kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/config.h>
  15 #include <linux/slab.h>
  16 #include <linux/init.h>
  17 #include <linux/unistd.h>
  18 #include <linux/smp_lock.h>
  19 #include <linux/module.h>
  20 #include <linux/vmalloc.h>
  21 #include <linux/completion.h>
  22 #include <linux/namespace.h>
  23 #include <linux/personality.h>
  24 #include <linux/sem.h>
  25 #include <linux/file.h>
  26 #include <linux/binfmts.h>
  27 #include <linux/mman.h>
  28 #include <linux/fs.h>
  29 #include <linux/security.h>
  30 #include <linux/syscalls.h>
  31 #include <linux/jiffies.h>
  32 #include <linux/futex.h>
  33 #include <linux/ptrace.h>
  34 #include <linux/mount.h>
  35 #include <linux/audit.h>
  36 #include <linux/vinline.h>
  37 #include <linux/ninline.h>
  38
  39 #include <asm/pgtable.h>
  40 #include <asm/pgalloc.h>
  41 #include <asm/uaccess.h>
  42 #include <asm/mmu_context.h>
  43 #include <asm/cacheflush.h>
  44 #include <asm/tlbflush.h>
  45
  46 /* The idle threads do not count..
  47  * Protected by write_lock_irq(&tasklist_lock)
  48  */
  49 int nr_threads;
  50
  51 int max_threads;
  52 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  53
  54 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  55
  56 rwlock_t tasklist_lock __cacheline_aligned = RW_LOCK_UNLOCKED;  /* outer */
  57
  58 EXPORT_SYMBOL(tasklist_lock);
  59
  60 int nr_processes(void)
  61 {
  62         int cpu;
  63         int total = 0;
  64
  65         for_each_cpu(cpu)
  66                 total += per_cpu(process_counts, cpu);
  67
  68         return total;
  69 }
  70
  71 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  72 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  73 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  74 static kmem_cache_t *task_struct_cachep;
  75 #endif
  76
  77 static void free_task(struct task_struct *tsk)
  78 {
  79         free_thread_info(tsk->thread_info);
  80         vxdprintk("freeing up task %p\n", tsk);
  81         clr_vx_info(&tsk->vx_info);
  82         clr_nx_info(&tsk->nx_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)
 205 {
 206         int ret = default_wake_function(wait, mode, sync);
 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",
 224                                   sizeof(struct task_struct),ARCH_MIN_TASKALIGN,
 225                                   0, NULL, NULL);
 226         if (!task_struct_cachep)
 227                 panic("fork_init(): cannot create task_struct SLAB cache");
 228 #endif
 229
 230         /*
 231          * The default maximum number of threads is set to a safe
 232          * value: the thread structures can take up at most half
 233          * of memory.
 234          */
 235         max_threads = mempages / (THREAD_SIZE/PAGE_SIZE) / 8;
 236         /*
 237          * we need to allow at least 20 threads to boot a system
 238          */
 239         if(max_threads < 20)
 240                 max_threads = 20;
 241
 242         init_task.rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 243         init_task.rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 244 }
 245
 246 static struct task_struct *dup_task_struct(struct task_struct *orig)
 247 {
 248         struct task_struct *tsk;
 249         struct thread_info *ti;
 250
 251         prepare_to_copy(orig);
 252
 253         tsk = alloc_task_struct();
 254         if (!tsk)
 255                 return NULL;
 256
 257         ti = alloc_thread_info(tsk);
 258         if (!ti) {
 259                 free_task_struct(tsk);
 260                 return NULL;
 261         }
 262
 263         *ti = *orig->thread_info;
 264         *tsk = *orig;
 265         tsk->thread_info = ti;
 266         ti->task = tsk;
 267
 268         /* One for us, one for whoever does the "release_task()" (usually parent) */
 269         atomic_set(&tsk->usage,2);
 270         return tsk;
 271 }
 272
 273 #ifdef CONFIG_MMU
 274 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
 275 {
 276         struct vm_area_struct * mpnt, *tmp, **pprev;
 277         struct rb_node **rb_link, *rb_parent;
 278         int retval;
 279         unsigned long charge = 0;
 280
 281         down_write(&oldmm->mmap_sem);
 282         flush_cache_mm(current->mm);
 283         mm->locked_vm = 0;
 284         mm->mmap = NULL;
 285         mm->mmap_cache = NULL;
 286         mm->free_area_cache = TASK_UNMAPPED_BASE;
 287         mm->map_count = 0;
 288         mm->rss = 0;
 289         cpus_clear(mm->cpu_vm_mask);
 290         mm->mm_rb = RB_ROOT;
 291         rb_link = &mm->mm_rb.rb_node;
 292         rb_parent = NULL;
 293         pprev = &mm->mmap;
 294
 295         /*
 296          * Add it to the mmlist after the parent.
 297          * Doing it this way means that we can order the list,
 298          * and fork() won't mess up the ordering significantly.
 299          * Add it first so that swapoff can see any swap entries.
 300          */
 301         spin_lock(&mmlist_lock);
 302         list_add(&mm->mmlist, &current->mm->mmlist);
 303         mmlist_nr++;
 304         spin_unlock(&mmlist_lock);
 305
 306         for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
 307                 struct file *file;
 308
 309                 if(mpnt->vm_flags & VM_DONTCOPY)
 310                         continue;
 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                 tmp->vm_flags &= ~VM_LOCKED;
 322                 tmp->vm_mm = mm;
 323                 tmp->vm_next = NULL;
 324                 file = tmp->vm_file;
 325                 INIT_LIST_HEAD(&tmp->shared);
 326                 if (file) {
 327                         struct inode *inode = file->f_dentry->d_inode;
 328                         get_file(file);
 329                         if (tmp->vm_flags & VM_DENYWRITE)
 330                                 atomic_dec(&inode->i_writecount);
 331
 332                         /* insert tmp into the share list, just after mpnt */
 333                         down(&file->f_mapping->i_shared_sem);
 334                         list_add(&tmp->shared, &mpnt->shared);
 335                         up(&file->f_mapping->i_shared_sem);
 336                 }
 337
 338                 /*
 339                  * Link in the new vma and copy the page table entries:
 340                  * link in first so that swapoff can see swap entries,
 341                  * and try_to_unmap_one's find_vma find the new vma.
 342                  */
 343                 spin_lock(&mm->page_table_lock);
 344                 *pprev = tmp;
 345                 pprev = &tmp->vm_next;
 346
 347                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 348                 rb_link = &tmp->vm_rb.rb_right;
 349                 rb_parent = &tmp->vm_rb;
 350
 351                 mm->map_count++;
 352                 retval = copy_page_range(mm, current->mm, tmp);
 353                 spin_unlock(&mm->page_table_lock);
 354
 355                 if (tmp->vm_ops && tmp->vm_ops->open)
 356                         tmp->vm_ops->open(tmp);
 357
 358                 if (retval)
 359                         goto fail;
 360         }
 361         retval = 0;
 362
 363 out:
 364         flush_tlb_mm(current->mm);
 365         up_write(&oldmm->mmap_sem);
 366         return retval;
 367 fail_nomem:
 368         retval = -ENOMEM;
 369 fail:
 370         vm_unacct_memory(charge);
 371         goto out;
 372 }
 373 static inline int mm_alloc_pgd(struct mm_struct * mm)
 374 {
 375         mm->pgd = pgd_alloc(mm);
 376         if (unlikely(!mm->pgd))
 377                 return -ENOMEM;
 378         return 0;
 379 }
 380
 381 static inline void mm_free_pgd(struct mm_struct * mm)
 382 {
 383         pgd_free(mm->pgd);
 384 }
 385 #else
 386 #define dup_mmap(mm, oldmm)     (0)
 387 #define mm_alloc_pgd(mm)        (0)
 388 #define mm_free_pgd(mm)
 389 #endif /* CONFIG_MMU */
 390
 391 spinlock_t mmlist_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
 392 int mmlist_nr;
 393
 394 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
 395 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 396
 397 #include <linux/init_task.h>
 398
 399 static struct mm_struct * mm_init(struct mm_struct * mm)
 400 {
 401         atomic_set(&mm->mm_users, 1);
 402         atomic_set(&mm->mm_count, 1);
 403         init_rwsem(&mm->mmap_sem);
 404         mm->core_waiters = 0;
 405         mm->page_table_lock = SPIN_LOCK_UNLOCKED;
 406         mm->ioctx_list_lock = RW_LOCK_UNLOCKED;
 407         mm->ioctx_list = NULL;
 408         mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
 409         mm->free_area_cache = TASK_UNMAPPED_BASE;
 410
 411         if (likely(!mm_alloc_pgd(mm))) {
 412                 mm->def_flags = 0;
 413                 set_vx_info(&mm->mm_vx_info, current->vx_info);
 414                 return mm;
 415         }
 416         free_mm(mm);
 417         return NULL;
 418 }
 419
 420 /*
 421  * Allocate and initialize an mm_struct.
 422  */
 423 struct mm_struct * mm_alloc(void)
 424 {
 425         struct mm_struct * mm;
 426
 427         mm = allocate_mm();
 428         if (mm) {
 429                 memset(mm, 0, sizeof(*mm));
 430                 return mm_init(mm);
 431         }
 432         return NULL;
 433 }
 434
 435 /*
 436  * Called when the last reference to the mm
 437  * is dropped: either by a lazy thread or by
 438  * mmput. Free the page directory and the mm.
 439  */
 440 void fastcall __mmdrop(struct mm_struct *mm)
 441 {
 442         BUG_ON(mm == &init_mm);
 443         mm_free_pgd(mm);
 444         destroy_context(mm);
 445         clr_vx_info(&mm->mm_vx_info);
 446         free_mm(mm);
 447 }
 448
 449 /*
 450  * Decrement the use count and release all resources for an mm.
 451  */
 452 void mmput(struct mm_struct *mm)
 453 {
 454         if (atomic_dec_and_lock(&mm->mm_users, &mmlist_lock)) {
 455                 list_del(&mm->mmlist);
 456                 mmlist_nr--;
 457                 spin_unlock(&mmlist_lock);
 458                 exit_aio(mm);
 459                 exit_mmap(mm);
 460                 mmdrop(mm);
 461         }
 462 }
 463
 464 /*
 465  * Checks if the use count of an mm is non-zero and if so
 466  * returns a reference to it after bumping up the use count.
 467  * If the use count is zero, it means this mm is going away,
 468  * so return NULL.
 469  */
 470 struct mm_struct *mmgrab(struct mm_struct *mm)
 471 {
 472         spin_lock(&mmlist_lock);
 473         if (!atomic_read(&mm->mm_users))
 474                 mm = NULL;
 475         else
 476                 atomic_inc(&mm->mm_users);
 477         spin_unlock(&mmlist_lock);
 478         return mm;
 479 }
 480
 481 /* Please note the differences between mmput and mm_release.
 482  * mmput is called whenever we stop holding onto a mm_struct,
 483  * error success whatever.
 484  *
 485  * mm_release is called after a mm_struct has been removed
 486  * from the current process.
 487  *
 488  * This difference is important for error handling, when we
 489  * only half set up a mm_struct for a new process and need to restore
 490  * the old one.  Because we mmput the new mm_struct before
 491  * restoring the old one. . .
 492  * Eric Biederman 10 January 1998
 493  */
 494 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 495 {
 496         struct completion *vfork_done = tsk->vfork_done;
 497
 498         /* Get rid of any cached register state */
 499         deactivate_mm(tsk, mm);
 500
 501         /* notify parent sleeping on vfork() */
 502         if (vfork_done) {
 503                 tsk->vfork_done = NULL;
 504                 complete(vfork_done);
 505         }
 506         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
 507                 u32 __user * tidptr = tsk->clear_child_tid;
 508                 tsk->clear_child_tid = NULL;
 509
 510                 /*
 511                  * We don't check the error code - if userspace has
 512                  * not set up a proper pointer then tough luck.
 513                  */
 514                 put_user(0, tidptr);
 515                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL);
 516         }
 517 }
 518
 519 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 520 {
 521         struct mm_struct * mm, *oldmm;
 522         int retval;
 523
 524         tsk->min_flt = tsk->maj_flt = 0;
 525         tsk->cmin_flt = tsk->cmaj_flt = 0;
 526         tsk->nvcsw = tsk->nivcsw = tsk->cnvcsw = tsk->cnivcsw = 0;
 527
 528         tsk->mm = NULL;
 529         tsk->active_mm = NULL;
 530
 531         /*
 532          * Are we cloning a kernel thread?
 533          *
 534          * We need to steal a active VM for that..
 535          */
 536         oldmm = current->mm;
 537         if (!oldmm)
 538                 return 0;
 539
 540         if (clone_flags & CLONE_VM) {
 541                 atomic_inc(&oldmm->mm_users);
 542                 mm = oldmm;
 543                 /*
 544                  * There are cases where the PTL is held to ensure no
 545                  * new threads start up in user mode using an mm, which
 546                  * allows optimizing out ipis; the tlb_gather_mmu code
 547                  * is an example.
 548                  */
 549                 spin_unlock_wait(&oldmm->page_table_lock);
 550                 goto good_mm;
 551         }
 552
 553         retval = -ENOMEM;
 554         mm = allocate_mm();
 555         if (!mm)
 556                 goto fail_nomem;
 557
 558         /* Copy the current MM stuff.. */
 559         memcpy(mm, oldmm, sizeof(*mm));
 560         mm->mm_vx_info = NULL;
 561         if (!mm_init(mm))
 562                 goto fail_nomem;
 563
 564         if (init_new_context(tsk,mm))
 565                 goto fail_nocontext;
 566
 567         retval = dup_mmap(mm, oldmm);
 568         if (retval)
 569                 goto free_pt;
 570
 571 good_mm:
 572         tsk->mm = mm;
 573         tsk->active_mm = mm;
 574         return 0;
 575
 576 free_pt:
 577         mmput(mm);
 578 fail_nomem:
 579         return retval;
 580
 581 fail_nocontext:
 582         /*
 583          * If init_new_context() failed, we cannot use mmput() to free the mm
 584          * because it calls destroy_context()
 585          */
 586         mm_free_pgd(mm);
 587         free_mm(mm);
 588         return retval;
 589 }
 590
 591 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 592 {
 593         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 594         /* We don't need to lock fs - think why ;-) */
 595         if (fs) {
 596                 atomic_set(&fs->count, 1);
 597                 fs->lock = RW_LOCK_UNLOCKED;
 598                 fs->umask = old->umask;
 599                 read_lock(&old->lock);
 600                 fs->rootmnt = mntget(old->rootmnt);
 601                 fs->root = dget(old->root);
 602                 fs->pwdmnt = mntget(old->pwdmnt);
 603                 fs->pwd = dget(old->pwd);
 604                 if (old->altroot) {
 605                         fs->altrootmnt = mntget(old->altrootmnt);
 606                         fs->altroot = dget(old->altroot);
 607                 } else {
 608                         fs->altrootmnt = NULL;
 609                         fs->altroot = NULL;
 610                 }
 611                 read_unlock(&old->lock);
 612         }
 613         return fs;
 614 }
 615
 616 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 617 {
 618         return __copy_fs_struct(old);
 619 }
 620
 621 EXPORT_SYMBOL_GPL(copy_fs_struct);
 622
 623 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 624 {
 625         if (clone_flags & CLONE_FS) {
 626                 atomic_inc(&current->fs->count);
 627                 return 0;
 628         }
 629         tsk->fs = __copy_fs_struct(current->fs);
 630         if (!tsk->fs)
 631                 return -ENOMEM;
 632         return 0;
 633 }
 634
 635 static int count_open_files(struct files_struct *files, int size)
 636 {
 637         int i;
 638
 639         /* Find the last open fd */
 640         for (i = size/(8*sizeof(long)); i > 0; ) {
 641                 if (files->open_fds->fds_bits[--i])
 642                         break;
 643         }
 644         i = (i+1) * 8 * sizeof(long);
 645         return i;
 646 }
 647
 648 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 649 {
 650         struct files_struct *oldf, *newf;
 651         struct file **old_fds, **new_fds;
 652         int open_files, nfds, size, i, error = 0;
 653
 654         /*
 655          * A background process may not have any files ...
 656          */
 657         oldf = current->files;
 658         if (!oldf)
 659                 goto out;
 660
 661         if (clone_flags & CLONE_FILES) {
 662                 atomic_inc(&oldf->count);
 663                 goto out;
 664         }
 665
 666         /*
 667          * Note: we may be using current for both targets (See exec.c)
 668          * This works because we cache current->files (old) as oldf. Don't
 669          * break this.
 670          */
 671         tsk->files = NULL;
 672         error = -ENOMEM;
 673         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
 674         if (!newf)
 675                 goto out;
 676
 677         atomic_set(&newf->count, 1);
 678
 679         newf->file_lock     = SPIN_LOCK_UNLOCKED;
 680         newf->next_fd       = 0;
 681         newf->max_fds       = NR_OPEN_DEFAULT;
 682         newf->max_fdset     = __FD_SETSIZE;
 683         newf->close_on_exec = &newf->close_on_exec_init;
 684         newf->open_fds      = &newf->open_fds_init;
 685         newf->fd            = &newf->fd_array[0];
 686
 687         /* We don't yet have the oldf readlock, but even if the old
 688            fdset gets grown now, we'll only copy up to "size" fds */
 689         size = oldf->max_fdset;
 690         if (size > __FD_SETSIZE) {
 691                 newf->max_fdset = 0;
 692                 spin_lock(&newf->file_lock);
 693                 error = expand_fdset(newf, size-1);
 694                 spin_unlock(&newf->file_lock);
 695                 if (error)
 696                         goto out_release;
 697         }
 698         spin_lock(&oldf->file_lock);
 699
 700         open_files = count_open_files(oldf, size);
 701
 702         /*
 703          * Check whether we need to allocate a larger fd array.
 704          * Note: we're not a clone task, so the open count won't
 705          * change.
 706          */
 707         nfds = NR_OPEN_DEFAULT;
 708         if (open_files > nfds) {
 709                 spin_unlock(&oldf->file_lock);
 710                 newf->max_fds = 0;
 711                 spin_lock(&newf->file_lock);
 712                 error = expand_fd_array(newf, open_files-1);
 713                 spin_unlock(&newf->file_lock);
 714                 if (error)
 715                         goto out_release;
 716                 nfds = newf->max_fds;
 717                 spin_lock(&oldf->file_lock);
 718         }
 719
 720         old_fds = oldf->fd;
 721         new_fds = newf->fd;
 722
 723         memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
 724         memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
 725
 726         for (i = open_files; i != 0; i--) {
 727                 struct file *f = *old_fds++;
 728                 if (f)
 729                         get_file(f);
 730                 *new_fds++ = f;
 731         }
 732         spin_unlock(&oldf->file_lock);
 733
 734         /* compute the remainder to be cleared */
 735         size = (newf->max_fds - open_files) * sizeof(struct file *);
 736
 737         /* This is long word aligned thus could use a optimized version */
 738         memset(new_fds, 0, size);
 739
 740         if (newf->max_fdset > open_files) {
 741                 int left = (newf->max_fdset-open_files)/8;
 742                 int start = open_files / (8 * sizeof(unsigned long));
 743
 744                 memset(&newf->open_fds->fds_bits[start], 0, left);
 745                 memset(&newf->close_on_exec->fds_bits[start], 0, left);
 746         }
 747
 748         tsk->files = newf;
 749         error = 0;
 750 out:
 751         return error;
 752
 753 out_release:
 754         free_fdset (newf->close_on_exec, newf->max_fdset);
 755         free_fdset (newf->open_fds, newf->max_fdset);
 756         kmem_cache_free(files_cachep, newf);
 757         goto out;
 758 }
 759
 760 /*
 761  *      Helper to unshare the files of the current task.
 762  *      We don't want to expose copy_files internals to
 763  *      the exec layer of the kernel.
 764  */
 765
 766 int unshare_files(void)
 767 {
 768         struct files_struct *files  = current->files;
 769         int rc;
 770
 771         if(!files)
 772                 BUG();
 773
 774         /* This can race but the race causes us to copy when we don't
 775            need to and drop the copy */
 776         if(atomic_read(&files->count) == 1)
 777         {
 778                 atomic_inc(&files->count);
 779                 return 0;
 780         }
 781         rc = copy_files(0, current);
 782         if(rc)
 783                 current->files = files;
 784         return rc;
 785 }
 786
 787 EXPORT_SYMBOL(unshare_files);
 788
 789 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 790 {
 791         struct sighand_struct *sig;
 792
 793         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 794                 atomic_inc(&current->sighand->count);
 795                 return 0;
 796         }
 797         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 798         tsk->sighand = sig;
 799         if (!sig)
 800                 return -ENOMEM;
 801         spin_lock_init(&sig->siglock);
 802         atomic_set(&sig->count, 1);
 803         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 804         return 0;
 805 }
 806
 807 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 808 {
 809         struct signal_struct *sig;
 810
 811         if (clone_flags & CLONE_THREAD) {
 812                 atomic_inc(&current->signal->count);
 813                 return 0;
 814         }
 815         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 816         tsk->signal = sig;
 817         if (!sig)
 818                 return -ENOMEM;
 819         atomic_set(&sig->count, 1);
 820         sig->group_exit = 0;
 821         sig->group_exit_code = 0;
 822         sig->group_exit_task = NULL;
 823         sig->group_stop_count = 0;
 824         sig->curr_target = NULL;
 825         init_sigpending(&sig->shared_pending);
 826         INIT_LIST_HEAD(&sig->posix_timers);
 827
 828         sig->tty = current->signal->tty;
 829         sig->pgrp = process_group(current);
 830         sig->session = current->signal->session;
 831         sig->leader = 0;        /* session leadership doesn't inherit */
 832         sig->tty_old_pgrp = 0;
 833
 834         return 0;
 835 }
 836
 837 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 838 {
 839         unsigned long new_flags = p->flags;
 840
 841         new_flags &= ~PF_SUPERPRIV;
 842         new_flags |= PF_FORKNOEXEC;
 843         if (!(clone_flags & CLONE_PTRACE))
 844                 p->ptrace = 0;
 845         p->flags = new_flags;
 846 }
 847
 848 asmlinkage long sys_set_tid_address(int __user *tidptr)
 849 {
 850         current->clear_child_tid = tidptr;
 851
 852         return current->pid;
 853 }
 854
 855 /*
 856  * This creates a new process as a copy of the old one,
 857  * but does not actually start it yet.
 858  *
 859  * It copies the registers, and all the appropriate
 860  * parts of the process environment (as per the clone
 861  * flags). The actual kick-off is left to the caller.
 862  */
 863 struct task_struct *copy_process(unsigned long clone_flags,
 864                                  unsigned long stack_start,
 865                                  struct pt_regs *regs,
 866                                  unsigned long stack_size,
 867                                  int __user *parent_tidptr,
 868                                  int __user *child_tidptr)
 869 {
 870         int retval;
 871         struct task_struct *p = NULL;
 872         struct vx_info *vxi;
 873
 874         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 875                 return ERR_PTR(-EINVAL);
 876
 877         /*
 878          * Thread groups must share signals as well, and detached threads
 879          * can only be started up within the thread group.
 880          */
 881         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 882                 return ERR_PTR(-EINVAL);
 883
 884         /*
 885          * Shared signal handlers imply shared VM. By way of the above,
 886          * thread groups also imply shared VM. Blocking this case allows
 887          * for various simplifications in other code.
 888          */
 889         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 890                 return ERR_PTR(-EINVAL);
 891
 892         retval = security_task_create(clone_flags);
 893         if (retval)
 894                 goto fork_out;
 895
 896         retval = -ENOMEM;
 897
 898         p = dup_task_struct(current);
 899         if (!p)
 900                 goto fork_out;
 901
 902         p->vx_info = NULL;
 903         set_vx_info(&p->vx_info, current->vx_info);
 904         p->nx_info = NULL;
 905         set_nx_info(&p->nx_info, current->nx_info);
 906
 907         /* check vserver memory */
 908         if (p->mm && !(clone_flags & CLONE_VM)) {
 909                 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
 910                         vx_pages_add(p->mm->mm_vx_info, RLIMIT_AS, p->mm->total_vm);
 911                 else
 912                         goto bad_fork_free;
 913         }
 914         if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
 915                 if (!vx_rsspages_avail(p->mm, p->mm->rss))
 916                         goto bad_fork_free;
 917         }
 918
 919         retval = -EAGAIN;
 920         vxi = current->vx_info;
 921         if (vxi && (atomic_read(&vxi->limit.res[RLIMIT_NPROC])
 922                 >= vxi->limit.rlim[RLIMIT_NPROC]))
 923                 goto bad_fork_free;
 924
 925         if (atomic_read(&p->user->processes) >=
 926                         p->rlim[RLIMIT_NPROC].rlim_cur) {
 927                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 928                                 p->user != &root_user)
 929                         goto bad_fork_free;
 930         }
 931
 932         atomic_inc(&p->user->__count);
 933         atomic_inc(&p->user->processes);
 934         get_group_info(p->group_info);
 935
 936         /*
 937          * If multiple threads are within copy_process(), then this check
 938          * triggers too late. This doesn't hurt, the check is only there
 939          * to stop root fork bombs.
 940          */
 941         if (nr_threads >= max_threads)
 942                 goto bad_fork_cleanup_count;
 943
 944         if (!try_module_get(p->thread_info->exec_domain->module))
 945                 goto bad_fork_cleanup_count;
 946
 947         if (p->binfmt && !try_module_get(p->binfmt->module))
 948                 goto bad_fork_cleanup_put_domain;
 949
 950         p->did_exec = 0;
 951         copy_flags(clone_flags, p);
 952         if (clone_flags & CLONE_IDLETASK)
 953                 p->pid = 0;
 954         else {
 955                 p->pid = alloc_pidmap();
 956                 if (p->pid == -1)
 957                         goto bad_fork_cleanup;
 958         }
 959         retval = -EFAULT;
 960         if (clone_flags & CLONE_PARENT_SETTID)
 961                 if (put_user(p->pid, parent_tidptr))
 962                         goto bad_fork_cleanup;
 963
 964         p->proc_dentry = NULL;
 965
 966         INIT_LIST_HEAD(&p->children);
 967         INIT_LIST_HEAD(&p->sibling);
 968         init_waitqueue_head(&p->wait_chldexit);
 969         p->vfork_done = NULL;
 970         spin_lock_init(&p->alloc_lock);
 971         spin_lock_init(&p->proc_lock);
 972
 973         clear_tsk_thread_flag(p, TIF_SIGPENDING);
 974         init_sigpending(&p->pending);
 975
 976         p->it_real_value = p->it_virt_value = p->it_prof_value = 0;
 977         p->it_real_incr = p->it_virt_incr = p->it_prof_incr = 0;
 978         init_timer(&p->real_timer);
 979         p->real_timer.data = (unsigned long) p;
 980
 981         p->utime = p->stime = 0;
 982         p->cutime = p->cstime = 0;
 983         p->lock_depth = -1;             /* -1 = no lock */
 984         p->start_time = get_jiffies_64();
 985         p->security = NULL;
 986         p->io_context = NULL;
 987         p->audit_context = NULL;
 988
 989         retval = -ENOMEM;
 990         if ((retval = security_task_alloc(p)))
 991                 goto bad_fork_cleanup;
 992         if ((retval = audit_alloc(p)))
 993                 goto bad_fork_cleanup_security;
 994         /* copy all the process information */
 995         if ((retval = copy_semundo(clone_flags, p)))
 996                 goto bad_fork_cleanup_audit;
 997         if ((retval = copy_files(clone_flags, p)))
 998                 goto bad_fork_cleanup_semundo;
 999         if ((retval = copy_fs(clone_flags, p)))
1000                 goto bad_fork_cleanup_files;
1001         if ((retval = copy_sighand(clone_flags, p)))
1002                 goto bad_fork_cleanup_fs;
1003         if ((retval = copy_signal(clone_flags, p)))
1004                 goto bad_fork_cleanup_sighand;
1005         if ((retval = copy_mm(clone_flags, p)))
1006                 goto bad_fork_cleanup_signal;
1007         if ((retval = copy_namespace(clone_flags, p)))
1008                 goto bad_fork_cleanup_mm;
1009         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1010         if (retval)
1011                 goto bad_fork_cleanup_namespace;
1012
1013         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1014         /*
1015          * Clear TID on mm_release()?
1016          */
1017         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1018
1019         /*
1020          * Syscall tracing should be turned off in the child regardless
1021          * of CLONE_PTRACE.
1022          */
1023         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1024
1025         /* Our parent execution domain becomes current domain
1026            These must match for thread signalling to apply */
1027
1028         p->parent_exec_id = p->self_exec_id;
1029
1030         /* ok, now we should be set up.. */
1031         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1032         p->pdeath_signal = 0;
1033
1034         /* Perform scheduler related setup */
1035         sched_fork(p);
1036
1037         /*
1038          * Ok, make it visible to the rest of the system.
1039          * We dont wake it up yet.
1040          */
1041         p->tgid = p->pid;
1042         p->group_leader = p;
1043         INIT_LIST_HEAD(&p->ptrace_children);
1044         INIT_LIST_HEAD(&p->ptrace_list);
1045
1046         /* Need tasklist lock for parent etc handling! */
1047         write_lock_irq(&tasklist_lock);
1048         /*
1049          * Check for pending SIGKILL! The new thread should not be allowed
1050          * to slip out of an OOM kill. (or normal SIGKILL.)
1051          */
1052         if (sigismember(&current->pending.signal, SIGKILL)) {
1053                 write_unlock_irq(&tasklist_lock);
1054                 retval = -EINTR;
1055                 goto bad_fork_cleanup_namespace;
1056         }
1057
1058         /* CLONE_PARENT re-uses the old parent */
1059         if (clone_flags & CLONE_PARENT)
1060                 p->real_parent = current->real_parent;
1061         else
1062                 p->real_parent = current;
1063         p->parent = p->real_parent;
1064
1065         if (clone_flags & CLONE_THREAD) {
1066                 spin_lock(&current->sighand->siglock);
1067                 /*
1068                  * Important: if an exit-all has been started then
1069                  * do not create this new thread - the whole thread
1070                  * group is supposed to exit anyway.
1071                  */
1072                 if (current->signal->group_exit) {
1073                         spin_unlock(&current->sighand->siglock);
1074                         write_unlock_irq(&tasklist_lock);
1075                         retval = -EAGAIN;
1076                         goto bad_fork_cleanup_namespace;
1077                 }
1078                 p->tgid = current->tgid;
1079                 p->group_leader = current->group_leader;
1080
1081                 if (current->signal->group_stop_count > 0) {
1082                         /*
1083                          * There is an all-stop in progress for the group.
1084                          * We ourselves will stop as soon as we check signals.
1085                          * Make the new thread part of that group stop too.
1086                          */
1087                         current->signal->group_stop_count++;
1088                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1089                 }
1090
1091                 spin_unlock(&current->sighand->siglock);
1092         }
1093
1094         SET_LINKS(p);
1095         if (p->ptrace & PT_PTRACED)
1096                 __ptrace_link(p, current->parent);
1097
1098         attach_pid(p, PIDTYPE_PID, p->pid);
1099         if (thread_group_leader(p)) {
1100                 attach_pid(p, PIDTYPE_TGID, p->tgid);
1101                 attach_pid(p, PIDTYPE_PGID, process_group(p));
1102                 attach_pid(p, PIDTYPE_SID, p->signal->session);
1103                 if (p->pid)
1104                         __get_cpu_var(process_counts)++;
1105         } else
1106                 link_pid(p, p->pids + PIDTYPE_TGID, &p->group_leader->pids[PIDTYPE_TGID].pid);
1107
1108         nr_threads++;
1109         if (vxi) {
1110                 atomic_inc(&vxi->cacct.nr_threads);
1111                 atomic_inc(&vxi->limit.res[RLIMIT_NPROC]);
1112         }
1113         write_unlock_irq(&tasklist_lock);
1114         retval = 0;
1115
1116 fork_out:
1117         if (retval)
1118                 return ERR_PTR(retval);
1119         return p;
1120
1121 bad_fork_cleanup_namespace:
1122         exit_namespace(p);
1123 bad_fork_cleanup_mm:
1124         exit_mm(p);
1125         if (p->active_mm)
1126                 mmdrop(p->active_mm);
1127 bad_fork_cleanup_signal:
1128         exit_signal(p);
1129 bad_fork_cleanup_sighand:
1130         exit_sighand(p);
1131 bad_fork_cleanup_fs:
1132         exit_fs(p); /* blocking */
1133 bad_fork_cleanup_files:
1134         exit_files(p); /* blocking */
1135 bad_fork_cleanup_semundo:
1136         exit_sem(p);
1137 bad_fork_cleanup_audit:
1138         audit_free(p);
1139 bad_fork_cleanup_security:
1140         security_task_free(p);
1141 bad_fork_cleanup:
1142         if (p->pid > 0)
1143                 free_pidmap(p->pid);
1144         if (p->binfmt)
1145                 module_put(p->binfmt->module);
1146 bad_fork_cleanup_put_domain:
1147         module_put(p->thread_info->exec_domain->module);
1148 bad_fork_cleanup_count:
1149         put_group_info(p->group_info);
1150         atomic_dec(&p->user->processes);
1151         free_uid(p->user);
1152 bad_fork_free:
1153         free_task(p);
1154         goto fork_out;
1155 }
1156
1157 static inline int fork_traceflag (unsigned clone_flags)
1158 {
1159         if (clone_flags & (CLONE_UNTRACED | CLONE_IDLETASK))
1160                 return 0;
1161         else if (clone_flags & CLONE_VFORK) {
1162                 if (current->ptrace & PT_TRACE_VFORK)
1163                         return PTRACE_EVENT_VFORK;
1164         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1165                 if (current->ptrace & PT_TRACE_CLONE)
1166                         return PTRACE_EVENT_CLONE;
1167         } else if (current->ptrace & PT_TRACE_FORK)
1168                 return PTRACE_EVENT_FORK;
1169
1170         return 0;
1171 }
1172
1173 /*
1174  *  Ok, this is the main fork-routine.
1175  *
1176  * It copies the process, and if successful kick-starts
1177  * it and waits for it to finish using the VM if required.
1178  */
1179 long do_fork(unsigned long clone_flags,
1180               unsigned long stack_start,
1181               struct pt_regs *regs,
1182               unsigned long stack_size,
1183               int __user *parent_tidptr,
1184               int __user *child_tidptr)
1185 {
1186         struct task_struct *p;
1187         int trace = 0;
1188         long pid;
1189
1190         if (unlikely(current->ptrace)) {
1191                 trace = fork_traceflag (clone_flags);
1192                 if (trace)
1193                         clone_flags |= CLONE_PTRACE;
1194         }
1195
1196         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr);
1197         /*
1198          * Do this prior waking up the new thread - the thread pointer
1199          * might get invalid after that point, if the thread exits quickly.
1200          */
1201         pid = IS_ERR(p) ? PTR_ERR(p) : p->pid;
1202
1203         if (!IS_ERR(p)) {
1204                 struct completion vfork;
1205
1206                 if (clone_flags & CLONE_VFORK) {
1207                         p->vfork_done = &vfork;
1208                         init_completion(&vfork);
1209                 }
1210
1211                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1212                         /*
1213                          * We'll start up with an immediate SIGSTOP.
1214                          */
1215                         sigaddset(&p->pending.signal, SIGSTOP);
1216                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1217                 }
1218
1219                 if (!(clone_flags & CLONE_STOPPED))
1220                         wake_up_forked_process(p);      /* do this last */
1221                 else
1222                         p->state = TASK_STOPPED;
1223                 ++total_forks;
1224
1225                 if (unlikely (trace)) {
1226                         current->ptrace_message = pid;
1227                         ptrace_notify ((trace << 8) | SIGTRAP);
1228                 }
1229
1230                 if (clone_flags & CLONE_VFORK) {
1231                         wait_for_completion(&vfork);
1232                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1233                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1234                 } else
1235                         /*
1236                          * Let the child process run first, to avoid most of the
1237                          * COW overhead when the child exec()s afterwards.
1238                          */
1239                         set_need_resched();
1240         }
1241         return pid;
1242 }
1243
1244 /* SLAB cache for signal_struct structures (tsk->signal) */
1245 kmem_cache_t *signal_cachep;
1246
1247 /* SLAB cache for sighand_struct structures (tsk->sighand) */
1248 kmem_cache_t *sighand_cachep;
1249
1250 /* SLAB cache for files_struct structures (tsk->files) */
1251 kmem_cache_t *files_cachep;
1252
1253 /* SLAB cache for fs_struct structures (tsk->fs) */
1254 kmem_cache_t *fs_cachep;
1255
1256 /* SLAB cache for vm_area_struct structures */
1257 kmem_cache_t *vm_area_cachep;
1258
1259 /* SLAB cache for mm_struct structures (tsk->mm) */
1260 kmem_cache_t *mm_cachep;
1261
1262 void __init proc_caches_init(void)
1263 {
1264         sighand_cachep = kmem_cache_create("sighand_cache",
1265                         sizeof(struct sighand_struct), 0,
1266                         SLAB_HWCACHE_ALIGN, NULL, NULL);
1267         if (!sighand_cachep)
1268                 panic("Cannot create sighand SLAB cache");
1269
1270         signal_cachep = kmem_cache_create("signal_cache",
1271                         sizeof(struct signal_struct), 0,
1272                         SLAB_HWCACHE_ALIGN, NULL, NULL);
1273         if (!signal_cachep)
1274                 panic("Cannot create signal SLAB cache");
1275
1276         files_cachep = kmem_cache_create("files_cache",
1277                          sizeof(struct files_struct), 0,
1278                          SLAB_HWCACHE_ALIGN, NULL, NULL);
1279         if (!files_cachep)
1280                 panic("Cannot create files SLAB cache");
1281
1282         fs_cachep = kmem_cache_create("fs_cache",
1283                          sizeof(struct fs_struct), 0,
1284                          SLAB_HWCACHE_ALIGN, NULL, NULL);
1285         if (!fs_cachep)
1286                 panic("Cannot create fs_struct SLAB cache");
1287
1288         vm_area_cachep = kmem_cache_create("vm_area_struct",
1289                         sizeof(struct vm_area_struct), 0,
1290                         0, NULL, NULL);
1291         if(!vm_area_cachep)
1292                 panic("vma_init: Cannot alloc vm_area_struct SLAB cache");
1293
1294         mm_cachep = kmem_cache_create("mm_struct",
1295                         sizeof(struct mm_struct), 0,
1296                         SLAB_HWCACHE_ALIGN, NULL, NULL);
1297         if(!mm_cachep)
1298                 panic("vma_init: Cannot alloc mm_struct SLAB cache");
1299 }