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/cpu.h>
  32 #include <linux/security.h>
  33 #include <linux/swap.h>
  34 #include <linux/syscalls.h>
  35 #include <linux/jiffies.h>
  36 #include <linux/futex.h>
  37 #include <linux/ptrace.h>
  38 #include <linux/mount.h>
  39 #include <linux/audit.h>
  40 #include <linux/profile.h>
  41 #include <linux/rmap.h>
  42
  43 #include <asm/pgtable.h>
  44 #include <asm/pgalloc.h>
  45 #include <asm/uaccess.h>
  46 #include <asm/mmu_context.h>
  47 #include <asm/cacheflush.h>
  48 #include <asm/tlbflush.h>
  49
  50 /* The idle threads do not count..
  51  * Protected by write_lock_irq(&tasklist_lock)
  52  */
  53 int nr_threads;
  54
  55 int max_threads;
  56 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  57
  58 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  59
  60 rwlock_t tasklist_lock __cacheline_aligned = RW_LOCK_UNLOCKED;  /* outer */
  61
  62 EXPORT_SYMBOL(tasklist_lock);
  63
  64 int nr_processes(void)
  65 {
  66         int cpu;
  67         int total = 0;
  68
  69         for_each_online_cpu(cpu)
  70                 total += per_cpu(process_counts, cpu);
  71
  72         return total;
  73 }
  74
  75 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  76 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  77 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  78 static kmem_cache_t *task_struct_cachep;
  79 #endif
  80
  81 void free_task(struct task_struct *tsk)
  82 {
  83         free_thread_info(tsk->thread_info);
  84         free_task_struct(tsk);
  85 }
  86 EXPORT_SYMBOL(free_task);
  87
  88 void __put_task_struct(struct task_struct *tsk)
  89 {
  90         WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
  91         WARN_ON(atomic_read(&tsk->usage));
  92         WARN_ON(tsk == current);
  93
  94         if (unlikely(tsk->audit_context))
  95                 audit_free(tsk);
  96         security_task_free(tsk);
  97         free_uid(tsk->user);
  98         put_group_info(tsk->group_info);
  99
 100         if (!profile_handoff_task(tsk))
 101                 free_task(tsk);
 102 }
 103
 104 void __init fork_init(unsigned long mempages)
 105 {
 106 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 107 #ifndef ARCH_MIN_TASKALIGN
 108 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 109 #endif
 110         /* create a slab on which task_structs can be allocated */
 111         task_struct_cachep =
 112                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 113                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
 114 #endif
 115
 116         /*
 117          * The default maximum number of threads is set to a safe
 118          * value: the thread structures can take up at most half
 119          * of memory.
 120          */
 121         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 122
 123         /*
 124          * we need to allow at least 20 threads to boot a system
 125          */
 126         if(max_threads < 20)
 127                 max_threads = 20;
 128
 129         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 130         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 131 }
 132
 133 static struct task_struct *dup_task_struct(struct task_struct *orig)
 134 {
 135         struct task_struct *tsk;
 136         struct thread_info *ti;
 137
 138         prepare_to_copy(orig);
 139
 140         tsk = alloc_task_struct();
 141         if (!tsk)
 142                 return NULL;
 143
 144         ti = alloc_thread_info(tsk);
 145         if (!ti) {
 146                 free_task_struct(tsk);
 147                 return NULL;
 148         }
 149
 150         *ti = *orig->thread_info;
 151         *tsk = *orig;
 152         tsk->thread_info = ti;
 153         ti->task = tsk;
 154
 155         /* One for us, one for whoever does the "release_task()" (usually parent) */
 156         atomic_set(&tsk->usage,2);
 157         return tsk;
 158 }
 159
 160 #ifdef CONFIG_MMU
 161 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
 162 {
 163         struct vm_area_struct * mpnt, *tmp, **pprev;
 164         struct rb_node **rb_link, *rb_parent;
 165         int retval;
 166         unsigned long charge;
 167         struct mempolicy *pol;
 168
 169         down_write(&oldmm->mmap_sem);
 170         flush_cache_mm(current->mm);
 171         mm->locked_vm = 0;
 172         mm->mmap = NULL;
 173         mm->mmap_cache = NULL;
 174         mm->free_area_cache = oldmm->mmap_base;
 175         mm->map_count = 0;
 176         mm->rss = 0;
 177         mm->anon_rss = 0;
 178         cpus_clear(mm->cpu_vm_mask);
 179         mm->mm_rb = RB_ROOT;
 180         rb_link = &mm->mm_rb.rb_node;
 181         rb_parent = NULL;
 182         pprev = &mm->mmap;
 183
 184         for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
 185                 struct file *file;
 186
 187                 if (mpnt->vm_flags & VM_DONTCOPY) {
 188                         __vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 189                                                         -vma_pages(mpnt));
 190                         continue;
 191                 }
 192                 charge = 0;
 193                 if (mpnt->vm_flags & VM_ACCOUNT) {
 194                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 195                         if (security_vm_enough_memory(len))
 196                                 goto fail_nomem;
 197                         charge = len;
 198                 }
 199                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 200                 if (!tmp)
 201                         goto fail_nomem;
 202                 *tmp = *mpnt;
 203                 pol = mpol_copy(vma_policy(mpnt));
 204                 retval = PTR_ERR(pol);
 205                 if (IS_ERR(pol))
 206                         goto fail_nomem_policy;
 207                 vma_set_policy(tmp, pol);
 208                 tmp->vm_flags &= ~VM_LOCKED;
 209                 tmp->vm_mm = mm;
 210                 tmp->vm_next = NULL;
 211                 anon_vma_link(tmp);
 212                 file = tmp->vm_file;
 213                 if (file) {
 214                         struct inode *inode = file->f_dentry->d_inode;
 215                         get_file(file);
 216                         if (tmp->vm_flags & VM_DENYWRITE)
 217                                 atomic_dec(&inode->i_writecount);
 218
 219                         /* insert tmp into the share list, just after mpnt */
 220                         spin_lock(&file->f_mapping->i_mmap_lock);
 221                         flush_dcache_mmap_lock(file->f_mapping);
 222                         vma_prio_tree_add(tmp, mpnt);
 223                         flush_dcache_mmap_unlock(file->f_mapping);
 224                         spin_unlock(&file->f_mapping->i_mmap_lock);
 225                 }
 226
 227                 /*
 228                  * Link in the new vma and copy the page table entries:
 229                  * link in first so that swapoff can see swap entries,
 230                  * and try_to_unmap_one's find_vma find the new vma.
 231                  */
 232                 spin_lock(&mm->page_table_lock);
 233                 *pprev = tmp;
 234                 pprev = &tmp->vm_next;
 235
 236                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 237                 rb_link = &tmp->vm_rb.rb_right;
 238                 rb_parent = &tmp->vm_rb;
 239
 240                 mm->map_count++;
 241                 retval = copy_page_range(mm, current->mm, tmp);
 242                 spin_unlock(&mm->page_table_lock);
 243
 244                 if (tmp->vm_ops && tmp->vm_ops->open)
 245                         tmp->vm_ops->open(tmp);
 246
 247                 if (retval)
 248                         goto out;
 249         }
 250         retval = 0;
 251
 252 out:
 253         flush_tlb_mm(current->mm);
 254         up_write(&oldmm->mmap_sem);
 255         return retval;
 256 fail_nomem_policy:
 257         kmem_cache_free(vm_area_cachep, tmp);
 258 fail_nomem:
 259         retval = -ENOMEM;
 260         vm_unacct_memory(charge);
 261         goto out;
 262 }
 263
 264 static inline int mm_alloc_pgd(struct mm_struct * mm)
 265 {
 266         mm->pgd = pgd_alloc(mm);
 267         if (unlikely(!mm->pgd))
 268                 return -ENOMEM;
 269         return 0;
 270 }
 271
 272 static inline void mm_free_pgd(struct mm_struct * mm)
 273 {
 274         pgd_free(mm->pgd);
 275 }
 276 #else
 277 #define dup_mmap(mm, oldmm)     (0)
 278 #define mm_alloc_pgd(mm)        (0)
 279 #define mm_free_pgd(mm)
 280 #endif /* CONFIG_MMU */
 281
 282 spinlock_t mmlist_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
 283
 284 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
 285 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 286
 287 #include <linux/init_task.h>
 288
 289 static struct mm_struct * mm_init(struct mm_struct * mm)
 290 {
 291         atomic_set(&mm->mm_users, 1);
 292         atomic_set(&mm->mm_count, 1);
 293         init_rwsem(&mm->mmap_sem);
 294         INIT_LIST_HEAD(&mm->mmlist);
 295         mm->core_waiters = 0;
 296         mm->nr_ptes = 0;
 297         spin_lock_init(&mm->page_table_lock);
 298         rwlock_init(&mm->ioctx_list_lock);
 299         mm->ioctx_list = NULL;
 300         mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
 301         mm->free_area_cache = TASK_UNMAPPED_BASE;
 302
 303         if (likely(!mm_alloc_pgd(mm))) {
 304                 mm->def_flags = 0;
 305                 return mm;
 306         }
 307         free_mm(mm);
 308         return NULL;
 309 }
 310
 311 /*
 312  * Allocate and initialize an mm_struct.
 313  */
 314 struct mm_struct * mm_alloc(void)
 315 {
 316         struct mm_struct * mm;
 317
 318         mm = allocate_mm();
 319         if (mm) {
 320                 memset(mm, 0, sizeof(*mm));
 321                 mm = mm_init(mm);
 322         }
 323         return mm;
 324 }
 325
 326 /*
 327  * Called when the last reference to the mm
 328  * is dropped: either by a lazy thread or by
 329  * mmput. Free the page directory and the mm.
 330  */
 331 void fastcall __mmdrop(struct mm_struct *mm)
 332 {
 333         BUG_ON(mm == &init_mm);
 334         mm_free_pgd(mm);
 335         destroy_context(mm);
 336         free_mm(mm);
 337 }
 338
 339 /*
 340  * Decrement the use count and release all resources for an mm.
 341  */
 342 void mmput(struct mm_struct *mm)
 343 {
 344         if (atomic_dec_and_test(&mm->mm_users)) {
 345                 exit_aio(mm);
 346                 exit_mmap(mm);
 347                 if (!list_empty(&mm->mmlist)) {
 348                         spin_lock(&mmlist_lock);
 349                         list_del(&mm->mmlist);
 350                         spin_unlock(&mmlist_lock);
 351                 }
 352                 put_swap_token(mm);
 353                 mmdrop(mm);
 354         }
 355 }
 356 EXPORT_SYMBOL_GPL(mmput);
 357
 358 /**
 359  * get_task_mm - acquire a reference to the task's mm
 360  *
 361  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
 362  * this kernel workthread has transiently adopted a user mm with use_mm,
 363  * to do its AIO) is not set and if so returns a reference to it, after
 364  * bumping up the use count.  User must release the mm via mmput()
 365  * after use.  Typically used by /proc and ptrace.
 366  */
 367 struct mm_struct *get_task_mm(struct task_struct *task)
 368 {
 369         struct mm_struct *mm;
 370
 371         task_lock(task);
 372         mm = task->mm;
 373         if (mm) {
 374                 if (task->flags & PF_BORROWED_MM)
 375                         mm = NULL;
 376                 else
 377                         atomic_inc(&mm->mm_users);
 378         }
 379         task_unlock(task);
 380         return mm;
 381 }
 382 EXPORT_SYMBOL_GPL(get_task_mm);
 383
 384 /* Please note the differences between mmput and mm_release.
 385  * mmput is called whenever we stop holding onto a mm_struct,
 386  * error success whatever.
 387  *
 388  * mm_release is called after a mm_struct has been removed
 389  * from the current process.
 390  *
 391  * This difference is important for error handling, when we
 392  * only half set up a mm_struct for a new process and need to restore
 393  * the old one.  Because we mmput the new mm_struct before
 394  * restoring the old one. . .
 395  * Eric Biederman 10 January 1998
 396  */
 397 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 398 {
 399         struct completion *vfork_done = tsk->vfork_done;
 400
 401         /* Get rid of any cached register state */
 402         deactivate_mm(tsk, mm);
 403
 404         /* notify parent sleeping on vfork() */
 405         if (vfork_done) {
 406                 tsk->vfork_done = NULL;
 407                 complete(vfork_done);
 408         }
 409         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
 410                 u32 __user * tidptr = tsk->clear_child_tid;
 411                 tsk->clear_child_tid = NULL;
 412
 413                 /*
 414                  * We don't check the error code - if userspace has
 415                  * not set up a proper pointer then tough luck.
 416                  */
 417                 put_user(0, tidptr);
 418                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 419         }
 420 }
 421
 422 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 423 {
 424         struct mm_struct * mm, *oldmm;
 425         int retval;
 426
 427         tsk->min_flt = tsk->maj_flt = 0;
 428         tsk->nvcsw = tsk->nivcsw = 0;
 429
 430         tsk->mm = NULL;
 431         tsk->active_mm = NULL;
 432
 433         /*
 434          * Are we cloning a kernel thread?
 435          *
 436          * We need to steal a active VM for that..
 437          */
 438         oldmm = current->mm;
 439         if (!oldmm)
 440                 return 0;
 441
 442         if (clone_flags & CLONE_VM) {
 443                 atomic_inc(&oldmm->mm_users);
 444                 mm = oldmm;
 445                 /*
 446                  * There are cases where the PTL is held to ensure no
 447                  * new threads start up in user mode using an mm, which
 448                  * allows optimizing out ipis; the tlb_gather_mmu code
 449                  * is an example.
 450                  */
 451                 spin_unlock_wait(&oldmm->page_table_lock);
 452                 goto good_mm;
 453         }
 454
 455         retval = -ENOMEM;
 456         mm = allocate_mm();
 457         if (!mm)
 458                 goto fail_nomem;
 459
 460         /* Copy the current MM stuff.. */
 461         memcpy(mm, oldmm, sizeof(*mm));
 462         if (!mm_init(mm))
 463                 goto fail_nomem;
 464
 465         if (init_new_context(tsk,mm))
 466                 goto fail_nocontext;
 467
 468         retval = dup_mmap(mm, oldmm);
 469         if (retval)
 470                 goto free_pt;
 471
 472 good_mm:
 473         tsk->mm = mm;
 474         tsk->active_mm = mm;
 475         return 0;
 476
 477 free_pt:
 478         mmput(mm);
 479 fail_nomem:
 480         return retval;
 481
 482 fail_nocontext:
 483         /*
 484          * If init_new_context() failed, we cannot use mmput() to free the mm
 485          * because it calls destroy_context()
 486          */
 487         mm_free_pgd(mm);
 488         free_mm(mm);
 489         return retval;
 490 }
 491
 492 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 493 {
 494         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 495         /* We don't need to lock fs - think why ;-) */
 496         if (fs) {
 497                 atomic_set(&fs->count, 1);
 498                 rwlock_init(&fs->lock);
 499                 fs->umask = old->umask;
 500                 read_lock(&old->lock);
 501                 fs->rootmnt = mntget(old->rootmnt);
 502                 fs->root = dget(old->root);
 503                 fs->pwdmnt = mntget(old->pwdmnt);
 504                 fs->pwd = dget(old->pwd);
 505                 if (old->altroot) {
 506                         fs->altrootmnt = mntget(old->altrootmnt);
 507                         fs->altroot = dget(old->altroot);
 508                 } else {
 509                         fs->altrootmnt = NULL;
 510                         fs->altroot = NULL;
 511                 }
 512                 read_unlock(&old->lock);
 513         }
 514         return fs;
 515 }
 516
 517 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 518 {
 519         return __copy_fs_struct(old);
 520 }
 521
 522 EXPORT_SYMBOL_GPL(copy_fs_struct);
 523
 524 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 525 {
 526         if (clone_flags & CLONE_FS) {
 527                 atomic_inc(&current->fs->count);
 528                 return 0;
 529         }
 530         tsk->fs = __copy_fs_struct(current->fs);
 531         if (!tsk->fs)
 532                 return -ENOMEM;
 533         return 0;
 534 }
 535
 536 static int count_open_files(struct files_struct *files, int size)
 537 {
 538         int i;
 539
 540         /* Find the last open fd */
 541         for (i = size/(8*sizeof(long)); i > 0; ) {
 542                 if (files->open_fds->fds_bits[--i])
 543                         break;
 544         }
 545         i = (i+1) * 8 * sizeof(long);
 546         return i;
 547 }
 548
 549 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 550 {
 551         struct files_struct *oldf, *newf;
 552         struct file **old_fds, **new_fds;
 553         int open_files, nfds, size, i, error = 0;
 554
 555         /*
 556          * A background process may not have any files ...
 557          */
 558         oldf = current->files;
 559         if (!oldf)
 560                 goto out;
 561
 562         if (clone_flags & CLONE_FILES) {
 563                 atomic_inc(&oldf->count);
 564                 goto out;
 565         }
 566
 567         /*
 568          * Note: we may be using current for both targets (See exec.c)
 569          * This works because we cache current->files (old) as oldf. Don't
 570          * break this.
 571          */
 572         tsk->files = NULL;
 573         error = -ENOMEM;
 574         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
 575         if (!newf)
 576                 goto out;
 577
 578         atomic_set(&newf->count, 1);
 579
 580         spin_lock_init(&newf->file_lock);
 581         newf->next_fd       = 0;
 582         newf->max_fds       = NR_OPEN_DEFAULT;
 583         newf->max_fdset     = __FD_SETSIZE;
 584         newf->close_on_exec = &newf->close_on_exec_init;
 585         newf->open_fds      = &newf->open_fds_init;
 586         newf->fd            = &newf->fd_array[0];
 587
 588         /* We don't yet have the oldf readlock, but even if the old
 589            fdset gets grown now, we'll only copy up to "size" fds */
 590         size = oldf->max_fdset;
 591         if (size > __FD_SETSIZE) {
 592                 newf->max_fdset = 0;
 593                 spin_lock(&newf->file_lock);
 594                 error = expand_fdset(newf, size-1);
 595                 spin_unlock(&newf->file_lock);
 596                 if (error)
 597                         goto out_release;
 598         }
 599         spin_lock(&oldf->file_lock);
 600
 601         open_files = count_open_files(oldf, size);
 602
 603         /*
 604          * Check whether we need to allocate a larger fd array.
 605          * Note: we're not a clone task, so the open count won't
 606          * change.
 607          */
 608         nfds = NR_OPEN_DEFAULT;
 609         if (open_files > nfds) {
 610                 spin_unlock(&oldf->file_lock);
 611                 newf->max_fds = 0;
 612                 spin_lock(&newf->file_lock);
 613                 error = expand_fd_array(newf, open_files-1);
 614                 spin_unlock(&newf->file_lock);
 615                 if (error)
 616                         goto out_release;
 617                 nfds = newf->max_fds;
 618                 spin_lock(&oldf->file_lock);
 619         }
 620
 621         old_fds = oldf->fd;
 622         new_fds = newf->fd;
 623
 624         memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
 625         memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
 626
 627         for (i = open_files; i != 0; i--) {
 628                 struct file *f = *old_fds++;
 629                 if (f) {
 630                         get_file(f);
 631                 } else {
 632                         /*
 633                          * The fd may be claimed in the fd bitmap but not yet
 634                          * instantiated in the files array if a sibling thread
 635                          * is partway through open().  So make sure that this
 636                          * fd is available to the new process.
 637                          */
 638                         FD_CLR(open_files - i, newf->open_fds);
 639                 }
 640                 *new_fds++ = f;
 641         }
 642         spin_unlock(&oldf->file_lock);
 643
 644         /* compute the remainder to be cleared */
 645         size = (newf->max_fds - open_files) * sizeof(struct file *);
 646
 647         /* This is long word aligned thus could use a optimized version */
 648         memset(new_fds, 0, size);
 649
 650         if (newf->max_fdset > open_files) {
 651                 int left = (newf->max_fdset-open_files)/8;
 652                 int start = open_files / (8 * sizeof(unsigned long));
 653
 654                 memset(&newf->open_fds->fds_bits[start], 0, left);
 655                 memset(&newf->close_on_exec->fds_bits[start], 0, left);
 656         }
 657
 658         tsk->files = newf;
 659         error = 0;
 660 out:
 661         return error;
 662
 663 out_release:
 664         free_fdset (newf->close_on_exec, newf->max_fdset);
 665         free_fdset (newf->open_fds, newf->max_fdset);
 666         kmem_cache_free(files_cachep, newf);
 667         goto out;
 668 }
 669
 670 /*
 671  *      Helper to unshare the files of the current task.
 672  *      We don't want to expose copy_files internals to
 673  *      the exec layer of the kernel.
 674  */
 675
 676 int unshare_files(void)
 677 {
 678         struct files_struct *files  = current->files;
 679         int rc;
 680
 681         if(!files)
 682                 BUG();
 683
 684         /* This can race but the race causes us to copy when we don't
 685            need to and drop the copy */
 686         if(atomic_read(&files->count) == 1)
 687         {
 688                 atomic_inc(&files->count);
 689                 return 0;
 690         }
 691         rc = copy_files(0, current);
 692         if(rc)
 693                 current->files = files;
 694         return rc;
 695 }
 696
 697 EXPORT_SYMBOL(unshare_files);
 698
 699 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 700 {
 701         struct sighand_struct *sig;
 702
 703         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 704                 atomic_inc(&current->sighand->count);
 705                 return 0;
 706         }
 707         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 708         tsk->sighand = sig;
 709         if (!sig)
 710                 return -ENOMEM;
 711         spin_lock_init(&sig->siglock);
 712         atomic_set(&sig->count, 1);
 713         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 714         return 0;
 715 }
 716
 717 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 718 {
 719         struct signal_struct *sig;
 720
 721         if (clone_flags & CLONE_THREAD) {
 722                 atomic_inc(&current->signal->count);
 723                 atomic_inc(&current->signal->live);
 724                 return 0;
 725         }
 726         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 727         tsk->signal = sig;
 728         if (!sig)
 729                 return -ENOMEM;
 730         atomic_set(&sig->count, 1);
 731         atomic_set(&sig->live, 1);
 732         sig->group_exit = 0;
 733         sig->group_exit_code = 0;
 734         sig->group_exit_task = NULL;
 735         sig->group_stop_count = 0;
 736         sig->stop_state = 0;
 737         sig->curr_target = NULL;
 738         init_sigpending(&sig->shared_pending);
 739         INIT_LIST_HEAD(&sig->posix_timers);
 740
 741         sig->tty = current->signal->tty;
 742         sig->pgrp = process_group(current);
 743         sig->session = current->signal->session;
 744         sig->leader = 0;        /* session leadership doesn't inherit */
 745         sig->tty_old_pgrp = 0;
 746
 747         sig->utime = sig->stime = sig->cutime = sig->cstime = 0;
 748         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 749         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 750
 751         task_lock(current->group_leader);
 752         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 753         task_unlock(current->group_leader);
 754
 755         return 0;
 756 }
 757
 758 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 759 {
 760         unsigned long new_flags = p->flags;
 761
 762         new_flags &= ~PF_SUPERPRIV;
 763         new_flags |= PF_FORKNOEXEC;
 764         if (!(clone_flags & CLONE_PTRACE))
 765                 p->ptrace = 0;
 766         p->flags = new_flags;
 767 }
 768
 769 asmlinkage long sys_set_tid_address(int __user *tidptr)
 770 {
 771         current->clear_child_tid = tidptr;
 772
 773         return current->pid;
 774 }
 775
 776 /*
 777  * This creates a new process as a copy of the old one,
 778  * but does not actually start it yet.
 779  *
 780  * It copies the registers, and all the appropriate
 781  * parts of the process environment (as per the clone
 782  * flags). The actual kick-off is left to the caller.
 783  */
 784 static task_t *copy_process(unsigned long clone_flags,
 785                                  unsigned long stack_start,
 786                                  struct pt_regs *regs,
 787                                  unsigned long stack_size,
 788                                  int __user *parent_tidptr,
 789                                  int __user *child_tidptr,
 790                                  int pid)
 791 {
 792         int retval;
 793         struct task_struct *p = NULL;
 794
 795         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 796                 return ERR_PTR(-EINVAL);
 797
 798         /*
 799          * Thread groups must share signals as well, and detached threads
 800          * can only be started up within the thread group.
 801          */
 802         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 803                 return ERR_PTR(-EINVAL);
 804
 805         /*
 806          * Shared signal handlers imply shared VM. By way of the above,
 807          * thread groups also imply shared VM. Blocking this case allows
 808          * for various simplifications in other code.
 809          */
 810         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 811                 return ERR_PTR(-EINVAL);
 812
 813         retval = security_task_create(clone_flags);
 814         if (retval)
 815                 goto fork_out;
 816
 817         retval = -ENOMEM;
 818         p = dup_task_struct(current);
 819         if (!p)
 820                 goto fork_out;
 821
 822         retval = -EAGAIN;
 823         if (atomic_read(&p->user->processes) >=
 824                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
 825                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 826                                 p->user != &root_user)
 827                         goto bad_fork_free;
 828         }
 829
 830         atomic_inc(&p->user->__count);
 831         atomic_inc(&p->user->processes);
 832         get_group_info(p->group_info);
 833
 834         /*
 835          * If multiple threads are within copy_process(), then this check
 836          * triggers too late. This doesn't hurt, the check is only there
 837          * to stop root fork bombs.
 838          */
 839         if (nr_threads >= max_threads)
 840                 goto bad_fork_cleanup_count;
 841
 842         if (!try_module_get(p->thread_info->exec_domain->module))
 843                 goto bad_fork_cleanup_count;
 844
 845         if (p->binfmt && !try_module_get(p->binfmt->module))
 846                 goto bad_fork_cleanup_put_domain;
 847
 848         p->did_exec = 0;
 849         copy_flags(clone_flags, p);
 850         p->pid = pid;
 851         retval = -EFAULT;
 852         if (clone_flags & CLONE_PARENT_SETTID)
 853                 if (put_user(p->pid, parent_tidptr))
 854                         goto bad_fork_cleanup;
 855
 856         p->proc_dentry = NULL;
 857
 858         INIT_LIST_HEAD(&p->children);
 859         INIT_LIST_HEAD(&p->sibling);
 860         init_waitqueue_head(&p->wait_chldexit);
 861         p->vfork_done = NULL;
 862         spin_lock_init(&p->alloc_lock);
 863         spin_lock_init(&p->proc_lock);
 864
 865         clear_tsk_thread_flag(p, TIF_SIGPENDING);
 866         init_sigpending(&p->pending);
 867
 868         p->it_real_value = p->it_virt_value = p->it_prof_value = 0;
 869         p->it_real_incr = p->it_virt_incr = p->it_prof_incr = 0;
 870         init_timer(&p->real_timer);
 871         p->real_timer.data = (unsigned long) p;
 872
 873         p->utime = p->stime = 0;
 874         p->lock_depth = -1;             /* -1 = no lock */
 875         do_posix_clock_monotonic_gettime(&p->start_time);
 876         p->security = NULL;
 877         p->io_context = NULL;
 878         p->io_wait = NULL;
 879         p->audit_context = NULL;
 880 #ifdef CONFIG_NUMA
 881         p->mempolicy = mpol_copy(p->mempolicy);
 882         if (IS_ERR(p->mempolicy)) {
 883                 retval = PTR_ERR(p->mempolicy);
 884                 p->mempolicy = NULL;
 885                 goto bad_fork_cleanup;
 886         }
 887 #endif
 888
 889         p->tgid = p->pid;
 890         if (clone_flags & CLONE_THREAD)
 891                 p->tgid = current->tgid;
 892
 893         if ((retval = security_task_alloc(p)))
 894                 goto bad_fork_cleanup_policy;
 895         if ((retval = audit_alloc(p)))
 896                 goto bad_fork_cleanup_security;
 897         /* copy all the process information */
 898         if ((retval = copy_semundo(clone_flags, p)))
 899                 goto bad_fork_cleanup_audit;
 900         if ((retval = copy_files(clone_flags, p)))
 901                 goto bad_fork_cleanup_semundo;
 902         if ((retval = copy_fs(clone_flags, p)))
 903                 goto bad_fork_cleanup_files;
 904         if ((retval = copy_sighand(clone_flags, p)))
 905                 goto bad_fork_cleanup_fs;
 906         if ((retval = copy_signal(clone_flags, p)))
 907                 goto bad_fork_cleanup_sighand;
 908         if ((retval = copy_mm(clone_flags, p)))
 909                 goto bad_fork_cleanup_signal;
 910         if ((retval = copy_keys(clone_flags, p)))
 911                 goto bad_fork_cleanup_mm;
 912         if ((retval = copy_namespace(clone_flags, p)))
 913                 goto bad_fork_cleanup_keys;
 914         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
 915         if (retval)
 916                 goto bad_fork_cleanup_namespace;
 917
 918         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
 919         /*
 920          * Clear TID on mm_release()?
 921          */
 922         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
 923
 924         /*
 925          * Syscall tracing should be turned off in the child regardless
 926          * of CLONE_PTRACE.
 927          */
 928         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
 929
 930         /* Our parent execution domain becomes current domain
 931            These must match for thread signalling to apply */
 932
 933         p->parent_exec_id = p->self_exec_id;
 934
 935         /* ok, now we should be set up.. */
 936         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
 937         p->pdeath_signal = 0;
 938         p->exit_state = 0;
 939
 940         /* Perform scheduler related setup */
 941         sched_fork(p);
 942
 943         /*
 944          * Ok, make it visible to the rest of the system.
 945          * We dont wake it up yet.
 946          */
 947         p->group_leader = p;
 948         INIT_LIST_HEAD(&p->ptrace_children);
 949         INIT_LIST_HEAD(&p->ptrace_list);
 950
 951         /* Need tasklist lock for parent etc handling! */
 952         write_lock_irq(&tasklist_lock);
 953
 954         /*
 955          * The task hasn't been attached yet, so cpus_allowed mask cannot
 956          * have changed. The cpus_allowed mask of the parent may have
 957          * changed after it was copied first time, and it may then move to
 958          * another CPU - so we re-copy it here and set the child's CPU to
 959          * the parent's CPU. This avoids alot of nasty races.
 960          */
 961         p->cpus_allowed = current->cpus_allowed;
 962         set_task_cpu(p, smp_processor_id());
 963
 964         /*
 965          * Check for pending SIGKILL! The new thread should not be allowed
 966          * to slip out of an OOM kill. (or normal SIGKILL.)
 967          */
 968         if (sigismember(&current->pending.signal, SIGKILL)) {
 969                 write_unlock_irq(&tasklist_lock);
 970                 retval = -EINTR;
 971                 goto bad_fork_cleanup_namespace;
 972         }
 973
 974         /* CLONE_PARENT re-uses the old parent */
 975         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
 976                 p->real_parent = current->real_parent;
 977         else
 978                 p->real_parent = current;
 979         p->parent = p->real_parent;
 980
 981         if (clone_flags & CLONE_THREAD) {
 982                 spin_lock(&current->sighand->siglock);
 983                 /*
 984                  * Important: if an exit-all has been started then
 985                  * do not create this new thread - the whole thread
 986                  * group is supposed to exit anyway.
 987                  */
 988                 if (current->signal->group_exit) {
 989                         spin_unlock(&current->sighand->siglock);
 990                         write_unlock_irq(&tasklist_lock);
 991                         retval = -EAGAIN;
 992                         goto bad_fork_cleanup_namespace;
 993                 }
 994                 p->group_leader = current->group_leader;
 995
 996                 if (current->signal->group_stop_count > 0) {
 997                         /*
 998                          * There is an all-stop in progress for the group.
 999                          * We ourselves will stop as soon as we check signals.
1000                          * Make the new thread part of that group stop too.
1001                          */
1002                         current->signal->group_stop_count++;
1003                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1004                 }
1005
1006                 spin_unlock(&current->sighand->siglock);
1007         }
1008
1009         SET_LINKS(p);
1010         if (unlikely(p->ptrace & PT_PTRACED))
1011                 __ptrace_link(p, current->parent);
1012
1013         attach_pid(p, PIDTYPE_PID, p->pid);
1014         attach_pid(p, PIDTYPE_TGID, p->tgid);
1015         if (thread_group_leader(p)) {
1016                 attach_pid(p, PIDTYPE_PGID, process_group(p));
1017                 attach_pid(p, PIDTYPE_SID, p->signal->session);
1018                 if (p->pid)
1019                         __get_cpu_var(process_counts)++;
1020         }
1021
1022         nr_threads++;
1023         write_unlock_irq(&tasklist_lock);
1024         retval = 0;
1025
1026 fork_out:
1027         if (retval)
1028                 return ERR_PTR(retval);
1029         return p;
1030
1031 bad_fork_cleanup_namespace:
1032         exit_namespace(p);
1033 bad_fork_cleanup_keys:
1034         exit_keys(p);
1035 bad_fork_cleanup_mm:
1036         if (p->mm)
1037                 mmput(p->mm);
1038 bad_fork_cleanup_signal:
1039         exit_signal(p);
1040 bad_fork_cleanup_sighand:
1041         exit_sighand(p);
1042 bad_fork_cleanup_fs:
1043         exit_fs(p); /* blocking */
1044 bad_fork_cleanup_files:
1045         exit_files(p); /* blocking */
1046 bad_fork_cleanup_semundo:
1047         exit_sem(p);
1048 bad_fork_cleanup_audit:
1049         audit_free(p);
1050 bad_fork_cleanup_security:
1051         security_task_free(p);
1052 bad_fork_cleanup_policy:
1053 #ifdef CONFIG_NUMA
1054         mpol_free(p->mempolicy);
1055 #endif
1056 bad_fork_cleanup:
1057         if (p->binfmt)
1058                 module_put(p->binfmt->module);
1059 bad_fork_cleanup_put_domain:
1060         module_put(p->thread_info->exec_domain->module);
1061 bad_fork_cleanup_count:
1062         put_group_info(p->group_info);
1063         atomic_dec(&p->user->processes);
1064         free_uid(p->user);
1065 bad_fork_free:
1066         free_task(p);
1067         goto fork_out;
1068 }
1069
1070 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1071 {
1072         memset(regs, 0, sizeof(struct pt_regs));
1073         return regs;
1074 }
1075
1076 task_t * __devinit fork_idle(int cpu)
1077 {
1078         task_t *task;
1079         struct pt_regs regs;
1080
1081         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1082         if (!task)
1083                 return ERR_PTR(-ENOMEM);
1084         init_idle(task, cpu);
1085         unhash_process(task);
1086         return task;
1087 }
1088
1089 static inline int fork_traceflag (unsigned clone_flags)
1090 {
1091         if (clone_flags & CLONE_UNTRACED)
1092                 return 0;
1093         else if (clone_flags & CLONE_VFORK) {
1094                 if (current->ptrace & PT_TRACE_VFORK)
1095                         return PTRACE_EVENT_VFORK;
1096         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1097                 if (current->ptrace & PT_TRACE_CLONE)
1098                         return PTRACE_EVENT_CLONE;
1099         } else if (current->ptrace & PT_TRACE_FORK)
1100                 return PTRACE_EVENT_FORK;
1101
1102         return 0;
1103 }
1104
1105 /*
1106  *  Ok, this is the main fork-routine.
1107  *
1108  * It copies the process, and if successful kick-starts
1109  * it and waits for it to finish using the VM if required.
1110  */
1111 long do_fork(unsigned long clone_flags,
1112               unsigned long stack_start,
1113               struct pt_regs *regs,
1114               unsigned long stack_size,
1115               int __user *parent_tidptr,
1116               int __user *child_tidptr)
1117 {
1118         struct task_struct *p;
1119         int trace = 0;
1120         long pid = alloc_pidmap();
1121
1122         if (pid < 0)
1123                 return -EAGAIN;
1124         if (unlikely(current->ptrace)) {
1125                 trace = fork_traceflag (clone_flags);
1126                 if (trace)
1127                         clone_flags |= CLONE_PTRACE;
1128         }
1129
1130         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
1131         /*
1132          * Do this prior waking up the new thread - the thread pointer
1133          * might get invalid after that point, if the thread exits quickly.
1134          */
1135         if (!IS_ERR(p)) {
1136                 struct completion vfork;
1137
1138                 if (clone_flags & CLONE_VFORK) {
1139                         p->vfork_done = &vfork;
1140                         init_completion(&vfork);
1141                 }
1142
1143                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1144                         /*
1145                          * We'll start up with an immediate SIGSTOP.
1146                          */
1147                         sigaddset(&p->pending.signal, SIGSTOP);
1148                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1149                 }
1150
1151                 if (!(clone_flags & CLONE_STOPPED))
1152                         wake_up_new_task(p, clone_flags);
1153                 else
1154                         p->state = TASK_STOPPED;
1155                 ++total_forks;
1156
1157                 if (unlikely (trace)) {
1158                         current->ptrace_message = pid;
1159                         ptrace_notify ((trace << 8) | SIGTRAP);
1160                 }
1161
1162                 if (clone_flags & CLONE_VFORK) {
1163                         wait_for_completion(&vfork);
1164                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1165                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1166                 }
1167         } else {
1168                 free_pidmap(pid);
1169                 pid = PTR_ERR(p);
1170         }
1171         return pid;
1172 }
1173
1174 /* SLAB cache for signal_struct structures (tsk->signal) */
1175 kmem_cache_t *signal_cachep;
1176
1177 /* SLAB cache for sighand_struct structures (tsk->sighand) */
1178 kmem_cache_t *sighand_cachep;
1179
1180 /* SLAB cache for files_struct structures (tsk->files) */
1181 kmem_cache_t *files_cachep;
1182
1183 /* SLAB cache for fs_struct structures (tsk->fs) */
1184 kmem_cache_t *fs_cachep;
1185
1186 /* SLAB cache for vm_area_struct structures */
1187 kmem_cache_t *vm_area_cachep;
1188
1189 /* SLAB cache for mm_struct structures (tsk->mm) */
1190 kmem_cache_t *mm_cachep;
1191
1192 void __init proc_caches_init(void)
1193 {
1194         sighand_cachep = kmem_cache_create("sighand_cache",
1195                         sizeof(struct sighand_struct), 0,
1196                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1197         signal_cachep = kmem_cache_create("signal_cache",
1198                         sizeof(struct signal_struct), 0,
1199                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1200         files_cachep = kmem_cache_create("files_cache",
1201                         sizeof(struct files_struct), 0,
1202                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1203         fs_cachep = kmem_cache_create("fs_cache",
1204                         sizeof(struct fs_struct), 0,
1205                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1206         vm_area_cachep = kmem_cache_create("vm_area_struct",
1207                         sizeof(struct vm_area_struct), 0,
1208                         SLAB_PANIC, NULL, NULL);
1209         mm_cachep = kmem_cache_create("mm_struct",
1210                         sizeof(struct mm_struct), 0,
1211                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1212 }