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         security_task_free(tsk);
 118         free_uid(tsk->user);
 119         put_group_info(tsk->group_info);
 120
 121         if (!profile_handoff_task(tsk))
 122                 free_task(tsk);
 123 }
 124
 125 void __init fork_init(unsigned long mempages)
 126 {
 127 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 128 #ifndef ARCH_MIN_TASKALIGN
 129 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 130 #endif
 131         /* create a slab on which task_structs can be allocated */
 132         task_struct_cachep =
 133                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 134                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
 135 #endif
 136
 137         /*
 138          * The default maximum number of threads is set to a safe
 139          * value: the thread structures can take up at most half
 140          * of memory.
 141          */
 142         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 143
 144         /*
 145          * we need to allow at least 20 threads to boot a system
 146          */
 147         if(max_threads < 20)
 148                 max_threads = 20;
 149
 150         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 151         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 152         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 153                 init_task.signal->rlim[RLIMIT_NPROC];
 154 }
 155
 156 static struct task_struct *dup_task_struct(struct task_struct *orig)
 157 {
 158         struct task_struct *tsk;
 159         struct thread_info *ti;
 160
 161         prepare_to_copy(orig);
 162
 163         tsk = alloc_task_struct();
 164         if (!tsk)
 165                 return NULL;
 166
 167         ti = alloc_thread_info(tsk);
 168         if (!ti) {
 169                 free_task_struct(tsk);
 170                 return NULL;
 171         }
 172
 173         *tsk = *orig;
 174         tsk->thread_info = ti;
 175         setup_thread_stack(tsk, orig);
 176
 177         /* One for us, one for whoever does the "release_task()" (usually parent) */
 178         atomic_set(&tsk->usage,2);
 179         atomic_set(&tsk->fs_excl, 0);
 180         tsk->btrace_seq = 0;
 181         tsk->splice_pipe = NULL;
 182         return tsk;
 183 }
 184
 185 #ifdef CONFIG_MMU
 186 static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 187 {
 188         struct vm_area_struct *mpnt, *tmp, **pprev;
 189         struct rb_node **rb_link, *rb_parent;
 190         int retval;
 191         unsigned long charge;
 192         struct mempolicy *pol;
 193
 194         down_write(&oldmm->mmap_sem);
 195         flush_cache_mm(oldmm);
 196         down_write(&mm->mmap_sem);
 197
 198         mm->locked_vm = 0;
 199         mm->mmap = NULL;
 200         mm->mmap_cache = NULL;
 201         mm->free_area_cache = oldmm->mmap_base;
 202         mm->cached_hole_size = ~0UL;
 203         mm->map_count = 0;
 204         cpus_clear(mm->cpu_vm_mask);
 205         mm->mm_rb = RB_ROOT;
 206         rb_link = &mm->mm_rb.rb_node;
 207         rb_parent = NULL;
 208         pprev = &mm->mmap;
 209
 210         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 211                 struct file *file;
 212
 213                 if (mpnt->vm_flags & VM_DONTCOPY) {
 214                         long pages = vma_pages(mpnt);
 215                         mm->total_vm -= pages;
 216                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 217                                                                 -pages);
 218                         continue;
 219                 }
 220                 charge = 0;
 221                 if (mpnt->vm_flags & VM_ACCOUNT) {
 222                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 223                         if (security_vm_enough_memory(len))
 224                                 goto fail_nomem;
 225                         charge = len;
 226                 }
 227                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 228                 if (!tmp)
 229                         goto fail_nomem;
 230                 *tmp = *mpnt;
 231                 pol = mpol_copy(vma_policy(mpnt));
 232                 retval = PTR_ERR(pol);
 233                 if (IS_ERR(pol))
 234                         goto fail_nomem_policy;
 235                 vma_set_policy(tmp, pol);
 236                 tmp->vm_flags &= ~VM_LOCKED;
 237                 tmp->vm_mm = mm;
 238                 tmp->vm_next = NULL;
 239                 anon_vma_link(tmp);
 240                 file = tmp->vm_file;
 241                 if (file) {
 242                         struct inode *inode = file->f_dentry->d_inode;
 243                         get_file(file);
 244                         if (tmp->vm_flags & VM_DENYWRITE)
 245                                 atomic_dec(&inode->i_writecount);
 246
 247                         /* insert tmp into the share list, just after mpnt */
 248                         spin_lock(&file->f_mapping->i_mmap_lock);
 249                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 250                         flush_dcache_mmap_lock(file->f_mapping);
 251                         vma_prio_tree_add(tmp, mpnt);
 252                         flush_dcache_mmap_unlock(file->f_mapping);
 253                         spin_unlock(&file->f_mapping->i_mmap_lock);
 254                 }
 255
 256                 /*
 257                  * Link in the new vma and copy the page table entries.
 258                  */
 259                 *pprev = tmp;
 260                 pprev = &tmp->vm_next;
 261
 262                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 263                 rb_link = &tmp->vm_rb.rb_right;
 264                 rb_parent = &tmp->vm_rb;
 265
 266                 mm->map_count++;
 267                 retval = copy_page_range(mm, oldmm, mpnt);
 268
 269                 if (tmp->vm_ops && tmp->vm_ops->open)
 270                         tmp->vm_ops->open(tmp);
 271
 272                 if (retval)
 273                         goto out;
 274         }
 275         retval = 0;
 276 out:
 277         up_write(&mm->mmap_sem);
 278         flush_tlb_mm(oldmm);
 279         up_write(&oldmm->mmap_sem);
 280         return retval;
 281 fail_nomem_policy:
 282         kmem_cache_free(vm_area_cachep, tmp);
 283 fail_nomem:
 284         retval = -ENOMEM;
 285         vm_unacct_memory(charge);
 286         goto out;
 287 }
 288
 289 static inline int mm_alloc_pgd(struct mm_struct * mm)
 290 {
 291         mm->pgd = pgd_alloc(mm);
 292         if (unlikely(!mm->pgd))
 293                 return -ENOMEM;
 294         return 0;
 295 }
 296
 297 static inline void mm_free_pgd(struct mm_struct * mm)
 298 {
 299         pgd_free(mm->pgd);
 300 }
 301 #else
 302 #define dup_mmap(mm, oldmm)     (0)
 303 #define mm_alloc_pgd(mm)        (0)
 304 #define mm_free_pgd(mm)
 305 #endif /* CONFIG_MMU */
 306
 307  __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 308
 309 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
 310 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 311
 312 #include <linux/init_task.h>
 313
 314 static struct mm_struct * mm_init(struct mm_struct * mm)
 315 {
 316         atomic_set(&mm->mm_users, 1);
 317         atomic_set(&mm->mm_count, 1);
 318         init_rwsem(&mm->mmap_sem);
 319         INIT_LIST_HEAD(&mm->mmlist);
 320         mm->core_waiters = 0;
 321         mm->nr_ptes = 0;
 322         set_mm_counter(mm, file_rss, 0);
 323         set_mm_counter(mm, anon_rss, 0);
 324         spin_lock_init(&mm->page_table_lock);
 325         rwlock_init(&mm->ioctx_list_lock);
 326         mm->ioctx_list = NULL;
 327         mm->free_area_cache = TASK_UNMAPPED_BASE;
 328         mm->cached_hole_size = ~0UL;
 329
 330         if (likely(!mm_alloc_pgd(mm))) {
 331                 mm->def_flags = 0;
 332                 return mm;
 333         }
 334         free_mm(mm);
 335         return NULL;
 336 }
 337
 338 /*
 339  * Allocate and initialize an mm_struct.
 340  */
 341 struct mm_struct * mm_alloc(void)
 342 {
 343         struct mm_struct * mm;
 344
 345         mm = allocate_mm();
 346         if (mm) {
 347                 memset(mm, 0, sizeof(*mm));
 348                 mm = mm_init(mm);
 349         }
 350         return mm;
 351 }
 352
 353 /*
 354  * Called when the last reference to the mm
 355  * is dropped: either by a lazy thread or by
 356  * mmput. Free the page directory and the mm.
 357  */
 358 void fastcall __mmdrop(struct mm_struct *mm)
 359 {
 360         BUG_ON(mm == &init_mm);
 361         mm_free_pgd(mm);
 362         destroy_context(mm);
 363         free_mm(mm);
 364 }
 365
 366 /*
 367  * Decrement the use count and release all resources for an mm.
 368  */
 369 void mmput(struct mm_struct *mm)
 370 {
 371         might_sleep();
 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  * errorp will be valid only when the returned files_struct is NULL.
 629  */
 630 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
 631 {
 632         struct files_struct *newf;
 633         struct file **old_fds, **new_fds;
 634         int open_files, size, i, expand;
 635         struct fdtable *old_fdt, *new_fdt;
 636
 637         *errorp = -ENOMEM;
 638         newf = alloc_files();
 639         if (!newf)
 640                 goto out;
 641
 642         spin_lock(&oldf->file_lock);
 643         old_fdt = files_fdtable(oldf);
 644         new_fdt = files_fdtable(newf);
 645         size = old_fdt->max_fdset;
 646         open_files = count_open_files(old_fdt);
 647         expand = 0;
 648
 649         /*
 650          * Check whether we need to allocate a larger fd array or fd set.
 651          * Note: we're not a clone task, so the open count won't  change.
 652          */
 653         if (open_files > new_fdt->max_fdset) {
 654                 new_fdt->max_fdset = 0;
 655                 expand = 1;
 656         }
 657         if (open_files > new_fdt->max_fds) {
 658                 new_fdt->max_fds = 0;
 659                 expand = 1;
 660         }
 661
 662         /* if the old fdset gets grown now, we'll only copy up to "size" fds */
 663         if (expand) {
 664                 spin_unlock(&oldf->file_lock);
 665                 spin_lock(&newf->file_lock);
 666                 *errorp = expand_files(newf, open_files-1);
 667                 spin_unlock(&newf->file_lock);
 668                 if (*errorp < 0)
 669                         goto out_release;
 670                 new_fdt = files_fdtable(newf);
 671                 /*
 672                  * Reacquire the oldf lock and a pointer to its fd table
 673                  * who knows it may have a new bigger fd table. We need
 674                  * the latest pointer.
 675                  */
 676                 spin_lock(&oldf->file_lock);
 677                 old_fdt = files_fdtable(oldf);
 678         }
 679
 680         old_fds = old_fdt->fd;
 681         new_fds = new_fdt->fd;
 682
 683         memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
 684         memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
 685
 686         for (i = open_files; i != 0; i--) {
 687                 struct file *f = *old_fds++;
 688                 if (f) {
 689                         get_file(f);
 690                 } else {
 691                         /*
 692                          * The fd may be claimed in the fd bitmap but not yet
 693                          * instantiated in the files array if a sibling thread
 694                          * is partway through open().  So make sure that this
 695                          * fd is available to the new process.
 696                          */
 697                         FD_CLR(open_files - i, new_fdt->open_fds);
 698                 }
 699                 rcu_assign_pointer(*new_fds++, f);
 700         }
 701         spin_unlock(&oldf->file_lock);
 702
 703         /* compute the remainder to be cleared */
 704         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
 705
 706         /* This is long word aligned thus could use a optimized version */
 707         memset(new_fds, 0, size);
 708
 709         if (new_fdt->max_fdset > open_files) {
 710                 int left = (new_fdt->max_fdset-open_files)/8;
 711                 int start = open_files / (8 * sizeof(unsigned long));
 712
 713                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
 714                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
 715         }
 716
 717 out:
 718         return newf;
 719
 720 out_release:
 721         free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
 722         free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
 723         free_fd_array(new_fdt->fd, new_fdt->max_fds);
 724         kmem_cache_free(files_cachep, newf);
 725         return NULL;
 726 }
 727
 728 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 729 {
 730         struct files_struct *oldf, *newf;
 731         int error = 0;
 732
 733         /*
 734          * A background process may not have any files ...
 735          */
 736         oldf = current->files;
 737         if (!oldf)
 738                 goto out;
 739
 740         if (clone_flags & CLONE_FILES) {
 741                 atomic_inc(&oldf->count);
 742                 goto out;
 743         }
 744
 745         /*
 746          * Note: we may be using current for both targets (See exec.c)
 747          * This works because we cache current->files (old) as oldf. Don't
 748          * break this.
 749          */
 750         tsk->files = NULL;
 751         newf = dup_fd(oldf, &error);
 752         if (!newf)
 753                 goto out;
 754
 755         tsk->files = newf;
 756         error = 0;
 757 out:
 758         return error;
 759 }
 760
 761 /*
 762  *      Helper to unshare the files of the current task.
 763  *      We don't want to expose copy_files internals to
 764  *      the exec layer of the kernel.
 765  */
 766
 767 int unshare_files(void)
 768 {
 769         struct files_struct *files  = current->files;
 770         int rc;
 771
 772         BUG_ON(!files);
 773
 774         /* This can race but the race causes us to copy when we don't
 775            need to and drop the copy */
 776         if(atomic_read(&files->count) == 1)
 777         {
 778                 atomic_inc(&files->count);
 779                 return 0;
 780         }
 781         rc = copy_files(0, current);
 782         if(rc)
 783                 current->files = files;
 784         return rc;
 785 }
 786
 787 EXPORT_SYMBOL(unshare_files);
 788
 789 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 790 {
 791         struct sighand_struct *sig;
 792
 793         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 794                 atomic_inc(&current->sighand->count);
 795                 return 0;
 796         }
 797         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 798         rcu_assign_pointer(tsk->sighand, sig);
 799         if (!sig)
 800                 return -ENOMEM;
 801         atomic_set(&sig->count, 1);
 802         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 803         return 0;
 804 }
 805
 806 void __cleanup_sighand(struct sighand_struct *sighand)
 807 {
 808         if (atomic_dec_and_test(&sighand->count))
 809                 kmem_cache_free(sighand_cachep, sighand);
 810 }
 811
 812 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 813 {
 814         struct signal_struct *sig;
 815         int ret;
 816
 817         if (clone_flags & CLONE_THREAD) {
 818                 atomic_inc(&current->signal->count);
 819                 atomic_inc(&current->signal->live);
 820                 return 0;
 821         }
 822         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 823         tsk->signal = sig;
 824         if (!sig)
 825                 return -ENOMEM;
 826
 827         ret = copy_thread_group_keys(tsk);
 828         if (ret < 0) {
 829                 kmem_cache_free(signal_cachep, sig);
 830                 return ret;
 831         }
 832
 833         atomic_set(&sig->count, 1);
 834         atomic_set(&sig->live, 1);
 835         init_waitqueue_head(&sig->wait_chldexit);
 836         sig->flags = 0;
 837         sig->group_exit_code = 0;
 838         sig->group_exit_task = NULL;
 839         sig->group_stop_count = 0;
 840         sig->curr_target = NULL;
 841         init_sigpending(&sig->shared_pending);
 842         INIT_LIST_HEAD(&sig->posix_timers);
 843
 844         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_REL);
 845         sig->it_real_incr.tv64 = 0;
 846         sig->real_timer.function = it_real_fn;
 847         sig->tsk = tsk;
 848
 849         sig->it_virt_expires = cputime_zero;
 850         sig->it_virt_incr = cputime_zero;
 851         sig->it_prof_expires = cputime_zero;
 852         sig->it_prof_incr = cputime_zero;
 853
 854         sig->leader = 0;        /* session leadership doesn't inherit */
 855         sig->tty_old_pgrp = 0;
 856
 857         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 858         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 859         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 860         sig->sched_time = 0;
 861         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 862         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 863         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 864
 865         task_lock(current->group_leader);
 866         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 867         task_unlock(current->group_leader);
 868
 869         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 870                 /*
 871                  * New sole thread in the process gets an expiry time
 872                  * of the whole CPU time limit.
 873                  */
 874                 tsk->it_prof_expires =
 875                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 876         }
 877         acct_init_pacct(&sig->pacct);
 878
 879         return 0;
 880 }
 881
 882 void __cleanup_signal(struct signal_struct *sig)
 883 {
 884         exit_thread_group_keys(sig);
 885         kmem_cache_free(signal_cachep, sig);
 886 }
 887
 888 static inline void cleanup_signal(struct task_struct *tsk)
 889 {
 890         struct signal_struct *sig = tsk->signal;
 891
 892         atomic_dec(&sig->live);
 893
 894         if (atomic_dec_and_test(&sig->count))
 895                 __cleanup_signal(sig);
 896 }
 897
 898 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 899 {
 900         unsigned long new_flags = p->flags;
 901
 902         new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
 903         new_flags |= PF_FORKNOEXEC;
 904         if (!(clone_flags & CLONE_PTRACE))
 905                 p->ptrace = 0;
 906         p->flags = new_flags;
 907 }
 908
 909 asmlinkage long sys_set_tid_address(int __user *tidptr)
 910 {
 911         current->clear_child_tid = tidptr;
 912
 913         return current->pid;
 914 }
 915
 916 /*
 917  * This creates a new process as a copy of the old one,
 918  * but does not actually start it yet.
 919  *
 920  * It copies the registers, and all the appropriate
 921  * parts of the process environment (as per the clone
 922  * flags). The actual kick-off is left to the caller.
 923  */
 924 static task_t *copy_process(unsigned long clone_flags,
 925                                  unsigned long stack_start,
 926                                  struct pt_regs *regs,
 927                                  unsigned long stack_size,
 928                                  int __user *parent_tidptr,
 929                                  int __user *child_tidptr,
 930                                  int pid)
 931 {
 932         int retval;
 933         struct task_struct *p = NULL;
 934
 935         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 936                 return ERR_PTR(-EINVAL);
 937
 938         /*
 939          * Thread groups must share signals as well, and detached threads
 940          * can only be started up within the thread group.
 941          */
 942         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 943                 return ERR_PTR(-EINVAL);
 944
 945         /*
 946          * Shared signal handlers imply shared VM. By way of the above,
 947          * thread groups also imply shared VM. Blocking this case allows
 948          * for various simplifications in other code.
 949          */
 950         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 951                 return ERR_PTR(-EINVAL);
 952
 953         retval = security_task_create(clone_flags);
 954         if (retval)
 955                 goto fork_out;
 956
 957         retval = -ENOMEM;
 958         p = dup_task_struct(current);
 959         if (!p)
 960                 goto fork_out;
 961
 962         retval = -EAGAIN;
 963         if (atomic_read(&p->user->processes) >=
 964                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
 965                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 966                                 p->user != &root_user)
 967                         goto bad_fork_free;
 968         }
 969
 970         atomic_inc(&p->user->__count);
 971         atomic_inc(&p->user->processes);
 972         get_group_info(p->group_info);
 973
 974         /*
 975          * If multiple threads are within copy_process(), then this check
 976          * triggers too late. This doesn't hurt, the check is only there
 977          * to stop root fork bombs.
 978          */
 979         if (nr_threads >= max_threads)
 980                 goto bad_fork_cleanup_count;
 981
 982         if (!try_module_get(task_thread_info(p)->exec_domain->module))
 983                 goto bad_fork_cleanup_count;
 984
 985         if (p->binfmt && !try_module_get(p->binfmt->module))
 986                 goto bad_fork_cleanup_put_domain;
 987
 988         p->did_exec = 0;
 989         copy_flags(clone_flags, p);
 990         p->pid = pid;
 991         retval = -EFAULT;
 992         if (clone_flags & CLONE_PARENT_SETTID)
 993                 if (put_user(p->pid, parent_tidptr))
 994                         goto bad_fork_cleanup;
 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
1001         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1002         init_sigpending(&p->pending);
1003
1004         p->utime = cputime_zero;
1005         p->stime = cputime_zero;
1006         p->sched_time = 0;
1007         p->rchar = 0;           /* I/O counter: bytes read */
1008         p->wchar = 0;           /* I/O counter: bytes written */
1009         p->syscr = 0;           /* I/O counter: read syscalls */
1010         p->syscw = 0;           /* I/O counter: write syscalls */
1011         acct_clear_integrals(p);
1012
1013         p->it_virt_expires = cputime_zero;
1014         p->it_prof_expires = cputime_zero;
1015         p->it_sched_expires = 0;
1016         INIT_LIST_HEAD(&p->cpu_timers[0]);
1017         INIT_LIST_HEAD(&p->cpu_timers[1]);
1018         INIT_LIST_HEAD(&p->cpu_timers[2]);
1019
1020         p->lock_depth = -1;             /* -1 = no lock */
1021         do_posix_clock_monotonic_gettime(&p->start_time);
1022         p->security = NULL;
1023         p->io_context = NULL;
1024         p->io_wait = NULL;
1025         p->audit_context = NULL;
1026         cpuset_fork(p);
1027 #ifdef CONFIG_NUMA
1028         p->mempolicy = mpol_copy(p->mempolicy);
1029         if (IS_ERR(p->mempolicy)) {
1030                 retval = PTR_ERR(p->mempolicy);
1031                 p->mempolicy = NULL;
1032                 goto bad_fork_cleanup_cpuset;
1033         }
1034         mpol_fix_fork_child_flag(p);
1035 #endif
1036
1037 #ifdef CONFIG_DEBUG_MUTEXES
1038         p->blocked_on = NULL; /* not blocked yet */
1039 #endif
1040
1041         p->tgid = p->pid;
1042         if (clone_flags & CLONE_THREAD)
1043                 p->tgid = current->tgid;
1044
1045         if ((retval = security_task_alloc(p)))
1046                 goto bad_fork_cleanup_policy;
1047         if ((retval = audit_alloc(p)))
1048                 goto bad_fork_cleanup_security;
1049         /* copy all the process information */
1050         if ((retval = copy_semundo(clone_flags, p)))
1051                 goto bad_fork_cleanup_audit;
1052         if ((retval = copy_files(clone_flags, p)))
1053                 goto bad_fork_cleanup_semundo;
1054         if ((retval = copy_fs(clone_flags, p)))
1055                 goto bad_fork_cleanup_files;
1056         if ((retval = copy_sighand(clone_flags, p)))
1057                 goto bad_fork_cleanup_fs;
1058         if ((retval = copy_signal(clone_flags, p)))
1059                 goto bad_fork_cleanup_sighand;
1060         if ((retval = copy_mm(clone_flags, p)))
1061                 goto bad_fork_cleanup_signal;
1062         if ((retval = copy_keys(clone_flags, p)))
1063                 goto bad_fork_cleanup_mm;
1064         if ((retval = copy_namespace(clone_flags, p)))
1065                 goto bad_fork_cleanup_keys;
1066         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1067         if (retval)
1068                 goto bad_fork_cleanup_namespace;
1069
1070         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1071         /*
1072          * Clear TID on mm_release()?
1073          */
1074         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1075         p->robust_list = NULL;
1076 #ifdef CONFIG_COMPAT
1077         p->compat_robust_list = NULL;
1078 #endif
1079         /*
1080          * sigaltstack should be cleared when sharing the same VM
1081          */
1082         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1083                 p->sas_ss_sp = p->sas_ss_size = 0;
1084
1085         /*
1086          * Syscall tracing should be turned off in the child regardless
1087          * of CLONE_PTRACE.
1088          */
1089         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1090 #ifdef TIF_SYSCALL_EMU
1091         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1092 #endif
1093
1094         /* Our parent execution domain becomes current domain
1095            These must match for thread signalling to apply */
1096
1097         p->parent_exec_id = p->self_exec_id;
1098
1099         /* ok, now we should be set up.. */
1100         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1101         p->pdeath_signal = 0;
1102         p->exit_state = 0;
1103
1104         /*
1105          * Ok, make it visible to the rest of the system.
1106          * We dont wake it up yet.
1107          */
1108         p->group_leader = p;
1109         INIT_LIST_HEAD(&p->thread_group);
1110         INIT_LIST_HEAD(&p->ptrace_children);
1111         INIT_LIST_HEAD(&p->ptrace_list);
1112
1113         /* Perform scheduler related setup. Assign this task to a CPU. */
1114         sched_fork(p, clone_flags);
1115
1116         /* Need tasklist lock for parent etc handling! */
1117         write_lock_irq(&tasklist_lock);
1118
1119         /*
1120          * The task hasn't been attached yet, so its cpus_allowed mask will
1121          * not be changed, nor will its assigned CPU.
1122          *
1123          * The cpus_allowed mask of the parent may have changed after it was
1124          * copied first time - so re-copy it here, then check the child's CPU
1125          * to ensure it is on a valid CPU (and if not, just force it back to
1126          * parent's CPU). This avoids alot of nasty races.
1127          */
1128         p->cpus_allowed = current->cpus_allowed;
1129         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1130                         !cpu_online(task_cpu(p))))
1131                 set_task_cpu(p, smp_processor_id());
1132
1133         /* CLONE_PARENT re-uses the old parent */
1134         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1135                 p->real_parent = current->real_parent;
1136         else
1137                 p->real_parent = current;
1138         p->parent = p->real_parent;
1139
1140         spin_lock(&current->sighand->siglock);
1141
1142         /*
1143          * Process group and session signals need to be delivered to just the
1144          * parent before the fork or both the parent and the child after the
1145          * fork. Restart if a signal comes in before we add the new process to
1146          * it's process group.
1147          * A fatal signal pending means that current will exit, so the new
1148          * thread can't slip out of an OOM kill (or normal SIGKILL).
1149          */
1150         recalc_sigpending();
1151         if (signal_pending(current)) {
1152                 spin_unlock(&current->sighand->siglock);
1153                 write_unlock_irq(&tasklist_lock);
1154                 retval = -ERESTARTNOINTR;
1155                 goto bad_fork_cleanup_namespace;
1156         }
1157
1158         if (clone_flags & CLONE_THREAD) {
1159                 p->group_leader = current->group_leader;
1160                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1161
1162                 if (!cputime_eq(current->signal->it_virt_expires,
1163                                 cputime_zero) ||
1164                     !cputime_eq(current->signal->it_prof_expires,
1165                                 cputime_zero) ||
1166                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1167                     !list_empty(&current->signal->cpu_timers[0]) ||
1168                     !list_empty(&current->signal->cpu_timers[1]) ||
1169                     !list_empty(&current->signal->cpu_timers[2])) {
1170                         /*
1171                          * Have child wake up on its first tick to check
1172                          * for process CPU timers.
1173                          */
1174                         p->it_prof_expires = jiffies_to_cputime(1);
1175                 }
1176         }
1177
1178         /*
1179          * inherit ioprio
1180          */
1181         p->ioprio = current->ioprio;
1182
1183         if (likely(p->pid)) {
1184                 add_parent(p);
1185                 if (unlikely(p->ptrace & PT_PTRACED))
1186                         __ptrace_link(p, current->parent);
1187
1188                 if (thread_group_leader(p)) {
1189                         p->signal->tty = current->signal->tty;
1190                         p->signal->pgrp = process_group(current);
1191                         p->signal->session = current->signal->session;
1192                         attach_pid(p, PIDTYPE_PGID, process_group(p));
1193                         attach_pid(p, PIDTYPE_SID, p->signal->session);
1194
1195                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1196                         __get_cpu_var(process_counts)++;
1197                 }
1198                 attach_pid(p, PIDTYPE_PID, p->pid);
1199                 nr_threads++;
1200         }
1201
1202         total_forks++;
1203         spin_unlock(&current->sighand->siglock);
1204         write_unlock_irq(&tasklist_lock);
1205         proc_fork_connector(p);
1206         return p;
1207
1208 bad_fork_cleanup_namespace:
1209         exit_namespace(p);
1210 bad_fork_cleanup_keys:
1211         exit_keys(p);
1212 bad_fork_cleanup_mm:
1213         if (p->mm)
1214                 mmput(p->mm);
1215 bad_fork_cleanup_signal:
1216         cleanup_signal(p);
1217 bad_fork_cleanup_sighand:
1218         __cleanup_sighand(p->sighand);
1219 bad_fork_cleanup_fs:
1220         exit_fs(p); /* blocking */
1221 bad_fork_cleanup_files:
1222         exit_files(p); /* blocking */
1223 bad_fork_cleanup_semundo:
1224         exit_sem(p);
1225 bad_fork_cleanup_audit:
1226         audit_free(p);
1227 bad_fork_cleanup_security:
1228         security_task_free(p);
1229 bad_fork_cleanup_policy:
1230 #ifdef CONFIG_NUMA
1231         mpol_free(p->mempolicy);
1232 bad_fork_cleanup_cpuset:
1233 #endif
1234         cpuset_exit(p);
1235 bad_fork_cleanup:
1236         if (p->binfmt)
1237                 module_put(p->binfmt->module);
1238 bad_fork_cleanup_put_domain:
1239         module_put(task_thread_info(p)->exec_domain->module);
1240 bad_fork_cleanup_count:
1241         put_group_info(p->group_info);
1242         atomic_dec(&p->user->processes);
1243         free_uid(p->user);
1244 bad_fork_free:
1245         free_task(p);
1246 fork_out:
1247         return ERR_PTR(retval);
1248 }
1249
1250 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1251 {
1252         memset(regs, 0, sizeof(struct pt_regs));
1253         return regs;
1254 }
1255
1256 task_t * __devinit fork_idle(int cpu)
1257 {
1258         task_t *task;
1259         struct pt_regs regs;
1260
1261         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1262         if (!task)
1263                 return ERR_PTR(-ENOMEM);
1264         init_idle(task, cpu);
1265
1266         return task;
1267 }
1268
1269 static inline int fork_traceflag (unsigned clone_flags)
1270 {
1271         if (clone_flags & CLONE_UNTRACED)
1272                 return 0;
1273         else if (clone_flags & CLONE_VFORK) {
1274                 if (current->ptrace & PT_TRACE_VFORK)
1275                         return PTRACE_EVENT_VFORK;
1276         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1277                 if (current->ptrace & PT_TRACE_CLONE)
1278                         return PTRACE_EVENT_CLONE;
1279         } else if (current->ptrace & PT_TRACE_FORK)
1280                 return PTRACE_EVENT_FORK;
1281
1282         return 0;
1283 }
1284
1285 /*
1286  *  Ok, this is the main fork-routine.
1287  *
1288  * It copies the process, and if successful kick-starts
1289  * it and waits for it to finish using the VM if required.
1290  */
1291 long do_fork(unsigned long clone_flags,
1292               unsigned long stack_start,
1293               struct pt_regs *regs,
1294               unsigned long stack_size,
1295               int __user *parent_tidptr,
1296               int __user *child_tidptr)
1297 {
1298         struct task_struct *p;
1299         int trace = 0;
1300         struct pid *pid = alloc_pid();
1301         long nr;
1302
1303         if (!pid)
1304                 return -EAGAIN;
1305         nr = pid->nr;
1306         if (unlikely(current->ptrace)) {
1307                 trace = fork_traceflag (clone_flags);
1308                 if (trace)
1309                         clone_flags |= CLONE_PTRACE;
1310         }
1311
1312         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, nr);
1313         /*
1314          * Do this prior waking up the new thread - the thread pointer
1315          * might get invalid after that point, if the thread exits quickly.
1316          */
1317         if (!IS_ERR(p)) {
1318                 struct completion vfork;
1319
1320                 if (clone_flags & CLONE_VFORK) {
1321                         p->vfork_done = &vfork;
1322                         init_completion(&vfork);
1323                 }
1324
1325                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1326                         /*
1327                          * We'll start up with an immediate SIGSTOP.
1328                          */
1329                         sigaddset(&p->pending.signal, SIGSTOP);
1330                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1331                 }
1332
1333                 if (!(clone_flags & CLONE_STOPPED))
1334                         wake_up_new_task(p, clone_flags);
1335                 else
1336                         p->state = TASK_STOPPED;
1337
1338                 if (unlikely (trace)) {
1339                         current->ptrace_message = nr;
1340                         ptrace_notify ((trace << 8) | SIGTRAP);
1341                 }
1342
1343                 if (clone_flags & CLONE_VFORK) {
1344                         wait_for_completion(&vfork);
1345                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1346                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1347                 }
1348         } else {
1349                 free_pid(pid);
1350                 nr = PTR_ERR(p);
1351         }
1352         return nr;
1353 }
1354
1355 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1356 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1357 #endif
1358
1359 static void sighand_ctor(void *data, kmem_cache_t *cachep, unsigned long flags)
1360 {
1361         struct sighand_struct *sighand = data;
1362
1363         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
1364                                         SLAB_CTOR_CONSTRUCTOR)
1365                 spin_lock_init(&sighand->siglock);
1366 }
1367
1368 void __init proc_caches_init(void)
1369 {
1370         sighand_cachep = kmem_cache_create("sighand_cache",
1371                         sizeof(struct sighand_struct), 0,
1372                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1373                         sighand_ctor, NULL);
1374         signal_cachep = kmem_cache_create("signal_cache",
1375                         sizeof(struct signal_struct), 0,
1376                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1377         files_cachep = kmem_cache_create("files_cache",
1378                         sizeof(struct files_struct), 0,
1379                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1380         fs_cachep = kmem_cache_create("fs_cache",
1381                         sizeof(struct fs_struct), 0,
1382                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1383         vm_area_cachep = kmem_cache_create("vm_area_struct",
1384                         sizeof(struct vm_area_struct), 0,
1385                         SLAB_PANIC, NULL, NULL);
1386         mm_cachep = kmem_cache_create("mm_struct",
1387                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1388                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1389 }
1390
1391
1392 /*
1393  * Check constraints on flags passed to the unshare system call and
1394  * force unsharing of additional process context as appropriate.
1395  */
1396 static inline void check_unshare_flags(unsigned long *flags_ptr)
1397 {
1398         /*
1399          * If unsharing a thread from a thread group, must also
1400          * unshare vm.
1401          */
1402         if (*flags_ptr & CLONE_THREAD)
1403                 *flags_ptr |= CLONE_VM;
1404
1405         /*
1406          * If unsharing vm, must also unshare signal handlers.
1407          */
1408         if (*flags_ptr & CLONE_VM)
1409                 *flags_ptr |= CLONE_SIGHAND;
1410
1411         /*
1412          * If unsharing signal handlers and the task was created
1413          * using CLONE_THREAD, then must unshare the thread
1414          */
1415         if ((*flags_ptr & CLONE_SIGHAND) &&
1416             (atomic_read(&current->signal->count) > 1))
1417                 *flags_ptr |= CLONE_THREAD;
1418
1419         /*
1420          * If unsharing namespace, must also unshare filesystem information.
1421          */
1422         if (*flags_ptr & CLONE_NEWNS)
1423                 *flags_ptr |= CLONE_FS;
1424 }
1425
1426 /*
1427  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1428  */
1429 static int unshare_thread(unsigned long unshare_flags)
1430 {
1431         if (unshare_flags & CLONE_THREAD)
1432                 return -EINVAL;
1433
1434         return 0;
1435 }
1436
1437 /*
1438  * Unshare the filesystem structure if it is being shared
1439  */
1440 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1441 {
1442         struct fs_struct *fs = current->fs;
1443
1444         if ((unshare_flags & CLONE_FS) &&
1445             (fs && atomic_read(&fs->count) > 1)) {
1446                 *new_fsp = __copy_fs_struct(current->fs);
1447                 if (!*new_fsp)
1448                         return -ENOMEM;
1449         }
1450
1451         return 0;
1452 }
1453
1454 /*
1455  * Unshare the namespace structure if it is being shared
1456  */
1457 static int unshare_namespace(unsigned long unshare_flags, struct namespace **new_nsp, struct fs_struct *new_fs)
1458 {
1459         struct namespace *ns = current->namespace;
1460
1461         if ((unshare_flags & CLONE_NEWNS) &&
1462             (ns && atomic_read(&ns->count) > 1)) {
1463                 if (!capable(CAP_SYS_ADMIN))
1464                         return -EPERM;
1465
1466                 *new_nsp = dup_namespace(current, new_fs ? new_fs : current->fs);
1467                 if (!*new_nsp)
1468                         return -ENOMEM;
1469         }
1470
1471         return 0;
1472 }
1473
1474 /*
1475  * Unsharing of sighand for tasks created with CLONE_SIGHAND is not
1476  * supported yet
1477  */
1478 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1479 {
1480         struct sighand_struct *sigh = current->sighand;
1481
1482         if ((unshare_flags & CLONE_SIGHAND) &&
1483             (sigh && atomic_read(&sigh->count) > 1))
1484                 return -EINVAL;
1485         else
1486                 return 0;
1487 }
1488
1489 /*
1490  * Unshare vm if it is being shared
1491  */
1492 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1493 {
1494         struct mm_struct *mm = current->mm;
1495
1496         if ((unshare_flags & CLONE_VM) &&
1497             (mm && atomic_read(&mm->mm_users) > 1)) {
1498                 return -EINVAL;
1499         }
1500
1501         return 0;
1502 }
1503
1504 /*
1505  * Unshare file descriptor table if it is being shared
1506  */
1507 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1508 {
1509         struct files_struct *fd = current->files;
1510         int error = 0;
1511
1512         if ((unshare_flags & CLONE_FILES) &&
1513             (fd && atomic_read(&fd->count) > 1)) {
1514                 *new_fdp = dup_fd(fd, &error);
1515                 if (!*new_fdp)
1516                         return error;
1517         }
1518
1519         return 0;
1520 }
1521
1522 /*
1523  * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1524  * supported yet
1525  */
1526 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1527 {
1528         if (unshare_flags & CLONE_SYSVSEM)
1529                 return -EINVAL;
1530
1531         return 0;
1532 }
1533
1534 /*
1535  * unshare allows a process to 'unshare' part of the process
1536  * context which was originally shared using clone.  copy_*
1537  * functions used by do_fork() cannot be used here directly
1538  * because they modify an inactive task_struct that is being
1539  * constructed. Here we are modifying the current, active,
1540  * task_struct.
1541  */
1542 asmlinkage long sys_unshare(unsigned long unshare_flags)
1543 {
1544         int err = 0;
1545         struct fs_struct *fs, *new_fs = NULL;
1546         struct namespace *ns, *new_ns = NULL;
1547         struct sighand_struct *sigh, *new_sigh = NULL;
1548         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1549         struct files_struct *fd, *new_fd = NULL;
1550         struct sem_undo_list *new_ulist = NULL;
1551
1552         check_unshare_flags(&unshare_flags);
1553
1554         /* Return -EINVAL for all unsupported flags */
1555         err = -EINVAL;
1556         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1557                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM))
1558                 goto bad_unshare_out;
1559
1560         if ((err = unshare_thread(unshare_flags)))
1561                 goto bad_unshare_out;
1562         if ((err = unshare_fs(unshare_flags, &new_fs)))
1563                 goto bad_unshare_cleanup_thread;
1564         if ((err = unshare_namespace(unshare_flags, &new_ns, new_fs)))
1565                 goto bad_unshare_cleanup_fs;
1566         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1567                 goto bad_unshare_cleanup_ns;
1568         if ((err = unshare_vm(unshare_flags, &new_mm)))
1569                 goto bad_unshare_cleanup_sigh;
1570         if ((err = unshare_fd(unshare_flags, &new_fd)))
1571                 goto bad_unshare_cleanup_vm;
1572         if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1573                 goto bad_unshare_cleanup_fd;
1574
1575         if (new_fs || new_ns || new_sigh || new_mm || new_fd || new_ulist) {
1576
1577                 task_lock(current);
1578
1579                 if (new_fs) {
1580                         fs = current->fs;
1581                         current->fs = new_fs;
1582                         new_fs = fs;
1583                 }
1584
1585                 if (new_ns) {
1586                         ns = current->namespace;
1587                         current->namespace = new_ns;
1588                         new_ns = ns;
1589                 }
1590
1591                 if (new_sigh) {
1592                         sigh = current->sighand;
1593                         rcu_assign_pointer(current->sighand, new_sigh);
1594                         new_sigh = sigh;
1595                 }
1596
1597                 if (new_mm) {
1598                         mm = current->mm;
1599                         active_mm = current->active_mm;
1600                         current->mm = new_mm;
1601                         current->active_mm = new_mm;
1602                         activate_mm(active_mm, new_mm);
1603                         new_mm = mm;
1604                 }
1605
1606                 if (new_fd) {
1607                         fd = current->files;
1608                         current->files = new_fd;
1609                         new_fd = fd;
1610                 }
1611
1612                 task_unlock(current);
1613         }
1614
1615 bad_unshare_cleanup_fd:
1616         if (new_fd)
1617                 put_files_struct(new_fd);
1618
1619 bad_unshare_cleanup_vm:
1620         if (new_mm)
1621                 mmput(new_mm);
1622
1623 bad_unshare_cleanup_sigh:
1624         if (new_sigh)
1625                 if (atomic_dec_and_test(&new_sigh->count))
1626                         kmem_cache_free(sighand_cachep, new_sigh);
1627
1628 bad_unshare_cleanup_ns:
1629         if (new_ns)
1630                 put_namespace(new_ns);
1631
1632 bad_unshare_cleanup_fs:
1633         if (new_fs)
1634                 put_fs_struct(new_fs);
1635
1636 bad_unshare_cleanup_thread:
1637 bad_unshare_out:
1638         return err;
1639 }