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