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         rt_mutex_debug_task_free(tsk);
 108         free_task_struct(tsk);
 109 }
 110 EXPORT_SYMBOL(free_task);
 111
 112 void __put_task_struct(struct task_struct *tsk)
 113 {
 114         WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
 115         WARN_ON(atomic_read(&tsk->usage));
 116         WARN_ON(tsk == current);
 117
 118         security_task_free(tsk);
 119         free_uid(tsk->user);
 120         put_group_info(tsk->group_info);
 121
 122         if (!profile_handoff_task(tsk))
 123                 free_task(tsk);
 124 }
 125
 126 void __init fork_init(unsigned long mempages)
 127 {
 128 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 129 #ifndef ARCH_MIN_TASKALIGN
 130 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 131 #endif
 132         /* create a slab on which task_structs can be allocated */
 133         task_struct_cachep =
 134                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 135                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
 136 #endif
 137
 138         /*
 139          * The default maximum number of threads is set to a safe
 140          * value: the thread structures can take up at most half
 141          * of memory.
 142          */
 143         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 144
 145         /*
 146          * we need to allow at least 20 threads to boot a system
 147          */
 148         if(max_threads < 20)
 149                 max_threads = 20;
 150
 151         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 152         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 153         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 154                 init_task.signal->rlim[RLIMIT_NPROC];
 155 }
 156
 157 static struct task_struct *dup_task_struct(struct task_struct *orig)
 158 {
 159         struct task_struct *tsk;
 160         struct thread_info *ti;
 161
 162         prepare_to_copy(orig);
 163
 164         tsk = alloc_task_struct();
 165         if (!tsk)
 166                 return NULL;
 167
 168         ti = alloc_thread_info(tsk);
 169         if (!ti) {
 170                 free_task_struct(tsk);
 171                 return NULL;
 172         }
 173
 174         *tsk = *orig;
 175         tsk->thread_info = ti;
 176         setup_thread_stack(tsk, orig);
 177
 178         /* One for us, one for whoever does the "release_task()" (usually parent) */
 179         atomic_set(&tsk->usage,2);
 180         atomic_set(&tsk->fs_excl, 0);
 181         tsk->btrace_seq = 0;
 182         tsk->splice_pipe = NULL;
 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         might_sleep();
 373
 374         if (atomic_dec_and_test(&mm->mm_users)) {
 375                 exit_aio(mm);
 376                 exit_mmap(mm);
 377                 if (!list_empty(&mm->mmlist)) {
 378                         spin_lock(&mmlist_lock);
 379                         list_del(&mm->mmlist);
 380                         spin_unlock(&mmlist_lock);
 381                 }
 382                 put_swap_token(mm);
 383                 mmdrop(mm);
 384         }
 385 }
 386 EXPORT_SYMBOL_GPL(mmput);
 387
 388 /**
 389  * get_task_mm - acquire a reference to the task's mm
 390  *
 391  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
 392  * this kernel workthread has transiently adopted a user mm with use_mm,
 393  * to do its AIO) is not set and if so returns a reference to it, after
 394  * bumping up the use count.  User must release the mm via mmput()
 395  * after use.  Typically used by /proc and ptrace.
 396  */
 397 struct mm_struct *get_task_mm(struct task_struct *task)
 398 {
 399         struct mm_struct *mm;
 400
 401         task_lock(task);
 402         mm = task->mm;
 403         if (mm) {
 404                 if (task->flags & PF_BORROWED_MM)
 405                         mm = NULL;
 406                 else
 407                         atomic_inc(&mm->mm_users);
 408         }
 409         task_unlock(task);
 410         return mm;
 411 }
 412 EXPORT_SYMBOL_GPL(get_task_mm);
 413
 414 /* Please note the differences between mmput and mm_release.
 415  * mmput is called whenever we stop holding onto a mm_struct,
 416  * error success whatever.
 417  *
 418  * mm_release is called after a mm_struct has been removed
 419  * from the current process.
 420  *
 421  * This difference is important for error handling, when we
 422  * only half set up a mm_struct for a new process and need to restore
 423  * the old one.  Because we mmput the new mm_struct before
 424  * restoring the old one. . .
 425  * Eric Biederman 10 January 1998
 426  */
 427 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 428 {
 429         struct completion *vfork_done = tsk->vfork_done;
 430
 431         /* Get rid of any cached register state */
 432         deactivate_mm(tsk, mm);
 433
 434         /* notify parent sleeping on vfork() */
 435         if (vfork_done) {
 436                 tsk->vfork_done = NULL;
 437                 complete(vfork_done);
 438         }
 439         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
 440                 u32 __user * tidptr = tsk->clear_child_tid;
 441                 tsk->clear_child_tid = NULL;
 442
 443                 /*
 444                  * We don't check the error code - if userspace has
 445                  * not set up a proper pointer then tough luck.
 446                  */
 447                 put_user(0, tidptr);
 448                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 449         }
 450 }
 451
 452 /*
 453  * Allocate a new mm structure and copy contents from the
 454  * mm structure of the passed in task structure.
 455  */
 456 static struct mm_struct *dup_mm(struct task_struct *tsk)
 457 {
 458         struct mm_struct *mm, *oldmm = current->mm;
 459         int err;
 460
 461         if (!oldmm)
 462                 return NULL;
 463
 464         mm = allocate_mm();
 465         if (!mm)
 466                 goto fail_nomem;
 467
 468         memcpy(mm, oldmm, sizeof(*mm));
 469
 470         if (!mm_init(mm))
 471                 goto fail_nomem;
 472
 473         if (init_new_context(tsk, mm))
 474                 goto fail_nocontext;
 475
 476         err = dup_mmap(mm, oldmm);
 477         if (err)
 478                 goto free_pt;
 479
 480         mm->hiwater_rss = get_mm_rss(mm);
 481         mm->hiwater_vm = mm->total_vm;
 482
 483         return mm;
 484
 485 free_pt:
 486         mmput(mm);
 487
 488 fail_nomem:
 489         return NULL;
 490
 491 fail_nocontext:
 492         /*
 493          * If init_new_context() failed, we cannot use mmput() to free the mm
 494          * because it calls destroy_context()
 495          */
 496         mm_free_pgd(mm);
 497         free_mm(mm);
 498         return NULL;
 499 }
 500
 501 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 502 {
 503         struct mm_struct * mm, *oldmm;
 504         int retval;
 505
 506         tsk->min_flt = tsk->maj_flt = 0;
 507         tsk->nvcsw = tsk->nivcsw = 0;
 508
 509         tsk->mm = NULL;
 510         tsk->active_mm = NULL;
 511
 512         /*
 513          * Are we cloning a kernel thread?
 514          *
 515          * We need to steal a active VM for that..
 516          */
 517         oldmm = current->mm;
 518         if (!oldmm)
 519                 return 0;
 520
 521         if (clone_flags & CLONE_VM) {
 522                 atomic_inc(&oldmm->mm_users);
 523                 mm = oldmm;
 524                 goto good_mm;
 525         }
 526
 527         retval = -ENOMEM;
 528         mm = dup_mm(tsk);
 529         if (!mm)
 530                 goto fail_nomem;
 531
 532 good_mm:
 533         tsk->mm = mm;
 534         tsk->active_mm = mm;
 535         return 0;
 536
 537 fail_nomem:
 538         return retval;
 539 }
 540
 541 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 542 {
 543         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 544         /* We don't need to lock fs - think why ;-) */
 545         if (fs) {
 546                 atomic_set(&fs->count, 1);
 547                 rwlock_init(&fs->lock);
 548                 fs->umask = old->umask;
 549                 read_lock(&old->lock);
 550                 fs->rootmnt = mntget(old->rootmnt);
 551                 fs->root = dget(old->root);
 552                 fs->pwdmnt = mntget(old->pwdmnt);
 553                 fs->pwd = dget(old->pwd);
 554                 if (old->altroot) {
 555                         fs->altrootmnt = mntget(old->altrootmnt);
 556                         fs->altroot = dget(old->altroot);
 557                 } else {
 558                         fs->altrootmnt = NULL;
 559                         fs->altroot = NULL;
 560                 }
 561                 read_unlock(&old->lock);
 562         }
 563         return fs;
 564 }
 565
 566 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 567 {
 568         return __copy_fs_struct(old);
 569 }
 570
 571 EXPORT_SYMBOL_GPL(copy_fs_struct);
 572
 573 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 574 {
 575         if (clone_flags & CLONE_FS) {
 576                 atomic_inc(&current->fs->count);
 577                 return 0;
 578         }
 579         tsk->fs = __copy_fs_struct(current->fs);
 580         if (!tsk->fs)
 581                 return -ENOMEM;
 582         return 0;
 583 }
 584
 585 static int count_open_files(struct fdtable *fdt)
 586 {
 587         int size = fdt->max_fdset;
 588         int i;
 589
 590         /* Find the last open fd */
 591         for (i = size/(8*sizeof(long)); i > 0; ) {
 592                 if (fdt->open_fds->fds_bits[--i])
 593                         break;
 594         }
 595         i = (i+1) * 8 * sizeof(long);
 596         return i;
 597 }
 598
 599 static struct files_struct *alloc_files(void)
 600 {
 601         struct files_struct *newf;
 602         struct fdtable *fdt;
 603
 604         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
 605         if (!newf)
 606                 goto out;
 607
 608         atomic_set(&newf->count, 1);
 609
 610         spin_lock_init(&newf->file_lock);
 611         newf->next_fd = 0;
 612         fdt = &newf->fdtab;
 613         fdt->max_fds = NR_OPEN_DEFAULT;
 614         fdt->max_fdset = EMBEDDED_FD_SET_SIZE;
 615         fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
 616         fdt->open_fds = (fd_set *)&newf->open_fds_init;
 617         fdt->fd = &newf->fd_array[0];
 618         INIT_RCU_HEAD(&fdt->rcu);
 619         fdt->free_files = NULL;
 620         fdt->next = NULL;
 621         rcu_assign_pointer(newf->fdt, fdt);
 622 out:
 623         return newf;
 624 }
 625
 626 /*
 627  * Allocate a new files structure and copy contents from the
 628  * passed in files structure.
 629  * errorp will be valid only when the returned files_struct is NULL.
 630  */
 631 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
 632 {
 633         struct files_struct *newf;
 634         struct file **old_fds, **new_fds;
 635         int open_files, size, i, expand;
 636         struct fdtable *old_fdt, *new_fdt;
 637
 638         *errorp = -ENOMEM;
 639         newf = alloc_files();
 640         if (!newf)
 641                 goto out;
 642
 643         spin_lock(&oldf->file_lock);
 644         old_fdt = files_fdtable(oldf);
 645         new_fdt = files_fdtable(newf);
 646         size = old_fdt->max_fdset;
 647         open_files = count_open_files(old_fdt);
 648         expand = 0;
 649
 650         /*
 651          * Check whether we need to allocate a larger fd array or fd set.
 652          * Note: we're not a clone task, so the open count won't  change.
 653          */
 654         if (open_files > new_fdt->max_fdset) {
 655                 new_fdt->max_fdset = 0;
 656                 expand = 1;
 657         }
 658         if (open_files > new_fdt->max_fds) {
 659                 new_fdt->max_fds = 0;
 660                 expand = 1;
 661         }
 662
 663         /* if the old fdset gets grown now, we'll only copy up to "size" fds */
 664         if (expand) {
 665                 spin_unlock(&oldf->file_lock);
 666                 spin_lock(&newf->file_lock);
 667                 *errorp = expand_files(newf, open_files-1);
 668                 spin_unlock(&newf->file_lock);
 669                 if (*errorp < 0)
 670                         goto out_release;
 671                 new_fdt = files_fdtable(newf);
 672                 /*
 673                  * Reacquire the oldf lock and a pointer to its fd table
 674                  * who knows it may have a new bigger fd table. We need
 675                  * the latest pointer.
 676                  */
 677                 spin_lock(&oldf->file_lock);
 678                 old_fdt = files_fdtable(oldf);
 679         }
 680
 681         old_fds = old_fdt->fd;
 682         new_fds = new_fdt->fd;
 683
 684         memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
 685         memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
 686
 687         for (i = open_files; i != 0; i--) {
 688                 struct file *f = *old_fds++;
 689                 if (f) {
 690                         get_file(f);
 691                 } else {
 692                         /*
 693                          * The fd may be claimed in the fd bitmap but not yet
 694                          * instantiated in the files array if a sibling thread
 695                          * is partway through open().  So make sure that this
 696                          * fd is available to the new process.
 697                          */
 698                         FD_CLR(open_files - i, new_fdt->open_fds);
 699                 }
 700                 rcu_assign_pointer(*new_fds++, f);
 701         }
 702         spin_unlock(&oldf->file_lock);
 703
 704         /* compute the remainder to be cleared */
 705         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
 706
 707         /* This is long word aligned thus could use a optimized version */
 708         memset(new_fds, 0, size);
 709
 710         if (new_fdt->max_fdset > open_files) {
 711                 int left = (new_fdt->max_fdset-open_files)/8;
 712                 int start = open_files / (8 * sizeof(unsigned long));
 713
 714                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
 715                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
 716         }
 717
 718 out:
 719         return newf;
 720
 721 out_release:
 722         free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
 723         free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
 724         free_fd_array(new_fdt->fd, new_fdt->max_fds);
 725         kmem_cache_free(files_cachep, newf);
 726         return NULL;
 727 }
 728
 729 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 730 {
 731         struct files_struct *oldf, *newf;
 732         int error = 0;
 733
 734         /*
 735          * A background process may not have any files ...
 736          */
 737         oldf = current->files;
 738         if (!oldf)
 739                 goto out;
 740
 741         if (clone_flags & CLONE_FILES) {
 742                 atomic_inc(&oldf->count);
 743                 goto out;
 744         }
 745
 746         /*
 747          * Note: we may be using current for both targets (See exec.c)
 748          * This works because we cache current->files (old) as oldf. Don't
 749          * break this.
 750          */
 751         tsk->files = NULL;
 752         newf = dup_fd(oldf, &error);
 753         if (!newf)
 754                 goto out;
 755
 756         tsk->files = newf;
 757         error = 0;
 758 out:
 759         return error;
 760 }
 761
 762 /*
 763  *      Helper to unshare the files of the current task.
 764  *      We don't want to expose copy_files internals to
 765  *      the exec layer of the kernel.
 766  */
 767
 768 int unshare_files(void)
 769 {
 770         struct files_struct *files  = current->files;
 771         int rc;
 772
 773         BUG_ON(!files);
 774
 775         /* This can race but the race causes us to copy when we don't
 776            need to and drop the copy */
 777         if(atomic_read(&files->count) == 1)
 778         {
 779                 atomic_inc(&files->count);
 780                 return 0;
 781         }
 782         rc = copy_files(0, current);
 783         if(rc)
 784                 current->files = files;
 785         return rc;
 786 }
 787
 788 EXPORT_SYMBOL(unshare_files);
 789
 790 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 791 {
 792         struct sighand_struct *sig;
 793
 794         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 795                 atomic_inc(&current->sighand->count);
 796                 return 0;
 797         }
 798         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 799         rcu_assign_pointer(tsk->sighand, sig);
 800         if (!sig)
 801                 return -ENOMEM;
 802         atomic_set(&sig->count, 1);
 803         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 804         return 0;
 805 }
 806
 807 void __cleanup_sighand(struct sighand_struct *sighand)
 808 {
 809         if (atomic_dec_and_test(&sighand->count))
 810                 kmem_cache_free(sighand_cachep, sighand);
 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->tsk = 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         acct_init_pacct(&sig->pacct);
 879
 880         return 0;
 881 }
 882
 883 void __cleanup_signal(struct signal_struct *sig)
 884 {
 885         exit_thread_group_keys(sig);
 886         kmem_cache_free(signal_cachep, sig);
 887 }
 888
 889 static inline void cleanup_signal(struct task_struct *tsk)
 890 {
 891         struct signal_struct *sig = tsk->signal;
 892
 893         atomic_dec(&sig->live);
 894
 895         if (atomic_dec_and_test(&sig->count))
 896                 __cleanup_signal(sig);
 897 }
 898
 899 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 900 {
 901         unsigned long new_flags = p->flags;
 902
 903         new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
 904         new_flags |= PF_FORKNOEXEC;
 905         if (!(clone_flags & CLONE_PTRACE))
 906                 p->ptrace = 0;
 907         p->flags = new_flags;
 908 }
 909
 910 asmlinkage long sys_set_tid_address(int __user *tidptr)
 911 {
 912         current->clear_child_tid = tidptr;
 913
 914         return current->pid;
 915 }
 916
 917 static inline void rt_mutex_init_task(struct task_struct *p)
 918 {
 919 #ifdef CONFIG_RT_MUTEXES
 920         spin_lock_init(&p->pi_lock);
 921         plist_head_init(&p->pi_waiters, &p->pi_lock);
 922         p->pi_blocked_on = NULL;
 923 # ifdef CONFIG_DEBUG_RT_MUTEXES
 924         spin_lock_init(&p->held_list_lock);
 925         INIT_LIST_HEAD(&p->held_list_head);
 926 # endif
 927 #endif
 928 }
 929
 930 /*
 931  * This creates a new process as a copy of the old one,
 932  * but does not actually start it yet.
 933  *
 934  * It copies the registers, and all the appropriate
 935  * parts of the process environment (as per the clone
 936  * flags). The actual kick-off is left to the caller.
 937  */
 938 static task_t *copy_process(unsigned long clone_flags,
 939                                  unsigned long stack_start,
 940                                  struct pt_regs *regs,
 941                                  unsigned long stack_size,
 942                                  int __user *parent_tidptr,
 943                                  int __user *child_tidptr,
 944                                  int pid)
 945 {
 946         int retval;
 947         struct task_struct *p = NULL;
 948
 949         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 950                 return ERR_PTR(-EINVAL);
 951
 952         /*
 953          * Thread groups must share signals as well, and detached threads
 954          * can only be started up within the thread group.
 955          */
 956         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 957                 return ERR_PTR(-EINVAL);
 958
 959         /*
 960          * Shared signal handlers imply shared VM. By way of the above,
 961          * thread groups also imply shared VM. Blocking this case allows
 962          * for various simplifications in other code.
 963          */
 964         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 965                 return ERR_PTR(-EINVAL);
 966
 967         retval = security_task_create(clone_flags);
 968         if (retval)
 969                 goto fork_out;
 970
 971         retval = -ENOMEM;
 972         p = dup_task_struct(current);
 973         if (!p)
 974                 goto fork_out;
 975
 976         retval = -EAGAIN;
 977         if (atomic_read(&p->user->processes) >=
 978                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
 979                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 980                                 p->user != &root_user)
 981                         goto bad_fork_free;
 982         }
 983
 984         atomic_inc(&p->user->__count);
 985         atomic_inc(&p->user->processes);
 986         get_group_info(p->group_info);
 987
 988         /*
 989          * If multiple threads are within copy_process(), then this check
 990          * triggers too late. This doesn't hurt, the check is only there
 991          * to stop root fork bombs.
 992          */
 993         if (nr_threads >= max_threads)
 994                 goto bad_fork_cleanup_count;
 995
 996         if (!try_module_get(task_thread_info(p)->exec_domain->module))
 997                 goto bad_fork_cleanup_count;
 998
 999         if (p->binfmt && !try_module_get(p->binfmt->module))
1000                 goto bad_fork_cleanup_put_domain;
1001
1002         p->did_exec = 0;
1003         copy_flags(clone_flags, p);
1004         p->pid = pid;
1005         retval = -EFAULT;
1006         if (clone_flags & CLONE_PARENT_SETTID)
1007                 if (put_user(p->pid, parent_tidptr))
1008                         goto bad_fork_cleanup;
1009
1010         INIT_LIST_HEAD(&p->children);
1011         INIT_LIST_HEAD(&p->sibling);
1012         p->vfork_done = NULL;
1013         spin_lock_init(&p->alloc_lock);
1014
1015         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1016         init_sigpending(&p->pending);
1017
1018         p->utime = cputime_zero;
1019         p->stime = cputime_zero;
1020         p->sched_time = 0;
1021         p->rchar = 0;           /* I/O counter: bytes read */
1022         p->wchar = 0;           /* I/O counter: bytes written */
1023         p->syscr = 0;           /* I/O counter: read syscalls */
1024         p->syscw = 0;           /* I/O counter: write syscalls */
1025         acct_clear_integrals(p);
1026
1027         p->it_virt_expires = cputime_zero;
1028         p->it_prof_expires = cputime_zero;
1029         p->it_sched_expires = 0;
1030         INIT_LIST_HEAD(&p->cpu_timers[0]);
1031         INIT_LIST_HEAD(&p->cpu_timers[1]);
1032         INIT_LIST_HEAD(&p->cpu_timers[2]);
1033
1034         p->lock_depth = -1;             /* -1 = no lock */
1035         do_posix_clock_monotonic_gettime(&p->start_time);
1036         p->security = NULL;
1037         p->io_context = NULL;
1038         p->io_wait = NULL;
1039         p->audit_context = NULL;
1040         cpuset_fork(p);
1041 #ifdef CONFIG_NUMA
1042         p->mempolicy = mpol_copy(p->mempolicy);
1043         if (IS_ERR(p->mempolicy)) {
1044                 retval = PTR_ERR(p->mempolicy);
1045                 p->mempolicy = NULL;
1046                 goto bad_fork_cleanup_cpuset;
1047         }
1048         mpol_fix_fork_child_flag(p);
1049 #endif
1050
1051         rt_mutex_init_task(p);
1052
1053 #ifdef CONFIG_DEBUG_MUTEXES
1054         p->blocked_on = NULL; /* not blocked yet */
1055 #endif
1056
1057         p->tgid = p->pid;
1058         if (clone_flags & CLONE_THREAD)
1059                 p->tgid = current->tgid;
1060
1061         if ((retval = security_task_alloc(p)))
1062                 goto bad_fork_cleanup_policy;
1063         if ((retval = audit_alloc(p)))
1064                 goto bad_fork_cleanup_security;
1065         /* copy all the process information */
1066         if ((retval = copy_semundo(clone_flags, p)))
1067                 goto bad_fork_cleanup_audit;
1068         if ((retval = copy_files(clone_flags, p)))
1069                 goto bad_fork_cleanup_semundo;
1070         if ((retval = copy_fs(clone_flags, p)))
1071                 goto bad_fork_cleanup_files;
1072         if ((retval = copy_sighand(clone_flags, p)))
1073                 goto bad_fork_cleanup_fs;
1074         if ((retval = copy_signal(clone_flags, p)))
1075                 goto bad_fork_cleanup_sighand;
1076         if ((retval = copy_mm(clone_flags, p)))
1077                 goto bad_fork_cleanup_signal;
1078         if ((retval = copy_keys(clone_flags, p)))
1079                 goto bad_fork_cleanup_mm;
1080         if ((retval = copy_namespace(clone_flags, p)))
1081                 goto bad_fork_cleanup_keys;
1082         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1083         if (retval)
1084                 goto bad_fork_cleanup_namespace;
1085
1086         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1087         /*
1088          * Clear TID on mm_release()?
1089          */
1090         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1091         p->robust_list = NULL;
1092 #ifdef CONFIG_COMPAT
1093         p->compat_robust_list = NULL;
1094 #endif
1095         INIT_LIST_HEAD(&p->pi_state_list);
1096         p->pi_state_cache = NULL;
1097
1098         /*
1099          * sigaltstack should be cleared when sharing the same VM
1100          */
1101         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1102                 p->sas_ss_sp = p->sas_ss_size = 0;
1103
1104         /*
1105          * Syscall tracing should be turned off in the child regardless
1106          * of CLONE_PTRACE.
1107          */
1108         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1109 #ifdef TIF_SYSCALL_EMU
1110         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1111 #endif
1112
1113         /* Our parent execution domain becomes current domain
1114            These must match for thread signalling to apply */
1115
1116         p->parent_exec_id = p->self_exec_id;
1117
1118         /* ok, now we should be set up.. */
1119         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1120         p->pdeath_signal = 0;
1121         p->exit_state = 0;
1122
1123         /*
1124          * Ok, make it visible to the rest of the system.
1125          * We dont wake it up yet.
1126          */
1127         p->group_leader = p;
1128         INIT_LIST_HEAD(&p->thread_group);
1129         INIT_LIST_HEAD(&p->ptrace_children);
1130         INIT_LIST_HEAD(&p->ptrace_list);
1131
1132         /* Perform scheduler related setup. Assign this task to a CPU. */
1133         sched_fork(p, clone_flags);
1134
1135         /* Need tasklist lock for parent etc handling! */
1136         write_lock_irq(&tasklist_lock);
1137
1138         /*
1139          * The task hasn't been attached yet, so its cpus_allowed mask will
1140          * not be changed, nor will its assigned CPU.
1141          *
1142          * The cpus_allowed mask of the parent may have changed after it was
1143          * copied first time - so re-copy it here, then check the child's CPU
1144          * to ensure it is on a valid CPU (and if not, just force it back to
1145          * parent's CPU). This avoids alot of nasty races.
1146          */
1147         p->cpus_allowed = current->cpus_allowed;
1148         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1149                         !cpu_online(task_cpu(p))))
1150                 set_task_cpu(p, smp_processor_id());
1151
1152         /* CLONE_PARENT re-uses the old parent */
1153         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1154                 p->real_parent = current->real_parent;
1155         else
1156                 p->real_parent = current;
1157         p->parent = p->real_parent;
1158
1159         spin_lock(&current->sighand->siglock);
1160
1161         /*
1162          * Process group and session signals need to be delivered to just the
1163          * parent before the fork or both the parent and the child after the
1164          * fork. Restart if a signal comes in before we add the new process to
1165          * it's process group.
1166          * A fatal signal pending means that current will exit, so the new
1167          * thread can't slip out of an OOM kill (or normal SIGKILL).
1168          */
1169         recalc_sigpending();
1170         if (signal_pending(current)) {
1171                 spin_unlock(&current->sighand->siglock);
1172                 write_unlock_irq(&tasklist_lock);
1173                 retval = -ERESTARTNOINTR;
1174                 goto bad_fork_cleanup_namespace;
1175         }
1176
1177         if (clone_flags & CLONE_THREAD) {
1178                 p->group_leader = current->group_leader;
1179                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1180
1181                 if (!cputime_eq(current->signal->it_virt_expires,
1182                                 cputime_zero) ||
1183                     !cputime_eq(current->signal->it_prof_expires,
1184                                 cputime_zero) ||
1185                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1186                     !list_empty(&current->signal->cpu_timers[0]) ||
1187                     !list_empty(&current->signal->cpu_timers[1]) ||
1188                     !list_empty(&current->signal->cpu_timers[2])) {
1189                         /*
1190                          * Have child wake up on its first tick to check
1191                          * for process CPU timers.
1192                          */
1193                         p->it_prof_expires = jiffies_to_cputime(1);
1194                 }
1195         }
1196
1197         /*
1198          * inherit ioprio
1199          */
1200         p->ioprio = current->ioprio;
1201
1202         if (likely(p->pid)) {
1203                 add_parent(p);
1204                 if (unlikely(p->ptrace & PT_PTRACED))
1205                         __ptrace_link(p, current->parent);
1206
1207                 if (thread_group_leader(p)) {
1208                         p->signal->tty = current->signal->tty;
1209                         p->signal->pgrp = process_group(current);
1210                         p->signal->session = current->signal->session;
1211                         attach_pid(p, PIDTYPE_PGID, process_group(p));
1212                         attach_pid(p, PIDTYPE_SID, p->signal->session);
1213
1214                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1215                         __get_cpu_var(process_counts)++;
1216                 }
1217                 attach_pid(p, PIDTYPE_PID, p->pid);
1218                 nr_threads++;
1219         }
1220
1221         total_forks++;
1222         spin_unlock(&current->sighand->siglock);
1223         write_unlock_irq(&tasklist_lock);
1224         proc_fork_connector(p);
1225         return p;
1226
1227 bad_fork_cleanup_namespace:
1228         exit_namespace(p);
1229 bad_fork_cleanup_keys:
1230         exit_keys(p);
1231 bad_fork_cleanup_mm:
1232         if (p->mm)
1233                 mmput(p->mm);
1234 bad_fork_cleanup_signal:
1235         cleanup_signal(p);
1236 bad_fork_cleanup_sighand:
1237         __cleanup_sighand(p->sighand);
1238 bad_fork_cleanup_fs:
1239         exit_fs(p); /* blocking */
1240 bad_fork_cleanup_files:
1241         exit_files(p); /* blocking */
1242 bad_fork_cleanup_semundo:
1243         exit_sem(p);
1244 bad_fork_cleanup_audit:
1245         audit_free(p);
1246 bad_fork_cleanup_security:
1247         security_task_free(p);
1248 bad_fork_cleanup_policy:
1249 #ifdef CONFIG_NUMA
1250         mpol_free(p->mempolicy);
1251 bad_fork_cleanup_cpuset:
1252 #endif
1253         cpuset_exit(p);
1254 bad_fork_cleanup:
1255         if (p->binfmt)
1256                 module_put(p->binfmt->module);
1257 bad_fork_cleanup_put_domain:
1258         module_put(task_thread_info(p)->exec_domain->module);
1259 bad_fork_cleanup_count:
1260         put_group_info(p->group_info);
1261         atomic_dec(&p->user->processes);
1262         free_uid(p->user);
1263 bad_fork_free:
1264         free_task(p);
1265 fork_out:
1266         return ERR_PTR(retval);
1267 }
1268
1269 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1270 {
1271         memset(regs, 0, sizeof(struct pt_regs));
1272         return regs;
1273 }
1274
1275 task_t * __devinit fork_idle(int cpu)
1276 {
1277         task_t *task;
1278         struct pt_regs regs;
1279
1280         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1281         if (!task)
1282                 return ERR_PTR(-ENOMEM);
1283         init_idle(task, cpu);
1284
1285         return task;
1286 }
1287
1288 static inline int fork_traceflag (unsigned clone_flags)
1289 {
1290         if (clone_flags & CLONE_UNTRACED)
1291                 return 0;
1292         else if (clone_flags & CLONE_VFORK) {
1293                 if (current->ptrace & PT_TRACE_VFORK)
1294                         return PTRACE_EVENT_VFORK;
1295         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1296                 if (current->ptrace & PT_TRACE_CLONE)
1297                         return PTRACE_EVENT_CLONE;
1298         } else if (current->ptrace & PT_TRACE_FORK)
1299                 return PTRACE_EVENT_FORK;
1300
1301         return 0;
1302 }
1303
1304 /*
1305  *  Ok, this is the main fork-routine.
1306  *
1307  * It copies the process, and if successful kick-starts
1308  * it and waits for it to finish using the VM if required.
1309  */
1310 long do_fork(unsigned long clone_flags,
1311               unsigned long stack_start,
1312               struct pt_regs *regs,
1313               unsigned long stack_size,
1314               int __user *parent_tidptr,
1315               int __user *child_tidptr)
1316 {
1317         struct task_struct *p;
1318         int trace = 0;
1319         struct pid *pid = alloc_pid();
1320         long nr;
1321
1322         if (!pid)
1323                 return -EAGAIN;
1324         nr = pid->nr;
1325         if (unlikely(current->ptrace)) {
1326                 trace = fork_traceflag (clone_flags);
1327                 if (trace)
1328                         clone_flags |= CLONE_PTRACE;
1329         }
1330
1331         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, nr);
1332         /*
1333          * Do this prior waking up the new thread - the thread pointer
1334          * might get invalid after that point, if the thread exits quickly.
1335          */
1336         if (!IS_ERR(p)) {
1337                 struct completion vfork;
1338
1339                 if (clone_flags & CLONE_VFORK) {
1340                         p->vfork_done = &vfork;
1341                         init_completion(&vfork);
1342                 }
1343
1344                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1345                         /*
1346                          * We'll start up with an immediate SIGSTOP.
1347                          */
1348                         sigaddset(&p->pending.signal, SIGSTOP);
1349                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1350                 }
1351
1352                 if (!(clone_flags & CLONE_STOPPED))
1353                         wake_up_new_task(p, clone_flags);
1354                 else
1355                         p->state = TASK_STOPPED;
1356
1357                 if (unlikely (trace)) {
1358                         current->ptrace_message = nr;
1359                         ptrace_notify ((trace << 8) | SIGTRAP);
1360                 }
1361
1362                 if (clone_flags & CLONE_VFORK) {
1363                         wait_for_completion(&vfork);
1364                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1365                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1366                 }
1367         } else {
1368                 free_pid(pid);
1369                 nr = PTR_ERR(p);
1370         }
1371         return nr;
1372 }
1373
1374 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1375 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1376 #endif
1377
1378 static void sighand_ctor(void *data, kmem_cache_t *cachep, unsigned long flags)
1379 {
1380         struct sighand_struct *sighand = data;
1381
1382         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
1383                                         SLAB_CTOR_CONSTRUCTOR)
1384                 spin_lock_init(&sighand->siglock);
1385 }
1386
1387 void __init proc_caches_init(void)
1388 {
1389         sighand_cachep = kmem_cache_create("sighand_cache",
1390                         sizeof(struct sighand_struct), 0,
1391                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1392                         sighand_ctor, NULL);
1393         signal_cachep = kmem_cache_create("signal_cache",
1394                         sizeof(struct signal_struct), 0,
1395                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1396         files_cachep = kmem_cache_create("files_cache",
1397                         sizeof(struct files_struct), 0,
1398                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1399         fs_cachep = kmem_cache_create("fs_cache",
1400                         sizeof(struct fs_struct), 0,
1401                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1402         vm_area_cachep = kmem_cache_create("vm_area_struct",
1403                         sizeof(struct vm_area_struct), 0,
1404                         SLAB_PANIC, NULL, NULL);
1405         mm_cachep = kmem_cache_create("mm_struct",
1406                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1407                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1408 }
1409
1410
1411 /*
1412  * Check constraints on flags passed to the unshare system call and
1413  * force unsharing of additional process context as appropriate.
1414  */
1415 static inline void check_unshare_flags(unsigned long *flags_ptr)
1416 {
1417         /*
1418          * If unsharing a thread from a thread group, must also
1419          * unshare vm.
1420          */
1421         if (*flags_ptr & CLONE_THREAD)
1422                 *flags_ptr |= CLONE_VM;
1423
1424         /*
1425          * If unsharing vm, must also unshare signal handlers.
1426          */
1427         if (*flags_ptr & CLONE_VM)
1428                 *flags_ptr |= CLONE_SIGHAND;
1429
1430         /*
1431          * If unsharing signal handlers and the task was created
1432          * using CLONE_THREAD, then must unshare the thread
1433          */
1434         if ((*flags_ptr & CLONE_SIGHAND) &&
1435             (atomic_read(&current->signal->count) > 1))
1436                 *flags_ptr |= CLONE_THREAD;
1437
1438         /*
1439          * If unsharing namespace, must also unshare filesystem information.
1440          */
1441         if (*flags_ptr & CLONE_NEWNS)
1442                 *flags_ptr |= CLONE_FS;
1443 }
1444
1445 /*
1446  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1447  */
1448 static int unshare_thread(unsigned long unshare_flags)
1449 {
1450         if (unshare_flags & CLONE_THREAD)
1451                 return -EINVAL;
1452
1453         return 0;
1454 }
1455
1456 /*
1457  * Unshare the filesystem structure if it is being shared
1458  */
1459 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1460 {
1461         struct fs_struct *fs = current->fs;
1462
1463         if ((unshare_flags & CLONE_FS) &&
1464             (fs && atomic_read(&fs->count) > 1)) {
1465                 *new_fsp = __copy_fs_struct(current->fs);
1466                 if (!*new_fsp)
1467                         return -ENOMEM;
1468         }
1469
1470         return 0;
1471 }
1472
1473 /*
1474  * Unshare the namespace structure if it is being shared
1475  */
1476 static int unshare_namespace(unsigned long unshare_flags, struct namespace **new_nsp, struct fs_struct *new_fs)
1477 {
1478         struct namespace *ns = current->namespace;
1479
1480         if ((unshare_flags & CLONE_NEWNS) &&
1481             (ns && atomic_read(&ns->count) > 1)) {
1482                 if (!capable(CAP_SYS_ADMIN))
1483                         return -EPERM;
1484
1485                 *new_nsp = dup_namespace(current, new_fs ? new_fs : current->fs);
1486                 if (!*new_nsp)
1487                         return -ENOMEM;
1488         }
1489
1490         return 0;
1491 }
1492
1493 /*
1494  * Unsharing of sighand for tasks created with CLONE_SIGHAND is not
1495  * supported yet
1496  */
1497 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1498 {
1499         struct sighand_struct *sigh = current->sighand;
1500
1501         if ((unshare_flags & CLONE_SIGHAND) &&
1502             (sigh && atomic_read(&sigh->count) > 1))
1503                 return -EINVAL;
1504         else
1505                 return 0;
1506 }
1507
1508 /*
1509  * Unshare vm if it is being shared
1510  */
1511 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1512 {
1513         struct mm_struct *mm = current->mm;
1514
1515         if ((unshare_flags & CLONE_VM) &&
1516             (mm && atomic_read(&mm->mm_users) > 1)) {
1517                 return -EINVAL;
1518         }
1519
1520         return 0;
1521 }
1522
1523 /*
1524  * Unshare file descriptor table if it is being shared
1525  */
1526 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1527 {
1528         struct files_struct *fd = current->files;
1529         int error = 0;
1530
1531         if ((unshare_flags & CLONE_FILES) &&
1532             (fd && atomic_read(&fd->count) > 1)) {
1533                 *new_fdp = dup_fd(fd, &error);
1534                 if (!*new_fdp)
1535                         return error;
1536         }
1537
1538         return 0;
1539 }
1540
1541 /*
1542  * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1543  * supported yet
1544  */
1545 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1546 {
1547         if (unshare_flags & CLONE_SYSVSEM)
1548                 return -EINVAL;
1549
1550         return 0;
1551 }
1552
1553 /*
1554  * unshare allows a process to 'unshare' part of the process
1555  * context which was originally shared using clone.  copy_*
1556  * functions used by do_fork() cannot be used here directly
1557  * because they modify an inactive task_struct that is being
1558  * constructed. Here we are modifying the current, active,
1559  * task_struct.
1560  */
1561 asmlinkage long sys_unshare(unsigned long unshare_flags)
1562 {
1563         int err = 0;
1564         struct fs_struct *fs, *new_fs = NULL;
1565         struct namespace *ns, *new_ns = NULL;
1566         struct sighand_struct *sigh, *new_sigh = NULL;
1567         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1568         struct files_struct *fd, *new_fd = NULL;
1569         struct sem_undo_list *new_ulist = NULL;
1570
1571         check_unshare_flags(&unshare_flags);
1572
1573         /* Return -EINVAL for all unsupported flags */
1574         err = -EINVAL;
1575         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1576                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM))
1577                 goto bad_unshare_out;
1578
1579         if ((err = unshare_thread(unshare_flags)))
1580                 goto bad_unshare_out;
1581         if ((err = unshare_fs(unshare_flags, &new_fs)))
1582                 goto bad_unshare_cleanup_thread;
1583         if ((err = unshare_namespace(unshare_flags, &new_ns, new_fs)))
1584                 goto bad_unshare_cleanup_fs;
1585         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1586                 goto bad_unshare_cleanup_ns;
1587         if ((err = unshare_vm(unshare_flags, &new_mm)))
1588                 goto bad_unshare_cleanup_sigh;
1589         if ((err = unshare_fd(unshare_flags, &new_fd)))
1590                 goto bad_unshare_cleanup_vm;
1591         if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1592                 goto bad_unshare_cleanup_fd;
1593
1594         if (new_fs || new_ns || new_sigh || new_mm || new_fd || new_ulist) {
1595
1596                 task_lock(current);
1597
1598                 if (new_fs) {
1599                         fs = current->fs;
1600                         current->fs = new_fs;
1601                         new_fs = fs;
1602                 }
1603
1604                 if (new_ns) {
1605                         ns = current->namespace;
1606                         current->namespace = new_ns;
1607                         new_ns = ns;
1608                 }
1609
1610                 if (new_sigh) {
1611                         sigh = current->sighand;
1612                         rcu_assign_pointer(current->sighand, new_sigh);
1613                         new_sigh = sigh;
1614                 }
1615
1616                 if (new_mm) {
1617                         mm = current->mm;
1618                         active_mm = current->active_mm;
1619                         current->mm = new_mm;
1620                         current->active_mm = new_mm;
1621                         activate_mm(active_mm, new_mm);
1622                         new_mm = mm;
1623                 }
1624
1625                 if (new_fd) {
1626                         fd = current->files;
1627                         current->files = new_fd;
1628                         new_fd = fd;
1629                 }
1630
1631                 task_unlock(current);
1632         }
1633
1634 bad_unshare_cleanup_fd:
1635         if (new_fd)
1636                 put_files_struct(new_fd);
1637
1638 bad_unshare_cleanup_vm:
1639         if (new_mm)
1640                 mmput(new_mm);
1641
1642 bad_unshare_cleanup_sigh:
1643         if (new_sigh)
1644                 if (atomic_dec_and_test(&new_sigh->count))
1645                         kmem_cache_free(sighand_cachep, new_sigh);
1646
1647 bad_unshare_cleanup_ns:
1648         if (new_ns)
1649                 put_namespace(new_ns);
1650
1651 bad_unshare_cleanup_fs:
1652         if (new_fs)
1653                 put_fs_struct(new_fs);
1654
1655 bad_unshare_cleanup_thread:
1656 bad_unshare_out:
1657         return err;
1658 }