kernel/fork.c

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