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/nsproxy.h>
  31 #include <linux/capability.h>
  32 #include <linux/cpu.h>
  33 #include <linux/cpuset.h>
  34 #include <linux/security.h>
  35 #include <linux/swap.h>
  36 #include <linux/syscalls.h>
  37 #include <linux/jiffies.h>
  38 #include <linux/futex.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_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_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         retval = 0;
 289 out:
 290         up_write(&mm->mmap_sem);
 291         flush_tlb_mm(oldmm);
 292         up_write(&oldmm->mmap_sem);
 293         return retval;
 294 fail_nomem_policy:
 295         kmem_cache_free(vm_area_cachep, tmp);
 296 fail_nomem:
 297         retval = -ENOMEM;
 298         vm_unacct_memory(charge);
 299         goto out;
 300 }
 301
 302 static inline int mm_alloc_pgd(struct mm_struct * mm)
 303 {
 304         mm->pgd = pgd_alloc(mm);
 305         if (unlikely(!mm->pgd))
 306                 return -ENOMEM;
 307         return 0;
 308 }
 309
 310 static inline void mm_free_pgd(struct mm_struct * mm)
 311 {
 312         pgd_free(mm->pgd);
 313 }
 314 #else
 315 #define dup_mmap(mm, oldmm)     (0)
 316 #define mm_alloc_pgd(mm)        (0)
 317 #define mm_free_pgd(mm)
 318 #endif /* CONFIG_MMU */
 319
 320  __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 321
 322 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 323 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 324
 325 #include <linux/init_task.h>
 326
 327 static struct mm_struct * mm_init(struct mm_struct * mm)
 328 {
 329         atomic_set(&mm->mm_users, 1);
 330         atomic_set(&mm->mm_count, 1);
 331         init_rwsem(&mm->mmap_sem);
 332         INIT_LIST_HEAD(&mm->mmlist);
 333         mm->core_waiters = 0;
 334         mm->nr_ptes = 0;
 335         set_mm_counter(mm, file_rss, 0);
 336         set_mm_counter(mm, anon_rss, 0);
 337         spin_lock_init(&mm->page_table_lock);
 338         rwlock_init(&mm->ioctx_list_lock);
 339         mm->ioctx_list = NULL;
 340         mm->free_area_cache = TASK_UNMAPPED_BASE;
 341         mm->cached_hole_size = ~0UL;
 342
 343         if (likely(!mm_alloc_pgd(mm))) {
 344                 mm->def_flags = 0;
 345                 return mm;
 346         }
 347         free_mm(mm);
 348         return NULL;
 349 }
 350
 351 /*
 352  * Allocate and initialize an mm_struct.
 353  */
 354 struct mm_struct * mm_alloc(void)
 355 {
 356         struct mm_struct * mm;
 357
 358         mm = allocate_mm();
 359         if (mm) {
 360                 memset(mm, 0, sizeof(*mm));
 361                 mm = mm_init(mm);
 362         }
 363         return mm;
 364 }
 365
 366 /*
 367  * Called when the last reference to the mm
 368  * is dropped: either by a lazy thread or by
 369  * mmput. Free the page directory and the mm.
 370  */
 371 void fastcall __mmdrop(struct mm_struct *mm)
 372 {
 373         BUG_ON(mm == &init_mm);
 374         mm_free_pgd(mm);
 375         destroy_context(mm);
 376         free_mm(mm);
 377 }
 378
 379 /*
 380  * Decrement the use count and release all resources for an mm.
 381  */
 382 void mmput(struct mm_struct *mm)
 383 {
 384         might_sleep();
 385
 386         if (atomic_dec_and_test(&mm->mm_users)) {
 387                 exit_aio(mm);
 388                 exit_mmap(mm);
 389                 if (!list_empty(&mm->mmlist)) {
 390                         spin_lock(&mmlist_lock);
 391                         list_del(&mm->mmlist);
 392                         spin_unlock(&mmlist_lock);
 393                 }
 394                 put_swap_token(mm);
 395                 mmdrop(mm);
 396         }
 397 }
 398 EXPORT_SYMBOL_GPL(mmput);
 399
 400 /**
 401  * get_task_mm - acquire a reference to the task's mm
 402  *
 403  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
 404  * this kernel workthread has transiently adopted a user mm with use_mm,
 405  * to do its AIO) is not set and if so returns a reference to it, after
 406  * bumping up the use count.  User must release the mm via mmput()
 407  * after use.  Typically used by /proc and ptrace.
 408  */
 409 struct mm_struct *get_task_mm(struct task_struct *task)
 410 {
 411         struct mm_struct *mm;
 412
 413         task_lock(task);
 414         mm = task->mm;
 415         if (mm) {
 416                 if (task->flags & PF_BORROWED_MM)
 417                         mm = NULL;
 418                 else
 419                         atomic_inc(&mm->mm_users);
 420         }
 421         task_unlock(task);
 422         return mm;
 423 }
 424 EXPORT_SYMBOL_GPL(get_task_mm);
 425
 426 /* Please note the differences between mmput and mm_release.
 427  * mmput is called whenever we stop holding onto a mm_struct,
 428  * error success whatever.
 429  *
 430  * mm_release is called after a mm_struct has been removed
 431  * from the current process.
 432  *
 433  * This difference is important for error handling, when we
 434  * only half set up a mm_struct for a new process and need to restore
 435  * the old one.  Because we mmput the new mm_struct before
 436  * restoring the old one. . .
 437  * Eric Biederman 10 January 1998
 438  */
 439 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 440 {
 441         struct completion *vfork_done = tsk->vfork_done;
 442
 443         /* Get rid of any cached register state */
 444         deactivate_mm(tsk, mm);
 445
 446         /* notify parent sleeping on vfork() */
 447         if (vfork_done) {
 448                 tsk->vfork_done = NULL;
 449                 complete(vfork_done);
 450         }
 451         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
 452                 u32 __user * tidptr = tsk->clear_child_tid;
 453                 tsk->clear_child_tid = NULL;
 454
 455                 /*
 456                  * We don't check the error code - if userspace has
 457                  * not set up a proper pointer then tough luck.
 458                  */
 459                 put_user(0, tidptr);
 460                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 461         }
 462 }
 463
 464 /*
 465  * Allocate a new mm structure and copy contents from the
 466  * mm structure of the passed in task structure.
 467  */
 468 static struct mm_struct *dup_mm(struct task_struct *tsk)
 469 {
 470         struct mm_struct *mm, *oldmm = current->mm;
 471         int err;
 472
 473         if (!oldmm)
 474                 return NULL;
 475
 476         mm = allocate_mm();
 477         if (!mm)
 478                 goto fail_nomem;
 479
 480         memcpy(mm, oldmm, sizeof(*mm));
 481
 482         /* Initializing for Swap token stuff */
 483         mm->token_priority = 0;
 484         mm->last_interval = 0;
 485
 486         if (!mm_init(mm))
 487                 goto fail_nomem;
 488
 489         if (init_new_context(tsk, mm))
 490                 goto fail_nocontext;
 491
 492         err = dup_mmap(mm, oldmm);
 493         if (err)
 494                 goto free_pt;
 495
 496         mm->hiwater_rss = get_mm_rss(mm);
 497         mm->hiwater_vm = mm->total_vm;
 498
 499         return mm;
 500
 501 free_pt:
 502         mmput(mm);
 503
 504 fail_nomem:
 505         return NULL;
 506
 507 fail_nocontext:
 508         /*
 509          * If init_new_context() failed, we cannot use mmput() to free the mm
 510          * because it calls destroy_context()
 511          */
 512         mm_free_pgd(mm);
 513         free_mm(mm);
 514         return NULL;
 515 }
 516
 517 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 518 {
 519         struct mm_struct * mm, *oldmm;
 520         int retval;
 521
 522         tsk->min_flt = tsk->maj_flt = 0;
 523         tsk->nvcsw = tsk->nivcsw = 0;
 524
 525         tsk->mm = NULL;
 526         tsk->active_mm = NULL;
 527
 528         /*
 529          * Are we cloning a kernel thread?
 530          *
 531          * We need to steal a active VM for that..
 532          */
 533         oldmm = current->mm;
 534         if (!oldmm)
 535                 return 0;
 536
 537         if (clone_flags & CLONE_VM) {
 538                 atomic_inc(&oldmm->mm_users);
 539                 mm = oldmm;
 540                 goto good_mm;
 541         }
 542
 543         retval = -ENOMEM;
 544         mm = dup_mm(tsk);
 545         if (!mm)
 546                 goto fail_nomem;
 547
 548 good_mm:
 549         /* Initializing for Swap token stuff */
 550         mm->token_priority = 0;
 551         mm->last_interval = 0;
 552
 553         tsk->mm = mm;
 554         tsk->active_mm = mm;
 555         return 0;
 556
 557 fail_nomem:
 558         return retval;
 559 }
 560
 561 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 562 {
 563         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 564         /* We don't need to lock fs - think why ;-) */
 565         if (fs) {
 566                 atomic_set(&fs->count, 1);
 567                 rwlock_init(&fs->lock);
 568                 fs->umask = old->umask;
 569                 read_lock(&old->lock);
 570                 fs->rootmnt = mntget(old->rootmnt);
 571                 fs->root = dget(old->root);
 572                 fs->pwdmnt = mntget(old->pwdmnt);
 573                 fs->pwd = dget(old->pwd);
 574                 if (old->altroot) {
 575                         fs->altrootmnt = mntget(old->altrootmnt);
 576                         fs->altroot = dget(old->altroot);
 577                 } else {
 578                         fs->altrootmnt = NULL;
 579                         fs->altroot = NULL;
 580                 }
 581                 read_unlock(&old->lock);
 582         }
 583         return fs;
 584 }
 585
 586 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 587 {
 588         return __copy_fs_struct(old);
 589 }
 590
 591 EXPORT_SYMBOL_GPL(copy_fs_struct);
 592
 593 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 594 {
 595         if (clone_flags & CLONE_FS) {
 596                 atomic_inc(&current->fs->count);
 597                 return 0;
 598         }
 599         tsk->fs = __copy_fs_struct(current->fs);
 600         if (!tsk->fs)
 601                 return -ENOMEM;
 602         return 0;
 603 }
 604
 605 static int count_open_files(struct fdtable *fdt)
 606 {
 607         int size = fdt->max_fdset;
 608         int i;
 609
 610         /* Find the last open fd */
 611         for (i = size/(8*sizeof(long)); i > 0; ) {
 612                 if (fdt->open_fds->fds_bits[--i])
 613                         break;
 614         }
 615         i = (i+1) * 8 * sizeof(long);
 616         return i;
 617 }
 618
 619 static struct files_struct *alloc_files(void)
 620 {
 621         struct files_struct *newf;
 622         struct fdtable *fdt;
 623
 624         newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
 625         if (!newf)
 626                 goto out;
 627
 628         atomic_set(&newf->count, 1);
 629
 630         spin_lock_init(&newf->file_lock);
 631         newf->next_fd = 0;
 632         fdt = &newf->fdtab;
 633         fdt->max_fds = NR_OPEN_DEFAULT;
 634         fdt->max_fdset = EMBEDDED_FD_SET_SIZE;
 635         fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
 636         fdt->open_fds = (fd_set *)&newf->open_fds_init;
 637         fdt->fd = &newf->fd_array[0];
 638         INIT_RCU_HEAD(&fdt->rcu);
 639         fdt->free_files = NULL;
 640         fdt->next = NULL;
 641         rcu_assign_pointer(newf->fdt, fdt);
 642 out:
 643         return newf;
 644 }
 645
 646 /*
 647  * Allocate a new files structure and copy contents from the
 648  * passed in files structure.
 649  * errorp will be valid only when the returned files_struct is NULL.
 650  */
 651 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
 652 {
 653         struct files_struct *newf;
 654         struct file **old_fds, **new_fds;
 655         int open_files, size, i, expand;
 656         struct fdtable *old_fdt, *new_fdt;
 657
 658         *errorp = -ENOMEM;
 659         newf = alloc_files();
 660         if (!newf)
 661                 goto out;
 662
 663         spin_lock(&oldf->file_lock);
 664         old_fdt = files_fdtable(oldf);
 665         new_fdt = files_fdtable(newf);
 666         size = old_fdt->max_fdset;
 667         open_files = count_open_files(old_fdt);
 668         expand = 0;
 669
 670         /*
 671          * Check whether we need to allocate a larger fd array or fd set.
 672          * Note: we're not a clone task, so the open count won't  change.
 673          */
 674         if (open_files > new_fdt->max_fdset) {
 675                 new_fdt->max_fdset = 0;
 676                 expand = 1;
 677         }
 678         if (open_files > new_fdt->max_fds) {
 679                 new_fdt->max_fds = 0;
 680                 expand = 1;
 681         }
 682
 683         /* if the old fdset gets grown now, we'll only copy up to "size" fds */
 684         if (expand) {
 685                 spin_unlock(&oldf->file_lock);
 686                 spin_lock(&newf->file_lock);
 687                 *errorp = expand_files(newf, open_files-1);
 688                 spin_unlock(&newf->file_lock);
 689                 if (*errorp < 0)
 690                         goto out_release;
 691                 new_fdt = files_fdtable(newf);
 692                 /*
 693                  * Reacquire the oldf lock and a pointer to its fd table
 694                  * who knows it may have a new bigger fd table. We need
 695                  * the latest pointer.
 696                  */
 697                 spin_lock(&oldf->file_lock);
 698                 old_fdt = files_fdtable(oldf);
 699         }
 700
 701         old_fds = old_fdt->fd;
 702         new_fds = new_fdt->fd;
 703
 704         memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
 705         memcpy(new_fdt->close_on_exec->fds_bits, 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_fdset > open_files) {
 731                 int left = (new_fdt->max_fdset-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 out:
 739         return newf;
 740
 741 out_release:
 742         free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
 743         free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
 744         free_fd_array(new_fdt->fd, new_fdt->max_fds);
 745         kmem_cache_free(files_cachep, newf);
 746         return NULL;
 747 }
 748
 749 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 750 {
 751         struct files_struct *oldf, *newf;
 752         int error = 0;
 753
 754         /*
 755          * A background process may not have any files ...
 756          */
 757         oldf = current->files;
 758         if (!oldf)
 759                 goto out;
 760
 761         if (clone_flags & CLONE_FILES) {
 762                 atomic_inc(&oldf->count);
 763                 goto out;
 764         }
 765
 766         /*
 767          * Note: we may be using current for both targets (See exec.c)
 768          * This works because we cache current->files (old) as oldf. Don't
 769          * break this.
 770          */
 771         tsk->files = NULL;
 772         newf = dup_fd(oldf, &error);
 773         if (!newf)
 774                 goto out;
 775
 776         tsk->files = newf;
 777         error = 0;
 778 out:
 779         return error;
 780 }
 781
 782 /*
 783  *      Helper to unshare the files of the current task.
 784  *      We don't want to expose copy_files internals to
 785  *      the exec layer of the kernel.
 786  */
 787
 788 int unshare_files(void)
 789 {
 790         struct files_struct *files  = current->files;
 791         int rc;
 792
 793         BUG_ON(!files);
 794
 795         /* This can race but the race causes us to copy when we don't
 796            need to and drop the copy */
 797         if(atomic_read(&files->count) == 1)
 798         {
 799                 atomic_inc(&files->count);
 800                 return 0;
 801         }
 802         rc = copy_files(0, current);
 803         if(rc)
 804                 current->files = files;
 805         return rc;
 806 }
 807
 808 EXPORT_SYMBOL(unshare_files);
 809
 810 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 811 {
 812         struct sighand_struct *sig;
 813
 814         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 815                 atomic_inc(&current->sighand->count);
 816                 return 0;
 817         }
 818         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 819         rcu_assign_pointer(tsk->sighand, sig);
 820         if (!sig)
 821                 return -ENOMEM;
 822         atomic_set(&sig->count, 1);
 823         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 824         return 0;
 825 }
 826
 827 void __cleanup_sighand(struct sighand_struct *sighand)
 828 {
 829         if (atomic_dec_and_test(&sighand->count))
 830                 kmem_cache_free(sighand_cachep, sighand);
 831 }
 832
 833 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 834 {
 835         struct signal_struct *sig;
 836         int ret;
 837
 838         if (clone_flags & CLONE_THREAD) {
 839                 atomic_inc(&current->signal->count);
 840                 atomic_inc(&current->signal->live);
 841                 taskstats_tgid_alloc(current);
 842                 return 0;
 843         }
 844         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 845         tsk->signal = sig;
 846         if (!sig)
 847                 return -ENOMEM;
 848
 849         ret = copy_thread_group_keys(tsk);
 850         if (ret < 0) {
 851                 kmem_cache_free(signal_cachep, sig);
 852                 return ret;
 853         }
 854
 855         atomic_set(&sig->count, 1);
 856         atomic_set(&sig->live, 1);
 857         init_waitqueue_head(&sig->wait_chldexit);
 858         sig->flags = 0;
 859         sig->group_exit_code = 0;
 860         sig->group_exit_task = NULL;
 861         sig->group_stop_count = 0;
 862         sig->curr_target = NULL;
 863         init_sigpending(&sig->shared_pending);
 864         INIT_LIST_HEAD(&sig->posix_timers);
 865
 866         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_REL);
 867         sig->it_real_incr.tv64 = 0;
 868         sig->real_timer.function = it_real_fn;
 869         sig->tsk = tsk;
 870
 871         sig->it_virt_expires = cputime_zero;
 872         sig->it_virt_incr = cputime_zero;
 873         sig->it_prof_expires = cputime_zero;
 874         sig->it_prof_incr = cputime_zero;
 875
 876         sig->leader = 0;        /* session leadership doesn't inherit */
 877         sig->tty_old_pgrp = 0;
 878
 879         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 880         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 881         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 882         sig->sched_time = 0;
 883         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 884         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 885         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 886         taskstats_tgid_init(sig);
 887
 888         task_lock(current->group_leader);
 889         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 890         task_unlock(current->group_leader);
 891
 892         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 893                 /*
 894                  * New sole thread in the process gets an expiry time
 895                  * of the whole CPU time limit.
 896                  */
 897                 tsk->it_prof_expires =
 898                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 899         }
 900         acct_init_pacct(&sig->pacct);
 901
 902         return 0;
 903 }
 904
 905 void __cleanup_signal(struct signal_struct *sig)
 906 {
 907         exit_thread_group_keys(sig);
 908         kmem_cache_free(signal_cachep, sig);
 909 }
 910
 911 static inline void cleanup_signal(struct task_struct *tsk)
 912 {
 913         struct signal_struct *sig = tsk->signal;
 914
 915         atomic_dec(&sig->live);
 916
 917         if (atomic_dec_and_test(&sig->count))
 918                 __cleanup_signal(sig);
 919 }
 920
 921 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 922 {
 923         unsigned long new_flags = p->flags;
 924
 925         new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
 926         new_flags |= PF_FORKNOEXEC;
 927         if (!(clone_flags & CLONE_PTRACE))
 928                 p->ptrace = 0;
 929         p->flags = new_flags;
 930 }
 931
 932 asmlinkage long sys_set_tid_address(int __user *tidptr)
 933 {
 934         current->clear_child_tid = tidptr;
 935
 936         return current->pid;
 937 }
 938
 939 static inline void rt_mutex_init_task(struct task_struct *p)
 940 {
 941 #ifdef CONFIG_RT_MUTEXES
 942         spin_lock_init(&p->pi_lock);
 943         plist_head_init(&p->pi_waiters, &p->pi_lock);
 944         p->pi_blocked_on = NULL;
 945 #endif
 946 }
 947
 948 /*
 949  * This creates a new process as a copy of the old one,
 950  * but does not actually start it yet.
 951  *
 952  * It copies the registers, and all the appropriate
 953  * parts of the process environment (as per the clone
 954  * flags). The actual kick-off is left to the caller.
 955  */
 956 static struct task_struct *copy_process(unsigned long clone_flags,
 957                                         unsigned long stack_start,
 958                                         struct pt_regs *regs,
 959                                         unsigned long stack_size,
 960                                         int __user *parent_tidptr,
 961                                         int __user *child_tidptr,
 962                                         int pid)
 963 {
 964         int retval;
 965         struct task_struct *p = NULL;
 966
 967         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 968                 return ERR_PTR(-EINVAL);
 969
 970         /*
 971          * Thread groups must share signals as well, and detached threads
 972          * can only be started up within the thread group.
 973          */
 974         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 975                 return ERR_PTR(-EINVAL);
 976
 977         /*
 978          * Shared signal handlers imply shared VM. By way of the above,
 979          * thread groups also imply shared VM. Blocking this case allows
 980          * for various simplifications in other code.
 981          */
 982         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 983                 return ERR_PTR(-EINVAL);
 984
 985         retval = security_task_create(clone_flags);
 986         if (retval)
 987                 goto fork_out;
 988
 989         retval = -ENOMEM;
 990         p = dup_task_struct(current);
 991         if (!p)
 992                 goto fork_out;
 993
 994         rt_mutex_init_task(p);
 995
 996 #ifdef CONFIG_TRACE_IRQFLAGS
 997         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
 998         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
 999 #endif
1000         retval = -EAGAIN;
1001         if (atomic_read(&p->user->processes) >=
1002                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1003                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1004                                 p->user != &root_user)
1005                         goto bad_fork_free;
1006         }
1007
1008         atomic_inc(&p->user->__count);
1009         atomic_inc(&p->user->processes);
1010         get_group_info(p->group_info);
1011
1012         /*
1013          * If multiple threads are within copy_process(), then this check
1014          * triggers too late. This doesn't hurt, the check is only there
1015          * to stop root fork bombs.
1016          */
1017         if (nr_threads >= max_threads)
1018                 goto bad_fork_cleanup_count;
1019
1020         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1021                 goto bad_fork_cleanup_count;
1022
1023         if (p->binfmt && !try_module_get(p->binfmt->module))
1024                 goto bad_fork_cleanup_put_domain;
1025
1026         p->did_exec = 0;
1027         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1028         copy_flags(clone_flags, p);
1029         p->pid = pid;
1030         retval = -EFAULT;
1031         if (clone_flags & CLONE_PARENT_SETTID)
1032                 if (put_user(p->pid, parent_tidptr))
1033                         goto bad_fork_cleanup_delays_binfmt;
1034
1035         INIT_LIST_HEAD(&p->children);
1036         INIT_LIST_HEAD(&p->sibling);
1037         p->vfork_done = NULL;
1038         spin_lock_init(&p->alloc_lock);
1039
1040         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1041         init_sigpending(&p->pending);
1042
1043         p->utime = cputime_zero;
1044         p->stime = cputime_zero;
1045         p->sched_time = 0;
1046         p->rchar = 0;           /* I/O counter: bytes read */
1047         p->wchar = 0;           /* I/O counter: bytes written */
1048         p->syscr = 0;           /* I/O counter: read syscalls */
1049         p->syscw = 0;           /* I/O counter: write syscalls */
1050         acct_clear_integrals(p);
1051
1052         p->it_virt_expires = cputime_zero;
1053         p->it_prof_expires = cputime_zero;
1054         p->it_sched_expires = 0;
1055         INIT_LIST_HEAD(&p->cpu_timers[0]);
1056         INIT_LIST_HEAD(&p->cpu_timers[1]);
1057         INIT_LIST_HEAD(&p->cpu_timers[2]);
1058
1059         p->lock_depth = -1;             /* -1 = no lock */
1060         do_posix_clock_monotonic_gettime(&p->start_time);
1061         p->security = NULL;
1062         p->io_context = NULL;
1063         p->io_wait = NULL;
1064         p->audit_context = NULL;
1065         cpuset_fork(p);
1066 #ifdef CONFIG_NUMA
1067         p->mempolicy = mpol_copy(p->mempolicy);
1068         if (IS_ERR(p->mempolicy)) {
1069                 retval = PTR_ERR(p->mempolicy);
1070                 p->mempolicy = NULL;
1071                 goto bad_fork_cleanup_cpuset;
1072         }
1073         mpol_fix_fork_child_flag(p);
1074 #endif
1075 #ifdef CONFIG_TRACE_IRQFLAGS
1076         p->irq_events = 0;
1077 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1078         p->hardirqs_enabled = 1;
1079 #else
1080         p->hardirqs_enabled = 0;
1081 #endif
1082         p->hardirq_enable_ip = 0;
1083         p->hardirq_enable_event = 0;
1084         p->hardirq_disable_ip = _THIS_IP_;
1085         p->hardirq_disable_event = 0;
1086         p->softirqs_enabled = 1;
1087         p->softirq_enable_ip = _THIS_IP_;
1088         p->softirq_enable_event = 0;
1089         p->softirq_disable_ip = 0;
1090         p->softirq_disable_event = 0;
1091         p->hardirq_context = 0;
1092         p->softirq_context = 0;
1093 #endif
1094 #ifdef CONFIG_LOCKDEP
1095         p->lockdep_depth = 0; /* no locks held yet */
1096         p->curr_chain_key = 0;
1097         p->lockdep_recursion = 0;
1098 #endif
1099
1100 #ifdef CONFIG_DEBUG_MUTEXES
1101         p->blocked_on = NULL; /* not blocked yet */
1102 #endif
1103
1104         p->tgid = p->pid;
1105         if (clone_flags & CLONE_THREAD)
1106                 p->tgid = current->tgid;
1107
1108         if ((retval = security_task_alloc(p)))
1109                 goto bad_fork_cleanup_policy;
1110         if ((retval = audit_alloc(p)))
1111                 goto bad_fork_cleanup_security;
1112         /* copy all the process information */
1113         if ((retval = copy_semundo(clone_flags, p)))
1114                 goto bad_fork_cleanup_audit;
1115         if ((retval = copy_files(clone_flags, p)))
1116                 goto bad_fork_cleanup_semundo;
1117         if ((retval = copy_fs(clone_flags, p)))
1118                 goto bad_fork_cleanup_files;
1119         if ((retval = copy_sighand(clone_flags, p)))
1120                 goto bad_fork_cleanup_fs;
1121         if ((retval = copy_signal(clone_flags, p)))
1122                 goto bad_fork_cleanup_sighand;
1123         if ((retval = copy_mm(clone_flags, p)))
1124                 goto bad_fork_cleanup_signal;
1125         if ((retval = copy_keys(clone_flags, p)))
1126                 goto bad_fork_cleanup_mm;
1127         if ((retval = copy_namespaces(clone_flags, p)))
1128                 goto bad_fork_cleanup_keys;
1129         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1130         if (retval)
1131                 goto bad_fork_cleanup_namespaces;
1132
1133         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1134         /*
1135          * Clear TID on mm_release()?
1136          */
1137         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1138         p->robust_list = NULL;
1139 #ifdef CONFIG_COMPAT
1140         p->compat_robust_list = NULL;
1141 #endif
1142         INIT_LIST_HEAD(&p->pi_state_list);
1143         p->pi_state_cache = NULL;
1144
1145         /*
1146          * sigaltstack should be cleared when sharing the same VM
1147          */
1148         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1149                 p->sas_ss_sp = p->sas_ss_size = 0;
1150
1151         /*
1152          * Syscall tracing should be turned off in the child regardless
1153          * of CLONE_PTRACE.
1154          */
1155         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1156 #ifdef TIF_SYSCALL_EMU
1157         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1158 #endif
1159
1160         /* Our parent execution domain becomes current domain
1161            These must match for thread signalling to apply */
1162         p->parent_exec_id = p->self_exec_id;
1163
1164         /* ok, now we should be set up.. */
1165         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1166         p->pdeath_signal = 0;
1167         p->exit_state = 0;
1168
1169         /*
1170          * Ok, make it visible to the rest of the system.
1171          * We dont wake it up yet.
1172          */
1173         p->group_leader = p;
1174         INIT_LIST_HEAD(&p->thread_group);
1175         INIT_LIST_HEAD(&p->ptrace_children);
1176         INIT_LIST_HEAD(&p->ptrace_list);
1177
1178         /* Perform scheduler related setup. Assign this task to a CPU. */
1179         sched_fork(p, clone_flags);
1180
1181         /* Need tasklist lock for parent etc handling! */
1182         write_lock_irq(&tasklist_lock);
1183
1184         /* for sys_ioprio_set(IOPRIO_WHO_PGRP) */
1185         p->ioprio = current->ioprio;
1186
1187         /*
1188          * The task hasn't been attached yet, so its cpus_allowed mask will
1189          * not be changed, nor will its assigned CPU.
1190          *
1191          * The cpus_allowed mask of the parent may have changed after it was
1192          * copied first time - so re-copy it here, then check the child's CPU
1193          * to ensure it is on a valid CPU (and if not, just force it back to
1194          * parent's CPU). This avoids alot of nasty races.
1195          */
1196         p->cpus_allowed = current->cpus_allowed;
1197         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1198                         !cpu_online(task_cpu(p))))
1199                 set_task_cpu(p, smp_processor_id());
1200
1201         /* CLONE_PARENT re-uses the old parent */
1202         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1203                 p->real_parent = current->real_parent;
1204         else
1205                 p->real_parent = current;
1206         p->parent = p->real_parent;
1207
1208         spin_lock(&current->sighand->siglock);
1209
1210         /*
1211          * Process group and session signals need to be delivered to just the
1212          * parent before the fork or both the parent and the child after the
1213          * fork. Restart if a signal comes in before we add the new process to
1214          * it's process group.
1215          * A fatal signal pending means that current will exit, so the new
1216          * thread can't slip out of an OOM kill (or normal SIGKILL).
1217          */
1218         recalc_sigpending();
1219         if (signal_pending(current)) {
1220                 spin_unlock(&current->sighand->siglock);
1221                 write_unlock_irq(&tasklist_lock);
1222                 retval = -ERESTARTNOINTR;
1223                 goto bad_fork_cleanup_namespaces;
1224         }
1225
1226         if (clone_flags & CLONE_THREAD) {
1227                 p->group_leader = current->group_leader;
1228                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1229
1230                 if (!cputime_eq(current->signal->it_virt_expires,
1231                                 cputime_zero) ||
1232                     !cputime_eq(current->signal->it_prof_expires,
1233                                 cputime_zero) ||
1234                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1235                     !list_empty(&current->signal->cpu_timers[0]) ||
1236                     !list_empty(&current->signal->cpu_timers[1]) ||
1237                     !list_empty(&current->signal->cpu_timers[2])) {
1238                         /*
1239                          * Have child wake up on its first tick to check
1240                          * for process CPU timers.
1241                          */
1242                         p->it_prof_expires = jiffies_to_cputime(1);
1243                 }
1244         }
1245
1246         if (likely(p->pid)) {
1247                 add_parent(p);
1248                 if (unlikely(p->ptrace & PT_PTRACED))
1249                         __ptrace_link(p, current->parent);
1250
1251                 if (thread_group_leader(p)) {
1252                         p->signal->tty = current->signal->tty;
1253                         p->signal->pgrp = process_group(current);
1254                         p->signal->session = current->signal->session;
1255                         attach_pid(p, PIDTYPE_PGID, process_group(p));
1256                         attach_pid(p, PIDTYPE_SID, p->signal->session);
1257
1258                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1259                         __get_cpu_var(process_counts)++;
1260                 }
1261                 attach_pid(p, PIDTYPE_PID, p->pid);
1262                 nr_threads++;
1263         }
1264
1265         total_forks++;
1266         spin_unlock(&current->sighand->siglock);
1267         write_unlock_irq(&tasklist_lock);
1268         proc_fork_connector(p);
1269         return p;
1270
1271 bad_fork_cleanup_namespaces:
1272         exit_task_namespaces(p);
1273 bad_fork_cleanup_keys:
1274         exit_keys(p);
1275 bad_fork_cleanup_mm:
1276         if (p->mm)
1277                 mmput(p->mm);
1278 bad_fork_cleanup_signal:
1279         cleanup_signal(p);
1280 bad_fork_cleanup_sighand:
1281         __cleanup_sighand(p->sighand);
1282 bad_fork_cleanup_fs:
1283         exit_fs(p); /* blocking */
1284 bad_fork_cleanup_files:
1285         exit_files(p); /* blocking */
1286 bad_fork_cleanup_semundo:
1287         exit_sem(p);
1288 bad_fork_cleanup_audit:
1289         audit_free(p);
1290 bad_fork_cleanup_security:
1291         security_task_free(p);
1292 bad_fork_cleanup_policy:
1293 #ifdef CONFIG_NUMA
1294         mpol_free(p->mempolicy);
1295 bad_fork_cleanup_cpuset:
1296 #endif
1297         cpuset_exit(p);
1298 bad_fork_cleanup_delays_binfmt:
1299         delayacct_tsk_free(p);
1300         if (p->binfmt)
1301                 module_put(p->binfmt->module);
1302 bad_fork_cleanup_put_domain:
1303         module_put(task_thread_info(p)->exec_domain->module);
1304 bad_fork_cleanup_count:
1305         put_group_info(p->group_info);
1306         atomic_dec(&p->user->processes);
1307         free_uid(p->user);
1308 bad_fork_free:
1309         free_task(p);
1310 fork_out:
1311         return ERR_PTR(retval);
1312 }
1313
1314 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1315 {
1316         memset(regs, 0, sizeof(struct pt_regs));
1317         return regs;
1318 }
1319
1320 struct task_struct * __devinit fork_idle(int cpu)
1321 {
1322         struct task_struct *task;
1323         struct pt_regs regs;
1324
1325         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1326         if (!IS_ERR(task))
1327                 init_idle(task, cpu);
1328
1329         return task;
1330 }
1331
1332 static inline int fork_traceflag (unsigned clone_flags)
1333 {
1334         if (clone_flags & CLONE_UNTRACED)
1335                 return 0;
1336         else if (clone_flags & CLONE_VFORK) {
1337                 if (current->ptrace & PT_TRACE_VFORK)
1338                         return PTRACE_EVENT_VFORK;
1339         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1340                 if (current->ptrace & PT_TRACE_CLONE)
1341                         return PTRACE_EVENT_CLONE;
1342         } else if (current->ptrace & PT_TRACE_FORK)
1343                 return PTRACE_EVENT_FORK;
1344
1345         return 0;
1346 }
1347
1348 /*
1349  *  Ok, this is the main fork-routine.
1350  *
1351  * It copies the process, and if successful kick-starts
1352  * it and waits for it to finish using the VM if required.
1353  */
1354 long do_fork(unsigned long clone_flags,
1355               unsigned long stack_start,
1356               struct pt_regs *regs,
1357               unsigned long stack_size,
1358               int __user *parent_tidptr,
1359               int __user *child_tidptr)
1360 {
1361         struct task_struct *p;
1362         int trace = 0;
1363         struct pid *pid = alloc_pid();
1364         long nr;
1365
1366         if (!pid)
1367                 return -EAGAIN;
1368         nr = pid->nr;
1369         if (unlikely(current->ptrace)) {
1370                 trace = fork_traceflag (clone_flags);
1371                 if (trace)
1372                         clone_flags |= CLONE_PTRACE;
1373         }
1374
1375         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, nr);
1376         /*
1377          * Do this prior waking up the new thread - the thread pointer
1378          * might get invalid after that point, if the thread exits quickly.
1379          */
1380         if (!IS_ERR(p)) {
1381                 struct completion vfork;
1382
1383                 if (clone_flags & CLONE_VFORK) {
1384                         p->vfork_done = &vfork;
1385                         init_completion(&vfork);
1386                 }
1387
1388                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1389                         /*
1390                          * We'll start up with an immediate SIGSTOP.
1391                          */
1392                         sigaddset(&p->pending.signal, SIGSTOP);
1393                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1394                 }
1395
1396                 if (!(clone_flags & CLONE_STOPPED))
1397                         wake_up_new_task(p, clone_flags);
1398                 else
1399                         p->state = TASK_STOPPED;
1400
1401                 if (unlikely (trace)) {
1402                         current->ptrace_message = nr;
1403                         ptrace_notify ((trace << 8) | SIGTRAP);
1404                 }
1405
1406                 if (clone_flags & CLONE_VFORK) {
1407                         wait_for_completion(&vfork);
1408                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1409                                 current->ptrace_message = nr;
1410                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1411                         }
1412                 }
1413         } else {
1414                 free_pid(pid);
1415                 nr = PTR_ERR(p);
1416         }
1417         return nr;
1418 }
1419
1420 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1421 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1422 #endif
1423
1424 static void sighand_ctor(void *data, struct kmem_cache *cachep, unsigned long flags)
1425 {
1426         struct sighand_struct *sighand = data;
1427
1428         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
1429                                         SLAB_CTOR_CONSTRUCTOR)
1430                 spin_lock_init(&sighand->siglock);
1431 }
1432
1433 void __init proc_caches_init(void)
1434 {
1435         sighand_cachep = kmem_cache_create("sighand_cache",
1436                         sizeof(struct sighand_struct), 0,
1437                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1438                         sighand_ctor, NULL);
1439         signal_cachep = kmem_cache_create("signal_cache",
1440                         sizeof(struct signal_struct), 0,
1441                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1442         files_cachep = kmem_cache_create("files_cache",
1443                         sizeof(struct files_struct), 0,
1444                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1445         fs_cachep = kmem_cache_create("fs_cache",
1446                         sizeof(struct fs_struct), 0,
1447                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1448         vm_area_cachep = kmem_cache_create("vm_area_struct",
1449                         sizeof(struct vm_area_struct), 0,
1450                         SLAB_PANIC, NULL, NULL);
1451         mm_cachep = kmem_cache_create("mm_struct",
1452                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1453                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1454 }
1455
1456
1457 /*
1458  * Check constraints on flags passed to the unshare system call and
1459  * force unsharing of additional process context as appropriate.
1460  */
1461 static inline void check_unshare_flags(unsigned long *flags_ptr)
1462 {
1463         /*
1464          * If unsharing a thread from a thread group, must also
1465          * unshare vm.
1466          */
1467         if (*flags_ptr & CLONE_THREAD)
1468                 *flags_ptr |= CLONE_VM;
1469
1470         /*
1471          * If unsharing vm, must also unshare signal handlers.
1472          */
1473         if (*flags_ptr & CLONE_VM)
1474                 *flags_ptr |= CLONE_SIGHAND;
1475
1476         /*
1477          * If unsharing signal handlers and the task was created
1478          * using CLONE_THREAD, then must unshare the thread
1479          */
1480         if ((*flags_ptr & CLONE_SIGHAND) &&
1481             (atomic_read(&current->signal->count) > 1))
1482                 *flags_ptr |= CLONE_THREAD;
1483
1484         /*
1485          * If unsharing namespace, must also unshare filesystem information.
1486          */
1487         if (*flags_ptr & CLONE_NEWNS)
1488                 *flags_ptr |= CLONE_FS;
1489 }
1490
1491 /*
1492  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1493  */
1494 static int unshare_thread(unsigned long unshare_flags)
1495 {
1496         if (unshare_flags & CLONE_THREAD)
1497                 return -EINVAL;
1498
1499         return 0;
1500 }
1501
1502 /*
1503  * Unshare the filesystem structure if it is being shared
1504  */
1505 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1506 {
1507         struct fs_struct *fs = current->fs;
1508
1509         if ((unshare_flags & CLONE_FS) &&
1510             (fs && atomic_read(&fs->count) > 1)) {
1511                 *new_fsp = __copy_fs_struct(current->fs);
1512                 if (!*new_fsp)
1513                         return -ENOMEM;
1514         }
1515
1516         return 0;
1517 }
1518
1519 /*
1520  * Unshare the namespace structure if it is being shared
1521  */
1522 static int unshare_namespace(unsigned long unshare_flags, struct namespace **new_nsp, struct fs_struct *new_fs)
1523 {
1524         struct namespace *ns = current->nsproxy->namespace;
1525
1526         if ((unshare_flags & CLONE_NEWNS) && ns) {
1527                 if (!capable(CAP_SYS_ADMIN))
1528                         return -EPERM;
1529
1530                 *new_nsp = dup_namespace(current, new_fs ? new_fs : current->fs);
1531                 if (!*new_nsp)
1532                         return -ENOMEM;
1533         }
1534
1535         return 0;
1536 }
1537
1538 /*
1539  * Unsharing of sighand for tasks created with CLONE_SIGHAND is not
1540  * supported yet
1541  */
1542 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1543 {
1544         struct sighand_struct *sigh = current->sighand;
1545
1546         if ((unshare_flags & CLONE_SIGHAND) &&
1547             (sigh && atomic_read(&sigh->count) > 1))
1548                 return -EINVAL;
1549         else
1550                 return 0;
1551 }
1552
1553 /*
1554  * Unshare vm if it is being shared
1555  */
1556 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1557 {
1558         struct mm_struct *mm = current->mm;
1559
1560         if ((unshare_flags & CLONE_VM) &&
1561             (mm && atomic_read(&mm->mm_users) > 1)) {
1562                 return -EINVAL;
1563         }
1564
1565         return 0;
1566 }
1567
1568 /*
1569  * Unshare file descriptor table if it is being shared
1570  */
1571 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1572 {
1573         struct files_struct *fd = current->files;
1574         int error = 0;
1575
1576         if ((unshare_flags & CLONE_FILES) &&
1577             (fd && atomic_read(&fd->count) > 1)) {
1578                 *new_fdp = dup_fd(fd, &error);
1579                 if (!*new_fdp)
1580                         return error;
1581         }
1582
1583         return 0;
1584 }
1585
1586 /*
1587  * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1588  * supported yet
1589  */
1590 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1591 {
1592         if (unshare_flags & CLONE_SYSVSEM)
1593                 return -EINVAL;
1594
1595         return 0;
1596 }
1597
1598 #ifndef CONFIG_IPC_NS
1599 static inline int unshare_ipcs(unsigned long flags, struct ipc_namespace **ns)
1600 {
1601         if (flags & CLONE_NEWIPC)
1602                 return -EINVAL;
1603
1604         return 0;
1605 }
1606 #endif
1607
1608 /*
1609  * unshare allows a process to 'unshare' part of the process
1610  * context which was originally shared using clone.  copy_*
1611  * functions used by do_fork() cannot be used here directly
1612  * because they modify an inactive task_struct that is being
1613  * constructed. Here we are modifying the current, active,
1614  * task_struct.
1615  */
1616 asmlinkage long sys_unshare(unsigned long unshare_flags)
1617 {
1618         int err = 0;
1619         struct fs_struct *fs, *new_fs = NULL;
1620         struct namespace *ns, *new_ns = NULL;
1621         struct sighand_struct *sigh, *new_sigh = NULL;
1622         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1623         struct files_struct *fd, *new_fd = NULL;
1624         struct sem_undo_list *new_ulist = NULL;
1625         struct nsproxy *new_nsproxy = NULL, *old_nsproxy = NULL;
1626         struct uts_namespace *uts, *new_uts = NULL;
1627         struct ipc_namespace *ipc, *new_ipc = NULL;
1628
1629         check_unshare_flags(&unshare_flags);
1630
1631         /* Return -EINVAL for all unsupported flags */
1632         err = -EINVAL;
1633         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1634                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1635                                 CLONE_NEWUTS|CLONE_NEWIPC))
1636                 goto bad_unshare_out;
1637
1638         if ((err = unshare_thread(unshare_flags)))
1639                 goto bad_unshare_out;
1640         if ((err = unshare_fs(unshare_flags, &new_fs)))
1641                 goto bad_unshare_cleanup_thread;
1642         if ((err = unshare_namespace(unshare_flags, &new_ns, new_fs)))
1643                 goto bad_unshare_cleanup_fs;
1644         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1645                 goto bad_unshare_cleanup_ns;
1646         if ((err = unshare_vm(unshare_flags, &new_mm)))
1647                 goto bad_unshare_cleanup_sigh;
1648         if ((err = unshare_fd(unshare_flags, &new_fd)))
1649                 goto bad_unshare_cleanup_vm;
1650         if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1651                 goto bad_unshare_cleanup_fd;
1652         if ((err = unshare_utsname(unshare_flags, &new_uts)))
1653                 goto bad_unshare_cleanup_semundo;
1654         if ((err = unshare_ipcs(unshare_flags, &new_ipc)))
1655                 goto bad_unshare_cleanup_uts;
1656
1657         if (new_ns || new_uts || new_ipc) {
1658                 old_nsproxy = current->nsproxy;
1659                 new_nsproxy = dup_namespaces(old_nsproxy);
1660                 if (!new_nsproxy) {
1661                         err = -ENOMEM;
1662                         goto bad_unshare_cleanup_ipc;
1663                 }
1664         }
1665
1666         if (new_fs || new_ns || new_sigh || new_mm || new_fd || new_ulist ||
1667                                 new_uts || new_ipc) {
1668
1669                 task_lock(current);
1670
1671                 if (new_nsproxy) {
1672                         current->nsproxy = new_nsproxy;
1673                         new_nsproxy = old_nsproxy;
1674                 }
1675
1676                 if (new_fs) {
1677                         fs = current->fs;
1678                         current->fs = new_fs;
1679                         new_fs = fs;
1680                 }
1681
1682                 if (new_ns) {
1683                         ns = current->nsproxy->namespace;
1684                         current->nsproxy->namespace = new_ns;
1685                         new_ns = ns;
1686                 }
1687
1688                 if (new_sigh) {
1689                         sigh = current->sighand;
1690                         rcu_assign_pointer(current->sighand, new_sigh);
1691                         new_sigh = sigh;
1692                 }
1693
1694                 if (new_mm) {
1695                         mm = current->mm;
1696                         active_mm = current->active_mm;
1697                         current->mm = new_mm;
1698                         current->active_mm = new_mm;
1699                         activate_mm(active_mm, new_mm);
1700                         new_mm = mm;
1701                 }
1702
1703                 if (new_fd) {
1704                         fd = current->files;
1705                         current->files = new_fd;
1706                         new_fd = fd;
1707                 }
1708
1709                 if (new_uts) {
1710                         uts = current->nsproxy->uts_ns;
1711                         current->nsproxy->uts_ns = new_uts;
1712                         new_uts = uts;
1713                 }
1714
1715                 if (new_ipc) {
1716                         ipc = current->nsproxy->ipc_ns;
1717                         current->nsproxy->ipc_ns = new_ipc;
1718                         new_ipc = ipc;
1719                 }
1720
1721                 task_unlock(current);
1722         }
1723
1724         if (new_nsproxy)
1725                 put_nsproxy(new_nsproxy);
1726
1727 bad_unshare_cleanup_ipc:
1728         if (new_ipc)
1729                 put_ipc_ns(new_ipc);
1730
1731 bad_unshare_cleanup_uts:
1732         if (new_uts)
1733                 put_uts_ns(new_uts);
1734
1735 bad_unshare_cleanup_semundo:
1736 bad_unshare_cleanup_fd:
1737         if (new_fd)
1738                 put_files_struct(new_fd);
1739
1740 bad_unshare_cleanup_vm:
1741         if (new_mm)
1742                 mmput(new_mm);
1743
1744 bad_unshare_cleanup_sigh:
1745         if (new_sigh)
1746                 if (atomic_dec_and_test(&new_sigh->count))
1747                         kmem_cache_free(sighand_cachep, new_sigh);
1748
1749 bad_unshare_cleanup_ns:
1750         if (new_ns)
1751                 put_namespace(new_ns);
1752
1753 bad_unshare_cleanup_fs:
1754         if (new_fs)
1755                 put_fs_struct(new_fs);
1756
1757 bad_unshare_cleanup_thread:
1758 bad_unshare_out:
1759         return err;
1760 }