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