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