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/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/smp_lock.h>
  18 #include <linux/module.h>
  19 #include <linux/vmalloc.h>
  20 #include <linux/completion.h>
  21 #include <linux/namespace.h>
  22 #include <linux/personality.h>
  23 #include <linux/mempolicy.h>
  24 #include <linux/sem.h>
  25 #include <linux/file.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/fs.h>
  30 #include <linux/capability.h>
  31 #include <linux/cpu.h>
  32 #include <linux/cpuset.h>
  33 #include <linux/security.h>
  34 #include <linux/swap.h>
  35 #include <linux/syscalls.h>
  36 #include <linux/jiffies.h>
  37 #include <linux/futex.h>
  38 #include <linux/rcupdate.h>
  39 #include <linux/ptrace.h>
  40 #include <linux/mount.h>
  41 #include <linux/audit.h>
  42 #include <linux/profile.h>
  43 #include <linux/rmap.h>
  44 #include <linux/acct.h>
  45 #include <linux/cn_proc.h>
  46 #include <linux/delayacct.h>
  47 #include <linux/taskstats_kern.h>
  48 #include <linux/random.h>
  49
  50 #include <asm/pgtable.h>
  51 #include <asm/pgalloc.h>
  52 #include <asm/uaccess.h>
  53 #include <asm/mmu_context.h>
  54 #include <asm/cacheflush.h>
  55 #include <asm/tlbflush.h>
  56
  57 /*
  58  * Protected counters by write_lock_irq(&tasklist_lock)
  59  */
  60 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  61 int nr_threads;                 /* The idle threads do not count.. */
  62
  63 int max_threads;                /* tunable limit on nr_threads */
  64
  65 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  66
  67 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  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 static 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         rt_mutex_debug_task_free(tsk);
 108         free_task_struct(tsk);
 109 }
 110 EXPORT_SYMBOL(free_task);
 111
 112 void __put_task_struct(struct task_struct *tsk)
 113 {
 114         WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
 115         WARN_ON(atomic_read(&tsk->usage));
 116         WARN_ON(tsk == current);
 117
 118         security_task_free(tsk);
 119         free_uid(tsk->user);
 120         put_group_info(tsk->group_info);
 121         delayacct_tsk_free(tsk);
 122
 123         if (!profile_handoff_task(tsk))
 124                 free_task(tsk);
 125 }
 126
 127 void __init fork_init(unsigned long mempages)
 128 {
 129 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 130 #ifndef ARCH_MIN_TASKALIGN
 131 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 132 #endif
 133         /* create a slab on which task_structs can be allocated */
 134         task_struct_cachep =
 135                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 136                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
 137 #endif
 138
 139         /*
 140          * The default maximum number of threads is set to a safe
 141          * value: the thread structures can take up at most half
 142          * of memory.
 143          */
 144         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 145
 146         /*
 147          * we need to allow at least 20 threads to boot a system
 148          */
 149         if(max_threads < 20)
 150                 max_threads = 20;
 151
 152         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 153         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 154         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 155                 init_task.signal->rlim[RLIMIT_NPROC];
 156 }
 157
 158 static struct task_struct *dup_task_struct(struct task_struct *orig)
 159 {
 160         struct task_struct *tsk;
 161         struct thread_info *ti;
 162
 163         prepare_to_copy(orig);
 164
 165         tsk = alloc_task_struct();
 166         if (!tsk)
 167                 return NULL;
 168
 169         ti = alloc_thread_info(tsk);
 170         if (!ti) {
 171                 free_task_struct(tsk);
 172                 return NULL;
 173         }
 174
 175         *tsk = *orig;
 176         tsk->thread_info = ti;
 177         setup_thread_stack(tsk, orig);
 178
 179 #ifdef CONFIG_CC_STACKPROTECTOR
 180         tsk->stack_canary = get_random_int();
 181 #endif
 182
 183         /* One for us, one for whoever does the "release_task()" (usually parent) */
 184         atomic_set(&tsk->usage,2);
 185         atomic_set(&tsk->fs_excl, 0);
 186 #ifdef CONFIG_BLK_DEV_IO_TRACE
 187         tsk->btrace_seq = 0;
 188 #endif
 189         tsk->splice_pipe = NULL;
 190         return tsk;
 191 }
 192
 193 #ifdef CONFIG_MMU
 194 static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 195 {
 196         struct vm_area_struct *mpnt, *tmp, **pprev;
 197         struct rb_node **rb_link, *rb_parent;
 198         int retval;
 199         unsigned long charge;
 200         struct mempolicy *pol;
 201
 202         down_write(&oldmm->mmap_sem);
 203         flush_cache_mm(oldmm);
 204         /*
 205          * Not linked in yet - no deadlock potential:
 206          */
 207         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 208
 209         mm->locked_vm = 0;
 210         mm->mmap = NULL;
 211         mm->mmap_cache = NULL;
 212         mm->free_area_cache = oldmm->mmap_base;
 213         mm->cached_hole_size = ~0UL;
 214         mm->map_count = 0;
 215         cpus_clear(mm->cpu_vm_mask);
 216         mm->mm_rb = RB_ROOT;
 217         rb_link = &mm->mm_rb.rb_node;
 218         rb_parent = NULL;
 219         pprev = &mm->mmap;
 220
 221         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 222                 struct file *file;
 223
 224                 if (mpnt->vm_flags & VM_DONTCOPY) {
 225                         long pages = vma_pages(mpnt);
 226                         mm->total_vm -= pages;
 227                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 228                                                                 -pages);
 229                         continue;
 230                 }
 231                 charge = 0;
 232                 if (mpnt->vm_flags & VM_ACCOUNT) {
 233                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 234                         if (security_vm_enough_memory(len))
 235                                 goto fail_nomem;
 236                         charge = len;
 237                 }
 238                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 239                 if (!tmp)
 240                         goto fail_nomem;
 241                 *tmp = *mpnt;
 242                 pol = mpol_copy(vma_policy(mpnt));
 243                 retval = PTR_ERR(pol);
 244                 if (IS_ERR(pol))
 245                         goto fail_nomem_policy;
 246                 vma_set_policy(tmp, pol);
 247                 tmp->vm_flags &= ~VM_LOCKED;
 248                 tmp->vm_mm = mm;
 249                 tmp->vm_next = NULL;
 250                 anon_vma_link(tmp);
 251                 file = tmp->vm_file;
 252                 if (file) {
 253                         struct inode *inode = file->f_dentry->d_inode;
 254                         get_file(file);
 255                         if (tmp->vm_flags & VM_DENYWRITE)
 256                                 atomic_dec(&inode->i_writecount);
 257
 258                         /* insert tmp into the share list, just after mpnt */
 259                         spin_lock(&file->f_mapping->i_mmap_lock);
 260                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 261                         flush_dcache_mmap_lock(file->f_mapping);
 262                         vma_prio_tree_add(tmp, mpnt);
 263                         flush_dcache_mmap_unlock(file->f_mapping);
 264                         spin_unlock(&file->f_mapping->i_mmap_lock);
 265                 }
 266
 267                 /*
 268                  * Link in the new vma and copy the page table entries.
 269                  */
 270                 *pprev = tmp;
 271                 pprev = &tmp->vm_next;
 272
 273                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 274                 rb_link = &tmp->vm_rb.rb_right;
 275                 rb_parent = &tmp->vm_rb;
 276
 277                 mm->map_count++;
 278                 retval = copy_page_range(mm, oldmm, mpnt);
 279
 280                 if (tmp->vm_ops && tmp->vm_ops->open)
 281                         tmp->vm_ops->open(tmp);
 282
 283                 if (retval)
 284                         goto out;
 285         }
 286         retval = 0;
 287 out:
 288         up_write(&mm->mmap_sem);
 289         flush_tlb_mm(oldmm);
 290         up_write(&oldmm->mmap_sem);
 291         return retval;
 292 fail_nomem_policy:
 293         kmem_cache_free(vm_area_cachep, tmp);
 294 fail_nomem:
 295         retval = -ENOMEM;
 296         vm_unacct_memory(charge);
 297         goto out;
 298 }
 299
 300 static inline int mm_alloc_pgd(struct mm_struct * mm)
 301 {
 302         mm->pgd = pgd_alloc(mm);
 303         if (unlikely(!mm->pgd))
 304                 return -ENOMEM;
 305         return 0;
 306 }
 307
 308 static inline void mm_free_pgd(struct mm_struct * mm)
 309 {
 310         pgd_free(mm->pgd);
 311 }
 312 #else
 313 #define dup_mmap(mm, oldmm)     (0)
 314 #define mm_alloc_pgd(mm)        (0)
 315 #define mm_free_pgd(mm)
 316 #endif /* CONFIG_MMU */
 317
 318  __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 319
 320 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
 321 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 322
 323 #include <linux/init_task.h>
 324
 325 static struct mm_struct * mm_init(struct mm_struct * mm)
 326 {
 327         atomic_set(&mm->mm_users, 1);
 328         atomic_set(&mm->mm_count, 1);
 329         init_rwsem(&mm->mmap_sem);
 330         INIT_LIST_HEAD(&mm->mmlist);
 331         mm->core_waiters = 0;
 332         mm->nr_ptes = 0;
 333         set_mm_counter(mm, file_rss, 0);
 334         set_mm_counter(mm, anon_rss, 0);
 335         spin_lock_init(&mm->page_table_lock);
 336         rwlock_init(&mm->ioctx_list_lock);
 337         mm->ioctx_list = NULL;
 338         mm->free_area_cache = TASK_UNMAPPED_BASE;
 339         mm->cached_hole_size = ~0UL;
 340
 341         if (likely(!mm_alloc_pgd(mm))) {
 342                 mm->def_flags = 0;
 343                 return mm;
 344         }
 345         free_mm(mm);
 346         return NULL;
 347 }
 348
 349 /*
 350  * Allocate and initialize an mm_struct.
 351  */
 352 struct mm_struct * mm_alloc(void)
 353 {
 354         struct mm_struct * mm;
 355
 356         mm = allocate_mm();
 357         if (mm) {
 358                 memset(mm, 0, sizeof(*mm));
 359                 mm = mm_init(mm);
 360         }
 361         return mm;
 362 }
 363
 364 /*
 365  * Called when the last reference to the mm
 366  * is dropped: either by a lazy thread or by
 367  * mmput. Free the page directory and the mm.
 368  */
 369 void fastcall __mmdrop(struct mm_struct *mm)
 370 {
 371         BUG_ON(mm == &init_mm);
 372         mm_free_pgd(mm);
 373         destroy_context(mm);
 374         free_mm(mm);
 375 }
 376
 377 /*
 378  * Decrement the use count and release all resources for an mm.
 379  */
 380 void mmput(struct mm_struct *mm)
 381 {
 382         might_sleep();
 383
 384         if (atomic_dec_and_test(&mm->mm_users)) {
 385                 exit_aio(mm);
 386                 exit_mmap(mm);
 387                 if (!list_empty(&mm->mmlist)) {
 388                         spin_lock(&mmlist_lock);
 389                         list_del(&mm->mmlist);
 390                         spin_unlock(&mmlist_lock);
 391                 }
 392                 put_swap_token(mm);
 393                 mmdrop(mm);
 394         }
 395 }
 396 EXPORT_SYMBOL_GPL(mmput);
 397
 398 /**
 399  * get_task_mm - acquire a reference to the task's mm
 400  *
 401  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
 402  * this kernel workthread has transiently adopted a user mm with use_mm,
 403  * to do its AIO) is not set and if so returns a reference to it, after
 404  * bumping up the use count.  User must release the mm via mmput()
 405  * after use.  Typically used by /proc and ptrace.
 406  */
 407 struct mm_struct *get_task_mm(struct task_struct *task)
 408 {
 409         struct mm_struct *mm;
 410
 411         task_lock(task);
 412         mm = task->mm;
 413         if (mm) {
 414                 if (task->flags & PF_BORROWED_MM)
 415                         mm = NULL;
 416                 else
 417                         atomic_inc(&mm->mm_users);
 418         }
 419         task_unlock(task);
 420         return mm;
 421 }
 422 EXPORT_SYMBOL_GPL(get_task_mm);
 423
 424 /* Please note the differences between mmput and mm_release.
 425  * mmput is called whenever we stop holding onto a mm_struct,
 426  * error success whatever.
 427  *
 428  * mm_release is called after a mm_struct has been removed
 429  * from the current process.
 430  *
 431  * This difference is important for error handling, when we
 432  * only half set up a mm_struct for a new process and need to restore
 433  * the old one.  Because we mmput the new mm_struct before
 434  * restoring the old one. . .
 435  * Eric Biederman 10 January 1998
 436  */
 437 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 438 {
 439         struct completion *vfork_done = tsk->vfork_done;
 440
 441         /* Get rid of any cached register state */
 442         deactivate_mm(tsk, mm);
 443
 444         /* notify parent sleeping on vfork() */
 445         if (vfork_done) {
 446                 tsk->vfork_done = NULL;
 447                 complete(vfork_done);
 448         }
 449         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
 450                 u32 __user * tidptr = tsk->clear_child_tid;
 451                 tsk->clear_child_tid = NULL;
 452
 453                 /*
 454                  * We don't check the error code - if userspace has
 455                  * not set up a proper pointer then tough luck.
 456                  */
 457                 put_user(0, tidptr);
 458                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 459         }
 460 }
 461
 462 /*
 463  * Allocate a new mm structure and copy contents from the
 464  * mm structure of the passed in task structure.
 465  */
 466 static struct mm_struct *dup_mm(struct task_struct *tsk)
 467 {
 468         struct mm_struct *mm, *oldmm = current->mm;
 469         int err;
 470
 471         if (!oldmm)
 472                 return NULL;
 473
 474         mm = allocate_mm();
 475         if (!mm)
 476                 goto fail_nomem;
 477
 478         memcpy(mm, oldmm, sizeof(*mm));
 479
 480         if (!mm_init(mm))
 481                 goto fail_nomem;
 482
 483         if (init_new_context(tsk, mm))
 484                 goto fail_nocontext;
 485
 486         err = dup_mmap(mm, oldmm);
 487         if (err)
 488                 goto free_pt;
 489
 490         mm->hiwater_rss = get_mm_rss(mm);
 491         mm->hiwater_vm = mm->total_vm;
 492
 493         return mm;
 494
 495 free_pt:
 496         mmput(mm);
 497
 498 fail_nomem:
 499         return NULL;
 500
 501 fail_nocontext:
 502         /*
 503          * If init_new_context() failed, we cannot use mmput() to free the mm
 504          * because it calls destroy_context()
 505          */
 506         mm_free_pgd(mm);
 507         free_mm(mm);
 508         return NULL;
 509 }
 510
 511 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 512 {
 513         struct mm_struct * mm, *oldmm;
 514         int retval;
 515
 516         tsk->min_flt = tsk->maj_flt = 0;
 517         tsk->nvcsw = tsk->nivcsw = 0;
 518
 519         tsk->mm = NULL;
 520         tsk->active_mm = NULL;
 521
 522         /*
 523          * Are we cloning a kernel thread?
 524          *
 525          * We need to steal a active VM for that..
 526          */
 527         oldmm = current->mm;
 528         if (!oldmm)
 529                 return 0;
 530
 531         if (clone_flags & CLONE_VM) {
 532                 atomic_inc(&oldmm->mm_users);
 533                 mm = oldmm;
 534                 goto good_mm;
 535         }
 536
 537         retval = -ENOMEM;
 538         mm = dup_mm(tsk);
 539         if (!mm)
 540                 goto fail_nomem;
 541
 542 good_mm:
 543         tsk->mm = mm;
 544         tsk->active_mm = mm;
 545         return 0;
 546
 547 fail_nomem:
 548         return retval;
 549 }
 550
 551 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 552 {
 553         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 554         /* We don't need to lock fs - think why ;-) */
 555         if (fs) {
 556                 atomic_set(&fs->count, 1);
 557                 rwlock_init(&fs->lock);
 558                 fs->umask = old->umask;
 559                 read_lock(&old->lock);
 560                 fs->rootmnt = mntget(old->rootmnt);
 561                 fs->root = dget(old->root);
 562                 fs->pwdmnt = mntget(old->pwdmnt);
 563                 fs->pwd = dget(old->pwd);
 564                 if (old->altroot) {
 565                         fs->altrootmnt = mntget(old->altrootmnt);
 566                         fs->altroot = dget(old->altroot);
 567                 } else {
 568                         fs->altrootmnt = NULL;
 569                         fs->altroot = NULL;
 570                 }
 571                 read_unlock(&old->lock);
 572         }
 573         return fs;
 574 }
 575
 576 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 577 {
 578         return __copy_fs_struct(old);
 579 }
 580
 581 EXPORT_SYMBOL_GPL(copy_fs_struct);
 582
 583 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 584 {
 585         if (clone_flags & CLONE_FS) {
 586                 atomic_inc(&current->fs->count);
 587                 return 0;
 588         }
 589         tsk->fs = __copy_fs_struct(current->fs);
 590         if (!tsk->fs)
 591                 return -ENOMEM;
 592         return 0;
 593 }
 594
 595 static int count_open_files(struct fdtable *fdt)
 596 {
 597         int size = fdt->max_fdset;
 598         int i;
 599
 600         /* Find the last open fd */
 601         for (i = size/(8*sizeof(long)); i > 0; ) {
 602                 if (fdt->open_fds->fds_bits[--i])
 603                         break;
 604         }
 605         i = (i+1) * 8 * sizeof(long);
 606         return i;
 607 }
 608
 609 static struct files_struct *alloc_files(void)
 610 {
 611         struct files_struct *newf;
 612         struct fdtable *fdt;
 613
 614         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
 615         if (!newf)
 616                 goto out;
 617
 618         atomic_set(&newf->count, 1);
 619
 620         spin_lock_init(&newf->file_lock);
 621         newf->next_fd = 0;
 622         fdt = &newf->fdtab;
 623         fdt->max_fds = NR_OPEN_DEFAULT;
 624         fdt->max_fdset = EMBEDDED_FD_SET_SIZE;
 625         fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
 626         fdt->open_fds = (fd_set *)&newf->open_fds_init;
 627         fdt->fd = &newf->fd_array[0];
 628         INIT_RCU_HEAD(&fdt->rcu);
 629         fdt->free_files = NULL;
 630         fdt->next = NULL;
 631         rcu_assign_pointer(newf->fdt, fdt);
 632 out:
 633         return newf;
 634 }
 635
 636 /*
 637  * Allocate a new files structure and copy contents from the
 638  * passed in files structure.
 639  * errorp will be valid only when the returned files_struct is NULL.
 640  */
 641 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
 642 {
 643         struct files_struct *newf;
 644         struct file **old_fds, **new_fds;
 645         int open_files, size, i, expand;
 646         struct fdtable *old_fdt, *new_fdt;
 647
 648         *errorp = -ENOMEM;
 649         newf = alloc_files();
 650         if (!newf)
 651                 goto out;
 652
 653         spin_lock(&oldf->file_lock);
 654         old_fdt = files_fdtable(oldf);
 655         new_fdt = files_fdtable(newf);
 656         size = old_fdt->max_fdset;
 657         open_files = count_open_files(old_fdt);
 658         expand = 0;
 659
 660         /*
 661          * Check whether we need to allocate a larger fd array or fd set.
 662          * Note: we're not a clone task, so the open count won't  change.
 663          */
 664         if (open_files > new_fdt->max_fdset) {
 665                 new_fdt->max_fdset = 0;
 666                 expand = 1;
 667         }
 668         if (open_files > new_fdt->max_fds) {
 669                 new_fdt->max_fds = 0;
 670                 expand = 1;
 671         }
 672
 673         /* if the old fdset gets grown now, we'll only copy up to "size" fds */
 674         if (expand) {
 675                 spin_unlock(&oldf->file_lock);
 676                 spin_lock(&newf->file_lock);
 677                 *errorp = expand_files(newf, open_files-1);
 678                 spin_unlock(&newf->file_lock);
 679                 if (*errorp < 0)
 680                         goto out_release;
 681                 new_fdt = files_fdtable(newf);
 682                 /*
 683                  * Reacquire the oldf lock and a pointer to its fd table
 684                  * who knows it may have a new bigger fd table. We need
 685                  * the latest pointer.
 686                  */
 687                 spin_lock(&oldf->file_lock);
 688                 old_fdt = files_fdtable(oldf);
 689         }
 690
 691         old_fds = old_fdt->fd;
 692         new_fds = new_fdt->fd;
 693
 694         memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
 695         memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
 696
 697         for (i = open_files; i != 0; i--) {
 698                 struct file *f = *old_fds++;
 699                 if (f) {
 700                         get_file(f);
 701                 } else {
 702                         /*
 703                          * The fd may be claimed in the fd bitmap but not yet
 704                          * instantiated in the files array if a sibling thread
 705                          * is partway through open().  So make sure that this
 706                          * fd is available to the new process.
 707                          */
 708                         FD_CLR(open_files - i, new_fdt->open_fds);
 709                 }
 710                 rcu_assign_pointer(*new_fds++, f);
 711         }
 712         spin_unlock(&oldf->file_lock);
 713
 714         /* compute the remainder to be cleared */
 715         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
 716
 717         /* This is long word aligned thus could use a optimized version */
 718         memset(new_fds, 0, size);
 719
 720         if (new_fdt->max_fdset > open_files) {
 721                 int left = (new_fdt->max_fdset-open_files)/8;
 722                 int start = open_files / (8 * sizeof(unsigned long));
 723
 724                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
 725                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
 726         }
 727
 728 out:
 729         return newf;
 730
 731 out_release:
 732         free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
 733         free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
 734         free_fd_array(new_fdt->fd, new_fdt->max_fds);
 735         kmem_cache_free(files_cachep, newf);
 736         return NULL;
 737 }
 738
 739 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 740 {
 741         struct files_struct *oldf, *newf;
 742         int error = 0;
 743
 744         /*
 745          * A background process may not have any files ...
 746          */
 747         oldf = current->files;
 748         if (!oldf)
 749                 goto out;
 750
 751         if (clone_flags & CLONE_FILES) {
 752                 atomic_inc(&oldf->count);
 753                 goto out;
 754         }
 755
 756         /*
 757          * Note: we may be using current for both targets (See exec.c)
 758          * This works because we cache current->files (old) as oldf. Don't
 759          * break this.
 760          */
 761         tsk->files = NULL;
 762         newf = dup_fd(oldf, &error);
 763         if (!newf)
 764                 goto out;
 765
 766         tsk->files = newf;
 767         error = 0;
 768 out:
 769         return error;
 770 }
 771
 772 /*
 773  *      Helper to unshare the files of the current task.
 774  *      We don't want to expose copy_files internals to
 775  *      the exec layer of the kernel.
 776  */
 777
 778 int unshare_files(void)
 779 {
 780         struct files_struct *files  = current->files;
 781         int rc;
 782
 783         BUG_ON(!files);
 784
 785         /* This can race but the race causes us to copy when we don't
 786            need to and drop the copy */
 787         if(atomic_read(&files->count) == 1)
 788         {
 789                 atomic_inc(&files->count);
 790                 return 0;
 791         }
 792         rc = copy_files(0, current);
 793         if(rc)
 794                 current->files = files;
 795         return rc;
 796 }
 797
 798 EXPORT_SYMBOL(unshare_files);
 799
 800 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 801 {
 802         struct sighand_struct *sig;
 803
 804         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 805                 atomic_inc(&current->sighand->count);
 806                 return 0;
 807         }
 808         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 809         rcu_assign_pointer(tsk->sighand, sig);
 810         if (!sig)
 811                 return -ENOMEM;
 812         atomic_set(&sig->count, 1);
 813         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 814         return 0;
 815 }
 816
 817 void __cleanup_sighand(struct sighand_struct *sighand)
 818 {
 819         if (atomic_dec_and_test(&sighand->count))
 820                 kmem_cache_free(sighand_cachep, sighand);
 821 }
 822
 823 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 824 {
 825         struct signal_struct *sig;
 826         int ret;
 827
 828         if (clone_flags & CLONE_THREAD) {
 829                 atomic_inc(&current->signal->count);
 830                 atomic_inc(&current->signal->live);
 831                 taskstats_tgid_alloc(current->signal);
 832                 return 0;
 833         }
 834         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 835         tsk->signal = sig;
 836         if (!sig)
 837                 return -ENOMEM;
 838
 839         ret = copy_thread_group_keys(tsk);
 840         if (ret < 0) {
 841                 kmem_cache_free(signal_cachep, sig);
 842                 return ret;
 843         }
 844
 845         atomic_set(&sig->count, 1);
 846         atomic_set(&sig->live, 1);
 847         init_waitqueue_head(&sig->wait_chldexit);
 848         sig->flags = 0;
 849         sig->group_exit_code = 0;
 850         sig->group_exit_task = NULL;
 851         sig->group_stop_count = 0;
 852         sig->curr_target = NULL;
 853         init_sigpending(&sig->shared_pending);
 854         INIT_LIST_HEAD(&sig->posix_timers);
 855
 856         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_REL);
 857         sig->it_real_incr.tv64 = 0;
 858         sig->real_timer.function = it_real_fn;
 859         sig->tsk = tsk;
 860
 861         sig->it_virt_expires = cputime_zero;
 862         sig->it_virt_incr = cputime_zero;
 863         sig->it_prof_expires = cputime_zero;
 864         sig->it_prof_incr = cputime_zero;
 865
 866         sig->leader = 0;        /* session leadership doesn't inherit */
 867         sig->tty_old_pgrp = 0;
 868
 869         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 870         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 871         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 872         sig->sched_time = 0;
 873         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 874         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 875         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 876         taskstats_tgid_init(sig);
 877
 878         task_lock(current->group_leader);
 879         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 880         task_unlock(current->group_leader);
 881
 882         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 883                 /*
 884                  * New sole thread in the process gets an expiry time
 885                  * of the whole CPU time limit.
 886                  */
 887                 tsk->it_prof_expires =
 888                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 889         }
 890         acct_init_pacct(&sig->pacct);
 891
 892         return 0;
 893 }
 894
 895 void __cleanup_signal(struct signal_struct *sig)
 896 {
 897         exit_thread_group_keys(sig);
 898         taskstats_tgid_free(sig);
 899         kmem_cache_free(signal_cachep, sig);
 900 }
 901
 902 static inline void cleanup_signal(struct task_struct *tsk)
 903 {
 904         struct signal_struct *sig = tsk->signal;
 905
 906         atomic_dec(&sig->live);
 907
 908         if (atomic_dec_and_test(&sig->count))
 909                 __cleanup_signal(sig);
 910 }
 911
 912 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 913 {
 914         unsigned long new_flags = p->flags;
 915
 916         new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
 917         new_flags |= PF_FORKNOEXEC;
 918         if (!(clone_flags & CLONE_PTRACE))
 919                 p->ptrace = 0;
 920         p->flags = new_flags;
 921 }
 922
 923 asmlinkage long sys_set_tid_address(int __user *tidptr)
 924 {
 925         current->clear_child_tid = tidptr;
 926
 927         return current->pid;
 928 }
 929
 930 static inline void rt_mutex_init_task(struct task_struct *p)
 931 {
 932 #ifdef CONFIG_RT_MUTEXES
 933         spin_lock_init(&p->pi_lock);
 934         plist_head_init(&p->pi_waiters, &p->pi_lock);
 935         p->pi_blocked_on = NULL;
 936 #endif
 937 }
 938
 939 /*
 940  * This creates a new process as a copy of the old one,
 941  * but does not actually start it yet.
 942  *
 943  * It copies the registers, and all the appropriate
 944  * parts of the process environment (as per the clone
 945  * flags). The actual kick-off is left to the caller.
 946  */
 947 static struct task_struct *copy_process(unsigned long clone_flags,
 948                                         unsigned long stack_start,
 949                                         struct pt_regs *regs,
 950                                         unsigned long stack_size,
 951                                         int __user *parent_tidptr,
 952                                         int __user *child_tidptr,
 953                                         int pid)
 954 {
 955         int retval;
 956         struct task_struct *p = NULL;
 957
 958         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 959                 return ERR_PTR(-EINVAL);
 960
 961         /*
 962          * Thread groups must share signals as well, and detached threads
 963          * can only be started up within the thread group.
 964          */
 965         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 966                 return ERR_PTR(-EINVAL);
 967
 968         /*
 969          * Shared signal handlers imply shared VM. By way of the above,
 970          * thread groups also imply shared VM. Blocking this case allows
 971          * for various simplifications in other code.
 972          */
 973         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 974                 return ERR_PTR(-EINVAL);
 975
 976         retval = security_task_create(clone_flags);
 977         if (retval)
 978                 goto fork_out;
 979
 980         retval = -ENOMEM;
 981         p = dup_task_struct(current);
 982         if (!p)
 983                 goto fork_out;
 984
 985 #ifdef CONFIG_TRACE_IRQFLAGS
 986         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
 987         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
 988 #endif
 989         retval = -EAGAIN;
 990         if (atomic_read(&p->user->processes) >=
 991                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
 992                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 993                                 p->user != &root_user)
 994                         goto bad_fork_free;
 995         }
 996
 997         atomic_inc(&p->user->__count);
 998         atomic_inc(&p->user->processes);
 999         get_group_info(p->group_info);
1000
1001         /*
1002          * If multiple threads are within copy_process(), then this check
1003          * triggers too late. This doesn't hurt, the check is only there
1004          * to stop root fork bombs.
1005          */
1006         if (nr_threads >= max_threads)
1007                 goto bad_fork_cleanup_count;
1008
1009         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1010                 goto bad_fork_cleanup_count;
1011
1012         if (p->binfmt && !try_module_get(p->binfmt->module))
1013                 goto bad_fork_cleanup_put_domain;
1014
1015         p->did_exec = 0;
1016         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1017         copy_flags(clone_flags, p);
1018         p->pid = pid;
1019         retval = -EFAULT;
1020         if (clone_flags & CLONE_PARENT_SETTID)
1021                 if (put_user(p->pid, parent_tidptr))
1022                         goto bad_fork_cleanup_delays_binfmt;
1023
1024         INIT_LIST_HEAD(&p->children);
1025         INIT_LIST_HEAD(&p->sibling);
1026         p->vfork_done = NULL;
1027         spin_lock_init(&p->alloc_lock);
1028
1029         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1030         init_sigpending(&p->pending);
1031
1032         p->utime = cputime_zero;
1033         p->stime = cputime_zero;
1034         p->sched_time = 0;
1035         p->rchar = 0;           /* I/O counter: bytes read */
1036         p->wchar = 0;           /* I/O counter: bytes written */
1037         p->syscr = 0;           /* I/O counter: read syscalls */
1038         p->syscw = 0;           /* I/O counter: write syscalls */
1039         acct_clear_integrals(p);
1040
1041         p->it_virt_expires = cputime_zero;
1042         p->it_prof_expires = cputime_zero;
1043         p->it_sched_expires = 0;
1044         INIT_LIST_HEAD(&p->cpu_timers[0]);
1045         INIT_LIST_HEAD(&p->cpu_timers[1]);
1046         INIT_LIST_HEAD(&p->cpu_timers[2]);
1047
1048         p->lock_depth = -1;             /* -1 = no lock */
1049         do_posix_clock_monotonic_gettime(&p->start_time);
1050         p->security = NULL;
1051         p->io_context = NULL;
1052         p->io_wait = NULL;
1053         p->audit_context = NULL;
1054         cpuset_fork(p);
1055 #ifdef CONFIG_NUMA
1056         p->mempolicy = mpol_copy(p->mempolicy);
1057         if (IS_ERR(p->mempolicy)) {
1058                 retval = PTR_ERR(p->mempolicy);
1059                 p->mempolicy = NULL;
1060                 goto bad_fork_cleanup_cpuset;
1061         }
1062         mpol_fix_fork_child_flag(p);
1063 #endif
1064 #ifdef CONFIG_TRACE_IRQFLAGS
1065         p->irq_events = 0;
1066 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1067         p->hardirqs_enabled = 1;
1068 #else
1069         p->hardirqs_enabled = 0;
1070 #endif
1071         p->hardirq_enable_ip = 0;
1072         p->hardirq_enable_event = 0;
1073         p->hardirq_disable_ip = _THIS_IP_;
1074         p->hardirq_disable_event = 0;
1075         p->softirqs_enabled = 1;
1076         p->softirq_enable_ip = _THIS_IP_;
1077         p->softirq_enable_event = 0;
1078         p->softirq_disable_ip = 0;
1079         p->softirq_disable_event = 0;
1080         p->hardirq_context = 0;
1081         p->softirq_context = 0;
1082 #endif
1083 #ifdef CONFIG_LOCKDEP
1084         p->lockdep_depth = 0; /* no locks held yet */
1085         p->curr_chain_key = 0;
1086         p->lockdep_recursion = 0;
1087 #endif
1088
1089         rt_mutex_init_task(p);
1090
1091 #ifdef CONFIG_DEBUG_MUTEXES
1092         p->blocked_on = NULL; /* not blocked yet */
1093 #endif
1094
1095         p->tgid = p->pid;
1096         if (clone_flags & CLONE_THREAD)
1097                 p->tgid = current->tgid;
1098
1099         if ((retval = security_task_alloc(p)))
1100                 goto bad_fork_cleanup_policy;
1101         if ((retval = audit_alloc(p)))
1102                 goto bad_fork_cleanup_security;
1103         /* copy all the process information */
1104         if ((retval = copy_semundo(clone_flags, p)))
1105                 goto bad_fork_cleanup_audit;
1106         if ((retval = copy_files(clone_flags, p)))
1107                 goto bad_fork_cleanup_semundo;
1108         if ((retval = copy_fs(clone_flags, p)))
1109                 goto bad_fork_cleanup_files;
1110         if ((retval = copy_sighand(clone_flags, p)))
1111                 goto bad_fork_cleanup_fs;
1112         if ((retval = copy_signal(clone_flags, p)))
1113                 goto bad_fork_cleanup_sighand;
1114         if ((retval = copy_mm(clone_flags, p)))
1115                 goto bad_fork_cleanup_signal;
1116         if ((retval = copy_keys(clone_flags, p)))
1117                 goto bad_fork_cleanup_mm;
1118         if ((retval = copy_namespace(clone_flags, p)))
1119                 goto bad_fork_cleanup_keys;
1120         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1121         if (retval)
1122                 goto bad_fork_cleanup_namespace;
1123
1124         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1125         /*
1126          * Clear TID on mm_release()?
1127          */
1128         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1129         p->robust_list = NULL;
1130 #ifdef CONFIG_COMPAT
1131         p->compat_robust_list = NULL;
1132 #endif
1133         INIT_LIST_HEAD(&p->pi_state_list);
1134         p->pi_state_cache = NULL;
1135
1136         /*
1137          * sigaltstack should be cleared when sharing the same VM
1138          */
1139         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1140                 p->sas_ss_sp = p->sas_ss_size = 0;
1141
1142         /*
1143          * Syscall tracing should be turned off in the child regardless
1144          * of CLONE_PTRACE.
1145          */
1146         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1147 #ifdef TIF_SYSCALL_EMU
1148         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1149 #endif
1150
1151         /* Our parent execution domain becomes current domain
1152            These must match for thread signalling to apply */
1153         p->parent_exec_id = p->self_exec_id;
1154
1155         /* ok, now we should be set up.. */
1156         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1157         p->pdeath_signal = 0;
1158         p->exit_state = 0;
1159
1160         /*
1161          * Ok, make it visible to the rest of the system.
1162          * We dont wake it up yet.
1163          */
1164         p->group_leader = p;
1165         INIT_LIST_HEAD(&p->thread_group);
1166         INIT_LIST_HEAD(&p->ptrace_children);
1167         INIT_LIST_HEAD(&p->ptrace_list);
1168
1169         /* Perform scheduler related setup. Assign this task to a CPU. */
1170         sched_fork(p, clone_flags);
1171
1172         /* Need tasklist lock for parent etc handling! */
1173         write_lock_irq(&tasklist_lock);
1174
1175         /* for sys_ioprio_set(IOPRIO_WHO_PGRP) */
1176         p->ioprio = current->ioprio;
1177
1178         /*
1179          * The task hasn't been attached yet, so its cpus_allowed mask will
1180          * not be changed, nor will its assigned CPU.
1181          *
1182          * The cpus_allowed mask of the parent may have changed after it was
1183          * copied first time - so re-copy it here, then check the child's CPU
1184          * to ensure it is on a valid CPU (and if not, just force it back to
1185          * parent's CPU). This avoids alot of nasty races.
1186          */
1187         p->cpus_allowed = current->cpus_allowed;
1188         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1189                         !cpu_online(task_cpu(p))))
1190                 set_task_cpu(p, smp_processor_id());
1191
1192         /* CLONE_PARENT re-uses the old parent */
1193         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1194                 p->real_parent = current->real_parent;
1195         else
1196                 p->real_parent = current;
1197         p->parent = p->real_parent;
1198
1199         spin_lock(&current->sighand->siglock);
1200
1201         /*
1202          * Process group and session signals need to be delivered to just the
1203          * parent before the fork or both the parent and the child after the
1204          * fork. Restart if a signal comes in before we add the new process to
1205          * it's process group.
1206          * A fatal signal pending means that current will exit, so the new
1207          * thread can't slip out of an OOM kill (or normal SIGKILL).
1208          */
1209         recalc_sigpending();
1210         if (signal_pending(current)) {
1211                 spin_unlock(&current->sighand->siglock);
1212                 write_unlock_irq(&tasklist_lock);
1213                 retval = -ERESTARTNOINTR;
1214                 goto bad_fork_cleanup_namespace;
1215         }
1216
1217         if (clone_flags & CLONE_THREAD) {
1218                 p->group_leader = current->group_leader;
1219                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1220
1221                 if (!cputime_eq(current->signal->it_virt_expires,
1222                                 cputime_zero) ||
1223                     !cputime_eq(current->signal->it_prof_expires,
1224                                 cputime_zero) ||
1225                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1226                     !list_empty(&current->signal->cpu_timers[0]) ||
1227                     !list_empty(&current->signal->cpu_timers[1]) ||
1228                     !list_empty(&current->signal->cpu_timers[2])) {
1229                         /*
1230                          * Have child wake up on its first tick to check
1231                          * for process CPU timers.
1232                          */
1233                         p->it_prof_expires = jiffies_to_cputime(1);
1234                 }
1235         }
1236
1237         if (likely(p->pid)) {
1238                 add_parent(p);
1239                 if (unlikely(p->ptrace & PT_PTRACED))
1240                         __ptrace_link(p, current->parent);
1241
1242                 if (thread_group_leader(p)) {
1243                         p->signal->tty = current->signal->tty;
1244                         p->signal->pgrp = process_group(current);
1245                         p->signal->session = current->signal->session;
1246                         attach_pid(p, PIDTYPE_PGID, process_group(p));
1247                         attach_pid(p, PIDTYPE_SID, p->signal->session);
1248
1249                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1250                         __get_cpu_var(process_counts)++;
1251                 }
1252                 attach_pid(p, PIDTYPE_PID, p->pid);
1253                 nr_threads++;
1254         }
1255
1256         total_forks++;
1257         spin_unlock(&current->sighand->siglock);
1258         write_unlock_irq(&tasklist_lock);
1259         proc_fork_connector(p);
1260         return p;
1261
1262 bad_fork_cleanup_namespace:
1263         exit_namespace(p);
1264 bad_fork_cleanup_keys:
1265         exit_keys(p);
1266 bad_fork_cleanup_mm:
1267         if (p->mm)
1268                 mmput(p->mm);
1269 bad_fork_cleanup_signal:
1270         cleanup_signal(p);
1271 bad_fork_cleanup_sighand:
1272         __cleanup_sighand(p->sighand);
1273 bad_fork_cleanup_fs:
1274         exit_fs(p); /* blocking */
1275 bad_fork_cleanup_files:
1276         exit_files(p); /* blocking */
1277 bad_fork_cleanup_semundo:
1278         exit_sem(p);
1279 bad_fork_cleanup_audit:
1280         audit_free(p);
1281 bad_fork_cleanup_security:
1282         security_task_free(p);
1283 bad_fork_cleanup_policy:
1284 #ifdef CONFIG_NUMA
1285         mpol_free(p->mempolicy);
1286 bad_fork_cleanup_cpuset:
1287 #endif
1288         cpuset_exit(p);
1289 bad_fork_cleanup_delays_binfmt:
1290         delayacct_tsk_free(p);
1291         if (p->binfmt)
1292                 module_put(p->binfmt->module);
1293 bad_fork_cleanup_put_domain:
1294         module_put(task_thread_info(p)->exec_domain->module);
1295 bad_fork_cleanup_count:
1296         put_group_info(p->group_info);
1297         atomic_dec(&p->user->processes);
1298         free_uid(p->user);
1299 bad_fork_free:
1300         free_task(p);
1301 fork_out:
1302         return ERR_PTR(retval);
1303 }
1304
1305 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1306 {
1307         memset(regs, 0, sizeof(struct pt_regs));
1308         return regs;
1309 }
1310
1311 struct task_struct * __devinit fork_idle(int cpu)
1312 {
1313         struct task_struct *task;
1314         struct pt_regs regs;
1315
1316         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1317         if (!task)
1318                 return ERR_PTR(-ENOMEM);
1319         init_idle(task, cpu);
1320
1321         return task;
1322 }
1323
1324 static inline int fork_traceflag (unsigned clone_flags)
1325 {
1326         if (clone_flags & CLONE_UNTRACED)
1327                 return 0;
1328         else if (clone_flags & CLONE_VFORK) {
1329                 if (current->ptrace & PT_TRACE_VFORK)
1330                         return PTRACE_EVENT_VFORK;
1331         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1332                 if (current->ptrace & PT_TRACE_CLONE)
1333                         return PTRACE_EVENT_CLONE;
1334         } else if (current->ptrace & PT_TRACE_FORK)
1335                 return PTRACE_EVENT_FORK;
1336
1337         return 0;
1338 }
1339
1340 /*
1341  *  Ok, this is the main fork-routine.
1342  *
1343  * It copies the process, and if successful kick-starts
1344  * it and waits for it to finish using the VM if required.
1345  */
1346 long do_fork(unsigned long clone_flags,
1347               unsigned long stack_start,
1348               struct pt_regs *regs,
1349               unsigned long stack_size,
1350               int __user *parent_tidptr,
1351               int __user *child_tidptr)
1352 {
1353         struct task_struct *p;
1354         int trace = 0;
1355         struct pid *pid = alloc_pid();
1356         long nr;
1357
1358         if (!pid)
1359                 return -EAGAIN;
1360         nr = pid->nr;
1361         if (unlikely(current->ptrace)) {
1362                 trace = fork_traceflag (clone_flags);
1363                 if (trace)
1364                         clone_flags |= CLONE_PTRACE;
1365         }
1366
1367         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, nr);
1368         /*
1369          * Do this prior waking up the new thread - the thread pointer
1370          * might get invalid after that point, if the thread exits quickly.
1371          */
1372         if (!IS_ERR(p)) {
1373                 struct completion vfork;
1374
1375                 if (clone_flags & CLONE_VFORK) {
1376                         p->vfork_done = &vfork;
1377                         init_completion(&vfork);
1378                 }
1379
1380                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1381                         /*
1382                          * We'll start up with an immediate SIGSTOP.
1383                          */
1384                         sigaddset(&p->pending.signal, SIGSTOP);
1385                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1386                 }
1387
1388                 if (!(clone_flags & CLONE_STOPPED))
1389                         wake_up_new_task(p, clone_flags);
1390                 else
1391                         p->state = TASK_STOPPED;
1392
1393                 if (unlikely (trace)) {
1394                         current->ptrace_message = nr;
1395                         ptrace_notify ((trace << 8) | SIGTRAP);
1396                 }
1397
1398                 if (clone_flags & CLONE_VFORK) {
1399                         wait_for_completion(&vfork);
1400                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1401                                 current->ptrace_message = nr;
1402                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1403                         }
1404                 }
1405         } else {
1406                 free_pid(pid);
1407                 nr = PTR_ERR(p);
1408         }
1409         return nr;
1410 }
1411
1412 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1413 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1414 #endif
1415
1416 static void sighand_ctor(void *data, kmem_cache_t *cachep, unsigned long flags)
1417 {
1418         struct sighand_struct *sighand = data;
1419
1420         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
1421                                         SLAB_CTOR_CONSTRUCTOR)
1422                 spin_lock_init(&sighand->siglock);
1423 }
1424
1425 void __init proc_caches_init(void)
1426 {
1427         sighand_cachep = kmem_cache_create("sighand_cache",
1428                         sizeof(struct sighand_struct), 0,
1429                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1430                         sighand_ctor, NULL);
1431         signal_cachep = kmem_cache_create("signal_cache",
1432                         sizeof(struct signal_struct), 0,
1433                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1434         files_cachep = kmem_cache_create("files_cache",
1435                         sizeof(struct files_struct), 0,
1436                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1437         fs_cachep = kmem_cache_create("fs_cache",
1438                         sizeof(struct fs_struct), 0,
1439                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1440         vm_area_cachep = kmem_cache_create("vm_area_struct",
1441                         sizeof(struct vm_area_struct), 0,
1442                         SLAB_PANIC, NULL, NULL);
1443         mm_cachep = kmem_cache_create("mm_struct",
1444                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1445                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1446 }
1447
1448
1449 /*
1450  * Check constraints on flags passed to the unshare system call and
1451  * force unsharing of additional process context as appropriate.
1452  */
1453 static inline void check_unshare_flags(unsigned long *flags_ptr)
1454 {
1455         /*
1456          * If unsharing a thread from a thread group, must also
1457          * unshare vm.
1458          */
1459         if (*flags_ptr & CLONE_THREAD)
1460                 *flags_ptr |= CLONE_VM;
1461
1462         /*
1463          * If unsharing vm, must also unshare signal handlers.
1464          */
1465         if (*flags_ptr & CLONE_VM)
1466                 *flags_ptr |= CLONE_SIGHAND;
1467
1468         /*
1469          * If unsharing signal handlers and the task was created
1470          * using CLONE_THREAD, then must unshare the thread
1471          */
1472         if ((*flags_ptr & CLONE_SIGHAND) &&
1473             (atomic_read(&current->signal->count) > 1))
1474                 *flags_ptr |= CLONE_THREAD;
1475
1476         /*
1477          * If unsharing namespace, must also unshare filesystem information.
1478          */
1479         if (*flags_ptr & CLONE_NEWNS)
1480                 *flags_ptr |= CLONE_FS;
1481 }
1482
1483 /*
1484  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1485  */
1486 static int unshare_thread(unsigned long unshare_flags)
1487 {
1488         if (unshare_flags & CLONE_THREAD)
1489                 return -EINVAL;
1490
1491         return 0;
1492 }
1493
1494 /*
1495  * Unshare the filesystem structure if it is being shared
1496  */
1497 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1498 {
1499         struct fs_struct *fs = current->fs;
1500
1501         if ((unshare_flags & CLONE_FS) &&
1502             (fs && atomic_read(&fs->count) > 1)) {
1503                 *new_fsp = __copy_fs_struct(current->fs);
1504                 if (!*new_fsp)
1505                         return -ENOMEM;
1506         }
1507
1508         return 0;
1509 }
1510
1511 /*
1512  * Unshare the namespace structure if it is being shared
1513  */
1514 static int unshare_namespace(unsigned long unshare_flags, struct namespace **new_nsp, struct fs_struct *new_fs)
1515 {
1516         struct namespace *ns = current->namespace;
1517
1518         if ((unshare_flags & CLONE_NEWNS) &&
1519             (ns && atomic_read(&ns->count) > 1)) {
1520                 if (!capable(CAP_SYS_ADMIN))
1521                         return -EPERM;
1522
1523                 *new_nsp = dup_namespace(current, new_fs ? new_fs : current->fs);
1524                 if (!*new_nsp)
1525                         return -ENOMEM;
1526         }
1527
1528         return 0;
1529 }
1530
1531 /*
1532  * Unsharing of sighand for tasks created with CLONE_SIGHAND is not
1533  * supported yet
1534  */
1535 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1536 {
1537         struct sighand_struct *sigh = current->sighand;
1538
1539         if ((unshare_flags & CLONE_SIGHAND) &&
1540             (sigh && atomic_read(&sigh->count) > 1))
1541                 return -EINVAL;
1542         else
1543                 return 0;
1544 }
1545
1546 /*
1547  * Unshare vm if it is being shared
1548  */
1549 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1550 {
1551         struct mm_struct *mm = current->mm;
1552
1553         if ((unshare_flags & CLONE_VM) &&
1554             (mm && atomic_read(&mm->mm_users) > 1)) {
1555                 return -EINVAL;
1556         }
1557
1558         return 0;
1559 }
1560
1561 /*
1562  * Unshare file descriptor table if it is being shared
1563  */
1564 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1565 {
1566         struct files_struct *fd = current->files;
1567         int error = 0;
1568
1569         if ((unshare_flags & CLONE_FILES) &&
1570             (fd && atomic_read(&fd->count) > 1)) {
1571                 *new_fdp = dup_fd(fd, &error);
1572                 if (!*new_fdp)
1573                         return error;
1574         }
1575
1576         return 0;
1577 }
1578
1579 /*
1580  * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1581  * supported yet
1582  */
1583 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1584 {
1585         if (unshare_flags & CLONE_SYSVSEM)
1586                 return -EINVAL;
1587
1588         return 0;
1589 }
1590
1591 /*
1592  * unshare allows a process to 'unshare' part of the process
1593  * context which was originally shared using clone.  copy_*
1594  * functions used by do_fork() cannot be used here directly
1595  * because they modify an inactive task_struct that is being
1596  * constructed. Here we are modifying the current, active,
1597  * task_struct.
1598  */
1599 asmlinkage long sys_unshare(unsigned long unshare_flags)
1600 {
1601         int err = 0;
1602         struct fs_struct *fs, *new_fs = NULL;
1603         struct namespace *ns, *new_ns = NULL;
1604         struct sighand_struct *sigh, *new_sigh = NULL;
1605         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1606         struct files_struct *fd, *new_fd = NULL;
1607         struct sem_undo_list *new_ulist = NULL;
1608
1609         check_unshare_flags(&unshare_flags);
1610
1611         /* Return -EINVAL for all unsupported flags */
1612         err = -EINVAL;
1613         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1614                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM))
1615                 goto bad_unshare_out;
1616
1617         if ((err = unshare_thread(unshare_flags)))
1618                 goto bad_unshare_out;
1619         if ((err = unshare_fs(unshare_flags, &new_fs)))
1620                 goto bad_unshare_cleanup_thread;
1621         if ((err = unshare_namespace(unshare_flags, &new_ns, new_fs)))
1622                 goto bad_unshare_cleanup_fs;
1623         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1624                 goto bad_unshare_cleanup_ns;
1625         if ((err = unshare_vm(unshare_flags, &new_mm)))
1626                 goto bad_unshare_cleanup_sigh;
1627         if ((err = unshare_fd(unshare_flags, &new_fd)))
1628                 goto bad_unshare_cleanup_vm;
1629         if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1630                 goto bad_unshare_cleanup_fd;
1631
1632         if (new_fs || new_ns || new_sigh || new_mm || new_fd || new_ulist) {
1633
1634                 task_lock(current);
1635
1636                 if (new_fs) {
1637                         fs = current->fs;
1638                         current->fs = new_fs;
1639                         new_fs = fs;
1640                 }
1641
1642                 if (new_ns) {
1643                         ns = current->namespace;
1644                         current->namespace = new_ns;
1645                         new_ns = ns;
1646                 }
1647
1648                 if (new_sigh) {
1649                         sigh = current->sighand;
1650                         rcu_assign_pointer(current->sighand, new_sigh);
1651                         new_sigh = sigh;
1652                 }
1653
1654                 if (new_mm) {
1655                         mm = current->mm;
1656                         active_mm = current->active_mm;
1657                         current->mm = new_mm;
1658                         current->active_mm = new_mm;
1659                         activate_mm(active_mm, new_mm);
1660                         new_mm = mm;
1661                 }
1662
1663                 if (new_fd) {
1664                         fd = current->files;
1665                         current->files = new_fd;
1666                         new_fd = fd;
1667                 }
1668
1669                 task_unlock(current);
1670         }
1671
1672 bad_unshare_cleanup_fd:
1673         if (new_fd)
1674                 put_files_struct(new_fd);
1675
1676 bad_unshare_cleanup_vm:
1677         if (new_mm)
1678                 mmput(new_mm);
1679
1680 bad_unshare_cleanup_sigh:
1681         if (new_sigh)
1682                 if (atomic_dec_and_test(&new_sigh->count))
1683                         kmem_cache_free(sighand_cachep, new_sigh);
1684
1685 bad_unshare_cleanup_ns:
1686         if (new_ns)
1687                 put_namespace(new_ns);
1688
1689 bad_unshare_cleanup_fs:
1690         if (new_fs)
1691                 put_fs_struct(new_fs);
1692
1693 bad_unshare_cleanup_thread:
1694 bad_unshare_out:
1695         return err;
1696 }