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/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/smp_lock.h>
  18 #include <linux/module.h>
  19 #include <linux/vmalloc.h>
  20 #include <linux/completion.h>
  21 #include <linux/namespace.h>
  22 #include <linux/personality.h>
  23 #include <linux/mempolicy.h>
  24 #include <linux/sem.h>
  25 #include <linux/file.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/fs.h>
  30 #include <linux/capability.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/rcupdate.h>
  39 #include <linux/ptrace.h>
  40 #include <linux/mount.h>
  41 #include <linux/audit.h>
  42 #include <linux/profile.h>
  43 #include <linux/rmap.h>
  44 #include <linux/acct.h>
  45 #include <linux/cn_proc.h>
  46 #include <linux/delayacct.h>
  47 #include <linux/taskstats_kern.h>
  48 #include <linux/vs_context.h>
  49 #include <linux/vs_network.h>
  50 #include <linux/vs_limit.h>
  51 #include <linux/vs_memory.h>
  52
  53 #include <asm/pgtable.h>
  54 #include <asm/pgalloc.h>
  55 #include <asm/uaccess.h>
  56 #include <asm/mmu_context.h>
  57 #include <asm/cacheflush.h>
  58 #include <asm/tlbflush.h>
  59
  60 /*
  61  * Protected counters by write_lock_irq(&tasklist_lock)
  62  */
  63 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  64 int nr_threads;                 /* The idle threads do not count.. */
  65
  66 int max_threads;                /* tunable limit on nr_threads */
  67
  68 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  69
  70 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  71
  72 int nr_processes(void)
  73 {
  74         int cpu;
  75         int total = 0;
  76
  77         for_each_online_cpu(cpu)
  78                 total += per_cpu(process_counts, cpu);
  79
  80         return total;
  81 }
  82
  83 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  84 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  85 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  86 static kmem_cache_t *task_struct_cachep;
  87 #endif
  88
  89 /* SLAB cache for signal_struct structures (tsk->signal) */
  90 static kmem_cache_t *signal_cachep;
  91
  92 /* SLAB cache for sighand_struct structures (tsk->sighand) */
  93 kmem_cache_t *sighand_cachep;
  94
  95 /* SLAB cache for files_struct structures (tsk->files) */
  96 kmem_cache_t *files_cachep;
  97
  98 /* SLAB cache for fs_struct structures (tsk->fs) */
  99 kmem_cache_t *fs_cachep;
 100
 101 /* SLAB cache for vm_area_struct structures */
 102 kmem_cache_t *vm_area_cachep;
 103
 104 /* SLAB cache for mm_struct structures (tsk->mm) */
 105 static kmem_cache_t *mm_cachep;
 106
 107 void free_task(struct task_struct *tsk)
 108 {
 109         free_thread_info(tsk->thread_info);
 110         rt_mutex_debug_task_free(tsk);
 111         clr_vx_info(&tsk->vx_info);
 112         clr_nx_info(&tsk->nx_info);
 113         free_task_struct(tsk);
 114 }
 115 EXPORT_SYMBOL(free_task);
 116
 117 void __put_task_struct(struct task_struct *tsk)
 118 {
 119         WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
 120         WARN_ON(atomic_read(&tsk->usage));
 121         WARN_ON(tsk == current);
 122
 123         security_task_free(tsk);
 124         free_uid(tsk->user);
 125         put_group_info(tsk->group_info);
 126         delayacct_tsk_free(tsk);
 127
 128         if (!profile_handoff_task(tsk))
 129                 free_task(tsk);
 130 }
 131
 132 void __init fork_init(unsigned long mempages)
 133 {
 134 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 135 #ifndef ARCH_MIN_TASKALIGN
 136 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 137 #endif
 138         /* create a slab on which task_structs can be allocated */
 139         task_struct_cachep =
 140                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 141                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
 142 #endif
 143
 144         /*
 145          * The default maximum number of threads is set to a safe
 146          * value: the thread structures can take up at most half
 147          * of memory.
 148          */
 149         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 150
 151         /*
 152          * we need to allow at least 20 threads to boot a system
 153          */
 154         if(max_threads < 20)
 155                 max_threads = 20;
 156
 157         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 158         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 159         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 160                 init_task.signal->rlim[RLIMIT_NPROC];
 161 }
 162
 163 static struct task_struct *dup_task_struct(struct task_struct *orig)
 164 {
 165         struct task_struct *tsk;
 166         struct thread_info *ti;
 167
 168         prepare_to_copy(orig);
 169
 170         tsk = alloc_task_struct();
 171         if (!tsk)
 172                 return NULL;
 173
 174         ti = alloc_thread_info(tsk);
 175         if (!ti) {
 176                 free_task_struct(tsk);
 177                 return NULL;
 178         }
 179
 180         *tsk = *orig;
 181         tsk->thread_info = ti;
 182         setup_thread_stack(tsk, orig);
 183
 184         /* One for us, one for whoever does the "release_task()" (usually parent) */
 185         atomic_set(&tsk->usage,2);
 186         atomic_set(&tsk->fs_excl, 0);
 187         tsk->btrace_seq = 0;
 188         tsk->splice_pipe = NULL;
 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         /*
 204          * Not linked in yet - no deadlock potential:
 205          */
 206         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 207
 208         mm->locked_vm = 0;
 209         mm->mmap = NULL;
 210         mm->mmap_cache = NULL;
 211         mm->free_area_cache = oldmm->mmap_base;
 212         mm->cached_hole_size = ~0UL;
 213         mm->map_count = 0;
 214         __set_mm_counter(mm, file_rss, 0);
 215         __set_mm_counter(mm, anon_rss, 0);
 216         cpus_clear(mm->cpu_vm_mask);
 217         mm->mm_rb = RB_ROOT;
 218         rb_link = &mm->mm_rb.rb_node;
 219         rb_parent = NULL;
 220         pprev = &mm->mmap;
 221
 222         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 223                 struct file *file;
 224
 225                 if (mpnt->vm_flags & VM_DONTCOPY) {
 226                         long pages = vma_pages(mpnt);
 227                         vx_vmpages_sub(mm, pages);
 228                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 229                                                                 -pages);
 230                         continue;
 231                 }
 232                 charge = 0;
 233                 if (mpnt->vm_flags & VM_ACCOUNT) {
 234                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 235                         if (security_vm_enough_memory(len))
 236                                 goto fail_nomem;
 237                         charge = len;
 238                 }
 239                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 240                 if (!tmp)
 241                         goto fail_nomem;
 242                 *tmp = *mpnt;
 243                 pol = mpol_copy(vma_policy(mpnt));
 244                 retval = PTR_ERR(pol);
 245                 if (IS_ERR(pol))
 246                         goto fail_nomem_policy;
 247                 vma_set_policy(tmp, pol);
 248                 tmp->vm_flags &= ~VM_LOCKED;
 249                 tmp->vm_mm = mm;
 250                 tmp->vm_next = NULL;
 251                 anon_vma_link(tmp);
 252                 file = tmp->vm_file;
 253                 if (file) {
 254                         struct inode *inode = file->f_dentry->d_inode;
 255                         get_file(file);
 256                         if (tmp->vm_flags & VM_DENYWRITE)
 257                                 atomic_dec(&inode->i_writecount);
 258
 259                         /* insert tmp into the share list, just after mpnt */
 260                         spin_lock(&file->f_mapping->i_mmap_lock);
 261                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 262                         flush_dcache_mmap_lock(file->f_mapping);
 263                         vma_prio_tree_add(tmp, mpnt);
 264                         flush_dcache_mmap_unlock(file->f_mapping);
 265                         spin_unlock(&file->f_mapping->i_mmap_lock);
 266                 }
 267
 268                 /*
 269                  * Link in the new vma and copy the page table entries.
 270                  */
 271                 *pprev = tmp;
 272                 pprev = &tmp->vm_next;
 273
 274                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 275                 rb_link = &tmp->vm_rb.rb_right;
 276                 rb_parent = &tmp->vm_rb;
 277
 278                 mm->map_count++;
 279                 retval = copy_page_range(mm, oldmm, mpnt);
 280
 281                 if (tmp->vm_ops && tmp->vm_ops->open)
 282                         tmp->vm_ops->open(tmp);
 283
 284                 if (retval)
 285                         goto out;
 286         }
 287 #ifdef arch_dup_mmap
 288         arch_dup_mmap(mm, oldmm);
 289 #endif
 290         retval = 0;
 291 out:
 292         up_write(&mm->mmap_sem);
 293         flush_tlb_mm(oldmm);
 294         up_write(&oldmm->mmap_sem);
 295         return retval;
 296 fail_nomem_policy:
 297         kmem_cache_free(vm_area_cachep, tmp);
 298 fail_nomem:
 299         retval = -ENOMEM;
 300         vm_unacct_memory(charge);
 301         goto out;
 302 }
 303
 304 static inline int mm_alloc_pgd(struct mm_struct * mm)
 305 {
 306         mm->pgd = pgd_alloc(mm);
 307         if (unlikely(!mm->pgd))
 308                 return -ENOMEM;
 309         return 0;
 310 }
 311
 312 static inline void mm_free_pgd(struct mm_struct * mm)
 313 {
 314         pgd_free(mm->pgd);
 315 }
 316 #else
 317 #define dup_mmap(mm, oldmm)     (0)
 318 #define mm_alloc_pgd(mm)        (0)
 319 #define mm_free_pgd(mm)
 320 #endif /* CONFIG_MMU */
 321
 322  __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 323
 324 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
 325 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 326
 327 #include <linux/init_task.h>
 328
 329 static struct mm_struct * mm_init(struct mm_struct * mm)
 330 {
 331         atomic_set(&mm->mm_users, 1);
 332         atomic_set(&mm->mm_count, 1);
 333         init_rwsem(&mm->mmap_sem);
 334         INIT_LIST_HEAD(&mm->mmlist);
 335         mm->core_waiters = 0;
 336         mm->nr_ptes = 0;
 337         spin_lock_init(&mm->page_table_lock);
 338         rwlock_init(&mm->ioctx_list_lock);
 339         mm->ioctx_list = NULL;
 340         mm->free_area_cache = TASK_UNMAPPED_BASE;
 341         mm->cached_hole_size = ~0UL;
 342
 343         if (likely(!mm_alloc_pgd(mm))) {
 344                 mm->def_flags = 0;
 345                 set_vx_info(&mm->mm_vx_info, current->vx_info);
 346                 return mm;
 347         }
 348         free_mm(mm);
 349         return NULL;
 350 }
 351
 352 /*
 353  * Allocate and initialize an mm_struct.
 354  */
 355 struct mm_struct * mm_alloc(void)
 356 {
 357         struct mm_struct * mm;
 358
 359         mm = allocate_mm();
 360         if (mm) {
 361                 memset(mm, 0, sizeof(*mm));
 362                 mm = mm_init(mm);
 363         }
 364         return mm;
 365 }
 366
 367 /*
 368  * Called when the last reference to the mm
 369  * is dropped: either by a lazy thread or by
 370  * mmput. Free the page directory and the mm.
 371  */
 372 void fastcall __mmdrop(struct mm_struct *mm)
 373 {
 374         BUG_ON(mm == &init_mm);
 375         mm_free_pgd(mm);
 376         destroy_context(mm);
 377         clr_vx_info(&mm->mm_vx_info);
 378         free_mm(mm);
 379 }
 380
 381 /*
 382  * Decrement the use count and release all resources for an mm.
 383  */
 384 void mmput(struct mm_struct *mm)
 385 {
 386         might_sleep();
 387
 388         if (atomic_dec_and_test(&mm->mm_users)) {
 389                 exit_aio(mm);
 390                 exit_mmap(mm);
 391                 if (!list_empty(&mm->mmlist)) {
 392                         spin_lock(&mmlist_lock);
 393                         list_del(&mm->mmlist);
 394                         spin_unlock(&mmlist_lock);
 395                 }
 396                 put_swap_token(mm);
 397                 mmdrop(mm);
 398         }
 399 }
 400 EXPORT_SYMBOL_GPL(mmput);
 401
 402 /**
 403  * get_task_mm - acquire a reference to the task's mm
 404  *
 405  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
 406  * this kernel workthread has transiently adopted a user mm with use_mm,
 407  * to do its AIO) is not set and if so returns a reference to it, after
 408  * bumping up the use count.  User must release the mm via mmput()
 409  * after use.  Typically used by /proc and ptrace.
 410  */
 411 struct mm_struct *get_task_mm(struct task_struct *task)
 412 {
 413         struct mm_struct *mm;
 414
 415         task_lock(task);
 416         mm = task->mm;
 417         if (mm) {
 418                 if (task->flags & PF_BORROWED_MM)
 419                         mm = NULL;
 420                 else
 421                         atomic_inc(&mm->mm_users);
 422         }
 423         task_unlock(task);
 424         return mm;
 425 }
 426 EXPORT_SYMBOL_GPL(get_task_mm);
 427
 428 /* Please note the differences between mmput and mm_release.
 429  * mmput is called whenever we stop holding onto a mm_struct,
 430  * error success whatever.
 431  *
 432  * mm_release is called after a mm_struct has been removed
 433  * from the current process.
 434  *
 435  * This difference is important for error handling, when we
 436  * only half set up a mm_struct for a new process and need to restore
 437  * the old one.  Because we mmput the new mm_struct before
 438  * restoring the old one. . .
 439  * Eric Biederman 10 January 1998
 440  */
 441 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 442 {
 443         struct completion *vfork_done = tsk->vfork_done;
 444
 445         /* Get rid of any cached register state */
 446         deactivate_mm(tsk, mm);
 447
 448         /* notify parent sleeping on vfork() */
 449         if (vfork_done) {
 450                 tsk->vfork_done = NULL;
 451                 complete(vfork_done);
 452         }
 453         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
 454                 u32 __user * tidptr = tsk->clear_child_tid;
 455                 tsk->clear_child_tid = NULL;
 456
 457                 /*
 458                  * We don't check the error code - if userspace has
 459                  * not set up a proper pointer then tough luck.
 460                  */
 461                 put_user(0, tidptr);
 462                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 463         }
 464 }
 465
 466 /*
 467  * Allocate a new mm structure and copy contents from the
 468  * mm structure of the passed in task structure.
 469  */
 470 static struct mm_struct *dup_mm(struct task_struct *tsk)
 471 {
 472         struct mm_struct *mm, *oldmm = current->mm;
 473         int err;
 474
 475         if (!oldmm)
 476                 return NULL;
 477
 478         mm = allocate_mm();
 479         if (!mm)
 480                 goto fail_nomem;
 481
 482         memcpy(mm, oldmm, sizeof(*mm));
 483         mm->mm_vx_info = NULL;
 484
 485         if (!mm_init(mm))
 486                 goto fail_nomem;
 487
 488         if (init_new_context(tsk, mm))
 489                 goto fail_nocontext;
 490
 491         err = dup_mmap(mm, oldmm);
 492         if (err)
 493                 goto free_pt;
 494
 495         mm->hiwater_rss = get_mm_rss(mm);
 496         mm->hiwater_vm = mm->total_vm;
 497
 498         return mm;
 499
 500 free_pt:
 501         mmput(mm);
 502
 503 fail_nomem:
 504         return NULL;
 505
 506 fail_nocontext:
 507         /*
 508          * If init_new_context() failed, we cannot use mmput() to free the mm
 509          * because it calls destroy_context()
 510          */
 511         clr_vx_info(&mm->mm_vx_info);
 512         mm_free_pgd(mm);
 513         free_mm(mm);
 514         return NULL;
 515 }
 516
 517 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 518 {
 519         struct mm_struct * mm, *oldmm;
 520         int retval;
 521
 522         tsk->min_flt = tsk->maj_flt = 0;
 523         tsk->nvcsw = tsk->nivcsw = 0;
 524
 525         tsk->mm = NULL;
 526         tsk->active_mm = NULL;
 527
 528         /*
 529          * Are we cloning a kernel thread?
 530          *
 531          * We need to steal a active VM for that..
 532          */
 533         oldmm = current->mm;
 534         if (!oldmm)
 535                 return 0;
 536
 537         if (clone_flags & CLONE_VM) {
 538                 atomic_inc(&oldmm->mm_users);
 539                 mm = oldmm;
 540                 goto good_mm;
 541         }
 542
 543         retval = -ENOMEM;
 544         mm = dup_mm(tsk);
 545         if (!mm)
 546                 goto fail_nomem;
 547
 548 good_mm:
 549         tsk->mm = mm;
 550         tsk->active_mm = mm;
 551         return 0;
 552
 553 fail_nomem:
 554         return retval;
 555 }
 556
 557 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 558 {
 559         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 560         /* We don't need to lock fs - think why ;-) */
 561         if (fs) {
 562                 atomic_set(&fs->count, 1);
 563                 rwlock_init(&fs->lock);
 564                 fs->umask = old->umask;
 565                 read_lock(&old->lock);
 566                 fs->rootmnt = mntget(old->rootmnt);
 567                 fs->root = dget(old->root);
 568                 fs->pwdmnt = mntget(old->pwdmnt);
 569                 fs->pwd = dget(old->pwd);
 570                 if (old->altroot) {
 571                         fs->altrootmnt = mntget(old->altrootmnt);
 572                         fs->altroot = dget(old->altroot);
 573                 } else {
 574                         fs->altrootmnt = NULL;
 575                         fs->altroot = NULL;
 576                 }
 577                 read_unlock(&old->lock);
 578         }
 579         return fs;
 580 }
 581
 582 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 583 {
 584         return __copy_fs_struct(old);
 585 }
 586
 587 EXPORT_SYMBOL_GPL(copy_fs_struct);
 588
 589 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
 590 {
 591         if (clone_flags & CLONE_FS) {
 592                 atomic_inc(&current->fs->count);
 593                 return 0;
 594         }
 595         tsk->fs = __copy_fs_struct(current->fs);
 596         if (!tsk->fs)
 597                 return -ENOMEM;
 598         return 0;
 599 }
 600
 601 static int count_open_files(struct fdtable *fdt)
 602 {
 603         int size = fdt->max_fdset;
 604         int i;
 605
 606         /* Find the last open fd */
 607         for (i = size/(8*sizeof(long)); i > 0; ) {
 608                 if (fdt->open_fds->fds_bits[--i])
 609                         break;
 610         }
 611         i = (i+1) * 8 * sizeof(long);
 612         return i;
 613 }
 614
 615 static struct files_struct *alloc_files(void)
 616 {
 617         struct files_struct *newf;
 618         struct fdtable *fdt;
 619
 620         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
 621         if (!newf)
 622                 goto out;
 623
 624         atomic_set(&newf->count, 1);
 625
 626         spin_lock_init(&newf->file_lock);
 627         newf->next_fd = 0;
 628         fdt = &newf->fdtab;
 629         fdt->max_fds = NR_OPEN_DEFAULT;
 630         fdt->max_fdset = EMBEDDED_FD_SET_SIZE;
 631         fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
 632         fdt->open_fds = (fd_set *)&newf->open_fds_init;
 633         fdt->fd = &newf->fd_array[0];
 634         INIT_RCU_HEAD(&fdt->rcu);
 635         fdt->free_files = NULL;
 636         fdt->next = NULL;
 637         rcu_assign_pointer(newf->fdt, fdt);
 638 out:
 639         return newf;
 640 }
 641
 642 /*
 643  * Allocate a new files structure and copy contents from the
 644  * passed in files structure.
 645  * errorp will be valid only when the returned files_struct is NULL.
 646  */
 647 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
 648 {
 649         struct files_struct *newf;
 650         struct file **old_fds, **new_fds;
 651         int open_files, size, i, expand;
 652         struct fdtable *old_fdt, *new_fdt;
 653
 654         *errorp = -ENOMEM;
 655         newf = alloc_files();
 656         if (!newf)
 657                 goto out;
 658
 659         spin_lock(&oldf->file_lock);
 660         old_fdt = files_fdtable(oldf);
 661         new_fdt = files_fdtable(newf);
 662         size = old_fdt->max_fdset;
 663         open_files = count_open_files(old_fdt);
 664         expand = 0;
 665
 666         /*
 667          * Check whether we need to allocate a larger fd array or fd set.
 668          * Note: we're not a clone task, so the open count won't  change.
 669          */
 670         if (open_files > new_fdt->max_fdset) {
 671                 new_fdt->max_fdset = 0;
 672                 expand = 1;
 673         }
 674         if (open_files > new_fdt->max_fds) {
 675                 new_fdt->max_fds = 0;
 676                 expand = 1;
 677         }
 678
 679         /* if the old fdset gets grown now, we'll only copy up to "size" fds */
 680         if (expand) {
 681                 spin_unlock(&oldf->file_lock);
 682                 spin_lock(&newf->file_lock);
 683                 *errorp = expand_files(newf, open_files-1);
 684                 spin_unlock(&newf->file_lock);
 685                 if (*errorp < 0)
 686                         goto out_release;
 687                 new_fdt = files_fdtable(newf);
 688                 /*
 689                  * Reacquire the oldf lock and a pointer to its fd table
 690                  * who knows it may have a new bigger fd table. We need
 691                  * the latest pointer.
 692                  */
 693                 spin_lock(&oldf->file_lock);
 694                 old_fdt = files_fdtable(oldf);
 695         }
 696
 697         old_fds = old_fdt->fd;
 698         new_fds = new_fdt->fd;
 699
 700         memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
 701         memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
 702
 703         for (i = open_files; i != 0; i--) {
 704                 struct file *f = *old_fds++;
 705                 if (f) {
 706                         get_file(f);
 707                         /* FIXME: sum it first for check and performance */
 708                         vx_openfd_inc(open_files - i);
 709                 } else {
 710                         /*
 711                          * The fd may be claimed in the fd bitmap but not yet
 712                          * instantiated in the files array if a sibling thread
 713                          * is partway through open().  So make sure that this
 714                          * fd is available to the new process.
 715                          */
 716                         FD_CLR(open_files - i, new_fdt->open_fds);
 717                 }
 718                 rcu_assign_pointer(*new_fds++, f);
 719         }
 720         spin_unlock(&oldf->file_lock);
 721
 722         /* compute the remainder to be cleared */
 723         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
 724
 725         /* This is long word aligned thus could use a optimized version */
 726         memset(new_fds, 0, size);
 727
 728         if (new_fdt->max_fdset > open_files) {
 729                 int left = (new_fdt->max_fdset-open_files)/8;
 730                 int start = open_files / (8 * sizeof(unsigned long));
 731
 732                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
 733                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
 734         }
 735
 736 out:
 737         return newf;
 738
 739 out_release:
 740         free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
 741         free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
 742         free_fd_array(new_fdt->fd, new_fdt->max_fds);
 743         kmem_cache_free(files_cachep, newf);
 744         return NULL;
 745 }
 746
 747 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 748 {
 749         struct files_struct *oldf, *newf;
 750         int error = 0;
 751
 752         /*
 753          * A background process may not have any files ...
 754          */
 755         oldf = current->files;
 756         if (!oldf)
 757                 goto out;
 758
 759         if (clone_flags & CLONE_FILES) {
 760                 atomic_inc(&oldf->count);
 761                 goto out;
 762         }
 763
 764         /*
 765          * Note: we may be using current for both targets (See exec.c)
 766          * This works because we cache current->files (old) as oldf. Don't
 767          * break this.
 768          */
 769         tsk->files = NULL;
 770         newf = dup_fd(oldf, &error);
 771         if (!newf)
 772                 goto out;
 773
 774         tsk->files = newf;
 775         error = 0;
 776 out:
 777         return error;
 778 }
 779
 780 /*
 781  *      Helper to unshare the files of the current task.
 782  *      We don't want to expose copy_files internals to
 783  *      the exec layer of the kernel.
 784  */
 785
 786 int unshare_files(void)
 787 {
 788         struct files_struct *files  = current->files;
 789         int rc;
 790
 791         BUG_ON(!files);
 792
 793         /* This can race but the race causes us to copy when we don't
 794            need to and drop the copy */
 795         if(atomic_read(&files->count) == 1)
 796         {
 797                 atomic_inc(&files->count);
 798                 return 0;
 799         }
 800         rc = copy_files(0, current);
 801         if(rc)
 802                 current->files = files;
 803         return rc;
 804 }
 805
 806 EXPORT_SYMBOL(unshare_files);
 807
 808 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
 809 {
 810         struct sighand_struct *sig;
 811
 812         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 813                 atomic_inc(&current->sighand->count);
 814                 return 0;
 815         }
 816         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 817         rcu_assign_pointer(tsk->sighand, sig);
 818         if (!sig)
 819                 return -ENOMEM;
 820         atomic_set(&sig->count, 1);
 821         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 822         return 0;
 823 }
 824
 825 void __cleanup_sighand(struct sighand_struct *sighand)
 826 {
 827         if (atomic_dec_and_test(&sighand->count))
 828                 kmem_cache_free(sighand_cachep, sighand);
 829 }
 830
 831 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
 832 {
 833         struct signal_struct *sig;
 834         int ret;
 835
 836         if (clone_flags & CLONE_THREAD) {
 837                 atomic_inc(&current->signal->count);
 838                 atomic_inc(&current->signal->live);
 839                 taskstats_tgid_alloc(current->signal);
 840                 return 0;
 841         }
 842         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 843         tsk->signal = sig;
 844         if (!sig)
 845                 return -ENOMEM;
 846
 847         ret = copy_thread_group_keys(tsk);
 848         if (ret < 0) {
 849                 kmem_cache_free(signal_cachep, sig);
 850                 return ret;
 851         }
 852
 853         atomic_set(&sig->count, 1);
 854         atomic_set(&sig->live, 1);
 855         init_waitqueue_head(&sig->wait_chldexit);
 856         sig->flags = 0;
 857         sig->group_exit_code = 0;
 858         sig->group_exit_task = NULL;
 859         sig->group_stop_count = 0;
 860         sig->curr_target = NULL;
 861         init_sigpending(&sig->shared_pending);
 862         INIT_LIST_HEAD(&sig->posix_timers);
 863
 864         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_REL);
 865         sig->it_real_incr.tv64 = 0;
 866         sig->real_timer.function = it_real_fn;
 867         sig->tsk = tsk;
 868
 869         sig->it_virt_expires = cputime_zero;
 870         sig->it_virt_incr = cputime_zero;
 871         sig->it_prof_expires = cputime_zero;
 872         sig->it_prof_incr = cputime_zero;
 873
 874         sig->leader = 0;        /* session leadership doesn't inherit */
 875         sig->tty_old_pgrp = 0;
 876
 877         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 878         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 879         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 880         sig->sched_time = 0;
 881         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 882         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 883         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 884         taskstats_tgid_init(sig);
 885
 886         task_lock(current->group_leader);
 887         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 888         task_unlock(current->group_leader);
 889
 890         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 891                 /*
 892                  * New sole thread in the process gets an expiry time
 893                  * of the whole CPU time limit.
 894                  */
 895                 tsk->it_prof_expires =
 896                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 897         }
 898         acct_init_pacct(&sig->pacct);
 899
 900         return 0;
 901 }
 902
 903 void __cleanup_signal(struct signal_struct *sig)
 904 {
 905         exit_thread_group_keys(sig);
 906         taskstats_tgid_free(sig);
 907         kmem_cache_free(signal_cachep, sig);
 908 }
 909
 910 static inline void cleanup_signal(struct task_struct *tsk)
 911 {
 912         struct signal_struct *sig = tsk->signal;
 913
 914         atomic_dec(&sig->live);
 915
 916         if (atomic_dec_and_test(&sig->count))
 917                 __cleanup_signal(sig);
 918 }
 919
 920 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
 921 {
 922         unsigned long new_flags = p->flags;
 923
 924         new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
 925         new_flags |= PF_FORKNOEXEC;
 926         if (!(clone_flags & CLONE_PTRACE))
 927                 p->ptrace = 0;
 928         p->flags = new_flags;
 929 }
 930
 931 asmlinkage long sys_set_tid_address(int __user *tidptr)
 932 {
 933         current->clear_child_tid = tidptr;
 934
 935         return current->pid;
 936 }
 937
 938 static inline void rt_mutex_init_task(struct task_struct *p)
 939 {
 940 #ifdef CONFIG_RT_MUTEXES
 941         spin_lock_init(&p->pi_lock);
 942         plist_head_init(&p->pi_waiters, &p->pi_lock);
 943         p->pi_blocked_on = NULL;
 944 #endif
 945 }
 946
 947 /*
 948  * This creates a new process as a copy of the old one,
 949  * but does not actually start it yet.
 950  *
 951  * It copies the registers, and all the appropriate
 952  * parts of the process environment (as per the clone
 953  * flags). The actual kick-off is left to the caller.
 954  */
 955 static struct task_struct *copy_process(unsigned long clone_flags,
 956                                         unsigned long stack_start,
 957                                         struct pt_regs *regs,
 958                                         unsigned long stack_size,
 959                                         int __user *parent_tidptr,
 960                                         int __user *child_tidptr,
 961                                         int pid)
 962 {
 963         int retval;
 964         struct task_struct *p = NULL;
 965         struct vx_info *vxi;
 966         struct nx_info *nxi;
 967
 968         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 969                 return ERR_PTR(-EINVAL);
 970
 971         /*
 972          * Thread groups must share signals as well, and detached threads
 973          * can only be started up within the thread group.
 974          */
 975         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 976                 return ERR_PTR(-EINVAL);
 977
 978         /*
 979          * Shared signal handlers imply shared VM. By way of the above,
 980          * thread groups also imply shared VM. Blocking this case allows
 981          * for various simplifications in other code.
 982          */
 983         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 984                 return ERR_PTR(-EINVAL);
 985
 986         retval = security_task_create(clone_flags);
 987         if (retval)
 988                 goto fork_out;
 989
 990         retval = -ENOMEM;
 991         p = dup_task_struct(current);
 992         if (!p)
 993                 goto fork_out;
 994
 995         rt_mutex_init_task(p);
 996
 997         p->tux_info = NULL;
 998
 999 #ifdef CONFIG_TRACE_IRQFLAGS
1000         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1001         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1002 #endif
1003         init_vx_info(&p->vx_info, current->vx_info);
1004         init_nx_info(&p->nx_info, current->nx_info);
1005
1006         /* check vserver memory */
1007         if (p->mm && !(clone_flags & CLONE_VM)) {
1008                 if (vx_vmpages_avail(p->mm, p->mm->total_vm))
1009                         vx_pages_add(p->vx_info, RLIMIT_AS, p->mm->total_vm);
1010                 else
1011                         goto bad_fork_free;
1012         }
1013         if (p->mm && vx_flags(VXF_FORK_RSS, 0)) {
1014                 if (!vx_rsspages_avail(p->mm, get_mm_counter(p->mm, file_rss)))
1015                         goto bad_fork_cleanup_vm;
1016         }
1017
1018         retval = -EAGAIN;
1019         if (!vx_nproc_avail(1))
1020                 goto bad_fork_cleanup_vm;
1021
1022         if (atomic_read(&p->user->processes) >=
1023                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1024                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1025                                 p->user != &root_user)
1026                         goto bad_fork_cleanup_vm;
1027         }
1028
1029         atomic_inc(&p->user->__count);
1030         atomic_inc(&p->user->processes);
1031         get_group_info(p->group_info);
1032
1033         /*
1034          * If multiple threads are within copy_process(), then this check
1035          * triggers too late. This doesn't hurt, the check is only there
1036          * to stop root fork bombs.
1037          */
1038         if (nr_threads >= max_threads)
1039                 goto bad_fork_cleanup_count;
1040
1041         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1042                 goto bad_fork_cleanup_count;
1043
1044         if (p->binfmt && !try_module_get(p->binfmt->module))
1045                 goto bad_fork_cleanup_put_domain;
1046
1047         p->did_exec = 0;
1048         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1049         copy_flags(clone_flags, p);
1050         p->pid = pid;
1051         retval = -EFAULT;
1052         if (clone_flags & CLONE_PARENT_SETTID)
1053                 if (put_user(p->pid, parent_tidptr))
1054                         goto bad_fork_cleanup_delays_binfmt;
1055
1056         INIT_LIST_HEAD(&p->children);
1057         INIT_LIST_HEAD(&p->sibling);
1058         p->vfork_done = NULL;
1059         spin_lock_init(&p->alloc_lock);
1060
1061         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1062         init_sigpending(&p->pending);
1063
1064         p->utime = cputime_zero;
1065         p->stime = cputime_zero;
1066         p->sched_time = 0;
1067         p->rchar = 0;           /* I/O counter: bytes read */
1068         p->wchar = 0;           /* I/O counter: bytes written */
1069         p->syscr = 0;           /* I/O counter: read syscalls */
1070         p->syscw = 0;           /* I/O counter: write syscalls */
1071         acct_clear_integrals(p);
1072
1073         p->it_virt_expires = cputime_zero;
1074         p->it_prof_expires = cputime_zero;
1075         p->it_sched_expires = 0;
1076         INIT_LIST_HEAD(&p->cpu_timers[0]);
1077         INIT_LIST_HEAD(&p->cpu_timers[1]);
1078         INIT_LIST_HEAD(&p->cpu_timers[2]);
1079
1080         p->lock_depth = -1;             /* -1 = no lock */
1081         do_posix_clock_monotonic_gettime(&p->start_time);
1082         p->security = NULL;
1083         p->io_context = NULL;
1084         p->io_wait = NULL;
1085         p->audit_context = NULL;
1086         cpuset_fork(p);
1087 #ifdef CONFIG_NUMA
1088         p->mempolicy = mpol_copy(p->mempolicy);
1089         if (IS_ERR(p->mempolicy)) {
1090                 retval = PTR_ERR(p->mempolicy);
1091                 p->mempolicy = NULL;
1092                 goto bad_fork_cleanup_cpuset;
1093         }
1094         mpol_fix_fork_child_flag(p);
1095 #endif
1096 #ifdef CONFIG_TRACE_IRQFLAGS
1097         p->irq_events = 0;
1098         p->hardirqs_enabled = 0;
1099         p->hardirq_enable_ip = 0;
1100         p->hardirq_enable_event = 0;
1101         p->hardirq_disable_ip = _THIS_IP_;
1102         p->hardirq_disable_event = 0;
1103         p->softirqs_enabled = 1;
1104         p->softirq_enable_ip = _THIS_IP_;
1105         p->softirq_enable_event = 0;
1106         p->softirq_disable_ip = 0;
1107         p->softirq_disable_event = 0;
1108         p->hardirq_context = 0;
1109         p->softirq_context = 0;
1110 #endif
1111 #ifdef CONFIG_LOCKDEP
1112         p->lockdep_depth = 0; /* no locks held yet */
1113         p->curr_chain_key = 0;
1114         p->lockdep_recursion = 0;
1115 #endif
1116
1117 #ifdef CONFIG_DEBUG_MUTEXES
1118         p->blocked_on = NULL; /* not blocked yet */
1119 #endif
1120
1121         p->tgid = p->pid;
1122         if (clone_flags & CLONE_THREAD)
1123                 p->tgid = current->tgid;
1124
1125         if ((retval = security_task_alloc(p)))
1126                 goto bad_fork_cleanup_policy;
1127         if ((retval = audit_alloc(p)))
1128                 goto bad_fork_cleanup_security;
1129         /* copy all the process information */
1130         if ((retval = copy_semundo(clone_flags, p)))
1131                 goto bad_fork_cleanup_audit;
1132         if ((retval = copy_files(clone_flags, p)))
1133                 goto bad_fork_cleanup_semundo;
1134         if ((retval = copy_fs(clone_flags, p)))
1135                 goto bad_fork_cleanup_files;
1136         if ((retval = copy_sighand(clone_flags, p)))
1137                 goto bad_fork_cleanup_fs;
1138         if ((retval = copy_signal(clone_flags, p)))
1139                 goto bad_fork_cleanup_sighand;
1140         if ((retval = copy_mm(clone_flags, p)))
1141                 goto bad_fork_cleanup_signal;
1142         if ((retval = copy_keys(clone_flags, p)))
1143                 goto bad_fork_cleanup_mm;
1144         if ((retval = copy_namespace(clone_flags, p)))
1145                 goto bad_fork_cleanup_keys;
1146         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1147         if (retval)
1148                 goto bad_fork_cleanup_namespace;
1149
1150         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1151         /*
1152          * Clear TID on mm_release()?
1153          */
1154         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1155         p->robust_list = NULL;
1156 #ifdef CONFIG_COMPAT
1157         p->compat_robust_list = NULL;
1158 #endif
1159         INIT_LIST_HEAD(&p->pi_state_list);
1160         p->pi_state_cache = NULL;
1161
1162         /*
1163          * sigaltstack should be cleared when sharing the same VM
1164          */
1165         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1166                 p->sas_ss_sp = p->sas_ss_size = 0;
1167
1168         /*
1169          * Syscall tracing should be turned off in the child regardless
1170          * of CLONE_PTRACE.
1171          */
1172         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1173 #ifdef TIF_SYSCALL_EMU
1174         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1175 #endif
1176
1177         /* Our parent execution domain becomes current domain
1178            These must match for thread signalling to apply */
1179
1180         p->parent_exec_id = p->self_exec_id;
1181
1182         /* ok, now we should be set up.. */
1183         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1184         p->pdeath_signal = 0;
1185         p->exit_state = 0;
1186
1187         /*
1188          * Ok, make it visible to the rest of the system.
1189          * We dont wake it up yet.
1190          */
1191         p->group_leader = p;
1192         INIT_LIST_HEAD(&p->thread_group);
1193         INIT_LIST_HEAD(&p->ptrace_children);
1194         INIT_LIST_HEAD(&p->ptrace_list);
1195
1196         /* Perform scheduler related setup. Assign this task to a CPU. */
1197         sched_fork(p, clone_flags);
1198
1199         /* Need tasklist lock for parent etc handling! */
1200         write_lock_irq(&tasklist_lock);
1201
1202         /*
1203          * The task hasn't been attached yet, so its cpus_allowed mask will
1204          * not be changed, nor will its assigned CPU.
1205          *
1206          * The cpus_allowed mask of the parent may have changed after it was
1207          * copied first time - so re-copy it here, then check the child's CPU
1208          * to ensure it is on a valid CPU (and if not, just force it back to
1209          * parent's CPU). This avoids alot of nasty races.
1210          */
1211         p->cpus_allowed = current->cpus_allowed;
1212         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1213                         !cpu_online(task_cpu(p))))
1214                 set_task_cpu(p, smp_processor_id());
1215
1216         /* CLONE_PARENT re-uses the old parent */
1217         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1218                 p->real_parent = current->real_parent;
1219         else
1220                 p->real_parent = current;
1221         p->parent = p->real_parent;
1222
1223         spin_lock(&current->sighand->siglock);
1224
1225         /*
1226          * Process group and session signals need to be delivered to just the
1227          * parent before the fork or both the parent and the child after the
1228          * fork. Restart if a signal comes in before we add the new process to
1229          * it's process group.
1230          * A fatal signal pending means that current will exit, so the new
1231          * thread can't slip out of an OOM kill (or normal SIGKILL).
1232          */
1233         recalc_sigpending();
1234         if (signal_pending(current)) {
1235                 spin_unlock(&current->sighand->siglock);
1236                 write_unlock_irq(&tasklist_lock);
1237                 retval = -ERESTARTNOINTR;
1238                 goto bad_fork_cleanup_namespace;
1239         }
1240
1241         if (clone_flags & CLONE_THREAD) {
1242                 p->group_leader = current->group_leader;
1243                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1244
1245                 if (!cputime_eq(current->signal->it_virt_expires,
1246                                 cputime_zero) ||
1247                     !cputime_eq(current->signal->it_prof_expires,
1248                                 cputime_zero) ||
1249                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1250                     !list_empty(&current->signal->cpu_timers[0]) ||
1251                     !list_empty(&current->signal->cpu_timers[1]) ||
1252                     !list_empty(&current->signal->cpu_timers[2])) {
1253                         /*
1254                          * Have child wake up on its first tick to check
1255                          * for process CPU timers.
1256                          */
1257                         p->it_prof_expires = jiffies_to_cputime(1);
1258                 }
1259         }
1260
1261         /*
1262          * inherit ioprio
1263          */
1264         p->ioprio = current->ioprio;
1265
1266         if (likely(p->pid)) {
1267                 add_parent(p);
1268                 if (unlikely(p->ptrace & PT_PTRACED))
1269                         __ptrace_link(p, current->parent);
1270
1271                 if (thread_group_leader(p)) {
1272                         p->signal->tty = current->signal->tty;
1273                         p->signal->pgrp = process_group(current);
1274                         p->signal->session = current->signal->session;
1275                         attach_pid(p, PIDTYPE_PGID, process_group(p));
1276                         attach_pid(p, PIDTYPE_SID, p->signal->session);
1277
1278                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1279                         __get_cpu_var(process_counts)++;
1280                 }
1281                 attach_pid(p, PIDTYPE_PID, p->pid);
1282                 nr_threads++;
1283         }
1284
1285         total_forks++;
1286         spin_unlock(&current->sighand->siglock);
1287
1288         /* p is copy of current */
1289         vxi = p->vx_info;
1290         if (vxi) {
1291                 claim_vx_info(vxi, p);
1292                 atomic_inc(&vxi->cvirt.nr_threads);
1293                 atomic_inc(&vxi->cvirt.total_forks);
1294                 vx_nproc_inc(p);
1295         }
1296         nxi = p->nx_info;
1297         if (nxi)
1298                 claim_nx_info(nxi, p);
1299         write_unlock_irq(&tasklist_lock);
1300         proc_fork_connector(p);
1301         return p;
1302
1303 bad_fork_cleanup_namespace:
1304         exit_namespace(p);
1305 bad_fork_cleanup_keys:
1306         exit_keys(p);
1307 bad_fork_cleanup_mm:
1308         if (p->mm)
1309                 mmput(p->mm);
1310 bad_fork_cleanup_signal:
1311         cleanup_signal(p);
1312 bad_fork_cleanup_sighand:
1313         __cleanup_sighand(p->sighand);
1314 bad_fork_cleanup_fs:
1315         exit_fs(p); /* blocking */
1316 bad_fork_cleanup_files:
1317         exit_files(p); /* blocking */
1318 bad_fork_cleanup_semundo:
1319         exit_sem(p);
1320 bad_fork_cleanup_audit:
1321         audit_free(p);
1322 bad_fork_cleanup_security:
1323         security_task_free(p);
1324 bad_fork_cleanup_policy:
1325 #ifdef CONFIG_NUMA
1326         mpol_free(p->mempolicy);
1327 bad_fork_cleanup_cpuset:
1328 #endif
1329         cpuset_exit(p);
1330 bad_fork_cleanup_delays_binfmt:
1331         delayacct_tsk_free(p);
1332         if (p->binfmt)
1333                 module_put(p->binfmt->module);
1334 bad_fork_cleanup_put_domain:
1335         module_put(task_thread_info(p)->exec_domain->module);
1336 bad_fork_cleanup_count:
1337         put_group_info(p->group_info);
1338         atomic_dec(&p->user->processes);
1339         free_uid(p->user);
1340 bad_fork_cleanup_vm:
1341         if (p->mm && !(clone_flags & CLONE_VM))
1342                 vx_pages_sub(p->vx_info, RLIMIT_AS, p->mm->total_vm);
1343 bad_fork_free:
1344         free_task(p);
1345 fork_out:
1346         return ERR_PTR(retval);
1347 }
1348
1349 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1350 {
1351         memset(regs, 0, sizeof(struct pt_regs));
1352         return regs;
1353 }
1354
1355 struct task_struct * __devinit fork_idle(int cpu)
1356 {
1357         struct task_struct *task;
1358         struct pt_regs regs;
1359
1360         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1361         if (!task)
1362                 return ERR_PTR(-ENOMEM);
1363         init_idle(task, cpu);
1364
1365         return task;
1366 }
1367
1368 static inline int fork_traceflag (unsigned clone_flags)
1369 {
1370         if (clone_flags & CLONE_UNTRACED)
1371                 return 0;
1372         else if (clone_flags & CLONE_VFORK) {
1373                 if (current->ptrace & PT_TRACE_VFORK)
1374                         return PTRACE_EVENT_VFORK;
1375         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1376                 if (current->ptrace & PT_TRACE_CLONE)
1377                         return PTRACE_EVENT_CLONE;
1378         } else if (current->ptrace & PT_TRACE_FORK)
1379                 return PTRACE_EVENT_FORK;
1380
1381         return 0;
1382 }
1383
1384 /*
1385  *  Ok, this is the main fork-routine.
1386  *
1387  * It copies the process, and if successful kick-starts
1388  * it and waits for it to finish using the VM if required.
1389  */
1390 long do_fork(unsigned long clone_flags,
1391               unsigned long stack_start,
1392               struct pt_regs *regs,
1393               unsigned long stack_size,
1394               int __user *parent_tidptr,
1395               int __user *child_tidptr)
1396 {
1397         struct task_struct *p;
1398         int trace = 0;
1399         struct pid *pid = alloc_pid();
1400         long nr;
1401
1402         if (!pid)
1403                 return -EAGAIN;
1404         nr = pid->nr;
1405         if (unlikely(current->ptrace)) {
1406                 trace = fork_traceflag (clone_flags);
1407                 if (trace)
1408                         clone_flags |= CLONE_PTRACE;
1409         }
1410
1411         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, nr);
1412         /*
1413          * Do this prior waking up the new thread - the thread pointer
1414          * might get invalid after that point, if the thread exits quickly.
1415          */
1416         if (!IS_ERR(p)) {
1417                 struct completion vfork;
1418
1419                 if (clone_flags & CLONE_VFORK) {
1420                         p->vfork_done = &vfork;
1421                         init_completion(&vfork);
1422                 }
1423
1424                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1425                         /*
1426                          * We'll start up with an immediate SIGSTOP.
1427                          */
1428                         sigaddset(&p->pending.signal, SIGSTOP);
1429                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1430                 }
1431
1432                 if (!(clone_flags & CLONE_STOPPED))
1433                         wake_up_new_task(p, clone_flags);
1434                 else
1435                         p->state = TASK_STOPPED;
1436
1437                 if (unlikely (trace)) {
1438                         current->ptrace_message = nr;
1439                         ptrace_notify ((trace << 8) | SIGTRAP);
1440                 }
1441
1442                 if (clone_flags & CLONE_VFORK) {
1443                         wait_for_completion(&vfork);
1444                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1445                                 current->ptrace_message = nr;
1446                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1447                         }
1448                 }
1449         } else {
1450                 free_pid(pid);
1451                 nr = PTR_ERR(p);
1452         }
1453         return nr;
1454 }
1455
1456 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1457 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1458 #endif
1459
1460 static void sighand_ctor(void *data, kmem_cache_t *cachep, unsigned long flags)
1461 {
1462         struct sighand_struct *sighand = data;
1463
1464         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
1465                                         SLAB_CTOR_CONSTRUCTOR)
1466                 spin_lock_init(&sighand->siglock);
1467 }
1468
1469 void __init proc_caches_init(void)
1470 {
1471         sighand_cachep = kmem_cache_create("sighand_cache",
1472                         sizeof(struct sighand_struct), 0,
1473                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1474                         sighand_ctor, NULL);
1475         signal_cachep = kmem_cache_create("signal_cache",
1476                         sizeof(struct signal_struct), 0,
1477                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1478         files_cachep = kmem_cache_create("files_cache",
1479                         sizeof(struct files_struct), 0,
1480                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1481         fs_cachep = kmem_cache_create("fs_cache",
1482                         sizeof(struct fs_struct), 0,
1483                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1484         vm_area_cachep = kmem_cache_create("vm_area_struct",
1485                         sizeof(struct vm_area_struct), 0,
1486                         SLAB_PANIC, NULL, NULL);
1487         mm_cachep = kmem_cache_create("mm_struct",
1488                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1489                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1490 }
1491
1492
1493 /*
1494  * Check constraints on flags passed to the unshare system call and
1495  * force unsharing of additional process context as appropriate.
1496  */
1497 static inline void check_unshare_flags(unsigned long *flags_ptr)
1498 {
1499         /*
1500          * If unsharing a thread from a thread group, must also
1501          * unshare vm.
1502          */
1503         if (*flags_ptr & CLONE_THREAD)
1504                 *flags_ptr |= CLONE_VM;
1505
1506         /*
1507          * If unsharing vm, must also unshare signal handlers.
1508          */
1509         if (*flags_ptr & CLONE_VM)
1510                 *flags_ptr |= CLONE_SIGHAND;
1511
1512         /*
1513          * If unsharing signal handlers and the task was created
1514          * using CLONE_THREAD, then must unshare the thread
1515          */
1516         if ((*flags_ptr & CLONE_SIGHAND) &&
1517             (atomic_read(&current->signal->count) > 1))
1518                 *flags_ptr |= CLONE_THREAD;
1519
1520         /*
1521          * If unsharing namespace, must also unshare filesystem information.
1522          */
1523         if (*flags_ptr & CLONE_NEWNS)
1524                 *flags_ptr |= CLONE_FS;
1525 }
1526
1527 /*
1528  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1529  */
1530 static int unshare_thread(unsigned long unshare_flags)
1531 {
1532         if (unshare_flags & CLONE_THREAD)
1533                 return -EINVAL;
1534
1535         return 0;
1536 }
1537
1538 /*
1539  * Unshare the filesystem structure if it is being shared
1540  */
1541 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1542 {
1543         struct fs_struct *fs = current->fs;
1544
1545         if ((unshare_flags & CLONE_FS) &&
1546             (fs && atomic_read(&fs->count) > 1)) {
1547                 *new_fsp = __copy_fs_struct(current->fs);
1548                 if (!*new_fsp)
1549                         return -ENOMEM;
1550         }
1551
1552         return 0;
1553 }
1554
1555 /*
1556  * Unshare the namespace structure if it is being shared
1557  */
1558 static int unshare_namespace(unsigned long unshare_flags, struct namespace **new_nsp, struct fs_struct *new_fs)
1559 {
1560         struct namespace *ns = current->namespace;
1561
1562         if ((unshare_flags & CLONE_NEWNS) &&
1563             (ns && atomic_read(&ns->count) > 1)) {
1564                 if (!capable(CAP_SYS_ADMIN))
1565                         return -EPERM;
1566
1567                 *new_nsp = dup_namespace(current, new_fs ? new_fs : current->fs);
1568                 if (!*new_nsp)
1569                         return -ENOMEM;
1570         }
1571
1572         return 0;
1573 }
1574
1575 /*
1576  * Unsharing of sighand for tasks created with CLONE_SIGHAND is not
1577  * supported yet
1578  */
1579 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1580 {
1581         struct sighand_struct *sigh = current->sighand;
1582
1583         if ((unshare_flags & CLONE_SIGHAND) &&
1584             (sigh && atomic_read(&sigh->count) > 1))
1585                 return -EINVAL;
1586         else
1587                 return 0;
1588 }
1589
1590 /*
1591  * Unshare vm if it is being shared
1592  */
1593 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1594 {
1595         struct mm_struct *mm = current->mm;
1596
1597         if ((unshare_flags & CLONE_VM) &&
1598             (mm && atomic_read(&mm->mm_users) > 1)) {
1599                 return -EINVAL;
1600         }
1601
1602         return 0;
1603 }
1604
1605 /*
1606  * Unshare file descriptor table if it is being shared
1607  */
1608 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1609 {
1610         struct files_struct *fd = current->files;
1611         int error = 0;
1612
1613         if ((unshare_flags & CLONE_FILES) &&
1614             (fd && atomic_read(&fd->count) > 1)) {
1615                 *new_fdp = dup_fd(fd, &error);
1616                 if (!*new_fdp)
1617                         return error;
1618         }
1619
1620         return 0;
1621 }
1622
1623 /*
1624  * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1625  * supported yet
1626  */
1627 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1628 {
1629         if (unshare_flags & CLONE_SYSVSEM)
1630                 return -EINVAL;
1631
1632         return 0;
1633 }
1634
1635 /*
1636  * unshare allows a process to 'unshare' part of the process
1637  * context which was originally shared using clone.  copy_*
1638  * functions used by do_fork() cannot be used here directly
1639  * because they modify an inactive task_struct that is being
1640  * constructed. Here we are modifying the current, active,
1641  * task_struct.
1642  */
1643 asmlinkage long sys_unshare(unsigned long unshare_flags)
1644 {
1645         int err = 0;
1646         struct fs_struct *fs, *new_fs = NULL;
1647         struct namespace *ns, *new_ns = NULL;
1648         struct sighand_struct *sigh, *new_sigh = NULL;
1649         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1650         struct files_struct *fd, *new_fd = NULL;
1651         struct sem_undo_list *new_ulist = NULL;
1652
1653         check_unshare_flags(&unshare_flags);
1654
1655         /* Return -EINVAL for all unsupported flags */
1656         err = -EINVAL;
1657         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1658                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM))
1659                 goto bad_unshare_out;
1660
1661         if ((err = unshare_thread(unshare_flags)))
1662                 goto bad_unshare_out;
1663         if ((err = unshare_fs(unshare_flags, &new_fs)))
1664                 goto bad_unshare_cleanup_thread;
1665         if ((err = unshare_namespace(unshare_flags, &new_ns, new_fs)))
1666                 goto bad_unshare_cleanup_fs;
1667         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1668                 goto bad_unshare_cleanup_ns;
1669         if ((err = unshare_vm(unshare_flags, &new_mm)))
1670                 goto bad_unshare_cleanup_sigh;
1671         if ((err = unshare_fd(unshare_flags, &new_fd)))
1672                 goto bad_unshare_cleanup_vm;
1673         if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1674                 goto bad_unshare_cleanup_fd;
1675
1676         if (new_fs || new_ns || new_sigh || new_mm || new_fd || new_ulist) {
1677
1678                 task_lock(current);
1679
1680                 if (new_fs) {
1681                         fs = current->fs;
1682                         current->fs = new_fs;
1683                         new_fs = fs;
1684                 }
1685
1686                 if (new_ns) {
1687                         ns = current->namespace;
1688                         current->namespace = new_ns;
1689                         new_ns = ns;
1690                 }
1691
1692                 if (new_sigh) {
1693                         sigh = current->sighand;
1694                         rcu_assign_pointer(current->sighand, new_sigh);
1695                         new_sigh = sigh;
1696                 }
1697
1698                 if (new_mm) {
1699                         mm = current->mm;
1700                         active_mm = current->active_mm;
1701                         current->mm = new_mm;
1702                         current->active_mm = new_mm;
1703                         activate_mm(active_mm, new_mm);
1704                         new_mm = mm;
1705                 }
1706
1707                 if (new_fd) {
1708                         fd = current->files;
1709                         current->files = new_fd;
1710                         new_fd = fd;
1711                 }
1712
1713                 task_unlock(current);
1714         }
1715
1716 bad_unshare_cleanup_fd:
1717         if (new_fd)
1718                 put_files_struct(new_fd);
1719
1720 bad_unshare_cleanup_vm:
1721         if (new_mm)
1722                 mmput(new_mm);
1723
1724 bad_unshare_cleanup_sigh:
1725         if (new_sigh)
1726                 if (atomic_dec_and_test(&new_sigh->count))
1727                         kmem_cache_free(sighand_cachep, new_sigh);
1728
1729 bad_unshare_cleanup_ns:
1730         if (new_ns)
1731                 put_namespace(new_ns);
1732
1733 bad_unshare_cleanup_fs:
1734         if (new_fs)
1735                 put_fs_struct(new_fs);
1736
1737 bad_unshare_cleanup_thread:
1738 bad_unshare_out:
1739         return err;
1740 }