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(unsigned long clone_flags, 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          * Share io context with parent, if CLONE_IO is set
 804          */
 805         if (clone_flags & CLONE_IO) {
 806                 tsk->io_context = ioc_task_link(ioc);
 807                 if (unlikely(!tsk->io_context))
 808                         return -ENOMEM;
 809         } else if (ioprio_valid(ioc->ioprio)) {
 810                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 811                 if (unlikely(!tsk->io_context))
 812                         return -ENOMEM;
 813
 814                 tsk->io_context->ioprio = ioc->ioprio;
 815         }
 816 #endif
 817         return 0;
 818 }
 819
 820 /*
 821  *      Helper to unshare the files of the current task.
 822  *      We don't want to expose copy_files internals to
 823  *      the exec layer of the kernel.
 824  */
 825
 826 int unshare_files(void)
 827 {
 828         struct files_struct *files  = current->files;
 829         int rc;
 830
 831         BUG_ON(!files);
 832
 833         /* This can race but the race causes us to copy when we don't
 834            need to and drop the copy */
 835         if(atomic_read(&files->count) == 1)
 836         {
 837                 atomic_inc(&files->count);
 838                 return 0;
 839         }
 840         rc = copy_files(0, current);
 841         if(rc)
 842                 current->files = files;
 843         return rc;
 844 }
 845
 846 EXPORT_SYMBOL(unshare_files);
 847
 848 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 849 {
 850         struct sighand_struct *sig;
 851
 852         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 853                 atomic_inc(&current->sighand->count);
 854                 return 0;
 855         }
 856         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 857         rcu_assign_pointer(tsk->sighand, sig);
 858         if (!sig)
 859                 return -ENOMEM;
 860         atomic_set(&sig->count, 1);
 861         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 862         return 0;
 863 }
 864
 865 void __cleanup_sighand(struct sighand_struct *sighand)
 866 {
 867         if (atomic_dec_and_test(&sighand->count))
 868                 kmem_cache_free(sighand_cachep, sighand);
 869 }
 870
 871 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 872 {
 873         struct signal_struct *sig;
 874         int ret;
 875
 876         if (clone_flags & CLONE_THREAD) {
 877                 atomic_inc(&current->signal->count);
 878                 atomic_inc(&current->signal->live);
 879                 return 0;
 880         }
 881         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 882         tsk->signal = sig;
 883         if (!sig)
 884                 return -ENOMEM;
 885
 886         ret = copy_thread_group_keys(tsk);
 887         if (ret < 0) {
 888                 kmem_cache_free(signal_cachep, sig);
 889                 return ret;
 890         }
 891
 892         atomic_set(&sig->count, 1);
 893         atomic_set(&sig->live, 1);
 894         init_waitqueue_head(&sig->wait_chldexit);
 895         sig->flags = 0;
 896         sig->group_exit_code = 0;
 897         sig->group_exit_task = NULL;
 898         sig->group_stop_count = 0;
 899         sig->curr_target = NULL;
 900         init_sigpending(&sig->shared_pending);
 901         INIT_LIST_HEAD(&sig->posix_timers);
 902
 903         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 904         sig->it_real_incr.tv64 = 0;
 905         sig->real_timer.function = it_real_fn;
 906         sig->tsk = tsk;
 907
 908         sig->it_virt_expires = cputime_zero;
 909         sig->it_virt_incr = cputime_zero;
 910         sig->it_prof_expires = cputime_zero;
 911         sig->it_prof_incr = cputime_zero;
 912
 913         sig->leader = 0;        /* session leadership doesn't inherit */
 914         sig->tty_old_pgrp = NULL;
 915
 916         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 917         sig->gtime = cputime_zero;
 918         sig->cgtime = cputime_zero;
 919         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 920         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 921         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
 922         sig->sum_sched_runtime = 0;
 923         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 924         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 925         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 926         taskstats_tgid_init(sig);
 927
 928         task_lock(current->group_leader);
 929         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 930         task_unlock(current->group_leader);
 931
 932         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 933                 /*
 934                  * New sole thread in the process gets an expiry time
 935                  * of the whole CPU time limit.
 936                  */
 937                 tsk->it_prof_expires =
 938                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 939         }
 940         acct_init_pacct(&sig->pacct);
 941
 942         tty_audit_fork(sig);
 943
 944         return 0;
 945 }
 946
 947 void __cleanup_signal(struct signal_struct *sig)
 948 {
 949         exit_thread_group_keys(sig);
 950         kmem_cache_free(signal_cachep, sig);
 951 }
 952
 953 static void cleanup_signal(struct task_struct *tsk)
 954 {
 955         struct signal_struct *sig = tsk->signal;
 956
 957         atomic_dec(&sig->live);
 958
 959         if (atomic_dec_and_test(&sig->count))
 960                 __cleanup_signal(sig);
 961 }
 962
 963 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 964 {
 965         unsigned long new_flags = p->flags;
 966
 967         new_flags &= ~PF_SUPERPRIV;
 968         new_flags |= PF_FORKNOEXEC;
 969         if (!(clone_flags & CLONE_PTRACE))
 970                 p->ptrace = 0;
 971         p->flags = new_flags;
 972         clear_freeze_flag(p);
 973 }
 974
 975 asmlinkage long sys_set_tid_address(int __user *tidptr)
 976 {
 977         current->clear_child_tid = tidptr;
 978
 979         return task_pid_vnr(current);
 980 }
 981
 982 static void rt_mutex_init_task(struct task_struct *p)
 983 {
 984         spin_lock_init(&p->pi_lock);
 985 #ifdef CONFIG_RT_MUTEXES
 986         plist_head_init(&p->pi_waiters, &p->pi_lock);
 987         p->pi_blocked_on = NULL;
 988 #endif
 989 }
 990
 991 /*
 992  * This creates a new process as a copy of the old one,
 993  * but does not actually start it yet.
 994  *
 995  * It copies the registers, and all the appropriate
 996  * parts of the process environment (as per the clone
 997  * flags). The actual kick-off is left to the caller.
 998  */
 999 static struct task_struct *copy_process(unsigned long clone_flags,
1000                                         unsigned long stack_start,
1001                                         struct pt_regs *regs,
1002                                         unsigned long stack_size,
1003                                         int __user *child_tidptr,
1004                                         struct pid *pid)
1005 {
1006         int retval;
1007         struct task_struct *p;
1008         int cgroup_callbacks_done = 0;
1009
1010         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1011                 return ERR_PTR(-EINVAL);
1012
1013         /*
1014          * Thread groups must share signals as well, and detached threads
1015          * can only be started up within the thread group.
1016          */
1017         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1018                 return ERR_PTR(-EINVAL);
1019
1020         /*
1021          * Shared signal handlers imply shared VM. By way of the above,
1022          * thread groups also imply shared VM. Blocking this case allows
1023          * for various simplifications in other code.
1024          */
1025         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1026                 return ERR_PTR(-EINVAL);
1027
1028         retval = security_task_create(clone_flags);
1029         if (retval)
1030                 goto fork_out;
1031
1032         retval = -ENOMEM;
1033         p = dup_task_struct(current);
1034         if (!p)
1035                 goto fork_out;
1036
1037         rt_mutex_init_task(p);
1038
1039 #ifdef CONFIG_TRACE_IRQFLAGS
1040         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1041         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1042 #endif
1043         retval = -EAGAIN;
1044         if (atomic_read(&p->user->processes) >=
1045                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1046                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1047                     p->user != current->nsproxy->user_ns->root_user)
1048                         goto bad_fork_free;
1049         }
1050
1051         atomic_inc(&p->user->__count);
1052         atomic_inc(&p->user->processes);
1053         get_group_info(p->group_info);
1054
1055         /*
1056          * If multiple threads are within copy_process(), then this check
1057          * triggers too late. This doesn't hurt, the check is only there
1058          * to stop root fork bombs.
1059          */
1060         if (nr_threads >= max_threads)
1061                 goto bad_fork_cleanup_count;
1062
1063         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1064                 goto bad_fork_cleanup_count;
1065
1066         if (p->binfmt && !try_module_get(p->binfmt->module))
1067                 goto bad_fork_cleanup_put_domain;
1068
1069         p->did_exec = 0;
1070         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1071         copy_flags(clone_flags, p);
1072         INIT_LIST_HEAD(&p->children);
1073         INIT_LIST_HEAD(&p->sibling);
1074 #ifdef CONFIG_PREEMPT_RCU
1075         p->rcu_read_lock_nesting = 0;
1076         p->rcu_flipctr_idx = 0;
1077 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1078         p->vfork_done = NULL;
1079         spin_lock_init(&p->alloc_lock);
1080
1081         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1082         init_sigpending(&p->pending);
1083
1084         p->utime = cputime_zero;
1085         p->stime = cputime_zero;
1086         p->gtime = cputime_zero;
1087         p->utimescaled = cputime_zero;
1088         p->stimescaled = cputime_zero;
1089         p->prev_utime = cputime_zero;
1090         p->prev_stime = cputime_zero;
1091
1092 #ifdef CONFIG_DETECT_SOFTLOCKUP
1093         p->last_switch_count = 0;
1094         p->last_switch_timestamp = 0;
1095 #endif
1096
1097 #ifdef CONFIG_TASK_XACCT
1098         p->rchar = 0;           /* I/O counter: bytes read */
1099         p->wchar = 0;           /* I/O counter: bytes written */
1100         p->syscr = 0;           /* I/O counter: read syscalls */
1101         p->syscw = 0;           /* I/O counter: write syscalls */
1102 #endif
1103         task_io_accounting_init(p);
1104         acct_clear_integrals(p);
1105
1106         p->it_virt_expires = cputime_zero;
1107         p->it_prof_expires = cputime_zero;
1108         p->it_sched_expires = 0;
1109         INIT_LIST_HEAD(&p->cpu_timers[0]);
1110         INIT_LIST_HEAD(&p->cpu_timers[1]);
1111         INIT_LIST_HEAD(&p->cpu_timers[2]);
1112
1113         p->lock_depth = -1;             /* -1 = no lock */
1114         do_posix_clock_monotonic_gettime(&p->start_time);
1115         p->real_start_time = p->start_time;
1116         monotonic_to_bootbased(&p->real_start_time);
1117 #ifdef CONFIG_SECURITY
1118         p->security = NULL;
1119 #endif
1120         p->io_context = NULL;
1121         p->audit_context = NULL;
1122         cgroup_fork(p);
1123 #ifdef CONFIG_NUMA
1124         p->mempolicy = mpol_copy(p->mempolicy);
1125         if (IS_ERR(p->mempolicy)) {
1126                 retval = PTR_ERR(p->mempolicy);
1127                 p->mempolicy = NULL;
1128                 goto bad_fork_cleanup_cgroup;
1129         }
1130         mpol_fix_fork_child_flag(p);
1131 #endif
1132 #ifdef CONFIG_TRACE_IRQFLAGS
1133         p->irq_events = 0;
1134 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1135         p->hardirqs_enabled = 1;
1136 #else
1137         p->hardirqs_enabled = 0;
1138 #endif
1139         p->hardirq_enable_ip = 0;
1140         p->hardirq_enable_event = 0;
1141         p->hardirq_disable_ip = _THIS_IP_;
1142         p->hardirq_disable_event = 0;
1143         p->softirqs_enabled = 1;
1144         p->softirq_enable_ip = _THIS_IP_;
1145         p->softirq_enable_event = 0;
1146         p->softirq_disable_ip = 0;
1147         p->softirq_disable_event = 0;
1148         p->hardirq_context = 0;
1149         p->softirq_context = 0;
1150 #endif
1151 #ifdef CONFIG_LOCKDEP
1152         p->lockdep_depth = 0; /* no locks held yet */
1153         p->curr_chain_key = 0;
1154         p->lockdep_recursion = 0;
1155 #endif
1156
1157 #ifdef CONFIG_DEBUG_MUTEXES
1158         p->blocked_on = NULL; /* not blocked yet */
1159 #endif
1160
1161         /* Perform scheduler related setup. Assign this task to a CPU. */
1162         sched_fork(p, clone_flags);
1163
1164         if ((retval = security_task_alloc(p)))
1165                 goto bad_fork_cleanup_policy;
1166         if ((retval = audit_alloc(p)))
1167                 goto bad_fork_cleanup_security;
1168         /* copy all the process information */
1169         if ((retval = copy_semundo(clone_flags, p)))
1170                 goto bad_fork_cleanup_audit;
1171         if ((retval = copy_files(clone_flags, p)))
1172                 goto bad_fork_cleanup_semundo;
1173         if ((retval = copy_fs(clone_flags, p)))
1174                 goto bad_fork_cleanup_files;
1175         if ((retval = copy_sighand(clone_flags, p)))
1176                 goto bad_fork_cleanup_fs;
1177         if ((retval = copy_signal(clone_flags, p)))
1178                 goto bad_fork_cleanup_sighand;
1179         if ((retval = copy_mm(clone_flags, p)))
1180                 goto bad_fork_cleanup_signal;
1181         if ((retval = copy_keys(clone_flags, p)))
1182                 goto bad_fork_cleanup_mm;
1183         if ((retval = copy_namespaces(clone_flags, p)))
1184                 goto bad_fork_cleanup_keys;
1185         if ((retval = copy_io(clone_flags, p)))
1186                 goto bad_fork_cleanup_namespaces;
1187         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1188         if (retval)
1189                 goto bad_fork_cleanup_io;
1190
1191         if (pid != &init_struct_pid) {
1192                 retval = -ENOMEM;
1193                 pid = alloc_pid(task_active_pid_ns(p));
1194                 if (!pid)
1195                         goto bad_fork_cleanup_io;
1196
1197                 if (clone_flags & CLONE_NEWPID) {
1198                         retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1199                         if (retval < 0)
1200                                 goto bad_fork_free_pid;
1201                 }
1202         }
1203
1204         p->pid = pid_nr(pid);
1205         p->tgid = p->pid;
1206         if (clone_flags & CLONE_THREAD)
1207                 p->tgid = current->tgid;
1208
1209         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1210         /*
1211          * Clear TID on mm_release()?
1212          */
1213         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1214 #ifdef CONFIG_FUTEX
1215         p->robust_list = NULL;
1216 #ifdef CONFIG_COMPAT
1217         p->compat_robust_list = NULL;
1218 #endif
1219         INIT_LIST_HEAD(&p->pi_state_list);
1220         p->pi_state_cache = NULL;
1221 #endif
1222         /*
1223          * sigaltstack should be cleared when sharing the same VM
1224          */
1225         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1226                 p->sas_ss_sp = p->sas_ss_size = 0;
1227
1228         /*
1229          * Syscall tracing should be turned off in the child regardless
1230          * of CLONE_PTRACE.
1231          */
1232         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1233 #ifdef TIF_SYSCALL_EMU
1234         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1235 #endif
1236         clear_all_latency_tracing(p);
1237
1238         /* Our parent execution domain becomes current domain
1239            These must match for thread signalling to apply */
1240         p->parent_exec_id = p->self_exec_id;
1241
1242         /* ok, now we should be set up.. */
1243         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1244         p->pdeath_signal = 0;
1245         p->exit_state = 0;
1246
1247         /*
1248          * Ok, make it visible to the rest of the system.
1249          * We dont wake it up yet.
1250          */
1251         p->group_leader = p;
1252         INIT_LIST_HEAD(&p->thread_group);
1253         INIT_LIST_HEAD(&p->ptrace_children);
1254         INIT_LIST_HEAD(&p->ptrace_list);
1255
1256         /* Now that the task is set up, run cgroup callbacks if
1257          * necessary. We need to run them before the task is visible
1258          * on the tasklist. */
1259         cgroup_fork_callbacks(p);
1260         cgroup_callbacks_done = 1;
1261
1262         /* Need tasklist lock for parent etc handling! */
1263         write_lock_irq(&tasklist_lock);
1264
1265         /*
1266          * The task hasn't been attached yet, so its cpus_allowed mask will
1267          * not be changed, nor will its assigned CPU.
1268          *
1269          * The cpus_allowed mask of the parent may have changed after it was
1270          * copied first time - so re-copy it here, then check the child's CPU
1271          * to ensure it is on a valid CPU (and if not, just force it back to
1272          * parent's CPU). This avoids alot of nasty races.
1273          */
1274         p->cpus_allowed = current->cpus_allowed;
1275         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1276         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1277                         !cpu_online(task_cpu(p))))
1278                 set_task_cpu(p, smp_processor_id());
1279
1280         /* CLONE_PARENT re-uses the old parent */
1281         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1282                 p->real_parent = current->real_parent;
1283         else
1284                 p->real_parent = current;
1285         p->parent = p->real_parent;
1286
1287         spin_lock(&current->sighand->siglock);
1288
1289         /*
1290          * Process group and session signals need to be delivered to just the
1291          * parent before the fork or both the parent and the child after the
1292          * fork. Restart if a signal comes in before we add the new process to
1293          * it's process group.
1294          * A fatal signal pending means that current will exit, so the new
1295          * thread can't slip out of an OOM kill (or normal SIGKILL).
1296          */
1297         recalc_sigpending();
1298         if (signal_pending(current)) {
1299                 spin_unlock(&current->sighand->siglock);
1300                 write_unlock_irq(&tasklist_lock);
1301                 retval = -ERESTARTNOINTR;
1302                 goto bad_fork_free_pid;
1303         }
1304
1305         if (clone_flags & CLONE_THREAD) {
1306                 p->group_leader = current->group_leader;
1307                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1308
1309                 if (!cputime_eq(current->signal->it_virt_expires,
1310                                 cputime_zero) ||
1311                     !cputime_eq(current->signal->it_prof_expires,
1312                                 cputime_zero) ||
1313                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1314                     !list_empty(&current->signal->cpu_timers[0]) ||
1315                     !list_empty(&current->signal->cpu_timers[1]) ||
1316                     !list_empty(&current->signal->cpu_timers[2])) {
1317                         /*
1318                          * Have child wake up on its first tick to check
1319                          * for process CPU timers.
1320                          */
1321                         p->it_prof_expires = jiffies_to_cputime(1);
1322                 }
1323         }
1324
1325         if (likely(p->pid)) {
1326                 add_parent(p);
1327                 if (unlikely(p->ptrace & PT_PTRACED))
1328                         __ptrace_link(p, current->parent);
1329
1330                 if (thread_group_leader(p)) {
1331                         if (clone_flags & CLONE_NEWPID)
1332                                 p->nsproxy->pid_ns->child_reaper = p;
1333
1334                         p->signal->tty = current->signal->tty;
1335                         set_task_pgrp(p, task_pgrp_nr(current));
1336                         set_task_session(p, task_session_nr(current));
1337                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1338                         attach_pid(p, PIDTYPE_SID, task_session(current));
1339                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1340                         __get_cpu_var(process_counts)++;
1341                 }
1342                 attach_pid(p, PIDTYPE_PID, pid);
1343                 nr_threads++;
1344         }
1345
1346         total_forks++;
1347         spin_unlock(&current->sighand->siglock);
1348         write_unlock_irq(&tasklist_lock);
1349         proc_fork_connector(p);
1350         cgroup_post_fork(p);
1351         return p;
1352
1353 bad_fork_free_pid:
1354         if (pid != &init_struct_pid)
1355                 free_pid(pid);
1356 bad_fork_cleanup_io:
1357         put_io_context(p->io_context);
1358 bad_fork_cleanup_namespaces:
1359         exit_task_namespaces(p);
1360 bad_fork_cleanup_keys:
1361         exit_keys(p);
1362 bad_fork_cleanup_mm:
1363         if (p->mm)
1364                 mmput(p->mm);
1365 bad_fork_cleanup_signal:
1366         cleanup_signal(p);
1367 bad_fork_cleanup_sighand:
1368         __cleanup_sighand(p->sighand);
1369 bad_fork_cleanup_fs:
1370         exit_fs(p); /* blocking */
1371 bad_fork_cleanup_files:
1372         exit_files(p); /* blocking */
1373 bad_fork_cleanup_semundo:
1374         exit_sem(p);
1375 bad_fork_cleanup_audit:
1376         audit_free(p);
1377 bad_fork_cleanup_security:
1378         security_task_free(p);
1379 bad_fork_cleanup_policy:
1380 #ifdef CONFIG_NUMA
1381         mpol_free(p->mempolicy);
1382 bad_fork_cleanup_cgroup:
1383 #endif
1384         cgroup_exit(p, cgroup_callbacks_done);
1385         delayacct_tsk_free(p);
1386         if (p->binfmt)
1387                 module_put(p->binfmt->module);
1388 bad_fork_cleanup_put_domain:
1389         module_put(task_thread_info(p)->exec_domain->module);
1390 bad_fork_cleanup_count:
1391         put_group_info(p->group_info);
1392         atomic_dec(&p->user->processes);
1393         free_uid(p->user);
1394 bad_fork_free:
1395         free_task(p);
1396 fork_out:
1397         return ERR_PTR(retval);
1398 }
1399
1400 noinline struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1401 {
1402         memset(regs, 0, sizeof(struct pt_regs));
1403         return regs;
1404 }
1405
1406 struct task_struct * __cpuinit fork_idle(int cpu)
1407 {
1408         struct task_struct *task;
1409         struct pt_regs regs;
1410
1411         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1412                                 &init_struct_pid);
1413         if (!IS_ERR(task))
1414                 init_idle(task, cpu);
1415
1416         return task;
1417 }
1418
1419 static int fork_traceflag(unsigned clone_flags)
1420 {
1421         if (clone_flags & CLONE_UNTRACED)
1422                 return 0;
1423         else if (clone_flags & CLONE_VFORK) {
1424                 if (current->ptrace & PT_TRACE_VFORK)
1425                         return PTRACE_EVENT_VFORK;
1426         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1427                 if (current->ptrace & PT_TRACE_CLONE)
1428                         return PTRACE_EVENT_CLONE;
1429         } else if (current->ptrace & PT_TRACE_FORK)
1430                 return PTRACE_EVENT_FORK;
1431
1432         return 0;
1433 }
1434
1435 /*
1436  *  Ok, this is the main fork-routine.
1437  *
1438  * It copies the process, and if successful kick-starts
1439  * it and waits for it to finish using the VM if required.
1440  */
1441 long do_fork(unsigned long clone_flags,
1442               unsigned long stack_start,
1443               struct pt_regs *regs,
1444               unsigned long stack_size,
1445               int __user *parent_tidptr,
1446               int __user *child_tidptr)
1447 {
1448         struct task_struct *p;
1449         int trace = 0;
1450         long nr;
1451
1452         if (unlikely(current->ptrace)) {
1453                 trace = fork_traceflag (clone_flags);
1454                 if (trace)
1455                         clone_flags |= CLONE_PTRACE;
1456         }
1457
1458         p = copy_process(clone_flags, stack_start, regs, stack_size,
1459                         child_tidptr, NULL);
1460         /*
1461          * Do this prior waking up the new thread - the thread pointer
1462          * might get invalid after that point, if the thread exits quickly.
1463          */
1464         if (!IS_ERR(p)) {
1465                 struct completion vfork;
1466
1467                 /*
1468                  * this is enough to call pid_nr_ns here, but this if
1469                  * improves optimisation of regular fork()
1470                  */
1471                 nr = (clone_flags & CLONE_NEWPID) ?
1472                         task_pid_nr_ns(p, current->nsproxy->pid_ns) :
1473                                 task_pid_vnr(p);
1474
1475                 if (clone_flags & CLONE_PARENT_SETTID)
1476                         put_user(nr, parent_tidptr);
1477
1478                 if (clone_flags & CLONE_VFORK) {
1479                         p->vfork_done = &vfork;
1480                         init_completion(&vfork);
1481                 }
1482
1483                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1484                         /*
1485                          * We'll start up with an immediate SIGSTOP.
1486                          */
1487                         sigaddset(&p->pending.signal, SIGSTOP);
1488                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1489                 }
1490
1491                 if (!(clone_flags & CLONE_STOPPED))
1492                         wake_up_new_task(p, clone_flags);
1493                 else
1494                         p->state = TASK_STOPPED;
1495
1496                 if (unlikely (trace)) {
1497                         current->ptrace_message = nr;
1498                         ptrace_notify ((trace << 8) | SIGTRAP);
1499                 }
1500
1501                 if (clone_flags & CLONE_VFORK) {
1502                         freezer_do_not_count();
1503                         wait_for_completion(&vfork);
1504                         freezer_count();
1505                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1506                                 current->ptrace_message = nr;
1507                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1508                         }
1509                 }
1510         } else {
1511                 nr = PTR_ERR(p);
1512         }
1513         return nr;
1514 }
1515
1516 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1517 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1518 #endif
1519
1520 static void sighand_ctor(struct kmem_cache *cachep, void *data)
1521 {
1522         struct sighand_struct *sighand = data;
1523
1524         spin_lock_init(&sighand->siglock);
1525         init_waitqueue_head(&sighand->signalfd_wqh);
1526 }
1527
1528 void __init proc_caches_init(void)
1529 {
1530         sighand_cachep = kmem_cache_create("sighand_cache",
1531                         sizeof(struct sighand_struct), 0,
1532                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1533                         sighand_ctor);
1534         signal_cachep = kmem_cache_create("signal_cache",
1535                         sizeof(struct signal_struct), 0,
1536                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1537         files_cachep = kmem_cache_create("files_cache",
1538                         sizeof(struct files_struct), 0,
1539                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1540         fs_cachep = kmem_cache_create("fs_cache",
1541                         sizeof(struct fs_struct), 0,
1542                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1543         vm_area_cachep = kmem_cache_create("vm_area_struct",
1544                         sizeof(struct vm_area_struct), 0,
1545                         SLAB_PANIC, NULL);
1546         mm_cachep = kmem_cache_create("mm_struct",
1547                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1548                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1549 }
1550
1551 /*
1552  * Check constraints on flags passed to the unshare system call and
1553  * force unsharing of additional process context as appropriate.
1554  */
1555 static void check_unshare_flags(unsigned long *flags_ptr)
1556 {
1557         /*
1558          * If unsharing a thread from a thread group, must also
1559          * unshare vm.
1560          */
1561         if (*flags_ptr & CLONE_THREAD)
1562                 *flags_ptr |= CLONE_VM;
1563
1564         /*
1565          * If unsharing vm, must also unshare signal handlers.
1566          */
1567         if (*flags_ptr & CLONE_VM)
1568                 *flags_ptr |= CLONE_SIGHAND;
1569
1570         /*
1571          * If unsharing signal handlers and the task was created
1572          * using CLONE_THREAD, then must unshare the thread
1573          */
1574         if ((*flags_ptr & CLONE_SIGHAND) &&
1575             (atomic_read(&current->signal->count) > 1))
1576                 *flags_ptr |= CLONE_THREAD;
1577
1578         /*
1579          * If unsharing namespace, must also unshare filesystem information.
1580          */
1581         if (*flags_ptr & CLONE_NEWNS)
1582                 *flags_ptr |= CLONE_FS;
1583 }
1584
1585 /*
1586  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1587  */
1588 static int unshare_thread(unsigned long unshare_flags)
1589 {
1590         if (unshare_flags & CLONE_THREAD)
1591                 return -EINVAL;
1592
1593         return 0;
1594 }
1595
1596 /*
1597  * Unshare the filesystem structure if it is being shared
1598  */
1599 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1600 {
1601         struct fs_struct *fs = current->fs;
1602
1603         if ((unshare_flags & CLONE_FS) &&
1604             (fs && atomic_read(&fs->count) > 1)) {
1605                 *new_fsp = __copy_fs_struct(current->fs);
1606                 if (!*new_fsp)
1607                         return -ENOMEM;
1608         }
1609
1610         return 0;
1611 }
1612
1613 /*
1614  * Unsharing of sighand is not supported yet
1615  */
1616 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1617 {
1618         struct sighand_struct *sigh = current->sighand;
1619
1620         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1621                 return -EINVAL;
1622         else
1623                 return 0;
1624 }
1625
1626 /*
1627  * Unshare vm if it is being shared
1628  */
1629 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1630 {
1631         struct mm_struct *mm = current->mm;
1632
1633         if ((unshare_flags & CLONE_VM) &&
1634             (mm && atomic_read(&mm->mm_users) > 1)) {
1635                 return -EINVAL;
1636         }
1637
1638         return 0;
1639 }
1640
1641 /*
1642  * Unshare file descriptor table if it is being shared
1643  */
1644 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1645 {
1646         struct files_struct *fd = current->files;
1647         int error = 0;
1648
1649         if ((unshare_flags & CLONE_FILES) &&
1650             (fd && atomic_read(&fd->count) > 1)) {
1651                 *new_fdp = dup_fd(fd, &error);
1652                 if (!*new_fdp)
1653                         return error;
1654         }
1655
1656         return 0;
1657 }
1658
1659 /*
1660  * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1661  * supported yet
1662  */
1663 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1664 {
1665         if (unshare_flags & CLONE_SYSVSEM)
1666                 return -EINVAL;
1667
1668         return 0;
1669 }
1670
1671 /*
1672  * unshare allows a process to 'unshare' part of the process
1673  * context which was originally shared using clone.  copy_*
1674  * functions used by do_fork() cannot be used here directly
1675  * because they modify an inactive task_struct that is being
1676  * constructed. Here we are modifying the current, active,
1677  * task_struct.
1678  */
1679 asmlinkage long sys_unshare(unsigned long unshare_flags)
1680 {
1681         int err = 0;
1682         struct fs_struct *fs, *new_fs = NULL;
1683         struct sighand_struct *new_sigh = NULL;
1684         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1685         struct files_struct *fd, *new_fd = NULL;
1686         struct sem_undo_list *new_ulist = NULL;
1687         struct nsproxy *new_nsproxy = NULL;
1688
1689         check_unshare_flags(&unshare_flags);
1690
1691         /* Return -EINVAL for all unsupported flags */
1692         err = -EINVAL;
1693         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1694                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1695                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
1696                                 CLONE_NEWNET))
1697                 goto bad_unshare_out;
1698
1699         if ((err = unshare_thread(unshare_flags)))
1700                 goto bad_unshare_out;
1701         if ((err = unshare_fs(unshare_flags, &new_fs)))
1702                 goto bad_unshare_cleanup_thread;
1703         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1704                 goto bad_unshare_cleanup_fs;
1705         if ((err = unshare_vm(unshare_flags, &new_mm)))
1706                 goto bad_unshare_cleanup_sigh;
1707         if ((err = unshare_fd(unshare_flags, &new_fd)))
1708                 goto bad_unshare_cleanup_vm;
1709         if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1710                 goto bad_unshare_cleanup_fd;
1711         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1712                         new_fs)))
1713                 goto bad_unshare_cleanup_semundo;
1714
1715         if (new_fs ||  new_mm || new_fd || new_ulist || new_nsproxy) {
1716
1717                 if (new_nsproxy) {
1718                         switch_task_namespaces(current, new_nsproxy);
1719                         new_nsproxy = NULL;
1720                 }
1721
1722                 task_lock(current);
1723
1724                 if (new_fs) {
1725                         fs = current->fs;
1726                         current->fs = new_fs;
1727                         new_fs = fs;
1728                 }
1729
1730                 if (new_mm) {
1731                         mm = current->mm;
1732                         active_mm = current->active_mm;
1733                         current->mm = new_mm;
1734                         current->active_mm = new_mm;
1735                         activate_mm(active_mm, new_mm);
1736                         new_mm = mm;
1737                 }
1738
1739                 if (new_fd) {
1740                         fd = current->files;
1741                         current->files = new_fd;
1742                         new_fd = fd;
1743                 }
1744
1745                 task_unlock(current);
1746         }
1747
1748         if (new_nsproxy)
1749                 put_nsproxy(new_nsproxy);
1750
1751 bad_unshare_cleanup_semundo:
1752 bad_unshare_cleanup_fd:
1753         if (new_fd)
1754                 put_files_struct(new_fd);
1755
1756 bad_unshare_cleanup_vm:
1757         if (new_mm)
1758                 mmput(new_mm);
1759
1760 bad_unshare_cleanup_sigh:
1761         if (new_sigh)
1762                 if (atomic_dec_and_test(&new_sigh->count))
1763                         kmem_cache_free(sighand_cachep, new_sigh);
1764
1765 bad_unshare_cleanup_fs:
1766         if (new_fs)
1767                 put_fs_struct(new_fs);
1768
1769 bad_unshare_cleanup_thread:
1770 bad_unshare_out:
1771         return err;
1772 }