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/file.h>
  25 #include <linux/binfmts.h>
  26 #include <linux/mman.h>
  27 #include <linux/fs.h>
  28 #include <linux/security.h>
  29 #include <linux/jiffies.h>
  30 #include <linux/futex.h>
  31 #include <linux/ptrace.h>
  32 #include <linux/mount.h>
  33
  34 #include <asm/pgtable.h>
  35 #include <asm/pgalloc.h>
  36 #include <asm/uaccess.h>
  37 #include <asm/mmu_context.h>
  38 #include <asm/cacheflush.h>
  39 #include <asm/tlbflush.h>
  40
  41 static kmem_cache_t *task_struct_cachep;
  42
  43 extern int copy_semundo(unsigned long clone_flags, struct task_struct *tsk);
  44 extern void exit_semundo(struct task_struct *tsk);
  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 /*
  59  * A per-CPU task cache - this relies on the fact that
  60  * the very last portion of sys_exit() is executed with
  61  * preemption turned off.
  62  */
  63 static task_t *task_cache[NR_CPUS] __cacheline_aligned;
  64
  65 int nr_processes(void)
  66 {
  67         int cpu;
  68         int total = 0;
  69
  70         for (cpu = 0; cpu < NR_CPUS; cpu++) {
  71                 if (cpu_online(cpu))
  72                         total += per_cpu(process_counts, cpu);
  73         }
  74         return total;
  75 }
  76
  77 static void free_task_struct(struct task_struct *tsk)
  78 {
  79         /*
  80          * The task cache is effectively disabled right now.
  81          * Do we want it? The slab cache already has per-cpu
  82          * stuff, but the thread info (usually a order-1 page
  83          * allocation) doesn't.
  84          */
  85         if (tsk != current) {
  86                 free_thread_info(tsk->thread_info);
  87                 kmem_cache_free(task_struct_cachep,tsk);
  88         } else {
  89                 int cpu = get_cpu();
  90
  91                 tsk = task_cache[cpu];
  92                 if (tsk) {
  93                         free_thread_info(tsk->thread_info);
  94                         kmem_cache_free(task_struct_cachep,tsk);
  95                 }
  96                 task_cache[cpu] = current;
  97                 put_cpu();
  98         }
  99 }
 100
 101 void __put_task_struct(struct task_struct *tsk)
 102 {
 103         WARN_ON(!(tsk->state & (TASK_DEAD | TASK_ZOMBIE)));
 104         WARN_ON(atomic_read(&tsk->usage));
 105         WARN_ON(tsk == current);
 106
 107         security_task_free(tsk);
 108         free_uid(tsk->user);
 109         free_task_struct(tsk);
 110 }
 111
 112 void add_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
 113 {
 114         unsigned long flags;
 115
 116         wait->flags &= ~WQ_FLAG_EXCLUSIVE;
 117         spin_lock_irqsave(&q->lock, flags);
 118         __add_wait_queue(q, wait);
 119         spin_unlock_irqrestore(&q->lock, flags);
 120 }
 121
 122 void add_wait_queue_exclusive(wait_queue_head_t *q, wait_queue_t * wait)
 123 {
 124         unsigned long flags;
 125
 126         wait->flags |= WQ_FLAG_EXCLUSIVE;
 127         spin_lock_irqsave(&q->lock, flags);
 128         __add_wait_queue_tail(q, wait);
 129         spin_unlock_irqrestore(&q->lock, flags);
 130 }
 131
 132 void remove_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
 133 {
 134         unsigned long flags;
 135
 136         spin_lock_irqsave(&q->lock, flags);
 137         __remove_wait_queue(q, wait);
 138         spin_unlock_irqrestore(&q->lock, flags);
 139 }
 140
 141 void prepare_to_wait(wait_queue_head_t *q, wait_queue_t *wait, int state)
 142 {
 143         unsigned long flags;
 144
 145         __set_current_state(state);
 146         wait->flags &= ~WQ_FLAG_EXCLUSIVE;
 147         spin_lock_irqsave(&q->lock, flags);
 148         if (list_empty(&wait->task_list))
 149                 __add_wait_queue(q, wait);
 150         spin_unlock_irqrestore(&q->lock, flags);
 151 }
 152
 153 void
 154 prepare_to_wait_exclusive(wait_queue_head_t *q, wait_queue_t *wait, int state)
 155 {
 156         unsigned long flags;
 157
 158         __set_current_state(state);
 159         wait->flags |= WQ_FLAG_EXCLUSIVE;
 160         spin_lock_irqsave(&q->lock, flags);
 161         if (list_empty(&wait->task_list))
 162                 __add_wait_queue_tail(q, wait);
 163         spin_unlock_irqrestore(&q->lock, flags);
 164 }
 165
 166 void finish_wait(wait_queue_head_t *q, wait_queue_t *wait)
 167 {
 168         unsigned long flags;
 169
 170         __set_current_state(TASK_RUNNING);
 171         if (!list_empty(&wait->task_list)) {
 172                 spin_lock_irqsave(&q->lock, flags);
 173                 list_del_init(&wait->task_list);
 174                 spin_unlock_irqrestore(&q->lock, flags);
 175         }
 176 }
 177
 178 int autoremove_wake_function(wait_queue_t *wait, unsigned mode, int sync)
 179 {
 180         int ret = default_wake_function(wait, mode, sync);
 181
 182         if (ret)
 183                 list_del_init(&wait->task_list);
 184         return ret;
 185 }
 186
 187 void __init fork_init(unsigned long mempages)
 188 {
 189         /* create a slab on which task_structs can be allocated */
 190         task_struct_cachep =
 191                 kmem_cache_create("task_struct",
 192                                   sizeof(struct task_struct),0,
 193                                   SLAB_MUST_HWCACHE_ALIGN, NULL, NULL);
 194         if (!task_struct_cachep)
 195                 panic("fork_init(): cannot create task_struct SLAB cache");
 196
 197         /*
 198          * The default maximum number of threads is set to a safe
 199          * value: the thread structures can take up at most half
 200          * of memory.
 201          */
 202         max_threads = mempages / (THREAD_SIZE/PAGE_SIZE) / 8;
 203         /*
 204          * we need to allow at least 20 threads to boot a system
 205          */
 206         if(max_threads < 20)
 207                 max_threads = 20;
 208
 209         init_task.rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 210         init_task.rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 211 }
 212
 213 static struct task_struct *dup_task_struct(struct task_struct *orig)
 214 {
 215         struct task_struct *tsk;
 216         struct thread_info *ti;
 217         int cpu = get_cpu();
 218
 219         prepare_to_copy(orig);
 220
 221         tsk = task_cache[cpu];
 222         task_cache[cpu] = NULL;
 223         put_cpu();
 224         if (!tsk) {
 225                 ti = alloc_thread_info();
 226                 if (!ti)
 227                         return NULL;
 228
 229                 tsk = kmem_cache_alloc(task_struct_cachep, GFP_KERNEL);
 230                 if (!tsk) {
 231                         free_thread_info(ti);
 232                         return NULL;
 233                 }
 234         } else
 235                 ti = tsk->thread_info;
 236
 237         *ti = *orig->thread_info;
 238         *tsk = *orig;
 239         tsk->thread_info = ti;
 240         ti->task = tsk;
 241
 242         /* One for us, one for whoever does the "release_task()" (usually parent) */
 243         atomic_set(&tsk->usage,2);
 244         return tsk;
 245 }
 246
 247 #ifdef CONFIG_MMU
 248 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
 249 {
 250         struct vm_area_struct * mpnt, *tmp, **pprev;
 251         int retval;
 252         unsigned long charge = 0;
 253
 254         down_write(&oldmm->mmap_sem);
 255         flush_cache_mm(current->mm);
 256         mm->locked_vm = 0;
 257         mm->mmap = NULL;
 258         mm->mmap_cache = NULL;
 259         mm->free_area_cache = TASK_UNMAPPED_BASE;
 260         mm->map_count = 0;
 261         mm->rss = 0;
 262         mm->cpu_vm_mask = 0;
 263         pprev = &mm->mmap;
 264
 265         /*
 266          * Add it to the mmlist after the parent.
 267          * Doing it this way means that we can order the list,
 268          * and fork() won't mess up the ordering significantly.
 269          * Add it first so that swapoff can see any swap entries.
 270          */
 271         spin_lock(&mmlist_lock);
 272         list_add(&mm->mmlist, &current->mm->mmlist);
 273         mmlist_nr++;
 274         spin_unlock(&mmlist_lock);
 275
 276         for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
 277                 struct file *file;
 278
 279                 if(mpnt->vm_flags & VM_DONTCOPY)
 280                         continue;
 281                 if (mpnt->vm_flags & VM_ACCOUNT) {
 282                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 283                         if (!vm_enough_memory(len))
 284                                 goto fail_nomem;
 285                         charge += len;
 286                 }
 287                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 288                 if (!tmp)
 289                         goto fail_nomem;
 290                 *tmp = *mpnt;
 291                 tmp->vm_flags &= ~VM_LOCKED;
 292                 tmp->vm_mm = mm;
 293                 tmp->vm_next = NULL;
 294                 file = tmp->vm_file;
 295                 INIT_LIST_HEAD(&tmp->shared);
 296                 if (file) {
 297                         struct inode *inode = file->f_dentry->d_inode;
 298                         get_file(file);
 299                         if (tmp->vm_flags & VM_DENYWRITE)
 300                                 atomic_dec(&inode->i_writecount);
 301
 302                         /* insert tmp into the share list, just after mpnt */
 303                         down(&inode->i_mapping->i_shared_sem);
 304                         list_add_tail(&tmp->shared, &mpnt->shared);
 305                         up(&inode->i_mapping->i_shared_sem);
 306                 }
 307
 308                 /*
 309                  * Link in the new vma and copy the page table entries:
 310                  * link in first so that swapoff can see swap entries.
 311                  */
 312                 spin_lock(&mm->page_table_lock);
 313                 *pprev = tmp;
 314                 pprev = &tmp->vm_next;
 315                 mm->map_count++;
 316                 retval = copy_page_range(mm, current->mm, tmp);
 317                 spin_unlock(&mm->page_table_lock);
 318
 319                 if (tmp->vm_ops && tmp->vm_ops->open)
 320                         tmp->vm_ops->open(tmp);
 321
 322                 if (retval)
 323                         goto fail;
 324         }
 325         retval = 0;
 326         build_mmap_rb(mm);
 327
 328 out:
 329         flush_tlb_mm(current->mm);
 330         up_write(&oldmm->mmap_sem);
 331         return retval;
 332 fail_nomem:
 333         retval = -ENOMEM;
 334   fail:
 335         vm_unacct_memory(charge);
 336         goto out;
 337 }
 338 static inline int mm_alloc_pgd(struct mm_struct * mm)
 339 {
 340         mm->pgd = pgd_alloc(mm);
 341         if (unlikely(!mm->pgd))
 342                 return -ENOMEM;
 343         return 0;
 344 }
 345
 346 static inline void mm_free_pgd(struct mm_struct * mm)
 347 {
 348         pgd_free(mm->pgd);
 349 }
 350 #else
 351 #define dup_mmap(mm, oldmm)     (0)
 352 #define mm_alloc_pgd(mm)        (0)
 353 #define mm_free_pgd(mm)
 354 #endif /* CONFIG_MMU */
 355
 356 spinlock_t mmlist_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
 357 int mmlist_nr;
 358
 359 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
 360 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 361
 362 #include <linux/init_task.h>
 363
 364 static struct mm_struct * mm_init(struct mm_struct * mm)
 365 {
 366         atomic_set(&mm->mm_users, 1);
 367         atomic_set(&mm->mm_count, 1);
 368         init_rwsem(&mm->mmap_sem);
 369         mm->core_waiters = 0;
 370         mm->page_table_lock = SPIN_LOCK_UNLOCKED;
 371         mm->ioctx_list_lock = RW_LOCK_UNLOCKED;
 372         mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
 373         mm->free_area_cache = TASK_UNMAPPED_BASE;
 374
 375         if (likely(!mm_alloc_pgd(mm))) {
 376                 mm->def_flags = 0;
 377                 return mm;
 378         }
 379         free_mm(mm);
 380         return NULL;
 381 }
 382
 383 /*
 384  * Allocate and initialize an mm_struct.
 385  */
 386 struct mm_struct * mm_alloc(void)
 387 {
 388         struct mm_struct * mm;
 389
 390         mm = allocate_mm();
 391         if (mm) {
 392                 memset(mm, 0, sizeof(*mm));
 393                 return mm_init(mm);
 394         }
 395         return NULL;
 396 }
 397
 398 /*
 399  * Called when the last reference to the mm
 400  * is dropped: either by a lazy thread or by
 401  * mmput. Free the page directory and the mm.
 402  */
 403 inline void __mmdrop(struct mm_struct *mm)
 404 {
 405         BUG_ON(mm == &init_mm);
 406         mm_free_pgd(mm);
 407         destroy_context(mm);
 408         free_mm(mm);
 409 }
 410
 411 /*
 412  * Decrement the use count and release all resources for an mm.
 413  */
 414 void mmput(struct mm_struct *mm)
 415 {
 416         if (atomic_dec_and_lock(&mm->mm_users, &mmlist_lock)) {
 417                 list_del(&mm->mmlist);
 418                 mmlist_nr--;
 419                 spin_unlock(&mmlist_lock);
 420                 exit_aio(mm);
 421                 exit_mmap(mm);
 422                 mmdrop(mm);
 423         }
 424 }
 425
 426 /* Please note the differences between mmput and mm_release.
 427  * mmput is called whenever we stop holding onto a mm_struct,
 428  * error success whatever.
 429  *
 430  * mm_release is called after a mm_struct has been removed
 431  * from the current process.
 432  *
 433  * This difference is important for error handling, when we
 434  * only half set up a mm_struct for a new process and need to restore
 435  * the old one.  Because we mmput the new mm_struct before
 436  * restoring the old one. . .
 437  * Eric Biederman 10 January 1998
 438  */
 439 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 440 {
 441         struct completion *vfork_done = tsk->vfork_done;
 442
 443         /* Get rid of any cached register state */
 444         deactivate_mm(tsk, mm);
 445
 446         /* notify parent sleeping on vfork() */
 447         if (vfork_done) {
 448                 tsk->vfork_done = NULL;
 449                 complete(vfork_done);
 450         }
 451         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
 452                 u32 __user * tidptr = tsk->clear_child_tid;
 453                 tsk->clear_child_tid = NULL;
 454
 455                 /*
 456                  * We don't check the error code - if userspace has
 457                  * not set up a proper pointer then tough luck.
 458                  */
 459                 put_user(0, tidptr);
 460                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL);
 461         }
 462 }
 463
 464 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 465 {
 466         struct mm_struct * mm, *oldmm;
 467         int retval;
 468
 469         tsk->min_flt = tsk->maj_flt = 0;
 470         tsk->cmin_flt = tsk->cmaj_flt = 0;
 471         tsk->nswap = tsk->cnswap = 0;
 472
 473         tsk->mm = NULL;
 474         tsk->active_mm = NULL;
 475
 476         /*
 477          * Are we cloning a kernel thread?
 478          *
 479          * We need to steal a active VM for that..
 480          */
 481         oldmm = current->mm;
 482         if (!oldmm)
 483                 return 0;
 484
 485         if (clone_flags & CLONE_VM) {
 486                 atomic_inc(&oldmm->mm_users);
 487                 mm = oldmm;
 488                 /*
 489                  * There are cases where the PTL is held to ensure no
 490                  * new threads start up in user mode using an mm, which
 491                  * allows optimizing out ipis; the tlb_gather_mmu code
 492                  * is an example.
 493                  */
 494                 spin_unlock_wait(&oldmm->page_table_lock);
 495                 goto good_mm;
 496         }
 497
 498         retval = -ENOMEM;
 499         mm = allocate_mm();
 500         if (!mm)
 501                 goto fail_nomem;
 502
 503         /* Copy the current MM stuff.. */
 504         memcpy(mm, oldmm, sizeof(*mm));
 505         if (!mm_init(mm))
 506                 goto fail_nomem;
 507
 508         if (init_new_context(tsk,mm))
 509                 goto free_pt;
 510
 511         retval = dup_mmap(mm, oldmm);
 512         if (retval)
 513                 goto free_pt;
 514
 515 good_mm:
 516         tsk->mm = mm;
 517         tsk->active_mm = mm;
 518         return 0;
 519
 520 free_pt:
 521         mmput(mm);
 522 fail_nomem:
 523         return retval;
 524 }
 525
 526 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 527 {
 528         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 529         /* We don't need to lock fs - think why ;-) */
 530         if (fs) {
 531                 atomic_set(&fs->count, 1);
 532                 fs->lock = RW_LOCK_UNLOCKED;
 533                 fs->umask = old->umask;
 534                 read_lock(&old->lock);
 535                 fs->rootmnt = mntget(old->rootmnt);
 536                 fs->root = dget(old->root);
 537                 fs->pwdmnt = mntget(old->pwdmnt);
 538                 fs->pwd = dget(old->pwd);
 539                 if (old->altroot) {
 540                         fs->altrootmnt = mntget(old->altrootmnt);
 541                         fs->altroot = dget(old->altroot);
 542                 } else {
 543                         fs->altrootmnt = NULL;
 544                         fs->altroot = NULL;
 545                 }
 546                 read_unlock(&old->lock);
 547         }
 548         return fs;
 549 }
 550
 551 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 552 {
 553         return __copy_fs_struct(old);
 554 }
 555
 556 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 557 {
 558         if (clone_flags & CLONE_FS) {
 559                 atomic_inc(&current->fs->count);
 560                 return 0;
 561         }
 562         tsk->fs = __copy_fs_struct(current->fs);
 563         if (!tsk->fs)
 564                 return -ENOMEM;
 565         return 0;
 566 }
 567
 568 static int count_open_files(struct files_struct *files, int size)
 569 {
 570         int i;
 571
 572         /* Find the last open fd */
 573         for (i = size/(8*sizeof(long)); i > 0; ) {
 574                 if (files->open_fds->fds_bits[--i])
 575                         break;
 576         }
 577         i = (i+1) * 8 * sizeof(long);
 578         return i;
 579 }
 580
 581 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 582 {
 583         struct files_struct *oldf, *newf;
 584         struct file **old_fds, **new_fds;
 585         int open_files, nfds, size, i, error = 0;
 586
 587         /*
 588          * A background process may not have any files ...
 589          */
 590         oldf = current->files;
 591         if (!oldf)
 592                 goto out;
 593
 594         if (clone_flags & CLONE_FILES) {
 595                 atomic_inc(&oldf->count);
 596                 goto out;
 597         }
 598
 599         tsk->files = NULL;
 600         error = -ENOMEM;
 601         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
 602         if (!newf)
 603                 goto out;
 604
 605         atomic_set(&newf->count, 1);
 606
 607         newf->file_lock     = SPIN_LOCK_UNLOCKED;
 608         newf->next_fd       = 0;
 609         newf->max_fds       = NR_OPEN_DEFAULT;
 610         newf->max_fdset     = __FD_SETSIZE;
 611         newf->close_on_exec = &newf->close_on_exec_init;
 612         newf->open_fds      = &newf->open_fds_init;
 613         newf->fd            = &newf->fd_array[0];
 614
 615         /* We don't yet have the oldf readlock, but even if the old
 616            fdset gets grown now, we'll only copy up to "size" fds */
 617         size = oldf->max_fdset;
 618         if (size > __FD_SETSIZE) {
 619                 newf->max_fdset = 0;
 620                 spin_lock(&newf->file_lock);
 621                 error = expand_fdset(newf, size-1);
 622                 spin_unlock(&newf->file_lock);
 623                 if (error)
 624                         goto out_release;
 625         }
 626         spin_lock(&oldf->file_lock);
 627
 628         open_files = count_open_files(oldf, size);
 629
 630         /*
 631          * Check whether we need to allocate a larger fd array.
 632          * Note: we're not a clone task, so the open count won't
 633          * change.
 634          */
 635         nfds = NR_OPEN_DEFAULT;
 636         if (open_files > nfds) {
 637                 spin_unlock(&oldf->file_lock);
 638                 newf->max_fds = 0;
 639                 spin_lock(&newf->file_lock);
 640                 error = expand_fd_array(newf, open_files-1);
 641                 spin_unlock(&newf->file_lock);
 642                 if (error)
 643                         goto out_release;
 644                 nfds = newf->max_fds;
 645                 spin_lock(&oldf->file_lock);
 646         }
 647
 648         old_fds = oldf->fd;
 649         new_fds = newf->fd;
 650
 651         memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
 652         memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
 653
 654         for (i = open_files; i != 0; i--) {
 655                 struct file *f = *old_fds++;
 656                 if (f)
 657                         get_file(f);
 658                 *new_fds++ = f;
 659         }
 660         spin_unlock(&oldf->file_lock);
 661
 662         /* compute the remainder to be cleared */
 663         size = (newf->max_fds - open_files) * sizeof(struct file *);
 664
 665         /* This is long word aligned thus could use a optimized version */
 666         memset(new_fds, 0, size);
 667
 668         if (newf->max_fdset > open_files) {
 669                 int left = (newf->max_fdset-open_files)/8;
 670                 int start = open_files / (8 * sizeof(unsigned long));
 671
 672                 memset(&newf->open_fds->fds_bits[start], 0, left);
 673                 memset(&newf->close_on_exec->fds_bits[start], 0, left);
 674         }
 675
 676         tsk->files = newf;
 677         error = 0;
 678 out:
 679         return error;
 680
 681 out_release:
 682         free_fdset (newf->close_on_exec, newf->max_fdset);
 683         free_fdset (newf->open_fds, newf->max_fdset);
 684         kmem_cache_free(files_cachep, newf);
 685         goto out;
 686 }
 687
 688 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 689 {
 690         struct sighand_struct *sig;
 691
 692         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 693                 atomic_inc(&current->sighand->count);
 694                 return 0;
 695         }
 696         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 697         tsk->sighand = sig;
 698         if (!sig)
 699                 return -ENOMEM;
 700         spin_lock_init(&sig->siglock);
 701         atomic_set(&sig->count, 1);
 702         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 703         return 0;
 704 }
 705
 706 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 707 {
 708         struct signal_struct *sig;
 709
 710         if (clone_flags & CLONE_THREAD) {
 711                 atomic_inc(&current->signal->count);
 712                 return 0;
 713         }
 714         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 715         tsk->signal = sig;
 716         if (!sig)
 717                 return -ENOMEM;
 718         atomic_set(&sig->count, 1);
 719         sig->group_exit = 0;
 720         sig->group_exit_code = 0;
 721         sig->group_exit_task = NULL;
 722         sig->group_stop_count = 0;
 723         sig->curr_target = NULL;
 724         init_sigpending(&sig->shared_pending);
 725
 726         return 0;
 727 }
 728
 729 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 730 {
 731         unsigned long new_flags = p->flags;
 732
 733         new_flags &= ~PF_SUPERPRIV;
 734         new_flags |= PF_FORKNOEXEC;
 735         if (!(clone_flags & CLONE_PTRACE))
 736                 p->ptrace = 0;
 737         p->flags = new_flags;
 738 }
 739
 740 asmlinkage long sys_set_tid_address(int __user *tidptr)
 741 {
 742         current->clear_child_tid = tidptr;
 743
 744         return current->pid;
 745 }
 746
 747 /*
 748  * This creates a new process as a copy of the old one,
 749  * but does not actually start it yet.
 750  *
 751  * It copies the registers, and all the appropriate
 752  * parts of the process environment (as per the clone
 753  * flags). The actual kick-off is left to the caller.
 754  */
 755 struct task_struct *copy_process(unsigned long clone_flags,
 756                                  unsigned long stack_start,
 757                                  struct pt_regs *regs,
 758                                  unsigned long stack_size,
 759                                  int __user *parent_tidptr,
 760                                  int __user *child_tidptr)
 761 {
 762         int retval;
 763         struct task_struct *p = NULL;
 764
 765         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 766                 return ERR_PTR(-EINVAL);
 767
 768         /*
 769          * Thread groups must share signals as well, and detached threads
 770          * can only be started up within the thread group.
 771          */
 772         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 773                 return ERR_PTR(-EINVAL);
 774         if ((clone_flags & CLONE_DETACHED) && !(clone_flags & CLONE_THREAD))
 775                 return ERR_PTR(-EINVAL);
 776
 777         retval = security_task_create(clone_flags);
 778         if (retval)
 779                 goto fork_out;
 780
 781         retval = -ENOMEM;
 782         p = dup_task_struct(current);
 783         if (!p)
 784                 goto fork_out;
 785
 786         retval = -EAGAIN;
 787         if (atomic_read(&p->user->processes) >= p->rlim[RLIMIT_NPROC].rlim_cur) {
 788                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE))
 789                         goto bad_fork_free;
 790         }
 791
 792         atomic_inc(&p->user->__count);
 793         atomic_inc(&p->user->processes);
 794
 795         /*
 796          * If multiple threads are within copy_process(), then this check
 797          * triggers too late. This doesn't hurt, the check is only there
 798          * to stop root fork bombs.
 799          */
 800         if (nr_threads >= max_threads)
 801                 goto bad_fork_cleanup_count;
 802
 803         if (!try_module_get(p->thread_info->exec_domain->module))
 804                 goto bad_fork_cleanup_count;
 805
 806         if (p->binfmt && !try_module_get(p->binfmt->module))
 807                 goto bad_fork_cleanup_put_domain;
 808
 809 #ifdef CONFIG_PREEMPT
 810         /*
 811          * schedule_tail drops this_rq()->lock so we compensate with a count
 812          * of 1.  Also, we want to start with kernel preemption disabled.
 813          */
 814         p->thread_info->preempt_count = 1;
 815 #endif
 816         p->did_exec = 0;
 817         p->state = TASK_UNINTERRUPTIBLE;
 818
 819         copy_flags(clone_flags, p);
 820         if (clone_flags & CLONE_IDLETASK)
 821                 p->pid = 0;
 822         else {
 823                 p->pid = alloc_pidmap();
 824                 if (p->pid == -1)
 825                         goto bad_fork_cleanup;
 826         }
 827         retval = -EFAULT;
 828         if (clone_flags & CLONE_PARENT_SETTID)
 829                 if (put_user(p->pid, parent_tidptr))
 830                         goto bad_fork_cleanup;
 831
 832         p->proc_dentry = NULL;
 833
 834         INIT_LIST_HEAD(&p->run_list);
 835
 836         INIT_LIST_HEAD(&p->children);
 837         INIT_LIST_HEAD(&p->sibling);
 838         INIT_LIST_HEAD(&p->posix_timers);
 839         init_waitqueue_head(&p->wait_chldexit);
 840         p->vfork_done = NULL;
 841         spin_lock_init(&p->alloc_lock);
 842         spin_lock_init(&p->switch_lock);
 843
 844         clear_tsk_thread_flag(p, TIF_SIGPENDING);
 845         init_sigpending(&p->pending);
 846
 847         p->it_real_value = p->it_virt_value = p->it_prof_value = 0;
 848         p->it_real_incr = p->it_virt_incr = p->it_prof_incr = 0;
 849         init_timer(&p->real_timer);
 850         p->real_timer.data = (unsigned long) p;
 851
 852         p->leader = 0;          /* session leadership doesn't inherit */
 853         p->tty_old_pgrp = 0;
 854         p->utime = p->stime = 0;
 855         p->cutime = p->cstime = 0;
 856         p->array = NULL;
 857         p->lock_depth = -1;             /* -1 = no lock */
 858         p->start_time = get_jiffies_64();
 859         p->security = NULL;
 860
 861         retval = -ENOMEM;
 862         if ((retval = security_task_alloc(p)))
 863                 goto bad_fork_cleanup;
 864         /* copy all the process information */
 865         if ((retval = copy_semundo(clone_flags, p)))
 866                 goto bad_fork_cleanup_security;
 867         if ((retval = copy_files(clone_flags, p)))
 868                 goto bad_fork_cleanup_semundo;
 869         if ((retval = copy_fs(clone_flags, p)))
 870                 goto bad_fork_cleanup_files;
 871         if ((retval = copy_sighand(clone_flags, p)))
 872                 goto bad_fork_cleanup_fs;
 873         if ((retval = copy_signal(clone_flags, p)))
 874                 goto bad_fork_cleanup_sighand;
 875         if ((retval = copy_mm(clone_flags, p)))
 876                 goto bad_fork_cleanup_signal;
 877         if ((retval = copy_namespace(clone_flags, p)))
 878                 goto bad_fork_cleanup_mm;
 879         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
 880         if (retval)
 881                 goto bad_fork_cleanup_namespace;
 882
 883         if (clone_flags & CLONE_CHILD_SETTID)
 884                 p->set_child_tid = child_tidptr;
 885         /*
 886          * Clear TID on mm_release()?
 887          */
 888         if (clone_flags & CLONE_CHILD_CLEARTID)
 889                 p->clear_child_tid = child_tidptr;
 890
 891         /*
 892          * Syscall tracing should be turned off in the child regardless
 893          * of CLONE_PTRACE.
 894          */
 895         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
 896
 897         /* Our parent execution domain becomes current domain
 898            These must match for thread signalling to apply */
 899
 900         p->parent_exec_id = p->self_exec_id;
 901
 902         /* ok, now we should be set up.. */
 903         if (clone_flags & CLONE_DETACHED)
 904                 p->exit_signal = -1;
 905         else
 906                 p->exit_signal = clone_flags & CSIGNAL;
 907         p->pdeath_signal = 0;
 908
 909         /*
 910          * Share the timeslice between parent and child, thus the
 911          * total amount of pending timeslices in the system doesn't change,
 912          * resulting in more scheduling fairness.
 913          */
 914         local_irq_disable();
 915         p->time_slice = (current->time_slice + 1) >> 1;
 916         /*
 917          * The remainder of the first timeslice might be recovered by
 918          * the parent if the child exits early enough.
 919          */
 920         p->first_time_slice = 1;
 921         current->time_slice >>= 1;
 922         p->last_run = jiffies;
 923         if (!current->time_slice) {
 924                 /*
 925                  * This case is rare, it happens when the parent has only
 926                  * a single jiffy left from its timeslice. Taking the
 927                  * runqueue lock is not a problem.
 928                  */
 929                 current->time_slice = 1;
 930                 preempt_disable();
 931                 scheduler_tick(0, 0);
 932                 local_irq_enable();
 933                 preempt_enable();
 934         } else
 935                 local_irq_enable();
 936         /*
 937          * Ok, add it to the run-queues and make it
 938          * visible to the rest of the system.
 939          *
 940          * Let it rip!
 941          */
 942         p->tgid = p->pid;
 943         p->group_leader = p;
 944         INIT_LIST_HEAD(&p->ptrace_children);
 945         INIT_LIST_HEAD(&p->ptrace_list);
 946
 947         /* Need tasklist lock for parent etc handling! */
 948         write_lock_irq(&tasklist_lock);
 949         /*
 950          * Check for pending SIGKILL! The new thread should not be allowed
 951          * to slip out of an OOM kill. (or normal SIGKILL.)
 952          */
 953         if (sigismember(&current->pending.signal, SIGKILL)) {
 954                 write_unlock_irq(&tasklist_lock);
 955                 retval = -EINTR;
 956                 goto bad_fork_cleanup_namespace;
 957         }
 958
 959         /* CLONE_PARENT re-uses the old parent */
 960         if (clone_flags & CLONE_PARENT)
 961                 p->real_parent = current->real_parent;
 962         else
 963                 p->real_parent = current;
 964         p->parent = p->real_parent;
 965
 966         if (clone_flags & CLONE_THREAD) {
 967                 spin_lock(&current->sighand->siglock);
 968                 /*
 969                  * Important: if an exit-all has been started then
 970                  * do not create this new thread - the whole thread
 971                  * group is supposed to exit anyway.
 972                  */
 973                 if (current->signal->group_exit) {
 974                         spin_unlock(&current->sighand->siglock);
 975                         write_unlock_irq(&tasklist_lock);
 976                         goto bad_fork_cleanup_namespace;
 977                 }
 978                 p->tgid = current->tgid;
 979                 p->group_leader = current->group_leader;
 980
 981                 if (current->signal->group_stop_count > 0) {
 982                         /*
 983                          * There is an all-stop in progress for the group.
 984                          * We ourselves will stop as soon as we check signals.
 985                          * Make the new thread part of that group stop too.
 986                          */
 987                         current->signal->group_stop_count++;
 988                         set_tsk_thread_flag(p, TIF_SIGPENDING);
 989                 }
 990
 991                 spin_unlock(&current->sighand->siglock);
 992         }
 993
 994         SET_LINKS(p);
 995         if (p->ptrace & PT_PTRACED)
 996                 __ptrace_link(p, current->parent);
 997
 998         attach_pid(p, PIDTYPE_PID, p->pid);
 999         if (thread_group_leader(p)) {
1000                 attach_pid(p, PIDTYPE_TGID, p->tgid);
1001                 attach_pid(p, PIDTYPE_PGID, p->pgrp);
1002                 attach_pid(p, PIDTYPE_SID, p->session);
1003                 if (p->pid)
1004                         per_cpu(process_counts, smp_processor_id())++;
1005         } else
1006                 link_pid(p, p->pids + PIDTYPE_TGID, &p->group_leader->pids[PIDTYPE_TGID].pid);
1007
1008         nr_threads++;
1009         write_unlock_irq(&tasklist_lock);
1010         retval = 0;
1011
1012 fork_out:
1013         if (retval)
1014                 return ERR_PTR(retval);
1015         return p;
1016
1017 bad_fork_cleanup_namespace:
1018         exit_namespace(p);
1019 bad_fork_cleanup_mm:
1020         exit_mm(p);
1021 bad_fork_cleanup_signal:
1022         exit_signal(p);
1023 bad_fork_cleanup_sighand:
1024         exit_sighand(p);
1025 bad_fork_cleanup_fs:
1026         exit_fs(p); /* blocking */
1027 bad_fork_cleanup_files:
1028         exit_files(p); /* blocking */
1029 bad_fork_cleanup_semundo:
1030         exit_semundo(p);
1031 bad_fork_cleanup_security:
1032         security_task_free(p);
1033 bad_fork_cleanup:
1034         if (p->pid > 0)
1035                 free_pidmap(p->pid);
1036         if (p->binfmt)
1037                 module_put(p->binfmt->module);
1038 bad_fork_cleanup_put_domain:
1039         module_put(p->thread_info->exec_domain->module);
1040 bad_fork_cleanup_count:
1041         atomic_dec(&p->user->processes);
1042         free_uid(p->user);
1043 bad_fork_free:
1044         free_task_struct(p);
1045         goto fork_out;
1046 }
1047
1048 static inline int fork_traceflag (unsigned clone_flags)
1049 {
1050         if (clone_flags & (CLONE_UNTRACED | CLONE_IDLETASK))
1051                 return 0;
1052         else if (clone_flags & CLONE_VFORK) {
1053                 if (current->ptrace & PT_TRACE_VFORK)
1054                         return PTRACE_EVENT_VFORK;
1055         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1056                 if (current->ptrace & PT_TRACE_CLONE)
1057                         return PTRACE_EVENT_CLONE;
1058         } else if (current->ptrace & PT_TRACE_FORK)
1059                 return PTRACE_EVENT_FORK;
1060
1061         return 0;
1062 }
1063
1064 /*
1065  *  Ok, this is the main fork-routine.
1066  *
1067  * It copies the process, and if successful kick-starts
1068  * it and waits for it to finish using the VM if required.
1069  */
1070 long do_fork(unsigned long clone_flags,
1071               unsigned long stack_start,
1072               struct pt_regs *regs,
1073               unsigned long stack_size,
1074               int __user *parent_tidptr,
1075               int __user *child_tidptr)
1076 {
1077         struct task_struct *p;
1078         int trace = 0;
1079         long pid;
1080
1081         if (unlikely(current->ptrace)) {
1082                 trace = fork_traceflag (clone_flags);
1083                 if (trace)
1084                         clone_flags |= CLONE_PTRACE;
1085         }
1086
1087         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr);
1088         /*
1089          * Do this prior waking up the new thread - the thread pointer
1090          * might get invalid after that point, if the thread exits quickly.
1091          */
1092         pid = IS_ERR(p) ? PTR_ERR(p) : p->pid;
1093
1094         if (!IS_ERR(p)) {
1095                 struct completion vfork;
1096
1097                 if (clone_flags & CLONE_VFORK) {
1098                         p->vfork_done = &vfork;
1099                         init_completion(&vfork);
1100                 }
1101
1102                 if (p->ptrace & PT_PTRACED) {
1103                         /*
1104                          * We'll start up with an immediate SIGSTOP.
1105                          */
1106                         sigaddset(&p->pending.signal, SIGSTOP);
1107                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1108                 }
1109
1110                 wake_up_forked_process(p);              /* do this last */
1111                 ++total_forks;
1112
1113                 if (unlikely (trace)) {
1114                         current->ptrace_message = pid;
1115                         ptrace_notify ((trace << 8) | SIGTRAP);
1116                 }
1117
1118                 if (clone_flags & CLONE_VFORK) {
1119                         wait_for_completion(&vfork);
1120                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1121                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1122                 } else
1123                         /*
1124                          * Let the child process run first, to avoid most of the
1125                          * COW overhead when the child exec()s afterwards.
1126                          */
1127                         set_need_resched();
1128         }
1129         return pid;
1130 }
1131
1132 /* SLAB cache for signal_struct structures (tsk->signal) */
1133 kmem_cache_t *signal_cachep;
1134
1135 /* SLAB cache for sighand_struct structures (tsk->sighand) */
1136 kmem_cache_t *sighand_cachep;
1137
1138 /* SLAB cache for files_struct structures (tsk->files) */
1139 kmem_cache_t *files_cachep;
1140
1141 /* SLAB cache for fs_struct structures (tsk->fs) */
1142 kmem_cache_t *fs_cachep;
1143
1144 /* SLAB cache for vm_area_struct structures */
1145 kmem_cache_t *vm_area_cachep;
1146
1147 /* SLAB cache for mm_struct structures (tsk->mm) */
1148 kmem_cache_t *mm_cachep;
1149
1150 void __init proc_caches_init(void)
1151 {
1152         sighand_cachep = kmem_cache_create("sighand_cache",
1153                         sizeof(struct sighand_struct), 0,
1154                         SLAB_HWCACHE_ALIGN, NULL, NULL);
1155         if (!sighand_cachep)
1156                 panic("Cannot create sighand SLAB cache");
1157
1158         signal_cachep = kmem_cache_create("signal_cache",
1159                         sizeof(struct signal_struct), 0,
1160                         SLAB_HWCACHE_ALIGN, NULL, NULL);
1161         if (!signal_cachep)
1162                 panic("Cannot create signal SLAB cache");
1163
1164         files_cachep = kmem_cache_create("files_cache",
1165                          sizeof(struct files_struct), 0,
1166                          SLAB_HWCACHE_ALIGN, NULL, NULL);
1167         if (!files_cachep)
1168                 panic("Cannot create files SLAB cache");
1169
1170         fs_cachep = kmem_cache_create("fs_cache",
1171                          sizeof(struct fs_struct), 0,
1172                          SLAB_HWCACHE_ALIGN, NULL, NULL);
1173         if (!fs_cachep)
1174                 panic("Cannot create fs_struct SLAB cache");
1175
1176         vm_area_cachep = kmem_cache_create("vm_area_struct",
1177                         sizeof(struct vm_area_struct), 0,
1178                         0, NULL, NULL);
1179         if(!vm_area_cachep)
1180                 panic("vma_init: Cannot alloc vm_area_struct SLAB cache");
1181
1182         mm_cachep = kmem_cache_create("mm_struct",
1183                         sizeof(struct mm_struct), 0,
1184                         SLAB_HWCACHE_ALIGN, NULL, NULL);
1185         if(!mm_cachep)
1186                 panic("vma_init: Cannot alloc mm_struct SLAB cache");
1187 }