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/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/fs.h>
  30 #include <linux/cpu.h>
  31 #include <linux/security.h>
  32 #include <linux/syscalls.h>
  33 #include <linux/jiffies.h>
  34 #include <linux/futex.h>
  35 #include <linux/ptrace.h>
  36 #include <linux/mount.h>
  37 #include <linux/audit.h>
  38 #include <linux/rmap.h>
  39 #include <linux/ckrm.h>
  40 #include <linux/ckrm_tsk.h>
  41
  42 #include <asm/pgtable.h>
  43 #include <asm/pgalloc.h>
  44 #include <asm/uaccess.h>
  45 #include <asm/mmu_context.h>
  46 #include <asm/cacheflush.h>
  47 #include <asm/tlbflush.h>
  48
  49 /* The idle threads do not count..
  50  * Protected by write_lock_irq(&tasklist_lock)
  51  */
  52 int nr_threads;
  53
  54 int max_threads;
  55 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  56
  57 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  58
  59 rwlock_t tasklist_lock __cacheline_aligned = RW_LOCK_UNLOCKED;  /* outer */
  60
  61 EXPORT_SYMBOL(tasklist_lock);
  62
  63 int nr_processes(void)
  64 {
  65         int cpu;
  66         int total = 0;
  67
  68         for_each_online_cpu(cpu)
  69                 total += per_cpu(process_counts, cpu);
  70
  71         return total;
  72 }
  73
  74 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  75 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  76 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  77 static kmem_cache_t *task_struct_cachep;
  78 #endif
  79
  80 static void free_task(struct task_struct *tsk)
  81 {
  82         free_thread_info(tsk->thread_info);
  83         free_task_struct(tsk);
  84 }
  85
  86 void __put_task_struct(struct task_struct *tsk)
  87 {
  88         WARN_ON(!(tsk->state & (TASK_DEAD | TASK_ZOMBIE)));
  89         WARN_ON(atomic_read(&tsk->usage));
  90         WARN_ON(tsk == current);
  91
  92         if (unlikely(tsk->audit_context))
  93                 audit_free(tsk);
  94         security_task_free(tsk);
  95         free_uid(tsk->user);
  96         put_group_info(tsk->group_info);
  97         free_task(tsk);
  98 }
  99
 100 void fastcall add_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
 101 {
 102         unsigned long flags;
 103
 104         wait->flags &= ~WQ_FLAG_EXCLUSIVE;
 105         spin_lock_irqsave(&q->lock, flags);
 106         __add_wait_queue(q, wait);
 107         spin_unlock_irqrestore(&q->lock, flags);
 108 }
 109
 110 EXPORT_SYMBOL(add_wait_queue);
 111
 112 void fastcall add_wait_queue_exclusive(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_tail(q, wait);
 119         spin_unlock_irqrestore(&q->lock, flags);
 120 }
 121
 122 EXPORT_SYMBOL(add_wait_queue_exclusive);
 123
 124 void fastcall remove_wait_queue(wait_queue_head_t *q, wait_queue_t * wait)
 125 {
 126         unsigned long flags;
 127
 128         spin_lock_irqsave(&q->lock, flags);
 129         __remove_wait_queue(q, wait);
 130         spin_unlock_irqrestore(&q->lock, flags);
 131 }
 132
 133 EXPORT_SYMBOL(remove_wait_queue);
 134
 135
 136 /*
 137  * Note: we use "set_current_state()" _after_ the wait-queue add,
 138  * because we need a memory barrier there on SMP, so that any
 139  * wake-function that tests for the wait-queue being active
 140  * will be guaranteed to see waitqueue addition _or_ subsequent
 141  * tests in this thread will see the wakeup having taken place.
 142  *
 143  * The spin_unlock() itself is semi-permeable and only protects
 144  * one way (it only protects stuff inside the critical region and
 145  * stops them from bleeding out - it would still allow subsequent
 146  * loads to move into the the critical region).
 147  */
 148 void fastcall prepare_to_wait(wait_queue_head_t *q, wait_queue_t *wait, int state)
 149 {
 150         unsigned long flags;
 151
 152         wait->flags &= ~WQ_FLAG_EXCLUSIVE;
 153         spin_lock_irqsave(&q->lock, flags);
 154         if (list_empty(&wait->task_list))
 155                 __add_wait_queue(q, wait);
 156         set_current_state(state);
 157         spin_unlock_irqrestore(&q->lock, flags);
 158 }
 159
 160 EXPORT_SYMBOL(prepare_to_wait);
 161
 162 void fastcall
 163 prepare_to_wait_exclusive(wait_queue_head_t *q, wait_queue_t *wait, int state)
 164 {
 165         unsigned long flags;
 166
 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         set_current_state(state);
 172         spin_unlock_irqrestore(&q->lock, flags);
 173 }
 174
 175 EXPORT_SYMBOL(prepare_to_wait_exclusive);
 176
 177 void fastcall finish_wait(wait_queue_head_t *q, wait_queue_t *wait)
 178 {
 179         unsigned long flags;
 180
 181         __set_current_state(TASK_RUNNING);
 182         /*
 183          * We can check for list emptiness outside the lock
 184          * IFF:
 185          *  - we use the "careful" check that verifies both
 186          *    the next and prev pointers, so that there cannot
 187          *    be any half-pending updates in progress on other
 188          *    CPU's that we haven't seen yet (and that might
 189          *    still change the stack area.
 190          * and
 191          *  - all other users take the lock (ie we can only
 192          *    have _one_ other CPU that looks at or modifies
 193          *    the list).
 194          */
 195         if (!list_empty_careful(&wait->task_list)) {
 196                 spin_lock_irqsave(&q->lock, flags);
 197                 list_del_init(&wait->task_list);
 198                 spin_unlock_irqrestore(&q->lock, flags);
 199         }
 200 }
 201
 202 EXPORT_SYMBOL(finish_wait);
 203
 204 int autoremove_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
 205 {
 206         int ret = default_wake_function(wait, mode, sync, key);
 207
 208         if (ret)
 209                 list_del_init(&wait->task_list);
 210         return ret;
 211 }
 212
 213 EXPORT_SYMBOL(autoremove_wake_function);
 214
 215 void __init fork_init(unsigned long mempages)
 216 {
 217 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 218 #ifndef ARCH_MIN_TASKALIGN
 219 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 220 #endif
 221         /* create a slab on which task_structs can be allocated */
 222         task_struct_cachep =
 223                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 224                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
 225 #endif
 226
 227         /*
 228          * The default maximum number of threads is set to a safe
 229          * value: the thread structures can take up at most half
 230          * of memory.
 231          */
 232         max_threads = mempages / (THREAD_SIZE/PAGE_SIZE) / 8;
 233         /*
 234          * we need to allow at least 20 threads to boot a system
 235          */
 236         if(max_threads < 20)
 237                 max_threads = 20;
 238
 239         init_task.rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 240         init_task.rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 241 }
 242
 243 static struct task_struct *dup_task_struct(struct task_struct *orig)
 244 {
 245         struct task_struct *tsk;
 246         struct thread_info *ti;
 247
 248         prepare_to_copy(orig);
 249
 250         tsk = alloc_task_struct();
 251         if (!tsk)
 252                 return NULL;
 253
 254         ti = alloc_thread_info(tsk);
 255         if (!ti) {
 256                 free_task_struct(tsk);
 257                 return NULL;
 258         }
 259
 260         *ti = *orig->thread_info;
 261         *tsk = *orig;
 262         tsk->thread_info = ti;
 263         ti->task = tsk;
 264
 265         ckrm_cb_newtask(tsk);
 266         /* One for us, one for whoever does the "release_task()" (usually parent) */
 267         atomic_set(&tsk->usage,2);
 268         return tsk;
 269 }
 270
 271 #ifdef CONFIG_MMU
 272 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
 273 {
 274         struct vm_area_struct * mpnt, *tmp, **pprev;
 275         struct rb_node **rb_link, *rb_parent;
 276         int retval;
 277         unsigned long charge;
 278         struct mempolicy *pol;
 279
 280         down_write(&oldmm->mmap_sem);
 281         flush_cache_mm(current->mm);
 282         mm->locked_vm = 0;
 283         mm->mmap = NULL;
 284         mm->mmap_cache = NULL;
 285         mm->free_area_cache = TASK_UNMAPPED_BASE;
 286         mm->map_count = 0;
 287         mm->rss = 0;
 288         cpus_clear(mm->cpu_vm_mask);
 289         mm->mm_rb = RB_ROOT;
 290         rb_link = &mm->mm_rb.rb_node;
 291         rb_parent = NULL;
 292         pprev = &mm->mmap;
 293
 294         /*
 295          * Add it to the mmlist after the parent.
 296          * Doing it this way means that we can order the list,
 297          * and fork() won't mess up the ordering significantly.
 298          * Add it first so that swapoff can see any swap entries.
 299          */
 300         spin_lock(&mmlist_lock);
 301         list_add(&mm->mmlist, &current->mm->mmlist);
 302         mmlist_nr++;
 303         spin_unlock(&mmlist_lock);
 304
 305         for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
 306                 struct file *file;
 307
 308                 if(mpnt->vm_flags & VM_DONTCOPY)
 309                         continue;
 310                 charge = 0;
 311                 if (mpnt->vm_flags & VM_ACCOUNT) {
 312                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 313                         if (security_vm_enough_memory(len))
 314                                 goto fail_nomem;
 315                         charge = len;
 316                 }
 317                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 318                 if (!tmp)
 319                         goto fail_nomem;
 320                 *tmp = *mpnt;
 321                 pol = mpol_copy(vma_policy(mpnt));
 322                 retval = PTR_ERR(pol);
 323                 if (IS_ERR(pol))
 324                         goto fail_nomem_policy;
 325                 vma_set_policy(tmp, pol);
 326                 tmp->vm_flags &= ~VM_LOCKED;
 327                 tmp->vm_mm = mm;
 328                 tmp->vm_next = NULL;
 329                 anon_vma_link(tmp);
 330                 vma_prio_tree_init(tmp);
 331                 file = tmp->vm_file;
 332                 if (file) {
 333                         struct inode *inode = file->f_dentry->d_inode;
 334                         get_file(file);
 335                         if (tmp->vm_flags & VM_DENYWRITE)
 336                                 atomic_dec(&inode->i_writecount);
 337
 338                         /* insert tmp into the share list, just after mpnt */
 339                         spin_lock(&file->f_mapping->i_mmap_lock);
 340                         flush_dcache_mmap_lock(file->f_mapping);
 341                         vma_prio_tree_add(tmp, mpnt);
 342                         flush_dcache_mmap_unlock(file->f_mapping);
 343                         spin_unlock(&file->f_mapping->i_mmap_lock);
 344                 }
 345
 346                 /*
 347                  * Link in the new vma and copy the page table entries:
 348                  * link in first so that swapoff can see swap entries,
 349                  * and try_to_unmap_one's find_vma find the new vma.
 350                  */
 351                 spin_lock(&mm->page_table_lock);
 352                 *pprev = tmp;
 353                 pprev = &tmp->vm_next;
 354
 355                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 356                 rb_link = &tmp->vm_rb.rb_right;
 357                 rb_parent = &tmp->vm_rb;
 358
 359                 mm->map_count++;
 360                 retval = copy_page_range(mm, current->mm, tmp);
 361                 spin_unlock(&mm->page_table_lock);
 362
 363                 if (tmp->vm_ops && tmp->vm_ops->open)
 364                         tmp->vm_ops->open(tmp);
 365
 366                 if (retval)
 367                         goto out;
 368         }
 369         retval = 0;
 370
 371 out:
 372         flush_tlb_mm(current->mm);
 373         up_write(&oldmm->mmap_sem);
 374         return retval;
 375 fail_nomem_policy:
 376         kmem_cache_free(vm_area_cachep, tmp);
 377 fail_nomem:
 378         retval = -ENOMEM;
 379         vm_unacct_memory(charge);
 380         goto out;
 381 }
 382
 383 static inline int mm_alloc_pgd(struct mm_struct * mm)
 384 {
 385         mm->pgd = pgd_alloc(mm);
 386         if (unlikely(!mm->pgd))
 387                 return -ENOMEM;
 388         return 0;
 389 }
 390
 391 static inline void mm_free_pgd(struct mm_struct * mm)
 392 {
 393         pgd_free(mm->pgd);
 394 }
 395 #else
 396 #define dup_mmap(mm, oldmm)     (0)
 397 #define mm_alloc_pgd(mm)        (0)
 398 #define mm_free_pgd(mm)
 399 #endif /* CONFIG_MMU */
 400
 401 spinlock_t mmlist_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
 402 int mmlist_nr;
 403
 404 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
 405 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 406
 407 #include <linux/init_task.h>
 408
 409 static struct mm_struct * mm_init(struct mm_struct * mm)
 410 {
 411         atomic_set(&mm->mm_users, 1);
 412         atomic_set(&mm->mm_count, 1);
 413         init_rwsem(&mm->mmap_sem);
 414         mm->core_waiters = 0;
 415         mm->page_table_lock = SPIN_LOCK_UNLOCKED;
 416         mm->ioctx_list_lock = RW_LOCK_UNLOCKED;
 417         mm->ioctx_list = NULL;
 418         mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
 419         mm->free_area_cache = TASK_UNMAPPED_BASE;
 420
 421         if (likely(!mm_alloc_pgd(mm))) {
 422                 mm->def_flags = 0;
 423                 return mm;
 424         }
 425         free_mm(mm);
 426         return NULL;
 427 }
 428
 429 /*
 430  * Allocate and initialize an mm_struct.
 431  */
 432 struct mm_struct * mm_alloc(void)
 433 {
 434         struct mm_struct * mm;
 435
 436         mm = allocate_mm();
 437         if (mm) {
 438                 memset(mm, 0, sizeof(*mm));
 439                 mm = mm_init(mm);
 440         }
 441         return mm;
 442 }
 443
 444 /*
 445  * Called when the last reference to the mm
 446  * is dropped: either by a lazy thread or by
 447  * mmput. Free the page directory and the mm.
 448  */
 449 void fastcall __mmdrop(struct mm_struct *mm)
 450 {
 451         BUG_ON(mm == &init_mm);
 452         mm_free_pgd(mm);
 453         destroy_context(mm);
 454         free_mm(mm);
 455 }
 456
 457 /*
 458  * Decrement the use count and release all resources for an mm.
 459  */
 460 void mmput(struct mm_struct *mm)
 461 {
 462         if (atomic_dec_and_lock(&mm->mm_users, &mmlist_lock)) {
 463                 list_del(&mm->mmlist);
 464                 mmlist_nr--;
 465                 spin_unlock(&mmlist_lock);
 466                 exit_aio(mm);
 467                 exit_mmap(mm);
 468                 mmdrop(mm);
 469         }
 470 }
 471
 472 /*
 473  * Checks if the use count of an mm is non-zero and if so
 474  * returns a reference to it after bumping up the use count.
 475  * If the use count is zero, it means this mm is going away,
 476  * so return NULL.
 477  */
 478 struct mm_struct *mmgrab(struct mm_struct *mm)
 479 {
 480         spin_lock(&mmlist_lock);
 481         if (!atomic_read(&mm->mm_users))
 482                 mm = NULL;
 483         else
 484                 atomic_inc(&mm->mm_users);
 485         spin_unlock(&mmlist_lock);
 486         return mm;
 487 }
 488
 489 /* Please note the differences between mmput and mm_release.
 490  * mmput is called whenever we stop holding onto a mm_struct,
 491  * error success whatever.
 492  *
 493  * mm_release is called after a mm_struct has been removed
 494  * from the current process.
 495  *
 496  * This difference is important for error handling, when we
 497  * only half set up a mm_struct for a new process and need to restore
 498  * the old one.  Because we mmput the new mm_struct before
 499  * restoring the old one. . .
 500  * Eric Biederman 10 January 1998
 501  */
 502 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 503 {
 504         struct completion *vfork_done = tsk->vfork_done;
 505
 506         /* Get rid of any cached register state */
 507         deactivate_mm(tsk, mm);
 508
 509         /* notify parent sleeping on vfork() */
 510         if (vfork_done) {
 511                 tsk->vfork_done = NULL;
 512                 complete(vfork_done);
 513         }
 514         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
 515                 u32 __user * tidptr = tsk->clear_child_tid;
 516                 tsk->clear_child_tid = NULL;
 517
 518                 /*
 519                  * We don't check the error code - if userspace has
 520                  * not set up a proper pointer then tough luck.
 521                  */
 522                 put_user(0, tidptr);
 523                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 524         }
 525 }
 526
 527 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 528 {
 529         struct mm_struct * mm, *oldmm;
 530         int retval;
 531
 532         tsk->min_flt = tsk->maj_flt = 0;
 533         tsk->cmin_flt = tsk->cmaj_flt = 0;
 534         tsk->nvcsw = tsk->nivcsw = tsk->cnvcsw = tsk->cnivcsw = 0;
 535
 536         tsk->mm = NULL;
 537         tsk->active_mm = NULL;
 538
 539         /*
 540          * Are we cloning a kernel thread?
 541          *
 542          * We need to steal a active VM for that..
 543          */
 544         oldmm = current->mm;
 545         if (!oldmm)
 546                 return 0;
 547
 548         if (clone_flags & CLONE_VM) {
 549                 atomic_inc(&oldmm->mm_users);
 550                 mm = oldmm;
 551                 /*
 552                  * There are cases where the PTL is held to ensure no
 553                  * new threads start up in user mode using an mm, which
 554                  * allows optimizing out ipis; the tlb_gather_mmu code
 555                  * is an example.
 556                  */
 557                 spin_unlock_wait(&oldmm->page_table_lock);
 558                 goto good_mm;
 559         }
 560
 561         retval = -ENOMEM;
 562         mm = allocate_mm();
 563         if (!mm)
 564                 goto fail_nomem;
 565
 566         /* Copy the current MM stuff.. */
 567         memcpy(mm, oldmm, sizeof(*mm));
 568         if (!mm_init(mm))
 569                 goto fail_nomem;
 570
 571         if (init_new_context(tsk,mm))
 572                 goto fail_nocontext;
 573
 574         retval = dup_mmap(mm, oldmm);
 575         if (retval)
 576                 goto free_pt;
 577
 578 good_mm:
 579         tsk->mm = mm;
 580         tsk->active_mm = mm;
 581         return 0;
 582
 583 free_pt:
 584         mmput(mm);
 585 fail_nomem:
 586         return retval;
 587
 588 fail_nocontext:
 589         /*
 590          * If init_new_context() failed, we cannot use mmput() to free the mm
 591          * because it calls destroy_context()
 592          */
 593         mm_free_pgd(mm);
 594         free_mm(mm);
 595         return retval;
 596 }
 597
 598 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 599 {
 600         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 601         /* We don't need to lock fs - think why ;-) */
 602         if (fs) {
 603                 atomic_set(&fs->count, 1);
 604                 fs->lock = RW_LOCK_UNLOCKED;
 605                 fs->umask = old->umask;
 606                 read_lock(&old->lock);
 607                 fs->rootmnt = mntget(old->rootmnt);
 608                 fs->root = dget(old->root);
 609                 fs->pwdmnt = mntget(old->pwdmnt);
 610                 fs->pwd = dget(old->pwd);
 611                 if (old->altroot) {
 612                         fs->altrootmnt = mntget(old->altrootmnt);
 613                         fs->altroot = dget(old->altroot);
 614                 } else {
 615                         fs->altrootmnt = NULL;
 616                         fs->altroot = NULL;
 617                 }
 618                 read_unlock(&old->lock);
 619         }
 620         return fs;
 621 }
 622
 623 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 624 {
 625         return __copy_fs_struct(old);
 626 }
 627
 628 EXPORT_SYMBOL_GPL(copy_fs_struct);
 629
 630 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 631 {
 632         if (clone_flags & CLONE_FS) {
 633                 atomic_inc(&current->fs->count);
 634                 return 0;
 635         }
 636         tsk->fs = __copy_fs_struct(current->fs);
 637         if (!tsk->fs)
 638                 return -ENOMEM;
 639         return 0;
 640 }
 641
 642 static int count_open_files(struct files_struct *files, int size)
 643 {
 644         int i;
 645
 646         /* Find the last open fd */
 647         for (i = size/(8*sizeof(long)); i > 0; ) {
 648                 if (files->open_fds->fds_bits[--i])
 649                         break;
 650         }
 651         i = (i+1) * 8 * sizeof(long);
 652         return i;
 653 }
 654
 655 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 656 {
 657         struct files_struct *oldf, *newf;
 658         struct file **old_fds, **new_fds;
 659         int open_files, nfds, size, i, error = 0;
 660
 661         /*
 662          * A background process may not have any files ...
 663          */
 664         oldf = current->files;
 665         if (!oldf)
 666                 goto out;
 667
 668         if (clone_flags & CLONE_FILES) {
 669                 atomic_inc(&oldf->count);
 670                 goto out;
 671         }
 672
 673         /*
 674          * Note: we may be using current for both targets (See exec.c)
 675          * This works because we cache current->files (old) as oldf. Don't
 676          * break this.
 677          */
 678         tsk->files = NULL;
 679         error = -ENOMEM;
 680         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
 681         if (!newf)
 682                 goto out;
 683
 684         atomic_set(&newf->count, 1);
 685
 686         newf->file_lock     = SPIN_LOCK_UNLOCKED;
 687         newf->next_fd       = 0;
 688         newf->max_fds       = NR_OPEN_DEFAULT;
 689         newf->max_fdset     = __FD_SETSIZE;
 690         newf->close_on_exec = &newf->close_on_exec_init;
 691         newf->open_fds      = &newf->open_fds_init;
 692         newf->fd            = &newf->fd_array[0];
 693
 694         /* We don't yet have the oldf readlock, but even if the old
 695            fdset gets grown now, we'll only copy up to "size" fds */
 696         size = oldf->max_fdset;
 697         if (size > __FD_SETSIZE) {
 698                 newf->max_fdset = 0;
 699                 spin_lock(&newf->file_lock);
 700                 error = expand_fdset(newf, size-1);
 701                 spin_unlock(&newf->file_lock);
 702                 if (error)
 703                         goto out_release;
 704         }
 705         spin_lock(&oldf->file_lock);
 706
 707         open_files = count_open_files(oldf, size);
 708
 709         /*
 710          * Check whether we need to allocate a larger fd array.
 711          * Note: we're not a clone task, so the open count won't
 712          * change.
 713          */
 714         nfds = NR_OPEN_DEFAULT;
 715         if (open_files > nfds) {
 716                 spin_unlock(&oldf->file_lock);
 717                 newf->max_fds = 0;
 718                 spin_lock(&newf->file_lock);
 719                 error = expand_fd_array(newf, open_files-1);
 720                 spin_unlock(&newf->file_lock);
 721                 if (error)
 722                         goto out_release;
 723                 nfds = newf->max_fds;
 724                 spin_lock(&oldf->file_lock);
 725         }
 726
 727         old_fds = oldf->fd;
 728         new_fds = newf->fd;
 729
 730         memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
 731         memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
 732
 733         for (i = open_files; i != 0; i--) {
 734                 struct file *f = *old_fds++;
 735                 if (f)
 736                         get_file(f);
 737                 *new_fds++ = f;
 738         }
 739         spin_unlock(&oldf->file_lock);
 740
 741         /* compute the remainder to be cleared */
 742         size = (newf->max_fds - open_files) * sizeof(struct file *);
 743
 744         /* This is long word aligned thus could use a optimized version */
 745         memset(new_fds, 0, size);
 746
 747         if (newf->max_fdset > open_files) {
 748                 int left = (newf->max_fdset-open_files)/8;
 749                 int start = open_files / (8 * sizeof(unsigned long));
 750
 751                 memset(&newf->open_fds->fds_bits[start], 0, left);
 752                 memset(&newf->close_on_exec->fds_bits[start], 0, left);
 753         }
 754
 755         tsk->files = newf;
 756         error = 0;
 757 out:
 758         return error;
 759
 760 out_release:
 761         free_fdset (newf->close_on_exec, newf->max_fdset);
 762         free_fdset (newf->open_fds, newf->max_fdset);
 763         kmem_cache_free(files_cachep, newf);
 764         goto out;
 765 }
 766
 767 /*
 768  *      Helper to unshare the files of the current task.
 769  *      We don't want to expose copy_files internals to
 770  *      the exec layer of the kernel.
 771  */
 772
 773 int unshare_files(void)
 774 {
 775         struct files_struct *files  = current->files;
 776         int rc;
 777
 778         if(!files)
 779                 BUG();
 780
 781         /* This can race but the race causes us to copy when we don't
 782            need to and drop the copy */
 783         if(atomic_read(&files->count) == 1)
 784         {
 785                 atomic_inc(&files->count);
 786                 return 0;
 787         }
 788         rc = copy_files(0, current);
 789         if(rc)
 790                 current->files = files;
 791         return rc;
 792 }
 793
 794 EXPORT_SYMBOL(unshare_files);
 795
 796 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 797 {
 798         struct sighand_struct *sig;
 799
 800         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 801                 atomic_inc(&current->sighand->count);
 802                 return 0;
 803         }
 804         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 805         tsk->sighand = sig;
 806         if (!sig)
 807                 return -ENOMEM;
 808         spin_lock_init(&sig->siglock);
 809         atomic_set(&sig->count, 1);
 810         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 811         return 0;
 812 }
 813
 814 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 815 {
 816         struct signal_struct *sig;
 817
 818         if (clone_flags & CLONE_THREAD) {
 819                 atomic_inc(&current->signal->count);
 820                 return 0;
 821         }
 822         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 823         tsk->signal = sig;
 824         if (!sig)
 825                 return -ENOMEM;
 826         atomic_set(&sig->count, 1);
 827         sig->group_exit = 0;
 828         sig->group_exit_code = 0;
 829         sig->group_exit_task = NULL;
 830         sig->group_stop_count = 0;
 831         sig->curr_target = NULL;
 832         init_sigpending(&sig->shared_pending);
 833         INIT_LIST_HEAD(&sig->posix_timers);
 834
 835         sig->tty = current->signal->tty;
 836         sig->pgrp = process_group(current);
 837         sig->session = current->signal->session;
 838         sig->leader = 0;        /* session leadership doesn't inherit */
 839         sig->tty_old_pgrp = 0;
 840
 841         return 0;
 842 }
 843
 844 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 845 {
 846         unsigned long new_flags = p->flags;
 847
 848         new_flags &= ~PF_SUPERPRIV;
 849         new_flags |= PF_FORKNOEXEC;
 850         if (!(clone_flags & CLONE_PTRACE))
 851                 p->ptrace = 0;
 852         p->flags = new_flags;
 853 }
 854
 855 asmlinkage long sys_set_tid_address(int __user *tidptr)
 856 {
 857         current->clear_child_tid = tidptr;
 858
 859         return current->pid;
 860 }
 861
 862 /*
 863  * This creates a new process as a copy of the old one,
 864  * but does not actually start it yet.
 865  *
 866  * It copies the registers, and all the appropriate
 867  * parts of the process environment (as per the clone
 868  * flags). The actual kick-off is left to the caller.
 869  */
 870 struct task_struct *copy_process(unsigned long clone_flags,
 871                                  unsigned long stack_start,
 872                                  struct pt_regs *regs,
 873                                  unsigned long stack_size,
 874                                  int __user *parent_tidptr,
 875                                  int __user *child_tidptr)
 876 {
 877         int retval;
 878         struct task_struct *p = NULL;
 879
 880         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 881                 return ERR_PTR(-EINVAL);
 882
 883         /*
 884          * Thread groups must share signals as well, and detached threads
 885          * can only be started up within the thread group.
 886          */
 887         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 888                 return ERR_PTR(-EINVAL);
 889
 890         /*
 891          * Shared signal handlers imply shared VM. By way of the above,
 892          * thread groups also imply shared VM. Blocking this case allows
 893          * for various simplifications in other code.
 894          */
 895         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 896                 return ERR_PTR(-EINVAL);
 897
 898         retval = security_task_create(clone_flags);
 899         if (retval)
 900                 goto fork_out;
 901
 902         retval = -ENOMEM;
 903         p = dup_task_struct(current);
 904         if (!p)
 905                 goto fork_out;
 906
 907         retval = -EAGAIN;
 908         if (atomic_read(&p->user->processes) >=
 909                         p->rlim[RLIMIT_NPROC].rlim_cur) {
 910                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 911                                 p->user != &root_user)
 912                         goto bad_fork_free;
 913         }
 914
 915         atomic_inc(&p->user->__count);
 916         atomic_inc(&p->user->processes);
 917         get_group_info(p->group_info);
 918
 919         /*
 920          * If multiple threads are within copy_process(), then this check
 921          * triggers too late. This doesn't hurt, the check is only there
 922          * to stop root fork bombs.
 923          */
 924         if (nr_threads >= max_threads)
 925                 goto bad_fork_cleanup_count;
 926
 927         if (!try_module_get(p->thread_info->exec_domain->module))
 928                 goto bad_fork_cleanup_count;
 929
 930         if (p->binfmt && !try_module_get(p->binfmt->module))
 931                 goto bad_fork_cleanup_put_domain;
 932
 933         init_delays(p);
 934         p->did_exec = 0;
 935         copy_flags(clone_flags, p);
 936         if (clone_flags & CLONE_IDLETASK)
 937                 p->pid = 0;
 938         else {
 939                 p->pid = alloc_pidmap();
 940                 if (p->pid == -1)
 941                         goto bad_fork_cleanup;
 942         }
 943         retval = -EFAULT;
 944         if (clone_flags & CLONE_PARENT_SETTID)
 945                 if (put_user(p->pid, parent_tidptr))
 946                         goto bad_fork_cleanup;
 947
 948         p->proc_dentry = NULL;
 949
 950         INIT_LIST_HEAD(&p->children);
 951         INIT_LIST_HEAD(&p->sibling);
 952         init_waitqueue_head(&p->wait_chldexit);
 953         p->vfork_done = NULL;
 954         spin_lock_init(&p->alloc_lock);
 955         spin_lock_init(&p->proc_lock);
 956
 957         clear_tsk_thread_flag(p, TIF_SIGPENDING);
 958         init_sigpending(&p->pending);
 959
 960         p->it_real_value = p->it_virt_value = p->it_prof_value = 0;
 961         p->it_real_incr = p->it_virt_incr = p->it_prof_incr = 0;
 962         init_timer(&p->real_timer);
 963         p->real_timer.data = (unsigned long) p;
 964
 965         p->utime = p->stime = 0;
 966         p->cutime = p->cstime = 0;
 967         p->lock_depth = -1;             /* -1 = no lock */
 968         p->start_time = get_jiffies_64();
 969         p->security = NULL;
 970         p->io_context = NULL;
 971         p->audit_context = NULL;
 972 #ifdef CONFIG_NUMA
 973         p->mempolicy = mpol_copy(p->mempolicy);
 974         if (IS_ERR(p->mempolicy)) {
 975                 retval = PTR_ERR(p->mempolicy);
 976                 p->mempolicy = NULL;
 977                 goto bad_fork_cleanup;
 978         }
 979 #endif
 980
 981         if ((retval = security_task_alloc(p)))
 982                 goto bad_fork_cleanup_policy;
 983         if ((retval = audit_alloc(p)))
 984                 goto bad_fork_cleanup_security;
 985         /* copy all the process information */
 986         if ((retval = copy_semundo(clone_flags, p)))
 987                 goto bad_fork_cleanup_audit;
 988         if ((retval = copy_files(clone_flags, p)))
 989                 goto bad_fork_cleanup_semundo;
 990         if ((retval = copy_fs(clone_flags, p)))
 991                 goto bad_fork_cleanup_files;
 992         if ((retval = copy_sighand(clone_flags, p)))
 993                 goto bad_fork_cleanup_fs;
 994         if ((retval = copy_signal(clone_flags, p)))
 995                 goto bad_fork_cleanup_sighand;
 996         if ((retval = copy_mm(clone_flags, p)))
 997                 goto bad_fork_cleanup_signal;
 998         if ((retval = copy_namespace(clone_flags, p)))
 999                 goto bad_fork_cleanup_mm;
1000         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1001         if (retval)
1002                 goto bad_fork_cleanup_namespace;
1003
1004         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1005         /*
1006          * Clear TID on mm_release()?
1007          */
1008         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1009
1010         /*
1011          * Syscall tracing should be turned off in the child regardless
1012          * of CLONE_PTRACE.
1013          */
1014         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1015
1016         /* Our parent execution domain becomes current domain
1017            These must match for thread signalling to apply */
1018
1019         p->parent_exec_id = p->self_exec_id;
1020
1021         /* ok, now we should be set up.. */
1022         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1023         p->pdeath_signal = 0;
1024
1025         /* Perform scheduler related setup */
1026         sched_fork(p);
1027
1028         /*
1029          * Ok, make it visible to the rest of the system.
1030          * We dont wake it up yet.
1031          */
1032         p->tgid = p->pid;
1033         p->group_leader = p;
1034         INIT_LIST_HEAD(&p->ptrace_children);
1035         INIT_LIST_HEAD(&p->ptrace_list);
1036
1037         /* Need tasklist lock for parent etc handling! */
1038         write_lock_irq(&tasklist_lock);
1039         /*
1040          * Check for pending SIGKILL! The new thread should not be allowed
1041          * to slip out of an OOM kill. (or normal SIGKILL.)
1042          */
1043         if (sigismember(&current->pending.signal, SIGKILL)) {
1044                 write_unlock_irq(&tasklist_lock);
1045                 retval = -EINTR;
1046                 goto bad_fork_cleanup_namespace;
1047         }
1048
1049         /* CLONE_PARENT re-uses the old parent */
1050         if (clone_flags & CLONE_PARENT)
1051                 p->real_parent = current->real_parent;
1052         else
1053                 p->real_parent = current;
1054         p->parent = p->real_parent;
1055
1056         if (clone_flags & CLONE_THREAD) {
1057                 spin_lock(&current->sighand->siglock);
1058                 /*
1059                  * Important: if an exit-all has been started then
1060                  * do not create this new thread - the whole thread
1061                  * group is supposed to exit anyway.
1062                  */
1063                 if (current->signal->group_exit) {
1064                         spin_unlock(&current->sighand->siglock);
1065                         write_unlock_irq(&tasklist_lock);
1066                         retval = -EAGAIN;
1067                         goto bad_fork_cleanup_namespace;
1068                 }
1069                 p->tgid = current->tgid;
1070                 p->group_leader = current->group_leader;
1071
1072                 if (current->signal->group_stop_count > 0) {
1073                         /*
1074                          * There is an all-stop in progress for the group.
1075                          * We ourselves will stop as soon as we check signals.
1076                          * Make the new thread part of that group stop too.
1077                          */
1078                         current->signal->group_stop_count++;
1079                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1080                 }
1081
1082                 spin_unlock(&current->sighand->siglock);
1083         }
1084
1085         SET_LINKS(p);
1086         if (p->ptrace & PT_PTRACED)
1087                 __ptrace_link(p, current->parent);
1088
1089         attach_pid(p, PIDTYPE_PID, p->pid);
1090         if (thread_group_leader(p)) {
1091                 attach_pid(p, PIDTYPE_TGID, p->tgid);
1092                 attach_pid(p, PIDTYPE_PGID, process_group(p));
1093                 attach_pid(p, PIDTYPE_SID, p->signal->session);
1094                 if (p->pid)
1095                         __get_cpu_var(process_counts)++;
1096         } else
1097                 link_pid(p, p->pids + PIDTYPE_TGID, &p->group_leader->pids[PIDTYPE_TGID].pid);
1098
1099         nr_threads++;
1100         write_unlock_irq(&tasklist_lock);
1101         retval = 0;
1102
1103 fork_out:
1104         if (retval)
1105                 return ERR_PTR(retval);
1106         return p;
1107
1108 bad_fork_cleanup_namespace:
1109         exit_namespace(p);
1110 bad_fork_cleanup_mm:
1111         exit_mm(p);
1112         if (p->active_mm)
1113                 mmdrop(p->active_mm);
1114 bad_fork_cleanup_signal:
1115         exit_signal(p);
1116 bad_fork_cleanup_sighand:
1117         exit_sighand(p);
1118 bad_fork_cleanup_fs:
1119         exit_fs(p); /* blocking */
1120 bad_fork_cleanup_files:
1121         exit_files(p); /* blocking */
1122 bad_fork_cleanup_semundo:
1123         exit_sem(p);
1124 bad_fork_cleanup_audit:
1125         audit_free(p);
1126 bad_fork_cleanup_security:
1127         security_task_free(p);
1128 bad_fork_cleanup_policy:
1129 #ifdef CONFIG_NUMA
1130         mpol_free(p->mempolicy);
1131 #endif
1132 bad_fork_cleanup:
1133         if (p->pid > 0)
1134                 free_pidmap(p->pid);
1135         if (p->binfmt)
1136                 module_put(p->binfmt->module);
1137 bad_fork_cleanup_put_domain:
1138         module_put(p->thread_info->exec_domain->module);
1139 bad_fork_cleanup_count:
1140         put_group_info(p->group_info);
1141         atomic_dec(&p->user->processes);
1142         free_uid(p->user);
1143 bad_fork_free:
1144         free_task(p);
1145         goto fork_out;
1146 }
1147
1148 static inline int fork_traceflag (unsigned clone_flags)
1149 {
1150         if (clone_flags & (CLONE_UNTRACED | CLONE_IDLETASK))
1151                 return 0;
1152         else if (clone_flags & CLONE_VFORK) {
1153                 if (current->ptrace & PT_TRACE_VFORK)
1154                         return PTRACE_EVENT_VFORK;
1155         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1156                 if (current->ptrace & PT_TRACE_CLONE)
1157                         return PTRACE_EVENT_CLONE;
1158         } else if (current->ptrace & PT_TRACE_FORK)
1159                 return PTRACE_EVENT_FORK;
1160
1161         return 0;
1162 }
1163
1164 /*
1165  *  Ok, this is the main fork-routine.
1166  *
1167  * It copies the process, and if successful kick-starts
1168  * it and waits for it to finish using the VM if required.
1169  */
1170 long do_fork(unsigned long clone_flags,
1171               unsigned long stack_start,
1172               struct pt_regs *regs,
1173               unsigned long stack_size,
1174               int __user *parent_tidptr,
1175               int __user *child_tidptr)
1176 {
1177         struct task_struct *p;
1178         int trace = 0;
1179         long pid;
1180
1181         if (unlikely(current->ptrace)) {
1182                 trace = fork_traceflag (clone_flags);
1183                 if (trace)
1184                         clone_flags |= CLONE_PTRACE;
1185         }
1186
1187 #ifdef CONFIG_CKRM_TYPE_TASKCLASS
1188         if (numtasks_get_ref(current->taskclass, 0) == 0) {
1189                 return -ENOMEM;
1190         }
1191 #endif
1192
1193         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr);
1194         /*
1195          * Do this prior waking up the new thread - the thread pointer
1196          * might get invalid after that point, if the thread exits quickly.
1197          */
1198         pid = IS_ERR(p) ? PTR_ERR(p) : p->pid;
1199
1200         if (!IS_ERR(p)) {
1201                 struct completion vfork;
1202
1203                 ckrm_cb_fork(p);
1204
1205                 if (clone_flags & CLONE_VFORK) {
1206                         p->vfork_done = &vfork;
1207                         init_completion(&vfork);
1208                 }
1209
1210                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1211                         /*
1212                          * We'll start up with an immediate SIGSTOP.
1213                          */
1214                         sigaddset(&p->pending.signal, SIGSTOP);
1215                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1216                 }
1217
1218                 if (!(clone_flags & CLONE_STOPPED)) {
1219                         /*
1220                          * Do the wakeup last. On SMP we treat fork() and
1221                          * CLONE_VM separately, because fork() has already
1222                          * created cache footprint on this CPU (due to
1223                          * copying the pagetables), hence migration would
1224                          * probably be costy. Threads on the other hand
1225                          * have less traction to the current CPU, and if
1226                          * there's an imbalance then the scheduler can
1227                          * migrate this fresh thread now, before it
1228                          * accumulates a larger cache footprint:
1229                          */
1230                         if (clone_flags & CLONE_VM)
1231                                 wake_up_forked_thread(p);
1232                         else
1233                                 wake_up_forked_process(p);
1234                 } else {
1235                         int cpu = get_cpu();
1236
1237                         p->state = TASK_STOPPED;
1238                         if (cpu_is_offline(task_cpu(p)))
1239                                 set_task_cpu(p, cpu);
1240
1241                         put_cpu();
1242                 }
1243                 ++total_forks;
1244
1245                 if (unlikely (trace)) {
1246                         current->ptrace_message = pid;
1247                         ptrace_notify ((trace << 8) | SIGTRAP);
1248                 }
1249
1250                 if (clone_flags & CLONE_VFORK) {
1251                         wait_for_completion(&vfork);
1252                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1253                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1254                 } else
1255                         /*
1256                          * Let the child process run first, to avoid most of the
1257                          * COW overhead when the child exec()s afterwards.
1258                          */
1259                         set_need_resched();
1260         } else {
1261 #ifdef CONFIG_CKRM_TYPE_TASKCLASS
1262                 numtasks_put_ref(current->taskclass);
1263 #endif
1264         }
1265         return pid;
1266 }
1267
1268 /* SLAB cache for signal_struct structures (tsk->signal) */
1269 kmem_cache_t *signal_cachep;
1270
1271 /* SLAB cache for sighand_struct structures (tsk->sighand) */
1272 kmem_cache_t *sighand_cachep;
1273
1274 /* SLAB cache for files_struct structures (tsk->files) */
1275 kmem_cache_t *files_cachep;
1276
1277 /* SLAB cache for fs_struct structures (tsk->fs) */
1278 kmem_cache_t *fs_cachep;
1279
1280 /* SLAB cache for vm_area_struct structures */
1281 kmem_cache_t *vm_area_cachep;
1282
1283 /* SLAB cache for mm_struct structures (tsk->mm) */
1284 kmem_cache_t *mm_cachep;
1285
1286 void __init proc_caches_init(void)
1287 {
1288         sighand_cachep = kmem_cache_create("sighand_cache",
1289                         sizeof(struct sighand_struct), 0,
1290                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1291         signal_cachep = kmem_cache_create("signal_cache",
1292                         sizeof(struct signal_struct), 0,
1293                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1294         files_cachep = kmem_cache_create("files_cache",
1295                         sizeof(struct files_struct), 0,
1296                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1297         fs_cachep = kmem_cache_create("fs_cache",
1298                         sizeof(struct fs_struct), 0,
1299                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1300         vm_area_cachep = kmem_cache_create("vm_area_struct",
1301                         sizeof(struct vm_area_struct), 0,
1302                         SLAB_PANIC, NULL, NULL);
1303         mm_cachep = kmem_cache_create("mm_struct",
1304                         sizeof(struct mm_struct), 0,
1305                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1306 }