kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/mnt_namespace.h>
  21 #include <linux/personality.h>
  22 #include <linux/mempolicy.h>
  23 #include <linux/sem.h>
  24 #include <linux/file.h>
  25 #include <linux/key.h>
  26 #include <linux/binfmts.h>
  27 #include <linux/mman.h>
  28 #include <linux/fs.h>
  29 #include <linux/nsproxy.h>
  30 #include <linux/capability.h>
  31 #include <linux/cpu.h>
  32 #include <linux/cgroup.h>
  33 #include <linux/security.h>
  34 #include <linux/swap.h>
  35 #include <linux/syscalls.h>
  36 #include <linux/jiffies.h>
  37 #include <linux/futex.h>
  38 #include <linux/task_io_accounting_ops.h>
  39 #include <linux/rcupdate.h>
  40 #include <linux/ptrace.h>
  41 #include <linux/mount.h>
  42 #include <linux/audit.h>
  43 #include <linux/profile.h>
  44 #include <linux/rmap.h>
  45 #include <linux/acct.h>
  46 #include <linux/tsacct_kern.h>
  47 #include <linux/cn_proc.h>
  48 #include <linux/freezer.h>
  49 #include <linux/delayacct.h>
  50 #include <linux/taskstats_kern.h>
  51 #include <linux/random.h>
  52 #include <linux/tty.h>
  53 #include <linux/proc_fs.h>
  54 #include <linux/blkdev.h>
  55
  56 #include <asm/pgtable.h>
  57 #include <asm/pgalloc.h>
  58 #include <asm/uaccess.h>
  59 #include <asm/mmu_context.h>
  60 #include <asm/cacheflush.h>
  61 #include <asm/tlbflush.h>
  62
  63 /*
  64  * Protected counters by write_lock_irq(&tasklist_lock)
  65  */
  66 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  67 int nr_threads;                 /* The idle threads do not count.. */
  68
  69 int max_threads;                /* tunable limit on nr_threads */
  70
  71 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  72
  73 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  74
  75 int nr_processes(void)
  76 {
  77         int cpu;
  78         int total = 0;
  79
  80         for_each_online_cpu(cpu)
  81                 total += per_cpu(process_counts, cpu);
  82
  83         return total;
  84 }
  85
  86 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  87 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  88 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  89 static struct kmem_cache *task_struct_cachep;
  90 #endif
  91
  92 /* SLAB cache for signal_struct structures (tsk->signal) */
  93 static struct kmem_cache *signal_cachep;
  94
  95 /* SLAB cache for sighand_struct structures (tsk->sighand) */
  96 struct kmem_cache *sighand_cachep;
  97
  98 /* SLAB cache for files_struct structures (tsk->files) */
  99 struct kmem_cache *files_cachep;
 100
 101 /* SLAB cache for fs_struct structures (tsk->fs) */
 102 struct kmem_cache *fs_cachep;
 103
 104 /* SLAB cache for vm_area_struct structures */
 105 struct kmem_cache *vm_area_cachep;
 106
 107 /* SLAB cache for mm_struct structures (tsk->mm) */
 108 static struct kmem_cache *mm_cachep;
 109
 110 void free_task(struct task_struct *tsk)
 111 {
 112         prop_local_destroy_single(&tsk->dirties);
 113         free_thread_info(tsk->stack);
 114         rt_mutex_debug_task_free(tsk);
 115         free_task_struct(tsk);
 116 }
 117 EXPORT_SYMBOL(free_task);
 118
 119 void __put_task_struct(struct task_struct *tsk)
 120 {
 121         WARN_ON(!tsk->exit_state);
 122         WARN_ON(atomic_read(&tsk->usage));
 123         WARN_ON(tsk == current);
 124
 125         security_task_free(tsk);
 126         free_uid(tsk->user);
 127         put_group_info(tsk->group_info);
 128         delayacct_tsk_free(tsk);
 129
 130         if (!profile_handoff_task(tsk))
 131                 free_task(tsk);
 132 }
 133
 134 void __init fork_init(unsigned long mempages)
 135 {
 136 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 137 #ifndef ARCH_MIN_TASKALIGN
 138 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 139 #endif
 140         /* create a slab on which task_structs can be allocated */
 141         task_struct_cachep =
 142                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 143                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
 144 #endif
 145
 146         /*
 147          * The default maximum number of threads is set to a safe
 148          * value: the thread structures can take up at most half
 149          * of memory.
 150          */
 151         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 152
 153         /*
 154          * we need to allow at least 20 threads to boot a system
 155          */
 156         if(max_threads < 20)
 157                 max_threads = 20;
 158
 159         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 160         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 161         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 162                 init_task.signal->rlim[RLIMIT_NPROC];
 163 }
 164
 165 static struct task_struct *dup_task_struct(struct task_struct *orig)
 166 {
 167         struct task_struct *tsk;
 168         struct thread_info *ti;
 169         int err;
 170
 171         prepare_to_copy(orig);
 172
 173         tsk = alloc_task_struct();
 174         if (!tsk)
 175                 return NULL;
 176
 177         ti = alloc_thread_info(tsk);
 178         if (!ti) {
 179                 free_task_struct(tsk);
 180                 return NULL;
 181         }
 182
 183         *tsk = *orig;
 184         tsk->stack = ti;
 185
 186         err = prop_local_init_single(&tsk->dirties);
 187         if (err) {
 188                 free_thread_info(ti);
 189                 free_task_struct(tsk);
 190                 return NULL;
 191         }
 192
 193         setup_thread_stack(tsk, orig);
 194
 195 #ifdef CONFIG_CC_STACKPROTECTOR
 196         tsk->stack_canary = get_random_int();
 197 #endif
 198
 199         /* One for us, one for whoever does the "release_task()" (usually parent) */
 200         atomic_set(&tsk->usage,2);
 201         atomic_set(&tsk->fs_excl, 0);
 202 #ifdef CONFIG_BLK_DEV_IO_TRACE
 203         tsk->btrace_seq = 0;
 204 #endif
 205         tsk->splice_pipe = NULL;
 206         return tsk;
 207 }
 208
 209 #ifdef CONFIG_MMU
 210 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 211 {
 212         struct vm_area_struct *mpnt, *tmp, **pprev;
 213         struct rb_node **rb_link, *rb_parent;
 214         int retval;
 215         unsigned long charge;
 216         struct mempolicy *pol;
 217
 218         down_write(&oldmm->mmap_sem);
 219         flush_cache_dup_mm(oldmm);
 220         /*
 221          * Not linked in yet - no deadlock potential:
 222          */
 223         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 224
 225         mm->locked_vm = 0;
 226         mm->mmap = NULL;
 227         mm->mmap_cache = NULL;
 228         mm->free_area_cache = oldmm->mmap_base;
 229         mm->cached_hole_size = ~0UL;
 230         mm->map_count = 0;
 231         cpus_clear(mm->cpu_vm_mask);
 232         mm->mm_rb = RB_ROOT;
 233         rb_link = &mm->mm_rb.rb_node;
 234         rb_parent = NULL;
 235         pprev = &mm->mmap;
 236
 237         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 238                 struct file *file;
 239
 240                 if (mpnt->vm_flags & VM_DONTCOPY) {
 241                         long pages = vma_pages(mpnt);
 242                         mm->total_vm -= pages;
 243                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 244                                                                 -pages);
 245                         continue;
 246                 }
 247                 charge = 0;
 248                 if (mpnt->vm_flags & VM_ACCOUNT) {
 249                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 250                         if (security_vm_enough_memory(len))
 251                                 goto fail_nomem;
 252                         charge = len;
 253                 }
 254                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 255                 if (!tmp)
 256                         goto fail_nomem;
 257                 *tmp = *mpnt;
 258                 pol = mpol_copy(vma_policy(mpnt));
 259                 retval = PTR_ERR(pol);
 260                 if (IS_ERR(pol))
 261                         goto fail_nomem_policy;
 262                 vma_set_policy(tmp, pol);
 263                 tmp->vm_flags &= ~VM_LOCKED;
 264                 tmp->vm_mm = mm;
 265                 tmp->vm_next = NULL;
 266                 anon_vma_link(tmp);
 267                 file = tmp->vm_file;
 268                 if (file) {
 269                         struct inode *inode = file->f_path.dentry->d_inode;
 270                         get_file(file);
 271                         if (tmp->vm_flags & VM_DENYWRITE)
 272                                 atomic_dec(&inode->i_writecount);
 273
 274                         /* insert tmp into the share list, just after mpnt */
 275                         spin_lock(&file->f_mapping->i_mmap_lock);
 276                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 277                         flush_dcache_mmap_lock(file->f_mapping);
 278                         vma_prio_tree_add(tmp, mpnt);
 279                         flush_dcache_mmap_unlock(file->f_mapping);
 280                         spin_unlock(&file->f_mapping->i_mmap_lock);
 281                 }
 282
 283                 /*
 284                  * Link in the new vma and copy the page table entries.
 285                  */
 286                 *pprev = tmp;
 287                 pprev = &tmp->vm_next;
 288
 289                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 290                 rb_link = &tmp->vm_rb.rb_right;
 291                 rb_parent = &tmp->vm_rb;
 292
 293                 mm->map_count++;
 294                 retval = copy_page_range(mm, oldmm, mpnt);
 295
 296                 if (tmp->vm_ops && tmp->vm_ops->open)
 297                         tmp->vm_ops->open(tmp);
 298
 299                 if (retval)
 300                         goto out;
 301         }
 302         /* a new mm has just been created */
 303         arch_dup_mmap(oldmm, mm);
 304         retval = 0;
 305 out:
 306         up_write(&mm->mmap_sem);
 307         flush_tlb_mm(oldmm);
 308         up_write(&oldmm->mmap_sem);
 309         return retval;
 310 fail_nomem_policy:
 311         kmem_cache_free(vm_area_cachep, tmp);
 312 fail_nomem:
 313         retval = -ENOMEM;
 314         vm_unacct_memory(charge);
 315         goto out;
 316 }
 317
 318 static inline int mm_alloc_pgd(struct mm_struct * mm)
 319 {
 320         mm->pgd = pgd_alloc(mm);
 321         if (unlikely(!mm->pgd))
 322                 return -ENOMEM;
 323         return 0;
 324 }
 325
 326 static inline void mm_free_pgd(struct mm_struct * mm)
 327 {
 328         pgd_free(mm->pgd);
 329 }
 330 #else
 331 #define dup_mmap(mm, oldmm)     (0)
 332 #define mm_alloc_pgd(mm)        (0)
 333 #define mm_free_pgd(mm)
 334 #endif /* CONFIG_MMU */
 335
 336 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 337
 338 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 339 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 340
 341 #include <linux/init_task.h>
 342
 343 static struct mm_struct * mm_init(struct mm_struct * mm)
 344 {
 345         atomic_set(&mm->mm_users, 1);
 346         atomic_set(&mm->mm_count, 1);
 347         init_rwsem(&mm->mmap_sem);
 348         INIT_LIST_HEAD(&mm->mmlist);
 349         mm->flags = (current->mm) ? current->mm->flags
 350                                   : MMF_DUMP_FILTER_DEFAULT;
 351         mm->core_waiters = 0;
 352         mm->nr_ptes = 0;
 353         set_mm_counter(mm, file_rss, 0);
 354         set_mm_counter(mm, anon_rss, 0);
 355         spin_lock_init(&mm->page_table_lock);
 356         rwlock_init(&mm->ioctx_list_lock);
 357         mm->ioctx_list = NULL;
 358         mm->free_area_cache = TASK_UNMAPPED_BASE;
 359         mm->cached_hole_size = ~0UL;
 360
 361         if (likely(!mm_alloc_pgd(mm))) {
 362                 mm->def_flags = 0;
 363                 return mm;
 364         }
 365         free_mm(mm);
 366         return NULL;
 367 }
 368
 369 /*
 370  * Allocate and initialize an mm_struct.
 371  */
 372 struct mm_struct * mm_alloc(void)
 373 {
 374         struct mm_struct * mm;
 375
 376         mm = allocate_mm();
 377         if (mm) {
 378                 memset(mm, 0, sizeof(*mm));
 379                 mm = mm_init(mm);
 380         }
 381         return mm;
 382 }
 383
 384 /*
 385  * Called when the last reference to the mm
 386  * is dropped: either by a lazy thread or by
 387  * mmput. Free the page directory and the mm.
 388  */
 389 void fastcall __mmdrop(struct mm_struct *mm)
 390 {
 391         BUG_ON(mm == &init_mm);
 392         mm_free_pgd(mm);
 393         destroy_context(mm);
 394         free_mm(mm);
 395 }
 396
 397 /*
 398  * Decrement the use count and release all resources for an mm.
 399  */
 400 void mmput(struct mm_struct *mm)
 401 {
 402         might_sleep();
 403
 404         if (atomic_dec_and_test(&mm->mm_users)) {
 405                 exit_aio(mm);
 406                 exit_mmap(mm);
 407                 if (!list_empty(&mm->mmlist)) {
 408                         spin_lock(&mmlist_lock);
 409                         list_del(&mm->mmlist);
 410                         spin_unlock(&mmlist_lock);
 411                 }
 412                 put_swap_token(mm);
 413                 mmdrop(mm);
 414         }
 415 }
 416 EXPORT_SYMBOL_GPL(mmput);
 417
 418 /**
 419  * get_task_mm - acquire a reference to the task's mm
 420  *
 421  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
 422  * this kernel workthread has transiently adopted a user mm with use_mm,
 423  * to do its AIO) is not set and if so returns a reference to it, after
 424  * bumping up the use count.  User must release the mm via mmput()
 425  * after use.  Typically used by /proc and ptrace.
 426  */
 427 struct mm_struct *get_task_mm(struct task_struct *task)
 428 {
 429         struct mm_struct *mm;
 430
 431         task_lock(task);
 432         mm = task->mm;
 433         if (mm) {
 434                 if (task->flags & PF_BORROWED_MM)
 435                         mm = NULL;
 436                 else
 437                         atomic_inc(&mm->mm_users);
 438         }
 439         task_unlock(task);
 440         return mm;
 441 }
 442 EXPORT_SYMBOL_GPL(get_task_mm);
 443
 444 /* Please note the differences between mmput and mm_release.
 445  * mmput is called whenever we stop holding onto a mm_struct,
 446  * error success whatever.
 447  *
 448  * mm_release is called after a mm_struct has been removed
 449  * from the current process.
 450  *
 451  * This difference is important for error handling, when we
 452  * only half set up a mm_struct for a new process and need to restore
 453  * the old one.  Because we mmput the new mm_struct before
 454  * restoring the old one. . .
 455  * Eric Biederman 10 January 1998
 456  */
 457 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 458 {
 459         struct completion *vfork_done = tsk->vfork_done;
 460
 461         /* Get rid of any cached register state */
 462         deactivate_mm(tsk, mm);
 463
 464         /* notify parent sleeping on vfork() */
 465         if (vfork_done) {
 466                 tsk->vfork_done = NULL;
 467                 complete(vfork_done);
 468         }
 469
 470         /*
 471          * If we're exiting normally, clear a user-space tid field if
 472          * requested.  We leave this alone when dying by signal, to leave
 473          * the value intact in a core dump, and to save the unnecessary
 474          * trouble otherwise.  Userland only wants this done for a sys_exit.
 475          */
 476         if (tsk->clear_child_tid
 477             && !(tsk->flags & PF_SIGNALED)
 478             && atomic_read(&mm->mm_users) > 1) {
 479                 u32 __user * tidptr = tsk->clear_child_tid;
 480                 tsk->clear_child_tid = NULL;
 481
 482                 /*
 483                  * We don't check the error code - if userspace has
 484                  * not set up a proper pointer then tough luck.
 485                  */
 486                 put_user(0, tidptr);
 487                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 488         }
 489 }
 490
 491 /*
 492  * Allocate a new mm structure and copy contents from the
 493  * mm structure of the passed in task structure.
 494  */
 495 static struct mm_struct *dup_mm(struct task_struct *tsk)
 496 {
 497         struct mm_struct *mm, *oldmm = current->mm;
 498         int err;
 499
 500         if (!oldmm)
 501                 return NULL;
 502
 503         mm = allocate_mm();
 504         if (!mm)
 505                 goto fail_nomem;
 506
 507         memcpy(mm, oldmm, sizeof(*mm));
 508
 509         /* Initializing for Swap token stuff */
 510         mm->token_priority = 0;
 511         mm->last_interval = 0;
 512
 513         if (!mm_init(mm))
 514                 goto fail_nomem;
 515
 516         if (init_new_context(tsk, mm))
 517                 goto fail_nocontext;
 518
 519         err = dup_mmap(mm, oldmm);
 520         if (err)
 521                 goto free_pt;
 522
 523         mm->hiwater_rss = get_mm_rss(mm);
 524         mm->hiwater_vm = mm->total_vm;
 525
 526         return mm;
 527
 528 free_pt:
 529         mmput(mm);
 530
 531 fail_nomem:
 532         return NULL;
 533
 534 fail_nocontext:
 535         /*
 536          * If init_new_context() failed, we cannot use mmput() to free the mm
 537          * because it calls destroy_context()
 538          */
 539         mm_free_pgd(mm);
 540         free_mm(mm);
 541         return NULL;
 542 }
 543
 544 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 545 {
 546         struct mm_struct * mm, *oldmm;
 547         int retval;
 548
 549         tsk->min_flt = tsk->maj_flt = 0;
 550         tsk->nvcsw = tsk->nivcsw = 0;
 551
 552         tsk->mm = NULL;
 553         tsk->active_mm = NULL;
 554
 555         /*
 556          * Are we cloning a kernel thread?
 557          *
 558          * We need to steal a active VM for that..
 559          */
 560         oldmm = current->mm;
 561         if (!oldmm)
 562                 return 0;
 563
 564         if (clone_flags & CLONE_VM) {
 565                 atomic_inc(&oldmm->mm_users);
 566                 mm = oldmm;
 567                 goto good_mm;
 568         }
 569
 570         retval = -ENOMEM;
 571         mm = dup_mm(tsk);
 572         if (!mm)
 573                 goto fail_nomem;
 574
 575 good_mm:
 576         /* Initializing for Swap token stuff */
 577         mm->token_priority = 0;
 578         mm->last_interval = 0;
 579
 580         tsk->mm = mm;
 581         tsk->active_mm = mm;
 582         return 0;
 583
 584 fail_nomem:
 585         return retval;
 586 }
 587
 588 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 589 {
 590         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 591         /* We don't need to lock fs - think why ;-) */
 592         if (fs) {
 593                 atomic_set(&fs->count, 1);
 594                 rwlock_init(&fs->lock);
 595                 fs->umask = old->umask;
 596                 read_lock(&old->lock);
 597                 fs->rootmnt = mntget(old->rootmnt);
 598                 fs->root = dget(old->root);
 599                 fs->pwdmnt = mntget(old->pwdmnt);
 600                 fs->pwd = dget(old->pwd);
 601                 if (old->altroot) {
 602                         fs->altrootmnt = mntget(old->altrootmnt);
 603                         fs->altroot = dget(old->altroot);
 604                 } else {
 605                         fs->altrootmnt = NULL;
 606                         fs->altroot = NULL;
 607                 }
 608                 read_unlock(&old->lock);
 609         }
 610         return fs;
 611 }
 612
 613 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 614 {
 615         return __copy_fs_struct(old);
 616 }
 617
 618 EXPORT_SYMBOL_GPL(copy_fs_struct);
 619
 620 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 621 {
 622         if (clone_flags & CLONE_FS) {
 623                 atomic_inc(&current->fs->count);
 624                 return 0;
 625         }
 626         tsk->fs = __copy_fs_struct(current->fs);
 627         if (!tsk->fs)
 628                 return -ENOMEM;
 629         return 0;
 630 }
 631
 632 static int count_open_files(struct fdtable *fdt)
 633 {
 634         int size = fdt->max_fds;
 635         int i;
 636
 637         /* Find the last open fd */
 638         for (i = size/(8*sizeof(long)); i > 0; ) {
 639                 if (fdt->open_fds->fds_bits[--i])
 640                         break;
 641         }
 642         i = (i+1) * 8 * sizeof(long);
 643         return i;
 644 }
 645
 646 static struct files_struct *alloc_files(void)
 647 {
 648         struct files_struct *newf;
 649         struct fdtable *fdt;
 650
 651         newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
 652         if (!newf)
 653                 goto out;
 654
 655         atomic_set(&newf->count, 1);
 656
 657         spin_lock_init(&newf->file_lock);
 658         newf->next_fd = 0;
 659         fdt = &newf->fdtab;
 660         fdt->max_fds = NR_OPEN_DEFAULT;
 661         fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
 662         fdt->open_fds = (fd_set *)&newf->open_fds_init;
 663         fdt->fd = &newf->fd_array[0];
 664         INIT_RCU_HEAD(&fdt->rcu);
 665         fdt->next = NULL;
 666         rcu_assign_pointer(newf->fdt, fdt);
 667 out:
 668         return newf;
 669 }
 670
 671 /*
 672  * Allocate a new files structure and copy contents from the
 673  * passed in files structure.
 674  * errorp will be valid only when the returned files_struct is NULL.
 675  */
 676 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
 677 {
 678         struct files_struct *newf;
 679         struct file **old_fds, **new_fds;
 680         int open_files, size, i;
 681         struct fdtable *old_fdt, *new_fdt;
 682
 683         *errorp = -ENOMEM;
 684         newf = alloc_files();
 685         if (!newf)
 686                 goto out;
 687
 688         spin_lock(&oldf->file_lock);
 689         old_fdt = files_fdtable(oldf);
 690         new_fdt = files_fdtable(newf);
 691         open_files = count_open_files(old_fdt);
 692
 693         /*
 694          * Check whether we need to allocate a larger fd array and fd set.
 695          * Note: we're not a clone task, so the open count won't change.
 696          */
 697         if (open_files > new_fdt->max_fds) {
 698                 new_fdt->max_fds = 0;
 699                 spin_unlock(&oldf->file_lock);
 700                 spin_lock(&newf->file_lock);
 701                 *errorp = expand_files(newf, open_files-1);
 702                 spin_unlock(&newf->file_lock);
 703                 if (*errorp < 0)
 704                         goto out_release;
 705                 new_fdt = files_fdtable(newf);
 706                 /*
 707                  * Reacquire the oldf lock and a pointer to its fd table
 708                  * who knows it may have a new bigger fd table. We need
 709                  * the latest pointer.
 710                  */
 711                 spin_lock(&oldf->file_lock);
 712                 old_fdt = files_fdtable(oldf);
 713         }
 714
 715         old_fds = old_fdt->fd;
 716         new_fds = new_fdt->fd;
 717
 718         memcpy(new_fdt->open_fds->fds_bits,
 719                 old_fdt->open_fds->fds_bits, open_files/8);
 720         memcpy(new_fdt->close_on_exec->fds_bits,
 721                 old_fdt->close_on_exec->fds_bits, open_files/8);
 722
 723         for (i = open_files; i != 0; i--) {
 724                 struct file *f = *old_fds++;
 725                 if (f) {
 726                         get_file(f);
 727                 } else {
 728                         /*
 729                          * The fd may be claimed in the fd bitmap but not yet
 730                          * instantiated in the files array if a sibling thread
 731                          * is partway through open().  So make sure that this
 732                          * fd is available to the new process.
 733                          */
 734                         FD_CLR(open_files - i, new_fdt->open_fds);
 735                 }
 736                 rcu_assign_pointer(*new_fds++, f);
 737         }
 738         spin_unlock(&oldf->file_lock);
 739
 740         /* compute the remainder to be cleared */
 741         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
 742
 743         /* This is long word aligned thus could use a optimized version */
 744         memset(new_fds, 0, size);
 745
 746         if (new_fdt->max_fds > open_files) {
 747                 int left = (new_fdt->max_fds-open_files)/8;
 748                 int start = open_files / (8 * sizeof(unsigned long));
 749
 750                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
 751                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
 752         }
 753
 754         return newf;
 755
 756 out_release:
 757         kmem_cache_free(files_cachep, newf);
 758 out:
 759         return NULL;
 760 }
 761
 762 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 763 {
 764         struct files_struct *oldf, *newf;
 765         int error = 0;
 766
 767         /*
 768          * A background process may not have any files ...
 769          */
 770         oldf = current->files;
 771         if (!oldf)
 772                 goto out;
 773
 774         if (clone_flags & CLONE_FILES) {
 775                 atomic_inc(&oldf->count);
 776                 goto out;
 777         }
 778
 779         /*
 780          * Note: we may be using current for both targets (See exec.c)
 781          * This works because we cache current->files (old) as oldf. Don't
 782          * break this.
 783          */
 784         tsk->files = NULL;
 785         newf = dup_fd(oldf, &error);
 786         if (!newf)
 787                 goto out;
 788
 789         tsk->files = newf;
 790         error = 0;
 791 out:
 792         return error;
 793 }
 794
 795 static int copy_io(struct task_struct *tsk)
 796 {
 797 #ifdef CONFIG_BLOCK
 798         struct io_context *ioc = current->io_context;
 799
 800         if (!ioc)
 801                 return 0;
 802
 803         if (ioprio_valid(ioc->ioprio)) {
 804                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 805                 if (unlikely(!tsk->io_context))
 806                         return -ENOMEM;
 807
 808                 tsk->io_context->ioprio = ioc->ioprio;
 809         }
 810 #endif
 811         return 0;
 812 }
 813
 814 /*
 815  *      Helper to unshare the files of the current task.
 816  *      We don't want to expose copy_files internals to
 817  *      the exec layer of the kernel.
 818  */
 819
 820 int unshare_files(void)
 821 {
 822         struct files_struct *files  = current->files;
 823         int rc;
 824
 825         BUG_ON(!files);
 826
 827         /* This can race but the race causes us to copy when we don't
 828            need to and drop the copy */
 829         if(atomic_read(&files->count) == 1)
 830         {
 831                 atomic_inc(&files->count);
 832                 return 0;
 833         }
 834         rc = copy_files(0, current);
 835         if(rc)
 836                 current->files = files;
 837         return rc;
 838 }
 839
 840 EXPORT_SYMBOL(unshare_files);
 841
 842 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 843 {
 844         struct sighand_struct *sig;
 845
 846         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 847                 atomic_inc(&current->sighand->count);
 848                 return 0;
 849         }
 850         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 851         rcu_assign_pointer(tsk->sighand, sig);
 852         if (!sig)
 853                 return -ENOMEM;
 854         atomic_set(&sig->count, 1);
 855         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 856         return 0;
 857 }
 858
 859 void __cleanup_sighand(struct sighand_struct *sighand)
 860 {
 861         if (atomic_dec_and_test(&sighand->count))
 862                 kmem_cache_free(sighand_cachep, sighand);
 863 }
 864
 865 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 866 {
 867         struct signal_struct *sig;
 868         int ret;
 869
 870         if (clone_flags & CLONE_THREAD) {
 871                 atomic_inc(&current->signal->count);
 872                 atomic_inc(&current->signal->live);
 873                 return 0;
 874         }
 875         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 876         tsk->signal = sig;
 877         if (!sig)
 878                 return -ENOMEM;
 879
 880         ret = copy_thread_group_keys(tsk);
 881         if (ret < 0) {
 882                 kmem_cache_free(signal_cachep, sig);
 883                 return ret;
 884         }
 885
 886         atomic_set(&sig->count, 1);
 887         atomic_set(&sig->live, 1);
 888         init_waitqueue_head(&sig->wait_chldexit);
 889         sig->flags = 0;
 890         sig->group_exit_code = 0;
 891         sig->group_exit_task = NULL;
 892         sig->group_stop_count = 0;
 893         sig->curr_target = NULL;
 894         init_sigpending(&sig->shared_pending);
 895         INIT_LIST_HEAD(&sig->posix_timers);
 896
 897         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 898         sig->it_real_incr.tv64 = 0;
 899         sig->real_timer.function = it_real_fn;
 900         sig->tsk = tsk;
 901
 902         sig->it_virt_expires = cputime_zero;
 903         sig->it_virt_incr = cputime_zero;
 904         sig->it_prof_expires = cputime_zero;
 905         sig->it_prof_incr = cputime_zero;
 906
 907         sig->leader = 0;        /* session leadership doesn't inherit */
 908         sig->tty_old_pgrp = NULL;
 909
 910         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 911         sig->gtime = cputime_zero;
 912         sig->cgtime = cputime_zero;
 913         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 914         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 915         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
 916         sig->sum_sched_runtime = 0;
 917         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 918         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 919         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 920         taskstats_tgid_init(sig);
 921
 922         task_lock(current->group_leader);
 923         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 924         task_unlock(current->group_leader);
 925
 926         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 927                 /*
 928                  * New sole thread in the process gets an expiry time
 929                  * of the whole CPU time limit.
 930                  */
 931                 tsk->it_prof_expires =
 932                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 933         }
 934         acct_init_pacct(&sig->pacct);
 935
 936         tty_audit_fork(sig);
 937
 938         return 0;
 939 }
 940
 941 void __cleanup_signal(struct signal_struct *sig)
 942 {
 943         exit_thread_group_keys(sig);
 944         kmem_cache_free(signal_cachep, sig);
 945 }
 946
 947 static void cleanup_signal(struct task_struct *tsk)
 948 {
 949         struct signal_struct *sig = tsk->signal;
 950
 951         atomic_dec(&sig->live);
 952
 953         if (atomic_dec_and_test(&sig->count))
 954                 __cleanup_signal(sig);
 955 }
 956
 957 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 958 {
 959         unsigned long new_flags = p->flags;
 960
 961         new_flags &= ~PF_SUPERPRIV;
 962         new_flags |= PF_FORKNOEXEC;
 963         if (!(clone_flags & CLONE_PTRACE))
 964                 p->ptrace = 0;
 965         p->flags = new_flags;
 966         clear_freeze_flag(p);
 967 }
 968
 969 asmlinkage long sys_set_tid_address(int __user *tidptr)
 970 {
 971         current->clear_child_tid = tidptr;
 972
 973         return task_pid_vnr(current);
 974 }
 975
 976 static void rt_mutex_init_task(struct task_struct *p)
 977 {
 978         spin_lock_init(&p->pi_lock);
 979 #ifdef CONFIG_RT_MUTEXES
 980         plist_head_init(&p->pi_waiters, &p->pi_lock);
 981         p->pi_blocked_on = NULL;
 982 #endif
 983 }
 984
 985 /*
 986  * This creates a new process as a copy of the old one,
 987  * but does not actually start it yet.
 988  *
 989  * It copies the registers, and all the appropriate
 990  * parts of the process environment (as per the clone
 991  * flags). The actual kick-off is left to the caller.
 992  */
 993 static struct task_struct *copy_process(unsigned long clone_flags,
 994                                         unsigned long stack_start,
 995                                         struct pt_regs *regs,
 996                                         unsigned long stack_size,
 997                                         int __user *child_tidptr,
 998                                         struct pid *pid)
 999 {
1000         int retval;
1001         struct task_struct *p;
1002         int cgroup_callbacks_done = 0;
1003
1004         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1005                 return ERR_PTR(-EINVAL);
1006
1007         /*
1008          * Thread groups must share signals as well, and detached threads
1009          * can only be started up within the thread group.
1010          */
1011         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1012                 return ERR_PTR(-EINVAL);
1013
1014         /*
1015          * Shared signal handlers imply shared VM. By way of the above,
1016          * thread groups also imply shared VM. Blocking this case allows
1017          * for various simplifications in other code.
1018          */
1019         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1020                 return ERR_PTR(-EINVAL);
1021
1022         retval = security_task_create(clone_flags);
1023         if (retval)
1024                 goto fork_out;
1025
1026         retval = -ENOMEM;
1027         p = dup_task_struct(current);
1028         if (!p)
1029                 goto fork_out;
1030
1031         rt_mutex_init_task(p);
1032
1033 #ifdef CONFIG_TRACE_IRQFLAGS
1034         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1035         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1036 #endif
1037         retval = -EAGAIN;
1038         if (atomic_read(&p->user->processes) >=
1039                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1040                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1041                     p->user != current->nsproxy->user_ns->root_user)
1042                         goto bad_fork_free;
1043         }
1044
1045         atomic_inc(&p->user->__count);
1046         atomic_inc(&p->user->processes);
1047         get_group_info(p->group_info);
1048
1049         /*
1050          * If multiple threads are within copy_process(), then this check
1051          * triggers too late. This doesn't hurt, the check is only there
1052          * to stop root fork bombs.
1053          */
1054         if (nr_threads >= max_threads)
1055                 goto bad_fork_cleanup_count;
1056
1057         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1058                 goto bad_fork_cleanup_count;
1059
1060         if (p->binfmt && !try_module_get(p->binfmt->module))
1061                 goto bad_fork_cleanup_put_domain;
1062
1063         p->did_exec = 0;
1064         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1065         copy_flags(clone_flags, p);
1066         INIT_LIST_HEAD(&p->children);
1067         INIT_LIST_HEAD(&p->sibling);
1068 #ifdef CONFIG_PREEMPT_RCU
1069         p->rcu_read_lock_nesting = 0;
1070         p->rcu_flipctr_idx = 0;
1071 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1072         p->vfork_done = NULL;
1073         spin_lock_init(&p->alloc_lock);
1074
1075         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1076         init_sigpending(&p->pending);
1077
1078         p->utime = cputime_zero;
1079         p->stime = cputime_zero;
1080         p->gtime = cputime_zero;
1081         p->utimescaled = cputime_zero;
1082         p->stimescaled = cputime_zero;
1083         p->prev_utime = cputime_zero;
1084         p->prev_stime = cputime_zero;
1085
1086 #ifdef CONFIG_DETECT_SOFTLOCKUP
1087         p->last_switch_count = 0;
1088         p->last_switch_timestamp = 0;
1089 #endif
1090
1091 #ifdef CONFIG_TASK_XACCT
1092         p->rchar = 0;           /* I/O counter: bytes read */
1093         p->wchar = 0;           /* I/O counter: bytes written */
1094         p->syscr = 0;           /* I/O counter: read syscalls */
1095         p->syscw = 0;           /* I/O counter: write syscalls */
1096 #endif
1097         task_io_accounting_init(p);
1098         acct_clear_integrals(p);
1099
1100         p->it_virt_expires = cputime_zero;
1101         p->it_prof_expires = cputime_zero;
1102         p->it_sched_expires = 0;
1103         INIT_LIST_HEAD(&p->cpu_timers[0]);
1104         INIT_LIST_HEAD(&p->cpu_timers[1]);
1105         INIT_LIST_HEAD(&p->cpu_timers[2]);
1106
1107         p->lock_depth = -1;             /* -1 = no lock */
1108         do_posix_clock_monotonic_gettime(&p->start_time);
1109         p->real_start_time = p->start_time;
1110         monotonic_to_bootbased(&p->real_start_time);
1111 #ifdef CONFIG_SECURITY
1112         p->security = NULL;
1113 #endif
1114         p->io_context = NULL;
1115         p->audit_context = NULL;
1116         cgroup_fork(p);
1117 #ifdef CONFIG_NUMA
1118         p->mempolicy = mpol_copy(p->mempolicy);
1119         if (IS_ERR(p->mempolicy)) {
1120                 retval = PTR_ERR(p->mempolicy);
1121                 p->mempolicy = NULL;
1122                 goto bad_fork_cleanup_cgroup;
1123         }
1124         mpol_fix_fork_child_flag(p);
1125 #endif
1126 #ifdef CONFIG_TRACE_IRQFLAGS
1127         p->irq_events = 0;
1128 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1129         p->hardirqs_enabled = 1;
1130 #else
1131         p->hardirqs_enabled = 0;
1132 #endif
1133         p->hardirq_enable_ip = 0;
1134         p->hardirq_enable_event = 0;
1135         p->hardirq_disable_ip = _THIS_IP_;
1136         p->hardirq_disable_event = 0;
1137         p->softirqs_enabled = 1;
1138         p->softirq_enable_ip = _THIS_IP_;
1139         p->softirq_enable_event = 0;
1140         p->softirq_disable_ip = 0;
1141         p->softirq_disable_event = 0;
1142         p->hardirq_context = 0;
1143         p->softirq_context = 0;
1144 #endif
1145 #ifdef CONFIG_LOCKDEP
1146         p->lockdep_depth = 0; /* no locks held yet */
1147         p->curr_chain_key = 0;
1148         p->lockdep_recursion = 0;
1149 #endif
1150
1151 #ifdef CONFIG_DEBUG_MUTEXES
1152         p->blocked_on = NULL; /* not blocked yet */
1153 #endif
1154
1155         /* Perform scheduler related setup. Assign this task to a CPU. */
1156         sched_fork(p, clone_flags);
1157
1158         if ((retval = security_task_alloc(p)))
1159                 goto bad_fork_cleanup_policy;
1160         if ((retval = audit_alloc(p)))
1161                 goto bad_fork_cleanup_security;
1162         /* copy all the process information */
1163         if ((retval = copy_semundo(clone_flags, p)))
1164                 goto bad_fork_cleanup_audit;
1165         if ((retval = copy_files(clone_flags, p)))
1166                 goto bad_fork_cleanup_semundo;
1167         if ((retval = copy_fs(clone_flags, p)))
1168                 goto bad_fork_cleanup_files;
1169         if ((retval = copy_sighand(clone_flags, p)))
1170                 goto bad_fork_cleanup_fs;
1171         if ((retval = copy_signal(clone_flags, p)))
1172                 goto bad_fork_cleanup_sighand;
1173         if ((retval = copy_mm(clone_flags, p)))
1174                 goto bad_fork_cleanup_signal;
1175         if ((retval = copy_keys(clone_flags, p)))
1176                 goto bad_fork_cleanup_mm;
1177         if ((retval = copy_namespaces(clone_flags, p)))
1178                 goto bad_fork_cleanup_keys;
1179         if ((retval = copy_io(p)))
1180                 goto bad_fork_cleanup_namespaces;
1181         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1182         if (retval)
1183                 goto bad_fork_cleanup_io;
1184
1185         if (pid != &init_struct_pid) {
1186                 retval = -ENOMEM;
1187                 pid = alloc_pid(task_active_pid_ns(p));
1188                 if (!pid)
1189                         goto bad_fork_cleanup_io;
1190
1191                 if (clone_flags & CLONE_NEWPID) {
1192                         retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1193                         if (retval < 0)
1194                                 goto bad_fork_free_pid;
1195                 }
1196         }
1197
1198         p->pid = pid_nr(pid);
1199         p->tgid = p->pid;
1200         if (clone_flags & CLONE_THREAD)
1201                 p->tgid = current->tgid;
1202
1203         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1204         /*
1205          * Clear TID on mm_release()?
1206          */
1207         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1208 #ifdef CONFIG_FUTEX
1209         p->robust_list = NULL;
1210 #ifdef CONFIG_COMPAT
1211         p->compat_robust_list = NULL;
1212 #endif
1213         INIT_LIST_HEAD(&p->pi_state_list);
1214         p->pi_state_cache = NULL;
1215 #endif
1216         /*
1217          * sigaltstack should be cleared when sharing the same VM
1218          */
1219         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1220                 p->sas_ss_sp = p->sas_ss_size = 0;
1221
1222         /*
1223          * Syscall tracing should be turned off in the child regardless
1224          * of CLONE_PTRACE.
1225          */
1226         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1227 #ifdef TIF_SYSCALL_EMU
1228         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1229 #endif
1230         clear_all_latency_tracing(p);
1231
1232         /* Our parent execution domain becomes current domain
1233            These must match for thread signalling to apply */
1234         p->parent_exec_id = p->self_exec_id;
1235
1236         /* ok, now we should be set up.. */
1237         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1238         p->pdeath_signal = 0;
1239         p->exit_state = 0;
1240
1241         /*
1242          * Ok, make it visible to the rest of the system.
1243          * We dont wake it up yet.
1244          */
1245         p->group_leader = p;
1246         INIT_LIST_HEAD(&p->thread_group);
1247         INIT_LIST_HEAD(&p->ptrace_children);
1248         INIT_LIST_HEAD(&p->ptrace_list);
1249
1250         /* Now that the task is set up, run cgroup callbacks if
1251          * necessary. We need to run them before the task is visible
1252          * on the tasklist. */
1253         cgroup_fork_callbacks(p);
1254         cgroup_callbacks_done = 1;
1255
1256         /* Need tasklist lock for parent etc handling! */
1257         write_lock_irq(&tasklist_lock);
1258
1259         /*
1260          * The task hasn't been attached yet, so its cpus_allowed mask will
1261          * not be changed, nor will its assigned CPU.
1262          *
1263          * The cpus_allowed mask of the parent may have changed after it was
1264          * copied first time - so re-copy it here, then check the child's CPU
1265          * to ensure it is on a valid CPU (and if not, just force it back to
1266          * parent's CPU). This avoids alot of nasty races.
1267          */
1268         p->cpus_allowed = current->cpus_allowed;
1269         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1270         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1271                         !cpu_online(task_cpu(p))))
1272                 set_task_cpu(p, smp_processor_id());
1273
1274         /* CLONE_PARENT re-uses the old parent */
1275         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1276                 p->real_parent = current->real_parent;
1277         else
1278                 p->real_parent = current;
1279         p->parent = p->real_parent;
1280
1281         spin_lock(&current->sighand->siglock);
1282
1283         /*
1284          * Process group and session signals need to be delivered to just the
1285          * parent before the fork or both the parent and the child after the
1286          * fork. Restart if a signal comes in before we add the new process to
1287          * it's process group.
1288          * A fatal signal pending means that current will exit, so the new
1289          * thread can't slip out of an OOM kill (or normal SIGKILL).
1290          */
1291         recalc_sigpending();
1292         if (signal_pending(current)) {
1293                 spin_unlock(&current->sighand->siglock);
1294                 write_unlock_irq(&tasklist_lock);
1295                 retval = -ERESTARTNOINTR;
1296                 goto bad_fork_free_pid;
1297         }
1298
1299         if (clone_flags & CLONE_THREAD) {
1300                 p->group_leader = current->group_leader;
1301                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1302
1303                 if (!cputime_eq(current->signal->it_virt_expires,
1304                                 cputime_zero) ||
1305                     !cputime_eq(current->signal->it_prof_expires,
1306                                 cputime_zero) ||
1307                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1308                     !list_empty(&current->signal->cpu_timers[0]) ||
1309                     !list_empty(&current->signal->cpu_timers[1]) ||
1310                     !list_empty(&current->signal->cpu_timers[2])) {
1311                         /*
1312                          * Have child wake up on its first tick to check
1313                          * for process CPU timers.
1314                          */
1315                         p->it_prof_expires = jiffies_to_cputime(1);
1316                 }
1317         }
1318
1319         if (likely(p->pid)) {
1320                 add_parent(p);
1321                 if (unlikely(p->ptrace & PT_PTRACED))
1322                         __ptrace_link(p, current->parent);
1323
1324                 if (thread_group_leader(p)) {
1325                         if (clone_flags & CLONE_NEWPID)
1326                                 p->nsproxy->pid_ns->child_reaper = p;
1327
1328                         p->signal->tty = current->signal->tty;
1329                         set_task_pgrp(p, task_pgrp_nr(current));
1330                         set_task_session(p, task_session_nr(current));
1331                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1332                         attach_pid(p, PIDTYPE_SID, task_session(current));
1333                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1334                         __get_cpu_var(process_counts)++;
1335                 }
1336                 attach_pid(p, PIDTYPE_PID, pid);
1337                 nr_threads++;
1338         }
1339
1340         total_forks++;
1341         spin_unlock(&current->sighand->siglock);
1342         write_unlock_irq(&tasklist_lock);
1343         proc_fork_connector(p);
1344         cgroup_post_fork(p);
1345         return p;
1346
1347 bad_fork_free_pid:
1348         if (pid != &init_struct_pid)
1349                 free_pid(pid);
1350 bad_fork_cleanup_io:
1351         put_io_context(p->io_context);
1352 bad_fork_cleanup_namespaces:
1353         exit_task_namespaces(p);
1354 bad_fork_cleanup_keys:
1355         exit_keys(p);
1356 bad_fork_cleanup_mm:
1357         if (p->mm)
1358                 mmput(p->mm);
1359 bad_fork_cleanup_signal:
1360         cleanup_signal(p);
1361 bad_fork_cleanup_sighand:
1362         __cleanup_sighand(p->sighand);
1363 bad_fork_cleanup_fs:
1364         exit_fs(p); /* blocking */
1365 bad_fork_cleanup_files:
1366         exit_files(p); /* blocking */
1367 bad_fork_cleanup_semundo:
1368         exit_sem(p);
1369 bad_fork_cleanup_audit:
1370         audit_free(p);
1371 bad_fork_cleanup_security:
1372         security_task_free(p);
1373 bad_fork_cleanup_policy:
1374 #ifdef CONFIG_NUMA
1375         mpol_free(p->mempolicy);
1376 bad_fork_cleanup_cgroup:
1377 #endif
1378         cgroup_exit(p, cgroup_callbacks_done);
1379         delayacct_tsk_free(p);
1380         if (p->binfmt)
1381                 module_put(p->binfmt->module);
1382 bad_fork_cleanup_put_domain:
1383         module_put(task_thread_info(p)->exec_domain->module);
1384 bad_fork_cleanup_count:
1385         put_group_info(p->group_info);
1386         atomic_dec(&p->user->processes);
1387         free_uid(p->user);
1388 bad_fork_free:
1389         free_task(p);
1390 fork_out:
1391         return ERR_PTR(retval);
1392 }
1393
1394 noinline struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1395 {
1396         memset(regs, 0, sizeof(struct pt_regs));
1397         return regs;
1398 }
1399
1400 struct task_struct * __cpuinit fork_idle(int cpu)
1401 {
1402         struct task_struct *task;
1403         struct pt_regs regs;
1404
1405         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1406                                 &init_struct_pid);
1407         if (!IS_ERR(task))
1408                 init_idle(task, cpu);
1409
1410         return task;
1411 }
1412
1413 static int fork_traceflag(unsigned clone_flags)
1414 {
1415         if (clone_flags & CLONE_UNTRACED)
1416                 return 0;
1417         else if (clone_flags & CLONE_VFORK) {
1418                 if (current->ptrace & PT_TRACE_VFORK)
1419                         return PTRACE_EVENT_VFORK;
1420         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1421                 if (current->ptrace & PT_TRACE_CLONE)
1422                         return PTRACE_EVENT_CLONE;
1423         } else if (current->ptrace & PT_TRACE_FORK)
1424                 return PTRACE_EVENT_FORK;
1425
1426         return 0;
1427 }
1428
1429 /*
1430  *  Ok, this is the main fork-routine.
1431  *
1432  * It copies the process, and if successful kick-starts
1433  * it and waits for it to finish using the VM if required.
1434  */
1435 long do_fork(unsigned long clone_flags,
1436               unsigned long stack_start,
1437               struct pt_regs *regs,
1438               unsigned long stack_size,
1439               int __user *parent_tidptr,
1440               int __user *child_tidptr)
1441 {
1442         struct task_struct *p;
1443         int trace = 0;
1444         long nr;
1445
1446         if (unlikely(current->ptrace)) {
1447                 trace = fork_traceflag (clone_flags);
1448                 if (trace)
1449                         clone_flags |= CLONE_PTRACE;
1450         }
1451
1452         p = copy_process(clone_flags, stack_start, regs, stack_size,
1453                         child_tidptr, NULL);
1454         /*
1455          * Do this prior waking up the new thread - the thread pointer
1456          * might get invalid after that point, if the thread exits quickly.
1457          */
1458         if (!IS_ERR(p)) {
1459                 struct completion vfork;
1460
1461                 /*
1462                  * this is enough to call pid_nr_ns here, but this if
1463                  * improves optimisation of regular fork()
1464                  */
1465                 nr = (clone_flags & CLONE_NEWPID) ?
1466                         task_pid_nr_ns(p, current->nsproxy->pid_ns) :
1467                                 task_pid_vnr(p);
1468
1469                 if (clone_flags & CLONE_PARENT_SETTID)
1470                         put_user(nr, parent_tidptr);
1471
1472                 if (clone_flags & CLONE_VFORK) {
1473                         p->vfork_done = &vfork;
1474                         init_completion(&vfork);
1475                 }
1476
1477                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1478                         /*
1479                          * We'll start up with an immediate SIGSTOP.
1480                          */
1481                         sigaddset(&p->pending.signal, SIGSTOP);
1482                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1483                 }
1484
1485                 if (!(clone_flags & CLONE_STOPPED))
1486                         wake_up_new_task(p, clone_flags);
1487                 else
1488                         p->state = TASK_STOPPED;
1489
1490                 if (unlikely (trace)) {
1491                         current->ptrace_message = nr;
1492                         ptrace_notify ((trace << 8) | SIGTRAP);
1493                 }
1494
1495                 if (clone_flags & CLONE_VFORK) {
1496                         freezer_do_not_count();
1497                         wait_for_completion(&vfork);
1498                         freezer_count();
1499                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1500                                 current->ptrace_message = nr;
1501                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1502                         }
1503                 }
1504         } else {
1505                 nr = PTR_ERR(p);
1506         }
1507         return nr;
1508 }
1509
1510 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1511 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1512 #endif
1513
1514 static void sighand_ctor(struct kmem_cache *cachep, void *data)
1515 {
1516         struct sighand_struct *sighand = data;
1517
1518         spin_lock_init(&sighand->siglock);
1519         init_waitqueue_head(&sighand->signalfd_wqh);
1520 }
1521
1522 void __init proc_caches_init(void)
1523 {
1524         sighand_cachep = kmem_cache_create("sighand_cache",
1525                         sizeof(struct sighand_struct), 0,
1526                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1527                         sighand_ctor);
1528         signal_cachep = kmem_cache_create("signal_cache",
1529                         sizeof(struct signal_struct), 0,
1530                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1531         files_cachep = kmem_cache_create("files_cache",
1532                         sizeof(struct files_struct), 0,
1533                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1534         fs_cachep = kmem_cache_create("fs_cache",
1535                         sizeof(struct fs_struct), 0,
1536                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1537         vm_area_cachep = kmem_cache_create("vm_area_struct",
1538                         sizeof(struct vm_area_struct), 0,
1539                         SLAB_PANIC, NULL);
1540         mm_cachep = kmem_cache_create("mm_struct",
1541                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1542                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1543 }
1544
1545 /*
1546  * Check constraints on flags passed to the unshare system call and
1547  * force unsharing of additional process context as appropriate.
1548  */
1549 static void check_unshare_flags(unsigned long *flags_ptr)
1550 {
1551         /*
1552          * If unsharing a thread from a thread group, must also
1553          * unshare vm.
1554          */
1555         if (*flags_ptr & CLONE_THREAD)
1556                 *flags_ptr |= CLONE_VM;
1557
1558         /*
1559          * If unsharing vm, must also unshare signal handlers.
1560          */
1561         if (*flags_ptr & CLONE_VM)
1562                 *flags_ptr |= CLONE_SIGHAND;
1563
1564         /*
1565          * If unsharing signal handlers and the task was created
1566          * using CLONE_THREAD, then must unshare the thread
1567          */
1568         if ((*flags_ptr & CLONE_SIGHAND) &&
1569             (atomic_read(&current->signal->count) > 1))
1570                 *flags_ptr |= CLONE_THREAD;
1571
1572         /*
1573          * If unsharing namespace, must also unshare filesystem information.
1574          */
1575         if (*flags_ptr & CLONE_NEWNS)
1576                 *flags_ptr |= CLONE_FS;
1577 }
1578
1579 /*
1580  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1581  */
1582 static int unshare_thread(unsigned long unshare_flags)
1583 {
1584         if (unshare_flags & CLONE_THREAD)
1585                 return -EINVAL;
1586
1587         return 0;
1588 }
1589
1590 /*
1591  * Unshare the filesystem structure if it is being shared
1592  */
1593 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1594 {
1595         struct fs_struct *fs = current->fs;
1596
1597         if ((unshare_flags & CLONE_FS) &&
1598             (fs && atomic_read(&fs->count) > 1)) {
1599                 *new_fsp = __copy_fs_struct(current->fs);
1600                 if (!*new_fsp)
1601                         return -ENOMEM;
1602         }
1603
1604         return 0;
1605 }
1606
1607 /*
1608  * Unsharing of sighand is not supported yet
1609  */
1610 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1611 {
1612         struct sighand_struct *sigh = current->sighand;
1613
1614         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1615                 return -EINVAL;
1616         else
1617                 return 0;
1618 }
1619
1620 /*
1621  * Unshare vm if it is being shared
1622  */
1623 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1624 {
1625         struct mm_struct *mm = current->mm;
1626
1627         if ((unshare_flags & CLONE_VM) &&
1628             (mm && atomic_read(&mm->mm_users) > 1)) {
1629                 return -EINVAL;
1630         }
1631
1632         return 0;
1633 }
1634
1635 /*
1636  * Unshare file descriptor table if it is being shared
1637  */
1638 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1639 {
1640         struct files_struct *fd = current->files;
1641         int error = 0;
1642
1643         if ((unshare_flags & CLONE_FILES) &&
1644             (fd && atomic_read(&fd->count) > 1)) {
1645                 *new_fdp = dup_fd(fd, &error);
1646                 if (!*new_fdp)
1647                         return error;
1648         }
1649
1650         return 0;
1651 }
1652
1653 /*
1654  * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1655  * supported yet
1656  */
1657 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1658 {
1659         if (unshare_flags & CLONE_SYSVSEM)
1660                 return -EINVAL;
1661
1662         return 0;
1663 }
1664
1665 /*
1666  * unshare allows a process to 'unshare' part of the process
1667  * context which was originally shared using clone.  copy_*
1668  * functions used by do_fork() cannot be used here directly
1669  * because they modify an inactive task_struct that is being
1670  * constructed. Here we are modifying the current, active,
1671  * task_struct.
1672  */
1673 asmlinkage long sys_unshare(unsigned long unshare_flags)
1674 {
1675         int err = 0;
1676         struct fs_struct *fs, *new_fs = NULL;
1677         struct sighand_struct *new_sigh = NULL;
1678         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1679         struct files_struct *fd, *new_fd = NULL;
1680         struct sem_undo_list *new_ulist = NULL;
1681         struct nsproxy *new_nsproxy = NULL;
1682
1683         check_unshare_flags(&unshare_flags);
1684
1685         /* Return -EINVAL for all unsupported flags */
1686         err = -EINVAL;
1687         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1688                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1689                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
1690                                 CLONE_NEWNET))
1691                 goto bad_unshare_out;
1692
1693         if ((err = unshare_thread(unshare_flags)))
1694                 goto bad_unshare_out;
1695         if ((err = unshare_fs(unshare_flags, &new_fs)))
1696                 goto bad_unshare_cleanup_thread;
1697         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1698                 goto bad_unshare_cleanup_fs;
1699         if ((err = unshare_vm(unshare_flags, &new_mm)))
1700                 goto bad_unshare_cleanup_sigh;
1701         if ((err = unshare_fd(unshare_flags, &new_fd)))
1702                 goto bad_unshare_cleanup_vm;
1703         if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1704                 goto bad_unshare_cleanup_fd;
1705         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1706                         new_fs)))
1707                 goto bad_unshare_cleanup_semundo;
1708
1709         if (new_fs ||  new_mm || new_fd || new_ulist || new_nsproxy) {
1710
1711                 if (new_nsproxy) {
1712                         switch_task_namespaces(current, new_nsproxy);
1713                         new_nsproxy = NULL;
1714                 }
1715
1716                 task_lock(current);
1717
1718                 if (new_fs) {
1719                         fs = current->fs;
1720                         current->fs = new_fs;
1721                         new_fs = fs;
1722                 }
1723
1724                 if (new_mm) {
1725                         mm = current->mm;
1726                         active_mm = current->active_mm;
1727                         current->mm = new_mm;
1728                         current->active_mm = new_mm;
1729                         activate_mm(active_mm, new_mm);
1730                         new_mm = mm;
1731                 }
1732
1733                 if (new_fd) {
1734                         fd = current->files;
1735                         current->files = new_fd;
1736                         new_fd = fd;
1737                 }
1738
1739                 task_unlock(current);
1740         }
1741
1742         if (new_nsproxy)
1743                 put_nsproxy(new_nsproxy);
1744
1745 bad_unshare_cleanup_semundo:
1746 bad_unshare_cleanup_fd:
1747         if (new_fd)
1748                 put_files_struct(new_fd);
1749
1750 bad_unshare_cleanup_vm:
1751         if (new_mm)
1752                 mmput(new_mm);
1753
1754 bad_unshare_cleanup_sigh:
1755         if (new_sigh)
1756                 if (atomic_dec_and_test(&new_sigh->count))
1757                         kmem_cache_free(sighand_cachep, new_sigh);
1758
1759 bad_unshare_cleanup_fs:
1760         if (new_fs)
1761                 put_fs_struct(new_fs);
1762
1763 bad_unshare_cleanup_thread:
1764 bad_unshare_out:
1765         return err;
1766 }