kernel/fork.c

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