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 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         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         if (atomic_dec_and_test(&mm->mm_users)) {
 372                 exit_aio(mm);
 373                 exit_mmap(mm);
 374                 if (!list_empty(&mm->mmlist)) {
 375                         spin_lock(&mmlist_lock);
 376                         list_del(&mm->mmlist);
 377                         spin_unlock(&mmlist_lock);
 378                 }
 379                 put_swap_token(mm);
 380                 mmdrop(mm);
 381         }
 382 }
 383 EXPORT_SYMBOL_GPL(mmput);
 384
 385 /**
 386  * get_task_mm - acquire a reference to the task's mm
 387  *
 388  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
 389  * this kernel workthread has transiently adopted a user mm with use_mm,
 390  * to do its AIO) is not set and if so returns a reference to it, after
 391  * bumping up the use count.  User must release the mm via mmput()
 392  * after use.  Typically used by /proc and ptrace.
 393  */
 394 struct mm_struct *get_task_mm(struct task_struct *task)
 395 {
 396         struct mm_struct *mm;
 397
 398         task_lock(task);
 399         mm = task->mm;
 400         if (mm) {
 401                 if (task->flags & PF_BORROWED_MM)
 402                         mm = NULL;
 403                 else
 404                         atomic_inc(&mm->mm_users);
 405         }
 406         task_unlock(task);
 407         return mm;
 408 }
 409 EXPORT_SYMBOL_GPL(get_task_mm);
 410
 411 /* Please note the differences between mmput and mm_release.
 412  * mmput is called whenever we stop holding onto a mm_struct,
 413  * error success whatever.
 414  *
 415  * mm_release is called after a mm_struct has been removed
 416  * from the current process.
 417  *
 418  * This difference is important for error handling, when we
 419  * only half set up a mm_struct for a new process and need to restore
 420  * the old one.  Because we mmput the new mm_struct before
 421  * restoring the old one. . .
 422  * Eric Biederman 10 January 1998
 423  */
 424 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 425 {
 426         struct completion *vfork_done = tsk->vfork_done;
 427
 428         /* Get rid of any cached register state */
 429         deactivate_mm(tsk, mm);
 430
 431         /* notify parent sleeping on vfork() */
 432         if (vfork_done) {
 433                 tsk->vfork_done = NULL;
 434                 complete(vfork_done);
 435         }
 436         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
 437                 u32 __user * tidptr = tsk->clear_child_tid;
 438                 tsk->clear_child_tid = NULL;
 439
 440                 /*
 441                  * We don't check the error code - if userspace has
 442                  * not set up a proper pointer then tough luck.
 443                  */
 444                 put_user(0, tidptr);
 445                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 446         }
 447 }
 448
 449 /*
 450  * Allocate a new mm structure and copy contents from the
 451  * mm structure of the passed in task structure.
 452  */
 453 static struct mm_struct *dup_mm(struct task_struct *tsk)
 454 {
 455         struct mm_struct *mm, *oldmm = current->mm;
 456         int err;
 457
 458         if (!oldmm)
 459                 return NULL;
 460
 461         mm = allocate_mm();
 462         if (!mm)
 463                 goto fail_nomem;
 464
 465         memcpy(mm, oldmm, sizeof(*mm));
 466
 467         if (!mm_init(mm))
 468                 goto fail_nomem;
 469
 470         if (init_new_context(tsk, mm))
 471                 goto fail_nocontext;
 472
 473         err = dup_mmap(mm, oldmm);
 474         if (err)
 475                 goto free_pt;
 476
 477         mm->hiwater_rss = get_mm_rss(mm);
 478         mm->hiwater_vm = mm->total_vm;
 479
 480         return mm;
 481
 482 free_pt:
 483         mmput(mm);
 484
 485 fail_nomem:
 486         return NULL;
 487
 488 fail_nocontext:
 489         /*
 490          * If init_new_context() failed, we cannot use mmput() to free the mm
 491          * because it calls destroy_context()
 492          */
 493         mm_free_pgd(mm);
 494         free_mm(mm);
 495         return NULL;
 496 }
 497
 498 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 499 {
 500         struct mm_struct * mm, *oldmm;
 501         int retval;
 502
 503         tsk->min_flt = tsk->maj_flt = 0;
 504         tsk->nvcsw = tsk->nivcsw = 0;
 505
 506         tsk->mm = NULL;
 507         tsk->active_mm = NULL;
 508
 509         /*
 510          * Are we cloning a kernel thread?
 511          *
 512          * We need to steal a active VM for that..
 513          */
 514         oldmm = current->mm;
 515         if (!oldmm)
 516                 return 0;
 517
 518         if (clone_flags & CLONE_VM) {
 519                 atomic_inc(&oldmm->mm_users);
 520                 mm = oldmm;
 521                 goto good_mm;
 522         }
 523
 524         retval = -ENOMEM;
 525         mm = dup_mm(tsk);
 526         if (!mm)
 527                 goto fail_nomem;
 528
 529 good_mm:
 530         tsk->mm = mm;
 531         tsk->active_mm = mm;
 532         return 0;
 533
 534 fail_nomem:
 535         return retval;
 536 }
 537
 538 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 539 {
 540         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 541         /* We don't need to lock fs - think why ;-) */
 542         if (fs) {
 543                 atomic_set(&fs->count, 1);
 544                 rwlock_init(&fs->lock);
 545                 fs->umask = old->umask;
 546                 read_lock(&old->lock);
 547                 fs->rootmnt = mntget(old->rootmnt);
 548                 fs->root = dget(old->root);
 549                 fs->pwdmnt = mntget(old->pwdmnt);
 550                 fs->pwd = dget(old->pwd);
 551                 if (old->altroot) {
 552                         fs->altrootmnt = mntget(old->altrootmnt);
 553                         fs->altroot = dget(old->altroot);
 554                 } else {
 555                         fs->altrootmnt = NULL;
 556                         fs->altroot = NULL;
 557                 }
 558                 read_unlock(&old->lock);
 559         }
 560         return fs;
 561 }
 562
 563 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 564 {
 565         return __copy_fs_struct(old);
 566 }
 567
 568 EXPORT_SYMBOL_GPL(copy_fs_struct);
 569
 570 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 571 {
 572         if (clone_flags & CLONE_FS) {
 573                 atomic_inc(&current->fs->count);
 574                 return 0;
 575         }
 576         tsk->fs = __copy_fs_struct(current->fs);
 577         if (!tsk->fs)
 578                 return -ENOMEM;
 579         return 0;
 580 }
 581
 582 static int count_open_files(struct fdtable *fdt)
 583 {
 584         int size = fdt->max_fdset;
 585         int i;
 586
 587         /* Find the last open fd */
 588         for (i = size/(8*sizeof(long)); i > 0; ) {
 589                 if (fdt->open_fds->fds_bits[--i])
 590                         break;
 591         }
 592         i = (i+1) * 8 * sizeof(long);
 593         return i;
 594 }
 595
 596 static struct files_struct *alloc_files(void)
 597 {
 598         struct files_struct *newf;
 599         struct fdtable *fdt;
 600
 601         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
 602         if (!newf)
 603                 goto out;
 604
 605         atomic_set(&newf->count, 1);
 606
 607         spin_lock_init(&newf->file_lock);
 608         fdt = &newf->fdtab;
 609         fdt->next_fd = 0;
 610         fdt->max_fds = NR_OPEN_DEFAULT;
 611         fdt->max_fdset = __FD_SETSIZE;
 612         fdt->close_on_exec = &newf->close_on_exec_init;
 613         fdt->open_fds = &newf->open_fds_init;
 614         fdt->fd = &newf->fd_array[0];
 615         INIT_RCU_HEAD(&fdt->rcu);
 616         fdt->free_files = NULL;
 617         fdt->next = NULL;
 618         rcu_assign_pointer(newf->fdt, fdt);
 619 out:
 620         return newf;
 621 }
 622
 623 /*
 624  * Allocate a new files structure and copy contents from the
 625  * passed in files structure.
 626  */
 627 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
 628 {
 629         struct files_struct *newf;
 630         struct file **old_fds, **new_fds;
 631         int open_files, size, i, expand;
 632         struct fdtable *old_fdt, *new_fdt;
 633
 634         newf = alloc_files();
 635         if (!newf)
 636                 goto out;
 637
 638         spin_lock(&oldf->file_lock);
 639         old_fdt = files_fdtable(oldf);
 640         new_fdt = files_fdtable(newf);
 641         size = old_fdt->max_fdset;
 642         open_files = count_open_files(old_fdt);
 643         expand = 0;
 644
 645         /*
 646          * Check whether we need to allocate a larger fd array or fd set.
 647          * Note: we're not a clone task, so the open count won't  change.
 648          */
 649         if (open_files > new_fdt->max_fdset) {
 650                 new_fdt->max_fdset = 0;
 651                 expand = 1;
 652         }
 653         if (open_files > new_fdt->max_fds) {
 654                 new_fdt->max_fds = 0;
 655                 expand = 1;
 656         }
 657
 658         /* if the old fdset gets grown now, we'll only copy up to "size" fds */
 659         if (expand) {
 660                 spin_unlock(&oldf->file_lock);
 661                 spin_lock(&newf->file_lock);
 662                 *errorp = expand_files(newf, open_files-1);
 663                 spin_unlock(&newf->file_lock);
 664                 if (*errorp < 0)
 665                         goto out_release;
 666                 new_fdt = files_fdtable(newf);
 667                 /*
 668                  * Reacquire the oldf lock and a pointer to its fd table
 669                  * who knows it may have a new bigger fd table. We need
 670                  * the latest pointer.
 671                  */
 672                 spin_lock(&oldf->file_lock);
 673                 old_fdt = files_fdtable(oldf);
 674         }
 675
 676         old_fds = old_fdt->fd;
 677         new_fds = new_fdt->fd;
 678
 679         memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
 680         memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
 681
 682         for (i = open_files; i != 0; i--) {
 683                 struct file *f = *old_fds++;
 684                 if (f) {
 685                         get_file(f);
 686                 } else {
 687                         /*
 688                          * The fd may be claimed in the fd bitmap but not yet
 689                          * instantiated in the files array if a sibling thread
 690                          * is partway through open().  So make sure that this
 691                          * fd is available to the new process.
 692                          */
 693                         FD_CLR(open_files - i, new_fdt->open_fds);
 694                 }
 695                 rcu_assign_pointer(*new_fds++, f);
 696         }
 697         spin_unlock(&oldf->file_lock);
 698
 699         /* compute the remainder to be cleared */
 700         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
 701
 702         /* This is long word aligned thus could use a optimized version */
 703         memset(new_fds, 0, size);
 704
 705         if (new_fdt->max_fdset > open_files) {
 706                 int left = (new_fdt->max_fdset-open_files)/8;
 707                 int start = open_files / (8 * sizeof(unsigned long));
 708
 709                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
 710                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
 711         }
 712
 713 out:
 714         return newf;
 715
 716 out_release:
 717         free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
 718         free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
 719         free_fd_array(new_fdt->fd, new_fdt->max_fds);
 720         kmem_cache_free(files_cachep, newf);
 721         goto out;
 722 }
 723
 724 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 725 {
 726         struct files_struct *oldf, *newf;
 727         int error = 0;
 728
 729         /*
 730          * A background process may not have any files ...
 731          */
 732         oldf = current->files;
 733         if (!oldf)
 734                 goto out;
 735
 736         if (clone_flags & CLONE_FILES) {
 737                 atomic_inc(&oldf->count);
 738                 goto out;
 739         }
 740
 741         /*
 742          * Note: we may be using current for both targets (See exec.c)
 743          * This works because we cache current->files (old) as oldf. Don't
 744          * break this.
 745          */
 746         tsk->files = NULL;
 747         error = -ENOMEM;
 748         newf = dup_fd(oldf, &error);
 749         if (!newf)
 750                 goto out;
 751
 752         tsk->files = newf;
 753         error = 0;
 754 out:
 755         return error;
 756 }
 757
 758 /*
 759  *      Helper to unshare the files of the current task.
 760  *      We don't want to expose copy_files internals to
 761  *      the exec layer of the kernel.
 762  */
 763
 764 int unshare_files(void)
 765 {
 766         struct files_struct *files  = current->files;
 767         int rc;
 768
 769         if(!files)
 770                 BUG();
 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 void sighand_free_cb(struct rcu_head *rhp)
 788 {
 789         struct sighand_struct *sp;
 790
 791         sp = container_of(rhp, struct sighand_struct, rcu);
 792         kmem_cache_free(sighand_cachep, sp);
 793 }
 794
 795 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 796 {
 797         struct sighand_struct *sig;
 798
 799         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 800                 atomic_inc(&current->sighand->count);
 801                 return 0;
 802         }
 803         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 804         rcu_assign_pointer(tsk->sighand, sig);
 805         if (!sig)
 806                 return -ENOMEM;
 807         spin_lock_init(&sig->siglock);
 808         atomic_set(&sig->count, 1);
 809         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 810         return 0;
 811 }
 812
 813 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 814 {
 815         struct signal_struct *sig;
 816         int ret;
 817
 818         if (clone_flags & CLONE_THREAD) {
 819                 atomic_inc(&current->signal->count);
 820                 atomic_inc(&current->signal->live);
 821                 return 0;
 822         }
 823         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 824         tsk->signal = sig;
 825         if (!sig)
 826                 return -ENOMEM;
 827
 828         ret = copy_thread_group_keys(tsk);
 829         if (ret < 0) {
 830                 kmem_cache_free(signal_cachep, sig);
 831                 return ret;
 832         }
 833
 834         atomic_set(&sig->count, 1);
 835         atomic_set(&sig->live, 1);
 836         init_waitqueue_head(&sig->wait_chldexit);
 837         sig->flags = 0;
 838         sig->group_exit_code = 0;
 839         sig->group_exit_task = NULL;
 840         sig->group_stop_count = 0;
 841         sig->curr_target = NULL;
 842         init_sigpending(&sig->shared_pending);
 843         INIT_LIST_HEAD(&sig->posix_timers);
 844
 845         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_REL);
 846         sig->it_real_incr.tv64 = 0;
 847         sig->real_timer.function = it_real_fn;
 848         sig->real_timer.data = tsk;
 849
 850         sig->it_virt_expires = cputime_zero;
 851         sig->it_virt_incr = cputime_zero;
 852         sig->it_prof_expires = cputime_zero;
 853         sig->it_prof_incr = cputime_zero;
 854
 855         sig->leader = 0;        /* session leadership doesn't inherit */
 856         sig->tty_old_pgrp = 0;
 857
 858         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 859         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 860         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 861         sig->sched_time = 0;
 862         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 863         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 864         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 865
 866         task_lock(current->group_leader);
 867         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 868         task_unlock(current->group_leader);
 869
 870         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 871                 /*
 872                  * New sole thread in the process gets an expiry time
 873                  * of the whole CPU time limit.
 874                  */
 875                 tsk->it_prof_expires =
 876                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 877         }
 878
 879         return 0;
 880 }
 881
 882 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 883 {
 884         unsigned long new_flags = p->flags;
 885
 886         new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
 887         new_flags |= PF_FORKNOEXEC;
 888         if (!(clone_flags & CLONE_PTRACE))
 889                 p->ptrace = 0;
 890         p->flags = new_flags;
 891 }
 892
 893 asmlinkage long sys_set_tid_address(int __user *tidptr)
 894 {
 895         current->clear_child_tid = tidptr;
 896
 897         return current->pid;
 898 }
 899
 900 /*
 901  * This creates a new process as a copy of the old one,
 902  * but does not actually start it yet.
 903  *
 904  * It copies the registers, and all the appropriate
 905  * parts of the process environment (as per the clone
 906  * flags). The actual kick-off is left to the caller.
 907  */
 908 static task_t *copy_process(unsigned long clone_flags,
 909                                  unsigned long stack_start,
 910                                  struct pt_regs *regs,
 911                                  unsigned long stack_size,
 912                                  int __user *parent_tidptr,
 913                                  int __user *child_tidptr,
 914                                  int pid)
 915 {
 916         int retval;
 917         struct task_struct *p = NULL;
 918
 919         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 920                 return ERR_PTR(-EINVAL);
 921
 922         /*
 923          * Thread groups must share signals as well, and detached threads
 924          * can only be started up within the thread group.
 925          */
 926         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 927                 return ERR_PTR(-EINVAL);
 928
 929         /*
 930          * Shared signal handlers imply shared VM. By way of the above,
 931          * thread groups also imply shared VM. Blocking this case allows
 932          * for various simplifications in other code.
 933          */
 934         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 935                 return ERR_PTR(-EINVAL);
 936
 937         retval = security_task_create(clone_flags);
 938         if (retval)
 939                 goto fork_out;
 940
 941         retval = -ENOMEM;
 942         p = dup_task_struct(current);
 943         if (!p)
 944                 goto fork_out;
 945
 946         retval = -EAGAIN;
 947         if (atomic_read(&p->user->processes) >=
 948                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
 949                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 950                                 p->user != &root_user)
 951                         goto bad_fork_free;
 952         }
 953
 954         atomic_inc(&p->user->__count);
 955         atomic_inc(&p->user->processes);
 956         get_group_info(p->group_info);
 957
 958         /*
 959          * If multiple threads are within copy_process(), then this check
 960          * triggers too late. This doesn't hurt, the check is only there
 961          * to stop root fork bombs.
 962          */
 963         if (nr_threads >= max_threads)
 964                 goto bad_fork_cleanup_count;
 965
 966         if (!try_module_get(task_thread_info(p)->exec_domain->module))
 967                 goto bad_fork_cleanup_count;
 968
 969         if (p->binfmt && !try_module_get(p->binfmt->module))
 970                 goto bad_fork_cleanup_put_domain;
 971
 972         p->did_exec = 0;
 973         copy_flags(clone_flags, p);
 974         p->pid = pid;
 975         retval = -EFAULT;
 976         if (clone_flags & CLONE_PARENT_SETTID)
 977                 if (put_user(p->pid, parent_tidptr))
 978                         goto bad_fork_cleanup;
 979
 980         p->proc_dentry = NULL;
 981
 982         INIT_LIST_HEAD(&p->children);
 983         INIT_LIST_HEAD(&p->sibling);
 984         p->vfork_done = NULL;
 985         spin_lock_init(&p->alloc_lock);
 986         spin_lock_init(&p->proc_lock);
 987
 988         clear_tsk_thread_flag(p, TIF_SIGPENDING);
 989         init_sigpending(&p->pending);
 990
 991         p->utime = cputime_zero;
 992         p->stime = cputime_zero;
 993         p->sched_time = 0;
 994         p->rchar = 0;           /* I/O counter: bytes read */
 995         p->wchar = 0;           /* I/O counter: bytes written */
 996         p->syscr = 0;           /* I/O counter: read syscalls */
 997         p->syscw = 0;           /* I/O counter: write syscalls */
 998         acct_clear_integrals(p);
 999
1000         p->it_virt_expires = cputime_zero;
1001         p->it_prof_expires = cputime_zero;
1002         p->it_sched_expires = 0;
1003         INIT_LIST_HEAD(&p->cpu_timers[0]);
1004         INIT_LIST_HEAD(&p->cpu_timers[1]);
1005         INIT_LIST_HEAD(&p->cpu_timers[2]);
1006
1007         p->lock_depth = -1;             /* -1 = no lock */
1008         do_posix_clock_monotonic_gettime(&p->start_time);
1009         p->security = NULL;
1010         p->io_context = NULL;
1011         p->io_wait = NULL;
1012         p->audit_context = NULL;
1013         cpuset_fork(p);
1014 #ifdef CONFIG_NUMA
1015         p->mempolicy = mpol_copy(p->mempolicy);
1016         if (IS_ERR(p->mempolicy)) {
1017                 retval = PTR_ERR(p->mempolicy);
1018                 p->mempolicy = NULL;
1019                 goto bad_fork_cleanup_cpuset;
1020         }
1021 #endif
1022
1023 #ifdef CONFIG_DEBUG_MUTEXES
1024         p->blocked_on = NULL; /* not blocked yet */
1025 #endif
1026
1027         p->tgid = p->pid;
1028         if (clone_flags & CLONE_THREAD)
1029                 p->tgid = current->tgid;
1030
1031         if ((retval = security_task_alloc(p)))
1032                 goto bad_fork_cleanup_policy;
1033         if ((retval = audit_alloc(p)))
1034                 goto bad_fork_cleanup_security;
1035         /* copy all the process information */
1036         if ((retval = copy_semundo(clone_flags, p)))
1037                 goto bad_fork_cleanup_audit;
1038         if ((retval = copy_files(clone_flags, p)))
1039                 goto bad_fork_cleanup_semundo;
1040         if ((retval = copy_fs(clone_flags, p)))
1041                 goto bad_fork_cleanup_files;
1042         if ((retval = copy_sighand(clone_flags, p)))
1043                 goto bad_fork_cleanup_fs;
1044         if ((retval = copy_signal(clone_flags, p)))
1045                 goto bad_fork_cleanup_sighand;
1046         if ((retval = copy_mm(clone_flags, p)))
1047                 goto bad_fork_cleanup_signal;
1048         if ((retval = copy_keys(clone_flags, p)))
1049                 goto bad_fork_cleanup_mm;
1050         if ((retval = copy_namespace(clone_flags, p)))
1051                 goto bad_fork_cleanup_keys;
1052         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1053         if (retval)
1054                 goto bad_fork_cleanup_namespace;
1055
1056         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1057         /*
1058          * Clear TID on mm_release()?
1059          */
1060         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1061
1062         /*
1063          * Syscall tracing should be turned off in the child regardless
1064          * of CLONE_PTRACE.
1065          */
1066         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1067 #ifdef TIF_SYSCALL_EMU
1068         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1069 #endif
1070
1071         /* Our parent execution domain becomes current domain
1072            These must match for thread signalling to apply */
1073
1074         p->parent_exec_id = p->self_exec_id;
1075
1076         /* ok, now we should be set up.. */
1077         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1078         p->pdeath_signal = 0;
1079         p->exit_state = 0;
1080
1081         /*
1082          * Ok, make it visible to the rest of the system.
1083          * We dont wake it up yet.
1084          */
1085         p->group_leader = p;
1086         INIT_LIST_HEAD(&p->ptrace_children);
1087         INIT_LIST_HEAD(&p->ptrace_list);
1088
1089         /* Perform scheduler related setup. Assign this task to a CPU. */
1090         sched_fork(p, clone_flags);
1091
1092         /* Need tasklist lock for parent etc handling! */
1093         write_lock_irq(&tasklist_lock);
1094
1095         /*
1096          * The task hasn't been attached yet, so its cpus_allowed mask will
1097          * not be changed, nor will its assigned CPU.
1098          *
1099          * The cpus_allowed mask of the parent may have changed after it was
1100          * copied first time - so re-copy it here, then check the child's CPU
1101          * to ensure it is on a valid CPU (and if not, just force it back to
1102          * parent's CPU). This avoids alot of nasty races.
1103          */
1104         p->cpus_allowed = current->cpus_allowed;
1105         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1106                         !cpu_online(task_cpu(p))))
1107                 set_task_cpu(p, smp_processor_id());
1108
1109         /*
1110          * Check for pending SIGKILL! The new thread should not be allowed
1111          * to slip out of an OOM kill. (or normal SIGKILL.)
1112          */
1113         if (sigismember(&current->pending.signal, SIGKILL)) {
1114                 write_unlock_irq(&tasklist_lock);
1115                 retval = -EINTR;
1116                 goto bad_fork_cleanup_namespace;
1117         }
1118
1119         /* CLONE_PARENT re-uses the old parent */
1120         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1121                 p->real_parent = current->real_parent;
1122         else
1123                 p->real_parent = current;
1124         p->parent = p->real_parent;
1125
1126         if (clone_flags & CLONE_THREAD) {
1127                 spin_lock(&current->sighand->siglock);
1128                 /*
1129                  * Important: if an exit-all has been started then
1130                  * do not create this new thread - the whole thread
1131                  * group is supposed to exit anyway.
1132                  */
1133                 if (current->signal->flags & SIGNAL_GROUP_EXIT) {
1134                         spin_unlock(&current->sighand->siglock);
1135                         write_unlock_irq(&tasklist_lock);
1136                         retval = -EAGAIN;
1137                         goto bad_fork_cleanup_namespace;
1138                 }
1139                 p->group_leader = current->group_leader;
1140
1141                 if (current->signal->group_stop_count > 0) {
1142                         /*
1143                          * There is an all-stop in progress for the group.
1144                          * We ourselves will stop as soon as we check signals.
1145                          * Make the new thread part of that group stop too.
1146                          */
1147                         current->signal->group_stop_count++;
1148                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1149                 }
1150
1151                 if (!cputime_eq(current->signal->it_virt_expires,
1152                                 cputime_zero) ||
1153                     !cputime_eq(current->signal->it_prof_expires,
1154                                 cputime_zero) ||
1155                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1156                     !list_empty(&current->signal->cpu_timers[0]) ||
1157                     !list_empty(&current->signal->cpu_timers[1]) ||
1158                     !list_empty(&current->signal->cpu_timers[2])) {
1159                         /*
1160                          * Have child wake up on its first tick to check
1161                          * for process CPU timers.
1162                          */
1163                         p->it_prof_expires = jiffies_to_cputime(1);
1164                 }
1165
1166                 spin_unlock(&current->sighand->siglock);
1167         }
1168
1169         /*
1170          * inherit ioprio
1171          */
1172         p->ioprio = current->ioprio;
1173
1174         SET_LINKS(p);
1175         if (unlikely(p->ptrace & PT_PTRACED))
1176                 __ptrace_link(p, current->parent);
1177
1178         attach_pid(p, PIDTYPE_PID, p->pid);
1179         attach_pid(p, PIDTYPE_TGID, p->tgid);
1180         if (thread_group_leader(p)) {
1181                 p->signal->tty = current->signal->tty;
1182                 p->signal->pgrp = process_group(current);
1183                 p->signal->session = current->signal->session;
1184                 attach_pid(p, PIDTYPE_PGID, process_group(p));
1185                 attach_pid(p, PIDTYPE_SID, p->signal->session);
1186                 if (p->pid)
1187                         __get_cpu_var(process_counts)++;
1188         }
1189
1190         nr_threads++;
1191         total_forks++;
1192         write_unlock_irq(&tasklist_lock);
1193         proc_fork_connector(p);
1194         return p;
1195
1196 bad_fork_cleanup_namespace:
1197         exit_namespace(p);
1198 bad_fork_cleanup_keys:
1199         exit_keys(p);
1200 bad_fork_cleanup_mm:
1201         if (p->mm)
1202                 mmput(p->mm);
1203 bad_fork_cleanup_signal:
1204         exit_signal(p);
1205 bad_fork_cleanup_sighand:
1206         exit_sighand(p);
1207 bad_fork_cleanup_fs:
1208         exit_fs(p); /* blocking */
1209 bad_fork_cleanup_files:
1210         exit_files(p); /* blocking */
1211 bad_fork_cleanup_semundo:
1212         exit_sem(p);
1213 bad_fork_cleanup_audit:
1214         audit_free(p);
1215 bad_fork_cleanup_security:
1216         security_task_free(p);
1217 bad_fork_cleanup_policy:
1218 #ifdef CONFIG_NUMA
1219         mpol_free(p->mempolicy);
1220 bad_fork_cleanup_cpuset:
1221 #endif
1222         cpuset_exit(p);
1223 bad_fork_cleanup:
1224         if (p->binfmt)
1225                 module_put(p->binfmt->module);
1226 bad_fork_cleanup_put_domain:
1227         module_put(task_thread_info(p)->exec_domain->module);
1228 bad_fork_cleanup_count:
1229         put_group_info(p->group_info);
1230         atomic_dec(&p->user->processes);
1231         free_uid(p->user);
1232 bad_fork_free:
1233         free_task(p);
1234 fork_out:
1235         return ERR_PTR(retval);
1236 }
1237
1238 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1239 {
1240         memset(regs, 0, sizeof(struct pt_regs));
1241         return regs;
1242 }
1243
1244 task_t * __devinit fork_idle(int cpu)
1245 {
1246         task_t *task;
1247         struct pt_regs regs;
1248
1249         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1250         if (!task)
1251                 return ERR_PTR(-ENOMEM);
1252         init_idle(task, cpu);
1253         unhash_process(task);
1254         return task;
1255 }
1256
1257 static inline int fork_traceflag (unsigned clone_flags)
1258 {
1259         if (clone_flags & CLONE_UNTRACED)
1260                 return 0;
1261         else if (clone_flags & CLONE_VFORK) {
1262                 if (current->ptrace & PT_TRACE_VFORK)
1263                         return PTRACE_EVENT_VFORK;
1264         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1265                 if (current->ptrace & PT_TRACE_CLONE)
1266                         return PTRACE_EVENT_CLONE;
1267         } else if (current->ptrace & PT_TRACE_FORK)
1268                 return PTRACE_EVENT_FORK;
1269
1270         return 0;
1271 }
1272
1273 /*
1274  *  Ok, this is the main fork-routine.
1275  *
1276  * It copies the process, and if successful kick-starts
1277  * it and waits for it to finish using the VM if required.
1278  */
1279 long do_fork(unsigned long clone_flags,
1280               unsigned long stack_start,
1281               struct pt_regs *regs,
1282               unsigned long stack_size,
1283               int __user *parent_tidptr,
1284               int __user *child_tidptr)
1285 {
1286         struct task_struct *p;
1287         int trace = 0;
1288         long pid = alloc_pidmap();
1289
1290         if (pid < 0)
1291                 return -EAGAIN;
1292         if (unlikely(current->ptrace)) {
1293                 trace = fork_traceflag (clone_flags);
1294                 if (trace)
1295                         clone_flags |= CLONE_PTRACE;
1296         }
1297
1298         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
1299         /*
1300          * Do this prior waking up the new thread - the thread pointer
1301          * might get invalid after that point, if the thread exits quickly.
1302          */
1303         if (!IS_ERR(p)) {
1304                 struct completion vfork;
1305
1306                 if (clone_flags & CLONE_VFORK) {
1307                         p->vfork_done = &vfork;
1308                         init_completion(&vfork);
1309                 }
1310
1311                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1312                         /*
1313                          * We'll start up with an immediate SIGSTOP.
1314                          */
1315                         sigaddset(&p->pending.signal, SIGSTOP);
1316                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1317                 }
1318
1319                 if (!(clone_flags & CLONE_STOPPED))
1320                         wake_up_new_task(p, clone_flags);
1321                 else
1322                         p->state = TASK_STOPPED;
1323
1324                 if (unlikely (trace)) {
1325                         current->ptrace_message = pid;
1326                         ptrace_notify ((trace << 8) | SIGTRAP);
1327                 }
1328
1329                 if (clone_flags & CLONE_VFORK) {
1330                         wait_for_completion(&vfork);
1331                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1332                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1333                 }
1334         } else {
1335                 free_pidmap(pid);
1336                 pid = PTR_ERR(p);
1337         }
1338         return pid;
1339 }
1340
1341 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1342 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1343 #endif
1344
1345 void __init proc_caches_init(void)
1346 {
1347         sighand_cachep = kmem_cache_create("sighand_cache",
1348                         sizeof(struct sighand_struct), 0,
1349                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1350         signal_cachep = kmem_cache_create("signal_cache",
1351                         sizeof(struct signal_struct), 0,
1352                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1353         files_cachep = kmem_cache_create("files_cache",
1354                         sizeof(struct files_struct), 0,
1355                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1356         fs_cachep = kmem_cache_create("fs_cache",
1357                         sizeof(struct fs_struct), 0,
1358                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1359         vm_area_cachep = kmem_cache_create("vm_area_struct",
1360                         sizeof(struct vm_area_struct), 0,
1361                         SLAB_PANIC, NULL, NULL);
1362         mm_cachep = kmem_cache_create("mm_struct",
1363                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1364                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1365 }
1366
1367
1368 /*
1369  * Check constraints on flags passed to the unshare system call and
1370  * force unsharing of additional process context as appropriate.
1371  */
1372 static inline void check_unshare_flags(unsigned long *flags_ptr)
1373 {
1374         /*
1375          * If unsharing a thread from a thread group, must also
1376          * unshare vm.
1377          */
1378         if (*flags_ptr & CLONE_THREAD)
1379                 *flags_ptr |= CLONE_VM;
1380
1381         /*
1382          * If unsharing vm, must also unshare signal handlers.
1383          */
1384         if (*flags_ptr & CLONE_VM)
1385                 *flags_ptr |= CLONE_SIGHAND;
1386
1387         /*
1388          * If unsharing signal handlers and the task was created
1389          * using CLONE_THREAD, then must unshare the thread
1390          */
1391         if ((*flags_ptr & CLONE_SIGHAND) &&
1392             (atomic_read(&current->signal->count) > 1))
1393                 *flags_ptr |= CLONE_THREAD;
1394
1395         /*
1396          * If unsharing namespace, must also unshare filesystem information.
1397          */
1398         if (*flags_ptr & CLONE_NEWNS)
1399                 *flags_ptr |= CLONE_FS;
1400 }
1401
1402 /*
1403  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1404  */
1405 static int unshare_thread(unsigned long unshare_flags)
1406 {
1407         if (unshare_flags & CLONE_THREAD)
1408                 return -EINVAL;
1409
1410         return 0;
1411 }
1412
1413 /*
1414  * Unshare the filesystem structure if it is being shared
1415  */
1416 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1417 {
1418         struct fs_struct *fs = current->fs;
1419
1420         if ((unshare_flags & CLONE_FS) &&
1421             (fs && atomic_read(&fs->count) > 1)) {
1422                 *new_fsp = __copy_fs_struct(current->fs);
1423                 if (!*new_fsp)
1424                         return -ENOMEM;
1425         }
1426
1427         return 0;
1428 }
1429
1430 /*
1431  * Unshare the namespace structure if it is being shared
1432  */
1433 static int unshare_namespace(unsigned long unshare_flags, struct namespace **new_nsp, struct fs_struct *new_fs)
1434 {
1435         struct namespace *ns = current->namespace;
1436
1437         if ((unshare_flags & CLONE_NEWNS) &&
1438             (ns && atomic_read(&ns->count) > 1)) {
1439                 if (!capable(CAP_SYS_ADMIN))
1440                         return -EPERM;
1441
1442                 *new_nsp = dup_namespace(current, new_fs ? new_fs : current->fs);
1443                 if (!*new_nsp)
1444                         return -ENOMEM;
1445         }
1446
1447         return 0;
1448 }
1449
1450 /*
1451  * Unsharing of sighand for tasks created with CLONE_SIGHAND is not
1452  * supported yet
1453  */
1454 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1455 {
1456         struct sighand_struct *sigh = current->sighand;
1457
1458         if ((unshare_flags & CLONE_SIGHAND) &&
1459             (sigh && atomic_read(&sigh->count) > 1))
1460                 return -EINVAL;
1461         else
1462                 return 0;
1463 }
1464
1465 /*
1466  * Unshare vm if it is being shared
1467  */
1468 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1469 {
1470         struct mm_struct *mm = current->mm;
1471
1472         if ((unshare_flags & CLONE_VM) &&
1473             (mm && atomic_read(&mm->mm_users) > 1)) {
1474                 *new_mmp = dup_mm(current);
1475                 if (!*new_mmp)
1476                         return -ENOMEM;
1477         }
1478
1479         return 0;
1480 }
1481
1482 /*
1483  * Unshare file descriptor table if it is being shared
1484  */
1485 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1486 {
1487         struct files_struct *fd = current->files;
1488         int error = 0;
1489
1490         if ((unshare_flags & CLONE_FILES) &&
1491             (fd && atomic_read(&fd->count) > 1)) {
1492                 *new_fdp = dup_fd(fd, &error);
1493                 if (!*new_fdp)
1494                         return error;
1495         }
1496
1497         return 0;
1498 }
1499
1500 /*
1501  * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1502  * supported yet
1503  */
1504 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1505 {
1506         if (unshare_flags & CLONE_SYSVSEM)
1507                 return -EINVAL;
1508
1509         return 0;
1510 }
1511
1512 /*
1513  * unshare allows a process to 'unshare' part of the process
1514  * context which was originally shared using clone.  copy_*
1515  * functions used by do_fork() cannot be used here directly
1516  * because they modify an inactive task_struct that is being
1517  * constructed. Here we are modifying the current, active,
1518  * task_struct.
1519  */
1520 asmlinkage long sys_unshare(unsigned long unshare_flags)
1521 {
1522         int err = 0;
1523         struct fs_struct *fs, *new_fs = NULL;
1524         struct namespace *ns, *new_ns = NULL;
1525         struct sighand_struct *sigh, *new_sigh = NULL;
1526         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1527         struct files_struct *fd, *new_fd = NULL;
1528         struct sem_undo_list *new_ulist = NULL;
1529
1530         check_unshare_flags(&unshare_flags);
1531
1532         if ((err = unshare_thread(unshare_flags)))
1533                 goto bad_unshare_out;
1534         if ((err = unshare_fs(unshare_flags, &new_fs)))
1535                 goto bad_unshare_cleanup_thread;
1536         if ((err = unshare_namespace(unshare_flags, &new_ns, new_fs)))
1537                 goto bad_unshare_cleanup_fs;
1538         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1539                 goto bad_unshare_cleanup_ns;
1540         if ((err = unshare_vm(unshare_flags, &new_mm)))
1541                 goto bad_unshare_cleanup_sigh;
1542         if ((err = unshare_fd(unshare_flags, &new_fd)))
1543                 goto bad_unshare_cleanup_vm;
1544         if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1545                 goto bad_unshare_cleanup_fd;
1546
1547         if (new_fs || new_ns || new_sigh || new_mm || new_fd || new_ulist) {
1548
1549                 task_lock(current);
1550
1551                 if (new_fs) {
1552                         fs = current->fs;
1553                         current->fs = new_fs;
1554                         new_fs = fs;
1555                 }
1556
1557                 if (new_ns) {
1558                         ns = current->namespace;
1559                         current->namespace = new_ns;
1560                         new_ns = ns;
1561                 }
1562
1563                 if (new_sigh) {
1564                         sigh = current->sighand;
1565                         current->sighand = new_sigh;
1566                         new_sigh = sigh;
1567                 }
1568
1569                 if (new_mm) {
1570                         mm = current->mm;
1571                         active_mm = current->active_mm;
1572                         current->mm = new_mm;
1573                         current->active_mm = new_mm;
1574                         activate_mm(active_mm, new_mm);
1575                         new_mm = mm;
1576                 }
1577
1578                 if (new_fd) {
1579                         fd = current->files;
1580                         current->files = new_fd;
1581                         new_fd = fd;
1582                 }
1583
1584                 task_unlock(current);
1585         }
1586
1587 bad_unshare_cleanup_fd:
1588         if (new_fd)
1589                 put_files_struct(new_fd);
1590
1591 bad_unshare_cleanup_vm:
1592         if (new_mm)
1593                 mmput(new_mm);
1594
1595 bad_unshare_cleanup_sigh:
1596         if (new_sigh)
1597                 if (atomic_dec_and_test(&new_sigh->count))
1598                         kmem_cache_free(sighand_cachep, new_sigh);
1599
1600 bad_unshare_cleanup_ns:
1601         if (new_ns)
1602                 put_namespace(new_ns);
1603
1604 bad_unshare_cleanup_fs:
1605         if (new_fs)
1606                 put_fs_struct(new_fs);
1607
1608 bad_unshare_cleanup_thread:
1609 bad_unshare_out:
1610         return err;
1611 }