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/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/mnt_namespace.h>
  21 #include <linux/personality.h>
  22 #include <linux/mempolicy.h>
  23 #include <linux/sem.h>
  24 #include <linux/file.h>
  25 #include <linux/key.h>
  26 #include <linux/binfmts.h>
  27 #include <linux/mman.h>
  28 #include <linux/fs.h>
  29 #include <linux/nsproxy.h>
  30 #include <linux/capability.h>
  31 #include <linux/cpu.h>
  32 #include <linux/cpuset.h>
  33 #include <linux/security.h>
  34 #include <linux/swap.h>
  35 #include <linux/syscalls.h>
  36 #include <linux/jiffies.h>
  37 #include <linux/futex.h>
  38 #include <linux/task_io_accounting_ops.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/tsacct_kern.h>
  47 #include <linux/cn_proc.h>
  48 #include <linux/freezer.h>
  49 #include <linux/delayacct.h>
  50 #include <linux/taskstats_kern.h>
  51 #include <linux/random.h>
  52
  53 #include <asm/pgtable.h>
  54 #include <asm/pgalloc.h>
  55 #include <asm/uaccess.h>
  56 #include <asm/mmu_context.h>
  57 #include <asm/cacheflush.h>
  58 #include <asm/tlbflush.h>
  59
  60 /*
  61  * Protected counters by write_lock_irq(&tasklist_lock)
  62  */
  63 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  64 int nr_threads;                 /* The idle threads do not count.. */
  65
  66 int max_threads;                /* tunable limit on nr_threads */
  67
  68 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  69
  70 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  71
  72 int nr_processes(void)
  73 {
  74         int cpu;
  75         int total = 0;
  76
  77         for_each_online_cpu(cpu)
  78                 total += per_cpu(process_counts, cpu);
  79
  80         return total;
  81 }
  82
  83 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  84 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  85 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  86 static struct kmem_cache *task_struct_cachep;
  87 #endif
  88
  89 /* SLAB cache for signal_struct structures (tsk->signal) */
  90 static struct kmem_cache *signal_cachep;
  91
  92 /* SLAB cache for sighand_struct structures (tsk->sighand) */
  93 struct kmem_cache *sighand_cachep;
  94
  95 /* SLAB cache for files_struct structures (tsk->files) */
  96 struct kmem_cache *files_cachep;
  97
  98 /* SLAB cache for fs_struct structures (tsk->fs) */
  99 struct kmem_cache *fs_cachep;
 100
 101 /* SLAB cache for vm_area_struct structures */
 102 struct kmem_cache *vm_area_cachep;
 103
 104 /* SLAB cache for mm_struct structures (tsk->mm) */
 105 static struct kmem_cache *mm_cachep;
 106
 107 void free_task(struct task_struct *tsk)
 108 {
 109         free_thread_info(tsk->stack);
 110         rt_mutex_debug_task_free(tsk);
 111         free_task_struct(tsk);
 112 }
 113 EXPORT_SYMBOL(free_task);
 114
 115 void __put_task_struct(struct task_struct *tsk)
 116 {
 117         WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
 118         WARN_ON(atomic_read(&tsk->usage));
 119         WARN_ON(tsk == current);
 120
 121         security_task_free(tsk);
 122         free_uid(tsk->user);
 123         put_group_info(tsk->group_info);
 124         delayacct_tsk_free(tsk);
 125
 126         if (!profile_handoff_task(tsk))
 127                 free_task(tsk);
 128 }
 129
 130 void __init fork_init(unsigned long mempages)
 131 {
 132 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 133 #ifndef ARCH_MIN_TASKALIGN
 134 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 135 #endif
 136         /* create a slab on which task_structs can be allocated */
 137         task_struct_cachep =
 138                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 139                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
 140 #endif
 141
 142         /*
 143          * The default maximum number of threads is set to a safe
 144          * value: the thread structures can take up at most half
 145          * of memory.
 146          */
 147         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 148
 149         /*
 150          * we need to allow at least 20 threads to boot a system
 151          */
 152         if(max_threads < 20)
 153                 max_threads = 20;
 154
 155         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 156         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 157         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 158                 init_task.signal->rlim[RLIMIT_NPROC];
 159 }
 160
 161 static struct task_struct *dup_task_struct(struct task_struct *orig)
 162 {
 163         struct task_struct *tsk;
 164         struct thread_info *ti;
 165
 166         prepare_to_copy(orig);
 167
 168         tsk = alloc_task_struct();
 169         if (!tsk)
 170                 return NULL;
 171
 172         ti = alloc_thread_info(tsk);
 173         if (!ti) {
 174                 free_task_struct(tsk);
 175                 return NULL;
 176         }
 177
 178         *tsk = *orig;
 179         tsk->stack = ti;
 180         setup_thread_stack(tsk, orig);
 181
 182 #ifdef CONFIG_CC_STACKPROTECTOR
 183         tsk->stack_canary = get_random_int();
 184 #endif
 185
 186         /* One for us, one for whoever does the "release_task()" (usually parent) */
 187         atomic_set(&tsk->usage,2);
 188         atomic_set(&tsk->fs_excl, 0);
 189 #ifdef CONFIG_BLK_DEV_IO_TRACE
 190         tsk->btrace_seq = 0;
 191 #endif
 192         tsk->splice_pipe = NULL;
 193         return tsk;
 194 }
 195
 196 #ifdef CONFIG_MMU
 197 static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 198 {
 199         struct vm_area_struct *mpnt, *tmp, **pprev;
 200         struct rb_node **rb_link, *rb_parent;
 201         int retval;
 202         unsigned long charge;
 203         struct mempolicy *pol;
 204
 205         down_write(&oldmm->mmap_sem);
 206         flush_cache_dup_mm(oldmm);
 207         /*
 208          * Not linked in yet - no deadlock potential:
 209          */
 210         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 211
 212         mm->locked_vm = 0;
 213         mm->mmap = NULL;
 214         mm->mmap_cache = NULL;
 215         mm->free_area_cache = oldmm->mmap_base;
 216         mm->cached_hole_size = ~0UL;
 217         mm->map_count = 0;
 218         cpus_clear(mm->cpu_vm_mask);
 219         mm->mm_rb = RB_ROOT;
 220         rb_link = &mm->mm_rb.rb_node;
 221         rb_parent = NULL;
 222         pprev = &mm->mmap;
 223
 224         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 225                 struct file *file;
 226
 227                 if (mpnt->vm_flags & VM_DONTCOPY) {
 228                         long pages = vma_pages(mpnt);
 229                         mm->total_vm -= pages;
 230                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 231                                                                 -pages);
 232                         continue;
 233                 }
 234                 charge = 0;
 235                 if (mpnt->vm_flags & VM_ACCOUNT) {
 236                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 237                         if (security_vm_enough_memory(len))
 238                                 goto fail_nomem;
 239                         charge = len;
 240                 }
 241                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 242                 if (!tmp)
 243                         goto fail_nomem;
 244                 *tmp = *mpnt;
 245                 pol = mpol_copy(vma_policy(mpnt));
 246                 retval = PTR_ERR(pol);
 247                 if (IS_ERR(pol))
 248                         goto fail_nomem_policy;
 249                 vma_set_policy(tmp, pol);
 250                 tmp->vm_flags &= ~VM_LOCKED;
 251                 tmp->vm_mm = mm;
 252                 tmp->vm_next = NULL;
 253                 anon_vma_link(tmp);
 254                 file = tmp->vm_file;
 255                 if (file) {
 256                         struct inode *inode = file->f_path.dentry->d_inode;
 257                         get_file(file);
 258                         if (tmp->vm_flags & VM_DENYWRITE)
 259                                 atomic_dec(&inode->i_writecount);
 260
 261                         /* insert tmp into the share list, just after mpnt */
 262                         spin_lock(&file->f_mapping->i_mmap_lock);
 263                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 264                         flush_dcache_mmap_lock(file->f_mapping);
 265                         vma_prio_tree_add(tmp, mpnt);
 266                         flush_dcache_mmap_unlock(file->f_mapping);
 267                         spin_unlock(&file->f_mapping->i_mmap_lock);
 268                 }
 269
 270                 /*
 271                  * Link in the new vma and copy the page table entries.
 272                  */
 273                 *pprev = tmp;
 274                 pprev = &tmp->vm_next;
 275
 276                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 277                 rb_link = &tmp->vm_rb.rb_right;
 278                 rb_parent = &tmp->vm_rb;
 279
 280                 mm->map_count++;
 281                 retval = copy_page_range(mm, oldmm, mpnt);
 282
 283                 if (tmp->vm_ops && tmp->vm_ops->open)
 284                         tmp->vm_ops->open(tmp);
 285
 286                 if (retval)
 287                         goto out;
 288         }
 289         /* a new mm has just been created */
 290         arch_dup_mmap(oldmm, mm);
 291         retval = 0;
 292 out:
 293         up_write(&mm->mmap_sem);
 294         flush_tlb_mm(oldmm);
 295         up_write(&oldmm->mmap_sem);
 296         return retval;
 297 fail_nomem_policy:
 298         kmem_cache_free(vm_area_cachep, tmp);
 299 fail_nomem:
 300         retval = -ENOMEM;
 301         vm_unacct_memory(charge);
 302         goto out;
 303 }
 304
 305 static inline int mm_alloc_pgd(struct mm_struct * mm)
 306 {
 307         mm->pgd = pgd_alloc(mm);
 308         if (unlikely(!mm->pgd))
 309                 return -ENOMEM;
 310         return 0;
 311 }
 312
 313 static inline void mm_free_pgd(struct mm_struct * mm)
 314 {
 315         pgd_free(mm->pgd);
 316 }
 317 #else
 318 #define dup_mmap(mm, oldmm)     (0)
 319 #define mm_alloc_pgd(mm)        (0)
 320 #define mm_free_pgd(mm)
 321 #endif /* CONFIG_MMU */
 322
 323  __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 324
 325 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 326 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 327
 328 #include <linux/init_task.h>
 329
 330 static struct mm_struct * mm_init(struct mm_struct * mm)
 331 {
 332         atomic_set(&mm->mm_users, 1);
 333         atomic_set(&mm->mm_count, 1);
 334         init_rwsem(&mm->mmap_sem);
 335         INIT_LIST_HEAD(&mm->mmlist);
 336         mm->core_waiters = 0;
 337         mm->nr_ptes = 0;
 338         set_mm_counter(mm, file_rss, 0);
 339         set_mm_counter(mm, anon_rss, 0);
 340         spin_lock_init(&mm->page_table_lock);
 341         rwlock_init(&mm->ioctx_list_lock);
 342         mm->ioctx_list = NULL;
 343         mm->free_area_cache = TASK_UNMAPPED_BASE;
 344         mm->cached_hole_size = ~0UL;
 345
 346         if (likely(!mm_alloc_pgd(mm))) {
 347                 mm->def_flags = 0;
 348                 return mm;
 349         }
 350         free_mm(mm);
 351         return NULL;
 352 }
 353
 354 /*
 355  * Allocate and initialize an mm_struct.
 356  */
 357 struct mm_struct * mm_alloc(void)
 358 {
 359         struct mm_struct * mm;
 360
 361         mm = allocate_mm();
 362         if (mm) {
 363                 memset(mm, 0, sizeof(*mm));
 364                 mm = mm_init(mm);
 365         }
 366         return mm;
 367 }
 368
 369 /*
 370  * Called when the last reference to the mm
 371  * is dropped: either by a lazy thread or by
 372  * mmput. Free the page directory and the mm.
 373  */
 374 void fastcall __mmdrop(struct mm_struct *mm)
 375 {
 376         BUG_ON(mm == &init_mm);
 377         mm_free_pgd(mm);
 378         destroy_context(mm);
 379         free_mm(mm);
 380 }
 381
 382 /*
 383  * Decrement the use count and release all resources for an mm.
 384  */
 385 void mmput(struct mm_struct *mm)
 386 {
 387         might_sleep();
 388
 389         if (atomic_dec_and_test(&mm->mm_users)) {
 390                 exit_aio(mm);
 391                 exit_mmap(mm);
 392                 if (!list_empty(&mm->mmlist)) {
 393                         spin_lock(&mmlist_lock);
 394                         list_del(&mm->mmlist);
 395                         spin_unlock(&mmlist_lock);
 396                 }
 397                 put_swap_token(mm);
 398                 mmdrop(mm);
 399         }
 400 }
 401 EXPORT_SYMBOL_GPL(mmput);
 402
 403 /**
 404  * get_task_mm - acquire a reference to the task's mm
 405  *
 406  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
 407  * this kernel workthread has transiently adopted a user mm with use_mm,
 408  * to do its AIO) is not set and if so returns a reference to it, after
 409  * bumping up the use count.  User must release the mm via mmput()
 410  * after use.  Typically used by /proc and ptrace.
 411  */
 412 struct mm_struct *get_task_mm(struct task_struct *task)
 413 {
 414         struct mm_struct *mm;
 415
 416         task_lock(task);
 417         mm = task->mm;
 418         if (mm) {
 419                 if (task->flags & PF_BORROWED_MM)
 420                         mm = NULL;
 421                 else
 422                         atomic_inc(&mm->mm_users);
 423         }
 424         task_unlock(task);
 425         return mm;
 426 }
 427 EXPORT_SYMBOL_GPL(get_task_mm);
 428
 429 /* Please note the differences between mmput and mm_release.
 430  * mmput is called whenever we stop holding onto a mm_struct,
 431  * error success whatever.
 432  *
 433  * mm_release is called after a mm_struct has been removed
 434  * from the current process.
 435  *
 436  * This difference is important for error handling, when we
 437  * only half set up a mm_struct for a new process and need to restore
 438  * the old one.  Because we mmput the new mm_struct before
 439  * restoring the old one. . .
 440  * Eric Biederman 10 January 1998
 441  */
 442 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 443 {
 444         struct completion *vfork_done = tsk->vfork_done;
 445
 446         /* Get rid of any cached register state */
 447         deactivate_mm(tsk, mm);
 448
 449         /* notify parent sleeping on vfork() */
 450         if (vfork_done) {
 451                 tsk->vfork_done = NULL;
 452                 complete(vfork_done);
 453         }
 454
 455         /*
 456          * If we're exiting normally, clear a user-space tid field if
 457          * requested.  We leave this alone when dying by signal, to leave
 458          * the value intact in a core dump, and to save the unnecessary
 459          * trouble otherwise.  Userland only wants this done for a sys_exit.
 460          */
 461         if (tsk->clear_child_tid
 462             && !(tsk->flags & PF_SIGNALED)
 463             && atomic_read(&mm->mm_users) > 1) {
 464                 u32 __user * tidptr = tsk->clear_child_tid;
 465                 tsk->clear_child_tid = NULL;
 466
 467                 /*
 468                  * We don't check the error code - if userspace has
 469                  * not set up a proper pointer then tough luck.
 470                  */
 471                 put_user(0, tidptr);
 472                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 473         }
 474 }
 475
 476 /*
 477  * Allocate a new mm structure and copy contents from the
 478  * mm structure of the passed in task structure.
 479  */
 480 static struct mm_struct *dup_mm(struct task_struct *tsk)
 481 {
 482         struct mm_struct *mm, *oldmm = current->mm;
 483         int err;
 484
 485         if (!oldmm)
 486                 return NULL;
 487
 488         mm = allocate_mm();
 489         if (!mm)
 490                 goto fail_nomem;
 491
 492         memcpy(mm, oldmm, sizeof(*mm));
 493
 494         /* Initializing for Swap token stuff */
 495         mm->token_priority = 0;
 496         mm->last_interval = 0;
 497
 498         if (!mm_init(mm))
 499                 goto fail_nomem;
 500
 501         if (init_new_context(tsk, mm))
 502                 goto fail_nocontext;
 503
 504         err = dup_mmap(mm, oldmm);
 505         if (err)
 506                 goto free_pt;
 507
 508         mm->hiwater_rss = get_mm_rss(mm);
 509         mm->hiwater_vm = mm->total_vm;
 510
 511         return mm;
 512
 513 free_pt:
 514         mmput(mm);
 515
 516 fail_nomem:
 517         return NULL;
 518
 519 fail_nocontext:
 520         /*
 521          * If init_new_context() failed, we cannot use mmput() to free the mm
 522          * because it calls destroy_context()
 523          */
 524         mm_free_pgd(mm);
 525         free_mm(mm);
 526         return NULL;
 527 }
 528
 529 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 530 {
 531         struct mm_struct * mm, *oldmm;
 532         int retval;
 533
 534         tsk->min_flt = tsk->maj_flt = 0;
 535         tsk->nvcsw = tsk->nivcsw = 0;
 536
 537         tsk->mm = NULL;
 538         tsk->active_mm = NULL;
 539
 540         /*
 541          * Are we cloning a kernel thread?
 542          *
 543          * We need to steal a active VM for that..
 544          */
 545         oldmm = current->mm;
 546         if (!oldmm)
 547                 return 0;
 548
 549         if (clone_flags & CLONE_VM) {
 550                 atomic_inc(&oldmm->mm_users);
 551                 mm = oldmm;
 552                 goto good_mm;
 553         }
 554
 555         retval = -ENOMEM;
 556         mm = dup_mm(tsk);
 557         if (!mm)
 558                 goto fail_nomem;
 559
 560 good_mm:
 561         /* Initializing for Swap token stuff */
 562         mm->token_priority = 0;
 563         mm->last_interval = 0;
 564
 565         tsk->mm = mm;
 566         tsk->active_mm = mm;
 567         return 0;
 568
 569 fail_nomem:
 570         return retval;
 571 }
 572
 573 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 574 {
 575         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 576         /* We don't need to lock fs - think why ;-) */
 577         if (fs) {
 578                 atomic_set(&fs->count, 1);
 579                 rwlock_init(&fs->lock);
 580                 fs->umask = old->umask;
 581                 read_lock(&old->lock);
 582                 fs->rootmnt = mntget(old->rootmnt);
 583                 fs->root = dget(old->root);
 584                 fs->pwdmnt = mntget(old->pwdmnt);
 585                 fs->pwd = dget(old->pwd);
 586                 if (old->altroot) {
 587                         fs->altrootmnt = mntget(old->altrootmnt);
 588                         fs->altroot = dget(old->altroot);
 589                 } else {
 590                         fs->altrootmnt = NULL;
 591                         fs->altroot = NULL;
 592                 }
 593                 read_unlock(&old->lock);
 594         }
 595         return fs;
 596 }
 597
 598 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 599 {
 600         return __copy_fs_struct(old);
 601 }
 602
 603 EXPORT_SYMBOL_GPL(copy_fs_struct);
 604
 605 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 606 {
 607         if (clone_flags & CLONE_FS) {
 608                 atomic_inc(&current->fs->count);
 609                 return 0;
 610         }
 611         tsk->fs = __copy_fs_struct(current->fs);
 612         if (!tsk->fs)
 613                 return -ENOMEM;
 614         return 0;
 615 }
 616
 617 static int count_open_files(struct fdtable *fdt)
 618 {
 619         int size = fdt->max_fds;
 620         int i;
 621
 622         /* Find the last open fd */
 623         for (i = size/(8*sizeof(long)); i > 0; ) {
 624                 if (fdt->open_fds->fds_bits[--i])
 625                         break;
 626         }
 627         i = (i+1) * 8 * sizeof(long);
 628         return i;
 629 }
 630
 631 static struct files_struct *alloc_files(void)
 632 {
 633         struct files_struct *newf;
 634         struct fdtable *fdt;
 635
 636         newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
 637         if (!newf)
 638                 goto out;
 639
 640         atomic_set(&newf->count, 1);
 641
 642         spin_lock_init(&newf->file_lock);
 643         newf->next_fd = 0;
 644         fdt = &newf->fdtab;
 645         fdt->max_fds = NR_OPEN_DEFAULT;
 646         fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
 647         fdt->open_fds = (fd_set *)&newf->open_fds_init;
 648         fdt->fd = &newf->fd_array[0];
 649         INIT_RCU_HEAD(&fdt->rcu);
 650         fdt->next = NULL;
 651         rcu_assign_pointer(newf->fdt, fdt);
 652 out:
 653         return newf;
 654 }
 655
 656 /*
 657  * Allocate a new files structure and copy contents from the
 658  * passed in files structure.
 659  * errorp will be valid only when the returned files_struct is NULL.
 660  */
 661 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
 662 {
 663         struct files_struct *newf;
 664         struct file **old_fds, **new_fds;
 665         int open_files, size, i;
 666         struct fdtable *old_fdt, *new_fdt;
 667
 668         *errorp = -ENOMEM;
 669         newf = alloc_files();
 670         if (!newf)
 671                 goto out;
 672
 673         spin_lock(&oldf->file_lock);
 674         old_fdt = files_fdtable(oldf);
 675         new_fdt = files_fdtable(newf);
 676         open_files = count_open_files(old_fdt);
 677
 678         /*
 679          * Check whether we need to allocate a larger fd array and fd set.
 680          * Note: we're not a clone task, so the open count won't change.
 681          */
 682         if (open_files > new_fdt->max_fds) {
 683                 new_fdt->max_fds = 0;
 684                 spin_unlock(&oldf->file_lock);
 685                 spin_lock(&newf->file_lock);
 686                 *errorp = expand_files(newf, open_files-1);
 687                 spin_unlock(&newf->file_lock);
 688                 if (*errorp < 0)
 689                         goto out_release;
 690                 new_fdt = files_fdtable(newf);
 691                 /*
 692                  * Reacquire the oldf lock and a pointer to its fd table
 693                  * who knows it may have a new bigger fd table. We need
 694                  * the latest pointer.
 695                  */
 696                 spin_lock(&oldf->file_lock);
 697                 old_fdt = files_fdtable(oldf);
 698         }
 699
 700         old_fds = old_fdt->fd;
 701         new_fds = new_fdt->fd;
 702
 703         memcpy(new_fdt->open_fds->fds_bits,
 704                 old_fdt->open_fds->fds_bits, open_files/8);
 705         memcpy(new_fdt->close_on_exec->fds_bits,
 706                 old_fdt->close_on_exec->fds_bits, open_files/8);
 707
 708         for (i = open_files; i != 0; i--) {
 709                 struct file *f = *old_fds++;
 710                 if (f) {
 711                         get_file(f);
 712                 } else {
 713                         /*
 714                          * The fd may be claimed in the fd bitmap but not yet
 715                          * instantiated in the files array if a sibling thread
 716                          * is partway through open().  So make sure that this
 717                          * fd is available to the new process.
 718                          */
 719                         FD_CLR(open_files - i, new_fdt->open_fds);
 720                 }
 721                 rcu_assign_pointer(*new_fds++, f);
 722         }
 723         spin_unlock(&oldf->file_lock);
 724
 725         /* compute the remainder to be cleared */
 726         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
 727
 728         /* This is long word aligned thus could use a optimized version */
 729         memset(new_fds, 0, size);
 730
 731         if (new_fdt->max_fds > open_files) {
 732                 int left = (new_fdt->max_fds-open_files)/8;
 733                 int start = open_files / (8 * sizeof(unsigned long));
 734
 735                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
 736                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
 737         }
 738
 739         return newf;
 740
 741 out_release:
 742         kmem_cache_free(files_cachep, newf);
 743 out:
 744         return NULL;
 745 }
 746
 747 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 748 {
 749         struct files_struct *oldf, *newf;
 750         int error = 0;
 751
 752         /*
 753          * A background process may not have any files ...
 754          */
 755         oldf = current->files;
 756         if (!oldf)
 757                 goto out;
 758
 759         if (clone_flags & CLONE_FILES) {
 760                 atomic_inc(&oldf->count);
 761                 goto out;
 762         }
 763
 764         /*
 765          * Note: we may be using current for both targets (See exec.c)
 766          * This works because we cache current->files (old) as oldf. Don't
 767          * break this.
 768          */
 769         tsk->files = NULL;
 770         newf = dup_fd(oldf, &error);
 771         if (!newf)
 772                 goto out;
 773
 774         tsk->files = newf;
 775         error = 0;
 776 out:
 777         return error;
 778 }
 779
 780 /*
 781  *      Helper to unshare the files of the current task.
 782  *      We don't want to expose copy_files internals to
 783  *      the exec layer of the kernel.
 784  */
 785
 786 int unshare_files(void)
 787 {
 788         struct files_struct *files  = current->files;
 789         int rc;
 790
 791         BUG_ON(!files);
 792
 793         /* This can race but the race causes us to copy when we don't
 794            need to and drop the copy */
 795         if(atomic_read(&files->count) == 1)
 796         {
 797                 atomic_inc(&files->count);
 798                 return 0;
 799         }
 800         rc = copy_files(0, current);
 801         if(rc)
 802                 current->files = files;
 803         return rc;
 804 }
 805
 806 EXPORT_SYMBOL(unshare_files);
 807
 808 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 809 {
 810         struct sighand_struct *sig;
 811
 812         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 813                 atomic_inc(&current->sighand->count);
 814                 return 0;
 815         }
 816         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 817         rcu_assign_pointer(tsk->sighand, sig);
 818         if (!sig)
 819                 return -ENOMEM;
 820         atomic_set(&sig->count, 1);
 821         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 822         return 0;
 823 }
 824
 825 void __cleanup_sighand(struct sighand_struct *sighand)
 826 {
 827         if (atomic_dec_and_test(&sighand->count))
 828                 kmem_cache_free(sighand_cachep, sighand);
 829 }
 830
 831 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 832 {
 833         struct signal_struct *sig;
 834         int ret;
 835
 836         if (clone_flags & CLONE_THREAD) {
 837                 atomic_inc(&current->signal->count);
 838                 atomic_inc(&current->signal->live);
 839                 return 0;
 840         }
 841         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 842         tsk->signal = sig;
 843         if (!sig)
 844                 return -ENOMEM;
 845
 846         ret = copy_thread_group_keys(tsk);
 847         if (ret < 0) {
 848                 kmem_cache_free(signal_cachep, sig);
 849                 return ret;
 850         }
 851
 852         atomic_set(&sig->count, 1);
 853         atomic_set(&sig->live, 1);
 854         init_waitqueue_head(&sig->wait_chldexit);
 855         sig->flags = 0;
 856         sig->group_exit_code = 0;
 857         sig->group_exit_task = NULL;
 858         sig->group_stop_count = 0;
 859         sig->curr_target = NULL;
 860         init_sigpending(&sig->shared_pending);
 861         INIT_LIST_HEAD(&sig->posix_timers);
 862
 863         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 864         sig->it_real_incr.tv64 = 0;
 865         sig->real_timer.function = it_real_fn;
 866         sig->tsk = tsk;
 867
 868         sig->it_virt_expires = cputime_zero;
 869         sig->it_virt_incr = cputime_zero;
 870         sig->it_prof_expires = cputime_zero;
 871         sig->it_prof_incr = cputime_zero;
 872
 873         sig->leader = 0;        /* session leadership doesn't inherit */
 874         sig->tty_old_pgrp = NULL;
 875
 876         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 877         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 878         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 879         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
 880         sig->sum_sched_runtime = 0;
 881         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 882         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 883         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 884         taskstats_tgid_init(sig);
 885
 886         task_lock(current->group_leader);
 887         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 888         task_unlock(current->group_leader);
 889
 890         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 891                 /*
 892                  * New sole thread in the process gets an expiry time
 893                  * of the whole CPU time limit.
 894                  */
 895                 tsk->it_prof_expires =
 896                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 897         }
 898         acct_init_pacct(&sig->pacct);
 899
 900         return 0;
 901 }
 902
 903 void __cleanup_signal(struct signal_struct *sig)
 904 {
 905         exit_thread_group_keys(sig);
 906         kmem_cache_free(signal_cachep, sig);
 907 }
 908
 909 static inline void cleanup_signal(struct task_struct *tsk)
 910 {
 911         struct signal_struct *sig = tsk->signal;
 912
 913         atomic_dec(&sig->live);
 914
 915         if (atomic_dec_and_test(&sig->count))
 916                 __cleanup_signal(sig);
 917 }
 918
 919 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 920 {
 921         unsigned long new_flags = p->flags;
 922
 923         new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
 924         new_flags |= PF_FORKNOEXEC;
 925         if (!(clone_flags & CLONE_PTRACE))
 926                 p->ptrace = 0;
 927         p->flags = new_flags;
 928 }
 929
 930 asmlinkage long sys_set_tid_address(int __user *tidptr)
 931 {
 932         current->clear_child_tid = tidptr;
 933
 934         return current->pid;
 935 }
 936
 937 static inline void rt_mutex_init_task(struct task_struct *p)
 938 {
 939         spin_lock_init(&p->pi_lock);
 940 #ifdef CONFIG_RT_MUTEXES
 941         plist_head_init(&p->pi_waiters, &p->pi_lock);
 942         p->pi_blocked_on = NULL;
 943 #endif
 944 }
 945
 946 /*
 947  * This creates a new process as a copy of the old one,
 948  * but does not actually start it yet.
 949  *
 950  * It copies the registers, and all the appropriate
 951  * parts of the process environment (as per the clone
 952  * flags). The actual kick-off is left to the caller.
 953  */
 954 static struct task_struct *copy_process(unsigned long clone_flags,
 955                                         unsigned long stack_start,
 956                                         struct pt_regs *regs,
 957                                         unsigned long stack_size,
 958                                         int __user *parent_tidptr,
 959                                         int __user *child_tidptr,
 960                                         struct pid *pid)
 961 {
 962         int retval;
 963         struct task_struct *p = NULL;
 964
 965         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 966                 return ERR_PTR(-EINVAL);
 967
 968         /*
 969          * Thread groups must share signals as well, and detached threads
 970          * can only be started up within the thread group.
 971          */
 972         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 973                 return ERR_PTR(-EINVAL);
 974
 975         /*
 976          * Shared signal handlers imply shared VM. By way of the above,
 977          * thread groups also imply shared VM. Blocking this case allows
 978          * for various simplifications in other code.
 979          */
 980         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 981                 return ERR_PTR(-EINVAL);
 982
 983         retval = security_task_create(clone_flags);
 984         if (retval)
 985                 goto fork_out;
 986
 987         retval = -ENOMEM;
 988         p = dup_task_struct(current);
 989         if (!p)
 990                 goto fork_out;
 991
 992         rt_mutex_init_task(p);
 993
 994 #ifdef CONFIG_TRACE_IRQFLAGS
 995         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
 996         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
 997 #endif
 998         retval = -EAGAIN;
 999         if (atomic_read(&p->user->processes) >=
1000                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1001                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1002                                 p->user != &root_user)
1003                         goto bad_fork_free;
1004         }
1005
1006         atomic_inc(&p->user->__count);
1007         atomic_inc(&p->user->processes);
1008         get_group_info(p->group_info);
1009
1010         /*
1011          * If multiple threads are within copy_process(), then this check
1012          * triggers too late. This doesn't hurt, the check is only there
1013          * to stop root fork bombs.
1014          */
1015         if (nr_threads >= max_threads)
1016                 goto bad_fork_cleanup_count;
1017
1018         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1019                 goto bad_fork_cleanup_count;
1020
1021         if (p->binfmt && !try_module_get(p->binfmt->module))
1022                 goto bad_fork_cleanup_put_domain;
1023
1024         p->did_exec = 0;
1025         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1026         copy_flags(clone_flags, p);
1027         p->pid = pid_nr(pid);
1028         retval = -EFAULT;
1029         if (clone_flags & CLONE_PARENT_SETTID)
1030                 if (put_user(p->pid, parent_tidptr))
1031                         goto bad_fork_cleanup_delays_binfmt;
1032
1033         INIT_LIST_HEAD(&p->children);
1034         INIT_LIST_HEAD(&p->sibling);
1035         p->vfork_done = NULL;
1036         spin_lock_init(&p->alloc_lock);
1037
1038         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1039         init_sigpending(&p->pending);
1040
1041         p->utime = cputime_zero;
1042         p->stime = cputime_zero;
1043
1044 #ifdef CONFIG_TASK_XACCT
1045         p->rchar = 0;           /* I/O counter: bytes read */
1046         p->wchar = 0;           /* I/O counter: bytes written */
1047         p->syscr = 0;           /* I/O counter: read syscalls */
1048         p->syscw = 0;           /* I/O counter: write syscalls */
1049 #endif
1050         task_io_accounting_init(p);
1051         acct_clear_integrals(p);
1052
1053         p->it_virt_expires = cputime_zero;
1054         p->it_prof_expires = cputime_zero;
1055         p->it_sched_expires = 0;
1056         INIT_LIST_HEAD(&p->cpu_timers[0]);
1057         INIT_LIST_HEAD(&p->cpu_timers[1]);
1058         INIT_LIST_HEAD(&p->cpu_timers[2]);
1059
1060         p->lock_depth = -1;             /* -1 = no lock */
1061         do_posix_clock_monotonic_gettime(&p->start_time);
1062         p->real_start_time = p->start_time;
1063         monotonic_to_bootbased(&p->real_start_time);
1064         p->security = NULL;
1065         p->io_context = NULL;
1066         p->io_wait = NULL;
1067         p->audit_context = NULL;
1068         cpuset_fork(p);
1069 #ifdef CONFIG_NUMA
1070         p->mempolicy = mpol_copy(p->mempolicy);
1071         if (IS_ERR(p->mempolicy)) {
1072                 retval = PTR_ERR(p->mempolicy);
1073                 p->mempolicy = NULL;
1074                 goto bad_fork_cleanup_cpuset;
1075         }
1076         mpol_fix_fork_child_flag(p);
1077 #endif
1078 #ifdef CONFIG_TRACE_IRQFLAGS
1079         p->irq_events = 0;
1080 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1081         p->hardirqs_enabled = 1;
1082 #else
1083         p->hardirqs_enabled = 0;
1084 #endif
1085         p->hardirq_enable_ip = 0;
1086         p->hardirq_enable_event = 0;
1087         p->hardirq_disable_ip = _THIS_IP_;
1088         p->hardirq_disable_event = 0;
1089         p->softirqs_enabled = 1;
1090         p->softirq_enable_ip = _THIS_IP_;
1091         p->softirq_enable_event = 0;
1092         p->softirq_disable_ip = 0;
1093         p->softirq_disable_event = 0;
1094         p->hardirq_context = 0;
1095         p->softirq_context = 0;
1096 #endif
1097 #ifdef CONFIG_LOCKDEP
1098         p->lockdep_depth = 0; /* no locks held yet */
1099         p->curr_chain_key = 0;
1100         p->lockdep_recursion = 0;
1101 #endif
1102
1103 #ifdef CONFIG_DEBUG_MUTEXES
1104         p->blocked_on = NULL; /* not blocked yet */
1105 #endif
1106
1107         p->tgid = p->pid;
1108         if (clone_flags & CLONE_THREAD)
1109                 p->tgid = current->tgid;
1110
1111         if ((retval = security_task_alloc(p)))
1112                 goto bad_fork_cleanup_policy;
1113         if ((retval = audit_alloc(p)))
1114                 goto bad_fork_cleanup_security;
1115         /* copy all the process information */
1116         if ((retval = copy_semundo(clone_flags, p)))
1117                 goto bad_fork_cleanup_audit;
1118         if ((retval = copy_files(clone_flags, p)))
1119                 goto bad_fork_cleanup_semundo;
1120         if ((retval = copy_fs(clone_flags, p)))
1121                 goto bad_fork_cleanup_files;
1122         if ((retval = copy_sighand(clone_flags, p)))
1123                 goto bad_fork_cleanup_fs;
1124         if ((retval = copy_signal(clone_flags, p)))
1125                 goto bad_fork_cleanup_sighand;
1126         if ((retval = copy_mm(clone_flags, p)))
1127                 goto bad_fork_cleanup_signal;
1128         if ((retval = copy_keys(clone_flags, p)))
1129                 goto bad_fork_cleanup_mm;
1130         if ((retval = copy_namespaces(clone_flags, p)))
1131                 goto bad_fork_cleanup_keys;
1132         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1133         if (retval)
1134                 goto bad_fork_cleanup_namespaces;
1135
1136         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1137         /*
1138          * Clear TID on mm_release()?
1139          */
1140         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1141         p->robust_list = NULL;
1142 #ifdef CONFIG_COMPAT
1143         p->compat_robust_list = NULL;
1144 #endif
1145         INIT_LIST_HEAD(&p->pi_state_list);
1146         p->pi_state_cache = NULL;
1147
1148         /*
1149          * sigaltstack should be cleared when sharing the same VM
1150          */
1151         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1152                 p->sas_ss_sp = p->sas_ss_size = 0;
1153
1154         /*
1155          * Syscall tracing should be turned off in the child regardless
1156          * of CLONE_PTRACE.
1157          */
1158         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1159 #ifdef TIF_SYSCALL_EMU
1160         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1161 #endif
1162
1163         /* Our parent execution domain becomes current domain
1164            These must match for thread signalling to apply */
1165         p->parent_exec_id = p->self_exec_id;
1166
1167         /* ok, now we should be set up.. */
1168         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1169         p->pdeath_signal = 0;
1170         p->exit_state = 0;
1171
1172         /*
1173          * Ok, make it visible to the rest of the system.
1174          * We dont wake it up yet.
1175          */
1176         p->group_leader = p;
1177         INIT_LIST_HEAD(&p->thread_group);
1178         INIT_LIST_HEAD(&p->ptrace_children);
1179         INIT_LIST_HEAD(&p->ptrace_list);
1180
1181         /* Perform scheduler related setup. Assign this task to a CPU. */
1182         sched_fork(p, clone_flags);
1183
1184         /* Need tasklist lock for parent etc handling! */
1185         write_lock_irq(&tasklist_lock);
1186
1187         /* for sys_ioprio_set(IOPRIO_WHO_PGRP) */
1188         p->ioprio = current->ioprio;
1189
1190         /*
1191          * The task hasn't been attached yet, so its cpus_allowed mask will
1192          * not be changed, nor will its assigned CPU.
1193          *
1194          * The cpus_allowed mask of the parent may have changed after it was
1195          * copied first time - so re-copy it here, then check the child's CPU
1196          * to ensure it is on a valid CPU (and if not, just force it back to
1197          * parent's CPU). This avoids alot of nasty races.
1198          */
1199         p->cpus_allowed = current->cpus_allowed;
1200         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1201                         !cpu_online(task_cpu(p))))
1202                 set_task_cpu(p, smp_processor_id());
1203
1204         /* CLONE_PARENT re-uses the old parent */
1205         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1206                 p->real_parent = current->real_parent;
1207         else
1208                 p->real_parent = current;
1209         p->parent = p->real_parent;
1210
1211         spin_lock(&current->sighand->siglock);
1212
1213         /*
1214          * Process group and session signals need to be delivered to just the
1215          * parent before the fork or both the parent and the child after the
1216          * fork. Restart if a signal comes in before we add the new process to
1217          * it's process group.
1218          * A fatal signal pending means that current will exit, so the new
1219          * thread can't slip out of an OOM kill (or normal SIGKILL).
1220          */
1221         recalc_sigpending();
1222         if (signal_pending(current)) {
1223                 spin_unlock(&current->sighand->siglock);
1224                 write_unlock_irq(&tasklist_lock);
1225                 retval = -ERESTARTNOINTR;
1226                 goto bad_fork_cleanup_namespaces;
1227         }
1228
1229         if (clone_flags & CLONE_THREAD) {
1230                 p->group_leader = current->group_leader;
1231                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1232
1233                 if (!cputime_eq(current->signal->it_virt_expires,
1234                                 cputime_zero) ||
1235                     !cputime_eq(current->signal->it_prof_expires,
1236                                 cputime_zero) ||
1237                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1238                     !list_empty(&current->signal->cpu_timers[0]) ||
1239                     !list_empty(&current->signal->cpu_timers[1]) ||
1240                     !list_empty(&current->signal->cpu_timers[2])) {
1241                         /*
1242                          * Have child wake up on its first tick to check
1243                          * for process CPU timers.
1244                          */
1245                         p->it_prof_expires = jiffies_to_cputime(1);
1246                 }
1247         }
1248
1249         if (likely(p->pid)) {
1250                 add_parent(p);
1251                 if (unlikely(p->ptrace & PT_PTRACED))
1252                         __ptrace_link(p, current->parent);
1253
1254                 if (thread_group_leader(p)) {
1255                         p->signal->tty = current->signal->tty;
1256                         p->signal->pgrp = process_group(current);
1257                         set_signal_session(p->signal, process_session(current));
1258                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1259                         attach_pid(p, PIDTYPE_SID, task_session(current));
1260
1261                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1262                         __get_cpu_var(process_counts)++;
1263                 }
1264                 attach_pid(p, PIDTYPE_PID, pid);
1265                 nr_threads++;
1266         }
1267
1268         total_forks++;
1269         spin_unlock(&current->sighand->siglock);
1270         write_unlock_irq(&tasklist_lock);
1271         proc_fork_connector(p);
1272         return p;
1273
1274 bad_fork_cleanup_namespaces:
1275         exit_task_namespaces(p);
1276 bad_fork_cleanup_keys:
1277         exit_keys(p);
1278 bad_fork_cleanup_mm:
1279         if (p->mm)
1280                 mmput(p->mm);
1281 bad_fork_cleanup_signal:
1282         cleanup_signal(p);
1283 bad_fork_cleanup_sighand:
1284         __cleanup_sighand(p->sighand);
1285 bad_fork_cleanup_fs:
1286         exit_fs(p); /* blocking */
1287 bad_fork_cleanup_files:
1288         exit_files(p); /* blocking */
1289 bad_fork_cleanup_semundo:
1290         exit_sem(p);
1291 bad_fork_cleanup_audit:
1292         audit_free(p);
1293 bad_fork_cleanup_security:
1294         security_task_free(p);
1295 bad_fork_cleanup_policy:
1296 #ifdef CONFIG_NUMA
1297         mpol_free(p->mempolicy);
1298 bad_fork_cleanup_cpuset:
1299 #endif
1300         cpuset_exit(p);
1301 bad_fork_cleanup_delays_binfmt:
1302         delayacct_tsk_free(p);
1303         if (p->binfmt)
1304                 module_put(p->binfmt->module);
1305 bad_fork_cleanup_put_domain:
1306         module_put(task_thread_info(p)->exec_domain->module);
1307 bad_fork_cleanup_count:
1308         put_group_info(p->group_info);
1309         atomic_dec(&p->user->processes);
1310         free_uid(p->user);
1311 bad_fork_free:
1312         free_task(p);
1313 fork_out:
1314         return ERR_PTR(retval);
1315 }
1316
1317 noinline struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1318 {
1319         memset(regs, 0, sizeof(struct pt_regs));
1320         return regs;
1321 }
1322
1323 struct task_struct * __cpuinit fork_idle(int cpu)
1324 {
1325         struct task_struct *task;
1326         struct pt_regs regs;
1327
1328         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL,
1329                                 &init_struct_pid);
1330         if (!IS_ERR(task))
1331                 init_idle(task, cpu);
1332
1333         return task;
1334 }
1335
1336 static inline int fork_traceflag (unsigned clone_flags)
1337 {
1338         if (clone_flags & CLONE_UNTRACED)
1339                 return 0;
1340         else if (clone_flags & CLONE_VFORK) {
1341                 if (current->ptrace & PT_TRACE_VFORK)
1342                         return PTRACE_EVENT_VFORK;
1343         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1344                 if (current->ptrace & PT_TRACE_CLONE)
1345                         return PTRACE_EVENT_CLONE;
1346         } else if (current->ptrace & PT_TRACE_FORK)
1347                 return PTRACE_EVENT_FORK;
1348
1349         return 0;
1350 }
1351
1352 /*
1353  *  Ok, this is the main fork-routine.
1354  *
1355  * It copies the process, and if successful kick-starts
1356  * it and waits for it to finish using the VM if required.
1357  */
1358 long do_fork(unsigned long clone_flags,
1359               unsigned long stack_start,
1360               struct pt_regs *regs,
1361               unsigned long stack_size,
1362               int __user *parent_tidptr,
1363               int __user *child_tidptr)
1364 {
1365         struct task_struct *p;
1366         int trace = 0;
1367         struct pid *pid = alloc_pid();
1368         long nr;
1369
1370         if (!pid)
1371                 return -EAGAIN;
1372         nr = pid->nr;
1373         if (unlikely(current->ptrace)) {
1374                 trace = fork_traceflag (clone_flags);
1375                 if (trace)
1376                         clone_flags |= CLONE_PTRACE;
1377         }
1378
1379         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
1380         /*
1381          * Do this prior waking up the new thread - the thread pointer
1382          * might get invalid after that point, if the thread exits quickly.
1383          */
1384         if (!IS_ERR(p)) {
1385                 struct completion vfork;
1386
1387                 if (clone_flags & CLONE_VFORK) {
1388                         p->vfork_done = &vfork;
1389                         init_completion(&vfork);
1390                 }
1391
1392                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1393                         /*
1394                          * We'll start up with an immediate SIGSTOP.
1395                          */
1396                         sigaddset(&p->pending.signal, SIGSTOP);
1397                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1398                 }
1399
1400                 if (!(clone_flags & CLONE_STOPPED))
1401                         wake_up_new_task(p, clone_flags);
1402                 else
1403                         p->state = TASK_STOPPED;
1404
1405                 if (unlikely (trace)) {
1406                         current->ptrace_message = nr;
1407                         ptrace_notify ((trace << 8) | SIGTRAP);
1408                 }
1409
1410                 if (clone_flags & CLONE_VFORK) {
1411                         freezer_do_not_count();
1412                         wait_for_completion(&vfork);
1413                         freezer_count();
1414                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1415                                 current->ptrace_message = nr;
1416                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1417                         }
1418                 }
1419         } else {
1420                 free_pid(pid);
1421                 nr = PTR_ERR(p);
1422         }
1423         return nr;
1424 }
1425
1426 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1427 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1428 #endif
1429
1430 static void sighand_ctor(void *data, struct kmem_cache *cachep,
1431                         unsigned long flags)
1432 {
1433         struct sighand_struct *sighand = data;
1434
1435         spin_lock_init(&sighand->siglock);
1436         INIT_LIST_HEAD(&sighand->signalfd_list);
1437 }
1438
1439 void __init proc_caches_init(void)
1440 {
1441         sighand_cachep = kmem_cache_create("sighand_cache",
1442                         sizeof(struct sighand_struct), 0,
1443                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1444                         sighand_ctor, NULL);
1445         signal_cachep = kmem_cache_create("signal_cache",
1446                         sizeof(struct signal_struct), 0,
1447                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1448         files_cachep = kmem_cache_create("files_cache",
1449                         sizeof(struct files_struct), 0,
1450                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1451         fs_cachep = kmem_cache_create("fs_cache",
1452                         sizeof(struct fs_struct), 0,
1453                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1454         vm_area_cachep = kmem_cache_create("vm_area_struct",
1455                         sizeof(struct vm_area_struct), 0,
1456                         SLAB_PANIC, NULL, NULL);
1457         mm_cachep = kmem_cache_create("mm_struct",
1458                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1459                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1460 }
1461
1462 /*
1463  * Check constraints on flags passed to the unshare system call and
1464  * force unsharing of additional process context as appropriate.
1465  */
1466 static inline void check_unshare_flags(unsigned long *flags_ptr)
1467 {
1468         /*
1469          * If unsharing a thread from a thread group, must also
1470          * unshare vm.
1471          */
1472         if (*flags_ptr & CLONE_THREAD)
1473                 *flags_ptr |= CLONE_VM;
1474
1475         /*
1476          * If unsharing vm, must also unshare signal handlers.
1477          */
1478         if (*flags_ptr & CLONE_VM)
1479                 *flags_ptr |= CLONE_SIGHAND;
1480
1481         /*
1482          * If unsharing signal handlers and the task was created
1483          * using CLONE_THREAD, then must unshare the thread
1484          */
1485         if ((*flags_ptr & CLONE_SIGHAND) &&
1486             (atomic_read(&current->signal->count) > 1))
1487                 *flags_ptr |= CLONE_THREAD;
1488
1489         /*
1490          * If unsharing namespace, must also unshare filesystem information.
1491          */
1492         if (*flags_ptr & CLONE_NEWNS)
1493                 *flags_ptr |= CLONE_FS;
1494 }
1495
1496 /*
1497  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1498  */
1499 static int unshare_thread(unsigned long unshare_flags)
1500 {
1501         if (unshare_flags & CLONE_THREAD)
1502                 return -EINVAL;
1503
1504         return 0;
1505 }
1506
1507 /*
1508  * Unshare the filesystem structure if it is being shared
1509  */
1510 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1511 {
1512         struct fs_struct *fs = current->fs;
1513
1514         if ((unshare_flags & CLONE_FS) &&
1515             (fs && atomic_read(&fs->count) > 1)) {
1516                 *new_fsp = __copy_fs_struct(current->fs);
1517                 if (!*new_fsp)
1518                         return -ENOMEM;
1519         }
1520
1521         return 0;
1522 }
1523
1524 /*
1525  * Unsharing of sighand is not supported yet
1526  */
1527 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1528 {
1529         struct sighand_struct *sigh = current->sighand;
1530
1531         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1532                 return -EINVAL;
1533         else
1534                 return 0;
1535 }
1536
1537 /*
1538  * Unshare vm if it is being shared
1539  */
1540 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1541 {
1542         struct mm_struct *mm = current->mm;
1543
1544         if ((unshare_flags & CLONE_VM) &&
1545             (mm && atomic_read(&mm->mm_users) > 1)) {
1546                 return -EINVAL;
1547         }
1548
1549         return 0;
1550 }
1551
1552 /*
1553  * Unshare file descriptor table if it is being shared
1554  */
1555 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1556 {
1557         struct files_struct *fd = current->files;
1558         int error = 0;
1559
1560         if ((unshare_flags & CLONE_FILES) &&
1561             (fd && atomic_read(&fd->count) > 1)) {
1562                 *new_fdp = dup_fd(fd, &error);
1563                 if (!*new_fdp)
1564                         return error;
1565         }
1566
1567         return 0;
1568 }
1569
1570 /*
1571  * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1572  * supported yet
1573  */
1574 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1575 {
1576         if (unshare_flags & CLONE_SYSVSEM)
1577                 return -EINVAL;
1578
1579         return 0;
1580 }
1581
1582 /*
1583  * unshare allows a process to 'unshare' part of the process
1584  * context which was originally shared using clone.  copy_*
1585  * functions used by do_fork() cannot be used here directly
1586  * because they modify an inactive task_struct that is being
1587  * constructed. Here we are modifying the current, active,
1588  * task_struct.
1589  */
1590 asmlinkage long sys_unshare(unsigned long unshare_flags)
1591 {
1592         int err = 0;
1593         struct fs_struct *fs, *new_fs = NULL;
1594         struct sighand_struct *new_sigh = NULL;
1595         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1596         struct files_struct *fd, *new_fd = NULL;
1597         struct sem_undo_list *new_ulist = NULL;
1598         struct nsproxy *new_nsproxy = NULL, *old_nsproxy = NULL;
1599
1600         check_unshare_flags(&unshare_flags);
1601
1602         /* Return -EINVAL for all unsupported flags */
1603         err = -EINVAL;
1604         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1605                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1606                                 CLONE_NEWUTS|CLONE_NEWIPC))
1607                 goto bad_unshare_out;
1608
1609         if ((err = unshare_thread(unshare_flags)))
1610                 goto bad_unshare_out;
1611         if ((err = unshare_fs(unshare_flags, &new_fs)))
1612                 goto bad_unshare_cleanup_thread;
1613         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1614                 goto bad_unshare_cleanup_fs;
1615         if ((err = unshare_vm(unshare_flags, &new_mm)))
1616                 goto bad_unshare_cleanup_sigh;
1617         if ((err = unshare_fd(unshare_flags, &new_fd)))
1618                 goto bad_unshare_cleanup_vm;
1619         if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1620                 goto bad_unshare_cleanup_fd;
1621         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1622                         new_fs)))
1623                 goto bad_unshare_cleanup_semundo;
1624
1625         if (new_fs ||  new_mm || new_fd || new_ulist || new_nsproxy) {
1626
1627                 task_lock(current);
1628
1629                 if (new_nsproxy) {
1630                         old_nsproxy = current->nsproxy;
1631                         current->nsproxy = new_nsproxy;
1632                         new_nsproxy = old_nsproxy;
1633                 }
1634
1635                 if (new_fs) {
1636                         fs = current->fs;
1637                         current->fs = new_fs;
1638                         new_fs = fs;
1639                 }
1640
1641                 if (new_mm) {
1642                         mm = current->mm;
1643                         active_mm = current->active_mm;
1644                         current->mm = new_mm;
1645                         current->active_mm = new_mm;
1646                         activate_mm(active_mm, new_mm);
1647                         new_mm = mm;
1648                 }
1649
1650                 if (new_fd) {
1651                         fd = current->files;
1652                         current->files = new_fd;
1653                         new_fd = fd;
1654                 }
1655
1656                 task_unlock(current);
1657         }
1658
1659         if (new_nsproxy)
1660                 put_nsproxy(new_nsproxy);
1661
1662 bad_unshare_cleanup_semundo:
1663 bad_unshare_cleanup_fd:
1664         if (new_fd)
1665                 put_files_struct(new_fd);
1666
1667 bad_unshare_cleanup_vm:
1668         if (new_mm)
1669                 mmput(new_mm);
1670
1671 bad_unshare_cleanup_sigh:
1672         if (new_sigh)
1673                 if (atomic_dec_and_test(&new_sigh->count))
1674                         kmem_cache_free(sighand_cachep, new_sigh);
1675
1676 bad_unshare_cleanup_fs:
1677         if (new_fs)
1678                 put_fs_struct(new_fs);
1679
1680 bad_unshare_cleanup_thread:
1681 bad_unshare_out:
1682         return err;
1683 }