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/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/fs.h>
  30 #include <linux/cpu.h>
  31 #include <linux/security.h>
  32 #include <linux/syscalls.h>
  33 #include <linux/jiffies.h>
  34 #include <linux/futex.h>
  35 #include <linux/ptrace.h>
  36 #include <linux/mount.h>
  37 #include <linux/audit.h>
  38 #include <linux/rmap.h>
  39
  40 #include <asm/pgtable.h>
  41 #include <asm/pgalloc.h>
  42 #include <asm/uaccess.h>
  43 #include <asm/mmu_context.h>
  44 #include <asm/cacheflush.h>
  45 #include <asm/tlbflush.h>
  46
  47 /* The idle threads do not count..
  48  * Protected by write_lock_irq(&tasklist_lock)
  49  */
  50 int nr_threads;
  51
  52 int max_threads;
  53 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  54
  55 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  56
  57 rwlock_t tasklist_lock __cacheline_aligned = RW_LOCK_UNLOCKED;  /* outer */
  58
  59 EXPORT_SYMBOL(tasklist_lock);
  60
  61 int nr_processes(void)
  62 {
  63         int cpu;
  64         int total = 0;
  65
  66         for_each_online_cpu(cpu)
  67                 total += per_cpu(process_counts, cpu);
  68
  69         return total;
  70 }
  71
  72 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  73 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  74 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  75 static kmem_cache_t *task_struct_cachep;
  76 #endif
  77
  78 static void free_task(struct task_struct *tsk)
  79 {
  80         free_thread_info(tsk->thread_info);
  81         free_task_struct(tsk);
  82 }
  83
  84 void __put_task_struct(struct task_struct *tsk)
  85 {
  86         WARN_ON(!(tsk->state & (TASK_DEAD | TASK_ZOMBIE)));
  87         WARN_ON(atomic_read(&tsk->usage));
  88         WARN_ON(tsk == current);
  89
  90         if (unlikely(tsk->audit_context))
  91                 audit_free(tsk);
  92         security_task_free(tsk);
  93         free_uid(tsk->user);
  94         put_group_info(tsk->group_info);
  95         free_task(tsk);
  96 }
  97
  98 void fastcall add_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
  99 {
 100         unsigned long flags;
 101
 102         wait->flags &= ~WQ_FLAG_EXCLUSIVE;
 103         spin_lock_irqsave(&q->lock, flags);
 104         __add_wait_queue(q, wait);
 105         spin_unlock_irqrestore(&q->lock, flags);
 106 }
 107
 108 EXPORT_SYMBOL(add_wait_queue);
 109
 110 void fastcall add_wait_queue_exclusive(wait_queue_head_t *q, wait_queue_t * wait)
 111 {
 112         unsigned long flags;
 113
 114         wait->flags |= WQ_FLAG_EXCLUSIVE;
 115         spin_lock_irqsave(&q->lock, flags);
 116         __add_wait_queue_tail(q, wait);
 117         spin_unlock_irqrestore(&q->lock, flags);
 118 }
 119
 120 EXPORT_SYMBOL(add_wait_queue_exclusive);
 121
 122 void fastcall remove_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
 123 {
 124         unsigned long flags;
 125
 126         spin_lock_irqsave(&q->lock, flags);
 127         __remove_wait_queue(q, wait);
 128         spin_unlock_irqrestore(&q->lock, flags);
 129 }
 130
 131 EXPORT_SYMBOL(remove_wait_queue);
 132
 133
 134 /*
 135  * Note: we use "set_current_state()" _after_ the wait-queue add,
 136  * because we need a memory barrier there on SMP, so that any
 137  * wake-function that tests for the wait-queue being active
 138  * will be guaranteed to see waitqueue addition _or_ subsequent
 139  * tests in this thread will see the wakeup having taken place.
 140  *
 141  * The spin_unlock() itself is semi-permeable and only protects
 142  * one way (it only protects stuff inside the critical region and
 143  * stops them from bleeding out - it would still allow subsequent
 144  * loads to move into the the critical region).
 145  */
 146 void fastcall prepare_to_wait(wait_queue_head_t *q, wait_queue_t *wait, int state)
 147 {
 148         unsigned long flags;
 149
 150         wait->flags &= ~WQ_FLAG_EXCLUSIVE;
 151         spin_lock_irqsave(&q->lock, flags);
 152         if (list_empty(&wait->task_list))
 153                 __add_wait_queue(q, wait);
 154         set_current_state(state);
 155         spin_unlock_irqrestore(&q->lock, flags);
 156 }
 157
 158 EXPORT_SYMBOL(prepare_to_wait);
 159
 160 void fastcall
 161 prepare_to_wait_exclusive(wait_queue_head_t *q, wait_queue_t *wait, int state)
 162 {
 163         unsigned long flags;
 164
 165         wait->flags |= WQ_FLAG_EXCLUSIVE;
 166         spin_lock_irqsave(&q->lock, flags);
 167         if (list_empty(&wait->task_list))
 168                 __add_wait_queue_tail(q, wait);
 169         set_current_state(state);
 170         spin_unlock_irqrestore(&q->lock, flags);
 171 }
 172
 173 EXPORT_SYMBOL(prepare_to_wait_exclusive);
 174
 175 void fastcall finish_wait(wait_queue_head_t *q, wait_queue_t *wait)
 176 {
 177         unsigned long flags;
 178
 179         __set_current_state(TASK_RUNNING);
 180         /*
 181          * We can check for list emptiness outside the lock
 182          * IFF:
 183          *  - we use the "careful" check that verifies both
 184          *    the next and prev pointers, so that there cannot
 185          *    be any half-pending updates in progress on other
 186          *    CPU's that we haven't seen yet (and that might
 187          *    still change the stack area.
 188          * and
 189          *  - all other users take the lock (ie we can only
 190          *    have _one_ other CPU that looks at or modifies
 191          *    the list).
 192          */
 193         if (!list_empty_careful(&wait->task_list)) {
 194                 spin_lock_irqsave(&q->lock, flags);
 195                 list_del_init(&wait->task_list);
 196                 spin_unlock_irqrestore(&q->lock, flags);
 197         }
 198 }
 199
 200 EXPORT_SYMBOL(finish_wait);
 201
 202 int autoremove_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
 203 {
 204         int ret = default_wake_function(wait, mode, sync, key);
 205
 206         if (ret)
 207                 list_del_init(&wait->task_list);
 208         return ret;
 209 }
 210
 211 EXPORT_SYMBOL(autoremove_wake_function);
 212
 213 void __init fork_init(unsigned long mempages)
 214 {
 215 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 216 #ifndef ARCH_MIN_TASKALIGN
 217 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 218 #endif
 219         /* create a slab on which task_structs can be allocated */
 220         task_struct_cachep =
 221                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 222                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
 223 #endif
 224
 225         /*
 226          * The default maximum number of threads is set to a safe
 227          * value: the thread structures can take up at most half
 228          * of memory.
 229          */
 230         max_threads = mempages / (THREAD_SIZE/PAGE_SIZE) / 8;
 231         /*
 232          * we need to allow at least 20 threads to boot a system
 233          */
 234         if(max_threads < 20)
 235                 max_threads = 20;
 236
 237         init_task.rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 238         init_task.rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 239 }
 240
 241 static struct task_struct *dup_task_struct(struct task_struct *orig)
 242 {
 243         struct task_struct *tsk;
 244         struct thread_info *ti;
 245
 246         prepare_to_copy(orig);
 247
 248         tsk = alloc_task_struct();
 249         if (!tsk)
 250                 return NULL;
 251
 252         ti = alloc_thread_info(tsk);
 253         if (!ti) {
 254                 free_task_struct(tsk);
 255                 return NULL;
 256         }
 257
 258         *ti = *orig->thread_info;
 259         *tsk = *orig;
 260         tsk->thread_info = ti;
 261         ti->task = tsk;
 262
 263         /* One for us, one for whoever does the "release_task()" (usually parent) */
 264         atomic_set(&tsk->usage,2);
 265         return tsk;
 266 }
 267
 268 #ifdef CONFIG_MMU
 269 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
 270 {
 271         struct vm_area_struct * mpnt, *tmp, **pprev;
 272         struct rb_node **rb_link, *rb_parent;
 273         int retval;
 274         unsigned long charge = 0;
 275         struct mempolicy *pol;
 276
 277         down_write(&oldmm->mmap_sem);
 278         flush_cache_mm(current->mm);
 279         mm->locked_vm = 0;
 280         mm->mmap = NULL;
 281         mm->mmap_cache = NULL;
 282         mm->free_area_cache = TASK_UNMAPPED_BASE;
 283         mm->map_count = 0;
 284         mm->rss = 0;
 285         cpus_clear(mm->cpu_vm_mask);
 286         mm->mm_rb = RB_ROOT;
 287         rb_link = &mm->mm_rb.rb_node;
 288         rb_parent = NULL;
 289         pprev = &mm->mmap;
 290
 291         /*
 292          * Add it to the mmlist after the parent.
 293          * Doing it this way means that we can order the list,
 294          * and fork() won't mess up the ordering significantly.
 295          * Add it first so that swapoff can see any swap entries.
 296          */
 297         spin_lock(&mmlist_lock);
 298         list_add(&mm->mmlist, &current->mm->mmlist);
 299         mmlist_nr++;
 300         spin_unlock(&mmlist_lock);
 301
 302         for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
 303                 struct file *file;
 304
 305                 if(mpnt->vm_flags & VM_DONTCOPY)
 306                         continue;
 307                 if (mpnt->vm_flags & VM_ACCOUNT) {
 308                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 309                         if (security_vm_enough_memory(len))
 310                                 goto fail_nomem;
 311                         charge += len;
 312                 }
 313                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 314                 if (!tmp)
 315                         goto fail_nomem;
 316                 *tmp = *mpnt;
 317                 pol = mpol_copy(vma_policy(mpnt));
 318                 retval = PTR_ERR(pol);
 319                 if (IS_ERR(pol))
 320                         goto fail_nomem_policy;
 321                 vma_set_policy(tmp, pol);
 322                 tmp->vm_flags &= ~VM_LOCKED;
 323                 tmp->vm_mm = mm;
 324                 tmp->vm_next = NULL;
 325                 anon_vma_link(tmp);
 326                 vma_prio_tree_init(tmp);
 327                 file = tmp->vm_file;
 328                 if (file) {
 329                         struct inode *inode = file->f_dentry->d_inode;
 330                         get_file(file);
 331                         if (tmp->vm_flags & VM_DENYWRITE)
 332                                 atomic_dec(&inode->i_writecount);
 333
 334                         /* insert tmp into the share list, just after mpnt */
 335                         spin_lock(&file->f_mapping->i_mmap_lock);
 336                         flush_dcache_mmap_lock(file->f_mapping);
 337                         vma_prio_tree_add(tmp, mpnt);
 338                         flush_dcache_mmap_unlock(file->f_mapping);
 339                         spin_unlock(&file->f_mapping->i_mmap_lock);
 340                 }
 341
 342                 /*
 343                  * Link in the new vma and copy the page table entries:
 344                  * link in first so that swapoff can see swap entries,
 345                  * and try_to_unmap_one's find_vma find the new vma.
 346                  */
 347                 spin_lock(&mm->page_table_lock);
 348                 *pprev = tmp;
 349                 pprev = &tmp->vm_next;
 350
 351                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 352                 rb_link = &tmp->vm_rb.rb_right;
 353                 rb_parent = &tmp->vm_rb;
 354
 355                 mm->map_count++;
 356                 retval = copy_page_range(mm, current->mm, tmp);
 357                 spin_unlock(&mm->page_table_lock);
 358
 359                 if (tmp->vm_ops && tmp->vm_ops->open)
 360                         tmp->vm_ops->open(tmp);
 361
 362                 if (retval)
 363                         goto fail;
 364         }
 365         retval = 0;
 366
 367 out:
 368         flush_tlb_mm(current->mm);
 369         up_write(&oldmm->mmap_sem);
 370         return retval;
 371 fail_nomem_policy:
 372         kmem_cache_free(vm_area_cachep, tmp);
 373 fail_nomem:
 374         retval = -ENOMEM;
 375 fail:
 376         vm_unacct_memory(charge);
 377         goto out;
 378 }
 379 static inline int mm_alloc_pgd(struct mm_struct * mm)
 380 {
 381         mm->pgd = pgd_alloc(mm);
 382         if (unlikely(!mm->pgd))
 383                 return -ENOMEM;
 384         return 0;
 385 }
 386
 387 static inline void mm_free_pgd(struct mm_struct * mm)
 388 {
 389         pgd_free(mm->pgd);
 390 }
 391 #else
 392 #define dup_mmap(mm, oldmm)     (0)
 393 #define mm_alloc_pgd(mm)        (0)
 394 #define mm_free_pgd(mm)
 395 #endif /* CONFIG_MMU */
 396
 397 spinlock_t mmlist_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
 398 int mmlist_nr;
 399
 400 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
 401 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 402
 403 #include <linux/init_task.h>
 404
 405 static struct mm_struct * mm_init(struct mm_struct * mm)
 406 {
 407         atomic_set(&mm->mm_users, 1);
 408         atomic_set(&mm->mm_count, 1);
 409         init_rwsem(&mm->mmap_sem);
 410         mm->core_waiters = 0;
 411         mm->page_table_lock = SPIN_LOCK_UNLOCKED;
 412         mm->ioctx_list_lock = RW_LOCK_UNLOCKED;
 413         mm->ioctx_list = NULL;
 414         mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
 415         mm->free_area_cache = TASK_UNMAPPED_BASE;
 416
 417         if (likely(!mm_alloc_pgd(mm))) {
 418                 mm->def_flags = 0;
 419 #ifdef __HAVE_ARCH_MMAP_TOP
 420                 mm->mmap_top = mmap_top();
 421 #endif
 422                 return mm;
 423         }
 424         free_mm(mm);
 425         return NULL;
 426 }
 427
 428 /*
 429  * Allocate and initialize an mm_struct.
 430  */
 431 struct mm_struct * mm_alloc(void)
 432 {
 433         struct mm_struct * mm;
 434
 435         mm = allocate_mm();
 436         if (mm) {
 437                 memset(mm, 0, sizeof(*mm));
 438                 mm = mm_init(mm);
 439         }
 440         return mm;
 441 }
 442
 443 /*
 444  * Called when the last reference to the mm
 445  * is dropped: either by a lazy thread or by
 446  * mmput. Free the page directory and the mm.
 447  */
 448 void fastcall __mmdrop(struct mm_struct *mm)
 449 {
 450         BUG_ON(mm == &init_mm);
 451         mm_free_pgd(mm);
 452         destroy_context(mm);
 453         free_mm(mm);
 454 }
 455
 456 /*
 457  * Decrement the use count and release all resources for an mm.
 458  */
 459 void mmput(struct mm_struct *mm)
 460 {
 461         if (atomic_dec_and_lock(&mm->mm_users, &mmlist_lock)) {
 462                 list_del(&mm->mmlist);
 463                 mmlist_nr--;
 464                 spin_unlock(&mmlist_lock);
 465                 exit_aio(mm);
 466                 exit_mmap(mm);
 467                 mmdrop(mm);
 468         }
 469 }
 470
 471 /*
 472  * Checks if the use count of an mm is non-zero and if so
 473  * returns a reference to it after bumping up the use count.
 474  * If the use count is zero, it means this mm is going away,
 475  * so return NULL.
 476  */
 477 struct mm_struct *mmgrab(struct mm_struct *mm)
 478 {
 479         spin_lock(&mmlist_lock);
 480         if (!atomic_read(&mm->mm_users))
 481                 mm = NULL;
 482         else
 483                 atomic_inc(&mm->mm_users);
 484         spin_unlock(&mmlist_lock);
 485         return mm;
 486 }
 487
 488 /* Please note the differences between mmput and mm_release.
 489  * mmput is called whenever we stop holding onto a mm_struct,
 490  * error success whatever.
 491  *
 492  * mm_release is called after a mm_struct has been removed
 493  * from the current process.
 494  *
 495  * This difference is important for error handling, when we
 496  * only half set up a mm_struct for a new process and need to restore
 497  * the old one.  Because we mmput the new mm_struct before
 498  * restoring the old one. . .
 499  * Eric Biederman 10 January 1998
 500  */
 501 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 502 {
 503         struct completion *vfork_done = tsk->vfork_done;
 504
 505         /* Get rid of any cached register state */
 506         deactivate_mm(tsk, mm);
 507
 508         /* notify parent sleeping on vfork() */
 509         if (vfork_done) {
 510                 tsk->vfork_done = NULL;
 511                 complete(vfork_done);
 512         }
 513         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
 514                 u32 __user * tidptr = tsk->clear_child_tid;
 515                 tsk->clear_child_tid = NULL;
 516
 517                 /*
 518                  * We don't check the error code - if userspace has
 519                  * not set up a proper pointer then tough luck.
 520                  */
 521                 put_user(0, tidptr);
 522                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL);
 523         }
 524 }
 525
 526 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 527 {
 528         struct mm_struct * mm, *oldmm;
 529         int retval;
 530
 531         tsk->min_flt = tsk->maj_flt = 0;
 532         tsk->cmin_flt = tsk->cmaj_flt = 0;
 533         tsk->nvcsw = tsk->nivcsw = tsk->cnvcsw = tsk->cnivcsw = 0;
 534
 535         tsk->mm = NULL;
 536         tsk->active_mm = NULL;
 537
 538         /*
 539          * Are we cloning a kernel thread?
 540          *
 541          * We need to steal a active VM for that..
 542          */
 543         oldmm = current->mm;
 544         if (!oldmm)
 545                 return 0;
 546
 547         if (clone_flags & CLONE_VM) {
 548                 atomic_inc(&oldmm->mm_users);
 549                 mm = oldmm;
 550                 /*
 551                  * There are cases where the PTL is held to ensure no
 552                  * new threads start up in user mode using an mm, which
 553                  * allows optimizing out ipis; the tlb_gather_mmu code
 554                  * is an example.
 555                  */
 556                 spin_unlock_wait(&oldmm->page_table_lock);
 557                 goto good_mm;
 558         }
 559
 560         retval = -ENOMEM;
 561         mm = allocate_mm();
 562         if (!mm)
 563                 goto fail_nomem;
 564
 565         /* Copy the current MM stuff.. */
 566         memcpy(mm, oldmm, sizeof(*mm));
 567         if (!mm_init(mm))
 568                 goto fail_nomem;
 569
 570         if (init_new_context(tsk,mm))
 571                 goto fail_nocontext;
 572
 573         retval = dup_mmap(mm, oldmm);
 574         if (retval)
 575                 goto free_pt;
 576
 577 good_mm:
 578         tsk->mm = mm;
 579         tsk->active_mm = mm;
 580         return 0;
 581
 582 free_pt:
 583         mmput(mm);
 584 fail_nomem:
 585         return retval;
 586
 587 fail_nocontext:
 588         /*
 589          * If init_new_context() failed, we cannot use mmput() to free the mm
 590          * because it calls destroy_context()
 591          */
 592         mm_free_pgd(mm);
 593         free_mm(mm);
 594         return retval;
 595 }
 596
 597 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 598 {
 599         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 600         /* We don't need to lock fs - think why ;-) */
 601         if (fs) {
 602                 atomic_set(&fs->count, 1);
 603                 fs->lock = RW_LOCK_UNLOCKED;
 604                 fs->umask = old->umask;
 605                 read_lock(&old->lock);
 606                 fs->rootmnt = mntget(old->rootmnt);
 607                 fs->root = dget(old->root);
 608                 fs->pwdmnt = mntget(old->pwdmnt);
 609                 fs->pwd = dget(old->pwd);
 610                 if (old->altroot) {
 611                         fs->altrootmnt = mntget(old->altrootmnt);
 612                         fs->altroot = dget(old->altroot);
 613                 } else {
 614                         fs->altrootmnt = NULL;
 615                         fs->altroot = NULL;
 616                 }
 617                 read_unlock(&old->lock);
 618         }
 619         return fs;
 620 }
 621
 622 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 623 {
 624         return __copy_fs_struct(old);
 625 }
 626
 627 EXPORT_SYMBOL_GPL(copy_fs_struct);
 628
 629 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 630 {
 631         if (clone_flags & CLONE_FS) {
 632                 atomic_inc(&current->fs->count);
 633                 return 0;
 634         }
 635         tsk->fs = __copy_fs_struct(current->fs);
 636         if (!tsk->fs)
 637                 return -ENOMEM;
 638         return 0;
 639 }
 640
 641 static int count_open_files(struct files_struct *files, int size)
 642 {
 643         int i;
 644
 645         /* Find the last open fd */
 646         for (i = size/(8*sizeof(long)); i > 0; ) {
 647                 if (files->open_fds->fds_bits[--i])
 648                         break;
 649         }
 650         i = (i+1) * 8 * sizeof(long);
 651         return i;
 652 }
 653
 654 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 655 {
 656         struct files_struct *oldf, *newf;
 657         struct file **old_fds, **new_fds;
 658         int open_files, nfds, size, i, error = 0;
 659
 660         /*
 661          * A background process may not have any files ...
 662          */
 663         oldf = current->files;
 664         if (!oldf)
 665                 goto out;
 666
 667         if (clone_flags & CLONE_FILES) {
 668                 atomic_inc(&oldf->count);
 669                 goto out;
 670         }
 671
 672         /*
 673          * Note: we may be using current for both targets (See exec.c)
 674          * This works because we cache current->files (old) as oldf. Don't
 675          * break this.
 676          */
 677         tsk->files = NULL;
 678         error = -ENOMEM;
 679         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
 680         if (!newf)
 681                 goto out;
 682
 683         atomic_set(&newf->count, 1);
 684
 685         newf->file_lock     = SPIN_LOCK_UNLOCKED;
 686         newf->next_fd       = 0;
 687         newf->max_fds       = NR_OPEN_DEFAULT;
 688         newf->max_fdset     = __FD_SETSIZE;
 689         newf->close_on_exec = &newf->close_on_exec_init;
 690         newf->open_fds      = &newf->open_fds_init;
 691         newf->fd            = &newf->fd_array[0];
 692
 693         /* We don't yet have the oldf readlock, but even if the old
 694            fdset gets grown now, we'll only copy up to "size" fds */
 695         size = oldf->max_fdset;
 696         if (size > __FD_SETSIZE) {
 697                 newf->max_fdset = 0;
 698                 spin_lock(&newf->file_lock);
 699                 error = expand_fdset(newf, size-1);
 700                 spin_unlock(&newf->file_lock);
 701                 if (error)
 702                         goto out_release;
 703         }
 704         spin_lock(&oldf->file_lock);
 705
 706         open_files = count_open_files(oldf, size);
 707
 708         /*
 709          * Check whether we need to allocate a larger fd array.
 710          * Note: we're not a clone task, so the open count won't
 711          * change.
 712          */
 713         nfds = NR_OPEN_DEFAULT;
 714         if (open_files > nfds) {
 715                 spin_unlock(&oldf->file_lock);
 716                 newf->max_fds = 0;
 717                 spin_lock(&newf->file_lock);
 718                 error = expand_fd_array(newf, open_files-1);
 719                 spin_unlock(&newf->file_lock);
 720                 if (error)
 721                         goto out_release;
 722                 nfds = newf->max_fds;
 723                 spin_lock(&oldf->file_lock);
 724         }
 725
 726         old_fds = oldf->fd;
 727         new_fds = newf->fd;
 728
 729         memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
 730         memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
 731
 732         for (i = open_files; i != 0; i--) {
 733                 struct file *f = *old_fds++;
 734                 if (f)
 735                         get_file(f);
 736                 *new_fds++ = f;
 737         }
 738         spin_unlock(&oldf->file_lock);
 739
 740         /* compute the remainder to be cleared */
 741         size = (newf->max_fds - open_files) * sizeof(struct file *);
 742
 743         /* This is long word aligned thus could use a optimized version */
 744         memset(new_fds, 0, size);
 745
 746         if (newf->max_fdset > open_files) {
 747                 int left = (newf->max_fdset-open_files)/8;
 748                 int start = open_files / (8 * sizeof(unsigned long));
 749
 750                 memset(&newf->open_fds->fds_bits[start], 0, left);
 751                 memset(&newf->close_on_exec->fds_bits[start], 0, left);
 752         }
 753
 754         tsk->files = newf;
 755         error = 0;
 756 out:
 757         return error;
 758
 759 out_release:
 760         free_fdset (newf->close_on_exec, newf->max_fdset);
 761         free_fdset (newf->open_fds, newf->max_fdset);
 762         kmem_cache_free(files_cachep, newf);
 763         goto out;
 764 }
 765
 766 /*
 767  *      Helper to unshare the files of the current task.
 768  *      We don't want to expose copy_files internals to
 769  *      the exec layer of the kernel.
 770  */
 771
 772 int unshare_files(void)
 773 {
 774         struct files_struct *files  = current->files;
 775         int rc;
 776
 777         if(!files)
 778                 BUG();
 779
 780         /* This can race but the race causes us to copy when we don't
 781            need to and drop the copy */
 782         if(atomic_read(&files->count) == 1)
 783         {
 784                 atomic_inc(&files->count);
 785                 return 0;
 786         }
 787         rc = copy_files(0, current);
 788         if(rc)
 789                 current->files = files;
 790         return rc;
 791 }
 792
 793 EXPORT_SYMBOL(unshare_files);
 794
 795 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 796 {
 797         struct sighand_struct *sig;
 798
 799         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 800                 atomic_inc(&current->sighand->count);
 801                 return 0;
 802         }
 803         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 804         tsk->sighand = sig;
 805         if (!sig)
 806                 return -ENOMEM;
 807         spin_lock_init(&sig->siglock);
 808         atomic_set(&sig->count, 1);
 809         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 810         return 0;
 811 }
 812
 813 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 814 {
 815         struct signal_struct *sig;
 816
 817         if (clone_flags & CLONE_THREAD) {
 818                 atomic_inc(&current->signal->count);
 819                 return 0;
 820         }
 821         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 822         tsk->signal = sig;
 823         if (!sig)
 824                 return -ENOMEM;
 825         atomic_set(&sig->count, 1);
 826         sig->group_exit = 0;
 827         sig->group_exit_code = 0;
 828         sig->group_exit_task = NULL;
 829         sig->group_stop_count = 0;
 830         sig->curr_target = NULL;
 831         init_sigpending(&sig->shared_pending);
 832         INIT_LIST_HEAD(&sig->posix_timers);
 833
 834         sig->tty = current->signal->tty;
 835         sig->pgrp = process_group(current);
 836         sig->session = current->signal->session;
 837         sig->leader = 0;        /* session leadership doesn't inherit */
 838         sig->tty_old_pgrp = 0;
 839
 840         return 0;
 841 }
 842
 843 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 844 {
 845         unsigned long new_flags = p->flags;
 846
 847         new_flags &= ~PF_SUPERPRIV;
 848         new_flags |= PF_FORKNOEXEC;
 849         if (!(clone_flags & CLONE_PTRACE))
 850                 p->ptrace = 0;
 851         p->flags = new_flags;
 852 }
 853
 854 asmlinkage long sys_set_tid_address(int __user *tidptr)
 855 {
 856         current->clear_child_tid = tidptr;
 857
 858         return current->pid;
 859 }
 860
 861 /*
 862  * This creates a new process as a copy of the old one,
 863  * but does not actually start it yet.
 864  *
 865  * It copies the registers, and all the appropriate
 866  * parts of the process environment (as per the clone
 867  * flags). The actual kick-off is left to the caller.
 868  */
 869 struct task_struct *copy_process(unsigned long clone_flags,
 870                                  unsigned long stack_start,
 871                                  struct pt_regs *regs,
 872                                  unsigned long stack_size,
 873                                  int __user *parent_tidptr,
 874                                  int __user *child_tidptr)
 875 {
 876         int retval;
 877         struct task_struct *p = NULL;
 878
 879         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 880                 return ERR_PTR(-EINVAL);
 881
 882         /*
 883          * Thread groups must share signals as well, and detached threads
 884          * can only be started up within the thread group.
 885          */
 886         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 887                 return ERR_PTR(-EINVAL);
 888
 889         /*
 890          * Shared signal handlers imply shared VM. By way of the above,
 891          * thread groups also imply shared VM. Blocking this case allows
 892          * for various simplifications in other code.
 893          */
 894         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 895                 return ERR_PTR(-EINVAL);
 896
 897         retval = security_task_create(clone_flags);
 898         if (retval)
 899                 goto fork_out;
 900
 901         retval = -ENOMEM;
 902         p = dup_task_struct(current);
 903         if (!p)
 904                 goto fork_out;
 905         p->tux_info = NULL;
 906
 907         retval = -EAGAIN;
 908         if (atomic_read(&p->user->processes) >=
 909                         p->rlim[RLIMIT_NPROC].rlim_cur) {
 910                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 911                                 p->user != &root_user)
 912                         goto bad_fork_free;
 913         }
 914
 915         atomic_inc(&p->user->__count);
 916         atomic_inc(&p->user->processes);
 917         get_group_info(p->group_info);
 918
 919         /*
 920          * If multiple threads are within copy_process(), then this check
 921          * triggers too late. This doesn't hurt, the check is only there
 922          * to stop root fork bombs.
 923          */
 924         if (nr_threads >= max_threads)
 925                 goto bad_fork_cleanup_count;
 926
 927         if (!try_module_get(p->thread_info->exec_domain->module))
 928                 goto bad_fork_cleanup_count;
 929
 930         if (p->binfmt && !try_module_get(p->binfmt->module))
 931                 goto bad_fork_cleanup_put_domain;
 932
 933         p->did_exec = 0;
 934         copy_flags(clone_flags, p);
 935         if (clone_flags & CLONE_IDLETASK)
 936                 p->pid = 0;
 937         else {
 938                 p->pid = alloc_pidmap();
 939                 if (p->pid == -1)
 940                         goto bad_fork_cleanup;
 941         }
 942         retval = -EFAULT;
 943         if (clone_flags & CLONE_PARENT_SETTID)
 944                 if (put_user(p->pid, parent_tidptr))
 945                         goto bad_fork_cleanup;
 946
 947         p->proc_dentry = NULL;
 948
 949         INIT_LIST_HEAD(&p->children);
 950         INIT_LIST_HEAD(&p->sibling);
 951         init_waitqueue_head(&p->wait_chldexit);
 952         p->vfork_done = NULL;
 953         spin_lock_init(&p->alloc_lock);
 954         spin_lock_init(&p->proc_lock);
 955
 956         clear_tsk_thread_flag(p, TIF_SIGPENDING);
 957         init_sigpending(&p->pending);
 958
 959         p->it_real_value = p->it_virt_value = p->it_prof_value = 0;
 960         p->it_real_incr = p->it_virt_incr = p->it_prof_incr = 0;
 961         init_timer(&p->real_timer);
 962         p->real_timer.data = (unsigned long) p;
 963
 964         p->utime = p->stime = 0;
 965         p->cutime = p->cstime = 0;
 966         p->lock_depth = -1;             /* -1 = no lock */
 967         p->start_time = get_jiffies_64();
 968         p->security = NULL;
 969         p->io_context = NULL;
 970         p->audit_context = NULL;
 971 #ifdef CONFIG_NUMA
 972         p->mempolicy = mpol_copy(p->mempolicy);
 973         if (IS_ERR(p->mempolicy)) {
 974                 retval = PTR_ERR(p->mempolicy);
 975                 p->mempolicy = NULL;
 976                 goto bad_fork_cleanup;
 977         }
 978 #endif
 979
 980         retval = -ENOMEM;
 981         if ((retval = security_task_alloc(p)))
 982                 goto bad_fork_cleanup_policy;
 983         if ((retval = audit_alloc(p)))
 984                 goto bad_fork_cleanup_security;
 985         /* copy all the process information */
 986         if ((retval = copy_semundo(clone_flags, p)))
 987                 goto bad_fork_cleanup_audit;
 988         if ((retval = copy_files(clone_flags, p)))
 989                 goto bad_fork_cleanup_semundo;
 990         if ((retval = copy_fs(clone_flags, p)))
 991                 goto bad_fork_cleanup_files;
 992         if ((retval = copy_sighand(clone_flags, p)))
 993                 goto bad_fork_cleanup_fs;
 994         if ((retval = copy_signal(clone_flags, p)))
 995                 goto bad_fork_cleanup_sighand;
 996         if ((retval = copy_mm(clone_flags, p)))
 997                 goto bad_fork_cleanup_signal;
 998         if ((retval = copy_namespace(clone_flags, p)))
 999                 goto bad_fork_cleanup_mm;
1000         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1001         if (retval)
1002                 goto bad_fork_cleanup_namespace;
1003
1004         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1005         /*
1006          * Clear TID on mm_release()?
1007          */
1008         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1009
1010         /*
1011          * Syscall tracing should be turned off in the child regardless
1012          * of CLONE_PTRACE.
1013          */
1014         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1015
1016         /* Our parent execution domain becomes current domain
1017            These must match for thread signalling to apply */
1018
1019         p->parent_exec_id = p->self_exec_id;
1020
1021         /* ok, now we should be set up.. */
1022         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1023         p->pdeath_signal = 0;
1024
1025         /* Perform scheduler related setup */
1026         sched_fork(p);
1027
1028         /*
1029          * Ok, make it visible to the rest of the system.
1030          * We dont wake it up yet.
1031          */
1032         p->tgid = p->pid;
1033         p->group_leader = p;
1034         INIT_LIST_HEAD(&p->ptrace_children);
1035         INIT_LIST_HEAD(&p->ptrace_list);
1036
1037         /* Need tasklist lock for parent etc handling! */
1038         write_lock_irq(&tasklist_lock);
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)
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->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->tgid = current->tgid;
1070                 p->group_leader = current->group_leader;
1071
1072                 if (current->signal->group_stop_count > 0) {
1073                         /*
1074                          * There is an all-stop in progress for the group.
1075                          * We ourselves will stop as soon as we check signals.
1076                          * Make the new thread part of that group stop too.
1077                          */
1078                         current->signal->group_stop_count++;
1079                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1080                 }
1081
1082                 spin_unlock(&current->sighand->siglock);
1083         }
1084
1085         SET_LINKS(p);
1086         if (p->ptrace & PT_PTRACED)
1087                 __ptrace_link(p, current->parent);
1088
1089         attach_pid(p, PIDTYPE_PID, p->pid);
1090         if (thread_group_leader(p)) {
1091                 attach_pid(p, PIDTYPE_TGID, p->tgid);
1092                 attach_pid(p, PIDTYPE_PGID, process_group(p));
1093                 attach_pid(p, PIDTYPE_SID, p->signal->session);
1094                 if (p->pid)
1095                         __get_cpu_var(process_counts)++;
1096         } else
1097                 link_pid(p, p->pids + PIDTYPE_TGID, &p->group_leader->pids[PIDTYPE_TGID].pid);
1098
1099         nr_threads++;
1100         write_unlock_irq(&tasklist_lock);
1101         retval = 0;
1102
1103 fork_out:
1104         if (retval)
1105                 return ERR_PTR(retval);
1106         return p;
1107
1108 bad_fork_cleanup_namespace:
1109         exit_namespace(p);
1110 bad_fork_cleanup_mm:
1111         exit_mm(p);
1112         if (p->active_mm)
1113                 mmdrop(p->active_mm);
1114 bad_fork_cleanup_signal:
1115         exit_signal(p);
1116 bad_fork_cleanup_sighand:
1117         exit_sighand(p);
1118 bad_fork_cleanup_fs:
1119         exit_fs(p); /* blocking */
1120 bad_fork_cleanup_files:
1121         exit_files(p); /* blocking */
1122 bad_fork_cleanup_semundo:
1123         exit_sem(p);
1124 bad_fork_cleanup_audit:
1125         audit_free(p);
1126 bad_fork_cleanup_security:
1127         security_task_free(p);
1128 bad_fork_cleanup_policy:
1129 #ifdef CONFIG_NUMA
1130         mpol_free(p->mempolicy);
1131 #endif
1132 bad_fork_cleanup:
1133         if (p->pid > 0)
1134                 free_pidmap(p->pid);
1135         if (p->binfmt)
1136                 module_put(p->binfmt->module);
1137 bad_fork_cleanup_put_domain:
1138         module_put(p->thread_info->exec_domain->module);
1139 bad_fork_cleanup_count:
1140         put_group_info(p->group_info);
1141         atomic_dec(&p->user->processes);
1142         free_uid(p->user);
1143 bad_fork_free:
1144         free_task(p);
1145         goto fork_out;
1146 }
1147
1148 static inline int fork_traceflag (unsigned clone_flags)
1149 {
1150         if (clone_flags & (CLONE_UNTRACED | CLONE_IDLETASK))
1151                 return 0;
1152         else if (clone_flags & CLONE_VFORK) {
1153                 if (current->ptrace & PT_TRACE_VFORK)
1154                         return PTRACE_EVENT_VFORK;
1155         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1156                 if (current->ptrace & PT_TRACE_CLONE)
1157                         return PTRACE_EVENT_CLONE;
1158         } else if (current->ptrace & PT_TRACE_FORK)
1159                 return PTRACE_EVENT_FORK;
1160
1161         return 0;
1162 }
1163
1164 /*
1165  *  Ok, this is the main fork-routine.
1166  *
1167  * It copies the process, and if successful kick-starts
1168  * it and waits for it to finish using the VM if required.
1169  */
1170 long do_fork(unsigned long clone_flags,
1171               unsigned long stack_start,
1172               struct pt_regs *regs,
1173               unsigned long stack_size,
1174               int __user *parent_tidptr,
1175               int __user *child_tidptr)
1176 {
1177         struct task_struct *p;
1178         int trace = 0;
1179         long pid;
1180
1181         if (unlikely(current->ptrace)) {
1182                 trace = fork_traceflag (clone_flags);
1183                 if (trace)
1184                         clone_flags |= CLONE_PTRACE;
1185         }
1186
1187         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr);
1188         /*
1189          * Do this prior waking up the new thread - the thread pointer
1190          * might get invalid after that point, if the thread exits quickly.
1191          */
1192         pid = IS_ERR(p) ? PTR_ERR(p) : p->pid;
1193
1194         if (!IS_ERR(p)) {
1195                 struct completion vfork;
1196
1197                 if (clone_flags & CLONE_VFORK) {
1198                         p->vfork_done = &vfork;
1199                         init_completion(&vfork);
1200                 }
1201
1202                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1203                         /*
1204                          * We'll start up with an immediate SIGSTOP.
1205                          */
1206                         sigaddset(&p->pending.signal, SIGSTOP);
1207                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1208                 }
1209
1210                 if (!(clone_flags & CLONE_STOPPED)) {
1211                         /*
1212                          * Do the wakeup last. On SMP we treat fork() and
1213                          * CLONE_VM separately, because fork() has already
1214                          * created cache footprint on this CPU (due to
1215                          * copying the pagetables), hence migration would
1216                          * probably be costy. Threads on the other hand
1217                          * have less traction to the current CPU, and if
1218                          * there's an imbalance then the scheduler can
1219                          * migrate this fresh thread now, before it
1220                          * accumulates a larger cache footprint:
1221                          */
1222                         if (clone_flags & CLONE_VM)
1223                                 wake_up_forked_thread(p);
1224                         else
1225                                 wake_up_forked_process(p);
1226                 } else {
1227                         int cpu = get_cpu();
1228
1229                         p->state = TASK_STOPPED;
1230                         if (cpu_is_offline(task_cpu(p)))
1231                                 set_task_cpu(p, cpu);
1232
1233                         put_cpu();
1234                 }
1235                 ++total_forks;
1236
1237                 if (unlikely (trace)) {
1238                         current->ptrace_message = pid;
1239                         ptrace_notify ((trace << 8) | SIGTRAP);
1240                 }
1241
1242                 if (clone_flags & CLONE_VFORK) {
1243                         wait_for_completion(&vfork);
1244                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1245                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1246                 } else
1247                         /*
1248                          * Let the child process run first, to avoid most of the
1249                          * COW overhead when the child exec()s afterwards.
1250                          */
1251                         set_need_resched();
1252         }
1253         return pid;
1254 }
1255
1256 /* SLAB cache for signal_struct structures (tsk->signal) */
1257 kmem_cache_t *signal_cachep;
1258
1259 /* SLAB cache for sighand_struct structures (tsk->sighand) */
1260 kmem_cache_t *sighand_cachep;
1261
1262 /* SLAB cache for files_struct structures (tsk->files) */
1263 kmem_cache_t *files_cachep;
1264
1265 /* SLAB cache for fs_struct structures (tsk->fs) */
1266 kmem_cache_t *fs_cachep;
1267
1268 /* SLAB cache for vm_area_struct structures */
1269 kmem_cache_t *vm_area_cachep;
1270
1271 /* SLAB cache for mm_struct structures (tsk->mm) */
1272 kmem_cache_t *mm_cachep;
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 }