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