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 #include <linux/vinline.h>
  37 #include <linux/ninline.h>
  38
  39 #include <asm/pgtable.h>
  40 #include <asm/pgalloc.h>
  41 #include <asm/uaccess.h>
  42 #include <asm/mmu_context.h>
  43 #include <asm/cacheflush.h>
  44 #include <asm/tlbflush.h>
  45
  46 /* The idle threads do not count..
  47  * Protected by write_lock_irq(&tasklist_lock)
  48  */
  49 int nr_threads;
  50
  51 int max_threads;
  52 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  53
  54 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  55
  56 rwlock_t tasklist_lock __cacheline_aligned = RW_LOCK_UNLOCKED;  /* outer */
  57
  58 EXPORT_SYMBOL(tasklist_lock);
  59
  60 int nr_processes(void)
  61 {
  62         int cpu;
  63         int total = 0;
  64
  65         for_each_cpu(cpu)
  66                 total += per_cpu(process_counts, cpu);
  67
  68         return total;
  69 }
  70
  71 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  72 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  73 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  74 static kmem_cache_t *task_struct_cachep;
  75 #endif
  76
  77 static void free_task(struct task_struct *tsk)
  78 {
  79         free_thread_info(tsk->thread_info);
  80         clr_vx_info(&tsk->vx_info);
  81         clr_nx_info(&tsk->nx_info);
  82         free_task_struct(tsk);
  83 }
  84
  85 void __put_task_struct(struct task_struct *tsk)
  86 {
  87         WARN_ON(!(tsk->state & (TASK_DEAD | TASK_ZOMBIE)));
  88         WARN_ON(atomic_read(&tsk->usage));
  89         WARN_ON(tsk == current);
  90
  91         if (unlikely(tsk->audit_context))
  92                 audit_free(tsk);
  93         security_task_free(tsk);
  94         free_uid(tsk->user);
  95         put_group_info(tsk->group_info);
  96         free_task(tsk);
  97 }
  98
  99 void fastcall add_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
 100 {
 101         unsigned long flags;
 102
 103         wait->flags &= ~WQ_FLAG_EXCLUSIVE;
 104         spin_lock_irqsave(&q->lock, flags);
 105         __add_wait_queue(q, wait);
 106         spin_unlock_irqrestore(&q->lock, flags);
 107 }
 108
 109 EXPORT_SYMBOL(add_wait_queue);
 110
 111 void fastcall add_wait_queue_exclusive(wait_queue_head_t *q, wait_queue_t * wait)
 112 {
 113         unsigned long flags;
 114
 115         wait->flags |= WQ_FLAG_EXCLUSIVE;
 116         spin_lock_irqsave(&q->lock, flags);
 117         __add_wait_queue_tail(q, wait);
 118         spin_unlock_irqrestore(&q->lock, flags);
 119 }
 120
 121 EXPORT_SYMBOL(add_wait_queue_exclusive);
 122
 123 void fastcall remove_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
 124 {
 125         unsigned long flags;
 126
 127         spin_lock_irqsave(&q->lock, flags);
 128         __remove_wait_queue(q, wait);
 129         spin_unlock_irqrestore(&q->lock, flags);
 130 }
 131
 132 EXPORT_SYMBOL(remove_wait_queue);
 133
 134
 135 /*
 136  * Note: we use "set_current_state()" _after_ the wait-queue add,
 137  * because we need a memory barrier there on SMP, so that any
 138  * wake-function that tests for the wait-queue being active
 139  * will be guaranteed to see waitqueue addition _or_ subsequent
 140  * tests in this thread will see the wakeup having taken place.
 141  *
 142  * The spin_unlock() itself is semi-permeable and only protects
 143  * one way (it only protects stuff inside the critical region and
 144  * stops them from bleeding out - it would still allow subsequent
 145  * loads to move into the the critical region).
 146  */
 147 void fastcall prepare_to_wait(wait_queue_head_t *q, wait_queue_t *wait, int state)
 148 {
 149         unsigned long flags;
 150
 151         wait->flags &= ~WQ_FLAG_EXCLUSIVE;
 152         spin_lock_irqsave(&q->lock, flags);
 153         if (list_empty(&wait->task_list))
 154                 __add_wait_queue(q, wait);
 155         set_current_state(state);
 156         spin_unlock_irqrestore(&q->lock, flags);
 157 }
 158
 159 EXPORT_SYMBOL(prepare_to_wait);
 160
 161 void fastcall
 162 prepare_to_wait_exclusive(wait_queue_head_t *q, wait_queue_t *wait, int state)
 163 {
 164         unsigned long flags;
 165
 166         wait->flags |= WQ_FLAG_EXCLUSIVE;
 167         spin_lock_irqsave(&q->lock, flags);
 168         if (list_empty(&wait->task_list))
 169                 __add_wait_queue_tail(q, wait);
 170         set_current_state(state);
 171         spin_unlock_irqrestore(&q->lock, flags);
 172 }
 173
 174 EXPORT_SYMBOL(prepare_to_wait_exclusive);
 175
 176 void fastcall finish_wait(wait_queue_head_t *q, wait_queue_t *wait)
 177 {
 178         unsigned long flags;
 179
 180         __set_current_state(TASK_RUNNING);
 181         /*
 182          * We can check for list emptiness outside the lock
 183          * IFF:
 184          *  - we use the "careful" check that verifies both
 185          *    the next and prev pointers, so that there cannot
 186          *    be any half-pending updates in progress on other
 187          *    CPU's that we haven't seen yet (and that might
 188          *    still change the stack area.
 189          * and
 190          *  - all other users take the lock (ie we can only
 191          *    have _one_ other CPU that looks at or modifies
 192          *    the list).
 193          */
 194         if (!list_empty_careful(&wait->task_list)) {
 195                 spin_lock_irqsave(&q->lock, flags);
 196                 list_del_init(&wait->task_list);
 197                 spin_unlock_irqrestore(&q->lock, flags);
 198         }
 199 }
 200
 201 EXPORT_SYMBOL(finish_wait);
 202
 203 int autoremove_wake_function(wait_queue_t *wait, unsigned mode, int sync)
 204 {
 205         int ret = default_wake_function(wait, mode, sync);
 206
 207         if (ret)
 208                 list_del_init(&wait->task_list);
 209         return ret;
 210 }
 211
 212 EXPORT_SYMBOL(autoremove_wake_function);
 213
 214 void __init fork_init(unsigned long mempages)
 215 {
 216 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 217 #ifndef ARCH_MIN_TASKALIGN
 218 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 219 #endif
 220         /* create a slab on which task_structs can be allocated */
 221         task_struct_cachep =
 222                 kmem_cache_create("task_struct",
 223                                   sizeof(struct task_struct),ARCH_MIN_TASKALIGN,
 224                                   0, NULL, NULL);
 225         if (!task_struct_cachep)
 226                 panic("fork_init(): cannot create task_struct SLAB cache");
 227 #endif
 228
 229         /*
 230          * The default maximum number of threads is set to a safe
 231          * value: the thread structures can take up at most half
 232          * of memory.
 233          */
 234         max_threads = mempages / (THREAD_SIZE/PAGE_SIZE) / 8;
 235         /*
 236          * we need to allow at least 20 threads to boot a system
 237          */
 238         if(max_threads < 20)
 239                 max_threads = 20;
 240
 241         init_task.rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 242         init_task.rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 243 }
 244
 245 static struct task_struct *dup_task_struct(struct task_struct *orig)
 246 {
 247         struct task_struct *tsk;
 248         struct thread_info *ti;
 249
 250         prepare_to_copy(orig);
 251
 252         tsk = alloc_task_struct();
 253         if (!tsk)
 254                 return NULL;
 255
 256         ti = alloc_thread_info(tsk);
 257         if (!ti) {
 258                 free_task_struct(tsk);
 259                 return NULL;
 260         }
 261
 262         *ti = *orig->thread_info;
 263         *tsk = *orig;
 264         tsk->thread_info = ti;
 265         ti->task = tsk;
 266
 267         /* One for us, one for whoever does the "release_task()" (usually parent) */
 268         atomic_set(&tsk->usage,2);
 269         return tsk;
 270 }
 271
 272 #ifdef CONFIG_MMU
 273 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
 274 {
 275         struct vm_area_struct * mpnt, *tmp, **pprev;
 276         struct rb_node **rb_link, *rb_parent;
 277         int retval;
 278         unsigned long charge = 0;
 279
 280         down_write(&oldmm->mmap_sem);
 281         flush_cache_mm(current->mm);
 282         mm->locked_vm = 0;
 283         mm->mmap = NULL;
 284         mm->mmap_cache = NULL;
 285         mm->free_area_cache = TASK_UNMAPPED_BASE;
 286         mm->map_count = 0;
 287         mm->rss = 0;
 288         cpus_clear(mm->cpu_vm_mask);
 289         mm->mm_rb = RB_ROOT;
 290         rb_link = &mm->mm_rb.rb_node;
 291         rb_parent = NULL;
 292         pprev = &mm->mmap;
 293
 294         /*
 295          * Add it to the mmlist after the parent.
 296          * Doing it this way means that we can order the list,
 297          * and fork() won't mess up the ordering significantly.
 298          * Add it first so that swapoff can see any swap entries.
 299          */
 300         spin_lock(&mmlist_lock);
 301         list_add(&mm->mmlist, &current->mm->mmlist);
 302         mmlist_nr++;
 303         spin_unlock(&mmlist_lock);
 304
 305         for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
 306                 struct file *file;
 307
 308                 if(mpnt->vm_flags & VM_DONTCOPY)
 309                         continue;
 310                 if (mpnt->vm_flags & VM_ACCOUNT) {
 311                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 312                         if (security_vm_enough_memory(len))
 313                                 goto fail_nomem;
 314                         charge += len;
 315                 }
 316                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 317                 if (!tmp)
 318                         goto fail_nomem;
 319                 *tmp = *mpnt;
 320                 tmp->vm_flags &= ~VM_LOCKED;
 321                 tmp->vm_mm = mm;
 322                 tmp->vm_next = NULL;
 323                 file = tmp->vm_file;
 324                 INIT_LIST_HEAD(&tmp->shared);
 325                 if (file) {
 326                         struct inode *inode = file->f_dentry->d_inode;
 327                         get_file(file);
 328                         if (tmp->vm_flags & VM_DENYWRITE)
 329                                 atomic_dec(&inode->i_writecount);
 330
 331                         /* insert tmp into the share list, just after mpnt */
 332                         down(&file->f_mapping->i_shared_sem);
 333                         list_add(&tmp->shared, &mpnt->shared);
 334                         up(&file->f_mapping->i_shared_sem);
 335                 }
 336
 337                 /*
 338                  * Link in the new vma and copy the page table entries:
 339                  * link in first so that swapoff can see swap entries,
 340                  * and try_to_unmap_one's find_vma find the new vma.
 341                  */
 342                 spin_lock(&mm->page_table_lock);
 343                 *pprev = tmp;
 344                 pprev = &tmp->vm_next;
 345
 346                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 347                 rb_link = &tmp->vm_rb.rb_right;
 348                 rb_parent = &tmp->vm_rb;
 349
 350                 mm->map_count++;
 351                 retval = copy_page_range(mm, current->mm, tmp);
 352                 spin_unlock(&mm->page_table_lock);
 353
 354                 if (tmp->vm_ops && tmp->vm_ops->open)
 355                         tmp->vm_ops->open(tmp);
 356
 357                 if (retval)
 358                         goto fail;
 359         }
 360         retval = 0;
 361
 362 out:
 363         flush_tlb_mm(current->mm);
 364         up_write(&oldmm->mmap_sem);
 365         return retval;
 366 fail_nomem:
 367         retval = -ENOMEM;
 368 fail:
 369         vm_unacct_memory(charge);
 370         goto out;
 371 }
 372 static inline int mm_alloc_pgd(struct mm_struct * mm)
 373 {
 374         mm->pgd = pgd_alloc(mm);
 375         if (unlikely(!mm->pgd))
 376                 return -ENOMEM;
 377         return 0;
 378 }
 379
 380 static inline void mm_free_pgd(struct mm_struct * mm)
 381 {
 382         pgd_free(mm->pgd);
 383 }
 384 #else
 385 #define dup_mmap(mm, oldmm)     (0)
 386 #define mm_alloc_pgd(mm)        (0)
 387 #define mm_free_pgd(mm)
 388 #endif /* CONFIG_MMU */
 389
 390 spinlock_t mmlist_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
 391 int mmlist_nr;
 392
 393 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
 394 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 395
 396 #include <linux/init_task.h>
 397
 398 static struct mm_struct * mm_init(struct mm_struct * mm)
 399 {
 400         atomic_set(&mm->mm_users, 1);
 401         atomic_set(&mm->mm_count, 1);
 402         init_rwsem(&mm->mmap_sem);
 403         mm->core_waiters = 0;
 404         mm->page_table_lock = SPIN_LOCK_UNLOCKED;
 405         mm->ioctx_list_lock = RW_LOCK_UNLOCKED;
 406         mm->ioctx_list = NULL;
 407         mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
 408         mm->free_area_cache = TASK_UNMAPPED_BASE;
 409
 410         if (likely(!mm_alloc_pgd(mm))) {
 411                 mm->def_flags = 0;
 412                 set_vx_info(&mm->mm_vx_info, current->vx_info);
 413                 return mm;
 414         }
 415         free_mm(mm);
 416         return NULL;
 417 }
 418
 419 /*
 420  * Allocate and initialize an mm_struct.
 421  */
 422 struct mm_struct * mm_alloc(void)
 423 {
 424         struct mm_struct * mm;
 425
 426         mm = allocate_mm();
 427         if (mm) {
 428                 memset(mm, 0, sizeof(*mm));
 429                 return mm_init(mm);
 430         }
 431         return NULL;
 432 }
 433
 434 /*
 435  * Called when the last reference to the mm
 436  * is dropped: either by a lazy thread or by
 437  * mmput. Free the page directory and the mm.
 438  */
 439 void fastcall __mmdrop(struct mm_struct *mm)
 440 {
 441         BUG_ON(mm == &init_mm);
 442         mm_free_pgd(mm);
 443         destroy_context(mm);
 444         clr_vx_info(&mm->mm_vx_info);
 445         free_mm(mm);
 446 }
 447
 448 /*
 449  * Decrement the use count and release all resources for an mm.
 450  */
 451 void mmput(struct mm_struct *mm)
 452 {
 453         if (atomic_dec_and_lock(&mm->mm_users, &mmlist_lock)) {
 454                 list_del(&mm->mmlist);
 455                 mmlist_nr--;
 456                 spin_unlock(&mmlist_lock);
 457                 exit_aio(mm);
 458                 exit_mmap(mm);
 459                 mmdrop(mm);
 460         }
 461 }
 462
 463 /*
 464  * Checks if the use count of an mm is non-zero and if so
 465  * returns a reference to it after bumping up the use count.
 466  * If the use count is zero, it means this mm is going away,
 467  * so return NULL.
 468  */
 469 struct mm_struct *mmgrab(struct mm_struct *mm)
 470 {
 471         spin_lock(&mmlist_lock);
 472         if (!atomic_read(&mm->mm_users))
 473                 mm = NULL;
 474         else
 475                 atomic_inc(&mm->mm_users);
 476         spin_unlock(&mmlist_lock);
 477         return mm;
 478 }
 479
 480 /* Please note the differences between mmput and mm_release.
 481  * mmput is called whenever we stop holding onto a mm_struct,
 482  * error success whatever.
 483  *
 484  * mm_release is called after a mm_struct has been removed
 485  * from the current process.
 486  *
 487  * This difference is important for error handling, when we
 488  * only half set up a mm_struct for a new process and need to restore
 489  * the old one.  Because we mmput the new mm_struct before
 490  * restoring the old one. . .
 491  * Eric Biederman 10 January 1998
 492  */
 493 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 494 {
 495         struct completion *vfork_done = tsk->vfork_done;
 496
 497         /* Get rid of any cached register state */
 498         deactivate_mm(tsk, mm);
 499
 500         /* notify parent sleeping on vfork() */
 501         if (vfork_done) {
 502                 tsk->vfork_done = NULL;
 503                 complete(vfork_done);
 504         }
 505         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
 506                 u32 __user * tidptr = tsk->clear_child_tid;
 507                 tsk->clear_child_tid = NULL;
 508
 509                 /*
 510                  * We don't check the error code - if userspace has
 511                  * not set up a proper pointer then tough luck.
 512                  */
 513                 put_user(0, tidptr);
 514                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL);
 515         }
 516 }
 517
 518 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 519 {
 520         struct mm_struct * mm, *oldmm;
 521         int retval;
 522
 523         tsk->min_flt = tsk->maj_flt = 0;
 524         tsk->cmin_flt = tsk->cmaj_flt = 0;
 525         tsk->nvcsw = tsk->nivcsw = tsk->cnvcsw = tsk->cnivcsw = 0;
 526
 527         tsk->mm = NULL;
 528         tsk->active_mm = NULL;
 529
 530         /*
 531          * Are we cloning a kernel thread?
 532          *
 533          * We need to steal a active VM for that..
 534          */
 535         oldmm = current->mm;
 536         if (!oldmm)
 537                 return 0;
 538
 539         if (clone_flags & CLONE_VM) {
 540                 atomic_inc(&oldmm->mm_users);
 541                 mm = oldmm;
 542                 /*
 543                  * There are cases where the PTL is held to ensure no
 544                  * new threads start up in user mode using an mm, which
 545                  * allows optimizing out ipis; the tlb_gather_mmu code
 546                  * is an example.
 547                  */
 548                 spin_unlock_wait(&oldmm->page_table_lock);
 549                 goto good_mm;
 550         }
 551
 552         retval = -ENOMEM;
 553         mm = allocate_mm();
 554         if (!mm)
 555                 goto fail_nomem;
 556
 557         /* Copy the current MM stuff.. */
 558         memcpy(mm, oldmm, sizeof(*mm));
 559         mm->mm_vx_info = NULL;
 560         if (!mm_init(mm))
 561                 goto fail_nomem;
 562
 563         if (init_new_context(tsk,mm))
 564                 goto fail_nocontext;
 565
 566         retval = dup_mmap(mm, oldmm);
 567         if (retval)
 568                 goto free_pt;
 569
 570 good_mm:
 571         tsk->mm = mm;
 572         tsk->active_mm = mm;
 573         return 0;
 574
 575 free_pt:
 576         mmput(mm);
 577 fail_nomem:
 578         return retval;
 579
 580 fail_nocontext:
 581         /*
 582          * If init_new_context() failed, we cannot use mmput() to free the mm
 583          * because it calls destroy_context()
 584          */
 585         mm_free_pgd(mm);
 586         free_mm(mm);
 587         return retval;
 588 }
 589
 590 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 591 {
 592         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 593         /* We don't need to lock fs - think why ;-) */
 594         if (fs) {
 595                 atomic_set(&fs->count, 1);
 596                 fs->lock = RW_LOCK_UNLOCKED;
 597                 fs->umask = old->umask;
 598                 read_lock(&old->lock);
 599                 fs->rootmnt = mntget(old->rootmnt);
 600                 fs->root = dget(old->root);
 601                 fs->pwdmnt = mntget(old->pwdmnt);
 602                 fs->pwd = dget(old->pwd);
 603                 if (old->altroot) {
 604                         fs->altrootmnt = mntget(old->altrootmnt);
 605                         fs->altroot = dget(old->altroot);
 606                 } else {
 607                         fs->altrootmnt = NULL;
 608                         fs->altroot = NULL;
 609                 }
 610                 read_unlock(&old->lock);
 611         }
 612         return fs;
 613 }
 614
 615 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 616 {
 617         return __copy_fs_struct(old);
 618 }
 619
 620 EXPORT_SYMBOL_GPL(copy_fs_struct);
 621
 622 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 623 {
 624         if (clone_flags & CLONE_FS) {
 625                 atomic_inc(&current->fs->count);
 626                 return 0;
 627         }
 628         tsk->fs = __copy_fs_struct(current->fs);
 629         if (!tsk->fs)
 630                 return -ENOMEM;
 631         return 0;
 632 }
 633
 634 static int count_open_files(struct files_struct *files, int size)
 635 {
 636         int i;
 637
 638         /* Find the last open fd */
 639         for (i = size/(8*sizeof(long)); i > 0; ) {
 640                 if (files->open_fds->fds_bits[--i])
 641                         break;
 642         }
 643         i = (i+1) * 8 * sizeof(long);
 644         return i;
 645 }
 646
 647 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 648 {
 649         struct files_struct *oldf, *newf;
 650         struct file **old_fds, **new_fds;
 651         int open_files, nfds, size, i, error = 0;
 652
 653         /*
 654          * A background process may not have any files ...
 655          */
 656         oldf = current->files;
 657         if (!oldf)
 658                 goto out;
 659
 660         if (clone_flags & CLONE_FILES) {
 661                 atomic_inc(&oldf->count);
 662                 goto out;
 663         }
 664
 665         /*
 666          * Note: we may be using current for both targets (See exec.c)
 667          * This works because we cache current->files (old) as oldf. Don't
 668          * break this.
 669          */
 670         tsk->files = NULL;
 671         error = -ENOMEM;
 672         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
 673         if (!newf)
 674                 goto out;
 675
 676         atomic_set(&newf->count, 1);
 677
 678         newf->file_lock     = SPIN_LOCK_UNLOCKED;
 679         newf->next_fd       = 0;
 680         newf->max_fds       = NR_OPEN_DEFAULT;
 681         newf->max_fdset     = __FD_SETSIZE;
 682         newf->close_on_exec = &newf->close_on_exec_init;
 683         newf->open_fds      = &newf->open_fds_init;
 684         newf->fd            = &newf->fd_array[0];
 685
 686         /* We don't yet have the oldf readlock, but even if the old
 687            fdset gets grown now, we'll only copy up to "size" fds */
 688         size = oldf->max_fdset;
 689         if (size > __FD_SETSIZE) {
 690                 newf->max_fdset = 0;
 691                 spin_lock(&newf->file_lock);
 692                 error = expand_fdset(newf, size-1);
 693                 spin_unlock(&newf->file_lock);
 694                 if (error)
 695                         goto out_release;
 696         }
 697         spin_lock(&oldf->file_lock);
 698
 699         open_files = count_open_files(oldf, size);
 700
 701         /*
 702          * Check whether we need to allocate a larger fd array.
 703          * Note: we're not a clone task, so the open count won't
 704          * change.
 705          */
 706         nfds = NR_OPEN_DEFAULT;
 707         if (open_files > nfds) {
 708                 spin_unlock(&oldf->file_lock);
 709                 newf->max_fds = 0;
 710                 spin_lock(&newf->file_lock);
 711                 error = expand_fd_array(newf, open_files-1);
 712                 spin_unlock(&newf->file_lock);
 713                 if (error)
 714                         goto out_release;
 715                 nfds = newf->max_fds;
 716                 spin_lock(&oldf->file_lock);
 717         }
 718
 719         old_fds = oldf->fd;
 720         new_fds = newf->fd;
 721
 722         memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
 723         memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
 724
 725         for (i = open_files; i != 0; i--) {
 726                 struct file *f = *old_fds++;
 727                 if (f)
 728                         get_file(f);
 729                 *new_fds++ = f;
 730         }
 731         spin_unlock(&oldf->file_lock);
 732
 733         /* compute the remainder to be cleared */
 734         size = (newf->max_fds - open_files) * sizeof(struct file *);
 735
 736         /* This is long word aligned thus could use a optimized version */
 737         memset(new_fds, 0, size);
 738
 739         if (newf->max_fdset > open_files) {
 740                 int left = (newf->max_fdset-open_files)/8;
 741                 int start = open_files / (8 * sizeof(unsigned long));
 742
 743                 memset(&newf->open_fds->fds_bits[start], 0, left);
 744                 memset(&newf->close_on_exec->fds_bits[start], 0, left);
 745         }
 746
 747         tsk->files = newf;
 748         error = 0;
 749 out:
 750         return error;
 751
 752 out_release:
 753         free_fdset (newf->close_on_exec, newf->max_fdset);
 754         free_fdset (newf->open_fds, newf->max_fdset);
 755         kmem_cache_free(files_cachep, newf);
 756         goto out;
 757 }
 758
 759 /*
 760  *      Helper to unshare the files of the current task.
 761  *      We don't want to expose copy_files internals to
 762  *      the exec layer of the kernel.
 763  */
 764
 765 int unshare_files(void)
 766 {
 767         struct files_struct *files  = current->files;
 768         int rc;
 769
 770         if(!files)
 771                 BUG();
 772
 773         /* This can race but the race causes us to copy when we don't
 774            need to and drop the copy */
 775         if(atomic_read(&files->count) == 1)
 776         {
 777                 atomic_inc(&files->count);
 778                 return 0;
 779         }
 780         rc = copy_files(0, current);
 781         if(rc)
 782                 current->files = files;
 783         return rc;
 784 }
 785
 786 EXPORT_SYMBOL(unshare_files);
 787
 788 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 789 {
 790         struct sighand_struct *sig;
 791
 792         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 793                 atomic_inc(&current->sighand->count);
 794                 return 0;
 795         }
 796         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 797         tsk->sighand = sig;
 798         if (!sig)
 799                 return -ENOMEM;
 800         spin_lock_init(&sig->siglock);
 801         atomic_set(&sig->count, 1);
 802         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 803         return 0;
 804 }
 805
 806 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 807 {
 808         struct signal_struct *sig;
 809
 810         if (clone_flags & CLONE_THREAD) {
 811                 atomic_inc(&current->signal->count);
 812                 return 0;
 813         }
 814         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 815         tsk->signal = sig;
 816         if (!sig)
 817                 return -ENOMEM;
 818         atomic_set(&sig->count, 1);
 819         sig->group_exit = 0;
 820         sig->group_exit_code = 0;
 821         sig->group_exit_task = NULL;
 822         sig->group_stop_count = 0;
 823         sig->curr_target = NULL;
 824         init_sigpending(&sig->shared_pending);
 825         INIT_LIST_HEAD(&sig->posix_timers);
 826
 827         sig->tty = current->signal->tty;
 828         sig->pgrp = process_group(current);
 829         sig->session = current->signal->session;
 830         sig->leader = 0;        /* session leadership doesn't inherit */
 831         sig->tty_old_pgrp = 0;
 832
 833         return 0;
 834 }
 835
 836 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 837 {
 838         unsigned long new_flags = p->flags;
 839
 840         new_flags &= ~PF_SUPERPRIV;
 841         new_flags |= PF_FORKNOEXEC;
 842         if (!(clone_flags & CLONE_PTRACE))
 843                 p->ptrace = 0;
 844         p->flags = new_flags;
 845 }
 846
 847 asmlinkage long sys_set_tid_address(int __user *tidptr)
 848 {
 849         current->clear_child_tid = tidptr;
 850
 851         return current->pid;
 852 }
 853
 854 /*
 855  * This creates a new process as a copy of the old one,
 856  * but does not actually start it yet.
 857  *
 858  * It copies the registers, and all the appropriate
 859  * parts of the process environment (as per the clone
 860  * flags). The actual kick-off is left to the caller.
 861  */
 862 struct task_struct *copy_process(unsigned long clone_flags,
 863                                  unsigned long stack_start,
 864                                  struct pt_regs *regs,
 865                                  unsigned long stack_size,
 866                                  int __user *parent_tidptr,
 867                                  int __user *child_tidptr)
 868 {
 869         int retval;
 870         struct task_struct *p = NULL;
 871         struct vx_info *vxi;
 872         struct nx_info *nxi;
 873
 874         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 875                 return ERR_PTR(-EINVAL);
 876
 877         /*
 878          * Thread groups must share signals as well, and detached threads
 879          * can only be started up within the thread group.
 880          */
 881         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 882                 return ERR_PTR(-EINVAL);
 883
 884         /*
 885          * Shared signal handlers imply shared VM. By way of the above,
 886          * thread groups also imply shared VM. Blocking this case allows
 887          * for various simplifications in other code.
 888          */
 889         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 890                 return ERR_PTR(-EINVAL);
 891
 892         retval = security_task_create(clone_flags);
 893         if (retval)
 894                 goto fork_out;
 895
 896         retval = -ENOMEM;
 897
 898         p = dup_task_struct(current);
 899         if (!p)
 900                 goto fork_out;
 901
 902         vxi = get_vx_info(current->vx_info);
 903         nxi = get_nx_info(current->nx_info);
 904
 905         /* check vserver memory */
 906         if (p->mm && !(clone_flags & CLONE_VM)) {
 907                 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
 908                         vx_pages_add(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
 909                 else
 910                         goto bad_fork_free;
 911         }
 912         if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
 913                 if (!vx_rsspages_avail(p->mm, p->mm->rss))
 914                         goto bad_fork_free;
 915         }
 916
 917         retval = -EAGAIN;
 918         if (vxi && (atomic_read(&vxi->limit.res[RLIMIT_NPROC])
 919                 >= vxi->limit.rlim[RLIMIT_NPROC]))
 920                 goto bad_fork_free;
 921
 922         if (atomic_read(&p->user->processes) >=
 923                         p->rlim[RLIMIT_NPROC].rlim_cur) {
 924                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 925                                 p->user != &root_user)
 926                         goto bad_fork_free;
 927         }
 928
 929         atomic_inc(&p->user->__count);
 930         atomic_inc(&p->user->processes);
 931         get_group_info(p->group_info);
 932
 933         /*
 934          * If multiple threads are within copy_process(), then this check
 935          * triggers too late. This doesn't hurt, the check is only there
 936          * to stop root fork bombs.
 937          */
 938         if (nr_threads >= max_threads)
 939                 goto bad_fork_cleanup_count;
 940
 941         if (!try_module_get(p->thread_info->exec_domain->module))
 942                 goto bad_fork_cleanup_count;
 943
 944         if (p->binfmt && !try_module_get(p->binfmt->module))
 945                 goto bad_fork_cleanup_put_domain;
 946
 947         p->did_exec = 0;
 948         copy_flags(clone_flags, p);
 949         if (clone_flags & CLONE_IDLETASK)
 950                 p->pid = 0;
 951         else {
 952                 p->pid = alloc_pidmap();
 953                 if (p->pid == -1)
 954                         goto bad_fork_cleanup;
 955         }
 956         retval = -EFAULT;
 957         if (clone_flags & CLONE_PARENT_SETTID)
 958                 if (put_user(p->pid, parent_tidptr))
 959                         goto bad_fork_cleanup;
 960
 961         p->proc_dentry = NULL;
 962
 963         INIT_LIST_HEAD(&p->children);
 964         INIT_LIST_HEAD(&p->sibling);
 965         init_waitqueue_head(&p->wait_chldexit);
 966         p->vfork_done = NULL;
 967         spin_lock_init(&p->alloc_lock);
 968         spin_lock_init(&p->proc_lock);
 969
 970         clear_tsk_thread_flag(p, TIF_SIGPENDING);
 971         init_sigpending(&p->pending);
 972
 973         p->it_real_value = p->it_virt_value = p->it_prof_value = 0;
 974         p->it_real_incr = p->it_virt_incr = p->it_prof_incr = 0;
 975         init_timer(&p->real_timer);
 976         p->real_timer.data = (unsigned long) p;
 977
 978         p->utime = p->stime = 0;
 979         p->cutime = p->cstime = 0;
 980         p->lock_depth = -1;             /* -1 = no lock */
 981         p->start_time = get_jiffies_64();
 982         p->security = NULL;
 983         p->io_context = NULL;
 984         p->audit_context = NULL;
 985
 986         retval = -ENOMEM;
 987         if ((retval = security_task_alloc(p)))
 988                 goto bad_fork_cleanup;
 989         if ((retval = audit_alloc(p)))
 990                 goto bad_fork_cleanup_security;
 991         /* copy all the process information */
 992         if ((retval = copy_semundo(clone_flags, p)))
 993                 goto bad_fork_cleanup_audit;
 994         if ((retval = copy_files(clone_flags, p)))
 995                 goto bad_fork_cleanup_semundo;
 996         if ((retval = copy_fs(clone_flags, p)))
 997                 goto bad_fork_cleanup_files;
 998         if ((retval = copy_sighand(clone_flags, p)))
 999                 goto bad_fork_cleanup_fs;
1000         if ((retval = copy_signal(clone_flags, p)))
1001                 goto bad_fork_cleanup_sighand;
1002         if ((retval = copy_mm(clone_flags, p)))
1003                 goto bad_fork_cleanup_signal;
1004         if ((retval = copy_namespace(clone_flags, p)))
1005                 goto bad_fork_cleanup_mm;
1006         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1007         if (retval)
1008                 goto bad_fork_cleanup_namespace;
1009
1010         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1011         /*
1012          * Clear TID on mm_release()?
1013          */
1014         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1015
1016         /*
1017          * Syscall tracing should be turned off in the child regardless
1018          * of CLONE_PTRACE.
1019          */
1020         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1021
1022         /* Our parent execution domain becomes current domain
1023            These must match for thread signalling to apply */
1024
1025         p->parent_exec_id = p->self_exec_id;
1026
1027         /* ok, now we should be set up.. */
1028         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1029         p->pdeath_signal = 0;
1030
1031         /* Perform scheduler related setup */
1032         sched_fork(p);
1033
1034         /*
1035          * Ok, make it visible to the rest of the system.
1036          * We dont wake it up yet.
1037          */
1038         p->tgid = p->pid;
1039         p->group_leader = p;
1040         INIT_LIST_HEAD(&p->ptrace_children);
1041         INIT_LIST_HEAD(&p->ptrace_list);
1042
1043         /* Need tasklist lock for parent etc handling! */
1044         write_lock_irq(&tasklist_lock);
1045         /*
1046          * Check for pending SIGKILL! The new thread should not be allowed
1047          * to slip out of an OOM kill. (or normal SIGKILL.)
1048          */
1049         if (sigismember(&current->pending.signal, SIGKILL)) {
1050                 write_unlock_irq(&tasklist_lock);
1051                 retval = -EINTR;
1052                 goto bad_fork_cleanup_namespace;
1053         }
1054
1055         /* CLONE_PARENT re-uses the old parent */
1056         if (clone_flags & CLONE_PARENT)
1057                 p->real_parent = current->real_parent;
1058         else
1059                 p->real_parent = current;
1060         p->parent = p->real_parent;
1061
1062         if (clone_flags & CLONE_THREAD) {
1063                 spin_lock(&current->sighand->siglock);
1064                 /*
1065                  * Important: if an exit-all has been started then
1066                  * do not create this new thread - the whole thread
1067                  * group is supposed to exit anyway.
1068                  */
1069                 if (current->signal->group_exit) {
1070                         spin_unlock(&current->sighand->siglock);
1071                         write_unlock_irq(&tasklist_lock);
1072                         retval = -EAGAIN;
1073                         goto bad_fork_cleanup_namespace;
1074                 }
1075                 p->tgid = current->tgid;
1076                 p->group_leader = current->group_leader;
1077
1078                 if (current->signal->group_stop_count > 0) {
1079                         /*
1080                          * There is an all-stop in progress for the group.
1081                          * We ourselves will stop as soon as we check signals.
1082                          * Make the new thread part of that group stop too.
1083                          */
1084                         current->signal->group_stop_count++;
1085                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1086                 }
1087
1088                 spin_unlock(&current->sighand->siglock);
1089         }
1090
1091         SET_LINKS(p);
1092         if (p->ptrace & PT_PTRACED)
1093                 __ptrace_link(p, current->parent);
1094
1095         attach_pid(p, PIDTYPE_PID, p->pid);
1096         if (thread_group_leader(p)) {
1097                 attach_pid(p, PIDTYPE_TGID, p->tgid);
1098                 attach_pid(p, PIDTYPE_PGID, process_group(p));
1099                 attach_pid(p, PIDTYPE_SID, p->signal->session);
1100                 if (p->pid)
1101                         __get_cpu_var(process_counts)++;
1102         } else
1103                 link_pid(p, p->pids + PIDTYPE_TGID, &p->group_leader->pids[PIDTYPE_TGID].pid);
1104
1105         nr_threads++;
1106         if (vxi) {
1107                 atomic_inc(&vxi->cacct.nr_threads);
1108                 atomic_inc(&vxi->limit.res[RLIMIT_NPROC]);
1109         }
1110         write_unlock_irq(&tasklist_lock);
1111         retval = 0;
1112
1113 fork_out:
1114         if (retval)
1115                 return ERR_PTR(retval);
1116         return p;
1117
1118 bad_fork_cleanup_namespace:
1119         exit_namespace(p);
1120 bad_fork_cleanup_mm:
1121         exit_mm(p);
1122         if (p->active_mm)
1123                 mmdrop(p->active_mm);
1124 bad_fork_cleanup_signal:
1125         exit_signal(p);
1126 bad_fork_cleanup_sighand:
1127         exit_sighand(p);
1128 bad_fork_cleanup_fs:
1129         exit_fs(p); /* blocking */
1130 bad_fork_cleanup_files:
1131         exit_files(p); /* blocking */
1132 bad_fork_cleanup_semundo:
1133         exit_sem(p);
1134 bad_fork_cleanup_audit:
1135         audit_free(p);
1136 bad_fork_cleanup_security:
1137         security_task_free(p);
1138 bad_fork_cleanup:
1139         if (p->pid > 0)
1140                 free_pidmap(p->pid);
1141         if (p->binfmt)
1142                 module_put(p->binfmt->module);
1143 bad_fork_cleanup_put_domain:
1144         module_put(p->thread_info->exec_domain->module);
1145 bad_fork_cleanup_count:
1146         put_group_info(p->group_info);
1147         atomic_dec(&p->user->processes);
1148         free_uid(p->user);
1149 bad_fork_free:
1150         free_task(p);
1151         goto fork_out;
1152 }
1153
1154 static inline int fork_traceflag (unsigned clone_flags)
1155 {
1156         if (clone_flags & (CLONE_UNTRACED | CLONE_IDLETASK))
1157                 return 0;
1158         else if (clone_flags & CLONE_VFORK) {
1159                 if (current->ptrace & PT_TRACE_VFORK)
1160                         return PTRACE_EVENT_VFORK;
1161         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1162                 if (current->ptrace & PT_TRACE_CLONE)
1163                         return PTRACE_EVENT_CLONE;
1164         } else if (current->ptrace & PT_TRACE_FORK)
1165                 return PTRACE_EVENT_FORK;
1166
1167         return 0;
1168 }
1169
1170 /*
1171  *  Ok, this is the main fork-routine.
1172  *
1173  * It copies the process, and if successful kick-starts
1174  * it and waits for it to finish using the VM if required.
1175  */
1176 long do_fork(unsigned long clone_flags,
1177               unsigned long stack_start,
1178               struct pt_regs *regs,
1179               unsigned long stack_size,
1180               int __user *parent_tidptr,
1181               int __user *child_tidptr)
1182 {
1183         struct task_struct *p;
1184         int trace = 0;
1185         long pid;
1186
1187         if (unlikely(current->ptrace)) {
1188                 trace = fork_traceflag (clone_flags);
1189                 if (trace)
1190                         clone_flags |= CLONE_PTRACE;
1191         }
1192
1193         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr);
1194         /*
1195          * Do this prior waking up the new thread - the thread pointer
1196          * might get invalid after that point, if the thread exits quickly.
1197          */
1198         pid = IS_ERR(p) ? PTR_ERR(p) : p->pid;
1199
1200         if (!IS_ERR(p)) {
1201                 struct completion vfork;
1202
1203                 if (clone_flags & CLONE_VFORK) {
1204                         p->vfork_done = &vfork;
1205                         init_completion(&vfork);
1206                 }
1207
1208                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1209                         /*
1210                          * We'll start up with an immediate SIGSTOP.
1211                          */
1212                         sigaddset(&p->pending.signal, SIGSTOP);
1213                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1214                 }
1215
1216                 if (!(clone_flags & CLONE_STOPPED))
1217                         wake_up_forked_process(p);      /* do this last */
1218                 else
1219                         p->state = TASK_STOPPED;
1220                 ++total_forks;
1221
1222                 if (unlikely (trace)) {
1223                         current->ptrace_message = pid;
1224                         ptrace_notify ((trace << 8) | SIGTRAP);
1225                 }
1226
1227                 if (clone_flags & CLONE_VFORK) {
1228                         wait_for_completion(&vfork);
1229                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1230                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1231                 } else
1232                         /*
1233                          * Let the child process run first, to avoid most of the
1234                          * COW overhead when the child exec()s afterwards.
1235                          */
1236                         set_need_resched();
1237         }
1238         return pid;
1239 }
1240
1241 /* SLAB cache for signal_struct structures (tsk->signal) */
1242 kmem_cache_t *signal_cachep;
1243
1244 /* SLAB cache for sighand_struct structures (tsk->sighand) */
1245 kmem_cache_t *sighand_cachep;
1246
1247 /* SLAB cache for files_struct structures (tsk->files) */
1248 kmem_cache_t *files_cachep;
1249
1250 /* SLAB cache for fs_struct structures (tsk->fs) */
1251 kmem_cache_t *fs_cachep;
1252
1253 /* SLAB cache for vm_area_struct structures */
1254 kmem_cache_t *vm_area_cachep;
1255
1256 /* SLAB cache for mm_struct structures (tsk->mm) */
1257 kmem_cache_t *mm_cachep;
1258
1259 void __init proc_caches_init(void)
1260 {
1261         sighand_cachep = kmem_cache_create("sighand_cache",
1262                         sizeof(struct sighand_struct), 0,
1263                         SLAB_HWCACHE_ALIGN, NULL, NULL);
1264         if (!sighand_cachep)
1265                 panic("Cannot create sighand SLAB cache");
1266
1267         signal_cachep = kmem_cache_create("signal_cache",
1268                         sizeof(struct signal_struct), 0,
1269                         SLAB_HWCACHE_ALIGN, NULL, NULL);
1270         if (!signal_cachep)
1271                 panic("Cannot create signal SLAB cache");
1272
1273         files_cachep = kmem_cache_create("files_cache",
1274                          sizeof(struct files_struct), 0,
1275                          SLAB_HWCACHE_ALIGN, NULL, NULL);
1276         if (!files_cachep)
1277                 panic("Cannot create files SLAB cache");
1278
1279         fs_cachep = kmem_cache_create("fs_cache",
1280                          sizeof(struct fs_struct), 0,
1281                          SLAB_HWCACHE_ALIGN, NULL, NULL);
1282         if (!fs_cachep)
1283                 panic("Cannot create fs_struct SLAB cache");
1284
1285         vm_area_cachep = kmem_cache_create("vm_area_struct",
1286                         sizeof(struct vm_area_struct), 0,
1287                         0, NULL, NULL);
1288         if(!vm_area_cachep)
1289                 panic("vma_init: Cannot alloc vm_area_struct SLAB cache");
1290
1291         mm_cachep = kmem_cache_create("mm_struct",
1292                         sizeof(struct mm_struct), 0,
1293                         SLAB_HWCACHE_ALIGN, NULL, NULL);
1294         if(!mm_cachep)
1295                 panic("vma_init: Cannot alloc mm_struct SLAB cache");
1296 }