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 extern int copy_semundo(unsigned long clone_flags, struct task_struct *tsk);
  42 extern void exit_semundo(struct task_struct *tsk);
  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 int nr_processes(void)
  57 {
  58         int cpu;
  59         int total = 0;
  60
  61         for (cpu = 0; cpu < NR_CPUS; cpu++) {
  62                 if (cpu_online(cpu))
  63                         total += per_cpu(process_counts, cpu);
  64         }
  65         return total;
  66 }
  67
  68 #ifdef CONFIG_IA64
  69 # define HAVE_ARCH_DUP_TASK_STRUCT
  70 #endif
  71
  72 #ifdef HAVE_ARCH_DUP_TASK_STRUCT
  73 extern void free_task_struct (struct task_struct *tsk);
  74 #else
  75 static kmem_cache_t *task_struct_cachep;
  76
  77 /*
  78  * A per-CPU task cache - this relies on the fact that
  79  * the very last portion of sys_exit() is executed with
  80  * preemption turned off.
  81  */
  82 static task_t *task_cache[NR_CPUS] __cacheline_aligned;
  83
  84 static void free_task_struct(struct task_struct *tsk)
  85 {
  86         /*
  87          * The task cache is effectively disabled right now.
  88          * Do we want it? The slab cache already has per-cpu
  89          * stuff, but the thread info (usually a order-1 page
  90          * allocation) doesn't.
  91          */
  92         if (tsk != current) {
  93                 free_thread_info(tsk->thread_info);
  94                 kmem_cache_free(task_struct_cachep,tsk);
  95         } else {
  96                 int cpu = get_cpu();
  97
  98                 tsk = task_cache[cpu];
  99                 if (tsk) {
 100                         free_thread_info(tsk->thread_info);
 101                         kmem_cache_free(task_struct_cachep,tsk);
 102                 }
 103                 task_cache[cpu] = current;
 104                 put_cpu();
 105         }
 106 }
 107 #endif /* HAVE_ARCH_DUP_TASK_STRUCT */
 108
 109 void __put_task_struct(struct task_struct *tsk)
 110 {
 111         WARN_ON(!(tsk->state & (TASK_DEAD | TASK_ZOMBIE)));
 112         WARN_ON(atomic_read(&tsk->usage));
 113         WARN_ON(tsk == current);
 114
 115         security_task_free(tsk);
 116         free_uid(tsk->user);
 117         free_task_struct(tsk);
 118 }
 119
 120 void add_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
 121 {
 122         unsigned long flags;
 123
 124         wait->flags &= ~WQ_FLAG_EXCLUSIVE;
 125         spin_lock_irqsave(&q->lock, flags);
 126         __add_wait_queue(q, wait);
 127         spin_unlock_irqrestore(&q->lock, flags);
 128 }
 129
 130 void add_wait_queue_exclusive(wait_queue_head_t *q, wait_queue_t * wait)
 131 {
 132         unsigned long flags;
 133
 134         wait->flags |= WQ_FLAG_EXCLUSIVE;
 135         spin_lock_irqsave(&q->lock, flags);
 136         __add_wait_queue_tail(q, wait);
 137         spin_unlock_irqrestore(&q->lock, flags);
 138 }
 139
 140 void remove_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
 141 {
 142         unsigned long flags;
 143
 144         spin_lock_irqsave(&q->lock, flags);
 145         __remove_wait_queue(q, wait);
 146         spin_unlock_irqrestore(&q->lock, flags);
 147 }
 148
 149 void prepare_to_wait(wait_queue_head_t *q, wait_queue_t *wait, int state)
 150 {
 151         unsigned long flags;
 152
 153         __set_current_state(state);
 154         wait->flags &= ~WQ_FLAG_EXCLUSIVE;
 155         spin_lock_irqsave(&q->lock, flags);
 156         if (list_empty(&wait->task_list))
 157                 __add_wait_queue(q, wait);
 158         spin_unlock_irqrestore(&q->lock, flags);
 159 }
 160
 161 void
 162 prepare_to_wait_exclusive(wait_queue_head_t *q, wait_queue_t *wait, int state)
 163 {
 164         unsigned long flags;
 165
 166         __set_current_state(state);
 167         wait->flags |= WQ_FLAG_EXCLUSIVE;
 168         spin_lock_irqsave(&q->lock, flags);
 169         if (list_empty(&wait->task_list))
 170                 __add_wait_queue_tail(q, wait);
 171         spin_unlock_irqrestore(&q->lock, flags);
 172 }
 173
 174 void finish_wait(wait_queue_head_t *q, wait_queue_t *wait)
 175 {
 176         unsigned long flags;
 177
 178         __set_current_state(TASK_RUNNING);
 179         if (!list_empty(&wait->task_list)) {
 180                 spin_lock_irqsave(&q->lock, flags);
 181                 list_del_init(&wait->task_list);
 182                 spin_unlock_irqrestore(&q->lock, flags);
 183         }
 184 }
 185
 186 int autoremove_wake_function(wait_queue_t *wait, unsigned mode, int sync)
 187 {
 188         int ret = default_wake_function(wait, mode, sync);
 189
 190         if (ret)
 191                 list_del_init(&wait->task_list);
 192         return ret;
 193 }
 194
 195 void __init fork_init(unsigned long mempages)
 196 {
 197 #ifndef HAVE_ARCH_DUP_TASK_STRUCT
 198         /* create a slab on which task_structs can be allocated */
 199         task_struct_cachep =
 200                 kmem_cache_create("task_struct",
 201                                   sizeof(struct task_struct),0,
 202                                   SLAB_MUST_HWCACHE_ALIGN, NULL, NULL);
 203         if (!task_struct_cachep)
 204                 panic("fork_init(): cannot create task_struct SLAB cache");
 205 #endif
 206
 207         /*
 208          * The default maximum number of threads is set to a safe
 209          * value: the thread structures can take up at most half
 210          * of memory.
 211          */
 212         max_threads = mempages / (THREAD_SIZE/PAGE_SIZE) / 8;
 213         /*
 214          * we need to allow at least 20 threads to boot a system
 215          */
 216         if(max_threads < 20)
 217                 max_threads = 20;
 218
 219         init_task.rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 220         init_task.rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 221 }
 222
 223 #ifdef HAVE_ARCH_DUP_TASK_STRUCT
 224 extern struct task_struct *dup_task_struct (struct task_struct *orig);
 225 #else /* !HAVE_ARCH_DUP_TASK_STRUCT */
 226
 227 struct task_struct *dup_task_struct(struct task_struct *orig)
 228 {
 229         struct task_struct *tsk;
 230         struct thread_info *ti;
 231         int cpu = get_cpu();
 232
 233         prepare_to_copy(orig);
 234
 235         tsk = task_cache[cpu];
 236         task_cache[cpu] = NULL;
 237         put_cpu();
 238         if (!tsk) {
 239                 ti = alloc_thread_info();
 240                 if (!ti)
 241                         return NULL;
 242
 243                 tsk = kmem_cache_alloc(task_struct_cachep, GFP_KERNEL);
 244                 if (!tsk) {
 245                         free_thread_info(ti);
 246                         return NULL;
 247                 }
 248         } else
 249                 ti = tsk->thread_info;
 250
 251         *ti = *orig->thread_info;
 252         *tsk = *orig;
 253         tsk->thread_info = ti;
 254         ti->task = tsk;
 255
 256         /* One for us, one for whoever does the "release_task()" (usually parent) */
 257         atomic_set(&tsk->usage,2);
 258         return tsk;
 259 }
 260
 261 #endif /* !HAVE_ARCH_DUP_TASK_STRUCT */
 262
 263 #ifdef CONFIG_MMU
 264 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
 265 {
 266         struct vm_area_struct * mpnt, *tmp, **pprev;
 267         int retval;
 268         unsigned long charge = 0;
 269
 270         down_write(&oldmm->mmap_sem);
 271         flush_cache_mm(current->mm);
 272         mm->locked_vm = 0;
 273         mm->mmap = NULL;
 274         mm->mmap_cache = NULL;
 275         mm->free_area_cache = TASK_UNMAPPED_BASE;
 276         mm->map_count = 0;
 277         mm->rss = 0;
 278         mm->cpu_vm_mask = 0;
 279         pprev = &mm->mmap;
 280
 281         /*
 282          * Add it to the mmlist after the parent.
 283          * Doing it this way means that we can order the list,
 284          * and fork() won't mess up the ordering significantly.
 285          * Add it first so that swapoff can see any swap entries.
 286          */
 287         spin_lock(&mmlist_lock);
 288         list_add(&mm->mmlist, &current->mm->mmlist);
 289         mmlist_nr++;
 290         spin_unlock(&mmlist_lock);
 291
 292         for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
 293                 struct file *file;
 294
 295                 if(mpnt->vm_flags & VM_DONTCOPY)
 296                         continue;
 297                 if (mpnt->vm_flags & VM_ACCOUNT) {
 298                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 299                         if (!vm_enough_memory(len))
 300                                 goto fail_nomem;
 301                         charge += len;
 302                 }
 303                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 304                 if (!tmp)
 305                         goto fail_nomem;
 306                 *tmp = *mpnt;
 307                 tmp->vm_flags &= ~VM_LOCKED;
 308                 tmp->vm_mm = mm;
 309                 tmp->vm_next = NULL;
 310                 file = tmp->vm_file;
 311                 INIT_LIST_HEAD(&tmp->shared);
 312                 if (file) {
 313                         struct inode *inode = file->f_dentry->d_inode;
 314                         get_file(file);
 315                         if (tmp->vm_flags & VM_DENYWRITE)
 316                                 atomic_dec(&inode->i_writecount);
 317
 318                         /* insert tmp into the share list, just after mpnt */
 319                         down(&inode->i_mapping->i_shared_sem);
 320                         list_add_tail(&tmp->shared, &mpnt->shared);
 321                         up(&inode->i_mapping->i_shared_sem);
 322                 }
 323
 324                 /*
 325                  * Link in the new vma and copy the page table entries:
 326                  * link in first so that swapoff can see swap entries.
 327                  */
 328                 spin_lock(&mm->page_table_lock);
 329                 *pprev = tmp;
 330                 pprev = &tmp->vm_next;
 331                 mm->map_count++;
 332                 retval = copy_page_range(mm, current->mm, tmp);
 333                 spin_unlock(&mm->page_table_lock);
 334
 335                 if (tmp->vm_ops && tmp->vm_ops->open)
 336                         tmp->vm_ops->open(tmp);
 337
 338                 if (retval)
 339                         goto fail;
 340         }
 341         retval = 0;
 342         build_mmap_rb(mm);
 343
 344 out:
 345         flush_tlb_mm(current->mm);
 346         up_write(&oldmm->mmap_sem);
 347         return retval;
 348 fail_nomem:
 349         retval = -ENOMEM;
 350   fail:
 351         vm_unacct_memory(charge);
 352         goto out;
 353 }
 354 static inline int mm_alloc_pgd(struct mm_struct * mm)
 355 {
 356         mm->pgd = pgd_alloc(mm);
 357         if (unlikely(!mm->pgd))
 358                 return -ENOMEM;
 359         return 0;
 360 }
 361
 362 static inline void mm_free_pgd(struct mm_struct * mm)
 363 {
 364         pgd_free(mm->pgd);
 365 }
 366 #else
 367 #define dup_mmap(mm, oldmm)     (0)
 368 #define mm_alloc_pgd(mm)        (0)
 369 #define mm_free_pgd(mm)
 370 #endif /* CONFIG_MMU */
 371
 372 spinlock_t mmlist_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
 373 int mmlist_nr;
 374
 375 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
 376 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 377
 378 #include <linux/init_task.h>
 379
 380 static struct mm_struct * mm_init(struct mm_struct * mm)
 381 {
 382         atomic_set(&mm->mm_users, 1);
 383         atomic_set(&mm->mm_count, 1);
 384         init_rwsem(&mm->mmap_sem);
 385         mm->core_waiters = 0;
 386         mm->page_table_lock = SPIN_LOCK_UNLOCKED;
 387         mm->ioctx_list_lock = RW_LOCK_UNLOCKED;
 388         mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
 389         mm->free_area_cache = TASK_UNMAPPED_BASE;
 390
 391         if (likely(!mm_alloc_pgd(mm))) {
 392                 mm->def_flags = 0;
 393                 return mm;
 394         }
 395         free_mm(mm);
 396         return NULL;
 397 }
 398
 399
 400 /*
 401  * Allocate and initialize an mm_struct.
 402  */
 403 struct mm_struct * mm_alloc(void)
 404 {
 405         struct mm_struct * mm;
 406
 407         mm = allocate_mm();
 408         if (mm) {
 409                 memset(mm, 0, sizeof(*mm));
 410                 return mm_init(mm);
 411         }
 412         return NULL;
 413 }
 414
 415 /*
 416  * Called when the last reference to the mm
 417  * is dropped: either by a lazy thread or by
 418  * mmput. Free the page directory and the mm.
 419  */
 420 inline void __mmdrop(struct mm_struct *mm)
 421 {
 422         BUG_ON(mm == &init_mm);
 423         mm_free_pgd(mm);
 424         destroy_context(mm);
 425         free_mm(mm);
 426 }
 427
 428 /*
 429  * Decrement the use count and release all resources for an mm.
 430  */
 431 void mmput(struct mm_struct *mm)
 432 {
 433         if (atomic_dec_and_lock(&mm->mm_users, &mmlist_lock)) {
 434                 list_del(&mm->mmlist);
 435                 mmlist_nr--;
 436                 spin_unlock(&mmlist_lock);
 437                 exit_aio(mm);
 438                 exit_mmap(mm);
 439                 mmdrop(mm);
 440         }
 441 }
 442
 443 /* Please note the differences between mmput and mm_release.
 444  * mmput is called whenever we stop holding onto a mm_struct,
 445  * error success whatever.
 446  *
 447  * mm_release is called after a mm_struct has been removed
 448  * from the current process.
 449  *
 450  * This difference is important for error handling, when we
 451  * only half set up a mm_struct for a new process and need to restore
 452  * the old one.  Because we mmput the new mm_struct before
 453  * restoring the old one. . .
 454  * Eric Biederman 10 January 1998
 455  */
 456 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 457 {
 458         struct completion *vfork_done = tsk->vfork_done;
 459
 460         /* Get rid of any cached register state */
 461         deactivate_mm(tsk, mm);
 462
 463         /* notify parent sleeping on vfork() */
 464         if (vfork_done) {
 465                 tsk->vfork_done = NULL;
 466                 complete(vfork_done);
 467         }
 468         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
 469                 u32 * tidptr = tsk->clear_child_tid;
 470                 tsk->clear_child_tid = NULL;
 471
 472                 /*
 473                  * We don't check the error code - if userspace has
 474                  * not set up a proper pointer then tough luck.
 475                  */
 476                 put_user(0, tidptr);
 477                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL);
 478         }
 479 }
 480
 481 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 482 {
 483         struct mm_struct * mm, *oldmm;
 484         int retval;
 485
 486         tsk->min_flt = tsk->maj_flt = 0;
 487         tsk->cmin_flt = tsk->cmaj_flt = 0;
 488         tsk->nswap = tsk->cnswap = 0;
 489
 490         tsk->mm = NULL;
 491         tsk->active_mm = NULL;
 492
 493         /*
 494          * Are we cloning a kernel thread?
 495          *
 496          * We need to steal a active VM for that..
 497          */
 498         oldmm = current->mm;
 499         if (!oldmm)
 500                 return 0;
 501
 502         if (clone_flags & CLONE_VM) {
 503                 atomic_inc(&oldmm->mm_users);
 504                 mm = oldmm;
 505                 /*
 506                  * There are cases where the PTL is held to ensure no
 507                  * new threads start up in user mode using an mm, which
 508                  * allows optimizing out ipis; the tlb_gather_mmu code
 509                  * is an example.
 510                  */
 511                 spin_unlock_wait(&oldmm->page_table_lock);
 512                 goto good_mm;
 513         }
 514
 515         retval = -ENOMEM;
 516         mm = allocate_mm();
 517         if (!mm)
 518                 goto fail_nomem;
 519
 520         /* Copy the current MM stuff.. */
 521         memcpy(mm, oldmm, sizeof(*mm));
 522         if (!mm_init(mm))
 523                 goto fail_nomem;
 524
 525         if (init_new_context(tsk,mm))
 526                 goto free_pt;
 527
 528         retval = dup_mmap(mm, oldmm);
 529         if (retval)
 530                 goto free_pt;
 531
 532 good_mm:
 533         tsk->mm = mm;
 534         tsk->active_mm = mm;
 535         return 0;
 536
 537 free_pt:
 538         mmput(mm);
 539 fail_nomem:
 540         return retval;
 541 }
 542
 543 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 544 {
 545         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 546         /* We don't need to lock fs - think why ;-) */
 547         if (fs) {
 548                 atomic_set(&fs->count, 1);
 549                 fs->lock = RW_LOCK_UNLOCKED;
 550                 fs->umask = old->umask;
 551                 read_lock(&old->lock);
 552                 fs->rootmnt = mntget(old->rootmnt);
 553                 fs->root = dget(old->root);
 554                 fs->pwdmnt = mntget(old->pwdmnt);
 555                 fs->pwd = dget(old->pwd);
 556                 if (old->altroot) {
 557                         fs->altrootmnt = mntget(old->altrootmnt);
 558                         fs->altroot = dget(old->altroot);
 559                 } else {
 560                         fs->altrootmnt = NULL;
 561                         fs->altroot = NULL;
 562                 }
 563                 read_unlock(&old->lock);
 564         }
 565         return fs;
 566 }
 567
 568 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 569 {
 570         return __copy_fs_struct(old);
 571 }
 572
 573 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 574 {
 575         if (clone_flags & CLONE_FS) {
 576                 atomic_inc(&current->fs->count);
 577                 return 0;
 578         }
 579         tsk->fs = __copy_fs_struct(current->fs);
 580         if (!tsk->fs)
 581                 return -1;
 582         return 0;
 583 }
 584
 585 static int count_open_files(struct files_struct *files, int size)
 586 {
 587         int i;
 588
 589         /* Find the last open fd */
 590         for (i = size/(8*sizeof(long)); i > 0; ) {
 591                 if (files->open_fds->fds_bits[--i])
 592                         break;
 593         }
 594         i = (i+1) * 8 * sizeof(long);
 595         return i;
 596 }
 597
 598 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 599 {
 600         struct files_struct *oldf, *newf;
 601         struct file **old_fds, **new_fds;
 602         int open_files, nfds, size, i, error = 0;
 603
 604         /*
 605          * A background process may not have any files ...
 606          */
 607         oldf = current->files;
 608         if (!oldf)
 609                 goto out;
 610
 611         if (clone_flags & CLONE_FILES) {
 612                 atomic_inc(&oldf->count);
 613                 goto out;
 614         }
 615
 616         tsk->files = NULL;
 617         error = -ENOMEM;
 618         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
 619         if (!newf)
 620                 goto out;
 621
 622         atomic_set(&newf->count, 1);
 623
 624         newf->file_lock     = SPIN_LOCK_UNLOCKED;
 625         newf->next_fd       = 0;
 626         newf->max_fds       = NR_OPEN_DEFAULT;
 627         newf->max_fdset     = __FD_SETSIZE;
 628         newf->close_on_exec = &newf->close_on_exec_init;
 629         newf->open_fds      = &newf->open_fds_init;
 630         newf->fd            = &newf->fd_array[0];
 631
 632         /* We don't yet have the oldf readlock, but even if the old
 633            fdset gets grown now, we'll only copy up to "size" fds */
 634         size = oldf->max_fdset;
 635         if (size > __FD_SETSIZE) {
 636                 newf->max_fdset = 0;
 637                 spin_lock(&newf->file_lock);
 638                 error = expand_fdset(newf, size-1);
 639                 spin_unlock(&newf->file_lock);
 640                 if (error)
 641                         goto out_release;
 642         }
 643         spin_lock(&oldf->file_lock);
 644
 645         open_files = count_open_files(oldf, size);
 646
 647         /*
 648          * Check whether we need to allocate a larger fd array.
 649          * Note: we're not a clone task, so the open count won't
 650          * change.
 651          */
 652         nfds = NR_OPEN_DEFAULT;
 653         if (open_files > nfds) {
 654                 spin_unlock(&oldf->file_lock);
 655                 newf->max_fds = 0;
 656                 spin_lock(&newf->file_lock);
 657                 error = expand_fd_array(newf, open_files-1);
 658                 spin_unlock(&newf->file_lock);
 659                 if (error)
 660                         goto out_release;
 661                 nfds = newf->max_fds;
 662                 spin_lock(&oldf->file_lock);
 663         }
 664
 665         old_fds = oldf->fd;
 666         new_fds = newf->fd;
 667
 668         memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
 669         memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
 670
 671         for (i = open_files; i != 0; i--) {
 672                 struct file *f = *old_fds++;
 673                 if (f)
 674                         get_file(f);
 675                 *new_fds++ = f;
 676         }
 677         spin_unlock(&oldf->file_lock);
 678
 679         /* compute the remainder to be cleared */
 680         size = (newf->max_fds - open_files) * sizeof(struct file *);
 681
 682         /* This is long word aligned thus could use a optimized version */
 683         memset(new_fds, 0, size);
 684
 685         if (newf->max_fdset > open_files) {
 686                 int left = (newf->max_fdset-open_files)/8;
 687                 int start = open_files / (8 * sizeof(unsigned long));
 688
 689                 memset(&newf->open_fds->fds_bits[start], 0, left);
 690                 memset(&newf->close_on_exec->fds_bits[start], 0, left);
 691         }
 692
 693         tsk->files = newf;
 694         error = 0;
 695 out:
 696         return error;
 697
 698 out_release:
 699         free_fdset (newf->close_on_exec, newf->max_fdset);
 700         free_fdset (newf->open_fds, newf->max_fdset);
 701         kmem_cache_free(files_cachep, newf);
 702         goto out;
 703 }
 704
 705 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 706 {
 707         struct sighand_struct *sig;
 708
 709         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 710                 atomic_inc(&current->sighand->count);
 711                 return 0;
 712         }
 713         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 714         tsk->sighand = sig;
 715         if (!sig)
 716                 return -1;
 717         spin_lock_init(&sig->siglock);
 718         atomic_set(&sig->count, 1);
 719         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 720         return 0;
 721 }
 722
 723 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 724 {
 725         struct signal_struct *sig;
 726
 727         if (clone_flags & CLONE_THREAD) {
 728                 atomic_inc(&current->signal->count);
 729                 return 0;
 730         }
 731         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 732         tsk->signal = sig;
 733         if (!sig)
 734                 return -1;
 735         atomic_set(&sig->count, 1);
 736         sig->group_exit = 0;
 737         sig->group_exit_code = 0;
 738         sig->group_exit_task = NULL;
 739         sig->group_stop_count = 0;
 740         sig->curr_target = NULL;
 741         init_sigpending(&sig->shared_pending);
 742
 743         return 0;
 744 }
 745
 746 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 747 {
 748         unsigned long new_flags = p->flags;
 749
 750         new_flags &= ~PF_SUPERPRIV;
 751         new_flags |= PF_FORKNOEXEC;
 752         if (!(clone_flags & CLONE_PTRACE))
 753                 p->ptrace = 0;
 754         p->flags = new_flags;
 755 }
 756
 757 asmlinkage long sys_set_tid_address(int *tidptr)
 758 {
 759         current->clear_child_tid = tidptr;
 760
 761         return current->pid;
 762 }
 763
 764 /*
 765  * This creates a new process as a copy of the old one,
 766  * but does not actually start it yet.
 767  *
 768  * It copies the registers, and all the appropriate
 769  * parts of the process environment (as per the clone
 770  * flags). The actual kick-off is left to the caller.
 771  */
 772 static struct task_struct *copy_process(unsigned long clone_flags,
 773                             unsigned long stack_start,
 774                             struct pt_regs *regs,
 775                             unsigned long stack_size,
 776                             int *parent_tidptr,
 777                             int *child_tidptr)
 778 {
 779         int retval;
 780         struct task_struct *p = NULL;
 781
 782         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 783                 return ERR_PTR(-EINVAL);
 784
 785         /*
 786          * Thread groups must share signals as well, and detached threads
 787          * can only be started up within the thread group.
 788          */
 789         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 790                 return ERR_PTR(-EINVAL);
 791         if ((clone_flags & CLONE_DETACHED) && !(clone_flags & CLONE_THREAD))
 792                 return ERR_PTR(-EINVAL);
 793
 794         retval = security_task_create(clone_flags);
 795         if (retval)
 796                 goto fork_out;
 797
 798         retval = -ENOMEM;
 799         p = dup_task_struct(current);
 800         if (!p)
 801                 goto fork_out;
 802
 803         retval = -EAGAIN;
 804         if (atomic_read(&p->user->processes) >= p->rlim[RLIMIT_NPROC].rlim_cur) {
 805                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE))
 806                         goto bad_fork_free;
 807         }
 808
 809         atomic_inc(&p->user->__count);
 810         atomic_inc(&p->user->processes);
 811
 812         /*
 813          * If multiple threads are within copy_process(), then this check
 814          * triggers too late. This doesn't hurt, the check is only there
 815          * to stop root fork bombs.
 816          */
 817         if (nr_threads >= max_threads)
 818                 goto bad_fork_cleanup_count;
 819
 820         if (!try_module_get(p->thread_info->exec_domain->module))
 821                 goto bad_fork_cleanup_count;
 822
 823         if (p->binfmt && !try_module_get(p->binfmt->module))
 824                 goto bad_fork_cleanup_put_domain;
 825
 826 #ifdef CONFIG_PREEMPT
 827         /*
 828          * schedule_tail drops this_rq()->lock so we compensate with a count
 829          * of 1.  Also, we want to start with kernel preemption disabled.
 830          */
 831         p->thread_info->preempt_count = 1;
 832 #endif
 833         p->did_exec = 0;
 834         p->state = TASK_UNINTERRUPTIBLE;
 835
 836         copy_flags(clone_flags, p);
 837         if (clone_flags & CLONE_IDLETASK)
 838                 p->pid = 0;
 839         else {
 840                 p->pid = alloc_pidmap();
 841                 if (p->pid == -1)
 842                         goto bad_fork_cleanup;
 843         }
 844         retval = -EFAULT;
 845         if (clone_flags & CLONE_PARENT_SETTID)
 846                 if (put_user(p->pid, parent_tidptr))
 847                         goto bad_fork_cleanup;
 848
 849         p->proc_dentry = NULL;
 850
 851         INIT_LIST_HEAD(&p->run_list);
 852
 853         INIT_LIST_HEAD(&p->children);
 854         INIT_LIST_HEAD(&p->sibling);
 855         INIT_LIST_HEAD(&p->posix_timers);
 856         init_waitqueue_head(&p->wait_chldexit);
 857         p->vfork_done = NULL;
 858         spin_lock_init(&p->alloc_lock);
 859         spin_lock_init(&p->switch_lock);
 860
 861         clear_tsk_thread_flag(p, TIF_SIGPENDING);
 862         init_sigpending(&p->pending);
 863
 864         p->it_real_value = p->it_virt_value = p->it_prof_value = 0;
 865         p->it_real_incr = p->it_virt_incr = p->it_prof_incr = 0;
 866         init_timer(&p->real_timer);
 867         p->real_timer.data = (unsigned long) p;
 868
 869         p->leader = 0;          /* session leadership doesn't inherit */
 870         p->tty_old_pgrp = 0;
 871         p->utime = p->stime = 0;
 872         p->cutime = p->cstime = 0;
 873         p->array = NULL;
 874         p->lock_depth = -1;             /* -1 = no lock */
 875         p->start_time = get_jiffies_64();
 876         p->security = NULL;
 877
 878         retval = -ENOMEM;
 879         if (security_task_alloc(p))
 880                 goto bad_fork_cleanup;
 881         /* copy all the process information */
 882         if (copy_semundo(clone_flags, p))
 883                 goto bad_fork_cleanup_security;
 884         if (copy_files(clone_flags, p))
 885                 goto bad_fork_cleanup_semundo;
 886         if (copy_fs(clone_flags, p))
 887                 goto bad_fork_cleanup_files;
 888         if (copy_sighand(clone_flags, p))
 889                 goto bad_fork_cleanup_fs;
 890         if (copy_signal(clone_flags, p))
 891                 goto bad_fork_cleanup_sighand;
 892         if (copy_mm(clone_flags, p))
 893                 goto bad_fork_cleanup_signal;
 894         retval = copy_namespace(clone_flags, p);
 895         if (retval)
 896                 goto bad_fork_cleanup_mm;
 897         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
 898         if (retval)
 899                 goto bad_fork_cleanup_namespace;
 900
 901         if (clone_flags & CLONE_CHILD_SETTID)
 902                 p->set_child_tid = child_tidptr;
 903         else
 904                 p->set_child_tid = NULL;
 905         /*
 906          * Clear TID on mm_release()?
 907          */
 908         if (clone_flags & CLONE_CHILD_CLEARTID)
 909                 p->clear_child_tid = child_tidptr;
 910         else
 911                 p->clear_child_tid = NULL;
 912
 913         /*
 914          * Syscall tracing should be turned off in the child regardless
 915          * of CLONE_PTRACE.
 916          */
 917         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
 918
 919         /* Our parent execution domain becomes current domain
 920            These must match for thread signalling to apply */
 921
 922         p->parent_exec_id = p->self_exec_id;
 923
 924         /* ok, now we should be set up.. */
 925         if (clone_flags & CLONE_DETACHED)
 926                 p->exit_signal = -1;
 927         else
 928                 p->exit_signal = clone_flags & CSIGNAL;
 929         p->pdeath_signal = 0;
 930
 931         /*
 932          * Share the timeslice between parent and child, thus the
 933          * total amount of pending timeslices in the system doesn't change,
 934          * resulting in more scheduling fairness.
 935          */
 936         local_irq_disable();
 937         p->time_slice = (current->time_slice + 1) >> 1;
 938         /*
 939          * The remainder of the first timeslice might be recovered by
 940          * the parent if the child exits early enough.
 941          */
 942         p->first_time_slice = 1;
 943         current->time_slice >>= 1;
 944         p->last_run = jiffies;
 945         if (!current->time_slice) {
 946                 /*
 947                  * This case is rare, it happens when the parent has only
 948                  * a single jiffy left from its timeslice. Taking the
 949                  * runqueue lock is not a problem.
 950                  */
 951                 current->time_slice = 1;
 952                 preempt_disable();
 953                 scheduler_tick(0, 0);
 954                 local_irq_enable();
 955                 preempt_enable();
 956         } else
 957                 local_irq_enable();
 958         /*
 959          * Ok, add it to the run-queues and make it
 960          * visible to the rest of the system.
 961          *
 962          * Let it rip!
 963          */
 964         p->tgid = p->pid;
 965         p->group_leader = p;
 966         INIT_LIST_HEAD(&p->ptrace_children);
 967         INIT_LIST_HEAD(&p->ptrace_list);
 968
 969         /* Need tasklist lock for parent etc handling! */
 970         write_lock_irq(&tasklist_lock);
 971         /*
 972          * Check for pending SIGKILL! The new thread should not be allowed
 973          * to slip out of an OOM kill. (or normal SIGKILL.)
 974          */
 975         if (sigismember(&current->pending.signal, SIGKILL)) {
 976                 write_unlock_irq(&tasklist_lock);
 977                 retval = -EINTR;
 978                 goto bad_fork_cleanup_namespace;
 979         }
 980
 981         /* CLONE_PARENT re-uses the old parent */
 982         if (clone_flags & CLONE_PARENT)
 983                 p->real_parent = current->real_parent;
 984         else
 985                 p->real_parent = current;
 986         p->parent = p->real_parent;
 987
 988         if (clone_flags & CLONE_THREAD) {
 989                 spin_lock(&current->sighand->siglock);
 990                 /*
 991                  * Important: if an exit-all has been started then
 992                  * do not create this new thread - the whole thread
 993                  * group is supposed to exit anyway.
 994                  */
 995                 if (current->signal->group_exit) {
 996                         spin_unlock(&current->sighand->siglock);
 997                         write_unlock_irq(&tasklist_lock);
 998                         goto bad_fork_cleanup_namespace;
 999                 }
1000                 p->tgid = current->tgid;
1001                 p->group_leader = current->group_leader;
1002
1003                 if (current->signal->group_stop_count > 0) {
1004                         /*
1005                          * There is an all-stop in progress for the group.
1006                          * We ourselves will stop as soon as we check signals.
1007                          * Make the new thread part of that group stop too.
1008                          */
1009                         current->signal->group_stop_count++;
1010                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1011                 }
1012
1013                 spin_unlock(&current->sighand->siglock);
1014         }
1015
1016         SET_LINKS(p);
1017         if (p->ptrace & PT_PTRACED)
1018                 __ptrace_link(p, current->parent);
1019
1020         attach_pid(p, PIDTYPE_PID, p->pid);
1021         if (thread_group_leader(p)) {
1022                 attach_pid(p, PIDTYPE_TGID, p->tgid);
1023                 attach_pid(p, PIDTYPE_PGID, p->pgrp);
1024                 attach_pid(p, PIDTYPE_SID, p->session);
1025                 if (p->pid)
1026                         per_cpu(process_counts, smp_processor_id())++;
1027         } else
1028                 link_pid(p, p->pids + PIDTYPE_TGID, &p->group_leader->pids[PIDTYPE_TGID].pid);
1029
1030         nr_threads++;
1031         write_unlock_irq(&tasklist_lock);
1032         retval = 0;
1033
1034 fork_out:
1035         if (retval)
1036                 return ERR_PTR(retval);
1037         return p;
1038
1039 bad_fork_cleanup_namespace:
1040         exit_namespace(p);
1041 bad_fork_cleanup_mm:
1042         exit_mm(p);
1043 bad_fork_cleanup_signal:
1044         exit_signal(p);
1045 bad_fork_cleanup_sighand:
1046         exit_sighand(p);
1047 bad_fork_cleanup_fs:
1048         exit_fs(p); /* blocking */
1049 bad_fork_cleanup_files:
1050         exit_files(p); /* blocking */
1051 bad_fork_cleanup_semundo:
1052         exit_semundo(p);
1053 bad_fork_cleanup_security:
1054         security_task_free(p);
1055 bad_fork_cleanup:
1056         if (p->pid > 0)
1057                 free_pidmap(p->pid);
1058         if (p->binfmt)
1059                 module_put(p->binfmt->module);
1060 bad_fork_cleanup_put_domain:
1061         module_put(p->thread_info->exec_domain->module);
1062 bad_fork_cleanup_count:
1063         atomic_dec(&p->user->processes);
1064         free_uid(p->user);
1065 bad_fork_free:
1066         free_task_struct(p);
1067         goto fork_out;
1068 }
1069
1070 static inline int fork_traceflag (unsigned clone_flags)
1071 {
1072         if (clone_flags & (CLONE_UNTRACED | CLONE_IDLETASK))
1073                 return 0;
1074         else if (clone_flags & CLONE_VFORK) {
1075                 if (current->ptrace & PT_TRACE_VFORK)
1076                         return PTRACE_EVENT_VFORK;
1077         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1078                 if (current->ptrace & PT_TRACE_CLONE)
1079                         return PTRACE_EVENT_CLONE;
1080         } else if (current->ptrace & PT_TRACE_FORK)
1081                 return PTRACE_EVENT_FORK;
1082
1083         return 0;
1084 }
1085
1086 /*
1087  *  Ok, this is the main fork-routine.
1088  *
1089  * It copies the process, and if successful kick-starts
1090  * it and waits for it to finish using the VM if required.
1091  */
1092 struct task_struct *do_fork(unsigned long clone_flags,
1093                             unsigned long stack_start,
1094                             struct pt_regs *regs,
1095                             unsigned long stack_size,
1096                             int *parent_tidptr,
1097                             int *child_tidptr)
1098 {
1099         struct task_struct *p;
1100         int trace = 0;
1101
1102         if (unlikely(current->ptrace)) {
1103                 trace = fork_traceflag (clone_flags);
1104                 if (trace)
1105                         clone_flags |= CLONE_PTRACE;
1106         }
1107
1108         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr);
1109         if (!IS_ERR(p)) {
1110                 struct completion vfork;
1111
1112                 if (clone_flags & CLONE_VFORK) {
1113                         p->vfork_done = &vfork;
1114                         init_completion(&vfork);
1115                 }
1116
1117                 if (p->ptrace & PT_PTRACED) {
1118                         /*
1119                          * We'll start up with an immediate SIGSTOP.
1120                          */
1121                         sigaddset(&p->pending.signal, SIGSTOP);
1122                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1123                 }
1124
1125                 wake_up_forked_process(p);              /* do this last */
1126                 ++total_forks;
1127
1128                 if (unlikely (trace)) {
1129                         current->ptrace_message = (unsigned long) p->pid;
1130                         ptrace_notify ((trace << 8) | SIGTRAP);
1131                 }
1132
1133                 if (clone_flags & CLONE_VFORK) {
1134                         wait_for_completion(&vfork);
1135                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1136                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1137                 } else
1138                         /*
1139                          * Let the child process run first, to avoid most of the
1140                          * COW overhead when the child exec()s afterwards.
1141                          */
1142                         set_need_resched();
1143         }
1144         return p;
1145 }
1146
1147 /* SLAB cache for signal_struct structures (tsk->signal) */
1148 kmem_cache_t *signal_cachep;
1149
1150 /* SLAB cache for sighand_struct structures (tsk->sighand) */
1151 kmem_cache_t *sighand_cachep;
1152
1153 /* SLAB cache for files_struct structures (tsk->files) */
1154 kmem_cache_t *files_cachep;
1155
1156 /* SLAB cache for fs_struct structures (tsk->fs) */
1157 kmem_cache_t *fs_cachep;
1158
1159 /* SLAB cache for vm_area_struct structures */
1160 kmem_cache_t *vm_area_cachep;
1161
1162 /* SLAB cache for mm_struct structures (tsk->mm) */
1163 kmem_cache_t *mm_cachep;
1164
1165 void __init proc_caches_init(void)
1166 {
1167         sighand_cachep = kmem_cache_create("sighand_cache",
1168                         sizeof(struct sighand_struct), 0,
1169                         SLAB_HWCACHE_ALIGN, NULL, NULL);
1170         if (!sighand_cachep)
1171                 panic("Cannot create sighand SLAB cache");
1172
1173         signal_cachep = kmem_cache_create("signal_cache",
1174                         sizeof(struct signal_struct), 0,
1175                         SLAB_HWCACHE_ALIGN, NULL, NULL);
1176         if (!signal_cachep)
1177                 panic("Cannot create signal SLAB cache");
1178
1179         files_cachep = kmem_cache_create("files_cache",
1180                          sizeof(struct files_struct), 0,
1181                          SLAB_HWCACHE_ALIGN, NULL, NULL);
1182         if (!files_cachep)
1183                 panic("Cannot create files SLAB cache");
1184
1185         fs_cachep = kmem_cache_create("fs_cache",
1186                          sizeof(struct fs_struct), 0,
1187                          SLAB_HWCACHE_ALIGN, NULL, NULL);
1188         if (!fs_cachep)
1189                 panic("Cannot create fs_struct SLAB cache");
1190
1191         vm_area_cachep = kmem_cache_create("vm_area_struct",
1192                         sizeof(struct vm_area_struct), 0,
1193                         0, NULL, NULL);
1194         if(!vm_area_cachep)
1195                 panic("vma_init: Cannot alloc vm_area_struct SLAB cache");
1196
1197         mm_cachep = kmem_cache_create("mm_struct",
1198                         sizeof(struct mm_struct), 0,
1199                         SLAB_HWCACHE_ALIGN, NULL, NULL);
1200         if(!mm_cachep)
1201                 panic("vma_init: Cannot alloc mm_struct SLAB cache");
1202 }