* We don't need to lock fs - think why ;-) */
 613         if (fs) {
 614                 atomic_set(&fs->count, 1);
 615                 fs->lock = RW_LOCK_UNLOCKED;
 616                 fs->umask = old->umask;
 617                 read_lock(&old->lock);
 618                 fs->rootmnt = mntget(old->rootmnt);
 619                 fs->root = dget(old->root);
 620                 fs->pwdmnt = mntget(old->pwdmnt);
 621                 fs->pwd = dget(old->pwd);
 622                 if (old->altroot) {
 623                         fs->altrootmnt = mntget(old->altrootmnt);
 624                         fs->altroot = dget(old->altroot);
 625                 } else {
 626                         fs->altrootmnt = NULL;
 627                         fs->altroot = NULL;
 628                 }
 629                 read_unlock(&old->lock);
 630         }
 631         return fs;
 632 }
 633
 634 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 635 {
 636         return __copy_fs_struct(old);
 637 }
 638
 639 EXPORT_SYMBOL_GPL(copy_fs_struct);
 640
 641 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 642 {
 643         if (clone_flags & CLONE_FS) {
 644                 atomic_inc(&current->fs->count);
 645                 return 0;
 646         }
 647         tsk->fs = __copy_fs_struct(current->fs);
 648         if (!tsk->fs)
 649                 return -ENOMEM;
 650         return 0;
 651 }
 652
 653 static int count_open_files(struct files_struct *files, int size)
 654 {
 655         int i;
 656
 657         /* Find the last open fd */
 658         for (i = size/(8*sizeof(long)); i > 0; ) {
 659                 if (files->open_fds->fds_bits[--i])
 660                         break;
 661         }
 662         i = (i+1) * 8 * sizeof(long);
 663         return i;
 664 }
 665
 666 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 667 {
 668         struct files_struct *oldf, *newf;
 669         struct file **old_fds, **new_fds;
 670         int open_files, nfds, size, i, error = 0;
 671
 672         /*
 673          * A background process may not have any files ...
 674          */
 675         oldf = current->files;
 676         if (!oldf)
 677                 goto out;
 678
 679         if (clone_flags & CLONE_FILES) {
 680                 atomic_inc(&oldf->count);
 681                 goto out;
 682         }
 683
 684         /*
 685          * Note: we may be using current for both targets (See exec.c)
 686          * This works because we cache current->files (old) as oldf. Don't
 687          * break this.
 688          */
 689         tsk->files = NULL;
 690         error = -ENOMEM;
 691         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
 692         if (!newf)
 693                 goto out;
 694
 695         atomic_set(&newf->count, 1);
 696
 697         newf->file_lock     = SPIN_LOCK_UNLOCKED;
 698         newf->next_fd       = 0;
 699         newf->max_fds       = NR_OPEN_DEFAULT;
 700         newf->max_fdset     = __FD_SETSIZE;
 701         newf->close_on_exec = &newf->close_on_exec_init;
 702         newf->open_fds      = &newf->open_fds_init;
 703         newf->fd            = &newf->fd_array[0];
 704
 705         /* We don't yet have the oldf readlock, but even if the old
 706            fdset gets grown now, we'll only copy up to "size" fds */
 707         size = oldf->max_fdset;
 708         if (size > __FD_SETSIZE) {
 709                 newf->max_fdset = 0;
 710                 spin_lock(&newf->file_lock);
 711                 error = expand_fdset(newf, size-1);
 712                 spin_unlock(&newf->file_lock);
 713                 if (error)
 714                         goto out_release;
 715         }
 716         spin_lock(&oldf->file_lock);
 717
 718         open_files = count_open_files(oldf, size);
 719
 720         /*
 721          * Check whether we need to allocate a larger fd array.
 722          * Note: we're not a clone task, so the open count won't
 723          * change.
 724          */
 725         nfds = NR_OPEN_DEFAULT;
 726         if (open_files > nfds) {
 727                 spin_unlock(&oldf->file_lock);
 728                 newf->max_fds = 0;
 729                 spin_lock(&newf->file_lock);
 730                 error = expand_fd_array(newf, open_files-1);
 731                 spin_unlock(&newf->file_lock);
 732                 if (error)
 733                         goto out_release;
 734                 nfds = newf->max_fds;
 735                 spin_lock(&oldf->file_lock);
 736         }
 737
 738         old_fds = oldf->fd;
 739         new_fds = newf->fd;
 740
 741         memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
 742         memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
 743
 744         for (i = open_files; i != 0; i--) {
 745                 struct file *f = *old_fds++;
 746                 if (f)
 747                         get_file(f);
 748                 *new_fds++ = f;
 749         }
 750         spin_unlock(&oldf->file_lock);
 751
 752         /* compute the remainder to be cleared */
 753         size = (newf->max_fds - open_files) * sizeof(struct file *);
 754
 755         /* This is long word aligned thus could use a optimized version */
 756         memset(new_fds, 0, size);
 757
 758         if (newf->max_fdset > open_files) {
 759                 int left = (newf->max_fdset-open_files)/8;
 760                 int start = open_files / (8 * sizeof(unsigned long));
 761
 762                 memset(&newf->open_fds->fds_bits[start], 0, left);
 763                 memset(&newf->close_on_exec->fds_bits[start], 0, left);
 764         }
 765
 766         tsk->files = newf;
 767         error = 0;
 768 out:
 769         return error;
 770
 771 out_release:
 772         free_fdset (newf->close_on_exec, newf->max_fdset);
 773         free_fdset (newf->open_fds, newf->max_fdset);
 774         kmem_cache_free(files_cachep, newf);
 775         goto out;
 776 }
 777
 778 /*
 779  *      Helper to unshare the files of the current task.
 780  *      We don't want to expose copy_files internals to
 781  *      the exec layer of the kernel.
 782  */
 783
 784 int unshare_files(void)
 785 {
 786         struct files_struct *files  = current->files;
 787         int rc;
 788
 789         if(!files)
 790                 BUG();
 791
 792         /* This can race but the race causes us to copy when we don't
 793            need to and drop the copy */
 794         if(atomic_read(&files->count) == 1)
 795         {
 796                 atomic_inc(&files->count);
 797                 return 0;
 798         }
 799         rc = copy_files(0, current);
 800         if(rc)
 801                 current->files = files;
 802         return rc;
 803 }
 804
 805 EXPORT_SYMBOL(unshare_files);
 806
 807 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 808 {
 809         struct sighand_struct *sig;
 810
 811         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 812                 atomic_inc(&current->sighand->count);
 813                 return 0;
 814         }
 815         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 816         tsk->sighand = sig;
 817         if (!sig)
 818                 return -ENOMEM;
 819         spin_lock_init(&sig->siglock);
 820         atomic_set(&sig->count, 1);
 821         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 822         return 0;
 823 }
 824
 825 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 826 {
 827         struct signal_struct *sig;
 828
 829         if (clone_flags & CLONE_THREAD) {
 830                 atomic_inc(&current->signal->count);
 831                 return 0;
 832         }
 833         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 834         tsk->signal = sig;
 835         if (!sig)
 836                 return -ENOMEM;
 837         atomic_set(&sig->count, 1);
 838         sig->group_exit = 0;
 839         sig->group_exit_code = 0;
 840         sig->group_exit_task = NULL;
 841         sig->group_stop_count = 0;
 842         sig->curr_target = NULL;
 843         init_sigpending(&sig->shared_pending);
 844         INIT_LIST_HEAD(&sig->posix_timers);
 845
 846         sig->tty = current->signal->tty;
 847         sig->pgrp = process_group(current);
 848         sig->session = current->signal->session;
 849         sig->leader = 0;        /* session leadership doesn't inherit */
 850         sig->tty_old_pgrp = 0;
 851
 852         return 0;
 853 }
 854
 855 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 856 {
 857         unsigned long new_flags = p->flags;
 858
 859         new_flags &= ~PF_SUPERPRIV;
 860         new_flags |= PF_FORKNOEXEC;
 861         if (!(clone_flags & CLONE_PTRACE))
 862                 p->ptrace = 0;
 863         p->flags = new_flags;
 864 }
 865
 866 asmlinkage long sys_set_tid_address(int __user *tidptr)
 867 {
 868         current->clear_child_tid = tidptr;
 869
 870         return current->pid;
 871 }
 872
 873 /*
 874  * This creates a new process as a copy of the old one,
 875  * but does not actually start it yet.
 876  *
 877  * It copies the registers, and all the appropriate
 878  * parts of the process environment (as per the clone
 879  * flags). The actual kick-off is left to the caller.
 880  */
 881 struct task_struct *copy_process(unsigned long clone_flags,
 882                                  unsigned long stack_start,
 883                                  struct pt_regs *regs,
 884                                  unsigned long stack_size,
 885                                  int __user *parent_tidptr,
 886                                  int __user *child_tidptr)
 887 {
 888         int retval;
 889         struct task_struct *p = NULL;
 890         struct vx_info *vxi;
 891
 892         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 893                 return ERR_PTR(-EINVAL);
 894
 895         /*
 896          * Thread groups must share signals as well, and detached threads
 897          * can only be started up within the thread group.
 898          */
 899         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 900                 return ERR_PTR(-EINVAL);
 901
 902         /*
 903          * Shared signal handlers imply shared VM. By way of the above,
 904          * thread groups also imply shared VM. Blocking this case allows
 905          * for various simplifications in other code.
 906          */
 907         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 908                 return ERR_PTR(-EINVAL);
 909
 910         retval = security_task_create(clone_flags);
 911         if (retval)
 912                 goto fork_out;
 913
 914         retval = -ENOMEM;
 915
 916         p = dup_task_struct(current);
 917         if (!p)
 918                 goto fork_out;
 919
 920         p->vx_info = NULL;
 921         set_vx_info(&p->vx_info, current->vx_info);
 922         p->nx_info = NULL;
 923         set_nx_info(&p->nx_info, current->nx_info);
 924
 925         /* check vserver memory */
 926         if (p->mm && !(clone_flags & CLONE_VM)) {
 927                 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
 928                         vx_pages_add(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
 929                 else
 930                         goto bad_fork_free;
 931         }
 932         if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
 933                 if (!vx_rsspages_avail(p->mm, p->mm->rss))
 934                         goto bad_fork_free;
 935         }
 936
 937         retval = -EAGAIN;
 938         if (!vx_nproc_avail(1))
 939                 goto bad_fork_free;
 940
 941         if (atomic_read(&p->user->processes) >=
 942                         p->rlim[RLIMIT_NPROC].rlim_cur) {
 943                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 944                                 p->user != &root_user)
 945                         goto bad_fork_free;
 946         }
 947
 948         atomic_inc(&p->user->__count);
 949         atomic_inc(&p->user->processes);
 950         get_group_info(p->group_info);
 951
 952         /*
 953          * If multiple threads are within copy_process(), then this check
 954          * triggers too late. This doesn't hurt, the check is only there
 955          * to stop root fork bombs.
 956          */
 957         if (nr_threads >= max_threads)
 958                 goto bad_fork_cleanup_count;
 959
 960         if (!try_module_get(p->thread_info->exec_domain->module))
 961                 goto bad_fork_cleanup_count;
 962
 963         if (p->binfmt && !try_module_get(p->binfmt->module))
 964                 goto bad_fork_cleanup_put_domain;
 965
 966         init_delays(p);
 967         p->did_exec = 0;
 968         copy_flags(clone_flags, p);
 969         if (clone_flags & CLONE_IDLETASK)
 970                 p->pid = 0;
 971         else {
 972                 p->pid = alloc_pidmap();
 973                 if (p->pid == -1)
 974                         goto bad_fork_cleanup;
 975         }
 976         retval = -EFAULT;
 977         if (clone_flags & CLONE_PARENT_SETTID)
 978                 if (put_user(p->pid, parent_tidptr))
 979                         goto bad_fork_cleanup;
 980
 981         p->proc_dentry = NULL;
 982
 983         INIT_LIST_HEAD(&p->children);
 984         INIT_LIST_HEAD(&p->sibling);
 985         init_waitqueue_head(&p->wait_chldexit);
 986         p->vfork_done = NULL;
 987         spin_lock_init(&p->alloc_lock);
 988         spin_lock_init(&p->proc_lock);
 989
 990         clear_tsk_thread_flag(p, TIF_SIGPENDING);
 991         init_sigpending(&p->pending);
 992
 993         p->it_real_value = p->it_virt_value = p->it_prof_value = 0;
 994         p->it_real_incr = p->it_virt_incr = p->it_prof_incr = 0;
 995         init_timer(&p->real_timer);
 996         p->real_timer.data = (unsigned long) p;
 997
 998         p->utime = p->stime = 0;
 999         p->cutime = p->cstime = 0;
1000         p->lock_depth = -1;             /* -1 = no lock */
1001         p->start_time = get_jiffies_64();
1002         p->security = NULL;
1003         p->io_context = NULL;
1004         p->audit_context = NULL;
1005 #ifdef CONFIG_NUMA
1006         p->mempolicy = mpol_copy(p->mempolicy);
1007         if (IS_ERR(p->mempolicy)) {
1008                 retval = PTR_ERR(p->mempolicy);
1009                 p->mempolicy = NULL;
1010                 goto bad_fork_cleanup;
1011         }
1012 #endif
1013
1014         retval = -ENOMEM;
1015         if ((retval = security_task_alloc(p)))
1016                 goto bad_fork_cleanup_policy;
1017         if ((retval = audit_alloc(p)))
1018                 goto bad_fork_cleanup_security;
1019         /* copy all the process information */
1020         if ((retval = copy_semundo(clone_flags, p)))
1021                 goto bad_fork_cleanup_audit;
1022         if ((retval = copy_files(clone_flags, p)))
1023                 goto bad_fork_cleanup_semundo;
1024         if ((retval = copy_fs(clone_flags, p)))
1025                 goto bad_fork_cleanup_files;
1026         if ((retval = copy_sighand(clone_flags, p)))
1027                 goto bad_fork_cleanup_fs;
1028         if ((retval = copy_signal(clone_flags, p)))
1029                 goto bad_fork_cleanup_sighand;
1030         if ((retval = copy_mm(clone_flags, p)))
1031                 goto bad_fork_cleanup_signal;
1032         if ((retval = copy_namespace(clone_flags, p)))
1033                 goto bad_fork_cleanup_mm;
1034         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1035         if (retval)
1036                 goto bad_fork_cleanup_namespace;
1037
1038         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1039         /*
1040          * Clear TID on mm_release()?
1041          */
1042         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1043
1044         /*
1045          * Syscall tracing should be turned off in the child regardless
1046          * of CLONE_PTRACE.
1047          */
1048         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1049
1050         /* Our parent execution domain becomes current domain
1051            These must match for thread signalling to apply */
1052
1053         p->parent_exec_id = p->self_exec_id;
1054
1055         /* ok, now we should be set up.. */
1056         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1057         p->pdeath_signal = 0;
1058
1059         /* Perform scheduler related setup */
1060         sched_fork(p);
1061
1062         /*
1063          * Ok, make it visible to the rest of the system.
1064          * We dont wake it up yet.
1065          */
1066         p->tgid = p->pid;
1067         p->group_leader = p;
1068         INIT_LIST_HEAD(&p->ptrace_children);
1069         INIT_LIST_HEAD(&p->ptrace_list);
1070
1071         /* Need tasklist lock for parent etc handling! */
1072         write_lock_irq(&tasklist_lock);
1073         /*
1074          * Check for pending SIGKILL! The new thread should not be allowed
1075          * to slip out of an OOM kill. (or normal SIGKILL.)
1076          */
1077         if (sigismember(&current->pending.signal, SIGKILL)) {
1078                 write_unlock_irq(&tasklist_lock);
1079                 retval = -EINTR;
1080                 goto bad_fork_cleanup_namespace;
1081         }
1082
1083         /* CLONE_PARENT re-uses the old parent */
1084         if (clone_flags & CLONE_PARENT)
1085                 p->real_parent = current->real_parent;
1086         else
1087                 p->real_parent = current;
1088         p->parent = p->real_parent;
1089
1090         if (clone_flags & CLONE_THREAD) {
1091                 spin_lock(&current->sighand->siglock);
1092                 /*
1093                  * Important: if an exit-all has been started then
1094                  * do not create this new thread - the whole thread
1095                  * group is supposed to exit anyway.
1096                  */
1097                 if (current->signal->group_exit) {
1098                         spin_unlock(&current->sighand->siglock);
1099                         write_unlock_irq(&tasklist_lock);
1100                         retval = -EAGAIN;
1101                         goto bad_fork_cleanup_namespace;
1102                 }
1103                 p->tgid = current->tgid;
1104                 p->group_leader = current->group_leader;
1105
1106                 if (current->signal->group_stop_count > 0) {
1107                         /*
1108                          * There is an all-stop in progress for the group.
1109                          * We ourselves will stop as soon as we check signals.
1110                          * Make the new thread part of that group stop too.
1111                          */
1112                         current->signal->group_stop_count++;
1113                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1114                 }
1115
1116                 spin_unlock(&current->sighand->siglock);
1117         }
1118
1119         SET_LINKS(p);
1120         if (p->ptrace & PT_PTRACED)
1121                 __ptrace_link(p, current->parent);
1122
1123         attach_pid(p, PIDTYPE_PID, p->pid);
1124         if (thread_group_leader(p)) {
1125                 attach_pid(p, PIDTYPE_TGID, p->tgid);
1126                 attach_pid(p, PIDTYPE_PGID, process_group(p));
1127                 attach_pid(p, PIDTYPE_SID, p->signal->session);
1128                 if (p->pid)
1129                         __get_cpu_var(process_counts)++;
1130         } else
1131                 link_pid(p, p->pids + PIDTYPE_TGID, &p->group_leader->pids[PIDTYPE_TGID].pid);
1132
1133         nr_threads++;
1134         vxi = current->vx_info;
1135         if (vxi) {
1136                 atomic_inc(&vxi->cacct.nr_threads);
1137                 // atomic_inc(&vxi->limit.rcur[RLIMIT_NPROC]);
1138         }
1139         vx_nproc_inc();
1140         write_unlock_irq(&tasklist_lock);
1141         retval = 0;
1142
1143 fork_out:
1144         if (retval)
1145                 return ERR_PTR(retval);
1146         return p;
1147
1148 bad_fork_cleanup_namespace:
1149         exit_namespace(p);
1150 bad_fork_cleanup_mm:
1151         exit_mm(p);
1152         if (p->active_mm)
1153                 mmdrop(p->active_mm);
1154 bad_fork_cleanup_signal:
1155         exit_signal(p);
1156 bad_fork_cleanup_sighand:
1157         exit_sighand(p);
1158 bad_fork_cleanup_fs:
1159         exit_fs(p); /* blocking */
1160 bad_fork_cleanup_files:
1161         exit_files(p); /* blocking */
1162 bad_fork_cleanup_semundo:
1163         exit_sem(p);
1164 bad_fork_cleanup_audit:
1165         audit_free(p);
1166 bad_fork_cleanup_security:
1167         security_task_free(p);
1168 bad_fork_cleanup_policy:
1169 #ifdef CONFIG_NUMA
1170         mpol_free(p->mempolicy);
1171 #endif
1172 bad_fork_cleanup:
1173         if (p->pid > 0)
1174                 free_pidmap(p->pid);
1175         if (p->binfmt)
1176                 module_put(p->binfmt->module);
1177 bad_fork_cleanup_put_domain:
1178         module_put(p->thread_info->exec_domain->module);
1179 bad_fork_cleanup_count:
1180         put_group_info(p->group_info);
1181         atomic_dec(&p->user->processes);
1182         free_uid(p->user);
1183 bad_fork_free:
1184         free_task(p);
1185         goto fork_out;
1186 }
1187
1188 static inline int fork_traceflag (unsigned clone_flags)
1189 {
1190         if (clone_flags & (CLONE_UNTRACED | CLONE_IDLETASK))
1191                 return 0;
1192         else if (clone_flags & CLONE_VFORK) {
1193                 if (current->ptrace & PT_TRACE_VFORK)
1194                         return PTRACE_EVENT_VFORK;
1195         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1196                 if (current->ptrace & PT_TRACE_CLONE)
1197                         return PTRACE_EVENT_CLONE;
1198         } else if (current->ptrace & PT_TRACE_FORK)
1199                 return PTRACE_EVENT_FORK;
1200
1201         return 0;
1202 }
1203
1204 /*
1205  *  Ok, this is the main fork-routine.
1206  *
1207  * It copies the process, and if successful kick-starts
1208  * it and waits for it to finish using the VM if required.
1209  */
1210 long do_fork(unsigned long clone_flags,
1211               unsigned long stack_start,
1212               struct pt_regs *regs,
1213               unsigned long stack_size,
1214               int __user *parent_tidptr,
1215               int __user *child_tidptr)
1216 {
1217         struct task_struct *p;
1218         int trace = 0;
1219         long pid;
1220
1221         if (unlikely(current->ptrace)) {
1222                 trace = fork_traceflag (clone_flags);
1223                 if (trace)
1224                         clone_flags |= CLONE_PTRACE;
1225         }
1226
1227 #ifdef CONFIG_CKRM_TYPE_TASKCLASS
1228         if (numtasks_get_ref(current->taskclass, 0) == 0) {
1229                 return -ENOMEM;
1230         }
1231 #endif
1232
1233         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr);
1234         /*
1235          * Do this prior waking up the new thread - the thread pointer
1236          * might get invalid after that point, if the thread exits quickly.
1237          */
1238         pid = IS_ERR(p) ? PTR_ERR(p) : p->pid;
1239
1240         if (!IS_ERR(p)) {
1241                 struct completion vfork;
1242
1243                 ckrm_cb_fork(p);
1244
1245                 if (clone_flags & CLONE_VFORK) {
1246                         p->vfork_done = &vfork;
1247                         init_completion(&vfork);
1248                 }
1249
1250                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1251                         /*
1252                          * We'll start up with an immediate SIGSTOP.
1253                          */
1254                         sigaddset(&p->pending.signal, SIGSTOP);
1255                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1256                 }
1257
1258                 if (!(clone_flags & CLONE_STOPPED)) {
1259                         /*
1260                          * Do the wakeup last. On SMP we treat fork() and
1261                          * CLONE_VM separately, because fork() has already
1262                          * created cache footprint on this CPU (due to
1263                          * copying the pagetables), hence migration would
1264                          * probably be costy. Threads on the other hand
1265                          * have less traction to the current CPU, and if
1266                          * there's an imbalance then the scheduler can
1267                          * migrate this fresh thread now, before it
1268                          * accumulates a larger cache footprint:
1269                          */
1270                         if (clone_flags & CLONE_VM)
1271                                 wake_up_forked_thread(p);
1272                         else
1273                                 wake_up_forked_process(p);
1274                 } else {
1275                         int cpu = get_cpu();
1276
1277                         p->state = TASK_STOPPED;
1278                         if (cpu_is_offline(task_cpu(p)))
1279                                 set_task_cpu(p, cpu);
1280
1281                         put_cpu();
1282                 }
1283                 ++total_forks;
1284
1285                 if (unlikely (trace)) {
1286                         current->ptrace_message = pid;
1287                         ptrace_notify ((trace << 8) | SIGTRAP);
1288                 }
1289
1290                 if (clone_flags & CLONE_VFORK) {
1291                         wait_for_completion(&vfork);
1292                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1293                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1294                 } else
1295                         /*
1296                          * Let the child process run first, to avoid most of the
1297                          * COW overhead when the child exec()s afterwards.
1298                          */
1299                         set_need_resched();
1300         } else {
1301 #ifdef CONFIG_CKRM_TYPE_TASKCLASS
1302                 numtasks_put_ref(current->taskclass);
1303 #endif
1304         }
1305         return pid;
1306 }
1307
1308 /* SLAB cache for signal_struct structures (tsk->signal) */
1309 kmem_cache_t *signal_cachep;
1310
1311 /* SLAB cache for sighand_struct structures (tsk->sighand) */
1312 kmem_cache_t *sighand_cachep;
1313
1314 /* SLAB cache for files_struct structures (tsk->files) */
1315 kmem_cache_t *files_cachep;
1316
1317 /* SLAB cache for fs_struct structures (tsk->fs) */
1318 kmem_cache_t *fs_cachep;
1319
1320 /* SLAB cache for vm_area_struct structures */
1321 kmem_cache_t *vm_area_cachep;
1322
1323 /* SLAB cache for mm_struct structures (tsk->mm) */
1324 kmem_cache_t *mm_cachep;
1325
1326 void __init proc_caches_init(void)
1327 {
1328         sighand_cachep = kmem_cache_create("sighand_cache",
1329                         sizeof(struct sighand_struct), 0,
1330                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1331         signal_cachep = kmem_cache_create("signal_cache",
1332                         sizeof(struct signal_struct), 0,
1333                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1334         files_cachep = kmem_cache_create("files_cache",
1335                         sizeof(struct files_struct), 0,
1336                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1337         fs_cachep = kmem_cache_create("fs_cache",
1338                         sizeof(struct fs_struct), 0,
1339                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1340         vm_area_cachep = kmem_cache_create("vm_area_struct",
1341                         sizeof(struct vm_area_struct), 0,
1342                         SLAB_PANIC, NULL, NULL);
1343         mm_cachep = kmem_cache_create("mm_struct",
1344                         sizeof(struct mm_struct), 0,
1345                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1346 }