kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/config.h>
  15 #include <linux/slab.h>
  16 #include <linux/init.h>
  17 #include <linux/unistd.h>
  18 #include <linux/smp_lock.h>
  19 #include <linux/module.h>
  20 #include <linux/vmalloc.h>
  21 #include <linux/completion.h>
  22 #include <linux/namespace.h>
  23 #include <linux/personality.h>
  24 #include <linux/mempolicy.h>
  25 #include <linux/sem.h>
  26 #include <linux/file.h>
  27 #include <linux/key.h>
  28 #include <linux/binfmts.h>
  29 #include <linux/mman.h>
  30 #include <linux/fs.h>
  31 #include <linux/cpu.h>
  32 #include <linux/security.h>
  33 #include <linux/swap.h>
  34 #include <linux/syscalls.h>
  35 #include <linux/jiffies.h>
  36 #include <linux/futex.h>
  37 #include <linux/ptrace.h>
  38 #include <linux/mount.h>
  39 #include <linux/audit.h>
  40 #include <linux/profile.h>
  41 #include <linux/rmap.h>
  42 #include <linux/vs_network.h>
  43 #include <linux/vs_limit.h>
  44 #include <linux/vs_memory.h>
  45 #include <linux/ckrm.h>
  46 #include <linux/ckrm_tsk.h>
  47 #include <linux/ckrm_mem_inline.h>
  48
  49 #include <asm/pgtable.h>
  50 #include <asm/pgalloc.h>
  51 #include <asm/uaccess.h>
  52 #include <asm/mmu_context.h>
  53 #include <asm/cacheflush.h>
  54 #include <asm/tlbflush.h>
  55
  56 /* The idle threads do not count..
  57  * Protected by write_lock_irq(&tasklist_lock)
  58  */
  59 int nr_threads;
  60
  61 int max_threads;
  62 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  63
  64 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  65
  66 rwlock_t tasklist_lock __cacheline_aligned = RW_LOCK_UNLOCKED;  /* outer */
  67
  68 EXPORT_SYMBOL(tasklist_lock);
  69
  70 int nr_processes(void)
  71 {
  72         int cpu;
  73         int total = 0;
  74
  75         for_each_online_cpu(cpu)
  76                 total += per_cpu(process_counts, cpu);
  77
  78         return total;
  79 }
  80
  81 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  82 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  83 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  84 static kmem_cache_t *task_struct_cachep;
  85 #endif
  86
  87 void free_task(struct task_struct *tsk)
  88 {
  89         free_thread_info(tsk->thread_info);
  90         clr_vx_info(&tsk->vx_info);
  91         clr_nx_info(&tsk->nx_info);
  92         free_task_struct(tsk);
  93 }
  94 EXPORT_SYMBOL(free_task);
  95
  96 void __put_task_struct(struct task_struct *tsk)
  97 {
  98         WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
  99         WARN_ON(atomic_read(&tsk->usage));
 100         WARN_ON(tsk == current);
 101
 102         if (unlikely(tsk->audit_context))
 103                 audit_free(tsk);
 104         security_task_free(tsk);
 105         free_uid(tsk->user);
 106         put_group_info(tsk->group_info);
 107
 108         if (!profile_handoff_task(tsk))
 109                 free_task(tsk);
 110 }
 111
 112 void __init fork_init(unsigned long mempages)
 113 {
 114 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 115 #ifndef ARCH_MIN_TASKALIGN
 116 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 117 #endif
 118         /* create a slab on which task_structs can be allocated */
 119         task_struct_cachep =
 120                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 121                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
 122 #endif
 123
 124         /*
 125          * The default maximum number of threads is set to a safe
 126          * value: the thread structures can take up at most half
 127          * of memory.
 128          */
 129         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 130
 131         /*
 132          * we need to allow at least 20 threads to boot a system
 133          */
 134         if(max_threads < 20)
 135                 max_threads = 20;
 136
 137         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 138         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 139 }
 140
 141 static struct task_struct *dup_task_struct(struct task_struct *orig)
 142 {
 143         struct task_struct *tsk;
 144         struct thread_info *ti;
 145
 146         prepare_to_copy(orig);
 147
 148         tsk = alloc_task_struct();
 149         if (!tsk)
 150                 return NULL;
 151
 152         ti = alloc_thread_info(tsk);
 153         if (!ti) {
 154                 free_task_struct(tsk);
 155                 return NULL;
 156         }
 157
 158         *ti = *orig->thread_info;
 159         *tsk = *orig;
 160         tsk->thread_info = ti;
 161         ti->task = tsk;
 162
 163         ckrm_cb_newtask(tsk);
 164         /* One for us, one for whoever does the "release_task()" (usually parent) */
 165         atomic_set(&tsk->usage,2);
 166 #ifdef CONFIG_CKRM_RES_MEM
 167         INIT_LIST_HEAD(&tsk->mm_peers);
 168 #endif
 169         return tsk;
 170 }
 171
 172 #ifdef CONFIG_MMU
 173 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
 174 {
 175         struct vm_area_struct * mpnt, *tmp, **pprev;
 176         struct rb_node **rb_link, *rb_parent;
 177         int retval;
 178         unsigned long charge;
 179         struct mempolicy *pol;
 180
 181         down_write(&oldmm->mmap_sem);
 182         flush_cache_mm(current->mm);
 183         mm->locked_vm = 0;
 184         mm->mmap = NULL;
 185         mm->mmap_cache = NULL;
 186         mm->free_area_cache = oldmm->mmap_base;
 187         mm->map_count = 0;
 188         mm->rss = 0;
 189         mm->anon_rss = 0;
 190         cpus_clear(mm->cpu_vm_mask);
 191         mm->mm_rb = RB_ROOT;
 192         rb_link = &mm->mm_rb.rb_node;
 193         rb_parent = NULL;
 194         pprev = &mm->mmap;
 195
 196         for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
 197                 struct file *file;
 198
 199                 if (mpnt->vm_flags & VM_DONTCOPY) {
 200                         __vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 201                                                         -vma_pages(mpnt));
 202                         continue;
 203                 }
 204                 charge = 0;
 205                 if (mpnt->vm_flags & VM_ACCOUNT) {
 206                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 207                         if (security_vm_enough_memory(len))
 208                                 goto fail_nomem;
 209                         charge = len;
 210                 }
 211                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 212                 if (!tmp)
 213                         goto fail_nomem;
 214                 *tmp = *mpnt;
 215                 pol = mpol_copy(vma_policy(mpnt));
 216                 retval = PTR_ERR(pol);
 217                 if (IS_ERR(pol))
 218                         goto fail_nomem_policy;
 219                 vma_set_policy(tmp, pol);
 220                 tmp->vm_flags &= ~VM_LOCKED;
 221                 tmp->vm_mm = mm;
 222                 tmp->vm_next = NULL;
 223                 anon_vma_link(tmp);
 224                 file = tmp->vm_file;
 225                 if (file) {
 226                         struct inode *inode = file->f_dentry->d_inode;
 227                         get_file(file);
 228                         if (tmp->vm_flags & VM_DENYWRITE)
 229                                 atomic_dec(&inode->i_writecount);
 230
 231                         /* insert tmp into the share list, just after mpnt */
 232                         spin_lock(&file->f_mapping->i_mmap_lock);
 233                         flush_dcache_mmap_lock(file->f_mapping);
 234                         vma_prio_tree_add(tmp, mpnt);
 235                         flush_dcache_mmap_unlock(file->f_mapping);
 236                         spin_unlock(&file->f_mapping->i_mmap_lock);
 237                 }
 238
 239                 /*
 240                  * Link in the new vma and copy the page table entries:
 241                  * link in first so that swapoff can see swap entries,
 242                  * and try_to_unmap_one's find_vma find the new vma.
 243                  */
 244                 spin_lock(&mm->page_table_lock);
 245                 *pprev = tmp;
 246                 pprev = &tmp->vm_next;
 247
 248                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 249                 rb_link = &tmp->vm_rb.rb_right;
 250                 rb_parent = &tmp->vm_rb;
 251
 252                 mm->map_count++;
 253                 retval = copy_page_range(mm, current->mm, tmp);
 254                 spin_unlock(&mm->page_table_lock);
 255
 256                 if (tmp->vm_ops && tmp->vm_ops->open)
 257                         tmp->vm_ops->open(tmp);
 258
 259                 if (retval)
 260                         goto out;
 261         }
 262         retval = 0;
 263
 264 out:
 265         flush_tlb_mm(current->mm);
 266         up_write(&oldmm->mmap_sem);
 267         return retval;
 268 fail_nomem_policy:
 269         kmem_cache_free(vm_area_cachep, tmp);
 270 fail_nomem:
 271         retval = -ENOMEM;
 272         vm_unacct_memory(charge);
 273         goto out;
 274 }
 275
 276 static inline int mm_alloc_pgd(struct mm_struct * mm)
 277 {
 278         mm->pgd = pgd_alloc(mm);
 279         if (unlikely(!mm->pgd))
 280                 return -ENOMEM;
 281         return 0;
 282 }
 283
 284 static inline void mm_free_pgd(struct mm_struct * mm)
 285 {
 286         pgd_free(mm->pgd);
 287 }
 288 #else
 289 #define dup_mmap(mm, oldmm)     (0)
 290 #define mm_alloc_pgd(mm)        (0)
 291 #define mm_free_pgd(mm)
 292 #endif /* CONFIG_MMU */
 293
 294 spinlock_t mmlist_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
 295
 296 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
 297 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 298
 299 #include <linux/init_task.h>
 300
 301 static struct mm_struct * mm_init(struct mm_struct * mm)
 302 {
 303         atomic_set(&mm->mm_users, 1);
 304         atomic_set(&mm->mm_count, 1);
 305         init_rwsem(&mm->mmap_sem);
 306         INIT_LIST_HEAD(&mm->mmlist);
 307         mm->core_waiters = 0;
 308         mm->nr_ptes = 0;
 309         spin_lock_init(&mm->page_table_lock);
 310         rwlock_init(&mm->ioctx_list_lock);
 311         mm->ioctx_list = NULL;
 312         mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
 313         mm->free_area_cache = TASK_UNMAPPED_BASE;
 314 #ifdef CONFIG_CKRM_RES_MEM
 315         INIT_LIST_HEAD(&mm->tasklist);
 316         mm->peertask_lock = SPIN_LOCK_UNLOCKED;
 317 #endif
 318
 319         if (likely(!mm_alloc_pgd(mm))) {
 320                 mm->def_flags = 0;
 321                 set_vx_info(&mm->mm_vx_info, current->vx_info);
 322                 return mm;
 323         }
 324         free_mm(mm);
 325         return NULL;
 326 }
 327
 328 /*
 329  * Allocate and initialize an mm_struct.
 330  */
 331 struct mm_struct * mm_alloc(void)
 332 {
 333         struct mm_struct * mm;
 334
 335         mm = allocate_mm();
 336         if (mm) {
 337                 memset(mm, 0, sizeof(*mm));
 338                 mm = mm_init(mm);
 339 #ifdef CONFIG_CKRM_RES_MEM
 340                 mm->memclass = GET_MEM_CLASS(current);
 341                 mem_class_get(mm->memclass);
 342 #endif
 343         }
 344         return mm;
 345 }
 346
 347 /*
 348  * Called when the last reference to the mm
 349  * is dropped: either by a lazy thread or by
 350  * mmput. Free the page directory and the mm.
 351  */
 352 void fastcall __mmdrop(struct mm_struct *mm)
 353 {
 354         BUG_ON(mm == &init_mm);
 355         mm_free_pgd(mm);
 356         destroy_context(mm);
 357         clr_vx_info(&mm->mm_vx_info);
 358 #ifdef CONFIG_CKRM_RES_MEM
 359         /* class can be null and mm's tasklist can be empty here */
 360         if (mm->memclass) {
 361                 mem_class_put(mm->memclass);
 362                 mm->memclass = NULL;
 363         }
 364 #endif
 365         free_mm(mm);
 366 }
 367
 368 /*
 369  * Decrement the use count and release all resources for an mm.
 370  */
 371 void mmput(struct mm_struct *mm)
 372 {
 373         if (atomic_dec_and_test(&mm->mm_users)) {
 374                 exit_aio(mm);
 375                 exit_mmap(mm);
 376                 if (!list_empty(&mm->mmlist)) {
 377                         spin_lock(&mmlist_lock);
 378                         list_del(&mm->mmlist);
 379                         spin_unlock(&mmlist_lock);
 380                 }
 381                 put_swap_token(mm);
 382                 mmdrop(mm);
 383         }
 384 }
 385 EXPORT_SYMBOL_GPL(mmput);
 386
 387 /**
 388  * get_task_mm - acquire a reference to the task's mm
 389  *
 390  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
 391  * this kernel workthread has transiently adopted a user mm with use_mm,
 392  * to do its AIO) is not set and if so returns a reference to it, after
 393  * bumping up the use count.  User must release the mm via mmput()
 394  * after use.  Typically used by /proc and ptrace.
 395  */
 396 struct mm_struct *get_task_mm(struct task_struct *task)
 397 {
 398         struct mm_struct *mm;
 399
 400         task_lock(task);
 401         mm = task->mm;
 402         if (mm) {
 403                 if (task->flags & PF_BORROWED_MM)
 404                         mm = NULL;
 405                 else
 406                         atomic_inc(&mm->mm_users);
 407         }
 408         task_unlock(task);
 409         return mm;
 410 }
 411 EXPORT_SYMBOL_GPL(get_task_mm);
 412
 413 /* Please note the differences between mmput and mm_release.
 414  * mmput is called whenever we stop holding onto a mm_struct,
 415  * error success whatever.
 416  *
 417  * mm_release is called after a mm_struct has been removed
 418  * from the current process.
 419  *
 420  * This difference is important for error handling, when we
 421  * only half set up a mm_struct for a new process and need to restore
 422  * the old one.  Because we mmput the new mm_struct before
 423  * restoring the old one. . .
 424  * Eric Biederman 10 January 1998
 425  */
 426 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 427 {
 428         struct completion *vfork_done = tsk->vfork_done;
 429
 430         /* Get rid of any cached register state */
 431         deactivate_mm(tsk, mm);
 432
 433         /* notify parent sleeping on vfork() */
 434         if (vfork_done) {
 435                 tsk->vfork_done = NULL;
 436                 complete(vfork_done);
 437         }
 438         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
 439                 u32 __user * tidptr = tsk->clear_child_tid;
 440                 tsk->clear_child_tid = NULL;
 441
 442                 /*
 443                  * We don't check the error code - if userspace has
 444                  * not set up a proper pointer then tough luck.
 445                  */
 446                 put_user(0, tidptr);
 447                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 448         }
 449 }
 450
 451 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 452 {
 453         struct mm_struct * mm, *oldmm;
 454         int retval;
 455
 456         tsk->min_flt = tsk->maj_flt = 0;
 457         tsk->nvcsw = tsk->nivcsw = 0;
 458
 459         tsk->mm = NULL;
 460         tsk->active_mm = NULL;
 461
 462         /*
 463          * Are we cloning a kernel thread?
 464          *
 465          * We need to steal a active VM for that..
 466          */
 467         oldmm = current->mm;
 468         if (!oldmm)
 469                 return 0;
 470
 471         if (clone_flags & CLONE_VM) {
 472                 atomic_inc(&oldmm->mm_users);
 473                 mm = oldmm;
 474                 /*
 475                  * There are cases where the PTL is held to ensure no
 476                  * new threads start up in user mode using an mm, which
 477                  * allows optimizing out ipis; the tlb_gather_mmu code
 478                  * is an example.
 479                  */
 480                 spin_unlock_wait(&oldmm->page_table_lock);
 481                 goto good_mm;
 482         }
 483
 484         retval = -ENOMEM;
 485         mm = allocate_mm();
 486         if (!mm)
 487                 goto fail_nomem;
 488
 489         /* Copy the current MM stuff.. */
 490         memcpy(mm, oldmm, sizeof(*mm));
 491         mm->mm_vx_info = NULL;
 492         if (!mm_init(mm))
 493                 goto fail_nomem;
 494
 495         if (init_new_context(tsk,mm))
 496                 goto fail_nocontext;
 497
 498         retval = dup_mmap(mm, oldmm);
 499         if (retval)
 500                 goto free_pt;
 501
 502 good_mm:
 503         tsk->mm = mm;
 504         tsk->active_mm = mm;
 505         ckrm_init_mm_to_task(mm, tsk);
 506         return 0;
 507
 508 free_pt:
 509         mmput(mm);
 510 fail_nomem:
 511         return retval;
 512
 513 fail_nocontext:
 514         /*
 515          * If init_new_context() failed, we cannot use mmput() to free the mm
 516          * because it calls destroy_context()
 517          */
 518         mm_free_pgd(mm);
 519         free_mm(mm);
 520         return retval;
 521 }
 522
 523 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 524 {
 525         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 526         /* We don't need to lock fs - think why ;-) */
 527         if (fs) {
 528                 atomic_set(&fs->count, 1);
 529                 rwlock_init(&fs->lock);
 530                 fs->umask = old->umask;
 531                 read_lock(&old->lock);
 532                 fs->rootmnt = mntget(old->rootmnt);
 533                 fs->root = dget(old->root);
 534                 fs->pwdmnt = mntget(old->pwdmnt);
 535                 fs->pwd = dget(old->pwd);
 536                 if (old->altroot) {
 537                         fs->altrootmnt = mntget(old->altrootmnt);
 538                         fs->altroot = dget(old->altroot);
 539                 } else {
 540                         fs->altrootmnt = NULL;
 541                         fs->altroot = NULL;
 542                 }
 543                 read_unlock(&old->lock);
 544         }
 545         return fs;
 546 }
 547
 548 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 549 {
 550         return __copy_fs_struct(old);
 551 }
 552
 553 EXPORT_SYMBOL_GPL(copy_fs_struct);
 554
 555 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 556 {
 557         if (clone_flags & CLONE_FS) {
 558                 atomic_inc(&current->fs->count);
 559                 return 0;
 560         }
 561         tsk->fs = __copy_fs_struct(current->fs);
 562         if (!tsk->fs)
 563                 return -ENOMEM;
 564         return 0;
 565 }
 566
 567 static int count_open_files(struct files_struct *files, int size)
 568 {
 569         int i;
 570
 571         /* Find the last open fd */
 572         for (i = size/(8*sizeof(long)); i > 0; ) {
 573                 if (files->open_fds->fds_bits[--i])
 574                         break;
 575         }
 576         i = (i+1) * 8 * sizeof(long);
 577         return i;
 578 }
 579
 580 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 581 {
 582         struct files_struct *oldf, *newf;
 583         struct file **old_fds, **new_fds;
 584         int open_files, nfds, size, i, error = 0;
 585
 586         /*
 587          * A background process may not have any files ...
 588          */
 589         oldf = current->files;
 590         if (!oldf)
 591                 goto out;
 592
 593         if (clone_flags & CLONE_FILES) {
 594                 atomic_inc(&oldf->count);
 595                 goto out;
 596         }
 597
 598         /*
 599          * Note: we may be using current for both targets (See exec.c)
 600          * This works because we cache current->files (old) as oldf. Don't
 601          * break this.
 602          */
 603         tsk->files = NULL;
 604         error = -ENOMEM;
 605         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
 606         if (!newf)
 607                 goto out;
 608
 609         atomic_set(&newf->count, 1);
 610
 611         spin_lock_init(&newf->file_lock);
 612         newf->next_fd       = 0;
 613         newf->max_fds       = NR_OPEN_DEFAULT;
 614         newf->max_fdset     = __FD_SETSIZE;
 615         newf->close_on_exec = &newf->close_on_exec_init;
 616         newf->open_fds      = &newf->open_fds_init;
 617         newf->fd            = &newf->fd_array[0];
 618
 619         /* We don't yet have the oldf readlock, but even if the old
 620            fdset gets grown now, we'll only copy up to "size" fds */
 621         size = oldf->max_fdset;
 622         if (size > __FD_SETSIZE) {
 623                 newf->max_fdset = 0;
 624                 spin_lock(&newf->file_lock);
 625                 error = expand_fdset(newf, size-1);
 626                 spin_unlock(&newf->file_lock);
 627                 if (error)
 628                         goto out_release;
 629         }
 630         spin_lock(&oldf->file_lock);
 631
 632         open_files = count_open_files(oldf, size);
 633
 634         /*
 635          * Check whether we need to allocate a larger fd array.
 636          * Note: we're not a clone task, so the open count won't
 637          * change.
 638          */
 639         nfds = NR_OPEN_DEFAULT;
 640         if (open_files > nfds) {
 641                 spin_unlock(&oldf->file_lock);
 642                 newf->max_fds = 0;
 643                 spin_lock(&newf->file_lock);
 644                 error = expand_fd_array(newf, open_files-1);
 645                 spin_unlock(&newf->file_lock);
 646                 if (error)
 647                         goto out_release;
 648                 nfds = newf->max_fds;
 649                 spin_lock(&oldf->file_lock);
 650         }
 651
 652         old_fds = oldf->fd;
 653         new_fds = newf->fd;
 654
 655         memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
 656         memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
 657
 658         for (i = open_files; i != 0; i--) {
 659                 struct file *f = *old_fds++;
 660                 if (f) {
 661                         get_file(f);
 662                 } else {
 663                         /*
 664                          * The fd may be claimed in the fd bitmap but not yet
 665                          * instantiated in the files array if a sibling thread
 666                          * is partway through open().  So make sure that this
 667                          * fd is available to the new process.
 668                          */
 669                         FD_CLR(open_files - i, newf->open_fds);
 670                 }
 671                 *new_fds++ = f;
 672         }
 673         spin_unlock(&oldf->file_lock);
 674
 675         /* compute the remainder to be cleared */
 676         size = (newf->max_fds - open_files) * sizeof(struct file *);
 677
 678         /* This is long word aligned thus could use a optimized version */
 679         memset(new_fds, 0, size);
 680
 681         if (newf->max_fdset > open_files) {
 682                 int left = (newf->max_fdset-open_files)/8;
 683                 int start = open_files / (8 * sizeof(unsigned long));
 684
 685                 memset(&newf->open_fds->fds_bits[start], 0, left);
 686                 memset(&newf->close_on_exec->fds_bits[start], 0, left);
 687         }
 688
 689         tsk->files = newf;
 690         error = 0;
 691 out:
 692         return error;
 693
 694 out_release:
 695         free_fdset (newf->close_on_exec, newf->max_fdset);
 696         free_fdset (newf->open_fds, newf->max_fdset);
 697         kmem_cache_free(files_cachep, newf);
 698         goto out;
 699 }
 700
 701 /*
 702  *      Helper to unshare the files of the current task.
 703  *      We don't want to expose copy_files internals to
 704  *      the exec layer of the kernel.
 705  */
 706
 707 int unshare_files(void)
 708 {
 709         struct files_struct *files  = current->files;
 710         int rc;
 711
 712         if(!files)
 713                 BUG();
 714
 715         /* This can race but the race causes us to copy when we don't
 716            need to and drop the copy */
 717         if(atomic_read(&files->count) == 1)
 718         {
 719                 atomic_inc(&files->count);
 720                 return 0;
 721         }
 722         rc = copy_files(0, current);
 723         if(rc)
 724                 current->files = files;
 725         return rc;
 726 }
 727
 728 EXPORT_SYMBOL(unshare_files);
 729
 730 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 731 {
 732         struct sighand_struct *sig;
 733
 734         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 735                 atomic_inc(&current->sighand->count);
 736                 return 0;
 737         }
 738         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 739         tsk->sighand = sig;
 740         if (!sig)
 741                 return -ENOMEM;
 742         spin_lock_init(&sig->siglock);
 743         atomic_set(&sig->count, 1);
 744         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 745         return 0;
 746 }
 747
 748 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 749 {
 750         struct signal_struct *sig;
 751
 752         if (clone_flags & CLONE_THREAD) {
 753                 atomic_inc(&current->signal->count);
 754                 atomic_inc(&current->signal->live);
 755                 return 0;
 756         }
 757         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 758         tsk->signal = sig;
 759         if (!sig)
 760                 return -ENOMEM;
 761         atomic_set(&sig->count, 1);
 762         atomic_set(&sig->live, 1);
 763         sig->group_exit = 0;
 764         sig->group_exit_code = 0;
 765         sig->group_exit_task = NULL;
 766         sig->group_stop_count = 0;
 767         sig->stop_state = 0;
 768         sig->curr_target = NULL;
 769         init_sigpending(&sig->shared_pending);
 770         INIT_LIST_HEAD(&sig->posix_timers);
 771
 772         sig->tty = current->signal->tty;
 773         sig->pgrp = process_group(current);
 774         sig->session = current->signal->session;
 775         sig->leader = 0;        /* session leadership doesn't inherit */
 776         sig->tty_old_pgrp = 0;
 777
 778         sig->utime = sig->stime = sig->cutime = sig->cstime = 0;
 779         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 780         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 781
 782         task_lock(current->group_leader);
 783         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 784         task_unlock(current->group_leader);
 785
 786         return 0;
 787 }
 788
 789 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 790 {
 791         unsigned long new_flags = p->flags;
 792
 793         new_flags &= ~PF_SUPERPRIV;
 794         new_flags |= PF_FORKNOEXEC;
 795         if (!(clone_flags & CLONE_PTRACE))
 796                 p->ptrace = 0;
 797         p->flags = new_flags;
 798 }
 799
 800 asmlinkage long sys_set_tid_address(int __user *tidptr)
 801 {
 802         current->clear_child_tid = tidptr;
 803
 804         return current->pid;
 805 }
 806
 807 /*
 808  * This creates a new process as a copy of the old one,
 809  * but does not actually start it yet.
 810  *
 811  * It copies the registers, and all the appropriate
 812  * parts of the process environment (as per the clone
 813  * flags). The actual kick-off is left to the caller.
 814  */
 815 static task_t *copy_process(unsigned long clone_flags,
 816                                  unsigned long stack_start,
 817                                  struct pt_regs *regs,
 818                                  unsigned long stack_size,
 819                                  int __user *parent_tidptr,
 820                                  int __user *child_tidptr,
 821                                  int pid)
 822 {
 823         int retval;
 824         struct task_struct *p = NULL;
 825         struct vx_info *vxi;
 826
 827         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 828                 return ERR_PTR(-EINVAL);
 829
 830         /*
 831          * Thread groups must share signals as well, and detached threads
 832          * can only be started up within the thread group.
 833          */
 834         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 835                 return ERR_PTR(-EINVAL);
 836
 837         /*
 838          * Shared signal handlers imply shared VM. By way of the above,
 839          * thread groups also imply shared VM. Blocking this case allows
 840          * for various simplifications in other code.
 841          */
 842         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 843                 return ERR_PTR(-EINVAL);
 844
 845         retval = security_task_create(clone_flags);
 846         if (retval)
 847                 goto fork_out;
 848
 849         retval = -ENOMEM;
 850         p = dup_task_struct(current);
 851         if (!p)
 852                 goto fork_out;
 853         p->tux_info = NULL;
 854
 855         p->vx_info = NULL;
 856         set_vx_info(&p->vx_info, current->vx_info);
 857         p->nx_info = NULL;
 858         set_nx_info(&p->nx_info, current->nx_info);
 859
 860         /* check vserver memory */
 861         if (p->mm && !(clone_flags & CLONE_VM)) {
 862                 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
 863                         vx_pages_add(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
 864                 else
 865                         goto bad_fork_free;
 866         }
 867         if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
 868                 if (!vx_rsspages_avail(p->mm, p->mm->rss))
 869                         goto bad_fork_cleanup_vm;
 870         }
 871
 872         retval = -EAGAIN;
 873         if (!vx_nproc_avail(1))
 874                 goto bad_fork_cleanup_vm;
 875
 876         if (atomic_read(&p->user->processes) >=
 877                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
 878                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 879                                 p->user != &root_user)
 880                         goto bad_fork_cleanup_vm;
 881         }
 882
 883         atomic_inc(&p->user->__count);
 884         atomic_inc(&p->user->processes);
 885         get_group_info(p->group_info);
 886
 887         /*
 888          * If multiple threads are within copy_process(), then this check
 889          * triggers too late. This doesn't hurt, the check is only there
 890          * to stop root fork bombs.
 891          */
 892         if (nr_threads >= max_threads)
 893                 goto bad_fork_cleanup_count;
 894
 895         if (!try_module_get(p->thread_info->exec_domain->module))
 896                 goto bad_fork_cleanup_count;
 897
 898         if (p->binfmt && !try_module_get(p->binfmt->module))
 899                 goto bad_fork_cleanup_put_domain;
 900
 901         init_delays(p);
 902         p->did_exec = 0;
 903         copy_flags(clone_flags, p);
 904         p->pid = pid;
 905         retval = -EFAULT;
 906         if (clone_flags & CLONE_PARENT_SETTID)
 907                 if (put_user(p->pid, parent_tidptr))
 908                         goto bad_fork_cleanup;
 909
 910         p->proc_dentry = NULL;
 911
 912         INIT_LIST_HEAD(&p->children);
 913         INIT_LIST_HEAD(&p->sibling);
 914         init_waitqueue_head(&p->wait_chldexit);
 915         p->vfork_done = NULL;
 916         spin_lock_init(&p->alloc_lock);
 917         spin_lock_init(&p->proc_lock);
 918
 919         clear_tsk_thread_flag(p, TIF_SIGPENDING);
 920         init_sigpending(&p->pending);
 921
 922         p->it_real_value = p->it_virt_value = p->it_prof_value = 0;
 923         p->it_real_incr = p->it_virt_incr = p->it_prof_incr = 0;
 924         init_timer(&p->real_timer);
 925         p->real_timer.data = (unsigned long) p;
 926
 927         p->utime = p->stime = 0;
 928         p->lock_depth = -1;             /* -1 = no lock */
 929         do_posix_clock_monotonic_gettime(&p->start_time);
 930         p->security = NULL;
 931         p->io_context = NULL;
 932         p->io_wait = NULL;
 933         p->audit_context = NULL;
 934 #ifdef CONFIG_NUMA
 935         p->mempolicy = mpol_copy(p->mempolicy);
 936         if (IS_ERR(p->mempolicy)) {
 937                 retval = PTR_ERR(p->mempolicy);
 938                 p->mempolicy = NULL;
 939                 goto bad_fork_cleanup;
 940         }
 941 #endif
 942
 943         p->tgid = p->pid;
 944         if (clone_flags & CLONE_THREAD)
 945                 p->tgid = current->tgid;
 946
 947         if ((retval = security_task_alloc(p)))
 948                 goto bad_fork_cleanup_policy;
 949         if ((retval = audit_alloc(p)))
 950                 goto bad_fork_cleanup_security;
 951         /* copy all the process information */
 952         if ((retval = copy_semundo(clone_flags, p)))
 953                 goto bad_fork_cleanup_audit;
 954         if ((retval = copy_files(clone_flags, p)))
 955                 goto bad_fork_cleanup_semundo;
 956         if ((retval = copy_fs(clone_flags, p)))
 957                 goto bad_fork_cleanup_files;
 958         if ((retval = copy_sighand(clone_flags, p)))
 959                 goto bad_fork_cleanup_fs;
 960         if ((retval = copy_signal(clone_flags, p)))
 961                 goto bad_fork_cleanup_sighand;
 962         if ((retval = copy_mm(clone_flags, p)))
 963                 goto bad_fork_cleanup_signal;
 964         if ((retval = copy_keys(clone_flags, p)))
 965                 goto bad_fork_cleanup_mm;
 966         if ((retval = copy_namespace(clone_flags, p)))
 967                 goto bad_fork_cleanup_keys;
 968         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
 969         if (retval)
 970                 goto bad_fork_cleanup_namespace;
 971
 972         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
 973         /*
 974          * Clear TID on mm_release()?
 975          */
 976         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
 977
 978         /*
 979          * Syscall tracing should be turned off in the child regardless
 980          * of CLONE_PTRACE.
 981          */
 982         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
 983
 984         /* Our parent execution domain becomes current domain
 985            These must match for thread signalling to apply */
 986
 987         p->parent_exec_id = p->self_exec_id;
 988
 989         /* ok, now we should be set up.. */
 990         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
 991         p->pdeath_signal = 0;
 992         p->exit_state = 0;
 993
 994         /* Perform scheduler related setup */
 995         sched_fork(p);
 996
 997         /*
 998          * Ok, make it visible to the rest of the system.
 999          * We dont wake it up yet.
1000          */
1001         p->group_leader = p;
1002         INIT_LIST_HEAD(&p->ptrace_children);
1003         INIT_LIST_HEAD(&p->ptrace_list);
1004
1005         /* Need tasklist lock for parent etc handling! */
1006         write_lock_irq(&tasklist_lock);
1007
1008         /*
1009          * The task hasn't been attached yet, so cpus_allowed mask cannot
1010          * have changed. The cpus_allowed mask of the parent may have
1011          * changed after it was copied first time, and it may then move to
1012          * another CPU - so we re-copy it here and set the child's CPU to
1013          * the parent's CPU. This avoids alot of nasty races.
1014          */
1015         p->cpus_allowed = current->cpus_allowed;
1016         set_task_cpu(p, smp_processor_id());
1017
1018         /*
1019          * Check for pending SIGKILL! The new thread should not be allowed
1020          * to slip out of an OOM kill. (or normal SIGKILL.)
1021          */
1022         if (sigismember(&current->pending.signal, SIGKILL)) {
1023                 write_unlock_irq(&tasklist_lock);
1024                 retval = -EINTR;
1025                 goto bad_fork_cleanup_namespace;
1026         }
1027
1028         /* CLONE_PARENT re-uses the old parent */
1029         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1030                 p->real_parent = current->real_parent;
1031         else
1032                 p->real_parent = current;
1033         p->parent = p->real_parent;
1034
1035         if (clone_flags & CLONE_THREAD) {
1036                 spin_lock(&current->sighand->siglock);
1037                 /*
1038                  * Important: if an exit-all has been started then
1039                  * do not create this new thread - the whole thread
1040                  * group is supposed to exit anyway.
1041                  */
1042                 if (current->signal->group_exit) {
1043                         spin_unlock(&current->sighand->siglock);
1044                         write_unlock_irq(&tasklist_lock);
1045                         retval = -EAGAIN;
1046                         goto bad_fork_cleanup_namespace;
1047                 }
1048                 p->group_leader = current->group_leader;
1049
1050                 if (current->signal->group_stop_count > 0) {
1051                         /*
1052                          * There is an all-stop in progress for the group.
1053                          * We ourselves will stop as soon as we check signals.
1054                          * Make the new thread part of that group stop too.
1055                          */
1056                         current->signal->group_stop_count++;
1057                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1058                 }
1059
1060                 spin_unlock(&current->sighand->siglock);
1061         }
1062
1063         SET_LINKS(p);
1064         if (unlikely(p->ptrace & PT_PTRACED))
1065                 __ptrace_link(p, current->parent);
1066
1067         attach_pid(p, PIDTYPE_PID, p->pid);
1068         attach_pid(p, PIDTYPE_TGID, p->tgid);
1069         if (thread_group_leader(p)) {
1070                 attach_pid(p, PIDTYPE_PGID, process_group(p));
1071                 attach_pid(p, PIDTYPE_SID, p->signal->session);
1072                 if (p->pid)
1073                         __get_cpu_var(process_counts)++;
1074         }
1075
1076         p->ioprio = current->ioprio;
1077         nr_threads++;
1078         /* p is copy of current */
1079         vxi = p->vx_info;
1080         if (vxi) {
1081                 atomic_inc(&vxi->cvirt.nr_threads);
1082                 vx_nproc_inc(p);
1083         }
1084         write_unlock_irq(&tasklist_lock);
1085         retval = 0;
1086
1087 fork_out:
1088         if (retval)
1089                 return ERR_PTR(retval);
1090         return p;
1091
1092 bad_fork_cleanup_namespace:
1093         exit_namespace(p);
1094 bad_fork_cleanup_keys:
1095         exit_keys(p);
1096 bad_fork_cleanup_mm:
1097         if (p->mm)
1098                 mmput(p->mm);
1099 bad_fork_cleanup_signal:
1100         exit_signal(p);
1101 bad_fork_cleanup_sighand:
1102         exit_sighand(p);
1103 bad_fork_cleanup_fs:
1104         exit_fs(p); /* blocking */
1105 bad_fork_cleanup_files:
1106         exit_files(p); /* blocking */
1107 bad_fork_cleanup_semundo:
1108         exit_sem(p);
1109 bad_fork_cleanup_audit:
1110         audit_free(p);
1111 bad_fork_cleanup_security:
1112         security_task_free(p);
1113 bad_fork_cleanup_policy:
1114 #ifdef CONFIG_NUMA
1115         mpol_free(p->mempolicy);
1116 #endif
1117 bad_fork_cleanup:
1118         if (p->binfmt)
1119                 module_put(p->binfmt->module);
1120 bad_fork_cleanup_put_domain:
1121         module_put(p->thread_info->exec_domain->module);
1122 bad_fork_cleanup_count:
1123         put_group_info(p->group_info);
1124         atomic_dec(&p->user->processes);
1125         free_uid(p->user);
1126 bad_fork_cleanup_vm:
1127         if (p->mm && !(clone_flags & CLONE_VM))
1128                 vx_pages_sub(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
1129 bad_fork_free:
1130         free_task(p);
1131         goto fork_out;
1132 }
1133
1134 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1135 {
1136         memset(regs, 0, sizeof(struct pt_regs));
1137         return regs;
1138 }
1139
1140 task_t * __devinit fork_idle(int cpu)
1141 {
1142         task_t *task;
1143         struct pt_regs regs;
1144
1145         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1146         if (!task)
1147                 return ERR_PTR(-ENOMEM);
1148         init_idle(task, cpu);
1149         unhash_process(task);
1150         return task;
1151 }
1152
1153 static inline int fork_traceflag (unsigned clone_flags)
1154 {
1155         if (clone_flags & CLONE_UNTRACED)
1156                 return 0;
1157         else if (clone_flags & CLONE_VFORK) {
1158                 if (current->ptrace & PT_TRACE_VFORK)
1159                         return PTRACE_EVENT_VFORK;
1160         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1161                 if (current->ptrace & PT_TRACE_CLONE)
1162                         return PTRACE_EVENT_CLONE;
1163         } else if (current->ptrace & PT_TRACE_FORK)
1164                 return PTRACE_EVENT_FORK;
1165
1166         return 0;
1167 }
1168
1169 /*
1170  *  Ok, this is the main fork-routine.
1171  *
1172  * It copies the process, and if successful kick-starts
1173  * it and waits for it to finish using the VM if required.
1174  */
1175 long do_fork(unsigned long clone_flags,
1176               unsigned long stack_start,
1177               struct pt_regs *regs,
1178               unsigned long stack_size,
1179               int __user *parent_tidptr,
1180               int __user *child_tidptr)
1181 {
1182         struct task_struct *p;
1183         int trace = 0;
1184         long pid = alloc_pidmap();
1185
1186         if (pid < 0)
1187                 return -EAGAIN;
1188         if (unlikely(current->ptrace)) {
1189                 trace = fork_traceflag (clone_flags);
1190                 if (trace)
1191                         clone_flags |= CLONE_PTRACE;
1192         }
1193
1194 #ifdef CONFIG_CKRM_TYPE_TASKCLASS
1195         if (numtasks_get_ref(current->taskclass, 0) == 0) {
1196                 return -ENOMEM;
1197         }
1198 #endif
1199
1200         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
1201         /*
1202          * Do this prior waking up the new thread - the thread pointer
1203          * might get invalid after that point, if the thread exits quickly.
1204          */
1205         if (!IS_ERR(p)) {
1206                 struct completion vfork;
1207
1208                 ckrm_cb_fork(p);
1209
1210                 if (clone_flags & CLONE_VFORK) {
1211                         p->vfork_done = &vfork;
1212                         init_completion(&vfork);
1213                 }
1214
1215                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1216                         /*
1217                          * We'll start up with an immediate SIGSTOP.
1218                          */
1219                         sigaddset(&p->pending.signal, SIGSTOP);
1220                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1221                 }
1222
1223                 if (!(clone_flags & CLONE_STOPPED))
1224                         wake_up_new_task(p, clone_flags);
1225                 else
1226                         p->state = TASK_STOPPED;
1227                 ++total_forks;
1228
1229                 if (unlikely (trace)) {
1230                         current->ptrace_message = pid;
1231                         ptrace_notify ((trace << 8) | SIGTRAP);
1232                 }
1233
1234                 if (clone_flags & CLONE_VFORK) {
1235                         wait_for_completion(&vfork);
1236                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1237                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1238                 }
1239         } else {
1240 #ifdef CONFIG_CKRM_TYPE_TASKCLASS
1241                 numtasks_put_ref(current->taskclass);
1242 #endif
1243                 free_pidmap(pid);
1244                 pid = PTR_ERR(p);
1245         }
1246         return pid;
1247 }
1248
1249 /* SLAB cache for signal_struct structures (tsk->signal) */
1250 kmem_cache_t *signal_cachep;
1251
1252 /* SLAB cache for sighand_struct structures (tsk->sighand) */
1253 kmem_cache_t *sighand_cachep;
1254
1255 /* SLAB cache for files_struct structures (tsk->files) */
1256 kmem_cache_t *files_cachep;
1257
1258 /* SLAB cache for fs_struct structures (tsk->fs) */
1259 kmem_cache_t *fs_cachep;
1260
1261 /* SLAB cache for vm_area_struct structures */
1262 kmem_cache_t *vm_area_cachep;
1263
1264 /* SLAB cache for mm_struct structures (tsk->mm) */
1265 kmem_cache_t *mm_cachep;
1266
1267 void __init proc_caches_init(void)
1268 {
1269         sighand_cachep = kmem_cache_create("sighand_cache",
1270                         sizeof(struct sighand_struct), 0,
1271                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1272         signal_cachep = kmem_cache_create("signal_cache",
1273                         sizeof(struct signal_struct), 0,
1274                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1275         files_cachep = kmem_cache_create("files_cache",
1276                         sizeof(struct files_struct), 0,
1277                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1278         fs_cachep = kmem_cache_create("fs_cache",
1279                         sizeof(struct fs_struct), 0,
1280                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1281         vm_area_cachep = kmem_cache_create("vm_area_struct",
1282                         sizeof(struct vm_area_struct), 0,
1283                         SLAB_PANIC, NULL, NULL);
1284         mm_cachep = kmem_cache_create("mm_struct",
1285                         sizeof(struct mm_struct), 0,
1286                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1287 }