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