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/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/mnt_namespace.h>
  21 #include <linux/personality.h>
  22 #include <linux/mempolicy.h>
  23 #include <linux/sem.h>
  24 #include <linux/file.h>
  25 #include <linux/key.h>
  26 #include <linux/binfmts.h>
  27 #include <linux/mman.h>
  28 #include <linux/fs.h>
  29 #include <linux/nsproxy.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/task_io_accounting_ops.h>
  39 #include <linux/rcupdate.h>
  40 #include <linux/ptrace.h>
  41 #include <linux/mount.h>
  42 #include <linux/audit.h>
  43 #include <linux/profile.h>
  44 #include <linux/rmap.h>
  45 #include <linux/acct.h>
  46 #include <linux/tsacct_kern.h>
  47 #include <linux/cn_proc.h>
  48 #include <linux/delayacct.h>
  49 #include <linux/taskstats_kern.h>
  50 #include <linux/random.h>
  51
  52 #include <asm/pgtable.h>
  53 #include <asm/pgalloc.h>
  54 #include <asm/uaccess.h>
  55 #include <asm/mmu_context.h>
  56 #include <asm/cacheflush.h>
  57 #include <asm/tlbflush.h>
  58
  59 /*
  60  * Protected counters by write_lock_irq(&tasklist_lock)
  61  */
  62 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  63 int nr_threads;                 /* The idle threads do not count.. */
  64
  65 int max_threads;                /* tunable limit on nr_threads */
  66
  67 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  68
  69 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  70
  71 int nr_processes(void)
  72 {
  73         int cpu;
  74         int total = 0;
  75
  76         for_each_online_cpu(cpu)
  77                 total += per_cpu(process_counts, cpu);
  78
  79         return total;
  80 }
  81
  82 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  83 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  84 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  85 static struct kmem_cache *task_struct_cachep;
  86 #endif
  87
  88 /* SLAB cache for signal_struct structures (tsk->signal) */
  89 static struct kmem_cache *signal_cachep;
  90
  91 /* SLAB cache for sighand_struct structures (tsk->sighand) */
  92 struct kmem_cache *sighand_cachep;
  93
  94 /* SLAB cache for files_struct structures (tsk->files) */
  95 struct kmem_cache *files_cachep;
  96
  97 /* SLAB cache for fs_struct structures (tsk->fs) */
  98 struct kmem_cache *fs_cachep;
  99
 100 /* SLAB cache for vm_area_struct structures */
 101 struct kmem_cache *vm_area_cachep;
 102
 103 /* SLAB cache for mm_struct structures (tsk->mm) */
 104 static struct kmem_cache *mm_cachep;
 105
 106 void free_task(struct task_struct *tsk)
 107 {
 108         free_thread_info(tsk->thread_info);
 109         rt_mutex_debug_task_free(tsk);
 110         free_task_struct(tsk);
 111 }
 112 EXPORT_SYMBOL(free_task);
 113
 114 void __put_task_struct(struct task_struct *tsk)
 115 {
 116         WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
 117         WARN_ON(atomic_read(&tsk->usage));
 118         WARN_ON(tsk == current);
 119
 120         security_task_free(tsk);
 121         free_uid(tsk->user);
 122         put_group_info(tsk->group_info);
 123         delayacct_tsk_free(tsk);
 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         *tsk = *orig;
 178         tsk->thread_info = ti;
 179         setup_thread_stack(tsk, orig);
 180
 181 #ifdef CONFIG_CC_STACKPROTECTOR
 182         tsk->stack_canary = get_random_int();
 183 #endif
 184
 185         /* One for us, one for whoever does the "release_task()" (usually parent) */
 186         atomic_set(&tsk->usage,2);
 187         atomic_set(&tsk->fs_excl, 0);
 188 #ifdef CONFIG_BLK_DEV_IO_TRACE
 189         tsk->btrace_seq = 0;
 190 #endif
 191         tsk->splice_pipe = NULL;
 192         return tsk;
 193 }
 194
 195 #ifdef CONFIG_MMU
 196 static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 197 {
 198         struct vm_area_struct *mpnt, *tmp, **pprev;
 199         struct rb_node **rb_link, *rb_parent;
 200         int retval;
 201         unsigned long charge;
 202         struct mempolicy *pol;
 203
 204         down_write(&oldmm->mmap_sem);
 205         flush_cache_dup_mm(oldmm);
 206         /*
 207          * Not linked in yet - no deadlock potential:
 208          */
 209         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 210
 211         mm->locked_vm = 0;
 212         mm->mmap = NULL;
 213         mm->mmap_cache = NULL;
 214         mm->free_area_cache = oldmm->mmap_base;
 215         mm->cached_hole_size = ~0UL;
 216         mm->map_count = 0;
 217         cpus_clear(mm->cpu_vm_mask);
 218         mm->mm_rb = RB_ROOT;
 219         rb_link = &mm->mm_rb.rb_node;
 220         rb_parent = NULL;
 221         pprev = &mm->mmap;
 222
 223         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 224                 struct file *file;
 225
 226                 if (mpnt->vm_flags & VM_DONTCOPY) {
 227                         long pages = vma_pages(mpnt);
 228                         mm->total_vm -= pages;
 229                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 230                                                                 -pages);
 231                         continue;
 232                 }
 233                 charge = 0;
 234                 if (mpnt->vm_flags & VM_ACCOUNT) {
 235                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 236                         if (security_vm_enough_memory(len))
 237                                 goto fail_nomem;
 238                         charge = len;
 239                 }
 240                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 241                 if (!tmp)
 242                         goto fail_nomem;
 243                 *tmp = *mpnt;
 244                 pol = mpol_copy(vma_policy(mpnt));
 245                 retval = PTR_ERR(pol);
 246                 if (IS_ERR(pol))
 247                         goto fail_nomem_policy;
 248                 vma_set_policy(tmp, pol);
 249                 tmp->vm_flags &= ~VM_LOCKED;
 250                 tmp->vm_mm = mm;
 251                 tmp->vm_next = NULL;
 252                 anon_vma_link(tmp);
 253                 file = tmp->vm_file;
 254                 if (file) {
 255                         struct inode *inode = file->f_path.dentry->d_inode;
 256                         get_file(file);
 257                         if (tmp->vm_flags & VM_DENYWRITE)
 258                                 atomic_dec(&inode->i_writecount);
 259
 260                         /* insert tmp into the share list, just after mpnt */
 261                         spin_lock(&file->f_mapping->i_mmap_lock);
 262                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 263                         flush_dcache_mmap_lock(file->f_mapping);
 264                         vma_prio_tree_add(tmp, mpnt);
 265                         flush_dcache_mmap_unlock(file->f_mapping);
 266                         spin_unlock(&file->f_mapping->i_mmap_lock);
 267                 }
 268
 269                 /*
 270                  * Link in the new vma and copy the page table entries.
 271                  */
 272                 *pprev = tmp;
 273                 pprev = &tmp->vm_next;
 274
 275                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 276                 rb_link = &tmp->vm_rb.rb_right;
 277                 rb_parent = &tmp->vm_rb;
 278
 279                 mm->map_count++;
 280                 retval = copy_page_range(mm, oldmm, mpnt);
 281
 282                 if (tmp->vm_ops && tmp->vm_ops->open)
 283                         tmp->vm_ops->open(tmp);
 284
 285                 if (retval)
 286                         goto out;
 287         }
 288         /* a new mm has just been created */
 289         arch_dup_mmap(oldmm, mm);
 290         retval = 0;
 291 out:
 292         up_write(&mm->mmap_sem);
 293         flush_tlb_mm(oldmm);
 294         up_write(&oldmm->mmap_sem);
 295         return retval;
 296 fail_nomem_policy:
 297         kmem_cache_free(vm_area_cachep, tmp);
 298 fail_nomem:
 299         retval = -ENOMEM;
 300         vm_unacct_memory(charge);
 301         goto out;
 302 }
 303
 304 static inline int mm_alloc_pgd(struct mm_struct * mm)
 305 {
 306         mm->pgd = pgd_alloc(mm);
 307         if (unlikely(!mm->pgd))
 308                 return -ENOMEM;
 309         return 0;
 310 }
 311
 312 static inline void mm_free_pgd(struct mm_struct * mm)
 313 {
 314         pgd_free(mm->pgd);
 315 }
 316 #else
 317 #define dup_mmap(mm, oldmm)     (0)
 318 #define mm_alloc_pgd(mm)        (0)
 319 #define mm_free_pgd(mm)
 320 #endif /* CONFIG_MMU */
 321
 322  __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 323
 324 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 325 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 326
 327 #include <linux/init_task.h>
 328
 329 static struct mm_struct * mm_init(struct mm_struct * mm)
 330 {
 331         atomic_set(&mm->mm_users, 1);
 332         atomic_set(&mm->mm_count, 1);
 333         init_rwsem(&mm->mmap_sem);
 334         INIT_LIST_HEAD(&mm->mmlist);
 335         mm->core_waiters = 0;
 336         mm->nr_ptes = 0;
 337         set_mm_counter(mm, file_rss, 0);
 338         set_mm_counter(mm, anon_rss, 0);
 339         spin_lock_init(&mm->page_table_lock);
 340         rwlock_init(&mm->ioctx_list_lock);
 341         mm->ioctx_list = NULL;
 342         mm->free_area_cache = TASK_UNMAPPED_BASE;
 343         mm->cached_hole_size = ~0UL;
 344
 345         if (likely(!mm_alloc_pgd(mm))) {
 346                 mm->def_flags = 0;
 347                 return mm;
 348         }
 349         free_mm(mm);
 350         return NULL;
 351 }
 352
 353 /*
 354  * Allocate and initialize an mm_struct.
 355  */
 356 struct mm_struct * mm_alloc(void)
 357 {
 358         struct mm_struct * mm;
 359
 360         mm = allocate_mm();
 361         if (mm) {
 362                 memset(mm, 0, sizeof(*mm));
 363                 mm = mm_init(mm);
 364         }
 365         return mm;
 366 }
 367
 368 /*
 369  * Called when the last reference to the mm
 370  * is dropped: either by a lazy thread or by
 371  * mmput. Free the page directory and the mm.
 372  */
 373 void fastcall __mmdrop(struct mm_struct *mm)
 374 {
 375         BUG_ON(mm == &init_mm);
 376         mm_free_pgd(mm);
 377         destroy_context(mm);
 378         free_mm(mm);
 379 }
 380
 381 /*
 382  * Decrement the use count and release all resources for an mm.
 383  */
 384 void mmput(struct mm_struct *mm)
 385 {
 386         might_sleep();
 387
 388         if (atomic_dec_and_test(&mm->mm_users)) {
 389                 exit_aio(mm);
 390                 exit_mmap(mm);
 391                 if (!list_empty(&mm->mmlist)) {
 392                         spin_lock(&mmlist_lock);
 393                         list_del(&mm->mmlist);
 394                         spin_unlock(&mmlist_lock);
 395                 }
 396                 put_swap_token(mm);
 397                 mmdrop(mm);
 398         }
 399 }
 400 EXPORT_SYMBOL_GPL(mmput);
 401
 402 /**
 403  * get_task_mm - acquire a reference to the task's mm
 404  *
 405  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
 406  * this kernel workthread has transiently adopted a user mm with use_mm,
 407  * to do its AIO) is not set and if so returns a reference to it, after
 408  * bumping up the use count.  User must release the mm via mmput()
 409  * after use.  Typically used by /proc and ptrace.
 410  */
 411 struct mm_struct *get_task_mm(struct task_struct *task)
 412 {
 413         struct mm_struct *mm;
 414
 415         task_lock(task);
 416         mm = task->mm;
 417         if (mm) {
 418                 if (task->flags & PF_BORROWED_MM)
 419                         mm = NULL;
 420                 else
 421                         atomic_inc(&mm->mm_users);
 422         }
 423         task_unlock(task);
 424         return mm;
 425 }
 426 EXPORT_SYMBOL_GPL(get_task_mm);
 427
 428 /* Please note the differences between mmput and mm_release.
 429  * mmput is called whenever we stop holding onto a mm_struct,
 430  * error success whatever.
 431  *
 432  * mm_release is called after a mm_struct has been removed
 433  * from the current process.
 434  *
 435  * This difference is important for error handling, when we
 436  * only half set up a mm_struct for a new process and need to restore
 437  * the old one.  Because we mmput the new mm_struct before
 438  * restoring the old one. . .
 439  * Eric Biederman 10 January 1998
 440  */
 441 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 442 {
 443         struct completion *vfork_done = tsk->vfork_done;
 444
 445         /* Get rid of any cached register state */
 446         deactivate_mm(tsk, mm);
 447
 448         /* notify parent sleeping on vfork() */
 449         if (vfork_done) {
 450                 tsk->vfork_done = NULL;
 451                 complete(vfork_done);
 452         }
 453
 454         /*
 455          * If we're exiting normally, clear a user-space tid field if
 456          * requested.  We leave this alone when dying by signal, to leave
 457          * the value intact in a core dump, and to save the unnecessary
 458          * trouble otherwise.  Userland only wants this done for a sys_exit.
 459          */
 460         if (tsk->clear_child_tid
 461             && !(tsk->flags & PF_SIGNALED)
 462             && atomic_read(&mm->mm_users) > 1) {
 463                 u32 __user * tidptr = tsk->clear_child_tid;
 464                 tsk->clear_child_tid = NULL;
 465
 466                 /*
 467                  * We don't check the error code - if userspace has
 468                  * not set up a proper pointer then tough luck.
 469                  */
 470                 put_user(0, tidptr);
 471                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 472         }
 473 }
 474
 475 /*
 476  * Allocate a new mm structure and copy contents from the
 477  * mm structure of the passed in task structure.
 478  */
 479 static struct mm_struct *dup_mm(struct task_struct *tsk)
 480 {
 481         struct mm_struct *mm, *oldmm = current->mm;
 482         int err;
 483
 484         if (!oldmm)
 485                 return NULL;
 486
 487         mm = allocate_mm();
 488         if (!mm)
 489                 goto fail_nomem;
 490
 491         memcpy(mm, oldmm, sizeof(*mm));
 492
 493         /* Initializing for Swap token stuff */
 494         mm->token_priority = 0;
 495         mm->last_interval = 0;
 496
 497         if (!mm_init(mm))
 498                 goto fail_nomem;
 499
 500         if (init_new_context(tsk, mm))
 501                 goto fail_nocontext;
 502
 503         err = dup_mmap(mm, oldmm);
 504         if (err)
 505                 goto free_pt;
 506
 507         mm->hiwater_rss = get_mm_rss(mm);
 508         mm->hiwater_vm = mm->total_vm;
 509
 510         return mm;
 511
 512 free_pt:
 513         mmput(mm);
 514
 515 fail_nomem:
 516         return NULL;
 517
 518 fail_nocontext:
 519         /*
 520          * If init_new_context() failed, we cannot use mmput() to free the mm
 521          * because it calls destroy_context()
 522          */
 523         mm_free_pgd(mm);
 524         free_mm(mm);
 525         return NULL;
 526 }
 527
 528 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 529 {
 530         struct mm_struct * mm, *oldmm;
 531         int retval;
 532
 533         tsk->min_flt = tsk->maj_flt = 0;
 534         tsk->nvcsw = tsk->nivcsw = 0;
 535
 536         tsk->mm = NULL;
 537         tsk->active_mm = NULL;
 538
 539         /*
 540          * Are we cloning a kernel thread?
 541          *
 542          * We need to steal a active VM for that..
 543          */
 544         oldmm = current->mm;
 545         if (!oldmm)
 546                 return 0;
 547
 548         if (clone_flags & CLONE_VM) {
 549                 atomic_inc(&oldmm->mm_users);
 550                 mm = oldmm;
 551                 goto good_mm;
 552         }
 553
 554         retval = -ENOMEM;
 555         mm = dup_mm(tsk);
 556         if (!mm)
 557                 goto fail_nomem;
 558
 559 good_mm:
 560         /* Initializing for Swap token stuff */
 561         mm->token_priority = 0;
 562         mm->last_interval = 0;
 563
 564         tsk->mm = mm;
 565         tsk->active_mm = mm;
 566         return 0;
 567
 568 fail_nomem:
 569         return retval;
 570 }
 571
 572 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 573 {
 574         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 575         /* We don't need to lock fs - think why ;-) */
 576         if (fs) {
 577                 atomic_set(&fs->count, 1);
 578                 rwlock_init(&fs->lock);
 579                 fs->umask = old->umask;
 580                 read_lock(&old->lock);
 581                 fs->rootmnt = mntget(old->rootmnt);
 582                 fs->root = dget(old->root);
 583                 fs->pwdmnt = mntget(old->pwdmnt);
 584                 fs->pwd = dget(old->pwd);
 585                 if (old->altroot) {
 586                         fs->altrootmnt = mntget(old->altrootmnt);
 587                         fs->altroot = dget(old->altroot);
 588                 } else {
 589                         fs->altrootmnt = NULL;
 590                         fs->altroot = NULL;
 591                 }
 592                 read_unlock(&old->lock);
 593         }
 594         return fs;
 595 }
 596
 597 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 598 {
 599         return __copy_fs_struct(old);
 600 }
 601
 602 EXPORT_SYMBOL_GPL(copy_fs_struct);
 603
 604 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 605 {
 606         if (clone_flags & CLONE_FS) {
 607                 atomic_inc(&current->fs->count);
 608                 return 0;
 609         }
 610         tsk->fs = __copy_fs_struct(current->fs);
 611         if (!tsk->fs)
 612                 return -ENOMEM;
 613         return 0;
 614 }
 615
 616 static int count_open_files(struct fdtable *fdt)
 617 {
 618         int size = fdt->max_fds;
 619         int i;
 620
 621         /* Find the last open fd */
 622         for (i = size/(8*sizeof(long)); i > 0; ) {
 623                 if (fdt->open_fds->fds_bits[--i])
 624                         break;
 625         }
 626         i = (i+1) * 8 * sizeof(long);
 627         return i;
 628 }
 629
 630 static struct files_struct *alloc_files(void)
 631 {
 632         struct files_struct *newf;
 633         struct fdtable *fdt;
 634
 635         newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
 636         if (!newf)
 637                 goto out;
 638
 639         atomic_set(&newf->count, 1);
 640
 641         spin_lock_init(&newf->file_lock);
 642         newf->next_fd = 0;
 643         fdt = &newf->fdtab;
 644         fdt->max_fds = NR_OPEN_DEFAULT;
 645         fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
 646         fdt->open_fds = (fd_set *)&newf->open_fds_init;
 647         fdt->fd = &newf->fd_array[0];
 648         INIT_RCU_HEAD(&fdt->rcu);
 649         fdt->next = NULL;
 650         rcu_assign_pointer(newf->fdt, fdt);
 651 out:
 652         return newf;
 653 }
 654
 655 /*
 656  * Allocate a new files structure and copy contents from the
 657  * passed in files structure.
 658  * errorp will be valid only when the returned files_struct is NULL.
 659  */
 660 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
 661 {
 662         struct files_struct *newf;
 663         struct file **old_fds, **new_fds;
 664         int open_files, size, i;
 665         struct fdtable *old_fdt, *new_fdt;
 666
 667         *errorp = -ENOMEM;
 668         newf = alloc_files();
 669         if (!newf)
 670                 goto out;
 671
 672         spin_lock(&oldf->file_lock);
 673         old_fdt = files_fdtable(oldf);
 674         new_fdt = files_fdtable(newf);
 675         open_files = count_open_files(old_fdt);
 676
 677         /*
 678          * Check whether we need to allocate a larger fd array and fd set.
 679          * Note: we're not a clone task, so the open count won't change.
 680          */
 681         if (open_files > new_fdt->max_fds) {
 682                 new_fdt->max_fds = 0;
 683                 spin_unlock(&oldf->file_lock);
 684                 spin_lock(&newf->file_lock);
 685                 *errorp = expand_files(newf, open_files-1);
 686                 spin_unlock(&newf->file_lock);
 687                 if (*errorp < 0)
 688                         goto out_release;
 689                 new_fdt = files_fdtable(newf);
 690                 /*
 691                  * Reacquire the oldf lock and a pointer to its fd table
 692                  * who knows it may have a new bigger fd table. We need
 693                  * the latest pointer.
 694                  */
 695                 spin_lock(&oldf->file_lock);
 696                 old_fdt = files_fdtable(oldf);
 697         }
 698
 699         old_fds = old_fdt->fd;
 700         new_fds = new_fdt->fd;
 701
 702         memcpy(new_fdt->open_fds->fds_bits,
 703                 old_fdt->open_fds->fds_bits, open_files/8);
 704         memcpy(new_fdt->close_on_exec->fds_bits,
 705                 old_fdt->close_on_exec->fds_bits, open_files/8);
 706
 707         for (i = open_files; i != 0; i--) {
 708                 struct file *f = *old_fds++;
 709                 if (f) {
 710                         get_file(f);
 711                 } else {
 712                         /*
 713                          * The fd may be claimed in the fd bitmap but not yet
 714                          * instantiated in the files array if a sibling thread
 715                          * is partway through open().  So make sure that this
 716                          * fd is available to the new process.
 717                          */
 718                         FD_CLR(open_files - i, new_fdt->open_fds);
 719                 }
 720                 rcu_assign_pointer(*new_fds++, f);
 721         }
 722         spin_unlock(&oldf->file_lock);
 723
 724         /* compute the remainder to be cleared */
 725         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
 726
 727         /* This is long word aligned thus could use a optimized version */
 728         memset(new_fds, 0, size);
 729
 730         if (new_fdt->max_fds > open_files) {
 731                 int left = (new_fdt->max_fds-open_files)/8;
 732                 int start = open_files / (8 * sizeof(unsigned long));
 733
 734                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
 735                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
 736         }
 737
 738         return newf;
 739
 740 out_release:
 741         kmem_cache_free(files_cachep, newf);
 742 out:
 743         return NULL;
 744 }
 745
 746 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 747 {
 748         struct files_struct *oldf, *newf;
 749         int error = 0;
 750
 751         /*
 752          * A background process may not have any files ...
 753          */
 754         oldf = current->files;
 755         if (!oldf)
 756                 goto out;
 757
 758         if (clone_flags & CLONE_FILES) {
 759                 atomic_inc(&oldf->count);
 760                 goto out;
 761         }
 762
 763         /*
 764          * Note: we may be using current for both targets (See exec.c)
 765          * This works because we cache current->files (old) as oldf. Don't
 766          * break this.
 767          */
 768         tsk->files = NULL;
 769         newf = dup_fd(oldf, &error);
 770         if (!newf)
 771                 goto out;
 772
 773         tsk->files = newf;
 774         error = 0;
 775 out:
 776         return error;
 777 }
 778
 779 /*
 780  *      Helper to unshare the files of the current task.
 781  *      We don't want to expose copy_files internals to
 782  *      the exec layer of the kernel.
 783  */
 784
 785 int unshare_files(void)
 786 {
 787         struct files_struct *files  = current->files;
 788         int rc;
 789
 790         BUG_ON(!files);
 791
 792         /* This can race but the race causes us to copy when we don't
 793            need to and drop the copy */
 794         if(atomic_read(&files->count) == 1)
 795         {
 796                 atomic_inc(&files->count);
 797                 return 0;
 798         }
 799         rc = copy_files(0, current);
 800         if(rc)
 801                 current->files = files;
 802         return rc;
 803 }
 804
 805 EXPORT_SYMBOL(unshare_files);
 806
 807 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 808 {
 809         struct sighand_struct *sig;
 810
 811         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 812                 atomic_inc(&current->sighand->count);
 813                 return 0;
 814         }
 815         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 816         rcu_assign_pointer(tsk->sighand, sig);
 817         if (!sig)
 818                 return -ENOMEM;
 819         atomic_set(&sig->count, 1);
 820         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 821         return 0;
 822 }
 823
 824 void __cleanup_sighand(struct sighand_struct *sighand)
 825 {
 826         if (atomic_dec_and_test(&sighand->count))
 827                 kmem_cache_free(sighand_cachep, sighand);
 828 }
 829
 830 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 831 {
 832         struct signal_struct *sig;
 833         int ret;
 834
 835         if (clone_flags & CLONE_THREAD) {
 836                 atomic_inc(&current->signal->count);
 837                 atomic_inc(&current->signal->live);
 838                 return 0;
 839         }
 840         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 841         tsk->signal = sig;
 842         if (!sig)
 843                 return -ENOMEM;
 844
 845         ret = copy_thread_group_keys(tsk);
 846         if (ret < 0) {
 847                 kmem_cache_free(signal_cachep, sig);
 848                 return ret;
 849         }
 850
 851         atomic_set(&sig->count, 1);
 852         atomic_set(&sig->live, 1);
 853         init_waitqueue_head(&sig->wait_chldexit);
 854         sig->flags = 0;
 855         sig->group_exit_code = 0;
 856         sig->group_exit_task = NULL;
 857         sig->group_stop_count = 0;
 858         sig->curr_target = NULL;
 859         init_sigpending(&sig->shared_pending);
 860         INIT_LIST_HEAD(&sig->posix_timers);
 861
 862         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 863         sig->it_real_incr.tv64 = 0;
 864         sig->real_timer.function = it_real_fn;
 865         sig->tsk = tsk;
 866
 867         sig->it_virt_expires = cputime_zero;
 868         sig->it_virt_incr = cputime_zero;
 869         sig->it_prof_expires = cputime_zero;
 870         sig->it_prof_incr = cputime_zero;
 871
 872         sig->leader = 0;        /* session leadership doesn't inherit */
 873         sig->tty_old_pgrp = NULL;
 874
 875         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 876         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 877         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 878         sig->sched_time = 0;
 879         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 880         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 881         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 882         taskstats_tgid_init(sig);
 883
 884         task_lock(current->group_leader);
 885         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 886         task_unlock(current->group_leader);
 887
 888         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 889                 /*
 890                  * New sole thread in the process gets an expiry time
 891                  * of the whole CPU time limit.
 892                  */
 893                 tsk->it_prof_expires =
 894                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 895         }
 896         acct_init_pacct(&sig->pacct);
 897
 898         return 0;
 899 }
 900
 901 void __cleanup_signal(struct signal_struct *sig)
 902 {
 903         exit_thread_group_keys(sig);
 904         kmem_cache_free(signal_cachep, sig);
 905 }
 906
 907 static inline void cleanup_signal(struct task_struct *tsk)
 908 {
 909         struct signal_struct *sig = tsk->signal;
 910
 911         atomic_dec(&sig->live);
 912
 913         if (atomic_dec_and_test(&sig->count))
 914                 __cleanup_signal(sig);
 915 }
 916
 917 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 918 {
 919         unsigned long new_flags = p->flags;
 920
 921         new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
 922         new_flags |= PF_FORKNOEXEC;
 923         if (!(clone_flags & CLONE_PTRACE))
 924                 p->ptrace = 0;
 925         p->flags = new_flags;
 926 }
 927
 928 asmlinkage long sys_set_tid_address(int __user *tidptr)
 929 {
 930         current->clear_child_tid = tidptr;
 931
 932         return current->pid;
 933 }
 934
 935 static inline void rt_mutex_init_task(struct task_struct *p)
 936 {
 937         spin_lock_init(&p->pi_lock);
 938 #ifdef CONFIG_RT_MUTEXES
 939         plist_head_init(&p->pi_waiters, &p->pi_lock);
 940         p->pi_blocked_on = NULL;
 941 #endif
 942 }
 943
 944 /*
 945  * This creates a new process as a copy of the old one,
 946  * but does not actually start it yet.
 947  *
 948  * It copies the registers, and all the appropriate
 949  * parts of the process environment (as per the clone
 950  * flags). The actual kick-off is left to the caller.
 951  */
 952 static struct task_struct *copy_process(unsigned long clone_flags,
 953                                         unsigned long stack_start,
 954                                         struct pt_regs *regs,
 955                                         unsigned long stack_size,
 956                                         int __user *parent_tidptr,
 957                                         int __user *child_tidptr,
 958                                         int pid)
 959 {
 960         int retval;
 961         struct task_struct *p = NULL;
 962
 963         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 964                 return ERR_PTR(-EINVAL);
 965
 966         /*
 967          * Thread groups must share signals as well, and detached threads
 968          * can only be started up within the thread group.
 969          */
 970         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 971                 return ERR_PTR(-EINVAL);
 972
 973         /*
 974          * Shared signal handlers imply shared VM. By way of the above,
 975          * thread groups also imply shared VM. Blocking this case allows
 976          * for various simplifications in other code.
 977          */
 978         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 979                 return ERR_PTR(-EINVAL);
 980
 981         retval = security_task_create(clone_flags);
 982         if (retval)
 983                 goto fork_out;
 984
 985         retval = -ENOMEM;
 986         p = dup_task_struct(current);
 987         if (!p)
 988                 goto fork_out;
 989
 990         rt_mutex_init_task(p);
 991
 992 #ifdef CONFIG_TRACE_IRQFLAGS
 993         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
 994         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
 995 #endif
 996         retval = -EAGAIN;
 997         if (atomic_read(&p->user->processes) >=
 998                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
 999                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1000                                 p->user != &root_user)
1001                         goto bad_fork_free;
1002         }
1003
1004         atomic_inc(&p->user->__count);
1005         atomic_inc(&p->user->processes);
1006         get_group_info(p->group_info);
1007
1008         /*
1009          * If multiple threads are within copy_process(), then this check
1010          * triggers too late. This doesn't hurt, the check is only there
1011          * to stop root fork bombs.
1012          */
1013         if (nr_threads >= max_threads)
1014                 goto bad_fork_cleanup_count;
1015
1016         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1017                 goto bad_fork_cleanup_count;
1018
1019         if (p->binfmt && !try_module_get(p->binfmt->module))
1020                 goto bad_fork_cleanup_put_domain;
1021
1022         p->did_exec = 0;
1023         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1024         copy_flags(clone_flags, p);
1025         p->pid = pid;
1026         retval = -EFAULT;
1027         if (clone_flags & CLONE_PARENT_SETTID)
1028                 if (put_user(p->pid, parent_tidptr))
1029                         goto bad_fork_cleanup_delays_binfmt;
1030
1031         INIT_LIST_HEAD(&p->children);
1032         INIT_LIST_HEAD(&p->sibling);
1033         p->vfork_done = NULL;
1034         spin_lock_init(&p->alloc_lock);
1035
1036         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1037         init_sigpending(&p->pending);
1038
1039         p->utime = cputime_zero;
1040         p->stime = cputime_zero;
1041         p->sched_time = 0;
1042 #ifdef CONFIG_TASK_XACCT
1043         p->rchar = 0;           /* I/O counter: bytes read */
1044         p->wchar = 0;           /* I/O counter: bytes written */
1045         p->syscr = 0;           /* I/O counter: read syscalls */
1046         p->syscw = 0;           /* I/O counter: write syscalls */
1047 #endif
1048         task_io_accounting_init(p);
1049         acct_clear_integrals(p);
1050
1051         p->it_virt_expires = cputime_zero;
1052         p->it_prof_expires = cputime_zero;
1053         p->it_sched_expires = 0;
1054         INIT_LIST_HEAD(&p->cpu_timers[0]);
1055         INIT_LIST_HEAD(&p->cpu_timers[1]);
1056         INIT_LIST_HEAD(&p->cpu_timers[2]);
1057
1058         p->lock_depth = -1;             /* -1 = no lock */
1059         do_posix_clock_monotonic_gettime(&p->start_time);
1060         p->security = NULL;
1061         p->io_context = NULL;
1062         p->io_wait = NULL;
1063         p->audit_context = NULL;
1064         cpuset_fork(p);
1065 #ifdef CONFIG_NUMA
1066         p->mempolicy = mpol_copy(p->mempolicy);
1067         if (IS_ERR(p->mempolicy)) {
1068                 retval = PTR_ERR(p->mempolicy);
1069                 p->mempolicy = NULL;
1070                 goto bad_fork_cleanup_cpuset;
1071         }
1072         mpol_fix_fork_child_flag(p);
1073 #endif
1074 #ifdef CONFIG_TRACE_IRQFLAGS
1075         p->irq_events = 0;
1076 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1077         p->hardirqs_enabled = 1;
1078 #else
1079         p->hardirqs_enabled = 0;
1080 #endif
1081         p->hardirq_enable_ip = 0;
1082         p->hardirq_enable_event = 0;
1083         p->hardirq_disable_ip = _THIS_IP_;
1084         p->hardirq_disable_event = 0;
1085         p->softirqs_enabled = 1;
1086         p->softirq_enable_ip = _THIS_IP_;
1087         p->softirq_enable_event = 0;
1088         p->softirq_disable_ip = 0;
1089         p->softirq_disable_event = 0;
1090         p->hardirq_context = 0;
1091         p->softirq_context = 0;
1092 #endif
1093 #ifdef CONFIG_LOCKDEP
1094         p->lockdep_depth = 0; /* no locks held yet */
1095         p->curr_chain_key = 0;
1096         p->lockdep_recursion = 0;
1097 #endif
1098
1099 #ifdef CONFIG_DEBUG_MUTEXES
1100         p->blocked_on = NULL; /* not blocked yet */
1101 #endif
1102
1103         p->tgid = p->pid;
1104         if (clone_flags & CLONE_THREAD)
1105                 p->tgid = current->tgid;
1106
1107         if ((retval = security_task_alloc(p)))
1108                 goto bad_fork_cleanup_policy;
1109         if ((retval = audit_alloc(p)))
1110                 goto bad_fork_cleanup_security;
1111         /* copy all the process information */
1112         if ((retval = copy_semundo(clone_flags, p)))
1113                 goto bad_fork_cleanup_audit;
1114         if ((retval = copy_files(clone_flags, p)))
1115                 goto bad_fork_cleanup_semundo;
1116         if ((retval = copy_fs(clone_flags, p)))
1117                 goto bad_fork_cleanup_files;
1118         if ((retval = copy_sighand(clone_flags, p)))
1119                 goto bad_fork_cleanup_fs;
1120         if ((retval = copy_signal(clone_flags, p)))
1121                 goto bad_fork_cleanup_sighand;
1122         if ((retval = copy_mm(clone_flags, p)))
1123                 goto bad_fork_cleanup_signal;
1124         if ((retval = copy_keys(clone_flags, p)))
1125                 goto bad_fork_cleanup_mm;
1126         if ((retval = copy_namespaces(clone_flags, p)))
1127                 goto bad_fork_cleanup_keys;
1128         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1129         if (retval)
1130                 goto bad_fork_cleanup_namespaces;
1131
1132         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1133         /*
1134          * Clear TID on mm_release()?
1135          */
1136         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1137         p->robust_list = NULL;
1138 #ifdef CONFIG_COMPAT
1139         p->compat_robust_list = NULL;
1140 #endif
1141         INIT_LIST_HEAD(&p->pi_state_list);
1142         p->pi_state_cache = NULL;
1143
1144         /*
1145          * sigaltstack should be cleared when sharing the same VM
1146          */
1147         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1148                 p->sas_ss_sp = p->sas_ss_size = 0;
1149
1150         /*
1151          * Syscall tracing should be turned off in the child regardless
1152          * of CLONE_PTRACE.
1153          */
1154         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1155 #ifdef TIF_SYSCALL_EMU
1156         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1157 #endif
1158
1159         /* Our parent execution domain becomes current domain
1160            These must match for thread signalling to apply */
1161         p->parent_exec_id = p->self_exec_id;
1162
1163         /* ok, now we should be set up.. */
1164         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1165         p->pdeath_signal = 0;
1166         p->exit_state = 0;
1167
1168         /*
1169          * Ok, make it visible to the rest of the system.
1170          * We dont wake it up yet.
1171          */
1172         p->group_leader = p;
1173         INIT_LIST_HEAD(&p->thread_group);
1174         INIT_LIST_HEAD(&p->ptrace_children);
1175         INIT_LIST_HEAD(&p->ptrace_list);
1176
1177         /* Perform scheduler related setup. Assign this task to a CPU. */
1178         sched_fork(p, clone_flags);
1179
1180         /* Need tasklist lock for parent etc handling! */
1181         write_lock_irq(&tasklist_lock);
1182
1183         /* for sys_ioprio_set(IOPRIO_WHO_PGRP) */
1184         p->ioprio = current->ioprio;
1185
1186         /*
1187          * The task hasn't been attached yet, so its cpus_allowed mask will
1188          * not be changed, nor will its assigned CPU.
1189          *
1190          * The cpus_allowed mask of the parent may have changed after it was
1191          * copied first time - so re-copy it here, then check the child's CPU
1192          * to ensure it is on a valid CPU (and if not, just force it back to
1193          * parent's CPU). This avoids alot of nasty races.
1194          */
1195         p->cpus_allowed = current->cpus_allowed;
1196         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1197                         !cpu_online(task_cpu(p))))
1198                 set_task_cpu(p, smp_processor_id());
1199
1200         /* CLONE_PARENT re-uses the old parent */
1201         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1202                 p->real_parent = current->real_parent;
1203         else
1204                 p->real_parent = current;
1205         p->parent = p->real_parent;
1206
1207         spin_lock(&current->sighand->siglock);
1208
1209         /*
1210          * Process group and session signals need to be delivered to just the
1211          * parent before the fork or both the parent and the child after the
1212          * fork. Restart if a signal comes in before we add the new process to
1213          * it's process group.
1214          * A fatal signal pending means that current will exit, so the new
1215          * thread can't slip out of an OOM kill (or normal SIGKILL).
1216          */
1217         recalc_sigpending();
1218         if (signal_pending(current)) {
1219                 spin_unlock(&current->sighand->siglock);
1220                 write_unlock_irq(&tasklist_lock);
1221                 retval = -ERESTARTNOINTR;
1222                 goto bad_fork_cleanup_namespaces;
1223         }
1224
1225         if (clone_flags & CLONE_THREAD) {
1226                 p->group_leader = current->group_leader;
1227                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1228
1229                 if (!cputime_eq(current->signal->it_virt_expires,
1230                                 cputime_zero) ||
1231                     !cputime_eq(current->signal->it_prof_expires,
1232                                 cputime_zero) ||
1233                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1234                     !list_empty(&current->signal->cpu_timers[0]) ||
1235                     !list_empty(&current->signal->cpu_timers[1]) ||
1236                     !list_empty(&current->signal->cpu_timers[2])) {
1237                         /*
1238                          * Have child wake up on its first tick to check
1239                          * for process CPU timers.
1240                          */
1241                         p->it_prof_expires = jiffies_to_cputime(1);
1242                 }
1243         }
1244
1245         if (likely(p->pid)) {
1246                 add_parent(p);
1247                 if (unlikely(p->ptrace & PT_PTRACED))
1248                         __ptrace_link(p, current->parent);
1249
1250                 if (thread_group_leader(p)) {
1251                         p->signal->tty = current->signal->tty;
1252                         p->signal->pgrp = process_group(current);
1253                         set_signal_session(p->signal, process_session(current));
1254                         attach_pid(p, PIDTYPE_PGID, process_group(p));
1255                         attach_pid(p, PIDTYPE_SID, process_session(p));
1256
1257                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1258                         __get_cpu_var(process_counts)++;
1259                 }
1260                 attach_pid(p, PIDTYPE_PID, p->pid);
1261                 nr_threads++;
1262         }
1263
1264         total_forks++;
1265         spin_unlock(&current->sighand->siglock);
1266         write_unlock_irq(&tasklist_lock);
1267         proc_fork_connector(p);
1268         return p;
1269
1270 bad_fork_cleanup_namespaces:
1271         exit_task_namespaces(p);
1272 bad_fork_cleanup_keys:
1273         exit_keys(p);
1274 bad_fork_cleanup_mm:
1275         if (p->mm)
1276                 mmput(p->mm);
1277 bad_fork_cleanup_signal:
1278         cleanup_signal(p);
1279 bad_fork_cleanup_sighand:
1280         __cleanup_sighand(p->sighand);
1281 bad_fork_cleanup_fs:
1282         exit_fs(p); /* blocking */
1283 bad_fork_cleanup_files:
1284         exit_files(p); /* blocking */
1285 bad_fork_cleanup_semundo:
1286         exit_sem(p);
1287 bad_fork_cleanup_audit:
1288         audit_free(p);
1289 bad_fork_cleanup_security:
1290         security_task_free(p);
1291 bad_fork_cleanup_policy:
1292 #ifdef CONFIG_NUMA
1293         mpol_free(p->mempolicy);
1294 bad_fork_cleanup_cpuset:
1295 #endif
1296         cpuset_exit(p);
1297 bad_fork_cleanup_delays_binfmt:
1298         delayacct_tsk_free(p);
1299         if (p->binfmt)
1300                 module_put(p->binfmt->module);
1301 bad_fork_cleanup_put_domain:
1302         module_put(task_thread_info(p)->exec_domain->module);
1303 bad_fork_cleanup_count:
1304         put_group_info(p->group_info);
1305         atomic_dec(&p->user->processes);
1306         free_uid(p->user);
1307 bad_fork_free:
1308         free_task(p);
1309 fork_out:
1310         return ERR_PTR(retval);
1311 }
1312
1313 noinline struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1314 {
1315         memset(regs, 0, sizeof(struct pt_regs));
1316         return regs;
1317 }
1318
1319 struct task_struct * __cpuinit fork_idle(int cpu)
1320 {
1321         struct task_struct *task;
1322         struct pt_regs regs;
1323
1324         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1325         if (!IS_ERR(task))
1326                 init_idle(task, cpu);
1327
1328         return task;
1329 }
1330
1331 static inline int fork_traceflag (unsigned clone_flags)
1332 {
1333         if (clone_flags & CLONE_UNTRACED)
1334                 return 0;
1335         else if (clone_flags & CLONE_VFORK) {
1336                 if (current->ptrace & PT_TRACE_VFORK)
1337                         return PTRACE_EVENT_VFORK;
1338         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1339                 if (current->ptrace & PT_TRACE_CLONE)
1340                         return PTRACE_EVENT_CLONE;
1341         } else if (current->ptrace & PT_TRACE_FORK)
1342                 return PTRACE_EVENT_FORK;
1343
1344         return 0;
1345 }
1346
1347 /*
1348  *  Ok, this is the main fork-routine.
1349  *
1350  * It copies the process, and if successful kick-starts
1351  * it and waits for it to finish using the VM if required.
1352  */
1353 long do_fork(unsigned long clone_flags,
1354               unsigned long stack_start,
1355               struct pt_regs *regs,
1356               unsigned long stack_size,
1357               int __user *parent_tidptr,
1358               int __user *child_tidptr)
1359 {
1360         struct task_struct *p;
1361         int trace = 0;
1362         struct pid *pid = alloc_pid();
1363         long nr;
1364
1365         if (!pid)
1366                 return -EAGAIN;
1367         nr = pid->nr;
1368         if (unlikely(current->ptrace)) {
1369                 trace = fork_traceflag (clone_flags);
1370                 if (trace)
1371                         clone_flags |= CLONE_PTRACE;
1372         }
1373
1374         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, nr);
1375         /*
1376          * Do this prior waking up the new thread - the thread pointer
1377          * might get invalid after that point, if the thread exits quickly.
1378          */
1379         if (!IS_ERR(p)) {
1380                 struct completion vfork;
1381
1382                 if (clone_flags & CLONE_VFORK) {
1383                         p->vfork_done = &vfork;
1384                         init_completion(&vfork);
1385                 }
1386
1387                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1388                         /*
1389                          * We'll start up with an immediate SIGSTOP.
1390                          */
1391                         sigaddset(&p->pending.signal, SIGSTOP);
1392                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1393                 }
1394
1395                 if (!(clone_flags & CLONE_STOPPED))
1396                         wake_up_new_task(p, clone_flags);
1397                 else
1398                         p->state = TASK_STOPPED;
1399
1400                 if (unlikely (trace)) {
1401                         current->ptrace_message = nr;
1402                         ptrace_notify ((trace << 8) | SIGTRAP);
1403                 }
1404
1405                 if (clone_flags & CLONE_VFORK) {
1406                         wait_for_completion(&vfork);
1407                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1408                                 current->ptrace_message = nr;
1409                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1410                         }
1411                 }
1412         } else {
1413                 free_pid(pid);
1414                 nr = PTR_ERR(p);
1415         }
1416         return nr;
1417 }
1418
1419 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1420 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1421 #endif
1422
1423 static void sighand_ctor(void *data, struct kmem_cache *cachep, unsigned long flags)
1424 {
1425         struct sighand_struct *sighand = data;
1426
1427         if (flags & SLAB_CTOR_CONSTRUCTOR)
1428                 spin_lock_init(&sighand->siglock);
1429 }
1430
1431 void __init proc_caches_init(void)
1432 {
1433         sighand_cachep = kmem_cache_create("sighand_cache",
1434                         sizeof(struct sighand_struct), 0,
1435                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1436                         sighand_ctor, NULL);
1437         signal_cachep = kmem_cache_create("signal_cache",
1438                         sizeof(struct signal_struct), 0,
1439                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1440         files_cachep = kmem_cache_create("files_cache",
1441                         sizeof(struct files_struct), 0,
1442                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1443         fs_cachep = kmem_cache_create("fs_cache",
1444                         sizeof(struct fs_struct), 0,
1445                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1446         vm_area_cachep = kmem_cache_create("vm_area_struct",
1447                         sizeof(struct vm_area_struct), 0,
1448                         SLAB_PANIC, NULL, NULL);
1449         mm_cachep = kmem_cache_create("mm_struct",
1450                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1451                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1452 }
1453
1454
1455 /*
1456  * Check constraints on flags passed to the unshare system call and
1457  * force unsharing of additional process context as appropriate.
1458  */
1459 static inline void check_unshare_flags(unsigned long *flags_ptr)
1460 {
1461         /*
1462          * If unsharing a thread from a thread group, must also
1463          * unshare vm.
1464          */
1465         if (*flags_ptr & CLONE_THREAD)
1466                 *flags_ptr |= CLONE_VM;
1467
1468         /*
1469          * If unsharing vm, must also unshare signal handlers.
1470          */
1471         if (*flags_ptr & CLONE_VM)
1472                 *flags_ptr |= CLONE_SIGHAND;
1473
1474         /*
1475          * If unsharing signal handlers and the task was created
1476          * using CLONE_THREAD, then must unshare the thread
1477          */
1478         if ((*flags_ptr & CLONE_SIGHAND) &&
1479             (atomic_read(&current->signal->count) > 1))
1480                 *flags_ptr |= CLONE_THREAD;
1481
1482         /*
1483          * If unsharing namespace, must also unshare filesystem information.
1484          */
1485         if (*flags_ptr & CLONE_NEWNS)
1486                 *flags_ptr |= CLONE_FS;
1487 }
1488
1489 /*
1490  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1491  */
1492 static int unshare_thread(unsigned long unshare_flags)
1493 {
1494         if (unshare_flags & CLONE_THREAD)
1495                 return -EINVAL;
1496
1497         return 0;
1498 }
1499
1500 /*
1501  * Unshare the filesystem structure if it is being shared
1502  */
1503 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1504 {
1505         struct fs_struct *fs = current->fs;
1506
1507         if ((unshare_flags & CLONE_FS) &&
1508             (fs && atomic_read(&fs->count) > 1)) {
1509                 *new_fsp = __copy_fs_struct(current->fs);
1510                 if (!*new_fsp)
1511                         return -ENOMEM;
1512         }
1513
1514         return 0;
1515 }
1516
1517 /*
1518  * Unsharing of sighand is not supported yet
1519  */
1520 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1521 {
1522         struct sighand_struct *sigh = current->sighand;
1523
1524         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1525                 return -EINVAL;
1526         else
1527                 return 0;
1528 }
1529
1530 /*
1531  * Unshare vm if it is being shared
1532  */
1533 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1534 {
1535         struct mm_struct *mm = current->mm;
1536
1537         if ((unshare_flags & CLONE_VM) &&
1538             (mm && atomic_read(&mm->mm_users) > 1)) {
1539                 return -EINVAL;
1540         }
1541
1542         return 0;
1543 }
1544
1545 /*
1546  * Unshare file descriptor table if it is being shared
1547  */
1548 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1549 {
1550         struct files_struct *fd = current->files;
1551         int error = 0;
1552
1553         if ((unshare_flags & CLONE_FILES) &&
1554             (fd && atomic_read(&fd->count) > 1)) {
1555                 *new_fdp = dup_fd(fd, &error);
1556                 if (!*new_fdp)
1557                         return error;
1558         }
1559
1560         return 0;
1561 }
1562
1563 /*
1564  * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1565  * supported yet
1566  */
1567 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1568 {
1569         if (unshare_flags & CLONE_SYSVSEM)
1570                 return -EINVAL;
1571
1572         return 0;
1573 }
1574
1575 /*
1576  * unshare allows a process to 'unshare' part of the process
1577  * context which was originally shared using clone.  copy_*
1578  * functions used by do_fork() cannot be used here directly
1579  * because they modify an inactive task_struct that is being
1580  * constructed. Here we are modifying the current, active,
1581  * task_struct.
1582  */
1583 asmlinkage long sys_unshare(unsigned long unshare_flags)
1584 {
1585         int err = 0;
1586         struct fs_struct *fs, *new_fs = NULL;
1587         struct sighand_struct *new_sigh = NULL;
1588         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1589         struct files_struct *fd, *new_fd = NULL;
1590         struct sem_undo_list *new_ulist = NULL;
1591         struct nsproxy *new_nsproxy = NULL, *old_nsproxy = NULL;
1592
1593         check_unshare_flags(&unshare_flags);
1594
1595         /* Return -EINVAL for all unsupported flags */
1596         err = -EINVAL;
1597         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1598                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1599                                 CLONE_NEWUTS|CLONE_NEWIPC))
1600                 goto bad_unshare_out;
1601
1602         if ((err = unshare_thread(unshare_flags)))
1603                 goto bad_unshare_out;
1604         if ((err = unshare_fs(unshare_flags, &new_fs)))
1605                 goto bad_unshare_cleanup_thread;
1606         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1607                 goto bad_unshare_cleanup_fs;
1608         if ((err = unshare_vm(unshare_flags, &new_mm)))
1609                 goto bad_unshare_cleanup_sigh;
1610         if ((err = unshare_fd(unshare_flags, &new_fd)))
1611                 goto bad_unshare_cleanup_vm;
1612         if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1613                 goto bad_unshare_cleanup_fd;
1614         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1615                         new_fs)))
1616                 goto bad_unshare_cleanup_semundo;
1617
1618         if (new_fs ||  new_mm || new_fd || new_ulist || new_nsproxy) {
1619
1620                 task_lock(current);
1621
1622                 if (new_nsproxy) {
1623                         old_nsproxy = current->nsproxy;
1624                         current->nsproxy = new_nsproxy;
1625                         new_nsproxy = old_nsproxy;
1626                 }
1627
1628                 if (new_fs) {
1629                         fs = current->fs;
1630                         current->fs = new_fs;
1631                         new_fs = fs;
1632                 }
1633
1634                 if (new_mm) {
1635                         mm = current->mm;
1636                         active_mm = current->active_mm;
1637                         current->mm = new_mm;
1638                         current->active_mm = new_mm;
1639                         activate_mm(active_mm, new_mm);
1640                         new_mm = mm;
1641                 }
1642
1643                 if (new_fd) {
1644                         fd = current->files;
1645                         current->files = new_fd;
1646                         new_fd = fd;
1647                 }
1648
1649                 task_unlock(current);
1650         }
1651
1652         if (new_nsproxy)
1653                 put_nsproxy(new_nsproxy);
1654
1655 bad_unshare_cleanup_semundo:
1656 bad_unshare_cleanup_fd:
1657         if (new_fd)
1658                 put_files_struct(new_fd);
1659
1660 bad_unshare_cleanup_vm:
1661         if (new_mm)
1662                 mmput(new_mm);
1663
1664 bad_unshare_cleanup_sigh:
1665         if (new_sigh)
1666                 if (atomic_dec_and_test(&new_sigh->count))
1667                         kmem_cache_free(sighand_cachep, new_sigh);
1668
1669 bad_unshare_cleanup_fs:
1670         if (new_fs)
1671                 put_fs_struct(new_fs);
1672
1673 bad_unshare_cleanup_thread:
1674 bad_unshare_out:
1675         return err;
1676 }