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/cgroup.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 #include <linux/tty.h>
  53 #include <linux/proc_fs.h>
  54
  55 #include <asm/pgtable.h>
  56 #include <asm/pgalloc.h>
  57 #include <asm/uaccess.h>
  58 #include <asm/mmu_context.h>
  59 #include <asm/cacheflush.h>
  60 #include <asm/tlbflush.h>
  61
  62 /*
  63  * Protected counters by write_lock_irq(&tasklist_lock)
  64  */
  65 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  66 int nr_threads;                 /* The idle threads do not count.. */
  67
  68 int max_threads;                /* tunable limit on nr_threads */
  69
  70 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  71
  72 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  73
  74 int nr_processes(void)
  75 {
  76         int cpu;
  77         int total = 0;
  78
  79         for_each_online_cpu(cpu)
  80                 total += per_cpu(process_counts, cpu);
  81
  82         return total;
  83 }
  84
  85 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  86 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  87 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  88 static struct kmem_cache *task_struct_cachep;
  89 #endif
  90
  91 /* SLAB cache for signal_struct structures (tsk->signal) */
  92 static struct kmem_cache *signal_cachep;
  93
  94 /* SLAB cache for sighand_struct structures (tsk->sighand) */
  95 struct kmem_cache *sighand_cachep;
  96
  97 /* SLAB cache for files_struct structures (tsk->files) */
  98 struct kmem_cache *files_cachep;
  99
 100 /* SLAB cache for fs_struct structures (tsk->fs) */
 101 struct kmem_cache *fs_cachep;
 102
 103 /* SLAB cache for vm_area_struct structures */
 104 struct kmem_cache *vm_area_cachep;
 105
 106 /* SLAB cache for mm_struct structures (tsk->mm) */
 107 static struct kmem_cache *mm_cachep;
 108
 109 void free_task(struct task_struct *tsk)
 110 {
 111         prop_local_destroy_single(&tsk->dirties);
 112         free_thread_info(tsk->stack);
 113         rt_mutex_debug_task_free(tsk);
 114         free_task_struct(tsk);
 115 }
 116 EXPORT_SYMBOL(free_task);
 117
 118 void __put_task_struct(struct task_struct *tsk)
 119 {
 120         WARN_ON(!tsk->exit_state);
 121         WARN_ON(atomic_read(&tsk->usage));
 122         WARN_ON(tsk == current);
 123
 124         security_task_free(tsk);
 125         free_uid(tsk->user);
 126         put_group_info(tsk->group_info);
 127         delayacct_tsk_free(tsk);
 128
 129         if (!profile_handoff_task(tsk))
 130                 free_task(tsk);
 131 }
 132
 133 void __init fork_init(unsigned long mempages)
 134 {
 135 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 136 #ifndef ARCH_MIN_TASKALIGN
 137 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 138 #endif
 139         /* create a slab on which task_structs can be allocated */
 140         task_struct_cachep =
 141                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 142                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
 143 #endif
 144
 145         /*
 146          * The default maximum number of threads is set to a safe
 147          * value: the thread structures can take up at most half
 148          * of memory.
 149          */
 150         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 151
 152         /*
 153          * we need to allow at least 20 threads to boot a system
 154          */
 155         if(max_threads < 20)
 156                 max_threads = 20;
 157
 158         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 159         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 160         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 161                 init_task.signal->rlim[RLIMIT_NPROC];
 162 }
 163
 164 static struct task_struct *dup_task_struct(struct task_struct *orig)
 165 {
 166         struct task_struct *tsk;
 167         struct thread_info *ti;
 168         int err;
 169
 170         prepare_to_copy(orig);
 171
 172         tsk = alloc_task_struct();
 173         if (!tsk)
 174                 return NULL;
 175
 176         ti = alloc_thread_info(tsk);
 177         if (!ti) {
 178                 free_task_struct(tsk);
 179                 return NULL;
 180         }
 181
 182         *tsk = *orig;
 183         tsk->stack = ti;
 184
 185         err = prop_local_init_single(&tsk->dirties);
 186         if (err) {
 187                 free_thread_info(ti);
 188                 free_task_struct(tsk);
 189                 return NULL;
 190         }
 191
 192         setup_thread_stack(tsk, orig);
 193
 194 #ifdef CONFIG_CC_STACKPROTECTOR
 195         tsk->stack_canary = get_random_int();
 196 #endif
 197
 198         /* One for us, one for whoever does the "release_task()" (usually parent) */
 199         atomic_set(&tsk->usage,2);
 200         atomic_set(&tsk->fs_excl, 0);
 201 #ifdef CONFIG_BLK_DEV_IO_TRACE
 202         tsk->btrace_seq = 0;
 203 #endif
 204         tsk->splice_pipe = NULL;
 205         return tsk;
 206 }
 207
 208 #ifdef CONFIG_MMU
 209 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 210 {
 211         struct vm_area_struct *mpnt, *tmp, **pprev;
 212         struct rb_node **rb_link, *rb_parent;
 213         int retval;
 214         unsigned long charge;
 215         struct mempolicy *pol;
 216
 217         down_write(&oldmm->mmap_sem);
 218         flush_cache_dup_mm(oldmm);
 219         /*
 220          * Not linked in yet - no deadlock potential:
 221          */
 222         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 223
 224         mm->locked_vm = 0;
 225         mm->mmap = NULL;
 226         mm->mmap_cache = NULL;
 227         mm->free_area_cache = oldmm->mmap_base;
 228         mm->cached_hole_size = ~0UL;
 229         mm->map_count = 0;
 230         cpus_clear(mm->cpu_vm_mask);
 231         mm->mm_rb = RB_ROOT;
 232         rb_link = &mm->mm_rb.rb_node;
 233         rb_parent = NULL;
 234         pprev = &mm->mmap;
 235
 236         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 237                 struct file *file;
 238
 239                 if (mpnt->vm_flags & VM_DONTCOPY) {
 240                         long pages = vma_pages(mpnt);
 241                         mm->total_vm -= pages;
 242                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 243                                                                 -pages);
 244                         continue;
 245                 }
 246                 charge = 0;
 247                 if (mpnt->vm_flags & VM_ACCOUNT) {
 248                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 249                         if (security_vm_enough_memory(len))
 250                                 goto fail_nomem;
 251                         charge = len;
 252                 }
 253                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 254                 if (!tmp)
 255                         goto fail_nomem;
 256                 *tmp = *mpnt;
 257                 pol = mpol_copy(vma_policy(mpnt));
 258                 retval = PTR_ERR(pol);
 259                 if (IS_ERR(pol))
 260                         goto fail_nomem_policy;
 261                 vma_set_policy(tmp, pol);
 262                 tmp->vm_flags &= ~VM_LOCKED;
 263                 tmp->vm_mm = mm;
 264                 tmp->vm_next = NULL;
 265                 anon_vma_link(tmp);
 266                 file = tmp->vm_file;
 267                 if (file) {
 268                         struct inode *inode = file->f_path.dentry->d_inode;
 269                         get_file(file);
 270                         if (tmp->vm_flags & VM_DENYWRITE)
 271                                 atomic_dec(&inode->i_writecount);
 272
 273                         /* insert tmp into the share list, just after mpnt */
 274                         spin_lock(&file->f_mapping->i_mmap_lock);
 275                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 276                         flush_dcache_mmap_lock(file->f_mapping);
 277                         vma_prio_tree_add(tmp, mpnt);
 278                         flush_dcache_mmap_unlock(file->f_mapping);
 279                         spin_unlock(&file->f_mapping->i_mmap_lock);
 280                 }
 281
 282                 /*
 283                  * Link in the new vma and copy the page table entries.
 284                  */
 285                 *pprev = tmp;
 286                 pprev = &tmp->vm_next;
 287
 288                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 289                 rb_link = &tmp->vm_rb.rb_right;
 290                 rb_parent = &tmp->vm_rb;
 291
 292                 mm->map_count++;
 293                 retval = copy_page_range(mm, oldmm, mpnt);
 294
 295                 if (tmp->vm_ops && tmp->vm_ops->open)
 296                         tmp->vm_ops->open(tmp);
 297
 298                 if (retval)
 299                         goto out;
 300         }
 301         /* a new mm has just been created */
 302         arch_dup_mmap(oldmm, mm);
 303         retval = 0;
 304 out:
 305         up_write(&mm->mmap_sem);
 306         flush_tlb_mm(oldmm);
 307         up_write(&oldmm->mmap_sem);
 308         return retval;
 309 fail_nomem_policy:
 310         kmem_cache_free(vm_area_cachep, tmp);
 311 fail_nomem:
 312         retval = -ENOMEM;
 313         vm_unacct_memory(charge);
 314         goto out;
 315 }
 316
 317 static inline int mm_alloc_pgd(struct mm_struct * mm)
 318 {
 319         mm->pgd = pgd_alloc(mm);
 320         if (unlikely(!mm->pgd))
 321                 return -ENOMEM;
 322         return 0;
 323 }
 324
 325 static inline void mm_free_pgd(struct mm_struct * mm)
 326 {
 327         pgd_free(mm->pgd);
 328 }
 329 #else
 330 #define dup_mmap(mm, oldmm)     (0)
 331 #define mm_alloc_pgd(mm)        (0)
 332 #define mm_free_pgd(mm)
 333 #endif /* CONFIG_MMU */
 334
 335 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 336
 337 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 338 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 339
 340 #include <linux/init_task.h>
 341
 342 static struct mm_struct * mm_init(struct mm_struct * mm)
 343 {
 344         atomic_set(&mm->mm_users, 1);
 345         atomic_set(&mm->mm_count, 1);
 346         init_rwsem(&mm->mmap_sem);
 347         INIT_LIST_HEAD(&mm->mmlist);
 348         mm->flags = (current->mm) ? current->mm->flags
 349                                   : MMF_DUMP_FILTER_DEFAULT;
 350         mm->core_waiters = 0;
 351         mm->nr_ptes = 0;
 352         set_mm_counter(mm, file_rss, 0);
 353         set_mm_counter(mm, anon_rss, 0);
 354         spin_lock_init(&mm->page_table_lock);
 355         rwlock_init(&mm->ioctx_list_lock);
 356         mm->ioctx_list = NULL;
 357         mm->free_area_cache = TASK_UNMAPPED_BASE;
 358         mm->cached_hole_size = ~0UL;
 359
 360         if (likely(!mm_alloc_pgd(mm))) {
 361                 mm->def_flags = 0;
 362                 return mm;
 363         }
 364         free_mm(mm);
 365         return NULL;
 366 }
 367
 368 /*
 369  * Allocate and initialize an mm_struct.
 370  */
 371 struct mm_struct * mm_alloc(void)
 372 {
 373         struct mm_struct * mm;
 374
 375         mm = allocate_mm();
 376         if (mm) {
 377                 memset(mm, 0, sizeof(*mm));
 378                 mm = mm_init(mm);
 379         }
 380         return mm;
 381 }
 382
 383 /*
 384  * Called when the last reference to the mm
 385  * is dropped: either by a lazy thread or by
 386  * mmput. Free the page directory and the mm.
 387  */
 388 void fastcall __mmdrop(struct mm_struct *mm)
 389 {
 390         BUG_ON(mm == &init_mm);
 391         mm_free_pgd(mm);
 392         destroy_context(mm);
 393         free_mm(mm);
 394 }
 395
 396 /*
 397  * Decrement the use count and release all resources for an mm.
 398  */
 399 void mmput(struct mm_struct *mm)
 400 {
 401         might_sleep();
 402
 403         if (atomic_dec_and_test(&mm->mm_users)) {
 404                 exit_aio(mm);
 405                 exit_mmap(mm);
 406                 if (!list_empty(&mm->mmlist)) {
 407                         spin_lock(&mmlist_lock);
 408                         list_del(&mm->mmlist);
 409                         spin_unlock(&mmlist_lock);
 410                 }
 411                 put_swap_token(mm);
 412                 mmdrop(mm);
 413         }
 414 }
 415 EXPORT_SYMBOL_GPL(mmput);
 416
 417 /**
 418  * get_task_mm - acquire a reference to the task's mm
 419  *
 420  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
 421  * this kernel workthread has transiently adopted a user mm with use_mm,
 422  * to do its AIO) is not set and if so returns a reference to it, after
 423  * bumping up the use count.  User must release the mm via mmput()
 424  * after use.  Typically used by /proc and ptrace.
 425  */
 426 struct mm_struct *get_task_mm(struct task_struct *task)
 427 {
 428         struct mm_struct *mm;
 429
 430         task_lock(task);
 431         mm = task->mm;
 432         if (mm) {
 433                 if (task->flags & PF_BORROWED_MM)
 434                         mm = NULL;
 435                 else
 436                         atomic_inc(&mm->mm_users);
 437         }
 438         task_unlock(task);
 439         return mm;
 440 }
 441 EXPORT_SYMBOL_GPL(get_task_mm);
 442
 443 /* Please note the differences between mmput and mm_release.
 444  * mmput is called whenever we stop holding onto a mm_struct,
 445  * error success whatever.
 446  *
 447  * mm_release is called after a mm_struct has been removed
 448  * from the current process.
 449  *
 450  * This difference is important for error handling, when we
 451  * only half set up a mm_struct for a new process and need to restore
 452  * the old one.  Because we mmput the new mm_struct before
 453  * restoring the old one. . .
 454  * Eric Biederman 10 January 1998
 455  */
 456 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 457 {
 458         struct completion *vfork_done = tsk->vfork_done;
 459
 460         /* Get rid of any cached register state */
 461         deactivate_mm(tsk, mm);
 462
 463         /* notify parent sleeping on vfork() */
 464         if (vfork_done) {
 465                 tsk->vfork_done = NULL;
 466                 complete(vfork_done);
 467         }
 468
 469         /*
 470          * If we're exiting normally, clear a user-space tid field if
 471          * requested.  We leave this alone when dying by signal, to leave
 472          * the value intact in a core dump, and to save the unnecessary
 473          * trouble otherwise.  Userland only wants this done for a sys_exit.
 474          */
 475         if (tsk->clear_child_tid
 476             && !(tsk->flags & PF_SIGNALED)
 477             && atomic_read(&mm->mm_users) > 1) {
 478                 u32 __user * tidptr = tsk->clear_child_tid;
 479                 tsk->clear_child_tid = NULL;
 480
 481                 /*
 482                  * We don't check the error code - if userspace has
 483                  * not set up a proper pointer then tough luck.
 484                  */
 485                 put_user(0, tidptr);
 486                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 487         }
 488 }
 489
 490 /*
 491  * Allocate a new mm structure and copy contents from the
 492  * mm structure of the passed in task structure.
 493  */
 494 static struct mm_struct *dup_mm(struct task_struct *tsk)
 495 {
 496         struct mm_struct *mm, *oldmm = current->mm;
 497         int err;
 498
 499         if (!oldmm)
 500                 return NULL;
 501
 502         mm = allocate_mm();
 503         if (!mm)
 504                 goto fail_nomem;
 505
 506         memcpy(mm, oldmm, sizeof(*mm));
 507
 508         /* Initializing for Swap token stuff */
 509         mm->token_priority = 0;
 510         mm->last_interval = 0;
 511
 512         if (!mm_init(mm))
 513                 goto fail_nomem;
 514
 515         if (init_new_context(tsk, mm))
 516                 goto fail_nocontext;
 517
 518         err = dup_mmap(mm, oldmm);
 519         if (err)
 520                 goto free_pt;
 521
 522         mm->hiwater_rss = get_mm_rss(mm);
 523         mm->hiwater_vm = mm->total_vm;
 524
 525         return mm;
 526
 527 free_pt:
 528         mmput(mm);
 529
 530 fail_nomem:
 531         return NULL;
 532
 533 fail_nocontext:
 534         /*
 535          * If init_new_context() failed, we cannot use mmput() to free the mm
 536          * because it calls destroy_context()
 537          */
 538         mm_free_pgd(mm);
 539         free_mm(mm);
 540         return NULL;
 541 }
 542
 543 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 544 {
 545         struct mm_struct * mm, *oldmm;
 546         int retval;
 547
 548         tsk->min_flt = tsk->maj_flt = 0;
 549         tsk->nvcsw = tsk->nivcsw = 0;
 550
 551         tsk->mm = NULL;
 552         tsk->active_mm = NULL;
 553
 554         /*
 555          * Are we cloning a kernel thread?
 556          *
 557          * We need to steal a active VM for that..
 558          */
 559         oldmm = current->mm;
 560         if (!oldmm)
 561                 return 0;
 562
 563         if (clone_flags & CLONE_VM) {
 564                 atomic_inc(&oldmm->mm_users);
 565                 mm = oldmm;
 566                 goto good_mm;
 567         }
 568
 569         retval = -ENOMEM;
 570         mm = dup_mm(tsk);
 571         if (!mm)
 572                 goto fail_nomem;
 573
 574 good_mm:
 575         /* Initializing for Swap token stuff */
 576         mm->token_priority = 0;
 577         mm->last_interval = 0;
 578
 579         tsk->mm = mm;
 580         tsk->active_mm = mm;
 581         return 0;
 582
 583 fail_nomem:
 584         return retval;
 585 }
 586
 587 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 588 {
 589         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 590         /* We don't need to lock fs - think why ;-) */
 591         if (fs) {
 592                 atomic_set(&fs->count, 1);
 593                 rwlock_init(&fs->lock);
 594                 fs->umask = old->umask;
 595                 read_lock(&old->lock);
 596                 fs->rootmnt = mntget(old->rootmnt);
 597                 fs->root = dget(old->root);
 598                 fs->pwdmnt = mntget(old->pwdmnt);
 599                 fs->pwd = dget(old->pwd);
 600                 if (old->altroot) {
 601                         fs->altrootmnt = mntget(old->altrootmnt);
 602                         fs->altroot = dget(old->altroot);
 603                 } else {
 604                         fs->altrootmnt = NULL;
 605                         fs->altroot = NULL;
 606                 }
 607                 read_unlock(&old->lock);
 608         }
 609         return fs;
 610 }
 611
 612 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 613 {
 614         return __copy_fs_struct(old);
 615 }
 616
 617 EXPORT_SYMBOL_GPL(copy_fs_struct);
 618
 619 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 620 {
 621         if (clone_flags & CLONE_FS) {
 622                 atomic_inc(&current->fs->count);
 623                 return 0;
 624         }
 625         tsk->fs = __copy_fs_struct(current->fs);
 626         if (!tsk->fs)
 627                 return -ENOMEM;
 628         return 0;
 629 }
 630
 631 static int count_open_files(struct fdtable *fdt)
 632 {
 633         int size = fdt->max_fds;
 634         int i;
 635
 636         /* Find the last open fd */
 637         for (i = size/(8*sizeof(long)); i > 0; ) {
 638                 if (fdt->open_fds->fds_bits[--i])
 639                         break;
 640         }
 641         i = (i+1) * 8 * sizeof(long);
 642         return i;
 643 }
 644
 645 static struct files_struct *alloc_files(void)
 646 {
 647         struct files_struct *newf;
 648         struct fdtable *fdt;
 649
 650         newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
 651         if (!newf)
 652                 goto out;
 653
 654         atomic_set(&newf->count, 1);
 655
 656         spin_lock_init(&newf->file_lock);
 657         newf->next_fd = 0;
 658         fdt = &newf->fdtab;
 659         fdt->max_fds = NR_OPEN_DEFAULT;
 660         fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
 661         fdt->open_fds = (fd_set *)&newf->open_fds_init;
 662         fdt->fd = &newf->fd_array[0];
 663         INIT_RCU_HEAD(&fdt->rcu);
 664         fdt->next = NULL;
 665         rcu_assign_pointer(newf->fdt, fdt);
 666 out:
 667         return newf;
 668 }
 669
 670 /*
 671  * Allocate a new files structure and copy contents from the
 672  * passed in files structure.
 673  * errorp will be valid only when the returned files_struct is NULL.
 674  */
 675 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
 676 {
 677         struct files_struct *newf;
 678         struct file **old_fds, **new_fds;
 679         int open_files, size, i;
 680         struct fdtable *old_fdt, *new_fdt;
 681
 682         *errorp = -ENOMEM;
 683         newf = alloc_files();
 684         if (!newf)
 685                 goto out;
 686
 687         spin_lock(&oldf->file_lock);
 688         old_fdt = files_fdtable(oldf);
 689         new_fdt = files_fdtable(newf);
 690         open_files = count_open_files(old_fdt);
 691
 692         /*
 693          * Check whether we need to allocate a larger fd array and fd set.
 694          * Note: we're not a clone task, so the open count won't change.
 695          */
 696         if (open_files > new_fdt->max_fds) {
 697                 new_fdt->max_fds = 0;
 698                 spin_unlock(&oldf->file_lock);
 699                 spin_lock(&newf->file_lock);
 700                 *errorp = expand_files(newf, open_files-1);
 701                 spin_unlock(&newf->file_lock);
 702                 if (*errorp < 0)
 703                         goto out_release;
 704                 new_fdt = files_fdtable(newf);
 705                 /*
 706                  * Reacquire the oldf lock and a pointer to its fd table
 707                  * who knows it may have a new bigger fd table. We need
 708                  * the latest pointer.
 709                  */
 710                 spin_lock(&oldf->file_lock);
 711                 old_fdt = files_fdtable(oldf);
 712         }
 713
 714         old_fds = old_fdt->fd;
 715         new_fds = new_fdt->fd;
 716
 717         memcpy(new_fdt->open_fds->fds_bits,
 718                 old_fdt->open_fds->fds_bits, open_files/8);
 719         memcpy(new_fdt->close_on_exec->fds_bits,
 720                 old_fdt->close_on_exec->fds_bits, open_files/8);
 721
 722         for (i = open_files; i != 0; i--) {
 723                 struct file *f = *old_fds++;
 724                 if (f) {
 725                         get_file(f);
 726                 } else {
 727                         /*
 728                          * The fd may be claimed in the fd bitmap but not yet
 729                          * instantiated in the files array if a sibling thread
 730                          * is partway through open().  So make sure that this
 731                          * fd is available to the new process.
 732                          */
 733                         FD_CLR(open_files - i, new_fdt->open_fds);
 734                 }
 735                 rcu_assign_pointer(*new_fds++, f);
 736         }
 737         spin_unlock(&oldf->file_lock);
 738
 739         /* compute the remainder to be cleared */
 740         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
 741
 742         /* This is long word aligned thus could use a optimized version */
 743         memset(new_fds, 0, size);
 744
 745         if (new_fdt->max_fds > open_files) {
 746                 int left = (new_fdt->max_fds-open_files)/8;
 747                 int start = open_files / (8 * sizeof(unsigned long));
 748
 749                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
 750                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
 751         }
 752
 753         return newf;
 754
 755 out_release:
 756         kmem_cache_free(files_cachep, newf);
 757 out:
 758         return NULL;
 759 }
 760
 761 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 762 {
 763         struct files_struct *oldf, *newf;
 764         int error = 0;
 765
 766         /*
 767          * A background process may not have any files ...
 768          */
 769         oldf = current->files;
 770         if (!oldf)
 771                 goto out;
 772
 773         if (clone_flags & CLONE_FILES) {
 774                 atomic_inc(&oldf->count);
 775                 goto out;
 776         }
 777
 778         /*
 779          * Note: we may be using current for both targets (See exec.c)
 780          * This works because we cache current->files (old) as oldf. Don't
 781          * break this.
 782          */
 783         tsk->files = NULL;
 784         newf = dup_fd(oldf, &error);
 785         if (!newf)
 786                 goto out;
 787
 788         tsk->files = newf;
 789         error = 0;
 790 out:
 791         return error;
 792 }
 793
 794 /*
 795  *      Helper to unshare the files of the current task.
 796  *      We don't want to expose copy_files internals to
 797  *      the exec layer of the kernel.
 798  */
 799
 800 int unshare_files(void)
 801 {
 802         struct files_struct *files  = current->files;
 803         int rc;
 804
 805         BUG_ON(!files);
 806
 807         /* This can race but the race causes us to copy when we don't
 808            need to and drop the copy */
 809         if(atomic_read(&files->count) == 1)
 810         {
 811                 atomic_inc(&files->count);
 812                 return 0;
 813         }
 814         rc = copy_files(0, current);
 815         if(rc)
 816                 current->files = files;
 817         return rc;
 818 }
 819
 820 EXPORT_SYMBOL(unshare_files);
 821
 822 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 823 {
 824         struct sighand_struct *sig;
 825
 826         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 827                 atomic_inc(&current->sighand->count);
 828                 return 0;
 829         }
 830         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 831         rcu_assign_pointer(tsk->sighand, sig);
 832         if (!sig)
 833                 return -ENOMEM;
 834         atomic_set(&sig->count, 1);
 835         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 836         return 0;
 837 }
 838
 839 void __cleanup_sighand(struct sighand_struct *sighand)
 840 {
 841         if (atomic_dec_and_test(&sighand->count))
 842                 kmem_cache_free(sighand_cachep, sighand);
 843 }
 844
 845 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 846 {
 847         struct signal_struct *sig;
 848         int ret;
 849
 850         if (clone_flags & CLONE_THREAD) {
 851                 atomic_inc(&current->signal->count);
 852                 atomic_inc(&current->signal->live);
 853                 return 0;
 854         }
 855         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 856         tsk->signal = sig;
 857         if (!sig)
 858                 return -ENOMEM;
 859
 860         ret = copy_thread_group_keys(tsk);
 861         if (ret < 0) {
 862                 kmem_cache_free(signal_cachep, sig);
 863                 return ret;
 864         }
 865
 866         atomic_set(&sig->count, 1);
 867         atomic_set(&sig->live, 1);
 868         init_waitqueue_head(&sig->wait_chldexit);
 869         sig->flags = 0;
 870         sig->group_exit_code = 0;
 871         sig->group_exit_task = NULL;
 872         sig->group_stop_count = 0;
 873         sig->curr_target = NULL;
 874         init_sigpending(&sig->shared_pending);
 875         INIT_LIST_HEAD(&sig->posix_timers);
 876
 877         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 878         sig->it_real_incr.tv64 = 0;
 879         sig->real_timer.function = it_real_fn;
 880         sig->tsk = tsk;
 881
 882         sig->it_virt_expires = cputime_zero;
 883         sig->it_virt_incr = cputime_zero;
 884         sig->it_prof_expires = cputime_zero;
 885         sig->it_prof_incr = cputime_zero;
 886
 887         sig->leader = 0;        /* session leadership doesn't inherit */
 888         sig->tty_old_pgrp = NULL;
 889
 890         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 891         sig->gtime = cputime_zero;
 892         sig->cgtime = cputime_zero;
 893         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 894         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 895         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
 896         sig->sum_sched_runtime = 0;
 897         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 898         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 899         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 900         taskstats_tgid_init(sig);
 901
 902         task_lock(current->group_leader);
 903         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 904         task_unlock(current->group_leader);
 905
 906         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 907                 /*
 908                  * New sole thread in the process gets an expiry time
 909                  * of the whole CPU time limit.
 910                  */
 911                 tsk->it_prof_expires =
 912                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 913         }
 914         acct_init_pacct(&sig->pacct);
 915
 916         tty_audit_fork(sig);
 917
 918         return 0;
 919 }
 920
 921 void __cleanup_signal(struct signal_struct *sig)
 922 {
 923         exit_thread_group_keys(sig);
 924         kmem_cache_free(signal_cachep, sig);
 925 }
 926
 927 static void cleanup_signal(struct task_struct *tsk)
 928 {
 929         struct signal_struct *sig = tsk->signal;
 930
 931         atomic_dec(&sig->live);
 932
 933         if (atomic_dec_and_test(&sig->count))
 934                 __cleanup_signal(sig);
 935 }
 936
 937 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 938 {
 939         unsigned long new_flags = p->flags;
 940
 941         new_flags &= ~PF_SUPERPRIV;
 942         new_flags |= PF_FORKNOEXEC;
 943         if (!(clone_flags & CLONE_PTRACE))
 944                 p->ptrace = 0;
 945         p->flags = new_flags;
 946         clear_freeze_flag(p);
 947 }
 948
 949 asmlinkage long sys_set_tid_address(int __user *tidptr)
 950 {
 951         current->clear_child_tid = tidptr;
 952
 953         return task_pid_vnr(current);
 954 }
 955
 956 static void rt_mutex_init_task(struct task_struct *p)
 957 {
 958         spin_lock_init(&p->pi_lock);
 959 #ifdef CONFIG_RT_MUTEXES
 960         plist_head_init(&p->pi_waiters, &p->pi_lock);
 961         p->pi_blocked_on = NULL;
 962 #endif
 963 }
 964
 965 /*
 966  * This creates a new process as a copy of the old one,
 967  * but does not actually start it yet.
 968  *
 969  * It copies the registers, and all the appropriate
 970  * parts of the process environment (as per the clone
 971  * flags). The actual kick-off is left to the caller.
 972  */
 973 static struct task_struct *copy_process(unsigned long clone_flags,
 974                                         unsigned long stack_start,
 975                                         struct pt_regs *regs,
 976                                         unsigned long stack_size,
 977                                         int __user *child_tidptr,
 978                                         struct pid *pid)
 979 {
 980         int retval;
 981         struct task_struct *p;
 982         int cgroup_callbacks_done = 0;
 983
 984         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 985                 return ERR_PTR(-EINVAL);
 986
 987         /*
 988          * Thread groups must share signals as well, and detached threads
 989          * can only be started up within the thread group.
 990          */
 991         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 992                 return ERR_PTR(-EINVAL);
 993
 994         /*
 995          * Shared signal handlers imply shared VM. By way of the above,
 996          * thread groups also imply shared VM. Blocking this case allows
 997          * for various simplifications in other code.
 998          */
 999         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1000                 return ERR_PTR(-EINVAL);
1001
1002         retval = security_task_create(clone_flags);
1003         if (retval)
1004                 goto fork_out;
1005
1006         retval = -ENOMEM;
1007         p = dup_task_struct(current);
1008         if (!p)
1009                 goto fork_out;
1010
1011         rt_mutex_init_task(p);
1012
1013 #ifdef CONFIG_TRACE_IRQFLAGS
1014         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1015         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1016 #endif
1017         retval = -EAGAIN;
1018         if (atomic_read(&p->user->processes) >=
1019                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1020                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1021                     p->user != current->nsproxy->user_ns->root_user)
1022                         goto bad_fork_free;
1023         }
1024
1025         atomic_inc(&p->user->__count);
1026         atomic_inc(&p->user->processes);
1027         get_group_info(p->group_info);
1028
1029         /*
1030          * If multiple threads are within copy_process(), then this check
1031          * triggers too late. This doesn't hurt, the check is only there
1032          * to stop root fork bombs.
1033          */
1034         if (nr_threads >= max_threads)
1035                 goto bad_fork_cleanup_count;
1036
1037         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1038                 goto bad_fork_cleanup_count;
1039
1040         if (p->binfmt && !try_module_get(p->binfmt->module))
1041                 goto bad_fork_cleanup_put_domain;
1042
1043         p->did_exec = 0;
1044         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1045         copy_flags(clone_flags, p);
1046         INIT_LIST_HEAD(&p->children);
1047         INIT_LIST_HEAD(&p->sibling);
1048 #ifdef CONFIG_PREEMPT_RCU
1049         p->rcu_read_lock_nesting = 0;
1050         p->rcu_flipctr_idx = 0;
1051 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1052         p->vfork_done = NULL;
1053         spin_lock_init(&p->alloc_lock);
1054
1055         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1056         init_sigpending(&p->pending);
1057
1058         p->utime = cputime_zero;
1059         p->stime = cputime_zero;
1060         p->gtime = cputime_zero;
1061         p->utimescaled = cputime_zero;
1062         p->stimescaled = cputime_zero;
1063         p->prev_utime = cputime_zero;
1064         p->prev_stime = cputime_zero;
1065
1066 #ifdef CONFIG_DETECT_SOFTLOCKUP
1067         p->last_switch_count = 0;
1068         p->last_switch_timestamp = 0;
1069 #endif
1070
1071 #ifdef CONFIG_TASK_XACCT
1072         p->rchar = 0;           /* I/O counter: bytes read */
1073         p->wchar = 0;           /* I/O counter: bytes written */
1074         p->syscr = 0;           /* I/O counter: read syscalls */
1075         p->syscw = 0;           /* I/O counter: write syscalls */
1076 #endif
1077         task_io_accounting_init(p);
1078         acct_clear_integrals(p);
1079
1080         p->it_virt_expires = cputime_zero;
1081         p->it_prof_expires = cputime_zero;
1082         p->it_sched_expires = 0;
1083         INIT_LIST_HEAD(&p->cpu_timers[0]);
1084         INIT_LIST_HEAD(&p->cpu_timers[1]);
1085         INIT_LIST_HEAD(&p->cpu_timers[2]);
1086
1087         p->lock_depth = -1;             /* -1 = no lock */
1088         do_posix_clock_monotonic_gettime(&p->start_time);
1089         p->real_start_time = p->start_time;
1090         monotonic_to_bootbased(&p->real_start_time);
1091 #ifdef CONFIG_SECURITY
1092         p->security = NULL;
1093 #endif
1094         p->io_context = NULL;
1095         p->audit_context = NULL;
1096         cgroup_fork(p);
1097 #ifdef CONFIG_NUMA
1098         p->mempolicy = mpol_copy(p->mempolicy);
1099         if (IS_ERR(p->mempolicy)) {
1100                 retval = PTR_ERR(p->mempolicy);
1101                 p->mempolicy = NULL;
1102                 goto bad_fork_cleanup_cgroup;
1103         }
1104         mpol_fix_fork_child_flag(p);
1105 #endif
1106 #ifdef CONFIG_TRACE_IRQFLAGS
1107         p->irq_events = 0;
1108 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1109         p->hardirqs_enabled = 1;
1110 #else
1111         p->hardirqs_enabled = 0;
1112 #endif
1113         p->hardirq_enable_ip = 0;
1114         p->hardirq_enable_event = 0;
1115         p->hardirq_disable_ip = _THIS_IP_;
1116         p->hardirq_disable_event = 0;
1117         p->softirqs_enabled = 1;
1118         p->softirq_enable_ip = _THIS_IP_;
1119         p->softirq_enable_event = 0;
1120         p->softirq_disable_ip = 0;
1121         p->softirq_disable_event = 0;
1122         p->hardirq_context = 0;
1123         p->softirq_context = 0;
1124 #endif
1125 #ifdef CONFIG_LOCKDEP
1126         p->lockdep_depth = 0; /* no locks held yet */
1127         p->curr_chain_key = 0;
1128         p->lockdep_recursion = 0;
1129 #endif
1130
1131 #ifdef CONFIG_DEBUG_MUTEXES
1132         p->blocked_on = NULL; /* not blocked yet */
1133 #endif
1134
1135         /* Perform scheduler related setup. Assign this task to a CPU. */
1136         sched_fork(p, clone_flags);
1137
1138         if ((retval = security_task_alloc(p)))
1139                 goto bad_fork_cleanup_policy;
1140         if ((retval = audit_alloc(p)))
1141                 goto bad_fork_cleanup_security;
1142         /* copy all the process information */
1143         if ((retval = copy_semundo(clone_flags, p)))
1144                 goto bad_fork_cleanup_audit;
1145         if ((retval = copy_files(clone_flags, p)))
1146                 goto bad_fork_cleanup_semundo;
1147         if ((retval = copy_fs(clone_flags, p)))
1148                 goto bad_fork_cleanup_files;
1149         if ((retval = copy_sighand(clone_flags, p)))
1150                 goto bad_fork_cleanup_fs;
1151         if ((retval = copy_signal(clone_flags, p)))
1152                 goto bad_fork_cleanup_sighand;
1153         if ((retval = copy_mm(clone_flags, p)))
1154                 goto bad_fork_cleanup_signal;
1155         if ((retval = copy_keys(clone_flags, p)))
1156                 goto bad_fork_cleanup_mm;
1157         if ((retval = copy_namespaces(clone_flags, p)))
1158                 goto bad_fork_cleanup_keys;
1159         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1160         if (retval)
1161                 goto bad_fork_cleanup_namespaces;
1162
1163         if (pid != &init_struct_pid) {
1164                 retval = -ENOMEM;
1165                 pid = alloc_pid(task_active_pid_ns(p));
1166                 if (!pid)
1167                         goto bad_fork_cleanup_namespaces;
1168
1169                 if (clone_flags & CLONE_NEWPID) {
1170                         retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1171                         if (retval < 0)
1172                                 goto bad_fork_free_pid;
1173                 }
1174         }
1175
1176         p->pid = pid_nr(pid);
1177         p->tgid = p->pid;
1178         if (clone_flags & CLONE_THREAD)
1179                 p->tgid = current->tgid;
1180
1181         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1182         /*
1183          * Clear TID on mm_release()?
1184          */
1185         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1186 #ifdef CONFIG_FUTEX
1187         p->robust_list = NULL;
1188 #ifdef CONFIG_COMPAT
1189         p->compat_robust_list = NULL;
1190 #endif
1191         INIT_LIST_HEAD(&p->pi_state_list);
1192         p->pi_state_cache = NULL;
1193 #endif
1194         /*
1195          * sigaltstack should be cleared when sharing the same VM
1196          */
1197         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1198                 p->sas_ss_sp = p->sas_ss_size = 0;
1199
1200         /*
1201          * Syscall tracing should be turned off in the child regardless
1202          * of CLONE_PTRACE.
1203          */
1204         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1205 #ifdef TIF_SYSCALL_EMU
1206         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1207 #endif
1208         clear_all_latency_tracing(p);
1209
1210         /* Our parent execution domain becomes current domain
1211            These must match for thread signalling to apply */
1212         p->parent_exec_id = p->self_exec_id;
1213
1214         /* ok, now we should be set up.. */
1215         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1216         p->pdeath_signal = 0;
1217         p->exit_state = 0;
1218
1219         /*
1220          * Ok, make it visible to the rest of the system.
1221          * We dont wake it up yet.
1222          */
1223         p->group_leader = p;
1224         INIT_LIST_HEAD(&p->thread_group);
1225         INIT_LIST_HEAD(&p->ptrace_children);
1226         INIT_LIST_HEAD(&p->ptrace_list);
1227
1228         /* Now that the task is set up, run cgroup callbacks if
1229          * necessary. We need to run them before the task is visible
1230          * on the tasklist. */
1231         cgroup_fork_callbacks(p);
1232         cgroup_callbacks_done = 1;
1233
1234         /* Need tasklist lock for parent etc handling! */
1235         write_lock_irq(&tasklist_lock);
1236
1237         /* for sys_ioprio_set(IOPRIO_WHO_PGRP) */
1238         p->ioprio = current->ioprio;
1239
1240         /*
1241          * The task hasn't been attached yet, so its cpus_allowed mask will
1242          * not be changed, nor will its assigned CPU.
1243          *
1244          * The cpus_allowed mask of the parent may have changed after it was
1245          * copied first time - so re-copy it here, then check the child's CPU
1246          * to ensure it is on a valid CPU (and if not, just force it back to
1247          * parent's CPU). This avoids alot of nasty races.
1248          */
1249         p->cpus_allowed = current->cpus_allowed;
1250         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1251         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1252                         !cpu_online(task_cpu(p))))
1253                 set_task_cpu(p, smp_processor_id());
1254
1255         /* CLONE_PARENT re-uses the old parent */
1256         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1257                 p->real_parent = current->real_parent;
1258         else
1259                 p->real_parent = current;
1260         p->parent = p->real_parent;
1261
1262         spin_lock(&current->sighand->siglock);
1263
1264         /*
1265          * Process group and session signals need to be delivered to just the
1266          * parent before the fork or both the parent and the child after the
1267          * fork. Restart if a signal comes in before we add the new process to
1268          * it's process group.
1269          * A fatal signal pending means that current will exit, so the new
1270          * thread can't slip out of an OOM kill (or normal SIGKILL).
1271          */
1272         recalc_sigpending();
1273         if (signal_pending(current)) {
1274                 spin_unlock(&current->sighand->siglock);
1275                 write_unlock_irq(&tasklist_lock);
1276                 retval = -ERESTARTNOINTR;
1277                 goto bad_fork_free_pid;
1278         }
1279
1280         if (clone_flags & CLONE_THREAD) {
1281                 p->group_leader = current->group_leader;
1282                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1283
1284                 if (!cputime_eq(current->signal->it_virt_expires,
1285                                 cputime_zero) ||
1286                     !cputime_eq(current->signal->it_prof_expires,
1287                                 cputime_zero) ||
1288                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1289                     !list_empty(&current->signal->cpu_timers[0]) ||
1290                     !list_empty(&current->signal->cpu_timers[1]) ||
1291                     !list_empty(&current->signal->cpu_timers[2])) {
1292                         /*
1293                          * Have child wake up on its first tick to check
1294                          * for process CPU timers.
1295                          */
1296                         p->it_prof_expires = jiffies_to_cputime(1);
1297                 }
1298         }
1299
1300         if (likely(p->pid)) {
1301                 add_parent(p);
1302                 if (unlikely(p->ptrace & PT_PTRACED))
1303                         __ptrace_link(p, current->parent);
1304
1305                 if (thread_group_leader(p)) {
1306                         if (clone_flags & CLONE_NEWPID)
1307                                 p->nsproxy->pid_ns->child_reaper = p;
1308
1309                         p->signal->tty = current->signal->tty;
1310                         set_task_pgrp(p, task_pgrp_nr(current));
1311                         set_task_session(p, task_session_nr(current));
1312                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1313                         attach_pid(p, PIDTYPE_SID, task_session(current));
1314                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1315                         __get_cpu_var(process_counts)++;
1316                 }
1317                 attach_pid(p, PIDTYPE_PID, pid);
1318                 nr_threads++;
1319         }
1320
1321         total_forks++;
1322         spin_unlock(&current->sighand->siglock);
1323         write_unlock_irq(&tasklist_lock);
1324         proc_fork_connector(p);
1325         cgroup_post_fork(p);
1326         return p;
1327
1328 bad_fork_free_pid:
1329         if (pid != &init_struct_pid)
1330                 free_pid(pid);
1331 bad_fork_cleanup_namespaces:
1332         exit_task_namespaces(p);
1333 bad_fork_cleanup_keys:
1334         exit_keys(p);
1335 bad_fork_cleanup_mm:
1336         if (p->mm)
1337                 mmput(p->mm);
1338 bad_fork_cleanup_signal:
1339         cleanup_signal(p);
1340 bad_fork_cleanup_sighand:
1341         __cleanup_sighand(p->sighand);
1342 bad_fork_cleanup_fs:
1343         exit_fs(p); /* blocking */
1344 bad_fork_cleanup_files:
1345         exit_files(p); /* blocking */
1346 bad_fork_cleanup_semundo:
1347         exit_sem(p);
1348 bad_fork_cleanup_audit:
1349         audit_free(p);
1350 bad_fork_cleanup_security:
1351         security_task_free(p);
1352 bad_fork_cleanup_policy:
1353 #ifdef CONFIG_NUMA
1354         mpol_free(p->mempolicy);
1355 bad_fork_cleanup_cgroup:
1356 #endif
1357         cgroup_exit(p, cgroup_callbacks_done);
1358         delayacct_tsk_free(p);
1359         if (p->binfmt)
1360                 module_put(p->binfmt->module);
1361 bad_fork_cleanup_put_domain:
1362         module_put(task_thread_info(p)->exec_domain->module);
1363 bad_fork_cleanup_count:
1364         put_group_info(p->group_info);
1365         atomic_dec(&p->user->processes);
1366         free_uid(p->user);
1367 bad_fork_free:
1368         free_task(p);
1369 fork_out:
1370         return ERR_PTR(retval);
1371 }
1372
1373 noinline struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1374 {
1375         memset(regs, 0, sizeof(struct pt_regs));
1376         return regs;
1377 }
1378
1379 struct task_struct * __cpuinit fork_idle(int cpu)
1380 {
1381         struct task_struct *task;
1382         struct pt_regs regs;
1383
1384         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1385                                 &init_struct_pid);
1386         if (!IS_ERR(task))
1387                 init_idle(task, cpu);
1388
1389         return task;
1390 }
1391
1392 static int fork_traceflag(unsigned clone_flags)
1393 {
1394         if (clone_flags & CLONE_UNTRACED)
1395                 return 0;
1396         else if (clone_flags & CLONE_VFORK) {
1397                 if (current->ptrace & PT_TRACE_VFORK)
1398                         return PTRACE_EVENT_VFORK;
1399         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1400                 if (current->ptrace & PT_TRACE_CLONE)
1401                         return PTRACE_EVENT_CLONE;
1402         } else if (current->ptrace & PT_TRACE_FORK)
1403                 return PTRACE_EVENT_FORK;
1404
1405         return 0;
1406 }
1407
1408 /*
1409  *  Ok, this is the main fork-routine.
1410  *
1411  * It copies the process, and if successful kick-starts
1412  * it and waits for it to finish using the VM if required.
1413  */
1414 long do_fork(unsigned long clone_flags,
1415               unsigned long stack_start,
1416               struct pt_regs *regs,
1417               unsigned long stack_size,
1418               int __user *parent_tidptr,
1419               int __user *child_tidptr)
1420 {
1421         struct task_struct *p;
1422         int trace = 0;
1423         long nr;
1424
1425         if (unlikely(current->ptrace)) {
1426                 trace = fork_traceflag (clone_flags);
1427                 if (trace)
1428                         clone_flags |= CLONE_PTRACE;
1429         }
1430
1431         p = copy_process(clone_flags, stack_start, regs, stack_size,
1432                         child_tidptr, NULL);
1433         /*
1434          * Do this prior waking up the new thread - the thread pointer
1435          * might get invalid after that point, if the thread exits quickly.
1436          */
1437         if (!IS_ERR(p)) {
1438                 struct completion vfork;
1439
1440                 /*
1441                  * this is enough to call pid_nr_ns here, but this if
1442                  * improves optimisation of regular fork()
1443                  */
1444                 nr = (clone_flags & CLONE_NEWPID) ?
1445                         task_pid_nr_ns(p, current->nsproxy->pid_ns) :
1446                                 task_pid_vnr(p);
1447
1448                 if (clone_flags & CLONE_PARENT_SETTID)
1449                         put_user(nr, parent_tidptr);
1450
1451                 if (clone_flags & CLONE_VFORK) {
1452                         p->vfork_done = &vfork;
1453                         init_completion(&vfork);
1454                 }
1455
1456                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1457                         /*
1458                          * We'll start up with an immediate SIGSTOP.
1459                          */
1460                         sigaddset(&p->pending.signal, SIGSTOP);
1461                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1462                 }
1463
1464                 if (!(clone_flags & CLONE_STOPPED))
1465                         wake_up_new_task(p, clone_flags);
1466                 else
1467                         p->state = TASK_STOPPED;
1468
1469                 if (unlikely (trace)) {
1470                         current->ptrace_message = nr;
1471                         ptrace_notify ((trace << 8) | SIGTRAP);
1472                 }
1473
1474                 if (clone_flags & CLONE_VFORK) {
1475                         freezer_do_not_count();
1476                         wait_for_completion(&vfork);
1477                         freezer_count();
1478                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1479                                 current->ptrace_message = nr;
1480                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1481                         }
1482                 }
1483         } else {
1484                 nr = PTR_ERR(p);
1485         }
1486         return nr;
1487 }
1488
1489 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1490 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1491 #endif
1492
1493 static void sighand_ctor(struct kmem_cache *cachep, void *data)
1494 {
1495         struct sighand_struct *sighand = data;
1496
1497         spin_lock_init(&sighand->siglock);
1498         init_waitqueue_head(&sighand->signalfd_wqh);
1499 }
1500
1501 void __init proc_caches_init(void)
1502 {
1503         sighand_cachep = kmem_cache_create("sighand_cache",
1504                         sizeof(struct sighand_struct), 0,
1505                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1506                         sighand_ctor);
1507         signal_cachep = kmem_cache_create("signal_cache",
1508                         sizeof(struct signal_struct), 0,
1509                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1510         files_cachep = kmem_cache_create("files_cache",
1511                         sizeof(struct files_struct), 0,
1512                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1513         fs_cachep = kmem_cache_create("fs_cache",
1514                         sizeof(struct fs_struct), 0,
1515                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1516         vm_area_cachep = kmem_cache_create("vm_area_struct",
1517                         sizeof(struct vm_area_struct), 0,
1518                         SLAB_PANIC, NULL);
1519         mm_cachep = kmem_cache_create("mm_struct",
1520                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1521                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1522 }
1523
1524 /*
1525  * Check constraints on flags passed to the unshare system call and
1526  * force unsharing of additional process context as appropriate.
1527  */
1528 static void check_unshare_flags(unsigned long *flags_ptr)
1529 {
1530         /*
1531          * If unsharing a thread from a thread group, must also
1532          * unshare vm.
1533          */
1534         if (*flags_ptr & CLONE_THREAD)
1535                 *flags_ptr |= CLONE_VM;
1536
1537         /*
1538          * If unsharing vm, must also unshare signal handlers.
1539          */
1540         if (*flags_ptr & CLONE_VM)
1541                 *flags_ptr |= CLONE_SIGHAND;
1542
1543         /*
1544          * If unsharing signal handlers and the task was created
1545          * using CLONE_THREAD, then must unshare the thread
1546          */
1547         if ((*flags_ptr & CLONE_SIGHAND) &&
1548             (atomic_read(&current->signal->count) > 1))
1549                 *flags_ptr |= CLONE_THREAD;
1550
1551         /*
1552          * If unsharing namespace, must also unshare filesystem information.
1553          */
1554         if (*flags_ptr & CLONE_NEWNS)
1555                 *flags_ptr |= CLONE_FS;
1556 }
1557
1558 /*
1559  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1560  */
1561 static int unshare_thread(unsigned long unshare_flags)
1562 {
1563         if (unshare_flags & CLONE_THREAD)
1564                 return -EINVAL;
1565
1566         return 0;
1567 }
1568
1569 /*
1570  * Unshare the filesystem structure if it is being shared
1571  */
1572 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1573 {
1574         struct fs_struct *fs = current->fs;
1575
1576         if ((unshare_flags & CLONE_FS) &&
1577             (fs && atomic_read(&fs->count) > 1)) {
1578                 *new_fsp = __copy_fs_struct(current->fs);
1579                 if (!*new_fsp)
1580                         return -ENOMEM;
1581         }
1582
1583         return 0;
1584 }
1585
1586 /*
1587  * Unsharing of sighand is not supported yet
1588  */
1589 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1590 {
1591         struct sighand_struct *sigh = current->sighand;
1592
1593         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1594                 return -EINVAL;
1595         else
1596                 return 0;
1597 }
1598
1599 /*
1600  * Unshare vm if it is being shared
1601  */
1602 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1603 {
1604         struct mm_struct *mm = current->mm;
1605
1606         if ((unshare_flags & CLONE_VM) &&
1607             (mm && atomic_read(&mm->mm_users) > 1)) {
1608                 return -EINVAL;
1609         }
1610
1611         return 0;
1612 }
1613
1614 /*
1615  * Unshare file descriptor table if it is being shared
1616  */
1617 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1618 {
1619         struct files_struct *fd = current->files;
1620         int error = 0;
1621
1622         if ((unshare_flags & CLONE_FILES) &&
1623             (fd && atomic_read(&fd->count) > 1)) {
1624                 *new_fdp = dup_fd(fd, &error);
1625                 if (!*new_fdp)
1626                         return error;
1627         }
1628
1629         return 0;
1630 }
1631
1632 /*
1633  * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1634  * supported yet
1635  */
1636 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1637 {
1638         if (unshare_flags & CLONE_SYSVSEM)
1639                 return -EINVAL;
1640
1641         return 0;
1642 }
1643
1644 /*
1645  * unshare allows a process to 'unshare' part of the process
1646  * context which was originally shared using clone.  copy_*
1647  * functions used by do_fork() cannot be used here directly
1648  * because they modify an inactive task_struct that is being
1649  * constructed. Here we are modifying the current, active,
1650  * task_struct.
1651  */
1652 asmlinkage long sys_unshare(unsigned long unshare_flags)
1653 {
1654         int err = 0;
1655         struct fs_struct *fs, *new_fs = NULL;
1656         struct sighand_struct *new_sigh = NULL;
1657         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1658         struct files_struct *fd, *new_fd = NULL;
1659         struct sem_undo_list *new_ulist = NULL;
1660         struct nsproxy *new_nsproxy = NULL;
1661
1662         check_unshare_flags(&unshare_flags);
1663
1664         /* Return -EINVAL for all unsupported flags */
1665         err = -EINVAL;
1666         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1667                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1668                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
1669                                 CLONE_NEWNET))
1670                 goto bad_unshare_out;
1671
1672         if ((err = unshare_thread(unshare_flags)))
1673                 goto bad_unshare_out;
1674         if ((err = unshare_fs(unshare_flags, &new_fs)))
1675                 goto bad_unshare_cleanup_thread;
1676         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1677                 goto bad_unshare_cleanup_fs;
1678         if ((err = unshare_vm(unshare_flags, &new_mm)))
1679                 goto bad_unshare_cleanup_sigh;
1680         if ((err = unshare_fd(unshare_flags, &new_fd)))
1681                 goto bad_unshare_cleanup_vm;
1682         if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1683                 goto bad_unshare_cleanup_fd;
1684         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1685                         new_fs)))
1686                 goto bad_unshare_cleanup_semundo;
1687
1688         if (new_fs ||  new_mm || new_fd || new_ulist || new_nsproxy) {
1689
1690                 if (new_nsproxy) {
1691                         switch_task_namespaces(current, new_nsproxy);
1692                         new_nsproxy = NULL;
1693                 }
1694
1695                 task_lock(current);
1696
1697                 if (new_fs) {
1698                         fs = current->fs;
1699                         current->fs = new_fs;
1700                         new_fs = fs;
1701                 }
1702
1703                 if (new_mm) {
1704                         mm = current->mm;
1705                         active_mm = current->active_mm;
1706                         current->mm = new_mm;
1707                         current->active_mm = new_mm;
1708                         activate_mm(active_mm, new_mm);
1709                         new_mm = mm;
1710                 }
1711
1712                 if (new_fd) {
1713                         fd = current->files;
1714                         current->files = new_fd;
1715                         new_fd = fd;
1716                 }
1717
1718                 task_unlock(current);
1719         }
1720
1721         if (new_nsproxy)
1722                 put_nsproxy(new_nsproxy);
1723
1724 bad_unshare_cleanup_semundo:
1725 bad_unshare_cleanup_fd:
1726         if (new_fd)
1727                 put_files_struct(new_fd);
1728
1729 bad_unshare_cleanup_vm:
1730         if (new_mm)
1731                 mmput(new_mm);
1732
1733 bad_unshare_cleanup_sigh:
1734         if (new_sigh)
1735                 if (atomic_dec_and_test(&new_sigh->count))
1736                         kmem_cache_free(sighand_cachep, new_sigh);
1737
1738 bad_unshare_cleanup_fs:
1739         if (new_fs)
1740                 put_fs_struct(new_fs);
1741
1742 bad_unshare_cleanup_thread:
1743 bad_unshare_out:
1744         return err;
1745 }