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/ptrace.h>
  39 #include <linux/mount.h>
  40 #include <linux/audit.h>
  41 #include <linux/profile.h>
  42 #include <linux/rmap.h>
  43 #include <linux/acct.h>
  44 #include <linux/vs_network.h>
  45 #include <linux/vs_limit.h>
  46 #include <linux/vs_memory.h>
  47
  48 #include <asm/pgtable.h>
  49 #include <asm/pgalloc.h>
  50 #include <asm/uaccess.h>
  51 #include <asm/mmu_context.h>
  52 #include <asm/cacheflush.h>
  53 #include <asm/tlbflush.h>
  54
  55 /*
  56  * Protected counters by write_lock_irq(&tasklist_lock)
  57  */
  58 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  59 int nr_threads;                 /* The idle threads do not count.. */
  60
  61 int max_threads;                /* tunable limit on nr_threads */
  62
  63 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  64
  65  __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  66
  67 EXPORT_SYMBOL(tasklist_lock);
  68
  69 int nr_processes(void)
  70 {
  71         int cpu;
  72         int total = 0;
  73
  74         for_each_online_cpu(cpu)
  75                 total += per_cpu(process_counts, cpu);
  76
  77         return total;
  78 }
  79
  80 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  81 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  82 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  83 static kmem_cache_t *task_struct_cachep;
  84 #endif
  85
  86 /* SLAB cache for signal_struct structures (tsk->signal) */
  87 kmem_cache_t *signal_cachep;
  88
  89 /* SLAB cache for sighand_struct structures (tsk->sighand) */
  90 kmem_cache_t *sighand_cachep;
  91
  92 /* SLAB cache for files_struct structures (tsk->files) */
  93 kmem_cache_t *files_cachep;
  94
  95 /* SLAB cache for fs_struct structures (tsk->fs) */
  96 kmem_cache_t *fs_cachep;
  97
  98 /* SLAB cache for vm_area_struct structures */
  99 kmem_cache_t *vm_area_cachep;
 100
 101 /* SLAB cache for mm_struct structures (tsk->mm) */
 102 static kmem_cache_t *mm_cachep;
 103
 104 void free_task(struct task_struct *tsk)
 105 {
 106         free_thread_info(tsk->thread_info);
 107         clr_vx_info(&tsk->vx_info);
 108         clr_nx_info(&tsk->nx_info);
 109         free_task_struct(tsk);
 110 }
 111 EXPORT_SYMBOL(free_task);
 112
 113 void __put_task_struct(struct task_struct *tsk)
 114 {
 115         WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
 116         WARN_ON(atomic_read(&tsk->usage));
 117         WARN_ON(tsk == current);
 118
 119         if (unlikely(tsk->audit_context))
 120                 audit_free(tsk);
 121         security_task_free(tsk);
 122         free_uid(tsk->user);
 123         put_group_info(tsk->group_info);
 124
 125         if (!profile_handoff_task(tsk))
 126                 free_task(tsk);
 127 }
 128
 129 void __init fork_init(unsigned long mempages)
 130 {
 131 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 132 #ifndef ARCH_MIN_TASKALIGN
 133 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 134 #endif
 135         /* create a slab on which task_structs can be allocated */
 136         task_struct_cachep =
 137                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 138                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
 139 #endif
 140
 141         /*
 142          * The default maximum number of threads is set to a safe
 143          * value: the thread structures can take up at most half
 144          * of memory.
 145          */
 146         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 147
 148         /*
 149          * we need to allow at least 20 threads to boot a system
 150          */
 151         if(max_threads < 20)
 152                 max_threads = 20;
 153
 154         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 155         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 156         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 157                 init_task.signal->rlim[RLIMIT_NPROC];
 158 }
 159
 160 static struct task_struct *dup_task_struct(struct task_struct *orig)
 161 {
 162         struct task_struct *tsk;
 163         struct thread_info *ti;
 164
 165         prepare_to_copy(orig);
 166
 167         tsk = alloc_task_struct();
 168         if (!tsk)
 169                 return NULL;
 170
 171         ti = alloc_thread_info(tsk);
 172         if (!ti) {
 173                 free_task_struct(tsk);
 174                 return NULL;
 175         }
 176
 177         *ti = *orig->thread_info;
 178         *tsk = *orig;
 179         tsk->thread_info = ti;
 180         ti->task = tsk;
 181
 182         /* One for us, one for whoever does the "release_task()" (usually parent) */
 183         atomic_set(&tsk->usage,2);
 184         return tsk;
 185 }
 186
 187 #ifdef CONFIG_MMU
 188 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
 189 {
 190         struct vm_area_struct * mpnt, *tmp, **pprev;
 191         struct rb_node **rb_link, *rb_parent;
 192         int retval;
 193         unsigned long charge;
 194         struct mempolicy *pol;
 195
 196         down_write(&oldmm->mmap_sem);
 197         flush_cache_mm(current->mm);
 198         mm->locked_vm = 0;
 199         mm->mmap = NULL;
 200         mm->mmap_cache = NULL;
 201         mm->free_area_cache = oldmm->mmap_base;
 202         mm->map_count = 0;
 203         __set_mm_counter(mm, rss, 0);
 204         __set_mm_counter(mm, anon_rss, 0);
 205         cpus_clear(mm->cpu_vm_mask);
 206         mm->mm_rb = RB_ROOT;
 207         rb_link = &mm->mm_rb.rb_node;
 208         rb_parent = NULL;
 209         pprev = &mm->mmap;
 210
 211         for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
 212                 struct file *file;
 213
 214                 if (mpnt->vm_flags & VM_DONTCOPY) {
 215                         __vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 216                                                         -vma_pages(mpnt));
 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                  * link in first so that swapoff can see swap entries,
 258                  * and try_to_unmap_one's find_vma find the new vma.
 259                  */
 260                 spin_lock(&mm->page_table_lock);
 261                 *pprev = tmp;
 262                 pprev = &tmp->vm_next;
 263
 264                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 265                 rb_link = &tmp->vm_rb.rb_right;
 266                 rb_parent = &tmp->vm_rb;
 267
 268                 mm->map_count++;
 269                 retval = copy_page_range(mm, current->mm, tmp);
 270                 spin_unlock(&mm->page_table_lock);
 271
 272                 if (tmp->vm_ops && tmp->vm_ops->open)
 273                         tmp->vm_ops->open(tmp);
 274
 275                 if (retval)
 276                         goto out;
 277         }
 278         retval = 0;
 279
 280 out:
 281         flush_tlb_mm(current->mm);
 282         up_write(&oldmm->mmap_sem);
 283         return retval;
 284 fail_nomem_policy:
 285         kmem_cache_free(vm_area_cachep, tmp);
 286 fail_nomem:
 287         retval = -ENOMEM;
 288         vm_unacct_memory(charge);
 289         goto out;
 290 }
 291
 292 static inline int mm_alloc_pgd(struct mm_struct * mm)
 293 {
 294         mm->pgd = pgd_alloc(mm);
 295         if (unlikely(!mm->pgd))
 296                 return -ENOMEM;
 297         return 0;
 298 }
 299
 300 static inline void mm_free_pgd(struct mm_struct * mm)
 301 {
 302         pgd_free(mm->pgd);
 303 }
 304 #else
 305 #define dup_mmap(mm, oldmm)     (0)
 306 #define mm_alloc_pgd(mm)        (0)
 307 #define mm_free_pgd(mm)
 308 #endif /* CONFIG_MMU */
 309
 310  __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 311
 312 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
 313 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 314
 315 #include <linux/init_task.h>
 316
 317 static struct mm_struct * mm_init(struct mm_struct * mm)
 318 {
 319         atomic_set(&mm->mm_users, 1);
 320         atomic_set(&mm->mm_count, 1);
 321         init_rwsem(&mm->mmap_sem);
 322         INIT_LIST_HEAD(&mm->mmlist);
 323         mm->core_waiters = 0;
 324         mm->nr_ptes = 0;
 325         spin_lock_init(&mm->page_table_lock);
 326         rwlock_init(&mm->ioctx_list_lock);
 327         mm->ioctx_list = NULL;
 328         mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
 329         mm->free_area_cache = TASK_UNMAPPED_BASE;
 330
 331         if (likely(!mm_alloc_pgd(mm))) {
 332                 mm->def_flags = 0;
 333                 set_vx_info(&mm->mm_vx_info, current->vx_info);
 334                 return mm;
 335         }
 336         free_mm(mm);
 337         return NULL;
 338 }
 339
 340 /*
 341  * Allocate and initialize an mm_struct.
 342  */
 343 struct mm_struct * mm_alloc(void)
 344 {
 345         struct mm_struct * mm;
 346
 347         mm = allocate_mm();
 348         if (mm) {
 349                 memset(mm, 0, sizeof(*mm));
 350                 mm = mm_init(mm);
 351         }
 352         return mm;
 353 }
 354
 355 /*
 356  * Called when the last reference to the mm
 357  * is dropped: either by a lazy thread or by
 358  * mmput. Free the page directory and the mm.
 359  */
 360 void fastcall __mmdrop(struct mm_struct *mm)
 361 {
 362         BUG_ON(mm == &init_mm);
 363         mm_free_pgd(mm);
 364         destroy_context(mm);
 365         clr_vx_info(&mm->mm_vx_info);
 366         free_mm(mm);
 367 }
 368
 369 /*
 370  * Decrement the use count and release all resources for an mm.
 371  */
 372 void mmput(struct mm_struct *mm)
 373 {
 374         if (atomic_dec_and_test(&mm->mm_users)) {
 375                 exit_aio(mm);
 376                 exit_mmap(mm);
 377                 if (!list_empty(&mm->mmlist)) {
 378                         spin_lock(&mmlist_lock);
 379                         list_del(&mm->mmlist);
 380                         spin_unlock(&mmlist_lock);
 381                 }
 382                 put_swap_token(mm);
 383                 mmdrop(mm);
 384         }
 385 }
 386 EXPORT_SYMBOL_GPL(mmput);
 387
 388 /**
 389  * get_task_mm - acquire a reference to the task's mm
 390  *
 391  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
 392  * this kernel workthread has transiently adopted a user mm with use_mm,
 393  * to do its AIO) is not set and if so returns a reference to it, after
 394  * bumping up the use count.  User must release the mm via mmput()
 395  * after use.  Typically used by /proc and ptrace.
 396  */
 397 struct mm_struct *get_task_mm(struct task_struct *task)
 398 {
 399         struct mm_struct *mm;
 400
 401         task_lock(task);
 402         mm = task->mm;
 403         if (mm) {
 404                 if (task->flags & PF_BORROWED_MM)
 405                         mm = NULL;
 406                 else
 407                         atomic_inc(&mm->mm_users);
 408         }
 409         task_unlock(task);
 410         return mm;
 411 }
 412 EXPORT_SYMBOL_GPL(get_task_mm);
 413
 414 /* Please note the differences between mmput and mm_release.
 415  * mmput is called whenever we stop holding onto a mm_struct,
 416  * error success whatever.
 417  *
 418  * mm_release is called after a mm_struct has been removed
 419  * from the current process.
 420  *
 421  * This difference is important for error handling, when we
 422  * only half set up a mm_struct for a new process and need to restore
 423  * the old one.  Because we mmput the new mm_struct before
 424  * restoring the old one. . .
 425  * Eric Biederman 10 January 1998
 426  */
 427 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 428 {
 429         struct completion *vfork_done = tsk->vfork_done;
 430
 431         /* Get rid of any cached register state */
 432         deactivate_mm(tsk, mm);
 433
 434         /* notify parent sleeping on vfork() */
 435         if (vfork_done) {
 436                 tsk->vfork_done = NULL;
 437                 complete(vfork_done);
 438         }
 439         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
 440                 u32 __user * tidptr = tsk->clear_child_tid;
 441                 tsk->clear_child_tid = NULL;
 442
 443                 /*
 444                  * We don't check the error code - if userspace has
 445                  * not set up a proper pointer then tough luck.
 446                  */
 447                 put_user(0, tidptr);
 448                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 449         }
 450 }
 451
 452 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 453 {
 454         struct mm_struct * mm, *oldmm;
 455         int retval;
 456
 457         tsk->min_flt = tsk->maj_flt = 0;
 458         tsk->nvcsw = tsk->nivcsw = 0;
 459
 460         tsk->mm = NULL;
 461         tsk->active_mm = NULL;
 462
 463         /*
 464          * Are we cloning a kernel thread?
 465          *
 466          * We need to steal a active VM for that..
 467          */
 468         oldmm = current->mm;
 469         if (!oldmm)
 470                 return 0;
 471
 472         if (clone_flags & CLONE_VM) {
 473                 atomic_inc(&oldmm->mm_users);
 474                 mm = oldmm;
 475                 /*
 476                  * There are cases where the PTL is held to ensure no
 477                  * new threads start up in user mode using an mm, which
 478                  * allows optimizing out ipis; the tlb_gather_mmu code
 479                  * is an example.
 480                  */
 481                 spin_unlock_wait(&oldmm->page_table_lock);
 482                 goto good_mm;
 483         }
 484
 485         retval = -ENOMEM;
 486         mm = allocate_mm();
 487         if (!mm)
 488                 goto fail_nomem;
 489
 490         /* Copy the current MM stuff.. */
 491         memcpy(mm, oldmm, sizeof(*mm));
 492         mm->mm_vx_info = NULL;
 493         if (!mm_init(mm))
 494                 goto fail_nomem;
 495
 496         if (init_new_context(tsk,mm))
 497                 goto fail_nocontext;
 498
 499         retval = dup_mmap(mm, oldmm);
 500         if (retval)
 501                 goto free_pt;
 502
 503         mm->hiwater_rss = get_mm_counter(mm,rss);
 504         mm->hiwater_vm = mm->total_vm;
 505
 506 good_mm:
 507         tsk->mm = mm;
 508         tsk->active_mm = mm;
 509         return 0;
 510
 511 free_pt:
 512         mmput(mm);
 513 fail_nomem:
 514         return retval;
 515
 516 fail_nocontext:
 517         /*
 518          * If init_new_context() failed, we cannot use mmput() to free the mm
 519          * because it calls destroy_context()
 520          */
 521         clr_vx_info(&mm->mm_vx_info);
 522         mm_free_pgd(mm);
 523         free_mm(mm);
 524         return retval;
 525 }
 526
 527 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 528 {
 529         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 530         /* We don't need to lock fs - think why ;-) */
 531         if (fs) {
 532                 atomic_set(&fs->count, 1);
 533                 rwlock_init(&fs->lock);
 534                 fs->umask = old->umask;
 535                 read_lock(&old->lock);
 536                 fs->rootmnt = mntget(old->rootmnt);
 537                 fs->root = dget(old->root);
 538                 fs->pwdmnt = mntget(old->pwdmnt);
 539                 fs->pwd = dget(old->pwd);
 540                 if (old->altroot) {
 541                         fs->altrootmnt = mntget(old->altrootmnt);
 542                         fs->altroot = dget(old->altroot);
 543                 } else {
 544                         fs->altrootmnt = NULL;
 545                         fs->altroot = NULL;
 546                 }
 547                 read_unlock(&old->lock);
 548         }
 549         return fs;
 550 }
 551
 552 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 553 {
 554         return __copy_fs_struct(old);
 555 }
 556
 557 EXPORT_SYMBOL_GPL(copy_fs_struct);
 558
 559 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 560 {
 561         if (clone_flags & CLONE_FS) {
 562                 atomic_inc(&current->fs->count);
 563                 return 0;
 564         }
 565         tsk->fs = __copy_fs_struct(current->fs);
 566         if (!tsk->fs)
 567                 return -ENOMEM;
 568         return 0;
 569 }
 570
 571 static int count_open_files(struct files_struct *files, int size)
 572 {
 573         int i;
 574
 575         /* Find the last open fd */
 576         for (i = size/(8*sizeof(long)); i > 0; ) {
 577                 if (files->open_fds->fds_bits[--i])
 578                         break;
 579         }
 580         i = (i+1) * 8 * sizeof(long);
 581         return i;
 582 }
 583
 584 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 585 {
 586         struct files_struct *oldf, *newf;
 587         struct file **old_fds, **new_fds;
 588         int open_files, size, i, error = 0, expand;
 589
 590         /*
 591          * A background process may not have any files ...
 592          */
 593         oldf = current->files;
 594         if (!oldf)
 595                 goto out;
 596
 597         if (clone_flags & CLONE_FILES) {
 598                 atomic_inc(&oldf->count);
 599                 goto out;
 600         }
 601
 602         /*
 603          * Note: we may be using current for both targets (See exec.c)
 604          * This works because we cache current->files (old) as oldf. Don't
 605          * break this.
 606          */
 607         tsk->files = NULL;
 608         error = -ENOMEM;
 609         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
 610         if (!newf)
 611                 goto out;
 612
 613         atomic_set(&newf->count, 1);
 614
 615         spin_lock_init(&newf->file_lock);
 616         newf->next_fd       = 0;
 617         newf->max_fds       = NR_OPEN_DEFAULT;
 618         newf->max_fdset     = __FD_SETSIZE;
 619         newf->close_on_exec = &newf->close_on_exec_init;
 620         newf->open_fds      = &newf->open_fds_init;
 621         newf->fd            = &newf->fd_array[0];
 622
 623         spin_lock(&oldf->file_lock);
 624
 625         open_files = count_open_files(oldf, oldf->max_fdset);
 626         expand = 0;
 627
 628         /*
 629          * Check whether we need to allocate a larger fd array or fd set.
 630          * Note: we're not a clone task, so the open count won't  change.
 631          */
 632         if (open_files > newf->max_fdset) {
 633                 newf->max_fdset = 0;
 634                 expand = 1;
 635         }
 636         if (open_files > newf->max_fds) {
 637                 newf->max_fds = 0;
 638                 expand = 1;
 639         }
 640
 641         /* if the old fdset gets grown now, we'll only copy up to "size" fds */
 642         if (expand) {
 643                 spin_unlock(&oldf->file_lock);
 644                 spin_lock(&newf->file_lock);
 645                 error = expand_files(newf, open_files-1);
 646                 spin_unlock(&newf->file_lock);
 647                 if (error < 0)
 648                         goto out_release;
 649                 spin_lock(&oldf->file_lock);
 650         }
 651
 652         old_fds = oldf->fd;
 653         new_fds = newf->fd;
 654
 655         memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
 656         memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
 657
 658         for (i = open_files; i != 0; i--) {
 659                 struct file *f = *old_fds++;
 660                 if (f) {
 661                         get_file(f);
 662                         /* FIXME sum it first for check and performance */
 663                         vx_openfd_inc(open_files - i);
 664                 } else {
 665                         /*
 666                          * The fd may be claimed in the fd bitmap but not yet
 667                          * instantiated in the files array if a sibling thread
 668                          * is partway through open().  So make sure that this
 669                          * fd is available to the new process.
 670                          */
 671                         FD_CLR(open_files - i, newf->open_fds);
 672                 }
 673                 *new_fds++ = f;
 674         }
 675         spin_unlock(&oldf->file_lock);
 676
 677         /* compute the remainder to be cleared */
 678         size = (newf->max_fds - open_files) * sizeof(struct file *);
 679
 680         /* This is long word aligned thus could use a optimized version */
 681         memset(new_fds, 0, size);
 682
 683         if (newf->max_fdset > open_files) {
 684                 int left = (newf->max_fdset-open_files)/8;
 685                 int start = open_files / (8 * sizeof(unsigned long));
 686
 687                 memset(&newf->open_fds->fds_bits[start], 0, left);
 688                 memset(&newf->close_on_exec->fds_bits[start], 0, left);
 689         }
 690
 691         tsk->files = newf;
 692         error = 0;
 693 out:
 694         return error;
 695
 696 out_release:
 697         free_fdset (newf->close_on_exec, newf->max_fdset);
 698         free_fdset (newf->open_fds, newf->max_fdset);
 699         free_fd_array(newf->fd, newf->max_fds);
 700         kmem_cache_free(files_cachep, newf);
 701         goto out;
 702 }
 703
 704 /*
 705  *      Helper to unshare the files of the current task.
 706  *      We don't want to expose copy_files internals to
 707  *      the exec layer of the kernel.
 708  */
 709
 710 int unshare_files(void)
 711 {
 712         struct files_struct *files  = current->files;
 713         int rc;
 714
 715         if(!files)
 716                 BUG();
 717
 718         /* This can race but the race causes us to copy when we don't
 719            need to and drop the copy */
 720         if(atomic_read(&files->count) == 1)
 721         {
 722                 atomic_inc(&files->count);
 723                 return 0;
 724         }
 725         rc = copy_files(0, current);
 726         if(rc)
 727                 current->files = files;
 728         return rc;
 729 }
 730
 731 EXPORT_SYMBOL(unshare_files);
 732
 733 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 734 {
 735         struct sighand_struct *sig;
 736
 737         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 738                 atomic_inc(&current->sighand->count);
 739                 return 0;
 740         }
 741         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 742         tsk->sighand = sig;
 743         if (!sig)
 744                 return -ENOMEM;
 745         spin_lock_init(&sig->siglock);
 746         atomic_set(&sig->count, 1);
 747         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 748         return 0;
 749 }
 750
 751 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 752 {
 753         struct signal_struct *sig;
 754         int ret;
 755
 756         if (clone_flags & CLONE_THREAD) {
 757                 atomic_inc(&current->signal->count);
 758                 atomic_inc(&current->signal->live);
 759                 return 0;
 760         }
 761         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 762         tsk->signal = sig;
 763         if (!sig)
 764                 return -ENOMEM;
 765
 766         ret = copy_thread_group_keys(tsk);
 767         if (ret < 0) {
 768                 kmem_cache_free(signal_cachep, sig);
 769                 return ret;
 770         }
 771
 772         atomic_set(&sig->count, 1);
 773         atomic_set(&sig->live, 1);
 774         init_waitqueue_head(&sig->wait_chldexit);
 775         sig->flags = 0;
 776         sig->group_exit_code = 0;
 777         sig->group_exit_task = NULL;
 778         sig->group_stop_count = 0;
 779         sig->curr_target = NULL;
 780         init_sigpending(&sig->shared_pending);
 781         INIT_LIST_HEAD(&sig->posix_timers);
 782
 783         sig->it_real_value = sig->it_real_incr = 0;
 784         sig->real_timer.function = it_real_fn;
 785         sig->real_timer.data = (unsigned long) tsk;
 786         init_timer(&sig->real_timer);
 787
 788         sig->it_virt_expires = cputime_zero;
 789         sig->it_virt_incr = cputime_zero;
 790         sig->it_prof_expires = cputime_zero;
 791         sig->it_prof_incr = cputime_zero;
 792
 793         sig->tty = current->signal->tty;
 794         sig->pgrp = process_group(current);
 795         sig->session = current->signal->session;
 796         sig->leader = 0;        /* session leadership doesn't inherit */
 797         sig->tty_old_pgrp = 0;
 798
 799         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 800         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 801         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 802         sig->sched_time = 0;
 803         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 804         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 805         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 806
 807         task_lock(current->group_leader);
 808         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 809         task_unlock(current->group_leader);
 810
 811         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 812                 /*
 813                  * New sole thread in the process gets an expiry time
 814                  * of the whole CPU time limit.
 815                  */
 816                 tsk->it_prof_expires =
 817                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 818         }
 819
 820         return 0;
 821 }
 822
 823 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 824 {
 825         unsigned long new_flags = p->flags;
 826
 827         new_flags &= ~PF_SUPERPRIV;
 828         new_flags |= PF_FORKNOEXEC;
 829         if (!(clone_flags & CLONE_PTRACE))
 830                 p->ptrace = 0;
 831         p->flags = new_flags;
 832 }
 833
 834 asmlinkage long sys_set_tid_address(int __user *tidptr)
 835 {
 836         current->clear_child_tid = tidptr;
 837
 838         return current->pid;
 839 }
 840
 841 /*
 842  * This creates a new process as a copy of the old one,
 843  * but does not actually start it yet.
 844  *
 845  * It copies the registers, and all the appropriate
 846  * parts of the process environment (as per the clone
 847  * flags). The actual kick-off is left to the caller.
 848  */
 849 static task_t *copy_process(unsigned long clone_flags,
 850                                  unsigned long stack_start,
 851                                  struct pt_regs *regs,
 852                                  unsigned long stack_size,
 853                                  int __user *parent_tidptr,
 854                                  int __user *child_tidptr,
 855                                  int pid)
 856 {
 857         int retval;
 858         struct task_struct *p = NULL;
 859         struct vx_info *vxi;
 860         struct nx_info *nxi;
 861
 862         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 863                 return ERR_PTR(-EINVAL);
 864
 865         /*
 866          * Thread groups must share signals as well, and detached threads
 867          * can only be started up within the thread group.
 868          */
 869         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 870                 return ERR_PTR(-EINVAL);
 871
 872         /*
 873          * Shared signal handlers imply shared VM. By way of the above,
 874          * thread groups also imply shared VM. Blocking this case allows
 875          * for various simplifications in other code.
 876          */
 877         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 878                 return ERR_PTR(-EINVAL);
 879
 880         retval = security_task_create(clone_flags);
 881         if (retval)
 882                 goto fork_out;
 883
 884         retval = -ENOMEM;
 885         p = dup_task_struct(current);
 886         if (!p)
 887                 goto fork_out;
 888         p->tux_info = NULL;
 889
 890         p->vx_info = NULL;
 891         set_vx_info(&p->vx_info, current->vx_info);
 892         p->nx_info = NULL;
 893         set_nx_info(&p->nx_info, current->nx_info);
 894
 895         /* check vserver memory */
 896         if (p->mm && !(clone_flags & CLONE_VM)) {
 897                 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
 898                         vx_pages_add(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
 899                 else
 900                         goto bad_fork_free;
 901         }
 902         if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
 903                 if (!vx_rsspages_avail(p->mm, get_mm_counter(p->mm, rss)))
 904                         goto bad_fork_cleanup_vm;
 905         }
 906
 907         p->vx_info = NULL;
 908         set_vx_info(&p->vx_info, current->vx_info);
 909         p->nx_info = NULL;
 910         set_nx_info(&p->nx_info, current->nx_info);
 911
 912         /* check vserver memory */
 913         if (p->mm && !(clone_flags & CLONE_VM)) {
 914                 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
 915                         vx_pages_add(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
 916                 else
 917                         goto bad_fork_free;
 918         }
 919         if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
 920                 if (!vx_rsspages_avail(p->mm, get_mm_counter(p->mm, rss)))
 921                         goto bad_fork_cleanup_vm;
 922         }
 923
 924         init_vx_info(&p->vx_info, current->vx_info);
 925         p->nx_info = NULL;
 926         set_nx_info(&p->nx_info, current->nx_info);
 927
 928         /* check vserver memory */
 929         if (p->mm && !(clone_flags & CLONE_VM)) {
 930                 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
 931                         vx_pages_add(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
 932                 else
 933                         goto bad_fork_free;
 934         }
 935         if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
 936                 if (!vx_rsspages_avail(p->mm, get_mm_counter(p->mm, rss)))
 937                         goto bad_fork_cleanup_vm;
 938         }
 939
 940         init_vx_info(&p->vx_info, current->vx_info);
 941         init_nx_info(&p->nx_info, current->nx_info);
 942
 943         /* check vserver memory */
 944         if (p->mm && !(clone_flags & CLONE_VM)) {
 945                 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
 946                         vx_pages_add(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
 947                 else
 948                         goto bad_fork_free;
 949         }
 950         if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
 951                 if (!vx_rsspages_avail(p->mm, get_mm_counter(p->mm, rss)))
 952                         goto bad_fork_cleanup_vm;
 953         }
 954
 955         init_vx_info(&p->vx_info, current->vx_info);
 956         init_nx_info(&p->nx_info, current->nx_info);
 957
 958         /* check vserver memory */
 959         if (p->mm && !(clone_flags & CLONE_VM)) {
 960                 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
 961                         vx_pages_add(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
 962                 else
 963                         goto bad_fork_free;
 964         }
 965         if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
 966                 if (!vx_rsspages_avail(p->mm, get_mm_counter(p->mm, rss)))
 967                         goto bad_fork_cleanup_vm;
 968         }
 969
 970         retval = -EAGAIN;
 971         if (!vx_nproc_avail(1))
 972                 goto bad_fork_cleanup_vm;
 973
 974         if (atomic_read(&p->user->processes) >=
 975                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
 976                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 977                                 p->user != &root_user)
 978                         goto bad_fork_cleanup_vm;
 979         }
 980
 981         atomic_inc(&p->user->__count);
 982         atomic_inc(&p->user->processes);
 983         get_group_info(p->group_info);
 984
 985         /*
 986          * If multiple threads are within copy_process(), then this check
 987          * triggers too late. This doesn't hurt, the check is only there
 988          * to stop root fork bombs.
 989          */
 990         if (nr_threads >= max_threads)
 991                 goto bad_fork_cleanup_count;
 992
 993         if (!try_module_get(p->thread_info->exec_domain->module))
 994                 goto bad_fork_cleanup_count;
 995
 996         if (p->binfmt && !try_module_get(p->binfmt->module))
 997                 goto bad_fork_cleanup_put_domain;
 998
 999         p->did_exec = 0;
1000         copy_flags(clone_flags, p);
1001         p->pid = pid;
1002         retval = -EFAULT;
1003         if (clone_flags & CLONE_PARENT_SETTID)
1004                 if (put_user(p->pid, parent_tidptr))
1005                         goto bad_fork_cleanup;
1006
1007         p->proc_dentry = NULL;
1008
1009         INIT_LIST_HEAD(&p->children);
1010         INIT_LIST_HEAD(&p->sibling);
1011         p->vfork_done = NULL;
1012         spin_lock_init(&p->alloc_lock);
1013         spin_lock_init(&p->proc_lock);
1014
1015         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1016         init_sigpending(&p->pending);
1017
1018         p->utime = cputime_zero;
1019         p->stime = cputime_zero;
1020         p->sched_time = 0;
1021         p->rchar = 0;           /* I/O counter: bytes read */
1022         p->wchar = 0;           /* I/O counter: bytes written */
1023         p->syscr = 0;           /* I/O counter: read syscalls */
1024         p->syscw = 0;           /* I/O counter: write syscalls */
1025         acct_clear_integrals(p);
1026
1027         p->it_virt_expires = cputime_zero;
1028         p->it_prof_expires = cputime_zero;
1029         p->it_sched_expires = 0;
1030         INIT_LIST_HEAD(&p->cpu_timers[0]);
1031         INIT_LIST_HEAD(&p->cpu_timers[1]);
1032         INIT_LIST_HEAD(&p->cpu_timers[2]);
1033
1034         p->lock_depth = -1;             /* -1 = no lock */
1035         do_posix_clock_monotonic_gettime(&p->start_time);
1036         p->security = NULL;
1037         p->io_context = NULL;
1038         p->io_wait = NULL;
1039         p->audit_context = NULL;
1040 #ifdef CONFIG_NUMA
1041         p->mempolicy = mpol_copy(p->mempolicy);
1042         if (IS_ERR(p->mempolicy)) {
1043                 retval = PTR_ERR(p->mempolicy);
1044                 p->mempolicy = NULL;
1045                 goto bad_fork_cleanup;
1046         }
1047 #endif
1048
1049         p->tgid = p->pid;
1050         if (clone_flags & CLONE_THREAD)
1051                 p->tgid = current->tgid;
1052
1053         if ((retval = security_task_alloc(p)))
1054                 goto bad_fork_cleanup_policy;
1055         if ((retval = audit_alloc(p)))
1056                 goto bad_fork_cleanup_security;
1057         /* copy all the process information */
1058         if ((retval = copy_semundo(clone_flags, p)))
1059                 goto bad_fork_cleanup_audit;
1060         if ((retval = copy_files(clone_flags, p)))
1061                 goto bad_fork_cleanup_semundo;
1062         if ((retval = copy_fs(clone_flags, p)))
1063                 goto bad_fork_cleanup_files;
1064         if ((retval = copy_sighand(clone_flags, p)))
1065                 goto bad_fork_cleanup_fs;
1066         if ((retval = copy_signal(clone_flags, p)))
1067                 goto bad_fork_cleanup_sighand;
1068         if ((retval = copy_mm(clone_flags, p)))
1069                 goto bad_fork_cleanup_signal;
1070         if ((retval = copy_keys(clone_flags, p)))
1071                 goto bad_fork_cleanup_mm;
1072         if ((retval = copy_namespace(clone_flags, p)))
1073                 goto bad_fork_cleanup_keys;
1074         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1075         if (retval)
1076                 goto bad_fork_cleanup_namespace;
1077
1078         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1079         /*
1080          * Clear TID on mm_release()?
1081          */
1082         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1083
1084         /*
1085          * Syscall tracing should be turned off in the child regardless
1086          * of CLONE_PTRACE.
1087          */
1088         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1089
1090         /* Our parent execution domain becomes current domain
1091            These must match for thread signalling to apply */
1092
1093         p->parent_exec_id = p->self_exec_id;
1094
1095         /* ok, now we should be set up.. */
1096         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1097         p->pdeath_signal = 0;
1098         p->exit_state = 0;
1099
1100         /* Perform scheduler related setup */
1101         sched_fork(p);
1102
1103         /*
1104          * Ok, make it visible to the rest of the system.
1105          * We dont wake it up yet.
1106          */
1107         p->group_leader = p;
1108         INIT_LIST_HEAD(&p->ptrace_children);
1109         INIT_LIST_HEAD(&p->ptrace_list);
1110
1111         /* Need tasklist lock for parent etc handling! */
1112         write_lock_irq(&tasklist_lock);
1113
1114         /*
1115          * The task hasn't been attached yet, so cpus_allowed mask cannot
1116          * have changed. The cpus_allowed mask of the parent may have
1117          * changed after it was copied first time, and it may then move to
1118          * another CPU - so we re-copy it here and set the child's CPU to
1119          * the parent's CPU. This avoids alot of nasty races.
1120          */
1121         p->cpus_allowed = current->cpus_allowed;
1122         set_task_cpu(p, smp_processor_id());
1123
1124         /*
1125          * Check for pending SIGKILL! The new thread should not be allowed
1126          * to slip out of an OOM kill. (or normal SIGKILL.)
1127          */
1128         if (sigismember(&current->pending.signal, SIGKILL)) {
1129                 write_unlock_irq(&tasklist_lock);
1130                 retval = -EINTR;
1131                 goto bad_fork_cleanup_namespace;
1132         }
1133
1134         /* CLONE_PARENT re-uses the old parent */
1135         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1136                 p->real_parent = current->real_parent;
1137         else
1138                 p->real_parent = current;
1139         p->parent = p->real_parent;
1140
1141         if (clone_flags & CLONE_THREAD) {
1142                 spin_lock(&current->sighand->siglock);
1143                 /*
1144                  * Important: if an exit-all has been started then
1145                  * do not create this new thread - the whole thread
1146                  * group is supposed to exit anyway.
1147                  */
1148                 if (current->signal->flags & SIGNAL_GROUP_EXIT) {
1149                         spin_unlock(&current->sighand->siglock);
1150                         write_unlock_irq(&tasklist_lock);
1151                         retval = -EAGAIN;
1152                         goto bad_fork_cleanup_namespace;
1153                 }
1154                 p->group_leader = current->group_leader;
1155
1156                 if (current->signal->group_stop_count > 0) {
1157                         /*
1158                          * There is an all-stop in progress for the group.
1159                          * We ourselves will stop as soon as we check signals.
1160                          * Make the new thread part of that group stop too.
1161                          */
1162                         current->signal->group_stop_count++;
1163                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1164                 }
1165
1166                 if (!cputime_eq(current->signal->it_virt_expires,
1167                                 cputime_zero) ||
1168                     !cputime_eq(current->signal->it_prof_expires,
1169                                 cputime_zero) ||
1170                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1171                     !list_empty(&current->signal->cpu_timers[0]) ||
1172                     !list_empty(&current->signal->cpu_timers[1]) ||
1173                     !list_empty(&current->signal->cpu_timers[2])) {
1174                         /*
1175                          * Have child wake up on its first tick to check
1176                          * for process CPU timers.
1177                          */
1178                         p->it_prof_expires = jiffies_to_cputime(1);
1179                 }
1180
1181                 spin_unlock(&current->sighand->siglock);
1182         }
1183
1184         SET_LINKS(p);
1185         if (unlikely(p->ptrace & PT_PTRACED))
1186                 __ptrace_link(p, current->parent);
1187
1188         cpuset_fork(p);
1189
1190         attach_pid(p, PIDTYPE_PID, p->pid);
1191         attach_pid(p, PIDTYPE_TGID, p->tgid);
1192         if (thread_group_leader(p)) {
1193                 attach_pid(p, PIDTYPE_PGID, process_group(p));
1194                 attach_pid(p, PIDTYPE_SID, p->signal->session);
1195                 if (p->pid)
1196                         __get_cpu_var(process_counts)++;
1197         }
1198
1199         p->ioprio = current->ioprio;
1200         nr_threads++;
1201         total_forks++;
1202
1203         /* p is copy of current */
1204         vxi = p->vx_info;
1205         if (vxi) {
1206                 claim_vx_info(vxi, p);
1207                 atomic_inc(&vxi->cvirt.nr_threads);
1208                 atomic_inc(&vxi->cvirt.total_forks);
1209                 vx_nproc_inc(p);
1210         }
1211         nxi = p->nx_info;
1212         if (nxi)
1213                 claim_nx_info(nxi, p);
1214         write_unlock_irq(&tasklist_lock);
1215         retval = 0;
1216
1217 fork_out:
1218         if (retval)
1219                 return ERR_PTR(retval);
1220         return p;
1221
1222 bad_fork_cleanup_namespace:
1223         exit_namespace(p);
1224 bad_fork_cleanup_keys:
1225         exit_keys(p);
1226 bad_fork_cleanup_mm:
1227         if (p->mm)
1228                 mmput(p->mm);
1229 bad_fork_cleanup_signal:
1230         exit_signal(p);
1231 bad_fork_cleanup_sighand:
1232         exit_sighand(p);
1233 bad_fork_cleanup_fs:
1234         exit_fs(p); /* blocking */
1235 bad_fork_cleanup_files:
1236         exit_files(p); /* blocking */
1237 bad_fork_cleanup_semundo:
1238         exit_sem(p);
1239 bad_fork_cleanup_audit:
1240         audit_free(p);
1241 bad_fork_cleanup_security:
1242         security_task_free(p);
1243 bad_fork_cleanup_policy:
1244 #ifdef CONFIG_NUMA
1245         mpol_free(p->mempolicy);
1246 #endif
1247 bad_fork_cleanup:
1248         if (p->binfmt)
1249                 module_put(p->binfmt->module);
1250 bad_fork_cleanup_put_domain:
1251         module_put(p->thread_info->exec_domain->module);
1252 bad_fork_cleanup_count:
1253         put_group_info(p->group_info);
1254         atomic_dec(&p->user->processes);
1255         free_uid(p->user);
1256 bad_fork_cleanup_vm:
1257         if (p->mm && !(clone_flags & CLONE_VM))
1258                 vx_pages_sub(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
1259 bad_fork_free:
1260         free_task(p);
1261         goto fork_out;
1262 }
1263
1264 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1265 {
1266         memset(regs, 0, sizeof(struct pt_regs));
1267         return regs;
1268 }
1269
1270 task_t * __devinit fork_idle(int cpu)
1271 {
1272         task_t *task;
1273         struct pt_regs regs;
1274
1275         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1276         if (!task)
1277                 return ERR_PTR(-ENOMEM);
1278         init_idle(task, cpu);
1279         unhash_process(task);
1280         return task;
1281 }
1282
1283 static inline int fork_traceflag (unsigned clone_flags)
1284 {
1285         if (clone_flags & CLONE_UNTRACED)
1286                 return 0;
1287         else if (clone_flags & CLONE_VFORK) {
1288                 if (current->ptrace & PT_TRACE_VFORK)
1289                         return PTRACE_EVENT_VFORK;
1290         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1291                 if (current->ptrace & PT_TRACE_CLONE)
1292                         return PTRACE_EVENT_CLONE;
1293         } else if (current->ptrace & PT_TRACE_FORK)
1294                 return PTRACE_EVENT_FORK;
1295
1296         return 0;
1297 }
1298
1299 /*
1300  *  Ok, this is the main fork-routine.
1301  *
1302  * It copies the process, and if successful kick-starts
1303  * it and waits for it to finish using the VM if required.
1304  */
1305 long do_fork(unsigned long clone_flags,
1306               unsigned long stack_start,
1307               struct pt_regs *regs,
1308               unsigned long stack_size,
1309               int __user *parent_tidptr,
1310               int __user *child_tidptr)
1311 {
1312         struct task_struct *p;
1313         int trace = 0;
1314         long pid = alloc_pidmap();
1315
1316         if (pid < 0)
1317                 return -EAGAIN;
1318         if (unlikely(current->ptrace)) {
1319                 trace = fork_traceflag (clone_flags);
1320                 if (trace)
1321                         clone_flags |= CLONE_PTRACE;
1322         }
1323
1324         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
1325
1326         /*
1327          * Do this prior waking up the new thread - the thread pointer
1328          * might get invalid after that point, if the thread exits quickly.
1329          */
1330         if (!IS_ERR(p)) {
1331                 struct completion vfork;
1332
1333                 if (clone_flags & CLONE_VFORK) {
1334                         p->vfork_done = &vfork;
1335                         init_completion(&vfork);
1336                 }
1337
1338                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1339                         /*
1340                          * We'll start up with an immediate SIGSTOP.
1341                          */
1342                         sigaddset(&p->pending.signal, SIGSTOP);
1343                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1344                 }
1345
1346                 if (!(clone_flags & CLONE_STOPPED))
1347                         wake_up_new_task(p, clone_flags);
1348                 else
1349                         p->state = TASK_STOPPED;
1350
1351                 if (unlikely (trace)) {
1352                         current->ptrace_message = pid;
1353                         ptrace_notify ((trace << 8) | SIGTRAP);
1354                 }
1355
1356                 if (clone_flags & CLONE_VFORK) {
1357                         wait_for_completion(&vfork);
1358                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1359                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1360                 }
1361         } else {
1362                 free_pidmap(pid);
1363                 pid = PTR_ERR(p);
1364         }
1365         return pid;
1366 }
1367
1368 void __init proc_caches_init(void)
1369 {
1370         sighand_cachep = kmem_cache_create("sighand_cache",
1371                         sizeof(struct sighand_struct), 0,
1372                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1373         signal_cachep = kmem_cache_create("signal_cache",
1374                         sizeof(struct signal_struct), 0,
1375                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1376         files_cachep = kmem_cache_create("files_cache",
1377                         sizeof(struct files_struct), 0,
1378                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1379         fs_cachep = kmem_cache_create("fs_cache",
1380                         sizeof(struct fs_struct), 0,
1381                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1382         vm_area_cachep = kmem_cache_create("vm_area_struct",
1383                         sizeof(struct vm_area_struct), 0,
1384                         SLAB_PANIC, NULL, NULL);
1385         mm_cachep = kmem_cache_create("mm_struct",
1386                         sizeof(struct mm_struct), 0,
1387                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1388 }