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