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/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/ptrace.h>
  39 #include <linux/mount.h>
  40 #include <linux/audit.h>
  41 #include <linux/profile.h>
  42 #include <linux/rmap.h>
  43 #include <linux/acct.h>
  44
  45 #include <asm/pgtable.h>
  46 #include <asm/pgalloc.h>
  47 #include <asm/uaccess.h>
  48 #include <asm/mmu_context.h>
  49 #include <asm/cacheflush.h>
  50 #include <asm/tlbflush.h>
  51
  52 /*
  53  * Protected counters by write_lock_irq(&tasklist_lock)
  54  */
  55 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  56 int nr_threads;                 /* The idle threads do not count.. */
  57
  58 int max_threads;                /* tunable limit on nr_threads */
  59
  60 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  61
  62  __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  63
  64 EXPORT_SYMBOL(tasklist_lock);
  65
  66 int nr_processes(void)
  67 {
  68         int cpu;
  69         int total = 0;
  70
  71         for_each_online_cpu(cpu)
  72                 total += per_cpu(process_counts, cpu);
  73
  74         return total;
  75 }
  76
  77 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  78 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  79 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  80 static kmem_cache_t *task_struct_cachep;
  81 #endif
  82
  83 /* SLAB cache for signal_struct structures (tsk->signal) */
  84 kmem_cache_t *signal_cachep;
  85
  86 /* SLAB cache for sighand_struct structures (tsk->sighand) */
  87 kmem_cache_t *sighand_cachep;
  88
  89 /* SLAB cache for files_struct structures (tsk->files) */
  90 kmem_cache_t *files_cachep;
  91
  92 /* SLAB cache for fs_struct structures (tsk->fs) */
  93 kmem_cache_t *fs_cachep;
  94
  95 /* SLAB cache for vm_area_struct structures */
  96 kmem_cache_t *vm_area_cachep;
  97
  98 /* SLAB cache for mm_struct structures (tsk->mm) */
  99 static kmem_cache_t *mm_cachep;
 100
 101 void free_task(struct task_struct *tsk)
 102 {
 103         free_thread_info(tsk->thread_info);
 104         free_task_struct(tsk);
 105 }
 106 EXPORT_SYMBOL(free_task);
 107
 108 void __put_task_struct(struct task_struct *tsk)
 109 {
 110         WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
 111         WARN_ON(atomic_read(&tsk->usage));
 112         WARN_ON(tsk == current);
 113
 114         if (unlikely(tsk->audit_context))
 115                 audit_free(tsk);
 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         *ti = *orig->thread_info;
 173         *tsk = *orig;
 174         tsk->thread_info = ti;
 175         ti->task = tsk;
 176
 177         /* One for us, one for whoever does the "release_task()" (usually parent) */
 178         atomic_set(&tsk->usage,2);
 179         atomic_set(&tsk->fs_excl, 0);
 180         return tsk;
 181 }
 182
 183 #ifdef CONFIG_MMU
 184 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
 185 {
 186         struct vm_area_struct * mpnt, *tmp, **pprev;
 187         struct rb_node **rb_link, *rb_parent;
 188         int retval;
 189         unsigned long charge;
 190         struct mempolicy *pol;
 191
 192         down_write(&oldmm->mmap_sem);
 193         flush_cache_mm(current->mm);
 194         mm->locked_vm = 0;
 195         mm->mmap = NULL;
 196         mm->mmap_cache = NULL;
 197         mm->free_area_cache = oldmm->mmap_base;
 198         mm->cached_hole_size = ~0UL;
 199         mm->map_count = 0;
 200         set_mm_counter(mm, rss, 0);
 201         set_mm_counter(mm, anon_rss, 0);
 202         cpus_clear(mm->cpu_vm_mask);
 203         mm->mm_rb = RB_ROOT;
 204         rb_link = &mm->mm_rb.rb_node;
 205         rb_parent = NULL;
 206         pprev = &mm->mmap;
 207
 208         for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
 209                 struct file *file;
 210
 211                 if (mpnt->vm_flags & VM_DONTCOPY) {
 212                         long pages = vma_pages(mpnt);
 213                         mm->total_vm -= pages;
 214                         __vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 215                                                                 -pages);
 216                         continue;
 217                 }
 218                 charge = 0;
 219                 if (mpnt->vm_flags & VM_ACCOUNT) {
 220                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 221                         if (security_vm_enough_memory(len))
 222                                 goto fail_nomem;
 223                         charge = len;
 224                 }
 225                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 226                 if (!tmp)
 227                         goto fail_nomem;
 228                 *tmp = *mpnt;
 229                 pol = mpol_copy(vma_policy(mpnt));
 230                 retval = PTR_ERR(pol);
 231                 if (IS_ERR(pol))
 232                         goto fail_nomem_policy;
 233                 vma_set_policy(tmp, pol);
 234                 tmp->vm_flags &= ~VM_LOCKED;
 235                 tmp->vm_mm = mm;
 236                 tmp->vm_next = NULL;
 237                 anon_vma_link(tmp);
 238                 file = tmp->vm_file;
 239                 if (file) {
 240                         struct inode *inode = file->f_dentry->d_inode;
 241                         get_file(file);
 242                         if (tmp->vm_flags & VM_DENYWRITE)
 243                                 atomic_dec(&inode->i_writecount);
 244
 245                         /* insert tmp into the share list, just after mpnt */
 246                         spin_lock(&file->f_mapping->i_mmap_lock);
 247                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 248                         flush_dcache_mmap_lock(file->f_mapping);
 249                         vma_prio_tree_add(tmp, mpnt);
 250                         flush_dcache_mmap_unlock(file->f_mapping);
 251                         spin_unlock(&file->f_mapping->i_mmap_lock);
 252                 }
 253
 254                 /*
 255                  * Link in the new vma and copy the page table entries:
 256                  * link in first so that swapoff can see swap entries.
 257                  * Note that, exceptionally, here the vma is inserted
 258                  * without holding mm->mmap_sem.
 259                  */
 260                 spin_lock(&mm->page_table_lock);
 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, current->mm, tmp);
 270                 spin_unlock(&mm->page_table_lock);
 271
 272                 if (tmp->vm_ops && tmp->vm_ops->open)
 273                         tmp->vm_ops->open(tmp);
 274
 275                 if (retval)
 276                         goto out;
 277         }
 278         retval = 0;
 279
 280 out:
 281         flush_tlb_mm(current->mm);
 282         up_write(&oldmm->mmap_sem);
 283         return retval;
 284 fail_nomem_policy:
 285         kmem_cache_free(vm_area_cachep, tmp);
 286 fail_nomem:
 287         retval = -ENOMEM;
 288         vm_unacct_memory(charge);
 289         goto out;
 290 }
 291
 292 static inline int mm_alloc_pgd(struct mm_struct * mm)
 293 {
 294         mm->pgd = pgd_alloc(mm);
 295         if (unlikely(!mm->pgd))
 296                 return -ENOMEM;
 297         return 0;
 298 }
 299
 300 static inline void mm_free_pgd(struct mm_struct * mm)
 301 {
 302         pgd_free(mm->pgd);
 303 }
 304 #else
 305 #define dup_mmap(mm, oldmm)     (0)
 306 #define mm_alloc_pgd(mm)        (0)
 307 #define mm_free_pgd(mm)
 308 #endif /* CONFIG_MMU */
 309
 310  __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 311
 312 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
 313 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 314
 315 #include <linux/init_task.h>
 316
 317 static struct mm_struct * mm_init(struct mm_struct * mm)
 318 {
 319         atomic_set(&mm->mm_users, 1);
 320         atomic_set(&mm->mm_count, 1);
 321         init_rwsem(&mm->mmap_sem);
 322         INIT_LIST_HEAD(&mm->mmlist);
 323         mm->core_waiters = 0;
 324         mm->nr_ptes = 0;
 325         spin_lock_init(&mm->page_table_lock);
 326         rwlock_init(&mm->ioctx_list_lock);
 327         mm->ioctx_list = NULL;
 328         mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
 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         if (atomic_dec_and_test(&mm->mm_users)) {
 374                 exit_aio(mm);
 375                 exit_mmap(mm);
 376                 if (!list_empty(&mm->mmlist)) {
 377                         spin_lock(&mmlist_lock);
 378                         list_del(&mm->mmlist);
 379                         spin_unlock(&mmlist_lock);
 380                 }
 381                 put_swap_token(mm);
 382                 mmdrop(mm);
 383         }
 384 }
 385 EXPORT_SYMBOL_GPL(mmput);
 386
 387 /**
 388  * get_task_mm - acquire a reference to the task's mm
 389  *
 390  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
 391  * this kernel workthread has transiently adopted a user mm with use_mm,
 392  * to do its AIO) is not set and if so returns a reference to it, after
 393  * bumping up the use count.  User must release the mm via mmput()
 394  * after use.  Typically used by /proc and ptrace.
 395  */
 396 struct mm_struct *get_task_mm(struct task_struct *task)
 397 {
 398         struct mm_struct *mm;
 399
 400         task_lock(task);
 401         mm = task->mm;
 402         if (mm) {
 403                 if (task->flags & PF_BORROWED_MM)
 404                         mm = NULL;
 405                 else
 406                         atomic_inc(&mm->mm_users);
 407         }
 408         task_unlock(task);
 409         return mm;
 410 }
 411 EXPORT_SYMBOL_GPL(get_task_mm);
 412
 413 /* Please note the differences between mmput and mm_release.
 414  * mmput is called whenever we stop holding onto a mm_struct,
 415  * error success whatever.
 416  *
 417  * mm_release is called after a mm_struct has been removed
 418  * from the current process.
 419  *
 420  * This difference is important for error handling, when we
 421  * only half set up a mm_struct for a new process and need to restore
 422  * the old one.  Because we mmput the new mm_struct before
 423  * restoring the old one. . .
 424  * Eric Biederman 10 January 1998
 425  */
 426 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 427 {
 428         struct completion *vfork_done = tsk->vfork_done;
 429
 430         /* Get rid of any cached register state */
 431         deactivate_mm(tsk, mm);
 432
 433         /* notify parent sleeping on vfork() */
 434         if (vfork_done) {
 435                 tsk->vfork_done = NULL;
 436                 complete(vfork_done);
 437         }
 438         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
 439                 u32 __user * tidptr = tsk->clear_child_tid;
 440                 tsk->clear_child_tid = NULL;
 441
 442                 /*
 443                  * We don't check the error code - if userspace has
 444                  * not set up a proper pointer then tough luck.
 445                  */
 446                 put_user(0, tidptr);
 447                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 448         }
 449 }
 450
 451 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 452 {
 453         struct mm_struct * mm, *oldmm;
 454         int retval;
 455
 456         tsk->min_flt = tsk->maj_flt = 0;
 457         tsk->nvcsw = tsk->nivcsw = 0;
 458
 459         tsk->mm = NULL;
 460         tsk->active_mm = NULL;
 461
 462         /*
 463          * Are we cloning a kernel thread?
 464          *
 465          * We need to steal a active VM for that..
 466          */
 467         oldmm = current->mm;
 468         if (!oldmm)
 469                 return 0;
 470
 471         if (clone_flags & CLONE_VM) {
 472                 atomic_inc(&oldmm->mm_users);
 473                 mm = oldmm;
 474                 /*
 475                  * There are cases where the PTL is held to ensure no
 476                  * new threads start up in user mode using an mm, which
 477                  * allows optimizing out ipis; the tlb_gather_mmu code
 478                  * is an example.
 479                  */
 480                 spin_unlock_wait(&oldmm->page_table_lock);
 481                 goto good_mm;
 482         }
 483
 484         retval = -ENOMEM;
 485         mm = allocate_mm();
 486         if (!mm)
 487                 goto fail_nomem;
 488
 489         /* Copy the current MM stuff.. */
 490         memcpy(mm, oldmm, sizeof(*mm));
 491         if (!mm_init(mm))
 492                 goto fail_nomem;
 493
 494         if (init_new_context(tsk,mm))
 495                 goto fail_nocontext;
 496
 497         retval = dup_mmap(mm, oldmm);
 498         if (retval)
 499                 goto free_pt;
 500
 501         mm->hiwater_rss = get_mm_counter(mm,rss);
 502         mm->hiwater_vm = mm->total_vm;
 503
 504 good_mm:
 505         tsk->mm = mm;
 506         tsk->active_mm = mm;
 507         return 0;
 508
 509 free_pt:
 510         mmput(mm);
 511 fail_nomem:
 512         return retval;
 513
 514 fail_nocontext:
 515         /*
 516          * If init_new_context() failed, we cannot use mmput() to free the mm
 517          * because it calls destroy_context()
 518          */
 519         mm_free_pgd(mm);
 520         free_mm(mm);
 521         return retval;
 522 }
 523
 524 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 525 {
 526         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 527         /* We don't need to lock fs - think why ;-) */
 528         if (fs) {
 529                 atomic_set(&fs->count, 1);
 530                 rwlock_init(&fs->lock);
 531                 fs->umask = old->umask;
 532                 read_lock(&old->lock);
 533                 fs->rootmnt = mntget(old->rootmnt);
 534                 fs->root = dget(old->root);
 535                 fs->pwdmnt = mntget(old->pwdmnt);
 536                 fs->pwd = dget(old->pwd);
 537                 if (old->altroot) {
 538                         fs->altrootmnt = mntget(old->altrootmnt);
 539                         fs->altroot = dget(old->altroot);
 540                 } else {
 541                         fs->altrootmnt = NULL;
 542                         fs->altroot = NULL;
 543                 }
 544                 read_unlock(&old->lock);
 545         }
 546         return fs;
 547 }
 548
 549 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 550 {
 551         return __copy_fs_struct(old);
 552 }
 553
 554 EXPORT_SYMBOL_GPL(copy_fs_struct);
 555
 556 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 557 {
 558         if (clone_flags & CLONE_FS) {
 559                 atomic_inc(&current->fs->count);
 560                 return 0;
 561         }
 562         tsk->fs = __copy_fs_struct(current->fs);
 563         if (!tsk->fs)
 564                 return -ENOMEM;
 565         return 0;
 566 }
 567
 568 static int count_open_files(struct files_struct *files, int size)
 569 {
 570         int i;
 571
 572         /* Find the last open fd */
 573         for (i = size/(8*sizeof(long)); i > 0; ) {
 574                 if (files->open_fds->fds_bits[--i])
 575                         break;
 576         }
 577         i = (i+1) * 8 * sizeof(long);
 578         return i;
 579 }
 580
 581 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 582 {
 583         struct files_struct *oldf, *newf;
 584         struct file **old_fds, **new_fds;
 585         int open_files, size, i, error = 0, expand;
 586
 587         /*
 588          * A background process may not have any files ...
 589          */
 590         oldf = current->files;
 591         if (!oldf)
 592                 goto out;
 593
 594         if (clone_flags & CLONE_FILES) {
 595                 atomic_inc(&oldf->count);
 596                 goto out;
 597         }
 598
 599         /*
 600          * Note: we may be using current for both targets (See exec.c)
 601          * This works because we cache current->files (old) as oldf. Don't
 602          * break this.
 603          */
 604         tsk->files = NULL;
 605         error = -ENOMEM;
 606         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
 607         if (!newf)
 608                 goto out;
 609
 610         atomic_set(&newf->count, 1);
 611
 612         spin_lock_init(&newf->file_lock);
 613         newf->next_fd       = 0;
 614         newf->max_fds       = NR_OPEN_DEFAULT;
 615         newf->max_fdset     = __FD_SETSIZE;
 616         newf->close_on_exec = &newf->close_on_exec_init;
 617         newf->open_fds      = &newf->open_fds_init;
 618         newf->fd            = &newf->fd_array[0];
 619
 620         spin_lock(&oldf->file_lock);
 621
 622         open_files = count_open_files(oldf, oldf->max_fdset);
 623         expand = 0;
 624
 625         /*
 626          * Check whether we need to allocate a larger fd array or fd set.
 627          * Note: we're not a clone task, so the open count won't  change.
 628          */
 629         if (open_files > newf->max_fdset) {
 630                 newf->max_fdset = 0;
 631                 expand = 1;
 632         }
 633         if (open_files > newf->max_fds) {
 634                 newf->max_fds = 0;
 635                 expand = 1;
 636         }
 637
 638         /* if the old fdset gets grown now, we'll only copy up to "size" fds */
 639         if (expand) {
 640                 spin_unlock(&oldf->file_lock);
 641                 spin_lock(&newf->file_lock);
 642                 error = expand_files(newf, open_files-1);
 643                 spin_unlock(&newf->file_lock);
 644                 if (error < 0)
 645                         goto out_release;
 646                 spin_lock(&oldf->file_lock);
 647         }
 648
 649         old_fds = oldf->fd;
 650         new_fds = newf->fd;
 651
 652         memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
 653         memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
 654
 655         for (i = open_files; i != 0; i--) {
 656                 struct file *f = *old_fds++;
 657                 if (f) {
 658                         get_file(f);
 659                 } else {
 660                         /*
 661                          * The fd may be claimed in the fd bitmap but not yet
 662                          * instantiated in the files array if a sibling thread
 663                          * is partway through open().  So make sure that this
 664                          * fd is available to the new process.
 665                          */
 666                         FD_CLR(open_files - i, newf->open_fds);
 667                 }
 668                 *new_fds++ = f;
 669         }
 670         spin_unlock(&oldf->file_lock);
 671
 672         /* compute the remainder to be cleared */
 673         size = (newf->max_fds - open_files) * sizeof(struct file *);
 674
 675         /* This is long word aligned thus could use a optimized version */
 676         memset(new_fds, 0, size);
 677
 678         if (newf->max_fdset > open_files) {
 679                 int left = (newf->max_fdset-open_files)/8;
 680                 int start = open_files / (8 * sizeof(unsigned long));
 681
 682                 memset(&newf->open_fds->fds_bits[start], 0, left);
 683                 memset(&newf->close_on_exec->fds_bits[start], 0, left);
 684         }
 685
 686         tsk->files = newf;
 687         error = 0;
 688 out:
 689         return error;
 690
 691 out_release:
 692         free_fdset (newf->close_on_exec, newf->max_fdset);
 693         free_fdset (newf->open_fds, newf->max_fdset);
 694         free_fd_array(newf->fd, newf->max_fds);
 695         kmem_cache_free(files_cachep, newf);
 696         goto out;
 697 }
 698
 699 /*
 700  *      Helper to unshare the files of the current task.
 701  *      We don't want to expose copy_files internals to
 702  *      the exec layer of the kernel.
 703  */
 704
 705 int unshare_files(void)
 706 {
 707         struct files_struct *files  = current->files;
 708         int rc;
 709
 710         if(!files)
 711                 BUG();
 712
 713         /* This can race but the race causes us to copy when we don't
 714            need to and drop the copy */
 715         if(atomic_read(&files->count) == 1)
 716         {
 717                 atomic_inc(&files->count);
 718                 return 0;
 719         }
 720         rc = copy_files(0, current);
 721         if(rc)
 722                 current->files = files;
 723         return rc;
 724 }
 725
 726 EXPORT_SYMBOL(unshare_files);
 727
 728 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 729 {
 730         struct sighand_struct *sig;
 731
 732         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 733                 atomic_inc(&current->sighand->count);
 734                 return 0;
 735         }
 736         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 737         tsk->sighand = sig;
 738         if (!sig)
 739                 return -ENOMEM;
 740         spin_lock_init(&sig->siglock);
 741         atomic_set(&sig->count, 1);
 742         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 743         return 0;
 744 }
 745
 746 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 747 {
 748         struct signal_struct *sig;
 749         int ret;
 750
 751         if (clone_flags & CLONE_THREAD) {
 752                 atomic_inc(&current->signal->count);
 753                 atomic_inc(&current->signal->live);
 754                 return 0;
 755         }
 756         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 757         tsk->signal = sig;
 758         if (!sig)
 759                 return -ENOMEM;
 760
 761         ret = copy_thread_group_keys(tsk);
 762         if (ret < 0) {
 763                 kmem_cache_free(signal_cachep, sig);
 764                 return ret;
 765         }
 766
 767         atomic_set(&sig->count, 1);
 768         atomic_set(&sig->live, 1);
 769         init_waitqueue_head(&sig->wait_chldexit);
 770         sig->flags = 0;
 771         sig->group_exit_code = 0;
 772         sig->group_exit_task = NULL;
 773         sig->group_stop_count = 0;
 774         sig->curr_target = NULL;
 775         init_sigpending(&sig->shared_pending);
 776         INIT_LIST_HEAD(&sig->posix_timers);
 777
 778         sig->it_real_value = sig->it_real_incr = 0;
 779         sig->real_timer.function = it_real_fn;
 780         sig->real_timer.data = (unsigned long) tsk;
 781         init_timer(&sig->real_timer);
 782
 783         sig->it_virt_expires = cputime_zero;
 784         sig->it_virt_incr = cputime_zero;
 785         sig->it_prof_expires = cputime_zero;
 786         sig->it_prof_incr = cputime_zero;
 787
 788         sig->tty = current->signal->tty;
 789         sig->pgrp = process_group(current);
 790         sig->session = current->signal->session;
 791         sig->leader = 0;        /* session leadership doesn't inherit */
 792         sig->tty_old_pgrp = 0;
 793
 794         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 795         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 796         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 797         sig->sched_time = 0;
 798         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 799         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 800         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 801
 802         task_lock(current->group_leader);
 803         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 804         task_unlock(current->group_leader);
 805
 806         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 807                 /*
 808                  * New sole thread in the process gets an expiry time
 809                  * of the whole CPU time limit.
 810                  */
 811                 tsk->it_prof_expires =
 812                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 813         }
 814
 815         return 0;
 816 }
 817
 818 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 819 {
 820         unsigned long new_flags = p->flags;
 821
 822         new_flags &= ~PF_SUPERPRIV;
 823         new_flags |= PF_FORKNOEXEC;
 824         if (!(clone_flags & CLONE_PTRACE))
 825                 p->ptrace = 0;
 826         p->flags = new_flags;
 827 }
 828
 829 asmlinkage long sys_set_tid_address(int __user *tidptr)
 830 {
 831         current->clear_child_tid = tidptr;
 832
 833         return current->pid;
 834 }
 835
 836 /*
 837  * This creates a new process as a copy of the old one,
 838  * but does not actually start it yet.
 839  *
 840  * It copies the registers, and all the appropriate
 841  * parts of the process environment (as per the clone
 842  * flags). The actual kick-off is left to the caller.
 843  */
 844 static task_t *copy_process(unsigned long clone_flags,
 845                                  unsigned long stack_start,
 846                                  struct pt_regs *regs,
 847                                  unsigned long stack_size,
 848                                  int __user *parent_tidptr,
 849                                  int __user *child_tidptr,
 850                                  int pid)
 851 {
 852         int retval;
 853         struct task_struct *p = NULL;
 854
 855         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 856                 return ERR_PTR(-EINVAL);
 857
 858         /*
 859          * Thread groups must share signals as well, and detached threads
 860          * can only be started up within the thread group.
 861          */
 862         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 863                 return ERR_PTR(-EINVAL);
 864
 865         /*
 866          * Shared signal handlers imply shared VM. By way of the above,
 867          * thread groups also imply shared VM. Blocking this case allows
 868          * for various simplifications in other code.
 869          */
 870         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 871                 return ERR_PTR(-EINVAL);
 872
 873         retval = security_task_create(clone_flags);
 874         if (retval)
 875                 goto fork_out;
 876
 877         retval = -ENOMEM;
 878         p = dup_task_struct(current);
 879         if (!p)
 880                 goto fork_out;
 881
 882         retval = -EAGAIN;
 883         if (atomic_read(&p->user->processes) >=
 884                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
 885                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 886                                 p->user != &root_user)
 887                         goto bad_fork_free;
 888         }
 889
 890         atomic_inc(&p->user->__count);
 891         atomic_inc(&p->user->processes);
 892         get_group_info(p->group_info);
 893
 894         /*
 895          * If multiple threads are within copy_process(), then this check
 896          * triggers too late. This doesn't hurt, the check is only there
 897          * to stop root fork bombs.
 898          */
 899         if (nr_threads >= max_threads)
 900                 goto bad_fork_cleanup_count;
 901
 902         if (!try_module_get(p->thread_info->exec_domain->module))
 903                 goto bad_fork_cleanup_count;
 904
 905         if (p->binfmt && !try_module_get(p->binfmt->module))
 906                 goto bad_fork_cleanup_put_domain;
 907
 908         p->did_exec = 0;
 909         copy_flags(clone_flags, p);
 910         p->pid = pid;
 911         retval = -EFAULT;
 912         if (clone_flags & CLONE_PARENT_SETTID)
 913                 if (put_user(p->pid, parent_tidptr))
 914                         goto bad_fork_cleanup;
 915
 916         p->proc_dentry = NULL;
 917
 918         INIT_LIST_HEAD(&p->children);
 919         INIT_LIST_HEAD(&p->sibling);
 920         p->vfork_done = NULL;
 921         spin_lock_init(&p->alloc_lock);
 922         spin_lock_init(&p->proc_lock);
 923
 924         clear_tsk_thread_flag(p, TIF_SIGPENDING);
 925         init_sigpending(&p->pending);
 926
 927         p->utime = cputime_zero;
 928         p->stime = cputime_zero;
 929         p->sched_time = 0;
 930         p->rchar = 0;           /* I/O counter: bytes read */
 931         p->wchar = 0;           /* I/O counter: bytes written */
 932         p->syscr = 0;           /* I/O counter: read syscalls */
 933         p->syscw = 0;           /* I/O counter: write syscalls */
 934         acct_clear_integrals(p);
 935
 936         p->it_virt_expires = cputime_zero;
 937         p->it_prof_expires = cputime_zero;
 938         p->it_sched_expires = 0;
 939         INIT_LIST_HEAD(&p->cpu_timers[0]);
 940         INIT_LIST_HEAD(&p->cpu_timers[1]);
 941         INIT_LIST_HEAD(&p->cpu_timers[2]);
 942
 943         p->lock_depth = -1;             /* -1 = no lock */
 944         do_posix_clock_monotonic_gettime(&p->start_time);
 945         p->security = NULL;
 946         p->io_context = NULL;
 947         p->io_wait = NULL;
 948         p->audit_context = NULL;
 949 #ifdef CONFIG_NUMA
 950         p->mempolicy = mpol_copy(p->mempolicy);
 951         if (IS_ERR(p->mempolicy)) {
 952                 retval = PTR_ERR(p->mempolicy);
 953                 p->mempolicy = NULL;
 954                 goto bad_fork_cleanup;
 955         }
 956 #endif
 957
 958         p->tgid = p->pid;
 959         if (clone_flags & CLONE_THREAD)
 960                 p->tgid = current->tgid;
 961
 962         if ((retval = security_task_alloc(p)))
 963                 goto bad_fork_cleanup_policy;
 964         if ((retval = audit_alloc(p)))
 965                 goto bad_fork_cleanup_security;
 966         /* copy all the process information */
 967         if ((retval = copy_semundo(clone_flags, p)))
 968                 goto bad_fork_cleanup_audit;
 969         if ((retval = copy_files(clone_flags, p)))
 970                 goto bad_fork_cleanup_semundo;
 971         if ((retval = copy_fs(clone_flags, p)))
 972                 goto bad_fork_cleanup_files;
 973         if ((retval = copy_sighand(clone_flags, p)))
 974                 goto bad_fork_cleanup_fs;
 975         if ((retval = copy_signal(clone_flags, p)))
 976                 goto bad_fork_cleanup_sighand;
 977         if ((retval = copy_mm(clone_flags, p)))
 978                 goto bad_fork_cleanup_signal;
 979         if ((retval = copy_keys(clone_flags, p)))
 980                 goto bad_fork_cleanup_mm;
 981         if ((retval = copy_namespace(clone_flags, p)))
 982                 goto bad_fork_cleanup_keys;
 983         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
 984         if (retval)
 985                 goto bad_fork_cleanup_namespace;
 986
 987         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
 988         /*
 989          * Clear TID on mm_release()?
 990          */
 991         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
 992
 993         /*
 994          * Syscall tracing should be turned off in the child regardless
 995          * of CLONE_PTRACE.
 996          */
 997         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
 998 #ifdef TIF_SYSCALL_EMU
 999         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1000 #endif
1001
1002         /* Our parent execution domain becomes current domain
1003            These must match for thread signalling to apply */
1004
1005         p->parent_exec_id = p->self_exec_id;
1006
1007         /* ok, now we should be set up.. */
1008         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1009         p->pdeath_signal = 0;
1010         p->exit_state = 0;
1011
1012         /*
1013          * Ok, make it visible to the rest of the system.
1014          * We dont wake it up yet.
1015          */
1016         p->group_leader = p;
1017         INIT_LIST_HEAD(&p->ptrace_children);
1018         INIT_LIST_HEAD(&p->ptrace_list);
1019
1020         /* Perform scheduler related setup. Assign this task to a CPU. */
1021         sched_fork(p, clone_flags);
1022
1023         /* Need tasklist lock for parent etc handling! */
1024         write_lock_irq(&tasklist_lock);
1025
1026         /*
1027          * The task hasn't been attached yet, so its cpus_allowed mask will
1028          * not be changed, nor will its assigned CPU.
1029          *
1030          * The cpus_allowed mask of the parent may have changed after it was
1031          * copied first time - so re-copy it here, then check the child's CPU
1032          * to ensure it is on a valid CPU (and if not, just force it back to
1033          * parent's CPU). This avoids alot of nasty races.
1034          */
1035         p->cpus_allowed = current->cpus_allowed;
1036         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed)))
1037                 set_task_cpu(p, smp_processor_id());
1038
1039         /*
1040          * Check for pending SIGKILL! The new thread should not be allowed
1041          * to slip out of an OOM kill. (or normal SIGKILL.)
1042          */
1043         if (sigismember(&current->pending.signal, SIGKILL)) {
1044                 write_unlock_irq(&tasklist_lock);
1045                 retval = -EINTR;
1046                 goto bad_fork_cleanup_namespace;
1047         }
1048
1049         /* CLONE_PARENT re-uses the old parent */
1050         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1051                 p->real_parent = current->real_parent;
1052         else
1053                 p->real_parent = current;
1054         p->parent = p->real_parent;
1055
1056         if (clone_flags & CLONE_THREAD) {
1057                 spin_lock(&current->sighand->siglock);
1058                 /*
1059                  * Important: if an exit-all has been started then
1060                  * do not create this new thread - the whole thread
1061                  * group is supposed to exit anyway.
1062                  */
1063                 if (current->signal->flags & SIGNAL_GROUP_EXIT) {
1064                         spin_unlock(&current->sighand->siglock);
1065                         write_unlock_irq(&tasklist_lock);
1066                         retval = -EAGAIN;
1067                         goto bad_fork_cleanup_namespace;
1068                 }
1069                 p->group_leader = current->group_leader;
1070
1071                 if (current->signal->group_stop_count > 0) {
1072                         /*
1073                          * There is an all-stop in progress for the group.
1074                          * We ourselves will stop as soon as we check signals.
1075                          * Make the new thread part of that group stop too.
1076                          */
1077                         current->signal->group_stop_count++;
1078                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1079                 }
1080
1081                 if (!cputime_eq(current->signal->it_virt_expires,
1082                                 cputime_zero) ||
1083                     !cputime_eq(current->signal->it_prof_expires,
1084                                 cputime_zero) ||
1085                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1086                     !list_empty(&current->signal->cpu_timers[0]) ||
1087                     !list_empty(&current->signal->cpu_timers[1]) ||
1088                     !list_empty(&current->signal->cpu_timers[2])) {
1089                         /*
1090                          * Have child wake up on its first tick to check
1091                          * for process CPU timers.
1092                          */
1093                         p->it_prof_expires = jiffies_to_cputime(1);
1094                 }
1095
1096                 spin_unlock(&current->sighand->siglock);
1097         }
1098
1099         /*
1100          * inherit ioprio
1101          */
1102         p->ioprio = current->ioprio;
1103
1104         SET_LINKS(p);
1105         if (unlikely(p->ptrace & PT_PTRACED))
1106                 __ptrace_link(p, current->parent);
1107
1108         cpuset_fork(p);
1109
1110         attach_pid(p, PIDTYPE_PID, p->pid);
1111         attach_pid(p, PIDTYPE_TGID, p->tgid);
1112         if (thread_group_leader(p)) {
1113                 attach_pid(p, PIDTYPE_PGID, process_group(p));
1114                 attach_pid(p, PIDTYPE_SID, p->signal->session);
1115                 if (p->pid)
1116                         __get_cpu_var(process_counts)++;
1117         }
1118
1119         if (!current->signal->tty && p->signal->tty)
1120                 p->signal->tty = NULL;
1121
1122         nr_threads++;
1123         total_forks++;
1124         write_unlock_irq(&tasklist_lock);
1125         retval = 0;
1126
1127 fork_out:
1128         if (retval)
1129                 return ERR_PTR(retval);
1130         return p;
1131
1132 bad_fork_cleanup_namespace:
1133         exit_namespace(p);
1134 bad_fork_cleanup_keys:
1135         exit_keys(p);
1136 bad_fork_cleanup_mm:
1137         if (p->mm)
1138                 mmput(p->mm);
1139 bad_fork_cleanup_signal:
1140         exit_signal(p);
1141 bad_fork_cleanup_sighand:
1142         exit_sighand(p);
1143 bad_fork_cleanup_fs:
1144         exit_fs(p); /* blocking */
1145 bad_fork_cleanup_files:
1146         exit_files(p); /* blocking */
1147 bad_fork_cleanup_semundo:
1148         exit_sem(p);
1149 bad_fork_cleanup_audit:
1150         audit_free(p);
1151 bad_fork_cleanup_security:
1152         security_task_free(p);
1153 bad_fork_cleanup_policy:
1154 #ifdef CONFIG_NUMA
1155         mpol_free(p->mempolicy);
1156 #endif
1157 bad_fork_cleanup:
1158         if (p->binfmt)
1159                 module_put(p->binfmt->module);
1160 bad_fork_cleanup_put_domain:
1161         module_put(p->thread_info->exec_domain->module);
1162 bad_fork_cleanup_count:
1163         put_group_info(p->group_info);
1164         atomic_dec(&p->user->processes);
1165         free_uid(p->user);
1166 bad_fork_free:
1167         free_task(p);
1168         goto fork_out;
1169 }
1170
1171 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1172 {
1173         memset(regs, 0, sizeof(struct pt_regs));
1174         return regs;
1175 }
1176
1177 task_t * __devinit fork_idle(int cpu)
1178 {
1179         task_t *task;
1180         struct pt_regs regs;
1181
1182         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1183         if (!task)
1184                 return ERR_PTR(-ENOMEM);
1185         init_idle(task, cpu);
1186         unhash_process(task);
1187         return task;
1188 }
1189
1190 static inline int fork_traceflag (unsigned clone_flags)
1191 {
1192         if (clone_flags & CLONE_UNTRACED)
1193                 return 0;
1194         else if (clone_flags & CLONE_VFORK) {
1195                 if (current->ptrace & PT_TRACE_VFORK)
1196                         return PTRACE_EVENT_VFORK;
1197         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1198                 if (current->ptrace & PT_TRACE_CLONE)
1199                         return PTRACE_EVENT_CLONE;
1200         } else if (current->ptrace & PT_TRACE_FORK)
1201                 return PTRACE_EVENT_FORK;
1202
1203         return 0;
1204 }
1205
1206 /*
1207  *  Ok, this is the main fork-routine.
1208  *
1209  * It copies the process, and if successful kick-starts
1210  * it and waits for it to finish using the VM if required.
1211  */
1212 long do_fork(unsigned long clone_flags,
1213               unsigned long stack_start,
1214               struct pt_regs *regs,
1215               unsigned long stack_size,
1216               int __user *parent_tidptr,
1217               int __user *child_tidptr)
1218 {
1219         struct task_struct *p;
1220         int trace = 0;
1221         long pid = alloc_pidmap();
1222
1223         if (pid < 0)
1224                 return -EAGAIN;
1225         if (unlikely(current->ptrace)) {
1226                 trace = fork_traceflag (clone_flags);
1227                 if (trace)
1228                         clone_flags |= CLONE_PTRACE;
1229         }
1230
1231         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
1232         /*
1233          * Do this prior waking up the new thread - the thread pointer
1234          * might get invalid after that point, if the thread exits quickly.
1235          */
1236         if (!IS_ERR(p)) {
1237                 struct completion vfork;
1238
1239                 if (clone_flags & CLONE_VFORK) {
1240                         p->vfork_done = &vfork;
1241                         init_completion(&vfork);
1242                 }
1243
1244                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1245                         /*
1246                          * We'll start up with an immediate SIGSTOP.
1247                          */
1248                         sigaddset(&p->pending.signal, SIGSTOP);
1249                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1250                 }
1251
1252                 if (!(clone_flags & CLONE_STOPPED))
1253                         wake_up_new_task(p, clone_flags);
1254                 else
1255                         p->state = TASK_STOPPED;
1256
1257                 if (unlikely (trace)) {
1258                         current->ptrace_message = pid;
1259                         ptrace_notify ((trace << 8) | SIGTRAP);
1260                 }
1261
1262                 if (clone_flags & CLONE_VFORK) {
1263                         wait_for_completion(&vfork);
1264                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1265                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1266                 }
1267         } else {
1268                 free_pidmap(pid);
1269                 pid = PTR_ERR(p);
1270         }
1271         return pid;
1272 }
1273
1274 void __init proc_caches_init(void)
1275 {
1276         sighand_cachep = kmem_cache_create("sighand_cache",
1277                         sizeof(struct sighand_struct), 0,
1278                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1279         signal_cachep = kmem_cache_create("signal_cache",
1280                         sizeof(struct signal_struct), 0,
1281                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1282         files_cachep = kmem_cache_create("files_cache",
1283                         sizeof(struct files_struct), 0,
1284                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1285         fs_cachep = kmem_cache_create("fs_cache",
1286                         sizeof(struct fs_struct), 0,
1287                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1288         vm_area_cachep = kmem_cache_create("vm_area_struct",
1289                         sizeof(struct vm_area_struct), 0,
1290                         SLAB_PANIC, NULL, NULL);
1291         mm_cachep = kmem_cache_create("mm_struct",
1292                         sizeof(struct mm_struct), 0,
1293                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1294 }