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/sem.h>
  25 #include <linux/file.h>
  26 #include <linux/binfmts.h>
  27 #include <linux/mman.h>
  28 #include <linux/fs.h>
  29 #include <linux/security.h>
  30 #include <linux/syscalls.h>
  31 #include <linux/jiffies.h>
  32 #include <linux/futex.h>
  33 #include <linux/ptrace.h>
  34 #include <linux/mount.h>
  35 #include <linux/audit.h>
  36
  37 #include <asm/pgtable.h>
  38 #include <asm/pgalloc.h>
  39 #include <asm/uaccess.h>
  40 #include <asm/mmu_context.h>
  41 #include <asm/cacheflush.h>
  42 #include <asm/tlbflush.h>
  43
  44 /* The idle threads do not count..
  45  * Protected by write_lock_irq(&tasklist_lock)
  46  */
  47 int nr_threads;
  48
  49 int max_threads;
  50 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  51
  52 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  53
  54 rwlock_t tasklist_lock __cacheline_aligned = RW_LOCK_UNLOCKED;  /* outer */
  55
  56 EXPORT_SYMBOL(tasklist_lock);
  57
  58 int nr_processes(void)
  59 {
  60         int cpu;
  61         int total = 0;
  62
  63         for_each_cpu(cpu)
  64                 total += per_cpu(process_counts, cpu);
  65
  66         return total;
  67 }
  68
  69 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  70 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  71 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  72 static kmem_cache_t *task_struct_cachep;
  73 #endif
  74
  75 static void free_task(struct task_struct *tsk)
  76 {
  77         free_thread_info(tsk->thread_info);
  78         free_task_struct(tsk);
  79 }
  80
  81 void __put_task_struct(struct task_struct *tsk)
  82 {
  83         WARN_ON(!(tsk->state & (TASK_DEAD | TASK_ZOMBIE)));
  84         WARN_ON(atomic_read(&tsk->usage));
  85         WARN_ON(tsk == current);
  86
  87         if (unlikely(tsk->audit_context))
  88                 audit_free(tsk);
  89         security_task_free(tsk);
  90         free_uid(tsk->user);
  91         put_group_info(tsk->group_info);
  92         free_task(tsk);
  93 }
  94
  95 void fastcall add_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
  96 {
  97         unsigned long flags;
  98
  99         wait->flags &= ~WQ_FLAG_EXCLUSIVE;
 100         spin_lock_irqsave(&q->lock, flags);
 101         __add_wait_queue(q, wait);
 102         spin_unlock_irqrestore(&q->lock, flags);
 103 }
 104
 105 EXPORT_SYMBOL(add_wait_queue);
 106
 107 void fastcall add_wait_queue_exclusive(wait_queue_head_t *q, wait_queue_t * wait)
 108 {
 109         unsigned long flags;
 110
 111         wait->flags |= WQ_FLAG_EXCLUSIVE;
 112         spin_lock_irqsave(&q->lock, flags);
 113         __add_wait_queue_tail(q, wait);
 114         spin_unlock_irqrestore(&q->lock, flags);
 115 }
 116
 117 EXPORT_SYMBOL(add_wait_queue_exclusive);
 118
 119 void fastcall remove_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
 120 {
 121         unsigned long flags;
 122
 123         spin_lock_irqsave(&q->lock, flags);
 124         __remove_wait_queue(q, wait);
 125         spin_unlock_irqrestore(&q->lock, flags);
 126 }
 127
 128 EXPORT_SYMBOL(remove_wait_queue);
 129
 130
 131 /*
 132  * Note: we use "set_current_state()" _after_ the wait-queue add,
 133  * because we need a memory barrier there on SMP, so that any
 134  * wake-function that tests for the wait-queue being active
 135  * will be guaranteed to see waitqueue addition _or_ subsequent
 136  * tests in this thread will see the wakeup having taken place.
 137  *
 138  * The spin_unlock() itself is semi-permeable and only protects
 139  * one way (it only protects stuff inside the critical region and
 140  * stops them from bleeding out - it would still allow subsequent
 141  * loads to move into the the critical region).
 142  */
 143 void fastcall prepare_to_wait(wait_queue_head_t *q, wait_queue_t *wait, int state)
 144 {
 145         unsigned long flags;
 146
 147         wait->flags &= ~WQ_FLAG_EXCLUSIVE;
 148         spin_lock_irqsave(&q->lock, flags);
 149         if (list_empty(&wait->task_list))
 150                 __add_wait_queue(q, wait);
 151         set_current_state(state);
 152         spin_unlock_irqrestore(&q->lock, flags);
 153 }
 154
 155 EXPORT_SYMBOL(prepare_to_wait);
 156
 157 void fastcall
 158 prepare_to_wait_exclusive(wait_queue_head_t *q, wait_queue_t *wait, int state)
 159 {
 160         unsigned long flags;
 161
 162         wait->flags |= WQ_FLAG_EXCLUSIVE;
 163         spin_lock_irqsave(&q->lock, flags);
 164         if (list_empty(&wait->task_list))
 165                 __add_wait_queue_tail(q, wait);
 166         set_current_state(state);
 167         spin_unlock_irqrestore(&q->lock, flags);
 168 }
 169
 170 EXPORT_SYMBOL(prepare_to_wait_exclusive);
 171
 172 void fastcall finish_wait(wait_queue_head_t *q, wait_queue_t *wait)
 173 {
 174         unsigned long flags;
 175
 176         __set_current_state(TASK_RUNNING);
 177         /*
 178          * We can check for list emptiness outside the lock
 179          * IFF:
 180          *  - we use the "careful" check that verifies both
 181          *    the next and prev pointers, so that there cannot
 182          *    be any half-pending updates in progress on other
 183          *    CPU's that we haven't seen yet (and that might
 184          *    still change the stack area.
 185          * and
 186          *  - all other users take the lock (ie we can only
 187          *    have _one_ other CPU that looks at or modifies
 188          *    the list).
 189          */
 190         if (!list_empty_careful(&wait->task_list)) {
 191                 spin_lock_irqsave(&q->lock, flags);
 192                 list_del_init(&wait->task_list);
 193                 spin_unlock_irqrestore(&q->lock, flags);
 194         }
 195 }
 196
 197 EXPORT_SYMBOL(finish_wait);
 198
 199 int autoremove_wake_function(wait_queue_t *wait, unsigned mode, int sync)
 200 {
 201         int ret = default_wake_function(wait, mode, sync);
 202
 203         if (ret)
 204                 list_del_init(&wait->task_list);
 205         return ret;
 206 }
 207
 208 EXPORT_SYMBOL(autoremove_wake_function);
 209
 210 void __init fork_init(unsigned long mempages)
 211 {
 212 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 213 #ifndef ARCH_MIN_TASKALIGN
 214 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 215 #endif
 216         /* create a slab on which task_structs can be allocated */
 217         task_struct_cachep =
 218                 kmem_cache_create("task_struct",
 219                                   sizeof(struct task_struct),ARCH_MIN_TASKALIGN,
 220                                   0, NULL, NULL);
 221         if (!task_struct_cachep)
 222                 panic("fork_init(): cannot create task_struct SLAB cache");
 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
 276         down_write(&oldmm->mmap_sem);
 277         flush_cache_mm(current->mm);
 278         mm->locked_vm = 0;
 279         mm->mmap = NULL;
 280         mm->mmap_cache = NULL;
 281         mm->free_area_cache = TASK_UNMAPPED_BASE;
 282         mm->map_count = 0;
 283         mm->rss = 0;
 284         cpus_clear(mm->cpu_vm_mask);
 285         mm->mm_rb = RB_ROOT;
 286         rb_link = &mm->mm_rb.rb_node;
 287         rb_parent = NULL;
 288         pprev = &mm->mmap;
 289
 290         /*
 291          * Add it to the mmlist after the parent.
 292          * Doing it this way means that we can order the list,
 293          * and fork() won't mess up the ordering significantly.
 294          * Add it first so that swapoff can see any swap entries.
 295          */
 296         spin_lock(&mmlist_lock);
 297         list_add(&mm->mmlist, &current->mm->mmlist);
 298         mmlist_nr++;
 299         spin_unlock(&mmlist_lock);
 300
 301         for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
 302                 struct file *file;
 303
 304                 if(mpnt->vm_flags & VM_DONTCOPY)
 305                         continue;
 306                 if (mpnt->vm_flags & VM_ACCOUNT) {
 307                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 308                         if (security_vm_enough_memory(len))
 309                                 goto fail_nomem;
 310                         charge += len;
 311                 }
 312                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 313                 if (!tmp)
 314                         goto fail_nomem;
 315                 *tmp = *mpnt;
 316                 tmp->vm_flags &= ~VM_LOCKED;
 317                 tmp->vm_mm = mm;
 318                 tmp->vm_next = NULL;
 319                 file = tmp->vm_file;
 320                 INIT_LIST_HEAD(&tmp->shared);
 321                 if (file) {
 322                         struct inode *inode = file->f_dentry->d_inode;
 323                         get_file(file);
 324                         if (tmp->vm_flags & VM_DENYWRITE)
 325                                 atomic_dec(&inode->i_writecount);
 326
 327                         /* insert tmp into the share list, just after mpnt */
 328                         down(&file->f_mapping->i_shared_sem);
 329                         list_add(&tmp->shared, &mpnt->shared);
 330                         up(&file->f_mapping->i_shared_sem);
 331                 }
 332
 333                 /*
 334                  * Link in the new vma and copy the page table entries:
 335                  * link in first so that swapoff can see swap entries,
 336                  * and try_to_unmap_one's find_vma find the new vma.
 337                  */
 338                 spin_lock(&mm->page_table_lock);
 339                 *pprev = tmp;
 340                 pprev = &tmp->vm_next;
 341
 342                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 343                 rb_link = &tmp->vm_rb.rb_right;
 344                 rb_parent = &tmp->vm_rb;
 345
 346                 mm->map_count++;
 347                 retval = copy_page_range(mm, current->mm, tmp);
 348                 spin_unlock(&mm->page_table_lock);
 349
 350                 if (tmp->vm_ops && tmp->vm_ops->open)
 351                         tmp->vm_ops->open(tmp);
 352
 353                 if (retval)
 354                         goto fail;
 355         }
 356         retval = 0;
 357
 358 out:
 359         flush_tlb_mm(current->mm);
 360         up_write(&oldmm->mmap_sem);
 361         return retval;
 362 fail_nomem:
 363         retval = -ENOMEM;
 364 fail:
 365         vm_unacct_memory(charge);
 366         goto out;
 367 }
 368 static inline int mm_alloc_pgd(struct mm_struct * mm)
 369 {
 370         mm->pgd = pgd_alloc(mm);
 371         if (unlikely(!mm->pgd))
 372                 return -ENOMEM;
 373         return 0;
 374 }
 375
 376 static inline void mm_free_pgd(struct mm_struct * mm)
 377 {
 378         pgd_free(mm->pgd);
 379 }
 380 #else
 381 #define dup_mmap(mm, oldmm)     (0)
 382 #define mm_alloc_pgd(mm)        (0)
 383 #define mm_free_pgd(mm)
 384 #endif /* CONFIG_MMU */
 385
 386 spinlock_t mmlist_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
 387 int mmlist_nr;
 388
 389 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
 390 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 391
 392 #include <linux/init_task.h>
 393
 394 static struct mm_struct * mm_init(struct mm_struct * mm)
 395 {
 396         atomic_set(&mm->mm_users, 1);
 397         atomic_set(&mm->mm_count, 1);
 398         init_rwsem(&mm->mmap_sem);
 399         mm->core_waiters = 0;
 400         mm->page_table_lock = SPIN_LOCK_UNLOCKED;
 401         mm->ioctx_list_lock = RW_LOCK_UNLOCKED;
 402         mm->ioctx_list = NULL;
 403         mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
 404         mm->free_area_cache = TASK_UNMAPPED_BASE;
 405
 406         if (likely(!mm_alloc_pgd(mm))) {
 407                 mm->def_flags = 0;
 408                 return mm;
 409         }
 410         free_mm(mm);
 411         return NULL;
 412 }
 413
 414 /*
 415  * Allocate and initialize an mm_struct.
 416  */
 417 struct mm_struct * mm_alloc(void)
 418 {
 419         struct mm_struct * mm;
 420
 421         mm = allocate_mm();
 422         if (mm) {
 423                 memset(mm, 0, sizeof(*mm));
 424                 return mm_init(mm);
 425         }
 426         return NULL;
 427 }
 428
 429 /*
 430  * Called when the last reference to the mm
 431  * is dropped: either by a lazy thread or by
 432  * mmput. Free the page directory and the mm.
 433  */
 434 void fastcall __mmdrop(struct mm_struct *mm)
 435 {
 436         BUG_ON(mm == &init_mm);
 437         mm_free_pgd(mm);
 438         destroy_context(mm);
 439         free_mm(mm);
 440 }
 441
 442 /*
 443  * Decrement the use count and release all resources for an mm.
 444  */
 445 void mmput(struct mm_struct *mm)
 446 {
 447         if (atomic_dec_and_lock(&mm->mm_users, &mmlist_lock)) {
 448                 list_del(&mm->mmlist);
 449                 mmlist_nr--;
 450                 spin_unlock(&mmlist_lock);
 451                 exit_aio(mm);
 452                 exit_mmap(mm);
 453                 mmdrop(mm);
 454         }
 455 }
 456
 457 /*
 458  * Checks if the use count of an mm is non-zero and if so
 459  * returns a reference to it after bumping up the use count.
 460  * If the use count is zero, it means this mm is going away,
 461  * so return NULL.
 462  */
 463 struct mm_struct *mmgrab(struct mm_struct *mm)
 464 {
 465         spin_lock(&mmlist_lock);
 466         if (!atomic_read(&mm->mm_users))
 467                 mm = NULL;
 468         else
 469                 atomic_inc(&mm->mm_users);
 470         spin_unlock(&mmlist_lock);
 471         return mm;
 472 }
 473
 474 /* Please note the differences between mmput and mm_release.
 475  * mmput is called whenever we stop holding onto a mm_struct,
 476  * error success whatever.
 477  *
 478  * mm_release is called after a mm_struct has been removed
 479  * from the current process.
 480  *
 481  * This difference is important for error handling, when we
 482  * only half set up a mm_struct for a new process and need to restore
 483  * the old one.  Because we mmput the new mm_struct before
 484  * restoring the old one. . .
 485  * Eric Biederman 10 January 1998
 486  */
 487 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 488 {
 489         struct completion *vfork_done = tsk->vfork_done;
 490
 491         /* Get rid of any cached register state */
 492         deactivate_mm(tsk, mm);
 493
 494         /* notify parent sleeping on vfork() */
 495         if (vfork_done) {
 496                 tsk->vfork_done = NULL;
 497                 complete(vfork_done);
 498         }
 499         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
 500                 u32 __user * tidptr = tsk->clear_child_tid;
 501                 tsk->clear_child_tid = NULL;
 502
 503                 /*
 504                  * We don't check the error code - if userspace has
 505                  * not set up a proper pointer then tough luck.
 506                  */
 507                 put_user(0, tidptr);
 508                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL);
 509         }
 510 }
 511
 512 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 513 {
 514         struct mm_struct * mm, *oldmm;
 515         int retval;
 516
 517         tsk->min_flt = tsk->maj_flt = 0;
 518         tsk->cmin_flt = tsk->cmaj_flt = 0;
 519         tsk->nvcsw = tsk->nivcsw = tsk->cnvcsw = tsk->cnivcsw = 0;
 520
 521         tsk->mm = NULL;
 522         tsk->active_mm = NULL;
 523
 524         /*
 525          * Are we cloning a kernel thread?
 526          *
 527          * We need to steal a active VM for that..
 528          */
 529         oldmm = current->mm;
 530         if (!oldmm)
 531                 return 0;
 532
 533         if (clone_flags & CLONE_VM) {
 534                 atomic_inc(&oldmm->mm_users);
 535                 mm = oldmm;
 536                 /*
 537                  * There are cases where the PTL is held to ensure no
 538                  * new threads start up in user mode using an mm, which
 539                  * allows optimizing out ipis; the tlb_gather_mmu code
 540                  * is an example.
 541                  */
 542                 spin_unlock_wait(&oldmm->page_table_lock);
 543                 goto good_mm;
 544         }
 545
 546         retval = -ENOMEM;
 547         mm = allocate_mm();
 548         if (!mm)
 549                 goto fail_nomem;
 550
 551         /* Copy the current MM stuff.. */
 552         memcpy(mm, oldmm, sizeof(*mm));
 553         if (!mm_init(mm))
 554                 goto fail_nomem;
 555
 556         if (init_new_context(tsk,mm))
 557                 goto fail_nocontext;
 558
 559         retval = dup_mmap(mm, oldmm);
 560         if (retval)
 561                 goto free_pt;
 562
 563 good_mm:
 564         tsk->mm = mm;
 565         tsk->active_mm = mm;
 566         return 0;
 567
 568 free_pt:
 569         mmput(mm);
 570 fail_nomem:
 571         return retval;
 572
 573 fail_nocontext:
 574         /*
 575          * If init_new_context() failed, we cannot use mmput() to free the mm
 576          * because it calls destroy_context()
 577          */
 578         mm_free_pgd(mm);
 579         free_mm(mm);
 580         return retval;
 581 }
 582
 583 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 584 {
 585         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 586         /* We don't need to lock fs - think why ;-) */
 587         if (fs) {
 588                 atomic_set(&fs->count, 1);
 589                 fs->lock = RW_LOCK_UNLOCKED;
 590                 fs->umask = old->umask;
 591                 read_lock(&old->lock);
 592                 fs->rootmnt = mntget(old->rootmnt);
 593                 fs->root = dget(old->root);
 594                 fs->pwdmnt = mntget(old->pwdmnt);
 595                 fs->pwd = dget(old->pwd);
 596                 if (old->altroot) {
 597                         fs->altrootmnt = mntget(old->altrootmnt);
 598                         fs->altroot = dget(old->altroot);
 599                 } else {
 600                         fs->altrootmnt = NULL;
 601                         fs->altroot = NULL;
 602                 }
 603                 read_unlock(&old->lock);
 604         }
 605         return fs;
 606 }
 607
 608 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 609 {
 610         return __copy_fs_struct(old);
 611 }
 612
 613 EXPORT_SYMBOL_GPL(copy_fs_struct);
 614
 615 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 616 {
 617         if (clone_flags & CLONE_FS) {
 618                 atomic_inc(&current->fs->count);
 619                 return 0;
 620         }
 621         tsk->fs = __copy_fs_struct(current->fs);
 622         if (!tsk->fs)
 623                 return -ENOMEM;
 624         return 0;
 625 }
 626
 627 static int count_open_files(struct files_struct *files, int size)
 628 {
 629         int i;
 630
 631         /* Find the last open fd */
 632         for (i = size/(8*sizeof(long)); i > 0; ) {
 633                 if (files->open_fds->fds_bits[--i])
 634                         break;
 635         }
 636         i = (i+1) * 8 * sizeof(long);
 637         return i;
 638 }
 639
 640 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 641 {
 642         struct files_struct *oldf, *newf;
 643         struct file **old_fds, **new_fds;
 644         int open_files, nfds, size, i, error = 0;
 645
 646         /*
 647          * A background process may not have any files ...
 648          */
 649         oldf = current->files;
 650         if (!oldf)
 651                 goto out;
 652
 653         if (clone_flags & CLONE_FILES) {
 654                 atomic_inc(&oldf->count);
 655                 goto out;
 656         }
 657
 658         /*
 659          * Note: we may be using current for both targets (See exec.c)
 660          * This works because we cache current->files (old) as oldf. Don't
 661          * break this.
 662          */
 663         tsk->files = NULL;
 664         error = -ENOMEM;
 665         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
 666         if (!newf)
 667                 goto out;
 668
 669         atomic_set(&newf->count, 1);
 670
 671         newf->file_lock     = SPIN_LOCK_UNLOCKED;
 672         newf->next_fd       = 0;
 673         newf->max_fds       = NR_OPEN_DEFAULT;
 674         newf->max_fdset     = __FD_SETSIZE;
 675         newf->close_on_exec = &newf->close_on_exec_init;
 676         newf->open_fds      = &newf->open_fds_init;
 677         newf->fd            = &newf->fd_array[0];
 678
 679         /* We don't yet have the oldf readlock, but even if the old
 680            fdset gets grown now, we'll only copy up to "size" fds */
 681         size = oldf->max_fdset;
 682         if (size > __FD_SETSIZE) {
 683                 newf->max_fdset = 0;
 684                 spin_lock(&newf->file_lock);
 685                 error = expand_fdset(newf, size-1);
 686                 spin_unlock(&newf->file_lock);
 687                 if (error)
 688                         goto out_release;
 689         }
 690         spin_lock(&oldf->file_lock);
 691
 692         open_files = count_open_files(oldf, size);
 693
 694         /*
 695          * Check whether we need to allocate a larger fd array.
 696          * Note: we're not a clone task, so the open count won't
 697          * change.
 698          */
 699         nfds = NR_OPEN_DEFAULT;
 700         if (open_files > nfds) {
 701                 spin_unlock(&oldf->file_lock);
 702                 newf->max_fds = 0;
 703                 spin_lock(&newf->file_lock);
 704                 error = expand_fd_array(newf, open_files-1);
 705                 spin_unlock(&newf->file_lock);
 706                 if (error)
 707                         goto out_release;
 708                 nfds = newf->max_fds;
 709                 spin_lock(&oldf->file_lock);
 710         }
 711
 712         old_fds = oldf->fd;
 713         new_fds = newf->fd;
 714
 715         memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
 716         memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
 717
 718         for (i = open_files; i != 0; i--) {
 719                 struct file *f = *old_fds++;
 720                 if (f)
 721                         get_file(f);
 722                 *new_fds++ = f;
 723         }
 724         spin_unlock(&oldf->file_lock);
 725
 726         /* compute the remainder to be cleared */
 727         size = (newf->max_fds - open_files) * sizeof(struct file *);
 728
 729         /* This is long word aligned thus could use a optimized version */
 730         memset(new_fds, 0, size);
 731
 732         if (newf->max_fdset > open_files) {
 733                 int left = (newf->max_fdset-open_files)/8;
 734                 int start = open_files / (8 * sizeof(unsigned long));
 735
 736                 memset(&newf->open_fds->fds_bits[start], 0, left);
 737                 memset(&newf->close_on_exec->fds_bits[start], 0, left);
 738         }
 739
 740         tsk->files = newf;
 741         error = 0;
 742 out:
 743         return error;
 744
 745 out_release:
 746         free_fdset (newf->close_on_exec, newf->max_fdset);
 747         free_fdset (newf->open_fds, newf->max_fdset);
 748         kmem_cache_free(files_cachep, newf);
 749         goto out;
 750 }
 751
 752 /*
 753  *      Helper to unshare the files of the current task.
 754  *      We don't want to expose copy_files internals to
 755  *      the exec layer of the kernel.
 756  */
 757
 758 int unshare_files(void)
 759 {
 760         struct files_struct *files  = current->files;
 761         int rc;
 762
 763         if(!files)
 764                 BUG();
 765
 766         /* This can race but the race causes us to copy when we don't
 767            need to and drop the copy */
 768         if(atomic_read(&files->count) == 1)
 769         {
 770                 atomic_inc(&files->count);
 771                 return 0;
 772         }
 773         rc = copy_files(0, current);
 774         if(rc)
 775                 current->files = files;
 776         return rc;
 777 }
 778
 779 EXPORT_SYMBOL(unshare_files);
 780
 781 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 782 {
 783         struct sighand_struct *sig;
 784
 785         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 786                 atomic_inc(&current->sighand->count);
 787                 return 0;
 788         }
 789         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 790         tsk->sighand = sig;
 791         if (!sig)
 792                 return -ENOMEM;
 793         spin_lock_init(&sig->siglock);
 794         atomic_set(&sig->count, 1);
 795         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 796         return 0;
 797 }
 798
 799 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 800 {
 801         struct signal_struct *sig;
 802
 803         if (clone_flags & CLONE_THREAD) {
 804                 atomic_inc(&current->signal->count);
 805                 return 0;
 806         }
 807         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 808         tsk->signal = sig;
 809         if (!sig)
 810                 return -ENOMEM;
 811         atomic_set(&sig->count, 1);
 812         sig->group_exit = 0;
 813         sig->group_exit_code = 0;
 814         sig->group_exit_task = NULL;
 815         sig->group_stop_count = 0;
 816         sig->curr_target = NULL;
 817         init_sigpending(&sig->shared_pending);
 818         INIT_LIST_HEAD(&sig->posix_timers);
 819
 820         sig->tty = current->signal->tty;
 821         sig->pgrp = process_group(current);
 822         sig->session = current->signal->session;
 823         sig->leader = 0;        /* session leadership doesn't inherit */
 824         sig->tty_old_pgrp = 0;
 825
 826         return 0;
 827 }
 828
 829 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 830 {
 831         unsigned long new_flags = p->flags;
 832
 833         new_flags &= ~PF_SUPERPRIV;
 834         new_flags |= PF_FORKNOEXEC;
 835         if (!(clone_flags & CLONE_PTRACE))
 836                 p->ptrace = 0;
 837         p->flags = new_flags;
 838 }
 839
 840 asmlinkage long sys_set_tid_address(int __user *tidptr)
 841 {
 842         current->clear_child_tid = tidptr;
 843
 844         return current->pid;
 845 }
 846
 847 /*
 848  * This creates a new process as a copy of the old one,
 849  * but does not actually start it yet.
 850  *
 851  * It copies the registers, and all the appropriate
 852  * parts of the process environment (as per the clone
 853  * flags). The actual kick-off is left to the caller.
 854  */
 855 struct task_struct *copy_process(unsigned long clone_flags,
 856                                  unsigned long stack_start,
 857                                  struct pt_regs *regs,
 858                                  unsigned long stack_size,
 859                                  int __user *parent_tidptr,
 860                                  int __user *child_tidptr)
 861 {
 862         int retval;
 863         struct task_struct *p = NULL;
 864
 865         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 866                 return ERR_PTR(-EINVAL);
 867
 868         /*
 869          * Thread groups must share signals as well, and detached threads
 870          * can only be started up within the thread group.
 871          */
 872         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 873                 return ERR_PTR(-EINVAL);
 874
 875         /*
 876          * Shared signal handlers imply shared VM. By way of the above,
 877          * thread groups also imply shared VM. Blocking this case allows
 878          * for various simplifications in other code.
 879          */
 880         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 881                 return ERR_PTR(-EINVAL);
 882
 883         retval = security_task_create(clone_flags);
 884         if (retval)
 885                 goto fork_out;
 886
 887         retval = -ENOMEM;
 888         p = dup_task_struct(current);
 889         if (!p)
 890                 goto fork_out;
 891
 892         retval = -EAGAIN;
 893         if (atomic_read(&p->user->processes) >=
 894                         p->rlim[RLIMIT_NPROC].rlim_cur) {
 895                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 896                                 p->user != &root_user)
 897                         goto bad_fork_free;
 898         }
 899
 900         atomic_inc(&p->user->__count);
 901         atomic_inc(&p->user->processes);
 902         get_group_info(p->group_info);
 903
 904         /*
 905          * If multiple threads are within copy_process(), then this check
 906          * triggers too late. This doesn't hurt, the check is only there
 907          * to stop root fork bombs.
 908          */
 909         if (nr_threads >= max_threads)
 910                 goto bad_fork_cleanup_count;
 911
 912         if (!try_module_get(p->thread_info->exec_domain->module))
 913                 goto bad_fork_cleanup_count;
 914
 915         if (p->binfmt && !try_module_get(p->binfmt->module))
 916                 goto bad_fork_cleanup_put_domain;
 917
 918         p->did_exec = 0;
 919         copy_flags(clone_flags, p);
 920         if (clone_flags & CLONE_IDLETASK)
 921                 p->pid = 0;
 922         else {
 923                 p->pid = alloc_pidmap();
 924                 if (p->pid == -1)
 925                         goto bad_fork_cleanup;
 926         }
 927         retval = -EFAULT;
 928         if (clone_flags & CLONE_PARENT_SETTID)
 929                 if (put_user(p->pid, parent_tidptr))
 930                         goto bad_fork_cleanup;
 931
 932         p->proc_dentry = NULL;
 933
 934         INIT_LIST_HEAD(&p->children);
 935         INIT_LIST_HEAD(&p->sibling);
 936         init_waitqueue_head(&p->wait_chldexit);
 937         p->vfork_done = NULL;
 938         spin_lock_init(&p->alloc_lock);
 939         spin_lock_init(&p->proc_lock);
 940
 941         clear_tsk_thread_flag(p, TIF_SIGPENDING);
 942         init_sigpending(&p->pending);
 943
 944         p->it_real_value = p->it_virt_value = p->it_prof_value = 0;
 945         p->it_real_incr = p->it_virt_incr = p->it_prof_incr = 0;
 946         init_timer(&p->real_timer);
 947         p->real_timer.data = (unsigned long) p;
 948
 949         p->utime = p->stime = 0;
 950         p->cutime = p->cstime = 0;
 951         p->lock_depth = -1;             /* -1 = no lock */
 952         p->start_time = get_jiffies_64();
 953         p->security = NULL;
 954         p->io_context = NULL;
 955         p->audit_context = NULL;
 956
 957         retval = -ENOMEM;
 958         if ((retval = security_task_alloc(p)))
 959                 goto bad_fork_cleanup;
 960         if ((retval = audit_alloc(p)))
 961                 goto bad_fork_cleanup_security;
 962         /* copy all the process information */
 963         if ((retval = copy_semundo(clone_flags, p)))
 964                 goto bad_fork_cleanup_audit;
 965         if ((retval = copy_files(clone_flags, p)))
 966                 goto bad_fork_cleanup_semundo;
 967         if ((retval = copy_fs(clone_flags, p)))
 968                 goto bad_fork_cleanup_files;
 969         if ((retval = copy_sighand(clone_flags, p)))
 970                 goto bad_fork_cleanup_fs;
 971         if ((retval = copy_signal(clone_flags, p)))
 972                 goto bad_fork_cleanup_sighand;
 973         if ((retval = copy_mm(clone_flags, p)))
 974                 goto bad_fork_cleanup_signal;
 975         if ((retval = copy_namespace(clone_flags, p)))
 976                 goto bad_fork_cleanup_mm;
 977         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
 978         if (retval)
 979                 goto bad_fork_cleanup_namespace;
 980
 981         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
 982         /*
 983          * Clear TID on mm_release()?
 984          */
 985         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
 986
 987         /*
 988          * Syscall tracing should be turned off in the child regardless
 989          * of CLONE_PTRACE.
 990          */
 991         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
 992
 993         /* Our parent execution domain becomes current domain
 994            These must match for thread signalling to apply */
 995
 996         p->parent_exec_id = p->self_exec_id;
 997
 998         /* ok, now we should be set up.. */
 999         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1000         p->pdeath_signal = 0;
1001
1002         /* Perform scheduler related setup */
1003         sched_fork(p);
1004
1005         /*
1006          * Ok, make it visible to the rest of the system.
1007          * We dont wake it up yet.
1008          */
1009         p->tgid = p->pid;
1010         p->group_leader = p;
1011         INIT_LIST_HEAD(&p->ptrace_children);
1012         INIT_LIST_HEAD(&p->ptrace_list);
1013
1014         /* Need tasklist lock for parent etc handling! */
1015         write_lock_irq(&tasklist_lock);
1016         /*
1017          * Check for pending SIGKILL! The new thread should not be allowed
1018          * to slip out of an OOM kill. (or normal SIGKILL.)
1019          */
1020         if (sigismember(&current->pending.signal, SIGKILL)) {
1021                 write_unlock_irq(&tasklist_lock);
1022                 retval = -EINTR;
1023                 goto bad_fork_cleanup_namespace;
1024         }
1025
1026         /* CLONE_PARENT re-uses the old parent */
1027         if (clone_flags & CLONE_PARENT)
1028                 p->real_parent = current->real_parent;
1029         else
1030                 p->real_parent = current;
1031         p->parent = p->real_parent;
1032
1033         if (clone_flags & CLONE_THREAD) {
1034                 spin_lock(&current->sighand->siglock);
1035                 /*
1036                  * Important: if an exit-all has been started then
1037                  * do not create this new thread - the whole thread
1038                  * group is supposed to exit anyway.
1039                  */
1040                 if (current->signal->group_exit) {
1041                         spin_unlock(&current->sighand->siglock);
1042                         write_unlock_irq(&tasklist_lock);
1043                         retval = -EAGAIN;
1044                         goto bad_fork_cleanup_namespace;
1045                 }
1046                 p->tgid = current->tgid;
1047                 p->group_leader = current->group_leader;
1048
1049                 if (current->signal->group_stop_count > 0) {
1050                         /*
1051                          * There is an all-stop in progress for the group.
1052                          * We ourselves will stop as soon as we check signals.
1053                          * Make the new thread part of that group stop too.
1054                          */
1055                         current->signal->group_stop_count++;
1056                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1057                 }
1058
1059                 spin_unlock(&current->sighand->siglock);
1060         }
1061
1062         SET_LINKS(p);
1063         if (p->ptrace & PT_PTRACED)
1064                 __ptrace_link(p, current->parent);
1065
1066         attach_pid(p, PIDTYPE_PID, p->pid);
1067         if (thread_group_leader(p)) {
1068                 attach_pid(p, PIDTYPE_TGID, p->tgid);
1069                 attach_pid(p, PIDTYPE_PGID, process_group(p));
1070                 attach_pid(p, PIDTYPE_SID, p->signal->session);
1071                 if (p->pid)
1072                         __get_cpu_var(process_counts)++;
1073         } else
1074                 link_pid(p, p->pids + PIDTYPE_TGID, &p->group_leader->pids[PIDTYPE_TGID].pid);
1075
1076         nr_threads++;
1077         write_unlock_irq(&tasklist_lock);
1078         retval = 0;
1079
1080 fork_out:
1081         if (retval)
1082                 return ERR_PTR(retval);
1083         return p;
1084
1085 bad_fork_cleanup_namespace:
1086         exit_namespace(p);
1087 bad_fork_cleanup_mm:
1088         exit_mm(p);
1089         if (p->active_mm)
1090                 mmdrop(p->active_mm);
1091 bad_fork_cleanup_signal:
1092         exit_signal(p);
1093 bad_fork_cleanup_sighand:
1094         exit_sighand(p);
1095 bad_fork_cleanup_fs:
1096         exit_fs(p); /* blocking */
1097 bad_fork_cleanup_files:
1098         exit_files(p); /* blocking */
1099 bad_fork_cleanup_semundo:
1100         exit_sem(p);
1101 bad_fork_cleanup_audit:
1102         audit_free(p);
1103 bad_fork_cleanup_security:
1104         security_task_free(p);
1105 bad_fork_cleanup:
1106         if (p->pid > 0)
1107                 free_pidmap(p->pid);
1108         if (p->binfmt)
1109                 module_put(p->binfmt->module);
1110 bad_fork_cleanup_put_domain:
1111         module_put(p->thread_info->exec_domain->module);
1112 bad_fork_cleanup_count:
1113         put_group_info(p->group_info);
1114         atomic_dec(&p->user->processes);
1115         free_uid(p->user);
1116 bad_fork_free:
1117         free_task(p);
1118         goto fork_out;
1119 }
1120
1121 static inline int fork_traceflag (unsigned clone_flags)
1122 {
1123         if (clone_flags & (CLONE_UNTRACED | CLONE_IDLETASK))
1124                 return 0;
1125         else if (clone_flags & CLONE_VFORK) {
1126                 if (current->ptrace & PT_TRACE_VFORK)
1127                         return PTRACE_EVENT_VFORK;
1128         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1129                 if (current->ptrace & PT_TRACE_CLONE)
1130                         return PTRACE_EVENT_CLONE;
1131         } else if (current->ptrace & PT_TRACE_FORK)
1132                 return PTRACE_EVENT_FORK;
1133
1134         return 0;
1135 }
1136
1137 /*
1138  *  Ok, this is the main fork-routine.
1139  *
1140  * It copies the process, and if successful kick-starts
1141  * it and waits for it to finish using the VM if required.
1142  */
1143 long do_fork(unsigned long clone_flags,
1144               unsigned long stack_start,
1145               struct pt_regs *regs,
1146               unsigned long stack_size,
1147               int __user *parent_tidptr,
1148               int __user *child_tidptr)
1149 {
1150         struct task_struct *p;
1151         int trace = 0;
1152         long pid;
1153
1154         if (unlikely(current->ptrace)) {
1155                 trace = fork_traceflag (clone_flags);
1156                 if (trace)
1157                         clone_flags |= CLONE_PTRACE;
1158         }
1159
1160         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr);
1161         /*
1162          * Do this prior waking up the new thread - the thread pointer
1163          * might get invalid after that point, if the thread exits quickly.
1164          */
1165         pid = IS_ERR(p) ? PTR_ERR(p) : p->pid;
1166
1167         if (!IS_ERR(p)) {
1168                 struct completion vfork;
1169
1170                 if (clone_flags & CLONE_VFORK) {
1171                         p->vfork_done = &vfork;
1172                         init_completion(&vfork);
1173                 }
1174
1175                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1176                         /*
1177                          * We'll start up with an immediate SIGSTOP.
1178                          */
1179                         sigaddset(&p->pending.signal, SIGSTOP);
1180                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1181                 }
1182
1183                 if (!(clone_flags & CLONE_STOPPED))
1184                         wake_up_forked_process(p);      /* do this last */
1185                 else
1186                         p->state = TASK_STOPPED;
1187                 ++total_forks;
1188
1189                 if (unlikely (trace)) {
1190                         current->ptrace_message = pid;
1191                         ptrace_notify ((trace << 8) | SIGTRAP);
1192                 }
1193
1194                 if (clone_flags & CLONE_VFORK) {
1195                         wait_for_completion(&vfork);
1196                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1197                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1198                 } else
1199                         /*
1200                          * Let the child process run first, to avoid most of the
1201                          * COW overhead when the child exec()s afterwards.
1202                          */
1203                         set_need_resched();
1204         }
1205         return pid;
1206 }
1207
1208 /* SLAB cache for signal_struct structures (tsk->signal) */
1209 kmem_cache_t *signal_cachep;
1210
1211 /* SLAB cache for sighand_struct structures (tsk->sighand) */
1212 kmem_cache_t *sighand_cachep;
1213
1214 /* SLAB cache for files_struct structures (tsk->files) */
1215 kmem_cache_t *files_cachep;
1216
1217 /* SLAB cache for fs_struct structures (tsk->fs) */
1218 kmem_cache_t *fs_cachep;
1219
1220 /* SLAB cache for vm_area_struct structures */
1221 kmem_cache_t *vm_area_cachep;
1222
1223 /* SLAB cache for mm_struct structures (tsk->mm) */
1224 kmem_cache_t *mm_cachep;
1225
1226 void __init proc_caches_init(void)
1227 {
1228         sighand_cachep = kmem_cache_create("sighand_cache",
1229                         sizeof(struct sighand_struct), 0,
1230                         SLAB_HWCACHE_ALIGN, NULL, NULL);
1231         if (!sighand_cachep)
1232                 panic("Cannot create sighand SLAB cache");
1233
1234         signal_cachep = kmem_cache_create("signal_cache",
1235                         sizeof(struct signal_struct), 0,
1236                         SLAB_HWCACHE_ALIGN, NULL, NULL);
1237         if (!signal_cachep)
1238                 panic("Cannot create signal SLAB cache");
1239
1240         files_cachep = kmem_cache_create("files_cache",
1241                          sizeof(struct files_struct), 0,
1242                          SLAB_HWCACHE_ALIGN, NULL, NULL);
1243         if (!files_cachep)
1244                 panic("Cannot create files SLAB cache");
1245
1246         fs_cachep = kmem_cache_create("fs_cache",
1247                          sizeof(struct fs_struct), 0,
1248                          SLAB_HWCACHE_ALIGN, NULL, NULL);
1249         if (!fs_cachep)
1250                 panic("Cannot create fs_struct SLAB cache");
1251
1252         vm_area_cachep = kmem_cache_create("vm_area_struct",
1253                         sizeof(struct vm_area_struct), 0,
1254                         0, NULL, NULL);
1255         if(!vm_area_cachep)
1256                 panic("vma_init: Cannot alloc vm_area_struct SLAB cache");
1257
1258         mm_cachep = kmem_cache_create("mm_struct",
1259                         sizeof(struct mm_struct), 0,
1260                         SLAB_HWCACHE_ALIGN, NULL, NULL);
1261         if(!mm_cachep)
1262                 panic("vma_init: Cannot alloc mm_struct SLAB cache");
1263 }