kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/smp_lock.h>
  18 #include <linux/module.h>
  19 #include <linux/vmalloc.h>
  20 #include <linux/completion.h>
  21 #include <linux/namespace.h>
  22 #include <linux/personality.h>
  23 #include <linux/mempolicy.h>
  24 #include <linux/sem.h>
  25 #include <linux/file.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/fs.h>
  30 #include <linux/capability.h>
  31 #include <linux/cpu.h>
  32 #include <linux/cpuset.h>
  33 #include <linux/security.h>
  34 #include <linux/swap.h>
  35 #include <linux/syscalls.h>
  36 #include <linux/jiffies.h>
  37 #include <linux/futex.h>
  38 #include <linux/rcupdate.h>
  39 #include <linux/ptrace.h>
  40 #include <linux/mount.h>
  41 #include <linux/audit.h>
  42 #include <linux/profile.h>
  43 #include <linux/rmap.h>
  44 #include <linux/acct.h>
  45 #include <linux/cn_proc.h>
  46 #include <linux/delayacct.h>
  47 #include <linux/taskstats_kern.h>
  48 #include <linux/random.h>
  49
  50 #include <asm/pgtable.h>
  51 #include <asm/pgalloc.h>
  52 #include <asm/uaccess.h>
  53 #include <asm/mmu_context.h>
  54 #include <asm/cacheflush.h>
  55 #include <asm/tlbflush.h>
  56
  57 /*
  58  * Protected counters by write_lock_irq(&tasklist_lock)
  59  */
  60 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  61 int nr_threads;                 /* The idle threads do not count.. */
  62
  63 int max_threads;                /* tunable limit on nr_threads */
  64
  65 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  66
  67 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  68
  69 int nr_processes(void)
  70 {
  71         int cpu;
  72         int total = 0;
  73
  74         for_each_online_cpu(cpu)
  75                 total += per_cpu(process_counts, cpu);
  76
  77         return total;
  78 }
  79
  80 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  81 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  82 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  83 static kmem_cache_t *task_struct_cachep;
  84 #endif
  85
  86 /* SLAB cache for signal_struct structures (tsk->signal) */
  87 static kmem_cache_t *signal_cachep;
  88
  89 /* SLAB cache for sighand_struct structures (tsk->sighand) */
  90 kmem_cache_t *sighand_cachep;
  91
  92 /* SLAB cache for files_struct structures (tsk->files) */
  93 kmem_cache_t *files_cachep;
  94
  95 /* SLAB cache for fs_struct structures (tsk->fs) */
  96 kmem_cache_t *fs_cachep;
  97
  98 /* SLAB cache for vm_area_struct structures */
  99 kmem_cache_t *vm_area_cachep;
 100
 101 /* SLAB cache for mm_struct structures (tsk->mm) */
 102 static kmem_cache_t *mm_cachep;
 103
 104 void free_task(struct task_struct *tsk)
 105 {
 106         free_thread_info(tsk->thread_info);
 107         rt_mutex_debug_task_free(tsk);
 108         free_task_struct(tsk);
 109 }
 110 EXPORT_SYMBOL(free_task);
 111
 112 void __put_task_struct(struct task_struct *tsk)
 113 {
 114         WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
 115         WARN_ON(atomic_read(&tsk->usage));
 116         WARN_ON(tsk == current);
 117
 118         security_task_free(tsk);
 119         free_uid(tsk->user);
 120         put_group_info(tsk->group_info);
 121         delayacct_tsk_free(tsk);
 122
 123         if (!profile_handoff_task(tsk))
 124                 free_task(tsk);
 125 }
 126
 127 void __init fork_init(unsigned long mempages)
 128 {
 129 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 130 #ifndef ARCH_MIN_TASKALIGN
 131 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 132 #endif
 133         /* create a slab on which task_structs can be allocated */
 134         task_struct_cachep =
 135                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 136                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
 137 #endif
 138
 139         /*
 140          * The default maximum number of threads is set to a safe
 141          * value: the thread structures can take up at most half
 142          * of memory.
 143          */
 144         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 145
 146         /*
 147          * we need to allow at least 20 threads to boot a system
 148          */
 149         if(max_threads < 20)
 150                 max_threads = 20;
 151
 152         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 153         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 154         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 155                 init_task.signal->rlim[RLIMIT_NPROC];
 156 }
 157
 158 static struct task_struct *dup_task_struct(struct task_struct *orig)
 159 {
 160         struct task_struct *tsk;
 161         struct thread_info *ti;
 162
 163         prepare_to_copy(orig);
 164
 165         tsk = alloc_task_struct();
 166         if (!tsk)
 167                 return NULL;
 168
 169         ti = alloc_thread_info(tsk);
 170         if (!ti) {
 171                 free_task_struct(tsk);
 172                 return NULL;
 173         }
 174
 175         *tsk = *orig;
 176         tsk->thread_info = ti;
 177         setup_thread_stack(tsk, orig);
 178
 179 #ifdef CONFIG_CC_STACKPROTECTOR
 180         tsk->stack_canary = get_random_int();
 181 #endif
 182
 183         /* One for us, one for whoever does the "release_task()" (usually parent) */
 184         atomic_set(&tsk->usage,2);
 185         atomic_set(&tsk->fs_excl, 0);
 186         tsk->btrace_seq = 0;
 187         tsk->splice_pipe = NULL;
 188         return tsk;
 189 }
 190
 191 #ifdef CONFIG_MMU
 192 static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 193 {
 194         struct vm_area_struct *mpnt, *tmp, **pprev;
 195         struct rb_node **rb_link, *rb_parent;
 196         int retval;
 197         unsigned long charge;
 198         struct mempolicy *pol;
 199
 200         down_write(&oldmm->mmap_sem);
 201         flush_cache_mm(oldmm);
 202         /*
 203          * Not linked in yet - no deadlock potential:
 204          */
 205         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 206
 207         mm->locked_vm = 0;
 208         mm->mmap = NULL;
 209         mm->mmap_cache = NULL;
 210         mm->free_area_cache = oldmm->mmap_base;
 211         mm->cached_hole_size = ~0UL;
 212         mm->map_count = 0;
 213         cpus_clear(mm->cpu_vm_mask);
 214         mm->mm_rb = RB_ROOT;
 215         rb_link = &mm->mm_rb.rb_node;
 216         rb_parent = NULL;
 217         pprev = &mm->mmap;
 218
 219         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 220                 struct file *file;
 221
 222                 if (mpnt->vm_flags & VM_DONTCOPY) {
 223                         long pages = vma_pages(mpnt);
 224                         mm->total_vm -= pages;
 225                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 226                                                                 -pages);
 227                         continue;
 228                 }
 229                 charge = 0;
 230                 if (mpnt->vm_flags & VM_ACCOUNT) {
 231                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 232                         if (security_vm_enough_memory(len))
 233                                 goto fail_nomem;
 234                         charge = len;
 235                 }
 236                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 237                 if (!tmp)
 238                         goto fail_nomem;
 239                 *tmp = *mpnt;
 240                 pol = mpol_copy(vma_policy(mpnt));
 241                 retval = PTR_ERR(pol);
 242                 if (IS_ERR(pol))
 243                         goto fail_nomem_policy;
 244                 vma_set_policy(tmp, pol);
 245                 tmp->vm_flags &= ~VM_LOCKED;
 246                 tmp->vm_mm = mm;
 247                 tmp->vm_next = NULL;
 248                 anon_vma_link(tmp);
 249                 file = tmp->vm_file;
 250                 if (file) {
 251                         struct inode *inode = file->f_dentry->d_inode;
 252                         get_file(file);
 253                         if (tmp->vm_flags & VM_DENYWRITE)
 254                                 atomic_dec(&inode->i_writecount);
 255
 256                         /* insert tmp into the share list, just after mpnt */
 257                         spin_lock(&file->f_mapping->i_mmap_lock);
 258                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 259                         flush_dcache_mmap_lock(file->f_mapping);
 260                         vma_prio_tree_add(tmp, mpnt);
 261                         flush_dcache_mmap_unlock(file->f_mapping);
 262                         spin_unlock(&file->f_mapping->i_mmap_lock);
 263                 }
 264
 265                 /*
 266                  * Link in the new vma and copy the page table entries.
 267                  */
 268                 *pprev = tmp;
 269                 pprev = &tmp->vm_next;
 270
 271                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 272                 rb_link = &tmp->vm_rb.rb_right;
 273                 rb_parent = &tmp->vm_rb;
 274
 275                 mm->map_count++;
 276                 retval = copy_page_range(mm, oldmm, mpnt);
 277
 278                 if (tmp->vm_ops && tmp->vm_ops->open)
 279                         tmp->vm_ops->open(tmp);
 280
 281                 if (retval)
 282                         goto out;
 283         }
 284         retval = 0;
 285 out:
 286         up_write(&mm->mmap_sem);
 287         flush_tlb_mm(oldmm);
 288         up_write(&oldmm->mmap_sem);
 289         return retval;
 290 fail_nomem_policy:
 291         kmem_cache_free(vm_area_cachep, tmp);
 292 fail_nomem:
 293         retval = -ENOMEM;
 294         vm_unacct_memory(charge);
 295         goto out;
 296 }
 297
 298 static inline int mm_alloc_pgd(struct mm_struct * mm)
 299 {
 300         mm->pgd = pgd_alloc(mm);
 301         if (unlikely(!mm->pgd))
 302                 return -ENOMEM;
 303         return 0;
 304 }
 305
 306 static inline void mm_free_pgd(struct mm_struct * mm)
 307 {
 308         pgd_free(mm->pgd);
 309 }
 310 #else
 311 #define dup_mmap(mm, oldmm)     (0)
 312 #define mm_alloc_pgd(mm)        (0)
 313 #define mm_free_pgd(mm)
 314 #endif /* CONFIG_MMU */
 315
 316  __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 317
 318 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
 319 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 320
 321 #include <linux/init_task.h>
 322
 323 static struct mm_struct * mm_init(struct mm_struct * mm)
 324 {
 325         atomic_set(&mm->mm_users, 1);
 326         atomic_set(&mm->mm_count, 1);
 327         init_rwsem(&mm->mmap_sem);
 328         INIT_LIST_HEAD(&mm->mmlist);
 329         mm->core_waiters = 0;
 330         mm->nr_ptes = 0;
 331         set_mm_counter(mm, file_rss, 0);
 332         set_mm_counter(mm, anon_rss, 0);
 333         spin_lock_init(&mm->page_table_lock);
 334         rwlock_init(&mm->ioctx_list_lock);
 335         mm->ioctx_list = NULL;
 336         mm->free_area_cache = TASK_UNMAPPED_BASE;
 337         mm->cached_hole_size = ~0UL;
 338
 339         if (likely(!mm_alloc_pgd(mm))) {
 340                 mm->def_flags = 0;
 341                 return mm;
 342         }
 343         free_mm(mm);
 344         return NULL;
 345 }
 346
 347 /*
 348  * Allocate and initialize an mm_struct.
 349  */
 350 struct mm_struct * mm_alloc(void)
 351 {
 352         struct mm_struct * mm;
 353
 354         mm = allocate_mm();
 355         if (mm) {
 356                 memset(mm, 0, sizeof(*mm));
 357                 mm = mm_init(mm);
 358         }
 359         return mm;
 360 }
 361
 362 /*
 363  * Called when the last reference to the mm
 364  * is dropped: either by a lazy thread or by
 365  * mmput. Free the page directory and the mm.
 366  */
 367 void fastcall __mmdrop(struct mm_struct *mm)
 368 {
 369         BUG_ON(mm == &init_mm);
 370         mm_free_pgd(mm);
 371         destroy_context(mm);
 372         free_mm(mm);
 373 }
 374
 375 /*
 376  * Decrement the use count and release all resources for an mm.
 377  */
 378 void mmput(struct mm_struct *mm)
 379 {
 380         might_sleep();
 381
 382         if (atomic_dec_and_test(&mm->mm_users)) {
 383                 exit_aio(mm);
 384                 exit_mmap(mm);
 385                 if (!list_empty(&mm->mmlist)) {
 386                         spin_lock(&mmlist_lock);
 387                         list_del(&mm->mmlist);
 388                         spin_unlock(&mmlist_lock);
 389                 }
 390                 put_swap_token(mm);
 391                 mmdrop(mm);
 392         }
 393 }
 394 EXPORT_SYMBOL_GPL(mmput);
 395
 396 /**
 397  * get_task_mm - acquire a reference to the task's mm
 398  *
 399  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
 400  * this kernel workthread has transiently adopted a user mm with use_mm,
 401  * to do its AIO) is not set and if so returns a reference to it, after
 402  * bumping up the use count.  User must release the mm via mmput()
 403  * after use.  Typically used by /proc and ptrace.
 404  */
 405 struct mm_struct *get_task_mm(struct task_struct *task)
 406 {
 407         struct mm_struct *mm;
 408
 409         task_lock(task);
 410         mm = task->mm;
 411         if (mm) {
 412                 if (task->flags & PF_BORROWED_MM)
 413                         mm = NULL;
 414                 else
 415                         atomic_inc(&mm->mm_users);
 416         }
 417         task_unlock(task);
 418         return mm;
 419 }
 420 EXPORT_SYMBOL_GPL(get_task_mm);
 421
 422 /* Please note the differences between mmput and mm_release.
 423  * mmput is called whenever we stop holding onto a mm_struct,
 424  * error success whatever.
 425  *
 426  * mm_release is called after a mm_struct has been removed
 427  * from the current process.
 428  *
 429  * This difference is important for error handling, when we
 430  * only half set up a mm_struct for a new process and need to restore
 431  * the old one.  Because we mmput the new mm_struct before
 432  * restoring the old one. . .
 433  * Eric Biederman 10 January 1998
 434  */
 435 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 436 {
 437         struct completion *vfork_done = tsk->vfork_done;
 438
 439         /* Get rid of any cached register state */
 440         deactivate_mm(tsk, mm);
 441
 442         /* notify parent sleeping on vfork() */
 443         if (vfork_done) {
 444                 tsk->vfork_done = NULL;
 445                 complete(vfork_done);
 446         }
 447         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
 448                 u32 __user * tidptr = tsk->clear_child_tid;
 449                 tsk->clear_child_tid = NULL;
 450
 451                 /*
 452                  * We don't check the error code - if userspace has
 453                  * not set up a proper pointer then tough luck.
 454                  */
 455                 put_user(0, tidptr);
 456                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 457         }
 458 }
 459
 460 /*
 461  * Allocate a new mm structure and copy contents from the
 462  * mm structure of the passed in task structure.
 463  */
 464 static struct mm_struct *dup_mm(struct task_struct *tsk)
 465 {
 466         struct mm_struct *mm, *oldmm = current->mm;
 467         int err;
 468
 469         if (!oldmm)
 470                 return NULL;
 471
 472         mm = allocate_mm();
 473         if (!mm)
 474                 goto fail_nomem;
 475
 476         memcpy(mm, oldmm, sizeof(*mm));
 477
 478         if (!mm_init(mm))
 479                 goto fail_nomem;
 480
 481         if (init_new_context(tsk, mm))
 482                 goto fail_nocontext;
 483
 484         err = dup_mmap(mm, oldmm);
 485         if (err)
 486                 goto free_pt;
 487
 488         mm->hiwater_rss = get_mm_rss(mm);
 489         mm->hiwater_vm = mm->total_vm;
 490
 491         return mm;
 492
 493 free_pt:
 494         mmput(mm);
 495
 496 fail_nomem:
 497         return NULL;
 498
 499 fail_nocontext:
 500         /*
 501          * If init_new_context() failed, we cannot use mmput() to free the mm
 502          * because it calls destroy_context()
 503          */
 504         mm_free_pgd(mm);
 505         free_mm(mm);
 506         return NULL;
 507 }
 508
 509 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 510 {
 511         struct mm_struct * mm, *oldmm;
 512         int retval;
 513
 514         tsk->min_flt = tsk->maj_flt = 0;
 515         tsk->nvcsw = tsk->nivcsw = 0;
 516
 517         tsk->mm = NULL;
 518         tsk->active_mm = NULL;
 519
 520         /*
 521          * Are we cloning a kernel thread?
 522          *
 523          * We need to steal a active VM for that..
 524          */
 525         oldmm = current->mm;
 526         if (!oldmm)
 527                 return 0;
 528
 529         if (clone_flags & CLONE_VM) {
 530                 atomic_inc(&oldmm->mm_users);
 531                 mm = oldmm;
 532                 goto good_mm;
 533         }
 534
 535         retval = -ENOMEM;
 536         mm = dup_mm(tsk);
 537         if (!mm)
 538                 goto fail_nomem;
 539
 540 good_mm:
 541         tsk->mm = mm;
 542         tsk->active_mm = mm;
 543         return 0;
 544
 545 fail_nomem:
 546         return retval;
 547 }
 548
 549 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 550 {
 551         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 552         /* We don't need to lock fs - think why ;-) */
 553         if (fs) {
 554                 atomic_set(&fs->count, 1);
 555                 rwlock_init(&fs->lock);
 556                 fs->umask = old->umask;
 557                 read_lock(&old->lock);
 558                 fs->rootmnt = mntget(old->rootmnt);
 559                 fs->root = dget(old->root);
 560                 fs->pwdmnt = mntget(old->pwdmnt);
 561                 fs->pwd = dget(old->pwd);
 562                 if (old->altroot) {
 563                         fs->altrootmnt = mntget(old->altrootmnt);
 564                         fs->altroot = dget(old->altroot);
 565                 } else {
 566                         fs->altrootmnt = NULL;
 567                         fs->altroot = NULL;
 568                 }
 569                 read_unlock(&old->lock);
 570         }
 571         return fs;
 572 }
 573
 574 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 575 {
 576         return __copy_fs_struct(old);
 577 }
 578
 579 EXPORT_SYMBOL_GPL(copy_fs_struct);
 580
 581 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 582 {
 583         if (clone_flags & CLONE_FS) {
 584                 atomic_inc(&current->fs->count);
 585                 return 0;
 586         }
 587         tsk->fs = __copy_fs_struct(current->fs);
 588         if (!tsk->fs)
 589                 return -ENOMEM;
 590         return 0;
 591 }
 592
 593 static int count_open_files(struct fdtable *fdt)
 594 {
 595         int size = fdt->max_fdset;
 596         int i;
 597
 598         /* Find the last open fd */
 599         for (i = size/(8*sizeof(long)); i > 0; ) {
 600                 if (fdt->open_fds->fds_bits[--i])
 601                         break;
 602         }
 603         i = (i+1) * 8 * sizeof(long);
 604         return i;
 605 }
 606
 607 static struct files_struct *alloc_files(void)
 608 {
 609         struct files_struct *newf;
 610         struct fdtable *fdt;
 611
 612         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
 613         if (!newf)
 614                 goto out;
 615
 616         atomic_set(&newf->count, 1);
 617
 618         spin_lock_init(&newf->file_lock);
 619         newf->next_fd = 0;
 620         fdt = &newf->fdtab;
 621         fdt->max_fds = NR_OPEN_DEFAULT;
 622         fdt->max_fdset = EMBEDDED_FD_SET_SIZE;
 623         fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
 624         fdt->open_fds = (fd_set *)&newf->open_fds_init;
 625         fdt->fd = &newf->fd_array[0];
 626         INIT_RCU_HEAD(&fdt->rcu);
 627         fdt->free_files = NULL;
 628         fdt->next = NULL;
 629         rcu_assign_pointer(newf->fdt, fdt);
 630 out:
 631         return newf;
 632 }
 633
 634 /*
 635  * Allocate a new files structure and copy contents from the
 636  * passed in files structure.
 637  * errorp will be valid only when the returned files_struct is NULL.
 638  */
 639 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
 640 {
 641         struct files_struct *newf;
 642         struct file **old_fds, **new_fds;
 643         int open_files, size, i, expand;
 644         struct fdtable *old_fdt, *new_fdt;
 645
 646         *errorp = -ENOMEM;
 647         newf = alloc_files();
 648         if (!newf)
 649                 goto out;
 650
 651         spin_lock(&oldf->file_lock);
 652         old_fdt = files_fdtable(oldf);
 653         new_fdt = files_fdtable(newf);
 654         size = old_fdt->max_fdset;
 655         open_files = count_open_files(old_fdt);
 656         expand = 0;
 657
 658         /*
 659          * Check whether we need to allocate a larger fd array or fd set.
 660          * Note: we're not a clone task, so the open count won't  change.
 661          */
 662         if (open_files > new_fdt->max_fdset) {
 663                 new_fdt->max_fdset = 0;
 664                 expand = 1;
 665         }
 666         if (open_files > new_fdt->max_fds) {
 667                 new_fdt->max_fds = 0;
 668                 expand = 1;
 669         }
 670
 671         /* if the old fdset gets grown now, we'll only copy up to "size" fds */
 672         if (expand) {
 673                 spin_unlock(&oldf->file_lock);
 674                 spin_lock(&newf->file_lock);
 675                 *errorp = expand_files(newf, open_files-1);
 676                 spin_unlock(&newf->file_lock);
 677                 if (*errorp < 0)
 678                         goto out_release;
 679                 new_fdt = files_fdtable(newf);
 680                 /*
 681                  * Reacquire the oldf lock and a pointer to its fd table
 682                  * who knows it may have a new bigger fd table. We need
 683                  * the latest pointer.
 684                  */
 685                 spin_lock(&oldf->file_lock);
 686                 old_fdt = files_fdtable(oldf);
 687         }
 688
 689         old_fds = old_fdt->fd;
 690         new_fds = new_fdt->fd;
 691
 692         memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
 693         memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
 694
 695         for (i = open_files; i != 0; i--) {
 696                 struct file *f = *old_fds++;
 697                 if (f) {
 698                         get_file(f);
 699                 } else {
 700                         /*
 701                          * The fd may be claimed in the fd bitmap but not yet
 702                          * instantiated in the files array if a sibling thread
 703                          * is partway through open().  So make sure that this
 704                          * fd is available to the new process.
 705                          */
 706                         FD_CLR(open_files - i, new_fdt->open_fds);
 707                 }
 708                 rcu_assign_pointer(*new_fds++, f);
 709         }
 710         spin_unlock(&oldf->file_lock);
 711
 712         /* compute the remainder to be cleared */
 713         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
 714
 715         /* This is long word aligned thus could use a optimized version */
 716         memset(new_fds, 0, size);
 717
 718         if (new_fdt->max_fdset > open_files) {
 719                 int left = (new_fdt->max_fdset-open_files)/8;
 720                 int start = open_files / (8 * sizeof(unsigned long));
 721
 722                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
 723                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
 724         }
 725
 726 out:
 727         return newf;
 728
 729 out_release:
 730         free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
 731         free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
 732         free_fd_array(new_fdt->fd, new_fdt->max_fds);
 733         kmem_cache_free(files_cachep, newf);
 734         return NULL;
 735 }
 736
 737 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 738 {
 739         struct files_struct *oldf, *newf;
 740         int error = 0;
 741
 742         /*
 743          * A background process may not have any files ...
 744          */
 745         oldf = current->files;
 746         if (!oldf)
 747                 goto out;
 748
 749         if (clone_flags & CLONE_FILES) {
 750                 atomic_inc(&oldf->count);
 751                 goto out;
 752         }
 753
 754         /*
 755          * Note: we may be using current for both targets (See exec.c)
 756          * This works because we cache current->files (old) as oldf. Don't
 757          * break this.
 758          */
 759         tsk->files = NULL;
 760         newf = dup_fd(oldf, &error);
 761         if (!newf)
 762                 goto out;
 763
 764         tsk->files = newf;
 765         error = 0;
 766 out:
 767         return error;
 768 }
 769
 770 /*
 771  *      Helper to unshare the files of the current task.
 772  *      We don't want to expose copy_files internals to
 773  *      the exec layer of the kernel.
 774  */
 775
 776 int unshare_files(void)
 777 {
 778         struct files_struct *files  = current->files;
 779         int rc;
 780
 781         BUG_ON(!files);
 782
 783         /* This can race but the race causes us to copy when we don't
 784            need to and drop the copy */
 785         if(atomic_read(&files->count) == 1)
 786         {
 787                 atomic_inc(&files->count);
 788                 return 0;
 789         }
 790         rc = copy_files(0, current);
 791         if(rc)
 792                 current->files = files;
 793         return rc;
 794 }
 795
 796 EXPORT_SYMBOL(unshare_files);
 797
 798 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 799 {
 800         struct sighand_struct *sig;
 801
 802         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 803                 atomic_inc(&current->sighand->count);
 804                 return 0;
 805         }
 806         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 807         rcu_assign_pointer(tsk->sighand, sig);
 808         if (!sig)
 809                 return -ENOMEM;
 810         atomic_set(&sig->count, 1);
 811         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 812         return 0;
 813 }
 814
 815 void __cleanup_sighand(struct sighand_struct *sighand)
 816 {
 817         if (atomic_dec_and_test(&sighand->count))
 818                 kmem_cache_free(sighand_cachep, sighand);
 819 }
 820
 821 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 822 {
 823         struct signal_struct *sig;
 824         int ret;
 825
 826         if (clone_flags & CLONE_THREAD) {
 827                 atomic_inc(&current->signal->count);
 828                 atomic_inc(&current->signal->live);
 829                 taskstats_tgid_alloc(current->signal);
 830                 return 0;
 831         }
 832         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 833         tsk->signal = sig;
 834         if (!sig)
 835                 return -ENOMEM;
 836
 837         ret = copy_thread_group_keys(tsk);
 838         if (ret < 0) {
 839                 kmem_cache_free(signal_cachep, sig);
 840                 return ret;
 841         }
 842
 843         atomic_set(&sig->count, 1);
 844         atomic_set(&sig->live, 1);
 845         init_waitqueue_head(&sig->wait_chldexit);
 846         sig->flags = 0;
 847         sig->group_exit_code = 0;
 848         sig->group_exit_task = NULL;
 849         sig->group_stop_count = 0;
 850         sig->curr_target = NULL;
 851         init_sigpending(&sig->shared_pending);
 852         INIT_LIST_HEAD(&sig->posix_timers);
 853
 854         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_REL);
 855         sig->it_real_incr.tv64 = 0;
 856         sig->real_timer.function = it_real_fn;
 857         sig->tsk = tsk;
 858
 859         sig->it_virt_expires = cputime_zero;
 860         sig->it_virt_incr = cputime_zero;
 861         sig->it_prof_expires = cputime_zero;
 862         sig->it_prof_incr = cputime_zero;
 863
 864         sig->leader = 0;        /* session leadership doesn't inherit */
 865         sig->tty_old_pgrp = 0;
 866
 867         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 868         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 869         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 870         sig->sched_time = 0;
 871         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 872         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 873         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 874         taskstats_tgid_init(sig);
 875
 876         task_lock(current->group_leader);
 877         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 878         task_unlock(current->group_leader);
 879
 880         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 881                 /*
 882                  * New sole thread in the process gets an expiry time
 883                  * of the whole CPU time limit.
 884                  */
 885                 tsk->it_prof_expires =
 886                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 887         }
 888         acct_init_pacct(&sig->pacct);
 889
 890         return 0;
 891 }
 892
 893 void __cleanup_signal(struct signal_struct *sig)
 894 {
 895         exit_thread_group_keys(sig);
 896         taskstats_tgid_free(sig);
 897         kmem_cache_free(signal_cachep, sig);
 898 }
 899
 900 static inline void cleanup_signal(struct task_struct *tsk)
 901 {
 902         struct signal_struct *sig = tsk->signal;
 903
 904         atomic_dec(&sig->live);
 905
 906         if (atomic_dec_and_test(&sig->count))
 907                 __cleanup_signal(sig);
 908 }
 909
 910 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 911 {
 912         unsigned long new_flags = p->flags;
 913
 914         new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
 915         new_flags |= PF_FORKNOEXEC;
 916         if (!(clone_flags & CLONE_PTRACE))
 917                 p->ptrace = 0;
 918         p->flags = new_flags;
 919 }
 920
 921 asmlinkage long sys_set_tid_address(int __user *tidptr)
 922 {
 923         current->clear_child_tid = tidptr;
 924
 925         return current->pid;
 926 }
 927
 928 static inline void rt_mutex_init_task(struct task_struct *p)
 929 {
 930 #ifdef CONFIG_RT_MUTEXES
 931         spin_lock_init(&p->pi_lock);
 932         plist_head_init(&p->pi_waiters, &p->pi_lock);
 933         p->pi_blocked_on = NULL;
 934 #endif
 935 }
 936
 937 /*
 938  * This creates a new process as a copy of the old one,
 939  * but does not actually start it yet.
 940  *
 941  * It copies the registers, and all the appropriate
 942  * parts of the process environment (as per the clone
 943  * flags). The actual kick-off is left to the caller.
 944  */
 945 static struct task_struct *copy_process(unsigned long clone_flags,
 946                                         unsigned long stack_start,
 947                                         struct pt_regs *regs,
 948                                         unsigned long stack_size,
 949                                         int __user *parent_tidptr,
 950                                         int __user *child_tidptr,
 951                                         int pid)
 952 {
 953         int retval;
 954         struct task_struct *p = NULL;
 955
 956         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 957                 return ERR_PTR(-EINVAL);
 958
 959         /*
 960          * Thread groups must share signals as well, and detached threads
 961          * can only be started up within the thread group.
 962          */
 963         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 964                 return ERR_PTR(-EINVAL);
 965
 966         /*
 967          * Shared signal handlers imply shared VM. By way of the above,
 968          * thread groups also imply shared VM. Blocking this case allows
 969          * for various simplifications in other code.
 970          */
 971         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 972                 return ERR_PTR(-EINVAL);
 973
 974         retval = security_task_create(clone_flags);
 975         if (retval)
 976                 goto fork_out;
 977
 978         retval = -ENOMEM;
 979         p = dup_task_struct(current);
 980         if (!p)
 981                 goto fork_out;
 982
 983 #ifdef CONFIG_TRACE_IRQFLAGS
 984         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
 985         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
 986 #endif
 987         retval = -EAGAIN;
 988         if (atomic_read(&p->user->processes) >=
 989                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
 990                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 991                                 p->user != &root_user)
 992                         goto bad_fork_free;
 993         }
 994
 995         atomic_inc(&p->user->__count);
 996         atomic_inc(&p->user->processes);
 997         get_group_info(p->group_info);
 998
 999         /*
1000          * If multiple threads are within copy_process(), then this check
1001          * triggers too late. This doesn't hurt, the check is only there
1002          * to stop root fork bombs.
1003          */
1004         if (nr_threads >= max_threads)
1005                 goto bad_fork_cleanup_count;
1006
1007         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1008                 goto bad_fork_cleanup_count;
1009
1010         if (p->binfmt && !try_module_get(p->binfmt->module))
1011                 goto bad_fork_cleanup_put_domain;
1012
1013         p->did_exec = 0;
1014         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1015         copy_flags(clone_flags, p);
1016         p->pid = pid;
1017         retval = -EFAULT;
1018         if (clone_flags & CLONE_PARENT_SETTID)
1019                 if (put_user(p->pid, parent_tidptr))
1020                         goto bad_fork_cleanup_delays_binfmt;
1021
1022         INIT_LIST_HEAD(&p->children);
1023         INIT_LIST_HEAD(&p->sibling);
1024         p->vfork_done = NULL;
1025         spin_lock_init(&p->alloc_lock);
1026
1027         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1028         init_sigpending(&p->pending);
1029
1030         p->utime = cputime_zero;
1031         p->stime = cputime_zero;
1032         p->sched_time = 0;
1033         p->rchar = 0;           /* I/O counter: bytes read */
1034         p->wchar = 0;           /* I/O counter: bytes written */
1035         p->syscr = 0;           /* I/O counter: read syscalls */
1036         p->syscw = 0;           /* I/O counter: write syscalls */
1037         acct_clear_integrals(p);
1038
1039         p->it_virt_expires = cputime_zero;
1040         p->it_prof_expires = cputime_zero;
1041         p->it_sched_expires = 0;
1042         INIT_LIST_HEAD(&p->cpu_timers[0]);
1043         INIT_LIST_HEAD(&p->cpu_timers[1]);
1044         INIT_LIST_HEAD(&p->cpu_timers[2]);
1045
1046         p->lock_depth = -1;             /* -1 = no lock */
1047         do_posix_clock_monotonic_gettime(&p->start_time);
1048         p->security = NULL;
1049         p->io_context = NULL;
1050         p->io_wait = NULL;
1051         p->audit_context = NULL;
1052         cpuset_fork(p);
1053 #ifdef CONFIG_NUMA
1054         p->mempolicy = mpol_copy(p->mempolicy);
1055         if (IS_ERR(p->mempolicy)) {
1056                 retval = PTR_ERR(p->mempolicy);
1057                 p->mempolicy = NULL;
1058                 goto bad_fork_cleanup_cpuset;
1059         }
1060         mpol_fix_fork_child_flag(p);
1061 #endif
1062 #ifdef CONFIG_TRACE_IRQFLAGS
1063         p->irq_events = 0;
1064         p->hardirqs_enabled = 0;
1065         p->hardirq_enable_ip = 0;
1066         p->hardirq_enable_event = 0;
1067         p->hardirq_disable_ip = _THIS_IP_;
1068         p->hardirq_disable_event = 0;
1069         p->softirqs_enabled = 1;
1070         p->softirq_enable_ip = _THIS_IP_;
1071         p->softirq_enable_event = 0;
1072         p->softirq_disable_ip = 0;
1073         p->softirq_disable_event = 0;
1074         p->hardirq_context = 0;
1075         p->softirq_context = 0;
1076 #endif
1077 #ifdef CONFIG_LOCKDEP
1078         p->lockdep_depth = 0; /* no locks held yet */
1079         p->curr_chain_key = 0;
1080         p->lockdep_recursion = 0;
1081 #endif
1082
1083         rt_mutex_init_task(p);
1084
1085 #ifdef CONFIG_DEBUG_MUTEXES
1086         p->blocked_on = NULL; /* not blocked yet */
1087 #endif
1088
1089         p->tgid = p->pid;
1090         if (clone_flags & CLONE_THREAD)
1091                 p->tgid = current->tgid;
1092
1093         if ((retval = security_task_alloc(p)))
1094                 goto bad_fork_cleanup_policy;
1095         if ((retval = audit_alloc(p)))
1096                 goto bad_fork_cleanup_security;
1097         /* copy all the process information */
1098         if ((retval = copy_semundo(clone_flags, p)))
1099                 goto bad_fork_cleanup_audit;
1100         if ((retval = copy_files(clone_flags, p)))
1101                 goto bad_fork_cleanup_semundo;
1102         if ((retval = copy_fs(clone_flags, p)))
1103                 goto bad_fork_cleanup_files;
1104         if ((retval = copy_sighand(clone_flags, p)))
1105                 goto bad_fork_cleanup_fs;
1106         if ((retval = copy_signal(clone_flags, p)))
1107                 goto bad_fork_cleanup_sighand;
1108         if ((retval = copy_mm(clone_flags, p)))
1109                 goto bad_fork_cleanup_signal;
1110         if ((retval = copy_keys(clone_flags, p)))
1111                 goto bad_fork_cleanup_mm;
1112         if ((retval = copy_namespace(clone_flags, p)))
1113                 goto bad_fork_cleanup_keys;
1114         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1115         if (retval)
1116                 goto bad_fork_cleanup_namespace;
1117
1118         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1119         /*
1120          * Clear TID on mm_release()?
1121          */
1122         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1123         p->robust_list = NULL;
1124 #ifdef CONFIG_COMPAT
1125         p->compat_robust_list = NULL;
1126 #endif
1127         INIT_LIST_HEAD(&p->pi_state_list);
1128         p->pi_state_cache = NULL;
1129
1130         /*
1131          * sigaltstack should be cleared when sharing the same VM
1132          */
1133         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1134                 p->sas_ss_sp = p->sas_ss_size = 0;
1135
1136         /*
1137          * Syscall tracing should be turned off in the child regardless
1138          * of CLONE_PTRACE.
1139          */
1140         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1141 #ifdef TIF_SYSCALL_EMU
1142         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1143 #endif
1144
1145         /* Our parent execution domain becomes current domain
1146            These must match for thread signalling to apply */
1147
1148         p->parent_exec_id = p->self_exec_id;
1149
1150         /* ok, now we should be set up.. */
1151         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1152         p->pdeath_signal = 0;
1153         p->exit_state = 0;
1154
1155         /*
1156          * Ok, make it visible to the rest of the system.
1157          * We dont wake it up yet.
1158          */
1159         p->group_leader = p;
1160         INIT_LIST_HEAD(&p->thread_group);
1161         INIT_LIST_HEAD(&p->ptrace_children);
1162         INIT_LIST_HEAD(&p->ptrace_list);
1163
1164         /* Perform scheduler related setup. Assign this task to a CPU. */
1165         sched_fork(p, clone_flags);
1166
1167         /* Need tasklist lock for parent etc handling! */
1168         write_lock_irq(&tasklist_lock);
1169
1170         /*
1171          * The task hasn't been attached yet, so its cpus_allowed mask will
1172          * not be changed, nor will its assigned CPU.
1173          *
1174          * The cpus_allowed mask of the parent may have changed after it was
1175          * copied first time - so re-copy it here, then check the child's CPU
1176          * to ensure it is on a valid CPU (and if not, just force it back to
1177          * parent's CPU). This avoids alot of nasty races.
1178          */
1179         p->cpus_allowed = current->cpus_allowed;
1180         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1181                         !cpu_online(task_cpu(p))))
1182                 set_task_cpu(p, smp_processor_id());
1183
1184         /* CLONE_PARENT re-uses the old parent */
1185         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1186                 p->real_parent = current->real_parent;
1187         else
1188                 p->real_parent = current;
1189         p->parent = p->real_parent;
1190
1191         spin_lock(&current->sighand->siglock);
1192
1193         /*
1194          * Process group and session signals need to be delivered to just the
1195          * parent before the fork or both the parent and the child after the
1196          * fork. Restart if a signal comes in before we add the new process to
1197          * it's process group.
1198          * A fatal signal pending means that current will exit, so the new
1199          * thread can't slip out of an OOM kill (or normal SIGKILL).
1200          */
1201         recalc_sigpending();
1202         if (signal_pending(current)) {
1203                 spin_unlock(&current->sighand->siglock);
1204                 write_unlock_irq(&tasklist_lock);
1205                 retval = -ERESTARTNOINTR;
1206                 goto bad_fork_cleanup_namespace;
1207         }
1208
1209         if (clone_flags & CLONE_THREAD) {
1210                 p->group_leader = current->group_leader;
1211                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1212
1213                 if (!cputime_eq(current->signal->it_virt_expires,
1214                                 cputime_zero) ||
1215                     !cputime_eq(current->signal->it_prof_expires,
1216                                 cputime_zero) ||
1217                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1218                     !list_empty(&current->signal->cpu_timers[0]) ||
1219                     !list_empty(&current->signal->cpu_timers[1]) ||
1220                     !list_empty(&current->signal->cpu_timers[2])) {
1221                         /*
1222                          * Have child wake up on its first tick to check
1223                          * for process CPU timers.
1224                          */
1225                         p->it_prof_expires = jiffies_to_cputime(1);
1226                 }
1227         }
1228
1229         /*
1230          * inherit ioprio
1231          */
1232         p->ioprio = current->ioprio;
1233
1234         if (likely(p->pid)) {
1235                 add_parent(p);
1236                 if (unlikely(p->ptrace & PT_PTRACED))
1237                         __ptrace_link(p, current->parent);
1238
1239                 if (thread_group_leader(p)) {
1240                         p->signal->tty = current->signal->tty;
1241                         p->signal->pgrp = process_group(current);
1242                         p->signal->session = current->signal->session;
1243                         attach_pid(p, PIDTYPE_PGID, process_group(p));
1244                         attach_pid(p, PIDTYPE_SID, p->signal->session);
1245
1246                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1247                         __get_cpu_var(process_counts)++;
1248                 }
1249                 attach_pid(p, PIDTYPE_PID, p->pid);
1250                 nr_threads++;
1251         }
1252
1253         total_forks++;
1254         spin_unlock(&current->sighand->siglock);
1255         write_unlock_irq(&tasklist_lock);
1256         proc_fork_connector(p);
1257         return p;
1258
1259 bad_fork_cleanup_namespace:
1260         exit_namespace(p);
1261 bad_fork_cleanup_keys:
1262         exit_keys(p);
1263 bad_fork_cleanup_mm:
1264         if (p->mm)
1265                 mmput(p->mm);
1266 bad_fork_cleanup_signal:
1267         cleanup_signal(p);
1268 bad_fork_cleanup_sighand:
1269         __cleanup_sighand(p->sighand);
1270 bad_fork_cleanup_fs:
1271         exit_fs(p); /* blocking */
1272 bad_fork_cleanup_files:
1273         exit_files(p); /* blocking */
1274 bad_fork_cleanup_semundo:
1275         exit_sem(p);
1276 bad_fork_cleanup_audit:
1277         audit_free(p);
1278 bad_fork_cleanup_security:
1279         security_task_free(p);
1280 bad_fork_cleanup_policy:
1281 #ifdef CONFIG_NUMA
1282         mpol_free(p->mempolicy);
1283 bad_fork_cleanup_cpuset:
1284 #endif
1285         cpuset_exit(p);
1286 bad_fork_cleanup_delays_binfmt:
1287         delayacct_tsk_free(p);
1288         if (p->binfmt)
1289                 module_put(p->binfmt->module);
1290 bad_fork_cleanup_put_domain:
1291         module_put(task_thread_info(p)->exec_domain->module);
1292 bad_fork_cleanup_count:
1293         put_group_info(p->group_info);
1294         atomic_dec(&p->user->processes);
1295         free_uid(p->user);
1296 bad_fork_free:
1297         free_task(p);
1298 fork_out:
1299         return ERR_PTR(retval);
1300 }
1301
1302 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1303 {
1304         memset(regs, 0, sizeof(struct pt_regs));
1305         return regs;
1306 }
1307
1308 struct task_struct * __devinit fork_idle(int cpu)
1309 {
1310         struct task_struct *task;
1311         struct pt_regs regs;
1312
1313         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1314         if (!task)
1315                 return ERR_PTR(-ENOMEM);
1316         init_idle(task, cpu);
1317
1318         return task;
1319 }
1320
1321 static inline int fork_traceflag (unsigned clone_flags)
1322 {
1323         if (clone_flags & CLONE_UNTRACED)
1324                 return 0;
1325         else if (clone_flags & CLONE_VFORK) {
1326                 if (current->ptrace & PT_TRACE_VFORK)
1327                         return PTRACE_EVENT_VFORK;
1328         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1329                 if (current->ptrace & PT_TRACE_CLONE)
1330                         return PTRACE_EVENT_CLONE;
1331         } else if (current->ptrace & PT_TRACE_FORK)
1332                 return PTRACE_EVENT_FORK;
1333
1334         return 0;
1335 }
1336
1337 /*
1338  *  Ok, this is the main fork-routine.
1339  *
1340  * It copies the process, and if successful kick-starts
1341  * it and waits for it to finish using the VM if required.
1342  */
1343 long do_fork(unsigned long clone_flags,
1344               unsigned long stack_start,
1345               struct pt_regs *regs,
1346               unsigned long stack_size,
1347               int __user *parent_tidptr,
1348               int __user *child_tidptr)
1349 {
1350         struct task_struct *p;
1351         int trace = 0;
1352         struct pid *pid = alloc_pid();
1353         long nr;
1354
1355         if (!pid)
1356                 return -EAGAIN;
1357         nr = pid->nr;
1358         if (unlikely(current->ptrace)) {
1359                 trace = fork_traceflag (clone_flags);
1360                 if (trace)
1361                         clone_flags |= CLONE_PTRACE;
1362         }
1363
1364         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, nr);
1365         /*
1366          * Do this prior waking up the new thread - the thread pointer
1367          * might get invalid after that point, if the thread exits quickly.
1368          */
1369         if (!IS_ERR(p)) {
1370                 struct completion vfork;
1371
1372                 if (clone_flags & CLONE_VFORK) {
1373                         p->vfork_done = &vfork;
1374                         init_completion(&vfork);
1375                 }
1376
1377                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1378                         /*
1379                          * We'll start up with an immediate SIGSTOP.
1380                          */
1381                         sigaddset(&p->pending.signal, SIGSTOP);
1382                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1383                 }
1384
1385                 if (!(clone_flags & CLONE_STOPPED))
1386                         wake_up_new_task(p, clone_flags);
1387                 else
1388                         p->state = TASK_STOPPED;
1389
1390                 if (unlikely (trace)) {
1391                         current->ptrace_message = nr;
1392                         ptrace_notify ((trace << 8) | SIGTRAP);
1393                 }
1394
1395                 if (clone_flags & CLONE_VFORK) {
1396                         wait_for_completion(&vfork);
1397                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1398                                 current->ptrace_message = nr;
1399                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1400                         }
1401                 }
1402         } else {
1403                 free_pid(pid);
1404                 nr = PTR_ERR(p);
1405         }
1406         return nr;
1407 }
1408
1409 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1410 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1411 #endif
1412
1413 static void sighand_ctor(void *data, kmem_cache_t *cachep, unsigned long flags)
1414 {
1415         struct sighand_struct *sighand = data;
1416
1417         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
1418                                         SLAB_CTOR_CONSTRUCTOR)
1419                 spin_lock_init(&sighand->siglock);
1420 }
1421
1422 void __init proc_caches_init(void)
1423 {
1424         sighand_cachep = kmem_cache_create("sighand_cache",
1425                         sizeof(struct sighand_struct), 0,
1426                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1427                         sighand_ctor, NULL);
1428         signal_cachep = kmem_cache_create("signal_cache",
1429                         sizeof(struct signal_struct), 0,
1430                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1431         files_cachep = kmem_cache_create("files_cache",
1432                         sizeof(struct files_struct), 0,
1433                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1434         fs_cachep = kmem_cache_create("fs_cache",
1435                         sizeof(struct fs_struct), 0,
1436                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1437         vm_area_cachep = kmem_cache_create("vm_area_struct",
1438                         sizeof(struct vm_area_struct), 0,
1439                         SLAB_PANIC, NULL, NULL);
1440         mm_cachep = kmem_cache_create("mm_struct",
1441                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1442                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1443 }
1444
1445
1446 /*
1447  * Check constraints on flags passed to the unshare system call and
1448  * force unsharing of additional process context as appropriate.
1449  */
1450 static inline void check_unshare_flags(unsigned long *flags_ptr)
1451 {
1452         /*
1453          * If unsharing a thread from a thread group, must also
1454          * unshare vm.
1455          */
1456         if (*flags_ptr & CLONE_THREAD)
1457                 *flags_ptr |= CLONE_VM;
1458
1459         /*
1460          * If unsharing vm, must also unshare signal handlers.
1461          */
1462         if (*flags_ptr & CLONE_VM)
1463                 *flags_ptr |= CLONE_SIGHAND;
1464
1465         /*
1466          * If unsharing signal handlers and the task was created
1467          * using CLONE_THREAD, then must unshare the thread
1468          */
1469         if ((*flags_ptr & CLONE_SIGHAND) &&
1470             (atomic_read(&current->signal->count) > 1))
1471                 *flags_ptr |= CLONE_THREAD;
1472
1473         /*
1474          * If unsharing namespace, must also unshare filesystem information.
1475          */
1476         if (*flags_ptr & CLONE_NEWNS)
1477                 *flags_ptr |= CLONE_FS;
1478 }
1479
1480 /*
1481  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1482  */
1483 static int unshare_thread(unsigned long unshare_flags)
1484 {
1485         if (unshare_flags & CLONE_THREAD)
1486                 return -EINVAL;
1487
1488         return 0;
1489 }
1490
1491 /*
1492  * Unshare the filesystem structure if it is being shared
1493  */
1494 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1495 {
1496         struct fs_struct *fs = current->fs;
1497
1498         if ((unshare_flags & CLONE_FS) &&
1499             (fs && atomic_read(&fs->count) > 1)) {
1500                 *new_fsp = __copy_fs_struct(current->fs);
1501                 if (!*new_fsp)
1502                         return -ENOMEM;
1503         }
1504
1505         return 0;
1506 }
1507
1508 /*
1509  * Unshare the namespace structure if it is being shared
1510  */
1511 static int unshare_namespace(unsigned long unshare_flags, struct namespace **new_nsp, struct fs_struct *new_fs)
1512 {
1513         struct namespace *ns = current->namespace;
1514
1515         if ((unshare_flags & CLONE_NEWNS) &&
1516             (ns && atomic_read(&ns->count) > 1)) {
1517                 if (!capable(CAP_SYS_ADMIN))
1518                         return -EPERM;
1519
1520                 *new_nsp = dup_namespace(current, new_fs ? new_fs : current->fs);
1521                 if (!*new_nsp)
1522                         return -ENOMEM;
1523         }
1524
1525         return 0;
1526 }
1527
1528 /*
1529  * Unsharing of sighand for tasks created with CLONE_SIGHAND is not
1530  * supported yet
1531  */
1532 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1533 {
1534         struct sighand_struct *sigh = current->sighand;
1535
1536         if ((unshare_flags & CLONE_SIGHAND) &&
1537             (sigh && atomic_read(&sigh->count) > 1))
1538                 return -EINVAL;
1539         else
1540                 return 0;
1541 }
1542
1543 /*
1544  * Unshare vm if it is being shared
1545  */
1546 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1547 {
1548         struct mm_struct *mm = current->mm;
1549
1550         if ((unshare_flags & CLONE_VM) &&
1551             (mm && atomic_read(&mm->mm_users) > 1)) {
1552                 return -EINVAL;
1553         }
1554
1555         return 0;
1556 }
1557
1558 /*
1559  * Unshare file descriptor table if it is being shared
1560  */
1561 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1562 {
1563         struct files_struct *fd = current->files;
1564         int error = 0;
1565
1566         if ((unshare_flags & CLONE_FILES) &&
1567             (fd && atomic_read(&fd->count) > 1)) {
1568                 *new_fdp = dup_fd(fd, &error);
1569                 if (!*new_fdp)
1570                         return error;
1571         }
1572
1573         return 0;
1574 }
1575
1576 /*
1577  * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1578  * supported yet
1579  */
1580 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1581 {
1582         if (unshare_flags & CLONE_SYSVSEM)
1583                 return -EINVAL;
1584
1585         return 0;
1586 }
1587
1588 /*
1589  * unshare allows a process to 'unshare' part of the process
1590  * context which was originally shared using clone.  copy_*
1591  * functions used by do_fork() cannot be used here directly
1592  * because they modify an inactive task_struct that is being
1593  * constructed. Here we are modifying the current, active,
1594  * task_struct.
1595  */
1596 asmlinkage long sys_unshare(unsigned long unshare_flags)
1597 {
1598         int err = 0;
1599         struct fs_struct *fs, *new_fs = NULL;
1600         struct namespace *ns, *new_ns = NULL;
1601         struct sighand_struct *sigh, *new_sigh = NULL;
1602         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1603         struct files_struct *fd, *new_fd = NULL;
1604         struct sem_undo_list *new_ulist = NULL;
1605
1606         check_unshare_flags(&unshare_flags);
1607
1608         /* Return -EINVAL for all unsupported flags */
1609         err = -EINVAL;
1610         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1611                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM))
1612                 goto bad_unshare_out;
1613
1614         if ((err = unshare_thread(unshare_flags)))
1615                 goto bad_unshare_out;
1616         if ((err = unshare_fs(unshare_flags, &new_fs)))
1617                 goto bad_unshare_cleanup_thread;
1618         if ((err = unshare_namespace(unshare_flags, &new_ns, new_fs)))
1619                 goto bad_unshare_cleanup_fs;
1620         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1621                 goto bad_unshare_cleanup_ns;
1622         if ((err = unshare_vm(unshare_flags, &new_mm)))
1623                 goto bad_unshare_cleanup_sigh;
1624         if ((err = unshare_fd(unshare_flags, &new_fd)))
1625                 goto bad_unshare_cleanup_vm;
1626         if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1627                 goto bad_unshare_cleanup_fd;
1628
1629         if (new_fs || new_ns || new_sigh || new_mm || new_fd || new_ulist) {
1630
1631                 task_lock(current);
1632
1633                 if (new_fs) {
1634                         fs = current->fs;
1635                         current->fs = new_fs;
1636                         new_fs = fs;
1637                 }
1638
1639                 if (new_ns) {
1640                         ns = current->namespace;
1641                         current->namespace = new_ns;
1642                         new_ns = ns;
1643                 }
1644
1645                 if (new_sigh) {
1646                         sigh = current->sighand;
1647                         rcu_assign_pointer(current->sighand, new_sigh);
1648                         new_sigh = sigh;
1649                 }
1650
1651                 if (new_mm) {
1652                         mm = current->mm;
1653                         active_mm = current->active_mm;
1654                         current->mm = new_mm;
1655                         current->active_mm = new_mm;
1656                         activate_mm(active_mm, new_mm);
1657                         new_mm = mm;
1658                 }
1659
1660                 if (new_fd) {
1661                         fd = current->files;
1662                         current->files = new_fd;
1663                         new_fd = fd;
1664                 }
1665
1666                 task_unlock(current);
1667         }
1668
1669 bad_unshare_cleanup_fd:
1670         if (new_fd)
1671                 put_files_struct(new_fd);
1672
1673 bad_unshare_cleanup_vm:
1674         if (new_mm)
1675                 mmput(new_mm);
1676
1677 bad_unshare_cleanup_sigh:
1678         if (new_sigh)
1679                 if (atomic_dec_and_test(&new_sigh->count))
1680                         kmem_cache_free(sighand_cachep, new_sigh);
1681
1682 bad_unshare_cleanup_ns:
1683         if (new_ns)
1684                 put_namespace(new_ns);
1685
1686 bad_unshare_cleanup_fs:
1687         if (new_fs)
1688                 put_fs_struct(new_fs);
1689
1690 bad_unshare_cleanup_thread:
1691 bad_unshare_out:
1692         return err;
1693 }