kernel/fork.c

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