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