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/smp_lock.h>
  18 #include <linux/module.h>
  19 #include <linux/vmalloc.h>
  20 #include <linux/completion.h>
  21 #include <linux/mnt_namespace.h>
  22 #include <linux/personality.h>
  23 #include <linux/mempolicy.h>
  24 #include <linux/sem.h>
  25 #include <linux/file.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/fs.h>
  30 #include <linux/nsproxy.h>
  31 #include <linux/capability.h>
  32 #include <linux/cpu.h>
  33 #include <linux/cpuset.h>
  34 #include <linux/security.h>
  35 #include <linux/swap.h>
  36 #include <linux/syscalls.h>
  37 #include <linux/jiffies.h>
  38 #include <linux/futex.h>
  39 #include <linux/task_io_accounting_ops.h>
  40 #include <linux/rcupdate.h>
  41 #include <linux/tracehook.h>
  42 #include <linux/mount.h>
  43 #include <linux/audit.h>
  44 #include <linux/profile.h>
  45 #include <linux/rmap.h>
  46 #include <linux/acct.h>
  47 #include <linux/tsacct_kern.h>
  48 #include <linux/cn_proc.h>
  49 #include <linux/delayacct.h>
  50 #include <linux/taskstats_kern.h>
  51 #include <linux/random.h>
  52 #include <linux/ptrace.h>
  53 #include <linux/vs_context.h>
  54 #include <linux/vs_network.h>
  55 #include <linux/vs_limit.h>
  56 #include <linux/vs_memory.h>
  57 #include <linux/vserver/global.h>
  58
  59 #include <asm/pgtable.h>
  60 #include <asm/pgalloc.h>
  61 #include <asm/uaccess.h>
  62 #include <asm/mmu_context.h>
  63 #include <asm/cacheflush.h>
  64 #include <asm/tlbflush.h>
  65
  66 /*
  67  * Protected counters by write_lock_irq(&tasklist_lock)
  68  */
  69 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  70 int nr_threads;                 /* The idle threads do not count.. */
  71
  72 int max_threads;                /* tunable limit on nr_threads */
  73
  74 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  75
  76 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  77
  78 int nr_processes(void)
  79 {
  80         int cpu;
  81         int total = 0;
  82
  83         for_each_online_cpu(cpu)
  84                 total += per_cpu(process_counts, cpu);
  85
  86         return total;
  87 }
  88
  89 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  90 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  91 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  92 static struct kmem_cache *task_struct_cachep;
  93 #endif
  94
  95 /* SLAB cache for signal_struct structures (tsk->signal) */
  96 static struct kmem_cache *signal_cachep;
  97
  98 /* SLAB cache for sighand_struct structures (tsk->sighand) */
  99 struct kmem_cache *sighand_cachep;
 100
 101 /* SLAB cache for files_struct structures (tsk->files) */
 102 struct kmem_cache *files_cachep;
 103
 104 /* SLAB cache for fs_struct structures (tsk->fs) */
 105 struct kmem_cache *fs_cachep;
 106
 107 /* SLAB cache for vm_area_struct structures */
 108 struct kmem_cache *vm_area_cachep;
 109
 110 /* SLAB cache for mm_struct structures (tsk->mm) */
 111 static struct kmem_cache *mm_cachep;
 112
 113 void free_task(struct task_struct *tsk)
 114 {
 115         free_thread_info(tsk->thread_info);
 116         rt_mutex_debug_task_free(tsk);
 117         clr_vx_info(&tsk->vx_info);
 118         clr_nx_info(&tsk->nx_info);
 119         free_task_struct(tsk);
 120 }
 121 EXPORT_SYMBOL(free_task);
 122
 123 void __put_task_struct(struct task_struct *tsk)
 124 {
 125         WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
 126         WARN_ON(atomic_read(&tsk->usage));
 127         WARN_ON(tsk == current);
 128
 129         security_task_free(tsk);
 130         free_uid(tsk->user);
 131         put_group_info(tsk->group_info);
 132         delayacct_tsk_free(tsk);
 133
 134         if (!profile_handoff_task(tsk))
 135                 free_task(tsk);
 136 }
 137
 138 void __init fork_init(unsigned long mempages)
 139 {
 140 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 141 #ifndef ARCH_MIN_TASKALIGN
 142 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 143 #endif
 144         /* create a slab on which task_structs can be allocated */
 145         task_struct_cachep =
 146                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 147                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
 148 #endif
 149
 150         /*
 151          * The default maximum number of threads is set to a safe
 152          * value: the thread structures can take up at most half
 153          * of memory.
 154          */
 155         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 156
 157         /*
 158          * we need to allow at least 20 threads to boot a system
 159          */
 160         if(max_threads < 20)
 161                 max_threads = 20;
 162
 163         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 164         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 165         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 166                 init_task.signal->rlim[RLIMIT_NPROC];
 167 }
 168
 169 static struct task_struct *dup_task_struct(struct task_struct *orig)
 170 {
 171         struct task_struct *tsk;
 172         struct thread_info *ti;
 173
 174         prepare_to_copy(orig);
 175
 176         tsk = alloc_task_struct();
 177         if (!tsk)
 178                 return NULL;
 179
 180         ti = alloc_thread_info(tsk);
 181         if (!ti) {
 182                 free_task_struct(tsk);
 183                 return NULL;
 184         }
 185
 186         *tsk = *orig;
 187         tsk->thread_info = ti;
 188         setup_thread_stack(tsk, orig);
 189
 190 #ifdef CONFIG_CC_STACKPROTECTOR
 191         tsk->stack_canary = get_random_int();
 192 #endif
 193
 194         /* One for us, one for whoever does the "release_task()" (usually parent) */
 195         atomic_set(&tsk->usage,2);
 196         atomic_set(&tsk->fs_excl, 0);
 197 #ifdef CONFIG_BLK_DEV_IO_TRACE
 198         tsk->btrace_seq = 0;
 199 #endif
 200         tsk->splice_pipe = NULL;
 201         return tsk;
 202 }
 203
 204 #ifdef CONFIG_MMU
 205 static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 206 {
 207         struct vm_area_struct *mpnt, *tmp, **pprev;
 208         struct rb_node **rb_link, *rb_parent;
 209         int retval;
 210         unsigned long charge;
 211         struct mempolicy *pol;
 212
 213         down_write(&oldmm->mmap_sem);
 214         flush_cache_dup_mm(oldmm);
 215         /*
 216          * Not linked in yet - no deadlock potential:
 217          */
 218         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 219
 220         mm->locked_vm = 0;
 221         mm->mmap = NULL;
 222         mm->mmap_cache = NULL;
 223         mm->free_area_cache = oldmm->mmap_base;
 224         mm->cached_hole_size = ~0UL;
 225         mm->map_count = 0;
 226         __set_mm_counter(mm, file_rss, 0);
 227         __set_mm_counter(mm, anon_rss, 0);
 228         cpus_clear(mm->cpu_vm_mask);
 229         mm->mm_rb = RB_ROOT;
 230         rb_link = &mm->mm_rb.rb_node;
 231         rb_parent = NULL;
 232         pprev = &mm->mmap;
 233
 234         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 235                 struct file *file;
 236
 237                 if (mpnt->vm_flags & VM_DONTCOPY) {
 238                         long pages = vma_pages(mpnt);
 239                         vx_vmpages_sub(mm, pages);
 240                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 241                                                                 -pages);
 242                         continue;
 243                 }
 244                 charge = 0;
 245                 if (mpnt->vm_flags & VM_ACCOUNT) {
 246                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 247                         if (security_vm_enough_memory(len))
 248                                 goto fail_nomem;
 249                         charge = len;
 250                 }
 251                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 252                 if (!tmp)
 253                         goto fail_nomem;
 254                 *tmp = *mpnt;
 255                 pol = mpol_copy(vma_policy(mpnt));
 256                 retval = PTR_ERR(pol);
 257                 if (IS_ERR(pol))
 258                         goto fail_nomem_policy;
 259                 vma_set_policy(tmp, pol);
 260                 tmp->vm_flags &= ~VM_LOCKED;
 261                 tmp->vm_mm = mm;
 262                 tmp->vm_next = NULL;
 263                 anon_vma_link(tmp);
 264                 file = tmp->vm_file;
 265                 if (file) {
 266                         struct inode *inode = file->f_path.dentry->d_inode;
 267                         get_file(file);
 268                         if (tmp->vm_flags & VM_DENYWRITE)
 269                                 atomic_dec(&inode->i_writecount);
 270
 271                         /* insert tmp into the share list, just after mpnt */
 272                         spin_lock(&file->f_mapping->i_mmap_lock);
 273                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 274                         flush_dcache_mmap_lock(file->f_mapping);
 275                         vma_prio_tree_add(tmp, mpnt);
 276                         flush_dcache_mmap_unlock(file->f_mapping);
 277                         spin_unlock(&file->f_mapping->i_mmap_lock);
 278                 }
 279
 280                 /*
 281                  * Link in the new vma and copy the page table entries.
 282                  */
 283                 *pprev = tmp;
 284                 pprev = &tmp->vm_next;
 285
 286                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 287                 rb_link = &tmp->vm_rb.rb_right;
 288                 rb_parent = &tmp->vm_rb;
 289
 290                 mm->map_count++;
 291                 retval = copy_page_range(mm, oldmm, mpnt);
 292
 293                 if (tmp->vm_ops && tmp->vm_ops->open)
 294                         tmp->vm_ops->open(tmp);
 295
 296                 if (retval)
 297                         goto out;
 298         }
 299 #ifdef arch_dup_mmap
 300         arch_dup_mmap(mm, oldmm);
 301 #endif
 302         retval = 0;
 303 out:
 304         up_write(&mm->mmap_sem);
 305         flush_tlb_mm(oldmm);
 306         up_write(&oldmm->mmap_sem);
 307         return retval;
 308 fail_nomem_policy:
 309         kmem_cache_free(vm_area_cachep, tmp);
 310 fail_nomem:
 311         retval = -ENOMEM;
 312         vm_unacct_memory(charge);
 313         goto out;
 314 }
 315
 316 static inline int mm_alloc_pgd(struct mm_struct * mm)
 317 {
 318         mm->pgd = pgd_alloc(mm);
 319         if (unlikely(!mm->pgd))
 320                 return -ENOMEM;
 321         return 0;
 322 }
 323
 324 static inline void mm_free_pgd(struct mm_struct * mm)
 325 {
 326         pgd_free(mm->pgd);
 327 }
 328 #else
 329 #define dup_mmap(mm, oldmm)     (0)
 330 #define mm_alloc_pgd(mm)        (0)
 331 #define mm_free_pgd(mm)
 332 #endif /* CONFIG_MMU */
 333
 334  __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 335
 336 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 337 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 338
 339 #include <linux/init_task.h>
 340
 341 static struct mm_struct * mm_init(struct mm_struct * mm)
 342 {
 343         atomic_set(&mm->mm_users, 1);
 344         atomic_set(&mm->mm_count, 1);
 345         init_rwsem(&mm->mmap_sem);
 346         INIT_LIST_HEAD(&mm->mmlist);
 347         mm->core_waiters = 0;
 348         mm->nr_ptes = 0;
 349         __set_mm_counter(mm, file_rss, 0);
 350         __set_mm_counter(mm, anon_rss, 0);
 351         spin_lock_init(&mm->page_table_lock);
 352         rwlock_init(&mm->ioctx_list_lock);
 353         mm->ioctx_list = NULL;
 354         mm->free_area_cache = TASK_UNMAPPED_BASE;
 355         mm->cached_hole_size = ~0UL;
 356
 357         if (likely(!mm_alloc_pgd(mm))) {
 358                 mm->def_flags = 0;
 359                 set_vx_info(&mm->mm_vx_info, current->vx_info);
 360                 return mm;
 361         }
 362         free_mm(mm);
 363         return NULL;
 364 }
 365
 366 /*
 367  * Allocate and initialize an mm_struct.
 368  */
 369 struct mm_struct * mm_alloc(void)
 370 {
 371         struct mm_struct * mm;
 372
 373         mm = allocate_mm();
 374         if (mm) {
 375                 memset(mm, 0, sizeof(*mm));
 376                 mm = mm_init(mm);
 377         }
 378         return mm;
 379 }
 380
 381 /*
 382  * Called when the last reference to the mm
 383  * is dropped: either by a lazy thread or by
 384  * mmput. Free the page directory and the mm.
 385  */
 386 void fastcall __mmdrop(struct mm_struct *mm)
 387 {
 388         BUG_ON(mm == &init_mm);
 389         mm_free_pgd(mm);
 390         destroy_context(mm);
 391         clr_vx_info(&mm->mm_vx_info);
 392         free_mm(mm);
 393 }
 394
 395 /*
 396  * Decrement the use count and release all resources for an mm.
 397  */
 398 void mmput(struct mm_struct *mm)
 399 {
 400         might_sleep();
 401
 402         if (atomic_dec_and_test(&mm->mm_users)) {
 403                 exit_aio(mm);
 404                 exit_mmap(mm);
 405                 if (!list_empty(&mm->mmlist)) {
 406                         spin_lock(&mmlist_lock);
 407                         list_del(&mm->mmlist);
 408                         spin_unlock(&mmlist_lock);
 409                 }
 410                 put_swap_token(mm);
 411                 mmdrop(mm);
 412         }
 413 }
 414 EXPORT_SYMBOL_GPL(mmput);
 415
 416 /**
 417  * get_task_mm - acquire a reference to the task's mm
 418  *
 419  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
 420  * this kernel workthread has transiently adopted a user mm with use_mm,
 421  * to do its AIO) is not set and if so returns a reference to it, after
 422  * bumping up the use count.  User must release the mm via mmput()
 423  * after use.  Typically used by /proc and ptrace.
 424  */
 425 struct mm_struct *get_task_mm(struct task_struct *task)
 426 {
 427         struct mm_struct *mm;
 428
 429         task_lock(task);
 430         mm = task->mm;
 431         if (mm) {
 432                 if (task->flags & PF_BORROWED_MM)
 433                         mm = NULL;
 434                 else
 435                         atomic_inc(&mm->mm_users);
 436         }
 437         task_unlock(task);
 438         return mm;
 439 }
 440 EXPORT_SYMBOL_GPL(get_task_mm);
 441
 442 /* Please note the differences between mmput and mm_release.
 443  * mmput is called whenever we stop holding onto a mm_struct,
 444  * error success whatever.
 445  *
 446  * mm_release is called after a mm_struct has been removed
 447  * from the current process.
 448  *
 449  * This difference is important for error handling, when we
 450  * only half set up a mm_struct for a new process and need to restore
 451  * the old one.  Because we mmput the new mm_struct before
 452  * restoring the old one. . .
 453  * Eric Biederman 10 January 1998
 454  */
 455 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 456 {
 457         struct completion *vfork_done = tsk->vfork_done;
 458
 459         /* Get rid of any cached register state */
 460         deactivate_mm(tsk, mm);
 461
 462         /* notify parent sleeping on vfork() */
 463         if (vfork_done) {
 464                 tsk->vfork_done = NULL;
 465                 complete(vfork_done);
 466         }
 467
 468         /*
 469          * If we're exiting normally, clear a user-space tid field if
 470          * requested.  We leave this alone when dying by signal, to leave
 471          * the value intact in a core dump, and to save the unnecessary
 472          * trouble otherwise.  Userland only wants this done for a sys_exit.
 473          */
 474         if (tsk->clear_child_tid
 475             && !(tsk->flags & PF_SIGNALED)
 476             && atomic_read(&mm->mm_users) > 1) {
 477                 u32 __user * tidptr = tsk->clear_child_tid;
 478                 tsk->clear_child_tid = NULL;
 479
 480                 /*
 481                  * We don't check the error code - if userspace has
 482                  * not set up a proper pointer then tough luck.
 483                  */
 484                 put_user(0, tidptr);
 485                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 486         }
 487 }
 488
 489 /*
 490  * Allocate a new mm structure and copy contents from the
 491  * mm structure of the passed in task structure.
 492  */
 493 static struct mm_struct *dup_mm(struct task_struct *tsk)
 494 {
 495         struct mm_struct *mm, *oldmm = current->mm;
 496         int err;
 497
 498         if (!oldmm)
 499                 return NULL;
 500
 501         mm = allocate_mm();
 502         if (!mm)
 503                 goto fail_nomem;
 504
 505         memcpy(mm, oldmm, sizeof(*mm));
 506         mm->mm_vx_info = NULL;
 507
 508         /* Initializing for Swap token stuff */
 509         mm->token_priority = 0;
 510         mm->last_interval = 0;
 511
 512         if (!mm_init(mm))
 513                 goto fail_nomem;
 514
 515         if (init_new_context(tsk, mm))
 516                 goto fail_nocontext;
 517
 518         err = dup_mmap(mm, oldmm);
 519         if (err)
 520                 goto free_pt;
 521
 522         mm->hiwater_rss = get_mm_rss(mm);
 523         mm->hiwater_vm = mm->total_vm;
 524
 525         return mm;
 526
 527 free_pt:
 528         mmput(mm);
 529
 530 fail_nomem:
 531         return NULL;
 532
 533 fail_nocontext:
 534         /*
 535          * If init_new_context() failed, we cannot use mmput() to free the mm
 536          * because it calls destroy_context()
 537          */
 538         clr_vx_info(&mm->mm_vx_info);
 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 inline 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                 atomic_inc(&vs_global_fs);
 610         }
 611         return fs;
 612 }
 613
 614 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 615 {
 616         return __copy_fs_struct(old);
 617 }
 618
 619 EXPORT_SYMBOL_GPL(copy_fs_struct);
 620
 621 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 622 {
 623         if (clone_flags & CLONE_FS) {
 624                 atomic_inc(&current->fs->count);
 625                 return 0;
 626         }
 627         tsk->fs = __copy_fs_struct(current->fs);
 628         if (!tsk->fs)
 629                 return -ENOMEM;
 630         return 0;
 631 }
 632
 633 static int count_open_files(struct fdtable *fdt)
 634 {
 635         int size = fdt->max_fds;
 636         int i;
 637
 638         /* Find the last open fd */
 639         for (i = size/(8*sizeof(long)); i > 0; ) {
 640                 if (fdt->open_fds->fds_bits[--i])
 641                         break;
 642         }
 643         i = (i+1) * 8 * sizeof(long);
 644         return i;
 645 }
 646
 647 static struct files_struct *alloc_files(void)
 648 {
 649         struct files_struct *newf;
 650         struct fdtable *fdt;
 651
 652         newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
 653         if (!newf)
 654                 goto out;
 655
 656         atomic_set(&newf->count, 1);
 657
 658         spin_lock_init(&newf->file_lock);
 659         newf->next_fd = 0;
 660         fdt = &newf->fdtab;
 661         fdt->max_fds = NR_OPEN_DEFAULT;
 662         fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
 663         fdt->open_fds = (fd_set *)&newf->open_fds_init;
 664         fdt->fd = &newf->fd_array[0];
 665         INIT_RCU_HEAD(&fdt->rcu);
 666         fdt->next = NULL;
 667         rcu_assign_pointer(newf->fdt, fdt);
 668 out:
 669         return newf;
 670 }
 671
 672 /*
 673  * Allocate a new files structure and copy contents from the
 674  * passed in files structure.
 675  * errorp will be valid only when the returned files_struct is NULL.
 676  */
 677 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
 678 {
 679         struct files_struct *newf;
 680         struct file **old_fds, **new_fds;
 681         int open_files, size, i;
 682         struct fdtable *old_fdt, *new_fdt;
 683
 684         *errorp = -ENOMEM;
 685         newf = alloc_files();
 686         if (!newf)
 687                 goto out;
 688
 689         spin_lock(&oldf->file_lock);
 690         old_fdt = files_fdtable(oldf);
 691         new_fdt = files_fdtable(newf);
 692         open_files = count_open_files(old_fdt);
 693
 694         /*
 695          * Check whether we need to allocate a larger fd array and fd set.
 696          * Note: we're not a clone task, so the open count won't change.
 697          */
 698         if (open_files > new_fdt->max_fds) {
 699                 new_fdt->max_fds = 0;
 700                 spin_unlock(&oldf->file_lock);
 701                 spin_lock(&newf->file_lock);
 702                 *errorp = expand_files(newf, open_files-1);
 703                 spin_unlock(&newf->file_lock);
 704                 if (*errorp < 0)
 705                         goto out_release;
 706                 new_fdt = files_fdtable(newf);
 707                 /*
 708                  * Reacquire the oldf lock and a pointer to its fd table
 709                  * who knows it may have a new bigger fd table. We need
 710                  * the latest pointer.
 711                  */
 712                 spin_lock(&oldf->file_lock);
 713                 old_fdt = files_fdtable(oldf);
 714         }
 715
 716         old_fds = old_fdt->fd;
 717         new_fds = new_fdt->fd;
 718
 719         memcpy(new_fdt->open_fds->fds_bits,
 720                 old_fdt->open_fds->fds_bits, open_files/8);
 721         memcpy(new_fdt->close_on_exec->fds_bits,
 722                 old_fdt->close_on_exec->fds_bits, open_files/8);
 723
 724         for (i = open_files; i != 0; i--) {
 725                 struct file *f = *old_fds++;
 726                 if (f) {
 727                         get_file(f);
 728                         /* TODO: sum it first for check and performance */
 729                         vx_openfd_inc(open_files - i);
 730                 } else {
 731                         /*
 732                          * The fd may be claimed in the fd bitmap but not yet
 733                          * instantiated in the files array if a sibling thread
 734                          * is partway through open().  So make sure that this
 735                          * fd is available to the new process.
 736                          */
 737                         FD_CLR(open_files - i, new_fdt->open_fds);
 738                 }
 739                 rcu_assign_pointer(*new_fds++, f);
 740         }
 741         spin_unlock(&oldf->file_lock);
 742
 743         /* compute the remainder to be cleared */
 744         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
 745
 746         /* This is long word aligned thus could use a optimized version */
 747         memset(new_fds, 0, size);
 748
 749         if (new_fdt->max_fds > open_files) {
 750                 int left = (new_fdt->max_fds-open_files)/8;
 751                 int start = open_files / (8 * sizeof(unsigned long));
 752
 753                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
 754                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
 755         }
 756
 757         return newf;
 758
 759 out_release:
 760         kmem_cache_free(files_cachep, newf);
 761 out:
 762         return NULL;
 763 }
 764
 765 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 766 {
 767         struct files_struct *oldf, *newf;
 768         int error = 0;
 769
 770         /*
 771          * A background process may not have any files ...
 772          */
 773         oldf = current->files;
 774         if (!oldf)
 775                 goto out;
 776
 777         if (clone_flags & CLONE_FILES) {
 778                 atomic_inc(&oldf->count);
 779                 goto out;
 780         }
 781
 782         /*
 783          * Note: we may be using current for both targets (See exec.c)
 784          * This works because we cache current->files (old) as oldf. Don't
 785          * break this.
 786          */
 787         tsk->files = NULL;
 788         newf = dup_fd(oldf, &error);
 789         if (!newf)
 790                 goto out;
 791
 792         tsk->files = newf;
 793         error = 0;
 794 out:
 795         return error;
 796 }
 797
 798 /*
 799  *      Helper to unshare the files of the current task.
 800  *      We don't want to expose copy_files internals to
 801  *      the exec layer of the kernel.
 802  */
 803
 804 int unshare_files(void)
 805 {
 806         struct files_struct *files  = current->files;
 807         int rc;
 808
 809         BUG_ON(!files);
 810
 811         /* This can race but the race causes us to copy when we don't
 812            need to and drop the copy */
 813         if(atomic_read(&files->count) == 1)
 814         {
 815                 atomic_inc(&files->count);
 816                 return 0;
 817         }
 818         rc = copy_files(0, current);
 819         if(rc)
 820                 current->files = files;
 821         return rc;
 822 }
 823
 824 EXPORT_SYMBOL(unshare_files);
 825
 826 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 827 {
 828         struct sighand_struct *sig;
 829
 830         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 831                 atomic_inc(&current->sighand->count);
 832                 return 0;
 833         }
 834         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 835         rcu_assign_pointer(tsk->sighand, sig);
 836         if (!sig)
 837                 return -ENOMEM;
 838         atomic_set(&sig->count, 1);
 839         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 840         return 0;
 841 }
 842
 843 void __cleanup_sighand(struct sighand_struct *sighand)
 844 {
 845         if (atomic_dec_and_test(&sighand->count))
 846                 kmem_cache_free(sighand_cachep, sighand);
 847 }
 848
 849 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 850 {
 851         struct signal_struct *sig;
 852         int ret;
 853
 854         if (clone_flags & CLONE_THREAD) {
 855                 atomic_inc(&current->signal->count);
 856                 atomic_inc(&current->signal->live);
 857                 return 0;
 858         }
 859         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 860         tsk->signal = sig;
 861         if (!sig)
 862                 return -ENOMEM;
 863
 864         ret = copy_thread_group_keys(tsk);
 865         if (ret < 0) {
 866                 kmem_cache_free(signal_cachep, sig);
 867                 return ret;
 868         }
 869
 870         atomic_set(&sig->count, 1);
 871         atomic_set(&sig->live, 1);
 872         init_waitqueue_head(&sig->wait_chldexit);
 873         sig->flags = 0;
 874         sig->group_exit_code = 0;
 875         sig->group_exit_task = NULL;
 876         sig->group_stop_count = 0;
 877         sig->curr_target = NULL;
 878         init_sigpending(&sig->shared_pending);
 879         INIT_LIST_HEAD(&sig->posix_timers);
 880
 881         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_REL);
 882         sig->it_real_incr.tv64 = 0;
 883         sig->real_timer.function = it_real_fn;
 884         sig->tsk = tsk;
 885
 886         sig->it_virt_expires = cputime_zero;
 887         sig->it_virt_incr = cputime_zero;
 888         sig->it_prof_expires = cputime_zero;
 889         sig->it_prof_incr = cputime_zero;
 890
 891         sig->leader = 0;        /* session leadership doesn't inherit */
 892         sig->tty_old_pgrp = 0;
 893
 894         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 895         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 896         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 897         sig->sched_time = 0;
 898         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 899         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 900         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 901         taskstats_tgid_init(sig);
 902
 903         task_lock(current->group_leader);
 904         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 905         task_unlock(current->group_leader);
 906
 907         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 908                 /*
 909                  * New sole thread in the process gets an expiry time
 910                  * of the whole CPU time limit.
 911                  */
 912                 tsk->it_prof_expires =
 913                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 914         }
 915         acct_init_pacct(&sig->pacct);
 916
 917         return 0;
 918 }
 919
 920 void __cleanup_signal(struct signal_struct *sig)
 921 {
 922         exit_thread_group_keys(sig);
 923         kmem_cache_free(signal_cachep, sig);
 924 }
 925
 926 static inline void cleanup_signal(struct task_struct *tsk)
 927 {
 928         struct signal_struct *sig = tsk->signal;
 929
 930         atomic_dec(&sig->live);
 931
 932         if (atomic_dec_and_test(&sig->count))
 933                 __cleanup_signal(sig);
 934 }
 935
 936 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 937 {
 938         unsigned long new_flags = p->flags;
 939
 940         new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
 941         new_flags |= PF_FORKNOEXEC;
 942         new_flags |= PF_STARTING;
 943         p->flags = new_flags;
 944 }
 945
 946 asmlinkage long sys_set_tid_address(int __user *tidptr)
 947 {
 948         current->clear_child_tid = tidptr;
 949
 950         return current->pid;
 951 }
 952
 953 static inline void rt_mutex_init_task(struct task_struct *p)
 954 {
 955         spin_lock_init(&p->pi_lock);
 956 #ifdef CONFIG_RT_MUTEXES
 957         plist_head_init(&p->pi_waiters, &p->pi_lock);
 958         p->pi_blocked_on = NULL;
 959 #endif
 960 }
 961
 962 /*
 963  * This creates a new process as a copy of the old one,
 964  * but does not actually start it yet.
 965  *
 966  * It copies the registers, and all the appropriate
 967  * parts of the process environment (as per the clone
 968  * flags). The actual kick-off is left to the caller.
 969  */
 970 static struct task_struct *copy_process(unsigned long clone_flags,
 971                                         unsigned long stack_start,
 972                                         struct pt_regs *regs,
 973                                         unsigned long stack_size,
 974                                         int __user *parent_tidptr,
 975                                         int __user *child_tidptr,
 976                                         int pid)
 977 {
 978         int retval;
 979         struct task_struct *p = NULL;
 980         struct vx_info *vxi;
 981         struct nx_info *nxi;
 982
 983         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 984                 return ERR_PTR(-EINVAL);
 985
 986         /*
 987          * Thread groups must share signals as well, and detached threads
 988          * can only be started up within the thread group.
 989          */
 990         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 991                 return ERR_PTR(-EINVAL);
 992
 993         /*
 994          * Shared signal handlers imply shared VM. By way of the above,
 995          * thread groups also imply shared VM. Blocking this case allows
 996          * for various simplifications in other code.
 997          */
 998         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 999                 return ERR_PTR(-EINVAL);
1000
1001         retval = security_task_create(clone_flags);
1002         if (retval)
1003                 goto fork_out;
1004
1005         retval = -ENOMEM;
1006         p = dup_task_struct(current);
1007         if (!p)
1008                 goto fork_out;
1009
1010         rt_mutex_init_task(p);
1011
1012 #ifdef CONFIG_TRACE_IRQFLAGS
1013         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1014         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1015 #endif
1016         init_vx_info(&p->vx_info, current->vx_info);
1017         init_nx_info(&p->nx_info, current->nx_info);
1018
1019         /* check vserver memory */
1020         if (p->mm && !(clone_flags & CLONE_VM)) {
1021                 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
1022                         vx_pages_add(p->vx_info, RLIMIT_AS, p->mm->total_vm);
1023                 else
1024                         goto bad_fork_free;
1025         }
1026         if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
1027                 if (!vx_rss_avail(p->mm, get_mm_counter(p->mm, file_rss)))
1028                         goto bad_fork_cleanup_vm;
1029         }
1030
1031         retval = -EAGAIN;
1032         if (!vx_nproc_avail(1))
1033                 goto bad_fork_cleanup_vm;
1034
1035         if (atomic_read(&p->user->processes) >=
1036                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1037                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1038                                 p->user != &root_user)
1039                         goto bad_fork_cleanup_vm;
1040         }
1041
1042         atomic_inc(&p->user->__count);
1043         atomic_inc(&p->user->processes);
1044         get_group_info(p->group_info);
1045
1046         /*
1047          * If multiple threads are within copy_process(), then this check
1048          * triggers too late. This doesn't hurt, the check is only there
1049          * to stop root fork bombs.
1050          */
1051         if (nr_threads >= max_threads)
1052                 goto bad_fork_cleanup_count;
1053
1054         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1055                 goto bad_fork_cleanup_count;
1056
1057         if (p->binfmt && !try_module_get(p->binfmt->module))
1058                 goto bad_fork_cleanup_put_domain;
1059
1060         p->did_exec = 0;
1061         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1062         copy_flags(clone_flags, p);
1063         p->pid = pid;
1064         retval = -EFAULT;
1065         if (clone_flags & CLONE_PARENT_SETTID)
1066                 if (put_user(p->pid, parent_tidptr))
1067                         goto bad_fork_cleanup_delays_binfmt;
1068
1069         INIT_LIST_HEAD(&p->children);
1070         INIT_LIST_HEAD(&p->sibling);
1071         p->vfork_done = NULL;
1072         spin_lock_init(&p->alloc_lock);
1073         ptrace_init_task(p);
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->sched_time = 0;
1081         p->rchar = 0;           /* I/O counter: bytes read */
1082         p->wchar = 0;           /* I/O counter: bytes written */
1083         p->syscr = 0;           /* I/O counter: read syscalls */
1084         p->syscw = 0;           /* I/O counter: write syscalls */
1085         task_io_accounting_init(p);
1086         acct_clear_integrals(p);
1087
1088         p->it_virt_expires = cputime_zero;
1089         p->it_prof_expires = cputime_zero;
1090         p->it_sched_expires = 0;
1091         INIT_LIST_HEAD(&p->cpu_timers[0]);
1092         INIT_LIST_HEAD(&p->cpu_timers[1]);
1093         INIT_LIST_HEAD(&p->cpu_timers[2]);
1094
1095         p->lock_depth = -1;             /* -1 = no lock */
1096         do_posix_clock_monotonic_gettime(&p->start_time);
1097         p->security = NULL;
1098         p->io_context = NULL;
1099         p->io_wait = NULL;
1100         p->audit_context = NULL;
1101         cpuset_fork(p);
1102 #ifdef CONFIG_NUMA
1103         p->mempolicy = mpol_copy(p->mempolicy);
1104         if (IS_ERR(p->mempolicy)) {
1105                 retval = PTR_ERR(p->mempolicy);
1106                 p->mempolicy = NULL;
1107                 goto bad_fork_cleanup_cpuset;
1108         }
1109         mpol_fix_fork_child_flag(p);
1110 #endif
1111 #ifdef CONFIG_TRACE_IRQFLAGS
1112         p->irq_events = 0;
1113 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1114         p->hardirqs_enabled = 1;
1115 #else
1116         p->hardirqs_enabled = 0;
1117 #endif
1118         p->hardirq_enable_ip = 0;
1119         p->hardirq_enable_event = 0;
1120         p->hardirq_disable_ip = _THIS_IP_;
1121         p->hardirq_disable_event = 0;
1122         p->softirqs_enabled = 1;
1123         p->softirq_enable_ip = _THIS_IP_;
1124         p->softirq_enable_event = 0;
1125         p->softirq_disable_ip = 0;
1126         p->softirq_disable_event = 0;
1127         p->hardirq_context = 0;
1128         p->softirq_context = 0;
1129 #endif
1130 #ifdef CONFIG_LOCKDEP
1131         p->lockdep_depth = 0; /* no locks held yet */
1132         p->curr_chain_key = 0;
1133         p->lockdep_recursion = 0;
1134 #endif
1135
1136 #ifdef CONFIG_DEBUG_MUTEXES
1137         p->blocked_on = NULL; /* not blocked yet */
1138 #endif
1139
1140         p->tgid = p->pid;
1141         if (clone_flags & CLONE_THREAD)
1142                 p->tgid = current->tgid;
1143
1144         if ((retval = security_task_alloc(p)))
1145                 goto bad_fork_cleanup_policy;
1146         if ((retval = audit_alloc(p)))
1147                 goto bad_fork_cleanup_security;
1148         /* copy all the process information */
1149         if ((retval = copy_semundo(clone_flags, p)))
1150                 goto bad_fork_cleanup_audit;
1151         if ((retval = copy_files(clone_flags, p)))
1152                 goto bad_fork_cleanup_semundo;
1153         if ((retval = copy_fs(clone_flags, p)))
1154                 goto bad_fork_cleanup_files;
1155         if ((retval = copy_sighand(clone_flags, p)))
1156                 goto bad_fork_cleanup_fs;
1157         if ((retval = copy_signal(clone_flags, p)))
1158                 goto bad_fork_cleanup_sighand;
1159         if ((retval = copy_mm(clone_flags, p)))
1160                 goto bad_fork_cleanup_signal;
1161         if ((retval = copy_keys(clone_flags, p)))
1162                 goto bad_fork_cleanup_mm;
1163         if ((retval = copy_namespaces(clone_flags, p)))
1164                 goto bad_fork_cleanup_keys;
1165         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1166         if (retval)
1167                 goto bad_fork_cleanup_namespaces;
1168
1169         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1170         /*
1171          * Clear TID on mm_release()?
1172          */
1173         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1174         p->robust_list = NULL;
1175 #ifdef CONFIG_COMPAT
1176         p->compat_robust_list = NULL;
1177 #endif
1178         INIT_LIST_HEAD(&p->pi_state_list);
1179         p->pi_state_cache = NULL;
1180
1181         /*
1182          * sigaltstack should be cleared when sharing the same VM
1183          */
1184         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1185                 p->sas_ss_sp = p->sas_ss_size = 0;
1186
1187         /*
1188          * Syscall tracing should be turned off in the child regardless
1189          * of CLONE_PTRACE.
1190          */
1191         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1192 #ifdef TIF_SYSCALL_EMU
1193         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1194 #endif
1195
1196         /* Our parent execution domain becomes current domain
1197            These must match for thread signalling to apply */
1198         p->parent_exec_id = p->self_exec_id;
1199
1200         /* ok, now we should be set up.. */
1201         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1202         p->pdeath_signal = 0;
1203         p->exit_state = 0;
1204
1205         /*
1206          * Ok, make it visible to the rest of the system.
1207          * We dont wake it up yet.
1208          */
1209         p->group_leader = p;
1210         INIT_LIST_HEAD(&p->thread_group);
1211
1212         /* Perform scheduler related setup. Assign this task to a CPU. */
1213         sched_fork(p, clone_flags);
1214
1215         /* Need tasklist lock for parent etc handling! */
1216         write_lock_irq(&tasklist_lock);
1217
1218         /* for sys_ioprio_set(IOPRIO_WHO_PGRP) */
1219         p->ioprio = current->ioprio;
1220
1221         /*
1222          * The task hasn't been attached yet, so its cpus_allowed mask will
1223          * not be changed, nor will its assigned CPU.
1224          *
1225          * The cpus_allowed mask of the parent may have changed after it was
1226          * copied first time - so re-copy it here, then check the child's CPU
1227          * to ensure it is on a valid CPU (and if not, just force it back to
1228          * parent's CPU). This avoids alot of nasty races.
1229          */
1230         p->cpus_allowed = current->cpus_allowed;
1231         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1232                         !cpu_online(task_cpu(p))))
1233                 set_task_cpu(p, smp_processor_id());
1234
1235         /* CLONE_PARENT re-uses the old parent */
1236         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1237                 p->parent = current->parent;
1238         else
1239                 p->parent = current;
1240
1241         spin_lock(&current->sighand->siglock);
1242
1243         /*
1244          * Process group and session signals need to be delivered to just the
1245          * parent before the fork or both the parent and the child after the
1246          * fork. Restart if a signal comes in before we add the new process to
1247          * it's process group.
1248          * A fatal signal pending means that current will exit, so the new
1249          * thread can't slip out of an OOM kill (or normal SIGKILL).
1250          */
1251         recalc_sigpending();
1252         if (signal_pending(current)) {
1253                 spin_unlock(&current->sighand->siglock);
1254                 write_unlock_irq(&tasklist_lock);
1255                 retval = -ERESTARTNOINTR;
1256                 goto bad_fork_cleanup_namespaces;
1257         }
1258
1259         if (clone_flags & CLONE_THREAD) {
1260                 p->group_leader = current->group_leader;
1261                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1262
1263                 if (!cputime_eq(current->signal->it_virt_expires,
1264                                 cputime_zero) ||
1265                     !cputime_eq(current->signal->it_prof_expires,
1266                                 cputime_zero) ||
1267                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1268                     !list_empty(&current->signal->cpu_timers[0]) ||
1269                     !list_empty(&current->signal->cpu_timers[1]) ||
1270                     !list_empty(&current->signal->cpu_timers[2])) {
1271                         /*
1272                          * Have child wake up on its first tick to check
1273                          * for process CPU timers.
1274                          */
1275                         p->it_prof_expires = jiffies_to_cputime(1);
1276                 }
1277         }
1278
1279         if (likely(p->pid)) {
1280                 add_parent(p);
1281                 tracehook_init_task(p);
1282
1283                 if (thread_group_leader(p)) {
1284                         p->signal->tty = current->signal->tty;
1285                         p->signal->pgrp = process_group(current);
1286                         set_signal_session(p->signal, process_session(current));
1287                         attach_pid(p, PIDTYPE_PGID, process_group(p));
1288                         attach_pid(p, PIDTYPE_SID, process_session(p));
1289
1290                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1291                         __get_cpu_var(process_counts)++;
1292                 }
1293                 attach_pid(p, PIDTYPE_PID, p->pid);
1294                 nr_threads++;
1295         }
1296
1297         total_forks++;
1298         spin_unlock(&current->sighand->siglock);
1299
1300         /* p is copy of current */
1301         vxi = p->vx_info;
1302         if (vxi) {
1303                 claim_vx_info(vxi, p);
1304                 atomic_inc(&vxi->cvirt.nr_threads);
1305                 atomic_inc(&vxi->cvirt.total_forks);
1306                 vx_nproc_inc(p);
1307         }
1308         nxi = p->nx_info;
1309         if (nxi)
1310                 claim_nx_info(nxi, p);
1311         write_unlock_irq(&tasklist_lock);
1312         proc_fork_connector(p);
1313         return p;
1314
1315 bad_fork_cleanup_namespaces:
1316         exit_task_namespaces(p);
1317 bad_fork_cleanup_keys:
1318         exit_keys(p);
1319 bad_fork_cleanup_mm:
1320         if (p->mm)
1321                 mmput(p->mm);
1322 bad_fork_cleanup_signal:
1323         cleanup_signal(p);
1324 bad_fork_cleanup_sighand:
1325         __cleanup_sighand(p->sighand);
1326 bad_fork_cleanup_fs:
1327         exit_fs(p); /* blocking */
1328 bad_fork_cleanup_files:
1329         exit_files(p); /* blocking */
1330 bad_fork_cleanup_semundo:
1331         exit_sem(p);
1332 bad_fork_cleanup_audit:
1333         audit_free(p);
1334 bad_fork_cleanup_security:
1335         security_task_free(p);
1336 bad_fork_cleanup_policy:
1337 #ifdef CONFIG_NUMA
1338         mpol_free(p->mempolicy);
1339 bad_fork_cleanup_cpuset:
1340 #endif
1341         cpuset_exit(p);
1342 bad_fork_cleanup_delays_binfmt:
1343         delayacct_tsk_free(p);
1344         if (p->binfmt)
1345                 module_put(p->binfmt->module);
1346 bad_fork_cleanup_put_domain:
1347         module_put(task_thread_info(p)->exec_domain->module);
1348 bad_fork_cleanup_count:
1349         put_group_info(p->group_info);
1350         atomic_dec(&p->user->processes);
1351         free_uid(p->user);
1352 bad_fork_cleanup_vm:
1353         if (p->mm && !(clone_flags & CLONE_VM))
1354                 vx_pages_sub(p->vx_info, RLIMIT_AS, p->mm->total_vm);
1355 bad_fork_free:
1356         free_task(p);
1357 fork_out:
1358         return ERR_PTR(retval);
1359 }
1360
1361 noinline struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1362 {
1363         memset(regs, 0, sizeof(struct pt_regs));
1364         return regs;
1365 }
1366
1367 struct task_struct * __cpuinit fork_idle(int cpu)
1368 {
1369         struct task_struct *task;
1370         struct pt_regs regs;
1371
1372         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1373         if (!IS_ERR(task))
1374                 init_idle(task, cpu);
1375
1376         return task;
1377 }
1378
1379 /*
1380  *  Ok, this is the main fork-routine.
1381  *
1382  * It copies the process, and if successful kick-starts
1383  * it and waits for it to finish using the VM if required.
1384  */
1385 long do_fork(unsigned long clone_flags,
1386               unsigned long stack_start,
1387               struct pt_regs *regs,
1388               unsigned long stack_size,
1389               int __user *parent_tidptr,
1390               int __user *child_tidptr)
1391 {
1392         struct task_struct *p;
1393         struct pid *pid = alloc_pid();
1394         long nr;
1395
1396         if (!pid)
1397                 return -EAGAIN;
1398
1399         /* kernel threads are host only */
1400         if ((clone_flags & CLONE_KTHREAD) && !vx_check(0, VS_ADMIN)) {
1401                 vxwprintk(1, "xid=%d tried to spawn a kernel thread.",
1402                         vx_current_xid());
1403                 free_pid(pid);
1404                 return -EPERM;
1405         }
1406
1407         nr = pid->nr;
1408
1409         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, nr);
1410         /*
1411          * Do this prior waking up the new thread - the thread pointer
1412          * might get invalid after that point, if the thread exits quickly.
1413          */
1414         if (!IS_ERR(p)) {
1415                 /*
1416                  * When called from kernel_thread, don't do user tracing stuff.
1417                  */
1418                 int is_user = likely(user_mode(regs));
1419                 struct completion vfork;
1420
1421                 if (clone_flags & CLONE_VFORK) {
1422                         p->vfork_done = &vfork;
1423                         init_completion(&vfork);
1424                 }
1425
1426                 if (likely(is_user))
1427                         tracehook_report_clone(clone_flags, p);
1428
1429                 p->flags &= ~PF_STARTING;
1430
1431                 if (clone_flags & CLONE_STOPPED) {
1432                         /*
1433                          * We'll start up with an immediate SIGSTOP.
1434                          */
1435                         sigaddset(&p->pending.signal, SIGSTOP);
1436                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1437                         p->state = TASK_STOPPED;
1438                 }
1439                 else
1440                         wake_up_new_task(p, clone_flags);
1441
1442                 if (likely(is_user))
1443                         tracehook_report_clone_complete(clone_flags, nr, p);
1444
1445                 if (clone_flags & CLONE_VFORK) {
1446                         wait_for_completion(&vfork);
1447                         if (likely(is_user))
1448                                 tracehook_report_vfork_done(p, nr);
1449                 }
1450         } else {
1451                 free_pid(pid);
1452                 nr = PTR_ERR(p);
1453         }
1454         return nr;
1455 }
1456
1457 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1458 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1459 #endif
1460
1461 static void sighand_ctor(void *data, struct kmem_cache *cachep, unsigned long flags)
1462 {
1463         struct sighand_struct *sighand = data;
1464
1465         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
1466                                         SLAB_CTOR_CONSTRUCTOR)
1467                 spin_lock_init(&sighand->siglock);
1468 }
1469
1470 void __init proc_caches_init(void)
1471 {
1472         sighand_cachep = kmem_cache_create("sighand_cache",
1473                         sizeof(struct sighand_struct), 0,
1474                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1475                         sighand_ctor, NULL);
1476         signal_cachep = kmem_cache_create("signal_cache",
1477                         sizeof(struct signal_struct), 0,
1478                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1479         files_cachep = kmem_cache_create("files_cache",
1480                         sizeof(struct files_struct), 0,
1481                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1482         fs_cachep = kmem_cache_create("fs_cache",
1483                         sizeof(struct fs_struct), 0,
1484                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1485         vm_area_cachep = kmem_cache_create("vm_area_struct",
1486                         sizeof(struct vm_area_struct), 0,
1487                         SLAB_PANIC, NULL, NULL);
1488         mm_cachep = kmem_cache_create("mm_struct",
1489                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1490                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1491 }
1492
1493
1494 /*
1495  * Check constraints on flags passed to the unshare system call and
1496  * force unsharing of additional process context as appropriate.
1497  */
1498 static inline void check_unshare_flags(unsigned long *flags_ptr)
1499 {
1500         /*
1501          * If unsharing a thread from a thread group, must also
1502          * unshare vm.
1503          */
1504         if (*flags_ptr & CLONE_THREAD)
1505                 *flags_ptr |= CLONE_VM;
1506
1507         /*
1508          * If unsharing vm, must also unshare signal handlers.
1509          */
1510         if (*flags_ptr & CLONE_VM)
1511                 *flags_ptr |= CLONE_SIGHAND;
1512
1513         /*
1514          * If unsharing signal handlers and the task was created
1515          * using CLONE_THREAD, then must unshare the thread
1516          */
1517         if ((*flags_ptr & CLONE_SIGHAND) &&
1518             (atomic_read(&current->signal->count) > 1))
1519                 *flags_ptr |= CLONE_THREAD;
1520
1521         /*
1522          * If unsharing namespace, must also unshare filesystem information.
1523          */
1524         if (*flags_ptr & CLONE_NEWNS)
1525                 *flags_ptr |= CLONE_FS;
1526 }
1527
1528 /*
1529  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1530  */
1531 static int unshare_thread(unsigned long unshare_flags)
1532 {
1533         if (unshare_flags & CLONE_THREAD)
1534                 return -EINVAL;
1535
1536         return 0;
1537 }
1538
1539 /*
1540  * Unshare the filesystem structure if it is being shared
1541  */
1542 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1543 {
1544         struct fs_struct *fs = current->fs;
1545
1546         if ((unshare_flags & CLONE_FS) &&
1547             (fs && atomic_read(&fs->count) > 1)) {
1548                 *new_fsp = __copy_fs_struct(current->fs);
1549                 if (!*new_fsp)
1550                         return -ENOMEM;
1551         }
1552
1553         return 0;
1554 }
1555
1556 /*
1557  * Unshare the mnt_namespace structure if it is being shared
1558  */
1559 static int unshare_mnt_namespace(unsigned long unshare_flags,
1560                 struct mnt_namespace **new_nsp, struct fs_struct *new_fs)
1561 {
1562         struct mnt_namespace *ns = current->nsproxy->mnt_ns;
1563
1564         if ((unshare_flags & CLONE_NEWNS) && ns) {
1565                 if (!capable(CAP_SYS_ADMIN))
1566                         return -EPERM;
1567
1568                 *new_nsp = dup_mnt_ns(current, new_fs ? new_fs : current->fs);
1569                 if (!*new_nsp)
1570                         return -ENOMEM;
1571         }
1572
1573         return 0;
1574 }
1575
1576 /*
1577  * Unsharing of sighand is not supported yet
1578  */
1579 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1580 {
1581         struct sighand_struct *sigh = current->sighand;
1582
1583         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1584                 return -EINVAL;
1585         else
1586                 return 0;
1587 }
1588
1589 /*
1590  * Unshare vm if it is being shared
1591  */
1592 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1593 {
1594         struct mm_struct *mm = current->mm;
1595
1596         if ((unshare_flags & CLONE_VM) &&
1597             (mm && atomic_read(&mm->mm_users) > 1)) {
1598                 return -EINVAL;
1599         }
1600
1601         return 0;
1602 }
1603
1604 /*
1605  * Unshare file descriptor table if it is being shared
1606  */
1607 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1608 {
1609         struct files_struct *fd = current->files;
1610         int error = 0;
1611
1612         if ((unshare_flags & CLONE_FILES) &&
1613             (fd && atomic_read(&fd->count) > 1)) {
1614                 *new_fdp = dup_fd(fd, &error);
1615                 if (!*new_fdp)
1616                         return error;
1617         }
1618
1619         return 0;
1620 }
1621
1622 /*
1623  * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1624  * supported yet
1625  */
1626 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1627 {
1628         if (unshare_flags & CLONE_SYSVSEM)
1629                 return -EINVAL;
1630
1631         return 0;
1632 }
1633
1634 #ifndef CONFIG_IPC_NS
1635 static inline int unshare_ipcs(unsigned long flags, struct ipc_namespace **ns)
1636 {
1637         if (flags & CLONE_NEWIPC)
1638                 return -EINVAL;
1639
1640         return 0;
1641 }
1642 #endif
1643
1644 /*
1645  * unshare allows a process to 'unshare' part of the process
1646  * context which was originally shared using clone.  copy_*
1647  * functions used by do_fork() cannot be used here directly
1648  * because they modify an inactive task_struct that is being
1649  * constructed. Here we are modifying the current, active,
1650  * task_struct.
1651  */
1652 asmlinkage long sys_unshare(unsigned long unshare_flags)
1653 {
1654         int err = 0;
1655         struct fs_struct *fs, *new_fs = NULL;
1656         struct mnt_namespace *ns, *new_ns = NULL;
1657         struct sighand_struct *new_sigh = NULL;
1658         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1659         struct files_struct *fd, *new_fd = NULL;
1660         struct sem_undo_list *new_ulist = NULL;
1661         struct nsproxy *new_nsproxy = NULL, *old_nsproxy = NULL;
1662         struct uts_namespace *uts, *new_uts = NULL;
1663         struct ipc_namespace *ipc, *new_ipc = NULL;
1664
1665         check_unshare_flags(&unshare_flags);
1666
1667         /* Return -EINVAL for all unsupported flags */
1668         err = -EINVAL;
1669         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1670                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1671                                 CLONE_NEWUTS|CLONE_NEWIPC))
1672                 goto bad_unshare_out;
1673
1674         if ((err = unshare_thread(unshare_flags)))
1675                 goto bad_unshare_out;
1676         if ((err = unshare_fs(unshare_flags, &new_fs)))
1677                 goto bad_unshare_cleanup_thread;
1678         if ((err = unshare_mnt_namespace(unshare_flags, &new_ns, new_fs)))
1679                 goto bad_unshare_cleanup_fs;
1680         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1681                 goto bad_unshare_cleanup_ns;
1682         if ((err = unshare_vm(unshare_flags, &new_mm)))
1683                 goto bad_unshare_cleanup_sigh;
1684         if ((err = unshare_fd(unshare_flags, &new_fd)))
1685                 goto bad_unshare_cleanup_vm;
1686         if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1687                 goto bad_unshare_cleanup_fd;
1688         if ((err = unshare_utsname(unshare_flags, &new_uts)))
1689                 goto bad_unshare_cleanup_semundo;
1690         if ((err = unshare_ipcs(unshare_flags, &new_ipc)))
1691                 goto bad_unshare_cleanup_uts;
1692
1693         if (new_ns || new_uts || new_ipc) {
1694                 old_nsproxy = current->nsproxy;
1695                 new_nsproxy = dup_namespaces(old_nsproxy);
1696                 if (!new_nsproxy) {
1697                         err = -ENOMEM;
1698                         goto bad_unshare_cleanup_ipc;
1699                 }
1700         }
1701
1702         if (new_fs || new_ns || new_mm || new_fd || new_ulist ||
1703                                 new_uts || new_ipc) {
1704
1705                 task_lock(current);
1706
1707                 if (new_nsproxy) {
1708                         current->nsproxy = new_nsproxy;
1709                         new_nsproxy = old_nsproxy;
1710                 }
1711
1712                 if (new_fs) {
1713                         fs = current->fs;
1714                         current->fs = new_fs;
1715                         new_fs = fs;
1716                 }
1717
1718                 if (new_ns) {
1719                         ns = current->nsproxy->mnt_ns;
1720                         current->nsproxy->mnt_ns = new_ns;
1721                         new_ns = ns;
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                 if (new_uts) {
1740                         uts = current->nsproxy->uts_ns;
1741                         current->nsproxy->uts_ns = new_uts;
1742                         new_uts = uts;
1743                 }
1744
1745                 if (new_ipc) {
1746                         ipc = current->nsproxy->ipc_ns;
1747                         current->nsproxy->ipc_ns = new_ipc;
1748                         new_ipc = ipc;
1749                 }
1750
1751                 task_unlock(current);
1752         }
1753
1754         if (new_nsproxy)
1755                 put_nsproxy(new_nsproxy);
1756
1757 bad_unshare_cleanup_ipc:
1758         if (new_ipc)
1759                 put_ipc_ns(new_ipc);
1760
1761 bad_unshare_cleanup_uts:
1762         if (new_uts)
1763                 put_uts_ns(new_uts);
1764
1765 bad_unshare_cleanup_semundo:
1766 bad_unshare_cleanup_fd:
1767         if (new_fd)
1768                 put_files_struct(new_fd);
1769
1770 bad_unshare_cleanup_vm:
1771         if (new_mm)
1772                 mmput(new_mm);
1773
1774 bad_unshare_cleanup_sigh:
1775         if (new_sigh)
1776                 if (atomic_dec_and_test(&new_sigh->count))
1777                         kmem_cache_free(sighand_cachep, new_sigh);
1778
1779 bad_unshare_cleanup_ns:
1780         if (new_ns)
1781                 put_mnt_ns(new_ns);
1782
1783 bad_unshare_cleanup_fs:
1784         if (new_fs)
1785                 put_fs_struct(new_fs);
1786
1787 bad_unshare_cleanup_thread:
1788 bad_unshare_out:
1789         return err;
1790 }