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         return tsk;
 180 }
 181
 182 #ifdef CONFIG_MMU
 183 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
 184 {
 185         struct vm_area_struct * mpnt, *tmp, **pprev;
 186         struct rb_node **rb_link, *rb_parent;
 187         int retval;
 188         unsigned long charge;
 189         struct mempolicy *pol;
 190
 191         down_write(&oldmm->mmap_sem);
 192         flush_cache_mm(current->mm);
 193         mm->locked_vm = 0;
 194         mm->mmap = NULL;
 195         mm->mmap_cache = NULL;
 196         mm->free_area_cache = oldmm->mmap_base;
 197         mm->cached_hole_size = ~0UL;
 198         mm->map_count = 0;
 199         set_mm_counter(mm, rss, 0);
 200         set_mm_counter(mm, anon_rss, 0);
 201         cpus_clear(mm->cpu_vm_mask);
 202         mm->mm_rb = RB_ROOT;
 203         rb_link = &mm->mm_rb.rb_node;
 204         rb_parent = NULL;
 205         pprev = &mm->mmap;
 206
 207         for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
 208                 struct file *file;
 209
 210                 if (mpnt->vm_flags & VM_DONTCOPY) {
 211                         __vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 212                                                         -vma_pages(mpnt));
 213                         continue;
 214                 }
 215                 charge = 0;
 216                 if (mpnt->vm_flags & VM_ACCOUNT) {
 217                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 218                         if (security_vm_enough_memory(len))
 219                                 goto fail_nomem;
 220                         charge = len;
 221                 }
 222                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 223                 if (!tmp)
 224                         goto fail_nomem;
 225                 *tmp = *mpnt;
 226                 pol = mpol_copy(vma_policy(mpnt));
 227                 retval = PTR_ERR(pol);
 228                 if (IS_ERR(pol))
 229                         goto fail_nomem_policy;
 230                 vma_set_policy(tmp, pol);
 231                 tmp->vm_flags &= ~VM_LOCKED;
 232                 tmp->vm_mm = mm;
 233                 tmp->vm_next = NULL;
 234                 anon_vma_link(tmp);
 235                 file = tmp->vm_file;
 236                 if (file) {
 237                         struct inode *inode = file->f_dentry->d_inode;
 238                         get_file(file);
 239                         if (tmp->vm_flags & VM_DENYWRITE)
 240                                 atomic_dec(&inode->i_writecount);
 241
 242                         /* insert tmp into the share list, just after mpnt */
 243                         spin_lock(&file->f_mapping->i_mmap_lock);
 244                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 245                         flush_dcache_mmap_lock(file->f_mapping);
 246                         vma_prio_tree_add(tmp, mpnt);
 247                         flush_dcache_mmap_unlock(file->f_mapping);
 248                         spin_unlock(&file->f_mapping->i_mmap_lock);
 249                 }
 250
 251                 /*
 252                  * Link in the new vma and copy the page table entries:
 253                  * link in first so that swapoff can see swap entries.
 254                  * Note that, exceptionally, here the vma is inserted
 255                  * without holding mm->mmap_sem.
 256                  */
 257                 spin_lock(&mm->page_table_lock);
 258                 *pprev = tmp;
 259                 pprev = &tmp->vm_next;
 260
 261                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 262                 rb_link = &tmp->vm_rb.rb_right;
 263                 rb_parent = &tmp->vm_rb;
 264
 265                 mm->map_count++;
 266                 retval = copy_page_range(mm, current->mm, tmp);
 267                 spin_unlock(&mm->page_table_lock);
 268
 269                 if (tmp->vm_ops && tmp->vm_ops->open)
 270                         tmp->vm_ops->open(tmp);
 271
 272                 if (retval)
 273                         goto out;
 274         }
 275         retval = 0;
 276
 277 out:
 278         flush_tlb_mm(current->mm);
 279         up_write(&oldmm->mmap_sem);
 280         return retval;
 281 fail_nomem_policy:
 282         kmem_cache_free(vm_area_cachep, tmp);
 283 fail_nomem:
 284         retval = -ENOMEM;
 285         vm_unacct_memory(charge);
 286         goto out;
 287 }
 288
 289 static inline int mm_alloc_pgd(struct mm_struct * mm)
 290 {
 291         mm->pgd = pgd_alloc(mm);
 292         if (unlikely(!mm->pgd))
 293                 return -ENOMEM;
 294         return 0;
 295 }
 296
 297 static inline void mm_free_pgd(struct mm_struct * mm)
 298 {
 299         pgd_free(mm->pgd);
 300 }
 301 #else
 302 #define dup_mmap(mm, oldmm)     (0)
 303 #define mm_alloc_pgd(mm)        (0)
 304 #define mm_free_pgd(mm)
 305 #endif /* CONFIG_MMU */
 306
 307  __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 308
 309 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
 310 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 311
 312 #include <linux/init_task.h>
 313
 314 static struct mm_struct * mm_init(struct mm_struct * mm)
 315 {
 316         atomic_set(&mm->mm_users, 1);
 317         atomic_set(&mm->mm_count, 1);
 318         init_rwsem(&mm->mmap_sem);
 319         INIT_LIST_HEAD(&mm->mmlist);
 320         mm->core_waiters = 0;
 321         mm->nr_ptes = 0;
 322         spin_lock_init(&mm->page_table_lock);
 323         rwlock_init(&mm->ioctx_list_lock);
 324         mm->ioctx_list = NULL;
 325         mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
 326         mm->free_area_cache = TASK_UNMAPPED_BASE;
 327         mm->cached_hole_size = ~0UL;
 328
 329         if (likely(!mm_alloc_pgd(mm))) {
 330                 mm->def_flags = 0;
 331                 return mm;
 332         }
 333         free_mm(mm);
 334         return NULL;
 335 }
 336
 337 /*
 338  * Allocate and initialize an mm_struct.
 339  */
 340 struct mm_struct * mm_alloc(void)
 341 {
 342         struct mm_struct * mm;
 343
 344         mm = allocate_mm();
 345         if (mm) {
 346                 memset(mm, 0, sizeof(*mm));
 347                 mm = mm_init(mm);
 348         }
 349         return mm;
 350 }
 351
 352 /*
 353  * Called when the last reference to the mm
 354  * is dropped: either by a lazy thread or by
 355  * mmput. Free the page directory and the mm.
 356  */
 357 void fastcall __mmdrop(struct mm_struct *mm)
 358 {
 359         BUG_ON(mm == &init_mm);
 360         mm_free_pgd(mm);
 361         destroy_context(mm);
 362         free_mm(mm);
 363 }
 364
 365 /*
 366  * Decrement the use count and release all resources for an mm.
 367  */
 368 void mmput(struct mm_struct *mm)
 369 {
 370         if (atomic_dec_and_test(&mm->mm_users)) {
 371                 exit_aio(mm);
 372                 exit_mmap(mm);
 373                 if (!list_empty(&mm->mmlist)) {
 374                         spin_lock(&mmlist_lock);
 375                         list_del(&mm->mmlist);
 376                         spin_unlock(&mmlist_lock);
 377                 }
 378                 put_swap_token(mm);
 379                 mmdrop(mm);
 380         }
 381 }
 382 EXPORT_SYMBOL_GPL(mmput);
 383
 384 /**
 385  * get_task_mm - acquire a reference to the task's mm
 386  *
 387  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
 388  * this kernel workthread has transiently adopted a user mm with use_mm,
 389  * to do its AIO) is not set and if so returns a reference to it, after
 390  * bumping up the use count.  User must release the mm via mmput()
 391  * after use.  Typically used by /proc and ptrace.
 392  */
 393 struct mm_struct *get_task_mm(struct task_struct *task)
 394 {
 395         struct mm_struct *mm;
 396
 397         task_lock(task);
 398         mm = task->mm;
 399         if (mm) {
 400                 if (task->flags & PF_BORROWED_MM)
 401                         mm = NULL;
 402                 else
 403                         atomic_inc(&mm->mm_users);
 404         }
 405         task_unlock(task);
 406         return mm;
 407 }
 408 EXPORT_SYMBOL_GPL(get_task_mm);
 409
 410 /* Please note the differences between mmput and mm_release.
 411  * mmput is called whenever we stop holding onto a mm_struct,
 412  * error success whatever.
 413  *
 414  * mm_release is called after a mm_struct has been removed
 415  * from the current process.
 416  *
 417  * This difference is important for error handling, when we
 418  * only half set up a mm_struct for a new process and need to restore
 419  * the old one.  Because we mmput the new mm_struct before
 420  * restoring the old one. . .
 421  * Eric Biederman 10 January 1998
 422  */
 423 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 424 {
 425         struct completion *vfork_done = tsk->vfork_done;
 426
 427         /* Get rid of any cached register state */
 428         deactivate_mm(tsk, mm);
 429
 430         /* notify parent sleeping on vfork() */
 431         if (vfork_done) {
 432                 tsk->vfork_done = NULL;
 433                 complete(vfork_done);
 434         }
 435         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
 436                 u32 __user * tidptr = tsk->clear_child_tid;
 437                 tsk->clear_child_tid = NULL;
 438
 439                 /*
 440                  * We don't check the error code - if userspace has
 441                  * not set up a proper pointer then tough luck.
 442                  */
 443                 put_user(0, tidptr);
 444                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 445         }
 446 }
 447
 448 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 449 {
 450         struct mm_struct * mm, *oldmm;
 451         int retval;
 452
 453         tsk->min_flt = tsk->maj_flt = 0;
 454         tsk->nvcsw = tsk->nivcsw = 0;
 455
 456         tsk->mm = NULL;
 457         tsk->active_mm = NULL;
 458
 459         /*
 460          * Are we cloning a kernel thread?
 461          *
 462          * We need to steal a active VM for that..
 463          */
 464         oldmm = current->mm;
 465         if (!oldmm)
 466                 return 0;
 467
 468         if (clone_flags & CLONE_VM) {
 469                 atomic_inc(&oldmm->mm_users);
 470                 mm = oldmm;
 471                 /*
 472                  * There are cases where the PTL is held to ensure no
 473                  * new threads start up in user mode using an mm, which
 474                  * allows optimizing out ipis; the tlb_gather_mmu code
 475                  * is an example.
 476                  */
 477                 spin_unlock_wait(&oldmm->page_table_lock);
 478                 goto good_mm;
 479         }
 480
 481         retval = -ENOMEM;
 482         mm = allocate_mm();
 483         if (!mm)
 484                 goto fail_nomem;
 485
 486         /* Copy the current MM stuff.. */
 487         memcpy(mm, oldmm, sizeof(*mm));
 488         if (!mm_init(mm))
 489                 goto fail_nomem;
 490
 491         if (init_new_context(tsk,mm))
 492                 goto fail_nocontext;
 493
 494         retval = dup_mmap(mm, oldmm);
 495         if (retval)
 496                 goto free_pt;
 497
 498         mm->hiwater_rss = get_mm_counter(mm,rss);
 499         mm->hiwater_vm = mm->total_vm;
 500
 501 good_mm:
 502         tsk->mm = mm;
 503         tsk->active_mm = mm;
 504         return 0;
 505
 506 free_pt:
 507         mmput(mm);
 508 fail_nomem:
 509         return retval;
 510
 511 fail_nocontext:
 512         /*
 513          * If init_new_context() failed, we cannot use mmput() to free the mm
 514          * because it calls destroy_context()
 515          */
 516         mm_free_pgd(mm);
 517         free_mm(mm);
 518         return retval;
 519 }
 520
 521 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 522 {
 523         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 524         /* We don't need to lock fs - think why ;-) */
 525         if (fs) {
 526                 atomic_set(&fs->count, 1);
 527                 rwlock_init(&fs->lock);
 528                 fs->umask = old->umask;
 529                 read_lock(&old->lock);
 530                 fs->rootmnt = mntget(old->rootmnt);
 531                 fs->root = dget(old->root);
 532                 fs->pwdmnt = mntget(old->pwdmnt);
 533                 fs->pwd = dget(old->pwd);
 534                 if (old->altroot) {
 535                         fs->altrootmnt = mntget(old->altrootmnt);
 536                         fs->altroot = dget(old->altroot);
 537                 } else {
 538                         fs->altrootmnt = NULL;
 539                         fs->altroot = NULL;
 540                 }
 541                 read_unlock(&old->lock);
 542         }
 543         return fs;
 544 }
 545
 546 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 547 {
 548         return __copy_fs_struct(old);
 549 }
 550
 551 EXPORT_SYMBOL_GPL(copy_fs_struct);
 552
 553 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 554 {
 555         if (clone_flags & CLONE_FS) {
 556                 atomic_inc(&current->fs->count);
 557                 return 0;
 558         }
 559         tsk->fs = __copy_fs_struct(current->fs);
 560         if (!tsk->fs)
 561                 return -ENOMEM;
 562         return 0;
 563 }
 564
 565 static int count_open_files(struct files_struct *files, int size)
 566 {
 567         int i;
 568
 569         /* Find the last open fd */
 570         for (i = size/(8*sizeof(long)); i > 0; ) {
 571                 if (files->open_fds->fds_bits[--i])
 572                         break;
 573         }
 574         i = (i+1) * 8 * sizeof(long);
 575         return i;
 576 }
 577
 578 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 579 {
 580         struct files_struct *oldf, *newf;
 581         struct file **old_fds, **new_fds;
 582         int open_files, size, i, error = 0, expand;
 583
 584         /*
 585          * A background process may not have any files ...
 586          */
 587         oldf = current->files;
 588         if (!oldf)
 589                 goto out;
 590
 591         if (clone_flags & CLONE_FILES) {
 592                 atomic_inc(&oldf->count);
 593                 goto out;
 594         }
 595
 596         /*
 597          * Note: we may be using current for both targets (See exec.c)
 598          * This works because we cache current->files (old) as oldf. Don't
 599          * break this.
 600          */
 601         tsk->files = NULL;
 602         error = -ENOMEM;
 603         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
 604         if (!newf)
 605                 goto out;
 606
 607         atomic_set(&newf->count, 1);
 608
 609         spin_lock_init(&newf->file_lock);
 610         newf->next_fd       = 0;
 611         newf->max_fds       = NR_OPEN_DEFAULT;
 612         newf->max_fdset     = __FD_SETSIZE;
 613         newf->close_on_exec = &newf->close_on_exec_init;
 614         newf->open_fds      = &newf->open_fds_init;
 615         newf->fd            = &newf->fd_array[0];
 616
 617         spin_lock(&oldf->file_lock);
 618
 619         open_files = count_open_files(oldf, oldf->max_fdset);
 620         expand = 0;
 621
 622         /*
 623          * Check whether we need to allocate a larger fd array or fd set.
 624          * Note: we're not a clone task, so the open count won't  change.
 625          */
 626         if (open_files > newf->max_fdset) {
 627                 newf->max_fdset = 0;
 628                 expand = 1;
 629         }
 630         if (open_files > newf->max_fds) {
 631                 newf->max_fds = 0;
 632                 expand = 1;
 633         }
 634
 635         /* if the old fdset gets grown now, we'll only copy up to "size" fds */
 636         if (expand) {
 637                 spin_unlock(&oldf->file_lock);
 638                 spin_lock(&newf->file_lock);
 639                 error = expand_files(newf, open_files-1);
 640                 spin_unlock(&newf->file_lock);
 641                 if (error < 0)
 642                         goto out_release;
 643                 spin_lock(&oldf->file_lock);
 644         }
 645
 646         old_fds = oldf->fd;
 647         new_fds = newf->fd;
 648
 649         memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
 650         memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
 651
 652         for (i = open_files; i != 0; i--) {
 653                 struct file *f = *old_fds++;
 654                 if (f) {
 655                         get_file(f);
 656                 } else {
 657                         /*
 658                          * The fd may be claimed in the fd bitmap but not yet
 659                          * instantiated in the files array if a sibling thread
 660                          * is partway through open().  So make sure that this
 661                          * fd is available to the new process.
 662                          */
 663                         FD_CLR(open_files - i, newf->open_fds);
 664                 }
 665                 *new_fds++ = f;
 666         }
 667         spin_unlock(&oldf->file_lock);
 668
 669         /* compute the remainder to be cleared */
 670         size = (newf->max_fds - open_files) * sizeof(struct file *);
 671
 672         /* This is long word aligned thus could use a optimized version */
 673         memset(new_fds, 0, size);
 674
 675         if (newf->max_fdset > open_files) {
 676                 int left = (newf->max_fdset-open_files)/8;
 677                 int start = open_files / (8 * sizeof(unsigned long));
 678
 679                 memset(&newf->open_fds->fds_bits[start], 0, left);
 680                 memset(&newf->close_on_exec->fds_bits[start], 0, left);
 681         }
 682
 683         tsk->files = newf;
 684         error = 0;
 685 out:
 686         return error;
 687
 688 out_release:
 689         free_fdset (newf->close_on_exec, newf->max_fdset);
 690         free_fdset (newf->open_fds, newf->max_fdset);
 691         free_fd_array(newf->fd, newf->max_fds);
 692         kmem_cache_free(files_cachep, newf);
 693         goto out;
 694 }
 695
 696 /*
 697  *      Helper to unshare the files of the current task.
 698  *      We don't want to expose copy_files internals to
 699  *      the exec layer of the kernel.
 700  */
 701
 702 int unshare_files(void)
 703 {
 704         struct files_struct *files  = current->files;
 705         int rc;
 706
 707         if(!files)
 708                 BUG();
 709
 710         /* This can race but the race causes us to copy when we don't
 711            need to and drop the copy */
 712         if(atomic_read(&files->count) == 1)
 713         {
 714                 atomic_inc(&files->count);
 715                 return 0;
 716         }
 717         rc = copy_files(0, current);
 718         if(rc)
 719                 current->files = files;
 720         return rc;
 721 }
 722
 723 EXPORT_SYMBOL(unshare_files);
 724
 725 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 726 {
 727         struct sighand_struct *sig;
 728
 729         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 730                 atomic_inc(&current->sighand->count);
 731                 return 0;
 732         }
 733         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 734         tsk->sighand = sig;
 735         if (!sig)
 736                 return -ENOMEM;
 737         spin_lock_init(&sig->siglock);
 738         atomic_set(&sig->count, 1);
 739         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 740         return 0;
 741 }
 742
 743 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 744 {
 745         struct signal_struct *sig;
 746         int ret;
 747
 748         if (clone_flags & CLONE_THREAD) {
 749                 atomic_inc(&current->signal->count);
 750                 atomic_inc(&current->signal->live);
 751                 return 0;
 752         }
 753         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 754         tsk->signal = sig;
 755         if (!sig)
 756                 return -ENOMEM;
 757
 758         ret = copy_thread_group_keys(tsk);
 759         if (ret < 0) {
 760                 kmem_cache_free(signal_cachep, sig);
 761                 return ret;
 762         }
 763
 764         atomic_set(&sig->count, 1);
 765         atomic_set(&sig->live, 1);
 766         init_waitqueue_head(&sig->wait_chldexit);
 767         sig->flags = 0;
 768         sig->group_exit_code = 0;
 769         sig->group_exit_task = NULL;
 770         sig->group_stop_count = 0;
 771         sig->curr_target = NULL;
 772         init_sigpending(&sig->shared_pending);
 773         INIT_LIST_HEAD(&sig->posix_timers);
 774
 775         sig->it_real_value = sig->it_real_incr = 0;
 776         sig->real_timer.function = it_real_fn;
 777         sig->real_timer.data = (unsigned long) tsk;
 778         init_timer(&sig->real_timer);
 779
 780         sig->it_virt_expires = cputime_zero;
 781         sig->it_virt_incr = cputime_zero;
 782         sig->it_prof_expires = cputime_zero;
 783         sig->it_prof_incr = cputime_zero;
 784
 785         sig->tty = current->signal->tty;
 786         sig->pgrp = process_group(current);
 787         sig->session = current->signal->session;
 788         sig->leader = 0;        /* session leadership doesn't inherit */
 789         sig->tty_old_pgrp = 0;
 790
 791         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 792         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 793         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 794         sig->sched_time = 0;
 795         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 796         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 797         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 798
 799         task_lock(current->group_leader);
 800         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 801         task_unlock(current->group_leader);
 802
 803         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 804                 /*
 805                  * New sole thread in the process gets an expiry time
 806                  * of the whole CPU time limit.
 807                  */
 808                 tsk->it_prof_expires =
 809                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 810         }
 811
 812         return 0;
 813 }
 814
 815 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 816 {
 817         unsigned long new_flags = p->flags;
 818
 819         new_flags &= ~PF_SUPERPRIV;
 820         new_flags |= PF_FORKNOEXEC;
 821         if (!(clone_flags & CLONE_PTRACE))
 822                 p->ptrace = 0;
 823         p->flags = new_flags;
 824 }
 825
 826 asmlinkage long sys_set_tid_address(int __user *tidptr)
 827 {
 828         current->clear_child_tid = tidptr;
 829
 830         return current->pid;
 831 }
 832
 833 /*
 834  * This creates a new process as a copy of the old one,
 835  * but does not actually start it yet.
 836  *
 837  * It copies the registers, and all the appropriate
 838  * parts of the process environment (as per the clone
 839  * flags). The actual kick-off is left to the caller.
 840  */
 841 static task_t *copy_process(unsigned long clone_flags,
 842                                  unsigned long stack_start,
 843                                  struct pt_regs *regs,
 844                                  unsigned long stack_size,
 845                                  int __user *parent_tidptr,
 846                                  int __user *child_tidptr,
 847                                  int pid)
 848 {
 849         int retval;
 850         struct task_struct *p = NULL;
 851
 852         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 853                 return ERR_PTR(-EINVAL);
 854
 855         /*
 856          * Thread groups must share signals as well, and detached threads
 857          * can only be started up within the thread group.
 858          */
 859         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 860                 return ERR_PTR(-EINVAL);
 861
 862         /*
 863          * Shared signal handlers imply shared VM. By way of the above,
 864          * thread groups also imply shared VM. Blocking this case allows
 865          * for various simplifications in other code.
 866          */
 867         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 868                 return ERR_PTR(-EINVAL);
 869
 870         retval = security_task_create(clone_flags);
 871         if (retval)
 872                 goto fork_out;
 873
 874         retval = -ENOMEM;
 875         p = dup_task_struct(current);
 876         if (!p)
 877                 goto fork_out;
 878
 879         retval = -EAGAIN;
 880         if (atomic_read(&p->user->processes) >=
 881                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
 882                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 883                                 p->user != &root_user)
 884                         goto bad_fork_free;
 885         }
 886
 887         atomic_inc(&p->user->__count);
 888         atomic_inc(&p->user->processes);
 889         get_group_info(p->group_info);
 890
 891         /*
 892          * If multiple threads are within copy_process(), then this check
 893          * triggers too late. This doesn't hurt, the check is only there
 894          * to stop root fork bombs.
 895          */
 896         if (nr_threads >= max_threads)
 897                 goto bad_fork_cleanup_count;
 898
 899         if (!try_module_get(p->thread_info->exec_domain->module))
 900                 goto bad_fork_cleanup_count;
 901
 902         if (p->binfmt && !try_module_get(p->binfmt->module))
 903                 goto bad_fork_cleanup_put_domain;
 904
 905         p->did_exec = 0;
 906         copy_flags(clone_flags, p);
 907         p->pid = pid;
 908         retval = -EFAULT;
 909         if (clone_flags & CLONE_PARENT_SETTID)
 910                 if (put_user(p->pid, parent_tidptr))
 911                         goto bad_fork_cleanup;
 912
 913         p->proc_dentry = NULL;
 914
 915         INIT_LIST_HEAD(&p->children);
 916         INIT_LIST_HEAD(&p->sibling);
 917         p->vfork_done = NULL;
 918         spin_lock_init(&p->alloc_lock);
 919         spin_lock_init(&p->proc_lock);
 920
 921         clear_tsk_thread_flag(p, TIF_SIGPENDING);
 922         init_sigpending(&p->pending);
 923
 924         p->utime = cputime_zero;
 925         p->stime = cputime_zero;
 926         p->sched_time = 0;
 927         p->rchar = 0;           /* I/O counter: bytes read */
 928         p->wchar = 0;           /* I/O counter: bytes written */
 929         p->syscr = 0;           /* I/O counter: read syscalls */
 930         p->syscw = 0;           /* I/O counter: write syscalls */
 931         acct_clear_integrals(p);
 932
 933         p->it_virt_expires = cputime_zero;
 934         p->it_prof_expires = cputime_zero;
 935         p->it_sched_expires = 0;
 936         INIT_LIST_HEAD(&p->cpu_timers[0]);
 937         INIT_LIST_HEAD(&p->cpu_timers[1]);
 938         INIT_LIST_HEAD(&p->cpu_timers[2]);
 939
 940         p->lock_depth = -1;             /* -1 = no lock */
 941         do_posix_clock_monotonic_gettime(&p->start_time);
 942         p->security = NULL;
 943         p->io_context = NULL;
 944         p->io_wait = NULL;
 945         p->audit_context = NULL;
 946 #ifdef CONFIG_NUMA
 947         p->mempolicy = mpol_copy(p->mempolicy);
 948         if (IS_ERR(p->mempolicy)) {
 949                 retval = PTR_ERR(p->mempolicy);
 950                 p->mempolicy = NULL;
 951                 goto bad_fork_cleanup;
 952         }
 953 #endif
 954
 955         p->tgid = p->pid;
 956         if (clone_flags & CLONE_THREAD)
 957                 p->tgid = current->tgid;
 958
 959         if ((retval = security_task_alloc(p)))
 960                 goto bad_fork_cleanup_policy;
 961         if ((retval = audit_alloc(p)))
 962                 goto bad_fork_cleanup_security;
 963         /* copy all the process information */
 964         if ((retval = copy_semundo(clone_flags, p)))
 965                 goto bad_fork_cleanup_audit;
 966         if ((retval = copy_files(clone_flags, p)))
 967                 goto bad_fork_cleanup_semundo;
 968         if ((retval = copy_fs(clone_flags, p)))
 969                 goto bad_fork_cleanup_files;
 970         if ((retval = copy_sighand(clone_flags, p)))
 971                 goto bad_fork_cleanup_fs;
 972         if ((retval = copy_signal(clone_flags, p)))
 973                 goto bad_fork_cleanup_sighand;
 974         if ((retval = copy_mm(clone_flags, p)))
 975                 goto bad_fork_cleanup_signal;
 976         if ((retval = copy_keys(clone_flags, p)))
 977                 goto bad_fork_cleanup_mm;
 978         if ((retval = copy_namespace(clone_flags, p)))
 979                 goto bad_fork_cleanup_keys;
 980         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
 981         if (retval)
 982                 goto bad_fork_cleanup_namespace;
 983
 984         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
 985         /*
 986          * Clear TID on mm_release()?
 987          */
 988         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
 989
 990         /*
 991          * Syscall tracing should be turned off in the child regardless
 992          * of CLONE_PTRACE.
 993          */
 994         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
 995
 996         /* Our parent execution domain becomes current domain
 997            These must match for thread signalling to apply */
 998
 999         p->parent_exec_id = p->self_exec_id;
1000
1001         /* ok, now we should be set up.. */
1002         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1003         p->pdeath_signal = 0;
1004         p->exit_state = 0;
1005
1006         /*
1007          * Ok, make it visible to the rest of the system.
1008          * We dont wake it up yet.
1009          */
1010         p->group_leader = p;
1011         INIT_LIST_HEAD(&p->ptrace_children);
1012         INIT_LIST_HEAD(&p->ptrace_list);
1013
1014         /* Perform scheduler related setup. Assign this task to a CPU. */
1015         sched_fork(p, clone_flags);
1016
1017         /* Need tasklist lock for parent etc handling! */
1018         write_lock_irq(&tasklist_lock);
1019
1020         /*
1021          * The task hasn't been attached yet, so its cpus_allowed mask will
1022          * not be changed, nor will its assigned CPU.
1023          *
1024          * The cpus_allowed mask of the parent may have changed after it was
1025          * copied first time - so re-copy it here, then check the child's CPU
1026          * to ensure it is on a valid CPU (and if not, just force it back to
1027          * parent's CPU). This avoids alot of nasty races.
1028          */
1029         p->cpus_allowed = current->cpus_allowed;
1030         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed)))
1031                 set_task_cpu(p, smp_processor_id());
1032
1033         /*
1034          * Check for pending SIGKILL! The new thread should not be allowed
1035          * to slip out of an OOM kill. (or normal SIGKILL.)
1036          */
1037         if (sigismember(&current->pending.signal, SIGKILL)) {
1038                 write_unlock_irq(&tasklist_lock);
1039                 retval = -EINTR;
1040                 goto bad_fork_cleanup_namespace;
1041         }
1042
1043         /* CLONE_PARENT re-uses the old parent */
1044         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1045                 p->real_parent = current->real_parent;
1046         else
1047                 p->real_parent = current;
1048         p->parent = p->real_parent;
1049
1050         if (clone_flags & CLONE_THREAD) {
1051                 spin_lock(&current->sighand->siglock);
1052                 /*
1053                  * Important: if an exit-all has been started then
1054                  * do not create this new thread - the whole thread
1055                  * group is supposed to exit anyway.
1056                  */
1057                 if (current->signal->flags & SIGNAL_GROUP_EXIT) {
1058                         spin_unlock(&current->sighand->siglock);
1059                         write_unlock_irq(&tasklist_lock);
1060                         retval = -EAGAIN;
1061                         goto bad_fork_cleanup_namespace;
1062                 }
1063                 p->group_leader = current->group_leader;
1064
1065                 if (current->signal->group_stop_count > 0) {
1066                         /*
1067                          * There is an all-stop in progress for the group.
1068                          * We ourselves will stop as soon as we check signals.
1069                          * Make the new thread part of that group stop too.
1070                          */
1071                         current->signal->group_stop_count++;
1072                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1073                 }
1074
1075                 if (!cputime_eq(current->signal->it_virt_expires,
1076                                 cputime_zero) ||
1077                     !cputime_eq(current->signal->it_prof_expires,
1078                                 cputime_zero) ||
1079                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1080                     !list_empty(&current->signal->cpu_timers[0]) ||
1081                     !list_empty(&current->signal->cpu_timers[1]) ||
1082                     !list_empty(&current->signal->cpu_timers[2])) {
1083                         /*
1084                          * Have child wake up on its first tick to check
1085                          * for process CPU timers.
1086                          */
1087                         p->it_prof_expires = jiffies_to_cputime(1);
1088                 }
1089
1090                 spin_unlock(&current->sighand->siglock);
1091         }
1092
1093         SET_LINKS(p);
1094         if (unlikely(p->ptrace & PT_PTRACED))
1095                 __ptrace_link(p, current->parent);
1096
1097         cpuset_fork(p);
1098
1099         attach_pid(p, PIDTYPE_PID, p->pid);
1100         attach_pid(p, PIDTYPE_TGID, p->tgid);
1101         if (thread_group_leader(p)) {
1102                 attach_pid(p, PIDTYPE_PGID, process_group(p));
1103                 attach_pid(p, PIDTYPE_SID, p->signal->session);
1104                 if (p->pid)
1105                         __get_cpu_var(process_counts)++;
1106         }
1107
1108         nr_threads++;
1109         total_forks++;
1110         write_unlock_irq(&tasklist_lock);
1111         retval = 0;
1112
1113 fork_out:
1114         if (retval)
1115                 return ERR_PTR(retval);
1116         return p;
1117
1118 bad_fork_cleanup_namespace:
1119         exit_namespace(p);
1120 bad_fork_cleanup_keys:
1121         exit_keys(p);
1122 bad_fork_cleanup_mm:
1123         if (p->mm)
1124                 mmput(p->mm);
1125 bad_fork_cleanup_signal:
1126         exit_signal(p);
1127 bad_fork_cleanup_sighand:
1128         exit_sighand(p);
1129 bad_fork_cleanup_fs:
1130         exit_fs(p); /* blocking */
1131 bad_fork_cleanup_files:
1132         exit_files(p); /* blocking */
1133 bad_fork_cleanup_semundo:
1134         exit_sem(p);
1135 bad_fork_cleanup_audit:
1136         audit_free(p);
1137 bad_fork_cleanup_security:
1138         security_task_free(p);
1139 bad_fork_cleanup_policy:
1140 #ifdef CONFIG_NUMA
1141         mpol_free(p->mempolicy);
1142 #endif
1143 bad_fork_cleanup:
1144         if (p->binfmt)
1145                 module_put(p->binfmt->module);
1146 bad_fork_cleanup_put_domain:
1147         module_put(p->thread_info->exec_domain->module);
1148 bad_fork_cleanup_count:
1149         put_group_info(p->group_info);
1150         atomic_dec(&p->user->processes);
1151         free_uid(p->user);
1152 bad_fork_free:
1153         free_task(p);
1154         goto fork_out;
1155 }
1156
1157 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1158 {
1159         memset(regs, 0, sizeof(struct pt_regs));
1160         return regs;
1161 }
1162
1163 task_t * __devinit fork_idle(int cpu)
1164 {
1165         task_t *task;
1166         struct pt_regs regs;
1167
1168         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1169         if (!task)
1170                 return ERR_PTR(-ENOMEM);
1171         init_idle(task, cpu);
1172         unhash_process(task);
1173         return task;
1174 }
1175
1176 static inline int fork_traceflag (unsigned clone_flags)
1177 {
1178         if (clone_flags & CLONE_UNTRACED)
1179                 return 0;
1180         else if (clone_flags & CLONE_VFORK) {
1181                 if (current->ptrace & PT_TRACE_VFORK)
1182                         return PTRACE_EVENT_VFORK;
1183         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1184                 if (current->ptrace & PT_TRACE_CLONE)
1185                         return PTRACE_EVENT_CLONE;
1186         } else if (current->ptrace & PT_TRACE_FORK)
1187                 return PTRACE_EVENT_FORK;
1188
1189         return 0;
1190 }
1191
1192 /*
1193  *  Ok, this is the main fork-routine.
1194  *
1195  * It copies the process, and if successful kick-starts
1196  * it and waits for it to finish using the VM if required.
1197  */
1198 long do_fork(unsigned long clone_flags,
1199               unsigned long stack_start,
1200               struct pt_regs *regs,
1201               unsigned long stack_size,
1202               int __user *parent_tidptr,
1203               int __user *child_tidptr)
1204 {
1205         struct task_struct *p;
1206         int trace = 0;
1207         long pid = alloc_pidmap();
1208
1209         if (pid < 0)
1210                 return -EAGAIN;
1211         if (unlikely(current->ptrace)) {
1212                 trace = fork_traceflag (clone_flags);
1213                 if (trace)
1214                         clone_flags |= CLONE_PTRACE;
1215         }
1216
1217         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
1218         /*
1219          * Do this prior waking up the new thread - the thread pointer
1220          * might get invalid after that point, if the thread exits quickly.
1221          */
1222         if (!IS_ERR(p)) {
1223                 struct completion vfork;
1224
1225                 if (clone_flags & CLONE_VFORK) {
1226                         p->vfork_done = &vfork;
1227                         init_completion(&vfork);
1228                 }
1229
1230                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1231                         /*
1232                          * We'll start up with an immediate SIGSTOP.
1233                          */
1234                         sigaddset(&p->pending.signal, SIGSTOP);
1235                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1236                 }
1237
1238                 if (!(clone_flags & CLONE_STOPPED))
1239                         wake_up_new_task(p, clone_flags);
1240                 else
1241                         p->state = TASK_STOPPED;
1242
1243                 if (unlikely (trace)) {
1244                         current->ptrace_message = pid;
1245                         ptrace_notify ((trace << 8) | SIGTRAP);
1246                 }
1247
1248                 if (clone_flags & CLONE_VFORK) {
1249                         wait_for_completion(&vfork);
1250                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1251                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1252                 }
1253         } else {
1254                 free_pidmap(pid);
1255                 pid = PTR_ERR(p);
1256         }
1257         return pid;
1258 }
1259
1260 void __init proc_caches_init(void)
1261 {
1262         sighand_cachep = kmem_cache_create("sighand_cache",
1263                         sizeof(struct sighand_struct), 0,
1264                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1265         signal_cachep = kmem_cache_create("signal_cache",
1266                         sizeof(struct signal_struct), 0,
1267                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1268         files_cachep = kmem_cache_create("files_cache",
1269                         sizeof(struct files_struct), 0,
1270                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1271         fs_cachep = kmem_cache_create("fs_cache",
1272                         sizeof(struct fs_struct), 0,
1273                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1274         vm_area_cachep = kmem_cache_create("vm_area_struct",
1275                         sizeof(struct vm_area_struct), 0,
1276                         SLAB_PANIC, NULL, NULL);
1277         mm_cachep = kmem_cache_create("mm_struct",
1278                         sizeof(struct mm_struct), 0,
1279                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1280 }