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/capability.h>
  32 #include <linux/cpu.h>
  33 #include <linux/cpuset.h>
  34 #include <linux/security.h>
  35 #include <linux/swap.h>
  36 #include <linux/syscalls.h>
  37 #include <linux/jiffies.h>
  38 #include <linux/futex.h>
  39 #include <linux/rcupdate.h>
  40 #include <linux/ptrace.h>
  41 #include <linux/mount.h>
  42 #include <linux/audit.h>
  43 #include <linux/profile.h>
  44 #include <linux/rmap.h>
  45 #include <linux/acct.h>
  46 #include <linux/cn_proc.h>
  47
  48 #include <asm/pgtable.h>
  49 #include <asm/pgalloc.h>
  50 #include <asm/uaccess.h>
  51 #include <asm/mmu_context.h>
  52 #include <asm/cacheflush.h>
  53 #include <asm/tlbflush.h>
  54
  55 /*
  56  * Protected counters by write_lock_irq(&tasklist_lock)
  57  */
  58 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  59 int nr_threads;                 /* The idle threads do not count.. */
  60
  61 int max_threads;                /* tunable limit on nr_threads */
  62
  63 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  64
  65  __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  66
  67 EXPORT_SYMBOL(tasklist_lock);
  68
  69 int nr_processes(void)
  70 {
  71         int cpu;
  72         int total = 0;
  73
  74         for_each_online_cpu(cpu)
  75                 total += per_cpu(process_counts, cpu);
  76
  77         return total;
  78 }
  79
  80 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  81 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  82 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  83 static kmem_cache_t *task_struct_cachep;
  84 #endif
  85
  86 /* SLAB cache for signal_struct structures (tsk->signal) */
  87 static kmem_cache_t *signal_cachep;
  88
  89 /* SLAB cache for sighand_struct structures (tsk->sighand) */
  90 kmem_cache_t *sighand_cachep;
  91
  92 /* SLAB cache for files_struct structures (tsk->files) */
  93 kmem_cache_t *files_cachep;
  94
  95 /* SLAB cache for fs_struct structures (tsk->fs) */
  96 kmem_cache_t *fs_cachep;
  97
  98 /* SLAB cache for vm_area_struct structures */
  99 kmem_cache_t *vm_area_cachep;
 100
 101 /* SLAB cache for mm_struct structures (tsk->mm) */
 102 static kmem_cache_t *mm_cachep;
 103
 104 void free_task(struct task_struct *tsk)
 105 {
 106         free_thread_info(tsk->thread_info);
 107         free_task_struct(tsk);
 108 }
 109 EXPORT_SYMBOL(free_task);
 110
 111 void __put_task_struct(struct task_struct *tsk)
 112 {
 113         WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
 114         WARN_ON(atomic_read(&tsk->usage));
 115         WARN_ON(tsk == current);
 116
 117         if (unlikely(tsk->audit_context))
 118                 audit_free(tsk);
 119         security_task_free(tsk);
 120         free_uid(tsk->user);
 121         put_group_info(tsk->group_info);
 122
 123         if (!profile_handoff_task(tsk))
 124                 free_task(tsk);
 125 }
 126
 127 void __put_task_struct_cb(struct rcu_head *rhp)
 128 {
 129         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
 130         __put_task_struct(tsk);
 131 }
 132
 133 void __init fork_init(unsigned long mempages)
 134 {
 135 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 136 #ifndef ARCH_MIN_TASKALIGN
 137 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 138 #endif
 139         /* create a slab on which task_structs can be allocated */
 140         task_struct_cachep =
 141                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 142                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
 143 #endif
 144
 145         /*
 146          * The default maximum number of threads is set to a safe
 147          * value: the thread structures can take up at most half
 148          * of memory.
 149          */
 150         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 151
 152         /*
 153          * we need to allow at least 20 threads to boot a system
 154          */
 155         if(max_threads < 20)
 156                 max_threads = 20;
 157
 158         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 159         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 160         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 161                 init_task.signal->rlim[RLIMIT_NPROC];
 162 }
 163
 164 static struct task_struct *dup_task_struct(struct task_struct *orig)
 165 {
 166         struct task_struct *tsk;
 167         struct thread_info *ti;
 168
 169         prepare_to_copy(orig);
 170
 171         tsk = alloc_task_struct();
 172         if (!tsk)
 173                 return NULL;
 174
 175         ti = alloc_thread_info(tsk);
 176         if (!ti) {
 177                 free_task_struct(tsk);
 178                 return NULL;
 179         }
 180
 181         *tsk = *orig;
 182         tsk->thread_info = ti;
 183         setup_thread_stack(tsk, orig);
 184
 185         /* One for us, one for whoever does the "release_task()" (usually parent) */
 186         atomic_set(&tsk->usage,2);
 187         atomic_set(&tsk->fs_excl, 0);
 188         tsk->btrace_seq = 0;
 189         return tsk;
 190 }
 191
 192 #ifdef CONFIG_MMU
 193 static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 194 {
 195         struct vm_area_struct *mpnt, *tmp, **pprev;
 196         struct rb_node **rb_link, *rb_parent;
 197         int retval;
 198         unsigned long charge;
 199         struct mempolicy *pol;
 200
 201         down_write(&oldmm->mmap_sem);
 202         flush_cache_mm(oldmm);
 203         down_write(&mm->mmap_sem);
 204
 205         mm->locked_vm = 0;
 206         mm->mmap = NULL;
 207         mm->mmap_cache = NULL;
 208         mm->free_area_cache = oldmm->mmap_base;
 209         mm->cached_hole_size = ~0UL;
 210         mm->map_count = 0;
 211         cpus_clear(mm->cpu_vm_mask);
 212         mm->mm_rb = RB_ROOT;
 213         rb_link = &mm->mm_rb.rb_node;
 214         rb_parent = NULL;
 215         pprev = &mm->mmap;
 216
 217         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 218                 struct file *file;
 219
 220                 if (mpnt->vm_flags & VM_DONTCOPY) {
 221                         long pages = vma_pages(mpnt);
 222                         mm->total_vm -= pages;
 223                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 224                                                                 -pages);
 225                         continue;
 226                 }
 227                 charge = 0;
 228                 if (mpnt->vm_flags & VM_ACCOUNT) {
 229                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 230                         if (security_vm_enough_memory(len))
 231                                 goto fail_nomem;
 232                         charge = len;
 233                 }
 234                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 235                 if (!tmp)
 236                         goto fail_nomem;
 237                 *tmp = *mpnt;
 238                 pol = mpol_copy(vma_policy(mpnt));
 239                 retval = PTR_ERR(pol);
 240                 if (IS_ERR(pol))
 241                         goto fail_nomem_policy;
 242                 vma_set_policy(tmp, pol);
 243                 tmp->vm_flags &= ~VM_LOCKED;
 244                 tmp->vm_mm = mm;
 245                 tmp->vm_next = NULL;
 246                 anon_vma_link(tmp);
 247                 file = tmp->vm_file;
 248                 if (file) {
 249                         struct inode *inode = file->f_dentry->d_inode;
 250                         get_file(file);
 251                         if (tmp->vm_flags & VM_DENYWRITE)
 252                                 atomic_dec(&inode->i_writecount);
 253
 254                         /* insert tmp into the share list, just after mpnt */
 255                         spin_lock(&file->f_mapping->i_mmap_lock);
 256                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 257                         flush_dcache_mmap_lock(file->f_mapping);
 258                         vma_prio_tree_add(tmp, mpnt);
 259                         flush_dcache_mmap_unlock(file->f_mapping);
 260                         spin_unlock(&file->f_mapping->i_mmap_lock);
 261                 }
 262
 263                 /*
 264                  * Link in the new vma and copy the page table entries.
 265                  */
 266                 *pprev = tmp;
 267                 pprev = &tmp->vm_next;
 268
 269                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 270                 rb_link = &tmp->vm_rb.rb_right;
 271                 rb_parent = &tmp->vm_rb;
 272
 273                 mm->map_count++;
 274                 retval = copy_page_range(mm, oldmm, mpnt);
 275
 276                 if (tmp->vm_ops && tmp->vm_ops->open)
 277                         tmp->vm_ops->open(tmp);
 278
 279                 if (retval)
 280                         goto out;
 281         }
 282         retval = 0;
 283 out:
 284         up_write(&mm->mmap_sem);
 285         flush_tlb_mm(oldmm);
 286         up_write(&oldmm->mmap_sem);
 287         return retval;
 288 fail_nomem_policy:
 289         kmem_cache_free(vm_area_cachep, tmp);
 290 fail_nomem:
 291         retval = -ENOMEM;
 292         vm_unacct_memory(charge);
 293         goto out;
 294 }
 295
 296 static inline int mm_alloc_pgd(struct mm_struct * mm)
 297 {
 298         mm->pgd = pgd_alloc(mm);
 299         if (unlikely(!mm->pgd))
 300                 return -ENOMEM;
 301         return 0;
 302 }
 303
 304 static inline void mm_free_pgd(struct mm_struct * mm)
 305 {
 306         pgd_free(mm->pgd);
 307 }
 308 #else
 309 #define dup_mmap(mm, oldmm)     (0)
 310 #define mm_alloc_pgd(mm)        (0)
 311 #define mm_free_pgd(mm)
 312 #endif /* CONFIG_MMU */
 313
 314  __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 315
 316 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
 317 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 318
 319 #include <linux/init_task.h>
 320
 321 static struct mm_struct * mm_init(struct mm_struct * mm)
 322 {
 323         atomic_set(&mm->mm_users, 1);
 324         atomic_set(&mm->mm_count, 1);
 325         init_rwsem(&mm->mmap_sem);
 326         INIT_LIST_HEAD(&mm->mmlist);
 327         mm->core_waiters = 0;
 328         mm->nr_ptes = 0;
 329         set_mm_counter(mm, file_rss, 0);
 330         set_mm_counter(mm, anon_rss, 0);
 331         spin_lock_init(&mm->page_table_lock);
 332         rwlock_init(&mm->ioctx_list_lock);
 333         mm->ioctx_list = NULL;
 334         mm->free_area_cache = TASK_UNMAPPED_BASE;
 335         mm->cached_hole_size = ~0UL;
 336
 337         if (likely(!mm_alloc_pgd(mm))) {
 338                 mm->def_flags = 0;
 339                 return mm;
 340         }
 341         free_mm(mm);
 342         return NULL;
 343 }
 344
 345 /*
 346  * Allocate and initialize an mm_struct.
 347  */
 348 struct mm_struct * mm_alloc(void)
 349 {
 350         struct mm_struct * mm;
 351
 352         mm = allocate_mm();
 353         if (mm) {
 354                 memset(mm, 0, sizeof(*mm));
 355                 mm = mm_init(mm);
 356         }
 357         return mm;
 358 }
 359
 360 /*
 361  * Called when the last reference to the mm
 362  * is dropped: either by a lazy thread or by
 363  * mmput. Free the page directory and the mm.
 364  */
 365 void fastcall __mmdrop(struct mm_struct *mm)
 366 {
 367         BUG_ON(mm == &init_mm);
 368         mm_free_pgd(mm);
 369         destroy_context(mm);
 370         free_mm(mm);
 371 }
 372
 373 /*
 374  * Decrement the use count and release all resources for an mm.
 375  */
 376 void mmput(struct mm_struct *mm)
 377 {
 378         if (atomic_dec_and_test(&mm->mm_users)) {
 379                 exit_aio(mm);
 380                 exit_mmap(mm);
 381                 if (!list_empty(&mm->mmlist)) {
 382                         spin_lock(&mmlist_lock);
 383                         list_del(&mm->mmlist);
 384                         spin_unlock(&mmlist_lock);
 385                 }
 386                 put_swap_token(mm);
 387                 mmdrop(mm);
 388         }
 389 }
 390 EXPORT_SYMBOL_GPL(mmput);
 391
 392 /**
 393  * get_task_mm - acquire a reference to the task's mm
 394  *
 395  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
 396  * this kernel workthread has transiently adopted a user mm with use_mm,
 397  * to do its AIO) is not set and if so returns a reference to it, after
 398  * bumping up the use count.  User must release the mm via mmput()
 399  * after use.  Typically used by /proc and ptrace.
 400  */
 401 struct mm_struct *get_task_mm(struct task_struct *task)
 402 {
 403         struct mm_struct *mm;
 404
 405         task_lock(task);
 406         mm = task->mm;
 407         if (mm) {
 408                 if (task->flags & PF_BORROWED_MM)
 409                         mm = NULL;
 410                 else
 411                         atomic_inc(&mm->mm_users);
 412         }
 413         task_unlock(task);
 414         return mm;
 415 }
 416 EXPORT_SYMBOL_GPL(get_task_mm);
 417
 418 /* Please note the differences between mmput and mm_release.
 419  * mmput is called whenever we stop holding onto a mm_struct,
 420  * error success whatever.
 421  *
 422  * mm_release is called after a mm_struct has been removed
 423  * from the current process.
 424  *
 425  * This difference is important for error handling, when we
 426  * only half set up a mm_struct for a new process and need to restore
 427  * the old one.  Because we mmput the new mm_struct before
 428  * restoring the old one. . .
 429  * Eric Biederman 10 January 1998
 430  */
 431 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 432 {
 433         struct completion *vfork_done = tsk->vfork_done;
 434
 435         /* Get rid of any cached register state */
 436         deactivate_mm(tsk, mm);
 437
 438         /* notify parent sleeping on vfork() */
 439         if (vfork_done) {
 440                 tsk->vfork_done = NULL;
 441                 complete(vfork_done);
 442         }
 443         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
 444                 u32 __user * tidptr = tsk->clear_child_tid;
 445                 tsk->clear_child_tid = NULL;
 446
 447                 /*
 448                  * We don't check the error code - if userspace has
 449                  * not set up a proper pointer then tough luck.
 450                  */
 451                 put_user(0, tidptr);
 452                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 453         }
 454 }
 455
 456 /*
 457  * Allocate a new mm structure and copy contents from the
 458  * mm structure of the passed in task structure.
 459  */
 460 static struct mm_struct *dup_mm(struct task_struct *tsk)
 461 {
 462         struct mm_struct *mm, *oldmm = current->mm;
 463         int err;
 464
 465         if (!oldmm)
 466                 return NULL;
 467
 468         mm = allocate_mm();
 469         if (!mm)
 470                 goto fail_nomem;
 471
 472         memcpy(mm, oldmm, sizeof(*mm));
 473
 474         if (!mm_init(mm))
 475                 goto fail_nomem;
 476
 477         if (init_new_context(tsk, mm))
 478                 goto fail_nocontext;
 479
 480         err = dup_mmap(mm, oldmm);
 481         if (err)
 482                 goto free_pt;
 483
 484         mm->hiwater_rss = get_mm_rss(mm);
 485         mm->hiwater_vm = mm->total_vm;
 486
 487         return mm;
 488
 489 free_pt:
 490         mmput(mm);
 491
 492 fail_nomem:
 493         return NULL;
 494
 495 fail_nocontext:
 496         /*
 497          * If init_new_context() failed, we cannot use mmput() to free the mm
 498          * because it calls destroy_context()
 499          */
 500         mm_free_pgd(mm);
 501         free_mm(mm);
 502         return NULL;
 503 }
 504
 505 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 506 {
 507         struct mm_struct * mm, *oldmm;
 508         int retval;
 509
 510         tsk->min_flt = tsk->maj_flt = 0;
 511         tsk->nvcsw = tsk->nivcsw = 0;
 512
 513         tsk->mm = NULL;
 514         tsk->active_mm = NULL;
 515
 516         /*
 517          * Are we cloning a kernel thread?
 518          *
 519          * We need to steal a active VM for that..
 520          */
 521         oldmm = current->mm;
 522         if (!oldmm)
 523                 return 0;
 524
 525         if (clone_flags & CLONE_VM) {
 526                 atomic_inc(&oldmm->mm_users);
 527                 mm = oldmm;
 528                 goto good_mm;
 529         }
 530
 531         retval = -ENOMEM;
 532         mm = dup_mm(tsk);
 533         if (!mm)
 534                 goto fail_nomem;
 535
 536 good_mm:
 537         tsk->mm = mm;
 538         tsk->active_mm = mm;
 539         return 0;
 540
 541 fail_nomem:
 542         return retval;
 543 }
 544
 545 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 546 {
 547         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 548         /* We don't need to lock fs - think why ;-) */
 549         if (fs) {
 550                 atomic_set(&fs->count, 1);
 551                 rwlock_init(&fs->lock);
 552                 fs->umask = old->umask;
 553                 read_lock(&old->lock);
 554                 fs->rootmnt = mntget(old->rootmnt);
 555                 fs->root = dget(old->root);
 556                 fs->pwdmnt = mntget(old->pwdmnt);
 557                 fs->pwd = dget(old->pwd);
 558                 if (old->altroot) {
 559                         fs->altrootmnt = mntget(old->altrootmnt);
 560                         fs->altroot = dget(old->altroot);
 561                 } else {
 562                         fs->altrootmnt = NULL;
 563                         fs->altroot = NULL;
 564                 }
 565                 read_unlock(&old->lock);
 566         }
 567         return fs;
 568 }
 569
 570 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 571 {
 572         return __copy_fs_struct(old);
 573 }
 574
 575 EXPORT_SYMBOL_GPL(copy_fs_struct);
 576
 577 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 578 {
 579         if (clone_flags & CLONE_FS) {
 580                 atomic_inc(&current->fs->count);
 581                 return 0;
 582         }
 583         tsk->fs = __copy_fs_struct(current->fs);
 584         if (!tsk->fs)
 585                 return -ENOMEM;
 586         return 0;
 587 }
 588
 589 static int count_open_files(struct fdtable *fdt)
 590 {
 591         int size = fdt->max_fdset;
 592         int i;
 593
 594         /* Find the last open fd */
 595         for (i = size/(8*sizeof(long)); i > 0; ) {
 596                 if (fdt->open_fds->fds_bits[--i])
 597                         break;
 598         }
 599         i = (i+1) * 8 * sizeof(long);
 600         return i;
 601 }
 602
 603 static struct files_struct *alloc_files(void)
 604 {
 605         struct files_struct *newf;
 606         struct fdtable *fdt;
 607
 608         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
 609         if (!newf)
 610                 goto out;
 611
 612         atomic_set(&newf->count, 1);
 613
 614         spin_lock_init(&newf->file_lock);
 615         newf->next_fd = 0;
 616         fdt = &newf->fdtab;
 617         fdt->max_fds = NR_OPEN_DEFAULT;
 618         fdt->max_fdset = EMBEDDED_FD_SET_SIZE;
 619         fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
 620         fdt->open_fds = (fd_set *)&newf->open_fds_init;
 621         fdt->fd = &newf->fd_array[0];
 622         INIT_RCU_HEAD(&fdt->rcu);
 623         fdt->free_files = NULL;
 624         fdt->next = NULL;
 625         rcu_assign_pointer(newf->fdt, fdt);
 626 out:
 627         return newf;
 628 }
 629
 630 /*
 631  * Allocate a new files structure and copy contents from the
 632  * passed in files structure.
 633  */
 634 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
 635 {
 636         struct files_struct *newf;
 637         struct file **old_fds, **new_fds;
 638         int open_files, size, i, expand;
 639         struct fdtable *old_fdt, *new_fdt;
 640
 641         newf = alloc_files();
 642         if (!newf)
 643                 goto out;
 644
 645         spin_lock(&oldf->file_lock);
 646         old_fdt = files_fdtable(oldf);
 647         new_fdt = files_fdtable(newf);
 648         size = old_fdt->max_fdset;
 649         open_files = count_open_files(old_fdt);
 650         expand = 0;
 651
 652         /*
 653          * Check whether we need to allocate a larger fd array or fd set.
 654          * Note: we're not a clone task, so the open count won't  change.
 655          */
 656         if (open_files > new_fdt->max_fdset) {
 657                 new_fdt->max_fdset = 0;
 658                 expand = 1;
 659         }
 660         if (open_files > new_fdt->max_fds) {
 661                 new_fdt->max_fds = 0;
 662                 expand = 1;
 663         }
 664
 665         /* if the old fdset gets grown now, we'll only copy up to "size" fds */
 666         if (expand) {
 667                 spin_unlock(&oldf->file_lock);
 668                 spin_lock(&newf->file_lock);
 669                 *errorp = expand_files(newf, open_files-1);
 670                 spin_unlock(&newf->file_lock);
 671                 if (*errorp < 0)
 672                         goto out_release;
 673                 new_fdt = files_fdtable(newf);
 674                 /*
 675                  * Reacquire the oldf lock and a pointer to its fd table
 676                  * who knows it may have a new bigger fd table. We need
 677                  * the latest pointer.
 678                  */
 679                 spin_lock(&oldf->file_lock);
 680                 old_fdt = files_fdtable(oldf);
 681         }
 682
 683         old_fds = old_fdt->fd;
 684         new_fds = new_fdt->fd;
 685
 686         memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
 687         memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
 688
 689         for (i = open_files; i != 0; i--) {
 690                 struct file *f = *old_fds++;
 691                 if (f) {
 692                         get_file(f);
 693                 } else {
 694                         /*
 695                          * The fd may be claimed in the fd bitmap but not yet
 696                          * instantiated in the files array if a sibling thread
 697                          * is partway through open().  So make sure that this
 698                          * fd is available to the new process.
 699                          */
 700                         FD_CLR(open_files - i, new_fdt->open_fds);
 701                 }
 702                 rcu_assign_pointer(*new_fds++, f);
 703         }
 704         spin_unlock(&oldf->file_lock);
 705
 706         /* compute the remainder to be cleared */
 707         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
 708
 709         /* This is long word aligned thus could use a optimized version */
 710         memset(new_fds, 0, size);
 711
 712         if (new_fdt->max_fdset > open_files) {
 713                 int left = (new_fdt->max_fdset-open_files)/8;
 714                 int start = open_files / (8 * sizeof(unsigned long));
 715
 716                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
 717                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
 718         }
 719
 720 out:
 721         return newf;
 722
 723 out_release:
 724         free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
 725         free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
 726         free_fd_array(new_fdt->fd, new_fdt->max_fds);
 727         kmem_cache_free(files_cachep, newf);
 728         return NULL;
 729 }
 730
 731 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 732 {
 733         struct files_struct *oldf, *newf;
 734         int error = 0;
 735
 736         /*
 737          * A background process may not have any files ...
 738          */
 739         oldf = current->files;
 740         if (!oldf)
 741                 goto out;
 742
 743         if (clone_flags & CLONE_FILES) {
 744                 atomic_inc(&oldf->count);
 745                 goto out;
 746         }
 747
 748         /*
 749          * Note: we may be using current for both targets (See exec.c)
 750          * This works because we cache current->files (old) as oldf. Don't
 751          * break this.
 752          */
 753         tsk->files = NULL;
 754         error = -ENOMEM;
 755         newf = dup_fd(oldf, &error);
 756         if (!newf)
 757                 goto out;
 758
 759         tsk->files = newf;
 760         error = 0;
 761 out:
 762         return error;
 763 }
 764
 765 /*
 766  *      Helper to unshare the files of the current task.
 767  *      We don't want to expose copy_files internals to
 768  *      the exec layer of the kernel.
 769  */
 770
 771 int unshare_files(void)
 772 {
 773         struct files_struct *files  = current->files;
 774         int rc;
 775
 776         BUG_ON(!files);
 777
 778         /* This can race but the race causes us to copy when we don't
 779            need to and drop the copy */
 780         if(atomic_read(&files->count) == 1)
 781         {
 782                 atomic_inc(&files->count);
 783                 return 0;
 784         }
 785         rc = copy_files(0, current);
 786         if(rc)
 787                 current->files = files;
 788         return rc;
 789 }
 790
 791 EXPORT_SYMBOL(unshare_files);
 792
 793 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 794 {
 795         struct sighand_struct *sig;
 796
 797         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 798                 atomic_inc(&current->sighand->count);
 799                 return 0;
 800         }
 801         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 802         rcu_assign_pointer(tsk->sighand, sig);
 803         if (!sig)
 804                 return -ENOMEM;
 805         atomic_set(&sig->count, 1);
 806         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 807         return 0;
 808 }
 809
 810 void __cleanup_sighand(struct sighand_struct *sighand)
 811 {
 812         if (atomic_dec_and_test(&sighand->count))
 813                 kmem_cache_free(sighand_cachep, sighand);
 814 }
 815
 816 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 817 {
 818         struct signal_struct *sig;
 819         int ret;
 820
 821         if (clone_flags & CLONE_THREAD) {
 822                 atomic_inc(&current->signal->count);
 823                 atomic_inc(&current->signal->live);
 824                 return 0;
 825         }
 826         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 827         tsk->signal = sig;
 828         if (!sig)
 829                 return -ENOMEM;
 830
 831         ret = copy_thread_group_keys(tsk);
 832         if (ret < 0) {
 833                 kmem_cache_free(signal_cachep, sig);
 834                 return ret;
 835         }
 836
 837         atomic_set(&sig->count, 1);
 838         atomic_set(&sig->live, 1);
 839         init_waitqueue_head(&sig->wait_chldexit);
 840         sig->flags = 0;
 841         sig->group_exit_code = 0;
 842         sig->group_exit_task = NULL;
 843         sig->group_stop_count = 0;
 844         sig->curr_target = NULL;
 845         init_sigpending(&sig->shared_pending);
 846         INIT_LIST_HEAD(&sig->posix_timers);
 847
 848         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_REL);
 849         sig->it_real_incr.tv64 = 0;
 850         sig->real_timer.function = it_real_fn;
 851         sig->tsk = tsk;
 852
 853         sig->it_virt_expires = cputime_zero;
 854         sig->it_virt_incr = cputime_zero;
 855         sig->it_prof_expires = cputime_zero;
 856         sig->it_prof_incr = cputime_zero;
 857
 858         sig->leader = 0;        /* session leadership doesn't inherit */
 859         sig->tty_old_pgrp = 0;
 860
 861         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 862         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 863         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 864         sig->sched_time = 0;
 865         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 866         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 867         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 868
 869         task_lock(current->group_leader);
 870         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 871         task_unlock(current->group_leader);
 872
 873         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 874                 /*
 875                  * New sole thread in the process gets an expiry time
 876                  * of the whole CPU time limit.
 877                  */
 878                 tsk->it_prof_expires =
 879                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 880         }
 881
 882         return 0;
 883 }
 884
 885 void __cleanup_signal(struct signal_struct *sig)
 886 {
 887         exit_thread_group_keys(sig);
 888         kmem_cache_free(signal_cachep, sig);
 889 }
 890
 891 static inline void cleanup_signal(struct task_struct *tsk)
 892 {
 893         struct signal_struct *sig = tsk->signal;
 894
 895         atomic_dec(&sig->live);
 896
 897         if (atomic_dec_and_test(&sig->count))
 898                 __cleanup_signal(sig);
 899 }
 900
 901 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 902 {
 903         unsigned long new_flags = p->flags;
 904
 905         new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
 906         new_flags |= PF_FORKNOEXEC;
 907         if (!(clone_flags & CLONE_PTRACE))
 908                 p->ptrace = 0;
 909         p->flags = new_flags;
 910 }
 911
 912 asmlinkage long sys_set_tid_address(int __user *tidptr)
 913 {
 914         current->clear_child_tid = tidptr;
 915
 916         return current->pid;
 917 }
 918
 919 /*
 920  * This creates a new process as a copy of the old one,
 921  * but does not actually start it yet.
 922  *
 923  * It copies the registers, and all the appropriate
 924  * parts of the process environment (as per the clone
 925  * flags). The actual kick-off is left to the caller.
 926  */
 927 static task_t *copy_process(unsigned long clone_flags,
 928                                  unsigned long stack_start,
 929                                  struct pt_regs *regs,
 930                                  unsigned long stack_size,
 931                                  int __user *parent_tidptr,
 932                                  int __user *child_tidptr,
 933                                  int pid)
 934 {
 935         int retval;
 936         struct task_struct *p = NULL;
 937
 938         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 939                 return ERR_PTR(-EINVAL);
 940
 941         /*
 942          * Thread groups must share signals as well, and detached threads
 943          * can only be started up within the thread group.
 944          */
 945         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 946                 return ERR_PTR(-EINVAL);
 947
 948         /*
 949          * Shared signal handlers imply shared VM. By way of the above,
 950          * thread groups also imply shared VM. Blocking this case allows
 951          * for various simplifications in other code.
 952          */
 953         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 954                 return ERR_PTR(-EINVAL);
 955
 956         retval = security_task_create(clone_flags);
 957         if (retval)
 958                 goto fork_out;
 959
 960         retval = -ENOMEM;
 961         p = dup_task_struct(current);
 962         if (!p)
 963                 goto fork_out;
 964
 965         retval = -EAGAIN;
 966         if (atomic_read(&p->user->processes) >=
 967                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
 968                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 969                                 p->user != &root_user)
 970                         goto bad_fork_free;
 971         }
 972
 973         atomic_inc(&p->user->__count);
 974         atomic_inc(&p->user->processes);
 975         get_group_info(p->group_info);
 976
 977         /*
 978          * If multiple threads are within copy_process(), then this check
 979          * triggers too late. This doesn't hurt, the check is only there
 980          * to stop root fork bombs.
 981          */
 982         if (nr_threads >= max_threads)
 983                 goto bad_fork_cleanup_count;
 984
 985         if (!try_module_get(task_thread_info(p)->exec_domain->module))
 986                 goto bad_fork_cleanup_count;
 987
 988         if (p->binfmt && !try_module_get(p->binfmt->module))
 989                 goto bad_fork_cleanup_put_domain;
 990
 991         p->did_exec = 0;
 992         copy_flags(clone_flags, p);
 993         p->pid = pid;
 994         retval = -EFAULT;
 995         if (clone_flags & CLONE_PARENT_SETTID)
 996                 if (put_user(p->pid, parent_tidptr))
 997                         goto bad_fork_cleanup;
 998
 999         p->proc_dentry = NULL;
1000
1001         INIT_LIST_HEAD(&p->children);
1002         INIT_LIST_HEAD(&p->sibling);
1003         p->vfork_done = NULL;
1004         spin_lock_init(&p->alloc_lock);
1005         spin_lock_init(&p->proc_lock);
1006
1007         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1008         init_sigpending(&p->pending);
1009
1010         p->utime = cputime_zero;
1011         p->stime = cputime_zero;
1012         p->sched_time = 0;
1013         p->rchar = 0;           /* I/O counter: bytes read */
1014         p->wchar = 0;           /* I/O counter: bytes written */
1015         p->syscr = 0;           /* I/O counter: read syscalls */
1016         p->syscw = 0;           /* I/O counter: write syscalls */
1017         acct_clear_integrals(p);
1018
1019         p->it_virt_expires = cputime_zero;
1020         p->it_prof_expires = cputime_zero;
1021         p->it_sched_expires = 0;
1022         INIT_LIST_HEAD(&p->cpu_timers[0]);
1023         INIT_LIST_HEAD(&p->cpu_timers[1]);
1024         INIT_LIST_HEAD(&p->cpu_timers[2]);
1025
1026         p->lock_depth = -1;             /* -1 = no lock */
1027         do_posix_clock_monotonic_gettime(&p->start_time);
1028         p->security = NULL;
1029         p->io_context = NULL;
1030         p->io_wait = NULL;
1031         p->audit_context = NULL;
1032         cpuset_fork(p);
1033 #ifdef CONFIG_NUMA
1034         p->mempolicy = mpol_copy(p->mempolicy);
1035         if (IS_ERR(p->mempolicy)) {
1036                 retval = PTR_ERR(p->mempolicy);
1037                 p->mempolicy = NULL;
1038                 goto bad_fork_cleanup_cpuset;
1039         }
1040         mpol_fix_fork_child_flag(p);
1041 #endif
1042
1043 #ifdef CONFIG_DEBUG_MUTEXES
1044         p->blocked_on = NULL; /* not blocked yet */
1045 #endif
1046
1047         p->tgid = p->pid;
1048         if (clone_flags & CLONE_THREAD)
1049                 p->tgid = current->tgid;
1050
1051         if ((retval = security_task_alloc(p)))
1052                 goto bad_fork_cleanup_policy;
1053         if ((retval = audit_alloc(p)))
1054                 goto bad_fork_cleanup_security;
1055         /* copy all the process information */
1056         if ((retval = copy_semundo(clone_flags, p)))
1057                 goto bad_fork_cleanup_audit;
1058         if ((retval = copy_files(clone_flags, p)))
1059                 goto bad_fork_cleanup_semundo;
1060         if ((retval = copy_fs(clone_flags, p)))
1061                 goto bad_fork_cleanup_files;
1062         if ((retval = copy_sighand(clone_flags, p)))
1063                 goto bad_fork_cleanup_fs;
1064         if ((retval = copy_signal(clone_flags, p)))
1065                 goto bad_fork_cleanup_sighand;
1066         if ((retval = copy_mm(clone_flags, p)))
1067                 goto bad_fork_cleanup_signal;
1068         if ((retval = copy_keys(clone_flags, p)))
1069                 goto bad_fork_cleanup_mm;
1070         if ((retval = copy_namespace(clone_flags, p)))
1071                 goto bad_fork_cleanup_keys;
1072         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1073         if (retval)
1074                 goto bad_fork_cleanup_namespace;
1075
1076         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1077         /*
1078          * Clear TID on mm_release()?
1079          */
1080         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1081         p->robust_list = NULL;
1082 #ifdef CONFIG_COMPAT
1083         p->compat_robust_list = NULL;
1084 #endif
1085         /*
1086          * sigaltstack should be cleared when sharing the same VM
1087          */
1088         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1089                 p->sas_ss_sp = p->sas_ss_size = 0;
1090
1091         /*
1092          * Syscall tracing should be turned off in the child regardless
1093          * of CLONE_PTRACE.
1094          */
1095         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1096 #ifdef TIF_SYSCALL_EMU
1097         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1098 #endif
1099
1100         /* Our parent execution domain becomes current domain
1101            These must match for thread signalling to apply */
1102
1103         p->parent_exec_id = p->self_exec_id;
1104
1105         /* ok, now we should be set up.. */
1106         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1107         p->pdeath_signal = 0;
1108         p->exit_state = 0;
1109
1110         /*
1111          * Ok, make it visible to the rest of the system.
1112          * We dont wake it up yet.
1113          */
1114         p->group_leader = p;
1115         INIT_LIST_HEAD(&p->thread_group);
1116         INIT_LIST_HEAD(&p->ptrace_children);
1117         INIT_LIST_HEAD(&p->ptrace_list);
1118
1119         /* Perform scheduler related setup. Assign this task to a CPU. */
1120         sched_fork(p, clone_flags);
1121
1122         /* Need tasklist lock for parent etc handling! */
1123         write_lock_irq(&tasklist_lock);
1124
1125         /*
1126          * The task hasn't been attached yet, so its cpus_allowed mask will
1127          * not be changed, nor will its assigned CPU.
1128          *
1129          * The cpus_allowed mask of the parent may have changed after it was
1130          * copied first time - so re-copy it here, then check the child's CPU
1131          * to ensure it is on a valid CPU (and if not, just force it back to
1132          * parent's CPU). This avoids alot of nasty races.
1133          */
1134         p->cpus_allowed = current->cpus_allowed;
1135         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1136                         !cpu_online(task_cpu(p))))
1137                 set_task_cpu(p, smp_processor_id());
1138
1139         /* CLONE_PARENT re-uses the old parent */
1140         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1141                 p->real_parent = current->real_parent;
1142         else
1143                 p->real_parent = current;
1144         p->parent = p->real_parent;
1145
1146         spin_lock(&current->sighand->siglock);
1147
1148         /*
1149          * Process group and session signals need to be delivered to just the
1150          * parent before the fork or both the parent and the child after the
1151          * fork. Restart if a signal comes in before we add the new process to
1152          * it's process group.
1153          * A fatal signal pending means that current will exit, so the new
1154          * thread can't slip out of an OOM kill (or normal SIGKILL).
1155          */
1156         recalc_sigpending();
1157         if (signal_pending(current)) {
1158                 spin_unlock(&current->sighand->siglock);
1159                 write_unlock_irq(&tasklist_lock);
1160                 retval = -ERESTARTNOINTR;
1161                 goto bad_fork_cleanup_namespace;
1162         }
1163
1164         if (clone_flags & CLONE_THREAD) {
1165                 /*
1166                  * Important: if an exit-all has been started then
1167                  * do not create this new thread - the whole thread
1168                  * group is supposed to exit anyway.
1169                  */
1170                 if (current->signal->flags & SIGNAL_GROUP_EXIT) {
1171                         spin_unlock(&current->sighand->siglock);
1172                         write_unlock_irq(&tasklist_lock);
1173                         retval = -EAGAIN;
1174                         goto bad_fork_cleanup_namespace;
1175                 }
1176
1177                 p->group_leader = current->group_leader;
1178                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1179
1180                 if (!cputime_eq(current->signal->it_virt_expires,
1181                                 cputime_zero) ||
1182                     !cputime_eq(current->signal->it_prof_expires,
1183                                 cputime_zero) ||
1184                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1185                     !list_empty(&current->signal->cpu_timers[0]) ||
1186                     !list_empty(&current->signal->cpu_timers[1]) ||
1187                     !list_empty(&current->signal->cpu_timers[2])) {
1188                         /*
1189                          * Have child wake up on its first tick to check
1190                          * for process CPU timers.
1191                          */
1192                         p->it_prof_expires = jiffies_to_cputime(1);
1193                 }
1194         }
1195
1196         /*
1197          * inherit ioprio
1198          */
1199         p->ioprio = current->ioprio;
1200
1201         if (likely(p->pid)) {
1202                 add_parent(p);
1203                 if (unlikely(p->ptrace & PT_PTRACED))
1204                         __ptrace_link(p, current->parent);
1205
1206                 if (thread_group_leader(p)) {
1207                         p->signal->tty = current->signal->tty;
1208                         p->signal->pgrp = process_group(current);
1209                         p->signal->session = current->signal->session;
1210                         attach_pid(p, PIDTYPE_PGID, process_group(p));
1211                         attach_pid(p, PIDTYPE_SID, p->signal->session);
1212
1213                         list_add_tail(&p->tasks, &init_task.tasks);
1214                         __get_cpu_var(process_counts)++;
1215                 }
1216                 attach_pid(p, PIDTYPE_PID, p->pid);
1217                 nr_threads++;
1218         }
1219
1220         total_forks++;
1221         spin_unlock(&current->sighand->siglock);
1222         write_unlock_irq(&tasklist_lock);
1223         proc_fork_connector(p);
1224         return p;
1225
1226 bad_fork_cleanup_namespace:
1227         exit_namespace(p);
1228 bad_fork_cleanup_keys:
1229         exit_keys(p);
1230 bad_fork_cleanup_mm:
1231         if (p->mm)
1232                 mmput(p->mm);
1233 bad_fork_cleanup_signal:
1234         cleanup_signal(p);
1235 bad_fork_cleanup_sighand:
1236         __cleanup_sighand(p->sighand);
1237 bad_fork_cleanup_fs:
1238         exit_fs(p); /* blocking */
1239 bad_fork_cleanup_files:
1240         exit_files(p); /* blocking */
1241 bad_fork_cleanup_semundo:
1242         exit_sem(p);
1243 bad_fork_cleanup_audit:
1244         audit_free(p);
1245 bad_fork_cleanup_security:
1246         security_task_free(p);
1247 bad_fork_cleanup_policy:
1248 #ifdef CONFIG_NUMA
1249         mpol_free(p->mempolicy);
1250 bad_fork_cleanup_cpuset:
1251 #endif
1252         cpuset_exit(p);
1253 bad_fork_cleanup:
1254         if (p->binfmt)
1255                 module_put(p->binfmt->module);
1256 bad_fork_cleanup_put_domain:
1257         module_put(task_thread_info(p)->exec_domain->module);
1258 bad_fork_cleanup_count:
1259         put_group_info(p->group_info);
1260         atomic_dec(&p->user->processes);
1261         free_uid(p->user);
1262 bad_fork_free:
1263         free_task(p);
1264 fork_out:
1265         return ERR_PTR(retval);
1266 }
1267
1268 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1269 {
1270         memset(regs, 0, sizeof(struct pt_regs));
1271         return regs;
1272 }
1273
1274 task_t * __devinit fork_idle(int cpu)
1275 {
1276         task_t *task;
1277         struct pt_regs regs;
1278
1279         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1280         if (!task)
1281                 return ERR_PTR(-ENOMEM);
1282         init_idle(task, cpu);
1283
1284         return task;
1285 }
1286
1287 static inline int fork_traceflag (unsigned clone_flags)
1288 {
1289         if (clone_flags & CLONE_UNTRACED)
1290                 return 0;
1291         else if (clone_flags & CLONE_VFORK) {
1292                 if (current->ptrace & PT_TRACE_VFORK)
1293                         return PTRACE_EVENT_VFORK;
1294         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1295                 if (current->ptrace & PT_TRACE_CLONE)
1296                         return PTRACE_EVENT_CLONE;
1297         } else if (current->ptrace & PT_TRACE_FORK)
1298                 return PTRACE_EVENT_FORK;
1299
1300         return 0;
1301 }
1302
1303 /*
1304  *  Ok, this is the main fork-routine.
1305  *
1306  * It copies the process, and if successful kick-starts
1307  * it and waits for it to finish using the VM if required.
1308  */
1309 long do_fork(unsigned long clone_flags,
1310               unsigned long stack_start,
1311               struct pt_regs *regs,
1312               unsigned long stack_size,
1313               int __user *parent_tidptr,
1314               int __user *child_tidptr)
1315 {
1316         struct task_struct *p;
1317         int trace = 0;
1318         struct pid *pid = alloc_pid();
1319         long nr;
1320
1321         if (!pid)
1322                 return -EAGAIN;
1323         nr = pid->nr;
1324         if (unlikely(current->ptrace)) {
1325                 trace = fork_traceflag (clone_flags);
1326                 if (trace)
1327                         clone_flags |= CLONE_PTRACE;
1328         }
1329
1330         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, nr);
1331         /*
1332          * Do this prior waking up the new thread - the thread pointer
1333          * might get invalid after that point, if the thread exits quickly.
1334          */
1335         if (!IS_ERR(p)) {
1336                 struct completion vfork;
1337
1338                 if (clone_flags & CLONE_VFORK) {
1339                         p->vfork_done = &vfork;
1340                         init_completion(&vfork);
1341                 }
1342
1343                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1344                         /*
1345                          * We'll start up with an immediate SIGSTOP.
1346                          */
1347                         sigaddset(&p->pending.signal, SIGSTOP);
1348                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1349                 }
1350
1351                 if (!(clone_flags & CLONE_STOPPED))
1352                         wake_up_new_task(p, clone_flags);
1353                 else
1354                         p->state = TASK_STOPPED;
1355
1356                 if (unlikely (trace)) {
1357                         current->ptrace_message = nr;
1358                         ptrace_notify ((trace << 8) | SIGTRAP);
1359                 }
1360
1361                 if (clone_flags & CLONE_VFORK) {
1362                         wait_for_completion(&vfork);
1363                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1364                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1365                 }
1366         } else {
1367                 free_pid(pid);
1368                 nr = PTR_ERR(p);
1369         }
1370         return nr;
1371 }
1372
1373 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1374 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1375 #endif
1376
1377 static void sighand_ctor(void *data, kmem_cache_t *cachep, unsigned long flags)
1378 {
1379         struct sighand_struct *sighand = data;
1380
1381         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
1382                                         SLAB_CTOR_CONSTRUCTOR)
1383                 spin_lock_init(&sighand->siglock);
1384 }
1385
1386 void __init proc_caches_init(void)
1387 {
1388         sighand_cachep = kmem_cache_create("sighand_cache",
1389                         sizeof(struct sighand_struct), 0,
1390                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1391                         sighand_ctor, NULL);
1392         signal_cachep = kmem_cache_create("signal_cache",
1393                         sizeof(struct signal_struct), 0,
1394                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1395         files_cachep = kmem_cache_create("files_cache",
1396                         sizeof(struct files_struct), 0,
1397                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1398         fs_cachep = kmem_cache_create("fs_cache",
1399                         sizeof(struct fs_struct), 0,
1400                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1401         vm_area_cachep = kmem_cache_create("vm_area_struct",
1402                         sizeof(struct vm_area_struct), 0,
1403                         SLAB_PANIC, NULL, NULL);
1404         mm_cachep = kmem_cache_create("mm_struct",
1405                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1406                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1407 }
1408
1409
1410 /*
1411  * Check constraints on flags passed to the unshare system call and
1412  * force unsharing of additional process context as appropriate.
1413  */
1414 static inline void check_unshare_flags(unsigned long *flags_ptr)
1415 {
1416         /*
1417          * If unsharing a thread from a thread group, must also
1418          * unshare vm.
1419          */
1420         if (*flags_ptr & CLONE_THREAD)
1421                 *flags_ptr |= CLONE_VM;
1422
1423         /*
1424          * If unsharing vm, must also unshare signal handlers.
1425          */
1426         if (*flags_ptr & CLONE_VM)
1427                 *flags_ptr |= CLONE_SIGHAND;
1428
1429         /*
1430          * If unsharing signal handlers and the task was created
1431          * using CLONE_THREAD, then must unshare the thread
1432          */
1433         if ((*flags_ptr & CLONE_SIGHAND) &&
1434             (atomic_read(&current->signal->count) > 1))
1435                 *flags_ptr |= CLONE_THREAD;
1436
1437         /*
1438          * If unsharing namespace, must also unshare filesystem information.
1439          */
1440         if (*flags_ptr & CLONE_NEWNS)
1441                 *flags_ptr |= CLONE_FS;
1442 }
1443
1444 /*
1445  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1446  */
1447 static int unshare_thread(unsigned long unshare_flags)
1448 {
1449         if (unshare_flags & CLONE_THREAD)
1450                 return -EINVAL;
1451
1452         return 0;
1453 }
1454
1455 /*
1456  * Unshare the filesystem structure if it is being shared
1457  */
1458 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1459 {
1460         struct fs_struct *fs = current->fs;
1461
1462         if ((unshare_flags & CLONE_FS) &&
1463             (fs && atomic_read(&fs->count) > 1)) {
1464                 *new_fsp = __copy_fs_struct(current->fs);
1465                 if (!*new_fsp)
1466                         return -ENOMEM;
1467         }
1468
1469         return 0;
1470 }
1471
1472 /*
1473  * Unshare the namespace structure if it is being shared
1474  */
1475 static int unshare_namespace(unsigned long unshare_flags, struct namespace **new_nsp, struct fs_struct *new_fs)
1476 {
1477         struct namespace *ns = current->namespace;
1478
1479         if ((unshare_flags & CLONE_NEWNS) &&
1480             (ns && atomic_read(&ns->count) > 1)) {
1481                 if (!capable(CAP_SYS_ADMIN))
1482                         return -EPERM;
1483
1484                 *new_nsp = dup_namespace(current, new_fs ? new_fs : current->fs);
1485                 if (!*new_nsp)
1486                         return -ENOMEM;
1487         }
1488
1489         return 0;
1490 }
1491
1492 /*
1493  * Unsharing of sighand for tasks created with CLONE_SIGHAND is not
1494  * supported yet
1495  */
1496 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1497 {
1498         struct sighand_struct *sigh = current->sighand;
1499
1500         if ((unshare_flags & CLONE_SIGHAND) &&
1501             (sigh && atomic_read(&sigh->count) > 1))
1502                 return -EINVAL;
1503         else
1504                 return 0;
1505 }
1506
1507 /*
1508  * Unshare vm if it is being shared
1509  */
1510 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1511 {
1512         struct mm_struct *mm = current->mm;
1513
1514         if ((unshare_flags & CLONE_VM) &&
1515             (mm && atomic_read(&mm->mm_users) > 1)) {
1516                 return -EINVAL;
1517         }
1518
1519         return 0;
1520 }
1521
1522 /*
1523  * Unshare file descriptor table if it is being shared
1524  */
1525 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1526 {
1527         struct files_struct *fd = current->files;
1528         int error = 0;
1529
1530         if ((unshare_flags & CLONE_FILES) &&
1531             (fd && atomic_read(&fd->count) > 1)) {
1532                 *new_fdp = dup_fd(fd, &error);
1533                 if (!*new_fdp)
1534                         return error;
1535         }
1536
1537         return 0;
1538 }
1539
1540 /*
1541  * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1542  * supported yet
1543  */
1544 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1545 {
1546         if (unshare_flags & CLONE_SYSVSEM)
1547                 return -EINVAL;
1548
1549         return 0;
1550 }
1551
1552 /*
1553  * unshare allows a process to 'unshare' part of the process
1554  * context which was originally shared using clone.  copy_*
1555  * functions used by do_fork() cannot be used here directly
1556  * because they modify an inactive task_struct that is being
1557  * constructed. Here we are modifying the current, active,
1558  * task_struct.
1559  */
1560 asmlinkage long sys_unshare(unsigned long unshare_flags)
1561 {
1562         int err = 0;
1563         struct fs_struct *fs, *new_fs = NULL;
1564         struct namespace *ns, *new_ns = NULL;
1565         struct sighand_struct *sigh, *new_sigh = NULL;
1566         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1567         struct files_struct *fd, *new_fd = NULL;
1568         struct sem_undo_list *new_ulist = NULL;
1569
1570         check_unshare_flags(&unshare_flags);
1571
1572         /* Return -EINVAL for all unsupported flags */
1573         err = -EINVAL;
1574         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1575                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM))
1576                 goto bad_unshare_out;
1577
1578         if ((err = unshare_thread(unshare_flags)))
1579                 goto bad_unshare_out;
1580         if ((err = unshare_fs(unshare_flags, &new_fs)))
1581                 goto bad_unshare_cleanup_thread;
1582         if ((err = unshare_namespace(unshare_flags, &new_ns, new_fs)))
1583                 goto bad_unshare_cleanup_fs;
1584         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1585                 goto bad_unshare_cleanup_ns;
1586         if ((err = unshare_vm(unshare_flags, &new_mm)))
1587                 goto bad_unshare_cleanup_sigh;
1588         if ((err = unshare_fd(unshare_flags, &new_fd)))
1589                 goto bad_unshare_cleanup_vm;
1590         if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1591                 goto bad_unshare_cleanup_fd;
1592
1593         if (new_fs || new_ns || new_sigh || new_mm || new_fd || new_ulist) {
1594
1595                 task_lock(current);
1596
1597                 if (new_fs) {
1598                         fs = current->fs;
1599                         current->fs = new_fs;
1600                         new_fs = fs;
1601                 }
1602
1603                 if (new_ns) {
1604                         ns = current->namespace;
1605                         current->namespace = new_ns;
1606                         new_ns = ns;
1607                 }
1608
1609                 if (new_sigh) {
1610                         sigh = current->sighand;
1611                         rcu_assign_pointer(current->sighand, new_sigh);
1612                         new_sigh = sigh;
1613                 }
1614
1615                 if (new_mm) {
1616                         mm = current->mm;
1617                         active_mm = current->active_mm;
1618                         current->mm = new_mm;
1619                         current->active_mm = new_mm;
1620                         activate_mm(active_mm, new_mm);
1621                         new_mm = mm;
1622                 }
1623
1624                 if (new_fd) {
1625                         fd = current->files;
1626                         current->files = new_fd;
1627                         new_fd = fd;
1628                 }
1629
1630                 task_unlock(current);
1631         }
1632
1633 bad_unshare_cleanup_fd:
1634         if (new_fd)
1635                 put_files_struct(new_fd);
1636
1637 bad_unshare_cleanup_vm:
1638         if (new_mm)
1639                 mmput(new_mm);
1640
1641 bad_unshare_cleanup_sigh:
1642         if (new_sigh)
1643                 if (atomic_dec_and_test(&new_sigh->count))
1644                         kmem_cache_free(sighand_cachep, new_sigh);
1645
1646 bad_unshare_cleanup_ns:
1647         if (new_ns)
1648                 put_namespace(new_ns);
1649
1650 bad_unshare_cleanup_fs:
1651         if (new_fs)
1652                 put_fs_struct(new_fs);
1653
1654 bad_unshare_cleanup_thread:
1655 bad_unshare_out:
1656         return err;
1657 }