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/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/mnt_namespace.h>
  21 #include <linux/personality.h>
  22 #include <linux/mempolicy.h>
  23 #include <linux/sem.h>
  24 #include <linux/file.h>
  25 #include <linux/fdtable.h>
  26 #include <linux/iocontext.h>
  27 #include <linux/key.h>
  28 #include <linux/binfmts.h>
  29 #include <linux/mman.h>
  30 #include <linux/mmu_notifier.h>
  31 #include <linux/fs.h>
  32 #include <linux/nsproxy.h>
  33 #include <linux/capability.h>
  34 #include <linux/cpu.h>
  35 #include <linux/cgroup.h>
  36 #include <linux/security.h>
  37 #include <linux/hugetlb.h>
  38 #include <linux/swap.h>
  39 #include <linux/syscalls.h>
  40 #include <linux/jiffies.h>
  41 #include <linux/tracehook.h>
  42 #include <linux/futex.h>
  43 #include <linux/task_io_accounting_ops.h>
  44 #include <linux/rcupdate.h>
  45 #include <linux/ptrace.h>
  46 #include <linux/mount.h>
  47 #include <linux/audit.h>
  48 #include <linux/memcontrol.h>
  49 #include <linux/profile.h>
  50 #include <linux/rmap.h>
  51 #include <linux/acct.h>
  52 #include <linux/tsacct_kern.h>
  53 #include <linux/cn_proc.h>
  54 #include <linux/freezer.h>
  55 #include <linux/delayacct.h>
  56 #include <linux/taskstats_kern.h>
  57 #include <linux/random.h>
  58 #include <linux/tty.h>
  59 #include <linux/proc_fs.h>
  60 #include <linux/blkdev.h>
  61
  62 #include <asm/pgtable.h>
  63 #include <asm/pgalloc.h>
  64 #include <asm/uaccess.h>
  65 #include <asm/mmu_context.h>
  66 #include <asm/cacheflush.h>
  67 #include <asm/tlbflush.h>
  68
  69 /*
  70  * Protected counters by write_lock_irq(&tasklist_lock)
  71  */
  72 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  73 int nr_threads;                 /* The idle threads do not count.. */
  74
  75 int max_threads;                /* tunable limit on nr_threads */
  76
  77 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  78
  79 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  80
  81 int nr_processes(void)
  82 {
  83         int cpu;
  84         int total = 0;
  85
  86         for_each_online_cpu(cpu)
  87                 total += per_cpu(process_counts, cpu);
  88
  89         return total;
  90 }
  91
  92 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  93 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  94 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  95 static struct kmem_cache *task_struct_cachep;
  96 #endif
  97
  98 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
  99 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
 100 {
 101 #ifdef CONFIG_DEBUG_STACK_USAGE
 102         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 103 #else
 104         gfp_t mask = GFP_KERNEL;
 105 #endif
 106         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
 107 }
 108
 109 static inline void free_thread_info(struct thread_info *ti)
 110 {
 111         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 112 }
 113 #endif
 114
 115 /* SLAB cache for signal_struct structures (tsk->signal) */
 116 static struct kmem_cache *signal_cachep;
 117
 118 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 119 struct kmem_cache *sighand_cachep;
 120
 121 /* SLAB cache for files_struct structures (tsk->files) */
 122 struct kmem_cache *files_cachep;
 123
 124 /* SLAB cache for fs_struct structures (tsk->fs) */
 125 struct kmem_cache *fs_cachep;
 126
 127 /* SLAB cache for vm_area_struct structures */
 128 struct kmem_cache *vm_area_cachep;
 129
 130 /* SLAB cache for mm_struct structures (tsk->mm) */
 131 static struct kmem_cache *mm_cachep;
 132
 133 void free_task(struct task_struct *tsk)
 134 {
 135         prop_local_destroy_single(&tsk->dirties);
 136         free_thread_info(tsk->stack);
 137         rt_mutex_debug_task_free(tsk);
 138         free_task_struct(tsk);
 139 }
 140 EXPORT_SYMBOL(free_task);
 141
 142 void __put_task_struct(struct task_struct *tsk)
 143 {
 144         WARN_ON(!tsk->exit_state);
 145         WARN_ON(atomic_read(&tsk->usage));
 146         WARN_ON(tsk == current);
 147
 148         security_task_free(tsk);
 149         free_uid(tsk->user);
 150         put_group_info(tsk->group_info);
 151         delayacct_tsk_free(tsk);
 152
 153         if (!profile_handoff_task(tsk))
 154                 free_task(tsk);
 155 }
 156
 157 /*
 158  * macro override instead of weak attribute alias, to workaround
 159  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 160  */
 161 #ifndef arch_task_cache_init
 162 #define arch_task_cache_init()
 163 #endif
 164
 165 void __init fork_init(unsigned long mempages)
 166 {
 167 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 168 #ifndef ARCH_MIN_TASKALIGN
 169 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 170 #endif
 171         /* create a slab on which task_structs can be allocated */
 172         task_struct_cachep =
 173                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 174                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
 175 #endif
 176
 177         /* do the arch specific task caches init */
 178         arch_task_cache_init();
 179
 180         /*
 181          * The default maximum number of threads is set to a safe
 182          * value: the thread structures can take up at most half
 183          * of memory.
 184          */
 185         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 186
 187         /*
 188          * we need to allow at least 20 threads to boot a system
 189          */
 190         if(max_threads < 20)
 191                 max_threads = 20;
 192
 193         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 194         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 195         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 196                 init_task.signal->rlim[RLIMIT_NPROC];
 197 }
 198
 199 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 200                                                struct task_struct *src)
 201 {
 202         *dst = *src;
 203         return 0;
 204 }
 205
 206 static struct task_struct *dup_task_struct(struct task_struct *orig)
 207 {
 208         struct task_struct *tsk;
 209         struct thread_info *ti;
 210         int err;
 211
 212         prepare_to_copy(orig);
 213
 214         tsk = alloc_task_struct();
 215         if (!tsk)
 216                 return NULL;
 217
 218         ti = alloc_thread_info(tsk);
 219         if (!ti) {
 220                 free_task_struct(tsk);
 221                 return NULL;
 222         }
 223
 224         err = arch_dup_task_struct(tsk, orig);
 225         if (err)
 226                 goto out;
 227
 228         tsk->stack = ti;
 229
 230         err = prop_local_init_single(&tsk->dirties);
 231         if (err)
 232                 goto out;
 233
 234         setup_thread_stack(tsk, orig);
 235
 236 #ifdef CONFIG_CC_STACKPROTECTOR
 237         tsk->stack_canary = get_random_int();
 238 #endif
 239
 240         /* One for us, one for whoever does the "release_task()" (usually parent) */
 241         atomic_set(&tsk->usage,2);
 242         atomic_set(&tsk->fs_excl, 0);
 243 #ifdef CONFIG_BLK_DEV_IO_TRACE
 244         tsk->btrace_seq = 0;
 245 #endif
 246         tsk->splice_pipe = NULL;
 247         return tsk;
 248
 249 out:
 250         free_thread_info(ti);
 251         free_task_struct(tsk);
 252         return NULL;
 253 }
 254
 255 #ifdef CONFIG_MMU
 256 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 257 {
 258         struct vm_area_struct *mpnt, *tmp, **pprev;
 259         struct rb_node **rb_link, *rb_parent;
 260         int retval;
 261         unsigned long charge;
 262         struct mempolicy *pol;
 263
 264         down_write(&oldmm->mmap_sem);
 265         flush_cache_dup_mm(oldmm);
 266         /*
 267          * Not linked in yet - no deadlock potential:
 268          */
 269         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 270
 271         mm->locked_vm = 0;
 272         mm->mmap = NULL;
 273         mm->mmap_cache = NULL;
 274         mm->free_area_cache = oldmm->mmap_base;
 275         mm->cached_hole_size = ~0UL;
 276         mm->map_count = 0;
 277         cpus_clear(mm->cpu_vm_mask);
 278         mm->mm_rb = RB_ROOT;
 279         rb_link = &mm->mm_rb.rb_node;
 280         rb_parent = NULL;
 281         pprev = &mm->mmap;
 282
 283         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 284                 struct file *file;
 285
 286                 if (mpnt->vm_flags & VM_DONTCOPY) {
 287                         long pages = vma_pages(mpnt);
 288                         mm->total_vm -= pages;
 289                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 290                                                                 -pages);
 291                         continue;
 292                 }
 293                 charge = 0;
 294                 if (mpnt->vm_flags & VM_ACCOUNT) {
 295                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 296                         if (security_vm_enough_memory(len))
 297                                 goto fail_nomem;
 298                         charge = len;
 299                 }
 300                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 301                 if (!tmp)
 302                         goto fail_nomem;
 303                 *tmp = *mpnt;
 304                 pol = mpol_dup(vma_policy(mpnt));
 305                 retval = PTR_ERR(pol);
 306                 if (IS_ERR(pol))
 307                         goto fail_nomem_policy;
 308                 vma_set_policy(tmp, pol);
 309                 tmp->vm_flags &= ~VM_LOCKED;
 310                 tmp->vm_mm = mm;
 311                 tmp->vm_next = NULL;
 312                 anon_vma_link(tmp);
 313                 file = tmp->vm_file;
 314                 if (file) {
 315                         struct inode *inode = file->f_path.dentry->d_inode;
 316                         get_file(file);
 317                         if (tmp->vm_flags & VM_DENYWRITE)
 318                                 atomic_dec(&inode->i_writecount);
 319
 320                         /* insert tmp into the share list, just after mpnt */
 321                         spin_lock(&file->f_mapping->i_mmap_lock);
 322                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 323                         flush_dcache_mmap_lock(file->f_mapping);
 324                         vma_prio_tree_add(tmp, mpnt);
 325                         flush_dcache_mmap_unlock(file->f_mapping);
 326                         spin_unlock(&file->f_mapping->i_mmap_lock);
 327                 }
 328
 329                 /*
 330                  * Clear hugetlb-related page reserves for children. This only
 331                  * affects MAP_PRIVATE mappings. Faults generated by the child
 332                  * are not guaranteed to succeed, even if read-only
 333                  */
 334                 if (is_vm_hugetlb_page(tmp))
 335                         reset_vma_resv_huge_pages(tmp);
 336
 337                 /*
 338                  * Link in the new vma and copy the page table entries.
 339                  */
 340                 *pprev = tmp;
 341                 pprev = &tmp->vm_next;
 342
 343                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 344                 rb_link = &tmp->vm_rb.rb_right;
 345                 rb_parent = &tmp->vm_rb;
 346
 347                 mm->map_count++;
 348                 retval = copy_page_range(mm, oldmm, mpnt);
 349
 350                 if (tmp->vm_ops && tmp->vm_ops->open)
 351                         tmp->vm_ops->open(tmp);
 352
 353                 if (retval)
 354                         goto out;
 355         }
 356         /* a new mm has just been created */
 357         arch_dup_mmap(oldmm, mm);
 358         retval = 0;
 359 out:
 360         up_write(&mm->mmap_sem);
 361         flush_tlb_mm(oldmm);
 362         up_write(&oldmm->mmap_sem);
 363         return retval;
 364 fail_nomem_policy:
 365         kmem_cache_free(vm_area_cachep, tmp);
 366 fail_nomem:
 367         retval = -ENOMEM;
 368         vm_unacct_memory(charge);
 369         goto out;
 370 }
 371
 372 static inline int mm_alloc_pgd(struct mm_struct * mm)
 373 {
 374         mm->pgd = pgd_alloc(mm);
 375         if (unlikely(!mm->pgd))
 376                 return -ENOMEM;
 377         return 0;
 378 }
 379
 380 static inline void mm_free_pgd(struct mm_struct * mm)
 381 {
 382         pgd_free(mm, mm->pgd);
 383 }
 384 #else
 385 #define dup_mmap(mm, oldmm)     (0)
 386 #define mm_alloc_pgd(mm)        (0)
 387 #define mm_free_pgd(mm)
 388 #endif /* CONFIG_MMU */
 389
 390 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 391
 392 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 393 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 394
 395 #include <linux/init_task.h>
 396
 397 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
 398 {
 399         atomic_set(&mm->mm_users, 1);
 400         atomic_set(&mm->mm_count, 1);
 401         init_rwsem(&mm->mmap_sem);
 402         INIT_LIST_HEAD(&mm->mmlist);
 403         mm->flags = (current->mm) ? current->mm->flags
 404                                   : MMF_DUMP_FILTER_DEFAULT;
 405         mm->core_state = NULL;
 406         mm->nr_ptes = 0;
 407         set_mm_counter(mm, file_rss, 0);
 408         set_mm_counter(mm, anon_rss, 0);
 409         spin_lock_init(&mm->page_table_lock);
 410         rwlock_init(&mm->ioctx_list_lock);
 411         mm->ioctx_list = NULL;
 412         mm->free_area_cache = TASK_UNMAPPED_BASE;
 413         mm->cached_hole_size = ~0UL;
 414         mm_init_owner(mm, p);
 415
 416         if (likely(!mm_alloc_pgd(mm))) {
 417                 mm->def_flags = 0;
 418                 mmu_notifier_mm_init(mm);
 419                 return mm;
 420         }
 421
 422         free_mm(mm);
 423         return NULL;
 424 }
 425
 426 /*
 427  * Allocate and initialize an mm_struct.
 428  */
 429 struct mm_struct * mm_alloc(void)
 430 {
 431         struct mm_struct * mm;
 432
 433         mm = allocate_mm();
 434         if (mm) {
 435                 memset(mm, 0, sizeof(*mm));
 436                 mm = mm_init(mm, current);
 437         }
 438         return mm;
 439 }
 440
 441 /*
 442  * Called when the last reference to the mm
 443  * is dropped: either by a lazy thread or by
 444  * mmput. Free the page directory and the mm.
 445  */
 446 void __mmdrop(struct mm_struct *mm)
 447 {
 448         BUG_ON(mm == &init_mm);
 449         mm_free_pgd(mm);
 450         destroy_context(mm);
 451         mmu_notifier_mm_destroy(mm);
 452         free_mm(mm);
 453 }
 454 EXPORT_SYMBOL_GPL(__mmdrop);
 455
 456 /*
 457  * Decrement the use count and release all resources for an mm.
 458  */
 459 void mmput(struct mm_struct *mm)
 460 {
 461         might_sleep();
 462
 463         if (atomic_dec_and_test(&mm->mm_users)) {
 464                 exit_aio(mm);
 465                 exit_mmap(mm);
 466                 set_mm_exe_file(mm, NULL);
 467                 if (!list_empty(&mm->mmlist)) {
 468                         spin_lock(&mmlist_lock);
 469                         list_del(&mm->mmlist);
 470                         spin_unlock(&mmlist_lock);
 471                 }
 472                 put_swap_token(mm);
 473                 mmdrop(mm);
 474         }
 475 }
 476 EXPORT_SYMBOL_GPL(mmput);
 477
 478 /**
 479  * get_task_mm - acquire a reference to the task's mm
 480  *
 481  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 482  * this kernel workthread has transiently adopted a user mm with use_mm,
 483  * to do its AIO) is not set and if so returns a reference to it, after
 484  * bumping up the use count.  User must release the mm via mmput()
 485  * after use.  Typically used by /proc and ptrace.
 486  */
 487 struct mm_struct *get_task_mm(struct task_struct *task)
 488 {
 489         struct mm_struct *mm;
 490
 491         task_lock(task);
 492         mm = task->mm;
 493         if (mm) {
 494                 if (task->flags & PF_KTHREAD)
 495                         mm = NULL;
 496                 else
 497                         atomic_inc(&mm->mm_users);
 498         }
 499         task_unlock(task);
 500         return mm;
 501 }
 502 EXPORT_SYMBOL_GPL(get_task_mm);
 503
 504 /* Please note the differences between mmput and mm_release.
 505  * mmput is called whenever we stop holding onto a mm_struct,
 506  * error success whatever.
 507  *
 508  * mm_release is called after a mm_struct has been removed
 509  * from the current process.
 510  *
 511  * This difference is important for error handling, when we
 512  * only half set up a mm_struct for a new process and need to restore
 513  * the old one.  Because we mmput the new mm_struct before
 514  * restoring the old one. . .
 515  * Eric Biederman 10 January 1998
 516  */
 517 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 518 {
 519         struct completion *vfork_done = tsk->vfork_done;
 520
 521         /* Get rid of any cached register state */
 522         deactivate_mm(tsk, mm);
 523
 524         /* notify parent sleeping on vfork() */
 525         if (vfork_done) {
 526                 tsk->vfork_done = NULL;
 527                 complete(vfork_done);
 528         }
 529
 530         /*
 531          * If we're exiting normally, clear a user-space tid field if
 532          * requested.  We leave this alone when dying by signal, to leave
 533          * the value intact in a core dump, and to save the unnecessary
 534          * trouble otherwise.  Userland only wants this done for a sys_exit.
 535          */
 536         if (tsk->clear_child_tid
 537             && !(tsk->flags & PF_SIGNALED)
 538             && atomic_read(&mm->mm_users) > 1) {
 539                 u32 __user * tidptr = tsk->clear_child_tid;
 540                 tsk->clear_child_tid = NULL;
 541
 542                 /*
 543                  * We don't check the error code - if userspace has
 544                  * not set up a proper pointer then tough luck.
 545                  */
 546                 put_user(0, tidptr);
 547                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 548         }
 549 }
 550
 551 /*
 552  * Allocate a new mm structure and copy contents from the
 553  * mm structure of the passed in task structure.
 554  */
 555 struct mm_struct *dup_mm(struct task_struct *tsk)
 556 {
 557         struct mm_struct *mm, *oldmm = current->mm;
 558         int err;
 559
 560         if (!oldmm)
 561                 return NULL;
 562
 563         mm = allocate_mm();
 564         if (!mm)
 565                 goto fail_nomem;
 566
 567         memcpy(mm, oldmm, sizeof(*mm));
 568
 569         /* Initializing for Swap token stuff */
 570         mm->token_priority = 0;
 571         mm->last_interval = 0;
 572
 573         if (!mm_init(mm, tsk))
 574                 goto fail_nomem;
 575
 576         if (init_new_context(tsk, mm))
 577                 goto fail_nocontext;
 578
 579         dup_mm_exe_file(oldmm, mm);
 580
 581         err = dup_mmap(mm, oldmm);
 582         if (err)
 583                 goto free_pt;
 584
 585         mm->hiwater_rss = get_mm_rss(mm);
 586         mm->hiwater_vm = mm->total_vm;
 587
 588         return mm;
 589
 590 free_pt:
 591         mmput(mm);
 592
 593 fail_nomem:
 594         return NULL;
 595
 596 fail_nocontext:
 597         /*
 598          * If init_new_context() failed, we cannot use mmput() to free the mm
 599          * because it calls destroy_context()
 600          */
 601         mm_free_pgd(mm);
 602         free_mm(mm);
 603         return NULL;
 604 }
 605
 606 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 607 {
 608         struct mm_struct * mm, *oldmm;
 609         int retval;
 610
 611         tsk->min_flt = tsk->maj_flt = 0;
 612         tsk->nvcsw = tsk->nivcsw = 0;
 613
 614         tsk->mm = NULL;
 615         tsk->active_mm = NULL;
 616
 617         /*
 618          * Are we cloning a kernel thread?
 619          *
 620          * We need to steal a active VM for that..
 621          */
 622         oldmm = current->mm;
 623         if (!oldmm)
 624                 return 0;
 625
 626         if (clone_flags & CLONE_VM) {
 627                 atomic_inc(&oldmm->mm_users);
 628                 mm = oldmm;
 629                 goto good_mm;
 630         }
 631
 632         retval = -ENOMEM;
 633         mm = dup_mm(tsk);
 634         if (!mm)
 635                 goto fail_nomem;
 636
 637 good_mm:
 638         /* Initializing for Swap token stuff */
 639         mm->token_priority = 0;
 640         mm->last_interval = 0;
 641
 642         tsk->mm = mm;
 643         tsk->active_mm = mm;
 644         return 0;
 645
 646 fail_nomem:
 647         return retval;
 648 }
 649
 650 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 651 {
 652         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 653         /* We don't need to lock fs - think why ;-) */
 654         if (fs) {
 655                 atomic_set(&fs->count, 1);
 656                 rwlock_init(&fs->lock);
 657                 fs->umask = old->umask;
 658                 read_lock(&old->lock);
 659                 fs->root = old->root;
 660                 path_get(&old->root);
 661                 fs->pwd = old->pwd;
 662                 path_get(&old->pwd);
 663                 read_unlock(&old->lock);
 664         }
 665         return fs;
 666 }
 667
 668 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 669 {
 670         return __copy_fs_struct(old);
 671 }
 672
 673 EXPORT_SYMBOL_GPL(copy_fs_struct);
 674
 675 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 676 {
 677         if (clone_flags & CLONE_FS) {
 678                 atomic_inc(&current->fs->count);
 679                 return 0;
 680         }
 681         tsk->fs = __copy_fs_struct(current->fs);
 682         if (!tsk->fs)
 683                 return -ENOMEM;
 684         return 0;
 685 }
 686
 687 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 688 {
 689         struct files_struct *oldf, *newf;
 690         int error = 0;
 691
 692         /*
 693          * A background process may not have any files ...
 694          */
 695         oldf = current->files;
 696         if (!oldf)
 697                 goto out;
 698
 699         if (clone_flags & CLONE_FILES) {
 700                 atomic_inc(&oldf->count);
 701                 goto out;
 702         }
 703
 704         newf = dup_fd(oldf, &error);
 705         if (!newf)
 706                 goto out;
 707
 708         tsk->files = newf;
 709         error = 0;
 710 out:
 711         return error;
 712 }
 713
 714 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 715 {
 716 #ifdef CONFIG_BLOCK
 717         struct io_context *ioc = current->io_context;
 718
 719         if (!ioc)
 720                 return 0;
 721         /*
 722          * Share io context with parent, if CLONE_IO is set
 723          */
 724         if (clone_flags & CLONE_IO) {
 725                 tsk->io_context = ioc_task_link(ioc);
 726                 if (unlikely(!tsk->io_context))
 727                         return -ENOMEM;
 728         } else if (ioprio_valid(ioc->ioprio)) {
 729                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 730                 if (unlikely(!tsk->io_context))
 731                         return -ENOMEM;
 732
 733                 tsk->io_context->ioprio = ioc->ioprio;
 734         }
 735 #endif
 736         return 0;
 737 }
 738
 739 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 740 {
 741         struct sighand_struct *sig;
 742
 743         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 744                 atomic_inc(&current->sighand->count);
 745                 return 0;
 746         }
 747         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 748         rcu_assign_pointer(tsk->sighand, sig);
 749         if (!sig)
 750                 return -ENOMEM;
 751         atomic_set(&sig->count, 1);
 752         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 753         return 0;
 754 }
 755
 756 void __cleanup_sighand(struct sighand_struct *sighand)
 757 {
 758         if (atomic_dec_and_test(&sighand->count))
 759                 kmem_cache_free(sighand_cachep, sighand);
 760 }
 761
 762
 763 /*
 764  * Initialize POSIX timer handling for a thread group.
 765  */
 766 static void posix_cpu_timers_init_group(struct signal_struct *sig)
 767 {
 768         /* Thread group counters. */
 769         thread_group_cputime_init(sig);
 770
 771         /* Expiration times and increments. */
 772         sig->it_virt_expires = cputime_zero;
 773         sig->it_virt_incr = cputime_zero;
 774         sig->it_prof_expires = cputime_zero;
 775         sig->it_prof_incr = cputime_zero;
 776
 777         /* Cached expiration times. */
 778         sig->cputime_expires.prof_exp = cputime_zero;
 779         sig->cputime_expires.virt_exp = cputime_zero;
 780         sig->cputime_expires.sched_exp = 0;
 781
 782         /* The timer lists. */
 783         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 784         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 785         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 786 }
 787
 788 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 789 {
 790         struct signal_struct *sig;
 791         int ret;
 792
 793         if (clone_flags & CLONE_THREAD) {
 794                 ret = thread_group_cputime_clone_thread(current);
 795                 if (likely(!ret)) {
 796                         atomic_inc(&current->signal->count);
 797                         atomic_inc(&current->signal->live);
 798                 }
 799                 return ret;
 800         }
 801         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 802         tsk->signal = sig;
 803         if (!sig)
 804                 return -ENOMEM;
 805
 806         ret = copy_thread_group_keys(tsk);
 807         if (ret < 0) {
 808                 kmem_cache_free(signal_cachep, sig);
 809                 return ret;
 810         }
 811
 812         atomic_set(&sig->count, 1);
 813         atomic_set(&sig->live, 1);
 814         init_waitqueue_head(&sig->wait_chldexit);
 815         sig->flags = 0;
 816         sig->group_exit_code = 0;
 817         sig->group_exit_task = NULL;
 818         sig->group_stop_count = 0;
 819         sig->curr_target = tsk;
 820         init_sigpending(&sig->shared_pending);
 821         INIT_LIST_HEAD(&sig->posix_timers);
 822
 823         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 824         sig->it_real_incr.tv64 = 0;
 825         sig->real_timer.function = it_real_fn;
 826
 827         sig->leader = 0;        /* session leadership doesn't inherit */
 828         sig->tty_old_pgrp = NULL;
 829
 830         sig->cutime = sig->cstime = cputime_zero;
 831         sig->gtime = cputime_zero;
 832         sig->cgtime = cputime_zero;
 833         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 834         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 835         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
 836         task_io_accounting_init(&sig->ioac);
 837         taskstats_tgid_init(sig);
 838
 839         task_lock(current->group_leader);
 840         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 841         task_unlock(current->group_leader);
 842
 843         posix_cpu_timers_init_group(sig);
 844
 845         acct_init_pacct(&sig->pacct);
 846
 847         tty_audit_fork(sig);
 848
 849         return 0;
 850 }
 851
 852 void __cleanup_signal(struct signal_struct *sig)
 853 {
 854         thread_group_cputime_free(sig);
 855         exit_thread_group_keys(sig);
 856         kmem_cache_free(signal_cachep, sig);
 857 }
 858
 859 static void cleanup_signal(struct task_struct *tsk)
 860 {
 861         struct signal_struct *sig = tsk->signal;
 862
 863         atomic_dec(&sig->live);
 864
 865         if (atomic_dec_and_test(&sig->count))
 866                 __cleanup_signal(sig);
 867 }
 868
 869 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 870 {
 871         unsigned long new_flags = p->flags;
 872
 873         new_flags &= ~PF_SUPERPRIV;
 874         new_flags |= PF_FORKNOEXEC;
 875         new_flags |= PF_STARTING;
 876         p->flags = new_flags;
 877         clear_freeze_flag(p);
 878 }
 879
 880 asmlinkage long sys_set_tid_address(int __user *tidptr)
 881 {
 882         current->clear_child_tid = tidptr;
 883
 884         return task_pid_vnr(current);
 885 }
 886
 887 static void rt_mutex_init_task(struct task_struct *p)
 888 {
 889         spin_lock_init(&p->pi_lock);
 890 #ifdef CONFIG_RT_MUTEXES
 891         plist_head_init(&p->pi_waiters, &p->pi_lock);
 892         p->pi_blocked_on = NULL;
 893 #endif
 894 }
 895
 896 #ifdef CONFIG_MM_OWNER
 897 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 898 {
 899         mm->owner = p;
 900 }
 901 #endif /* CONFIG_MM_OWNER */
 902
 903 /*
 904  * Initialize POSIX timer handling for a single task.
 905  */
 906 static void posix_cpu_timers_init(struct task_struct *tsk)
 907 {
 908         tsk->cputime_expires.prof_exp = cputime_zero;
 909         tsk->cputime_expires.virt_exp = cputime_zero;
 910         tsk->cputime_expires.sched_exp = 0;
 911         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
 912         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
 913         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
 914 }
 915
 916 /*
 917  * This creates a new process as a copy of the old one,
 918  * but does not actually start it yet.
 919  *
 920  * It copies the registers, and all the appropriate
 921  * parts of the process environment (as per the clone
 922  * flags). The actual kick-off is left to the caller.
 923  */
 924 static struct task_struct *copy_process(unsigned long clone_flags,
 925                                         unsigned long stack_start,
 926                                         struct pt_regs *regs,
 927                                         unsigned long stack_size,
 928                                         int __user *child_tidptr,
 929                                         struct pid *pid,
 930                                         int trace)
 931 {
 932         int retval;
 933         struct task_struct *p;
 934         int cgroup_callbacks_done = 0;
 935
 936         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 937                 return ERR_PTR(-EINVAL);
 938
 939         /*
 940          * Thread groups must share signals as well, and detached threads
 941          * can only be started up within the thread group.
 942          */
 943         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 944                 return ERR_PTR(-EINVAL);
 945
 946         /*
 947          * Shared signal handlers imply shared VM. By way of the above,
 948          * thread groups also imply shared VM. Blocking this case allows
 949          * for various simplifications in other code.
 950          */
 951         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 952                 return ERR_PTR(-EINVAL);
 953
 954         retval = security_task_create(clone_flags);
 955         if (retval)
 956                 goto fork_out;
 957
 958         retval = -ENOMEM;
 959         p = dup_task_struct(current);
 960         if (!p)
 961                 goto fork_out;
 962
 963         rt_mutex_init_task(p);
 964
 965 #ifdef CONFIG_PROVE_LOCKING
 966         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
 967         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
 968 #endif
 969         retval = -EAGAIN;
 970         if (atomic_read(&p->user->processes) >=
 971                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
 972                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 973                     p->user != current->nsproxy->user_ns->root_user)
 974                         goto bad_fork_free;
 975         }
 976
 977         atomic_inc(&p->user->__count);
 978         atomic_inc(&p->user->processes);
 979         get_group_info(p->group_info);
 980
 981         /*
 982          * If multiple threads are within copy_process(), then this check
 983          * triggers too late. This doesn't hurt, the check is only there
 984          * to stop root fork bombs.
 985          */
 986         if (nr_threads >= max_threads)
 987                 goto bad_fork_cleanup_count;
 988
 989         if (!try_module_get(task_thread_info(p)->exec_domain->module))
 990                 goto bad_fork_cleanup_count;
 991
 992         if (p->binfmt && !try_module_get(p->binfmt->module))
 993                 goto bad_fork_cleanup_put_domain;
 994
 995         p->did_exec = 0;
 996         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
 997         copy_flags(clone_flags, p);
 998         INIT_LIST_HEAD(&p->children);
 999         INIT_LIST_HEAD(&p->sibling);
1000 #ifdef CONFIG_PREEMPT_RCU
1001         p->rcu_read_lock_nesting = 0;
1002         p->rcu_flipctr_idx = 0;
1003 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1004         p->vfork_done = NULL;
1005         spin_lock_init(&p->alloc_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->gtime = cputime_zero;
1013         p->utimescaled = cputime_zero;
1014         p->stimescaled = cputime_zero;
1015         p->prev_utime = cputime_zero;
1016         p->prev_stime = cputime_zero;
1017
1018 #ifdef CONFIG_DETECT_SOFTLOCKUP
1019         p->last_switch_count = 0;
1020         p->last_switch_timestamp = 0;
1021 #endif
1022
1023         task_io_accounting_init(&p->ioac);
1024         acct_clear_integrals(p);
1025
1026         posix_cpu_timers_init(p);
1027
1028         p->lock_depth = -1;             /* -1 = no lock */
1029         do_posix_clock_monotonic_gettime(&p->start_time);
1030         p->real_start_time = p->start_time;
1031         monotonic_to_bootbased(&p->real_start_time);
1032 #ifdef CONFIG_SECURITY
1033         p->security = NULL;
1034 #endif
1035         p->cap_bset = current->cap_bset;
1036         p->io_context = NULL;
1037         p->audit_context = NULL;
1038         cgroup_fork(p);
1039 #ifdef CONFIG_NUMA
1040         p->mempolicy = mpol_dup(p->mempolicy);
1041         if (IS_ERR(p->mempolicy)) {
1042                 retval = PTR_ERR(p->mempolicy);
1043                 p->mempolicy = NULL;
1044                 goto bad_fork_cleanup_cgroup;
1045         }
1046         mpol_fix_fork_child_flag(p);
1047 #endif
1048 #ifdef CONFIG_TRACE_IRQFLAGS
1049         p->irq_events = 0;
1050 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1051         p->hardirqs_enabled = 1;
1052 #else
1053         p->hardirqs_enabled = 0;
1054 #endif
1055         p->hardirq_enable_ip = 0;
1056         p->hardirq_enable_event = 0;
1057         p->hardirq_disable_ip = _THIS_IP_;
1058         p->hardirq_disable_event = 0;
1059         p->softirqs_enabled = 1;
1060         p->softirq_enable_ip = _THIS_IP_;
1061         p->softirq_enable_event = 0;
1062         p->softirq_disable_ip = 0;
1063         p->softirq_disable_event = 0;
1064         p->hardirq_context = 0;
1065         p->softirq_context = 0;
1066 #endif
1067 #ifdef CONFIG_LOCKDEP
1068         p->lockdep_depth = 0; /* no locks held yet */
1069         p->curr_chain_key = 0;
1070         p->lockdep_recursion = 0;
1071 #endif
1072
1073 #ifdef CONFIG_DEBUG_MUTEXES
1074         p->blocked_on = NULL; /* not blocked yet */
1075 #endif
1076
1077         /* Perform scheduler related setup. Assign this task to a CPU. */
1078         sched_fork(p, clone_flags);
1079
1080         if ((retval = security_task_alloc(p)))
1081                 goto bad_fork_cleanup_policy;
1082         if ((retval = audit_alloc(p)))
1083                 goto bad_fork_cleanup_security;
1084         /* copy all the process information */
1085         if ((retval = copy_semundo(clone_flags, p)))
1086                 goto bad_fork_cleanup_audit;
1087         if ((retval = copy_files(clone_flags, p)))
1088                 goto bad_fork_cleanup_semundo;
1089         if ((retval = copy_fs(clone_flags, p)))
1090                 goto bad_fork_cleanup_files;
1091         if ((retval = copy_sighand(clone_flags, p)))
1092                 goto bad_fork_cleanup_fs;
1093         if ((retval = copy_signal(clone_flags, p)))
1094                 goto bad_fork_cleanup_sighand;
1095         if ((retval = copy_mm(clone_flags, p)))
1096                 goto bad_fork_cleanup_signal;
1097         if ((retval = copy_keys(clone_flags, p)))
1098                 goto bad_fork_cleanup_mm;
1099         if ((retval = copy_namespaces(clone_flags, p)))
1100                 goto bad_fork_cleanup_keys;
1101         if ((retval = copy_io(clone_flags, p)))
1102                 goto bad_fork_cleanup_namespaces;
1103         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1104         if (retval)
1105                 goto bad_fork_cleanup_io;
1106
1107         if (pid != &init_struct_pid) {
1108                 retval = -ENOMEM;
1109                 pid = alloc_pid(task_active_pid_ns(p));
1110                 if (!pid)
1111                         goto bad_fork_cleanup_io;
1112
1113                 if (clone_flags & CLONE_NEWPID) {
1114                         retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1115                         if (retval < 0)
1116                                 goto bad_fork_free_pid;
1117                 }
1118         }
1119
1120         p->pid = pid_nr(pid);
1121         p->tgid = p->pid;
1122         if (clone_flags & CLONE_THREAD)
1123                 p->tgid = current->tgid;
1124
1125         if (current->nsproxy != p->nsproxy) {
1126                 retval = ns_cgroup_clone(p, pid);
1127                 if (retval)
1128                         goto bad_fork_free_pid;
1129         }
1130
1131         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1132         /*
1133          * Clear TID on mm_release()?
1134          */
1135         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1136 #ifdef CONFIG_FUTEX
1137         p->robust_list = NULL;
1138 #ifdef CONFIG_COMPAT
1139         p->compat_robust_list = NULL;
1140 #endif
1141         INIT_LIST_HEAD(&p->pi_state_list);
1142         p->pi_state_cache = NULL;
1143 #endif
1144         /*
1145          * sigaltstack should be cleared when sharing the same VM
1146          */
1147         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1148                 p->sas_ss_sp = p->sas_ss_size = 0;
1149
1150         /*
1151          * Syscall tracing should be turned off in the child regardless
1152          * of CLONE_PTRACE.
1153          */
1154         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1155 #ifdef TIF_SYSCALL_EMU
1156         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1157 #endif
1158         clear_all_latency_tracing(p);
1159
1160         /* Our parent execution domain becomes current domain
1161            These must match for thread signalling to apply */
1162         p->parent_exec_id = p->self_exec_id;
1163
1164         /* ok, now we should be set up.. */
1165         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1166         p->pdeath_signal = 0;
1167         p->exit_state = 0;
1168
1169         /*
1170          * Ok, make it visible to the rest of the system.
1171          * We dont wake it up yet.
1172          */
1173         p->group_leader = p;
1174         INIT_LIST_HEAD(&p->thread_group);
1175
1176         /* Now that the task is set up, run cgroup callbacks if
1177          * necessary. We need to run them before the task is visible
1178          * on the tasklist. */
1179         cgroup_fork_callbacks(p);
1180         cgroup_callbacks_done = 1;
1181
1182         /* Need tasklist lock for parent etc handling! */
1183         write_lock_irq(&tasklist_lock);
1184
1185         /*
1186          * The task hasn't been attached yet, so its cpus_allowed mask will
1187          * not be changed, nor will its assigned CPU.
1188          *
1189          * The cpus_allowed mask of the parent may have changed after it was
1190          * copied first time - so re-copy it here, then check the child's CPU
1191          * to ensure it is on a valid CPU (and if not, just force it back to
1192          * parent's CPU). This avoids alot of nasty races.
1193          */
1194         p->cpus_allowed = current->cpus_allowed;
1195         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1196         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1197                         !cpu_online(task_cpu(p))))
1198                 set_task_cpu(p, smp_processor_id());
1199
1200         /* CLONE_PARENT re-uses the old parent */
1201         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1202                 p->real_parent = current->real_parent;
1203         else
1204                 p->real_parent = current;
1205
1206         spin_lock(&current->sighand->siglock);
1207
1208         /*
1209          * Process group and session signals need to be delivered to just the
1210          * parent before the fork or both the parent and the child after the
1211          * fork. Restart if a signal comes in before we add the new process to
1212          * it's process group.
1213          * A fatal signal pending means that current will exit, so the new
1214          * thread can't slip out of an OOM kill (or normal SIGKILL).
1215          */
1216         recalc_sigpending();
1217         if (signal_pending(current)) {
1218                 spin_unlock(&current->sighand->siglock);
1219                 write_unlock_irq(&tasklist_lock);
1220                 retval = -ERESTARTNOINTR;
1221                 goto bad_fork_free_pid;
1222         }
1223
1224         if (clone_flags & CLONE_THREAD) {
1225                 p->group_leader = current->group_leader;
1226                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1227         }
1228
1229         if (likely(p->pid)) {
1230                 list_add_tail(&p->sibling, &p->real_parent->children);
1231                 tracehook_finish_clone(p, clone_flags, trace);
1232
1233                 if (thread_group_leader(p)) {
1234                         if (clone_flags & CLONE_NEWPID)
1235                                 p->nsproxy->pid_ns->child_reaper = p;
1236
1237                         p->signal->leader_pid = pid;
1238                         p->signal->tty = current->signal->tty;
1239                         set_task_pgrp(p, task_pgrp_nr(current));
1240                         set_task_session(p, task_session_nr(current));
1241                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1242                         attach_pid(p, PIDTYPE_SID, task_session(current));
1243                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1244                         __get_cpu_var(process_counts)++;
1245                 }
1246                 attach_pid(p, PIDTYPE_PID, pid);
1247                 nr_threads++;
1248         }
1249
1250         total_forks++;
1251         spin_unlock(&current->sighand->siglock);
1252         write_unlock_irq(&tasklist_lock);
1253         proc_fork_connector(p);
1254         cgroup_post_fork(p);
1255         return p;
1256
1257 bad_fork_free_pid:
1258         if (pid != &init_struct_pid)
1259                 free_pid(pid);
1260 bad_fork_cleanup_io:
1261         put_io_context(p->io_context);
1262 bad_fork_cleanup_namespaces:
1263         exit_task_namespaces(p);
1264 bad_fork_cleanup_keys:
1265         exit_keys(p);
1266 bad_fork_cleanup_mm:
1267         if (p->mm)
1268                 mmput(p->mm);
1269 bad_fork_cleanup_signal:
1270         cleanup_signal(p);
1271 bad_fork_cleanup_sighand:
1272         __cleanup_sighand(p->sighand);
1273 bad_fork_cleanup_fs:
1274         exit_fs(p); /* blocking */
1275 bad_fork_cleanup_files:
1276         exit_files(p); /* blocking */
1277 bad_fork_cleanup_semundo:
1278         exit_sem(p);
1279 bad_fork_cleanup_audit:
1280         audit_free(p);
1281 bad_fork_cleanup_security:
1282         security_task_free(p);
1283 bad_fork_cleanup_policy:
1284 #ifdef CONFIG_NUMA
1285         mpol_put(p->mempolicy);
1286 bad_fork_cleanup_cgroup:
1287 #endif
1288         cgroup_exit(p, cgroup_callbacks_done);
1289         delayacct_tsk_free(p);
1290         if (p->binfmt)
1291                 module_put(p->binfmt->module);
1292 bad_fork_cleanup_put_domain:
1293         module_put(task_thread_info(p)->exec_domain->module);
1294 bad_fork_cleanup_count:
1295         put_group_info(p->group_info);
1296         atomic_dec(&p->user->processes);
1297         free_uid(p->user);
1298 bad_fork_free:
1299         free_task(p);
1300 fork_out:
1301         return ERR_PTR(retval);
1302 }
1303
1304 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1305 {
1306         memset(regs, 0, sizeof(struct pt_regs));
1307         return regs;
1308 }
1309
1310 struct task_struct * __cpuinit fork_idle(int cpu)
1311 {
1312         struct task_struct *task;
1313         struct pt_regs regs;
1314
1315         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1316                             &init_struct_pid, 0);
1317         if (!IS_ERR(task))
1318                 init_idle(task, cpu);
1319
1320         return task;
1321 }
1322
1323 /*
1324  *  Ok, this is the main fork-routine.
1325  *
1326  * It copies the process, and if successful kick-starts
1327  * it and waits for it to finish using the VM if required.
1328  */
1329 long do_fork(unsigned long clone_flags,
1330               unsigned long stack_start,
1331               struct pt_regs *regs,
1332               unsigned long stack_size,
1333               int __user *parent_tidptr,
1334               int __user *child_tidptr)
1335 {
1336         struct task_struct *p;
1337         int trace = 0;
1338         long nr;
1339
1340         /*
1341          * We hope to recycle these flags after 2.6.26
1342          */
1343         if (unlikely(clone_flags & CLONE_STOPPED)) {
1344                 static int __read_mostly count = 100;
1345
1346                 if (count > 0 && printk_ratelimit()) {
1347                         char comm[TASK_COMM_LEN];
1348
1349                         count--;
1350                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1351                                         "clone flags 0x%lx\n",
1352                                 get_task_comm(comm, current),
1353                                 clone_flags & CLONE_STOPPED);
1354                 }
1355         }
1356
1357         /*
1358          * When called from kernel_thread, don't do user tracing stuff.
1359          */
1360         if (likely(user_mode(regs)))
1361                 trace = tracehook_prepare_clone(clone_flags);
1362
1363         p = copy_process(clone_flags, stack_start, regs, stack_size,
1364                          child_tidptr, NULL, trace);
1365         /*
1366          * Do this prior waking up the new thread - the thread pointer
1367          * might get invalid after that point, if the thread exits quickly.
1368          */
1369         if (!IS_ERR(p)) {
1370                 struct completion vfork;
1371
1372                 nr = task_pid_vnr(p);
1373
1374                 if (clone_flags & CLONE_PARENT_SETTID)
1375                         put_user(nr, parent_tidptr);
1376
1377                 if (clone_flags & CLONE_VFORK) {
1378                         p->vfork_done = &vfork;
1379                         init_completion(&vfork);
1380                 }
1381
1382                 tracehook_report_clone(trace, regs, clone_flags, nr, p);
1383
1384                 /*
1385                  * We set PF_STARTING at creation in case tracing wants to
1386                  * use this to distinguish a fully live task from one that
1387                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1388                  * clear it and set the child going.
1389                  */
1390                 p->flags &= ~PF_STARTING;
1391
1392                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1393                         /*
1394                          * We'll start up with an immediate SIGSTOP.
1395                          */
1396                         sigaddset(&p->pending.signal, SIGSTOP);
1397                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1398                         __set_task_state(p, TASK_STOPPED);
1399                 } else {
1400                         wake_up_new_task(p, clone_flags);
1401                 }
1402
1403                 tracehook_report_clone_complete(trace, regs,
1404                                                 clone_flags, nr, p);
1405
1406                 if (clone_flags & CLONE_VFORK) {
1407                         freezer_do_not_count();
1408                         wait_for_completion(&vfork);
1409                         freezer_count();
1410                         tracehook_report_vfork_done(p, nr);
1411                 }
1412         } else {
1413                 nr = PTR_ERR(p);
1414         }
1415         return nr;
1416 }
1417
1418 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1419 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1420 #endif
1421
1422 static void sighand_ctor(void *data)
1423 {
1424         struct sighand_struct *sighand = data;
1425
1426         spin_lock_init(&sighand->siglock);
1427         init_waitqueue_head(&sighand->signalfd_wqh);
1428 }
1429
1430 void __init proc_caches_init(void)
1431 {
1432         sighand_cachep = kmem_cache_create("sighand_cache",
1433                         sizeof(struct sighand_struct), 0,
1434                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1435                         sighand_ctor);
1436         signal_cachep = kmem_cache_create("signal_cache",
1437                         sizeof(struct signal_struct), 0,
1438                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1439         files_cachep = kmem_cache_create("files_cache",
1440                         sizeof(struct files_struct), 0,
1441                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1442         fs_cachep = kmem_cache_create("fs_cache",
1443                         sizeof(struct fs_struct), 0,
1444                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1445         vm_area_cachep = kmem_cache_create("vm_area_struct",
1446                         sizeof(struct vm_area_struct), 0,
1447                         SLAB_PANIC, NULL);
1448         mm_cachep = kmem_cache_create("mm_struct",
1449                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1450                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1451 }
1452
1453 /*
1454  * Check constraints on flags passed to the unshare system call and
1455  * force unsharing of additional process context as appropriate.
1456  */
1457 static void check_unshare_flags(unsigned long *flags_ptr)
1458 {
1459         /*
1460          * If unsharing a thread from a thread group, must also
1461          * unshare vm.
1462          */
1463         if (*flags_ptr & CLONE_THREAD)
1464                 *flags_ptr |= CLONE_VM;
1465
1466         /*
1467          * If unsharing vm, must also unshare signal handlers.
1468          */
1469         if (*flags_ptr & CLONE_VM)
1470                 *flags_ptr |= CLONE_SIGHAND;
1471
1472         /*
1473          * If unsharing signal handlers and the task was created
1474          * using CLONE_THREAD, then must unshare the thread
1475          */
1476         if ((*flags_ptr & CLONE_SIGHAND) &&
1477             (atomic_read(&current->signal->count) > 1))
1478                 *flags_ptr |= CLONE_THREAD;
1479
1480         /*
1481          * If unsharing namespace, must also unshare filesystem information.
1482          */
1483         if (*flags_ptr & CLONE_NEWNS)
1484                 *flags_ptr |= CLONE_FS;
1485 }
1486
1487 /*
1488  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1489  */
1490 static int unshare_thread(unsigned long unshare_flags)
1491 {
1492         if (unshare_flags & CLONE_THREAD)
1493                 return -EINVAL;
1494
1495         return 0;
1496 }
1497
1498 /*
1499  * Unshare the filesystem structure if it is being shared
1500  */
1501 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1502 {
1503         struct fs_struct *fs = current->fs;
1504
1505         if ((unshare_flags & CLONE_FS) &&
1506             (fs && atomic_read(&fs->count) > 1)) {
1507                 *new_fsp = __copy_fs_struct(current->fs);
1508                 if (!*new_fsp)
1509                         return -ENOMEM;
1510         }
1511
1512         return 0;
1513 }
1514
1515 /*
1516  * Unsharing of sighand is not supported yet
1517  */
1518 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1519 {
1520         struct sighand_struct *sigh = current->sighand;
1521
1522         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1523                 return -EINVAL;
1524         else
1525                 return 0;
1526 }
1527
1528 /*
1529  * Unshare vm if it is being shared
1530  */
1531 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1532 {
1533         struct mm_struct *mm = current->mm;
1534
1535         if ((unshare_flags & CLONE_VM) &&
1536             (mm && atomic_read(&mm->mm_users) > 1)) {
1537                 return -EINVAL;
1538         }
1539
1540         return 0;
1541 }
1542
1543 /*
1544  * Unshare file descriptor table if it is being shared
1545  */
1546 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1547 {
1548         struct files_struct *fd = current->files;
1549         int error = 0;
1550
1551         if ((unshare_flags & CLONE_FILES) &&
1552             (fd && atomic_read(&fd->count) > 1)) {
1553                 *new_fdp = dup_fd(fd, &error);
1554                 if (!*new_fdp)
1555                         return error;
1556         }
1557
1558         return 0;
1559 }
1560
1561 /*
1562  * unshare allows a process to 'unshare' part of the process
1563  * context which was originally shared using clone.  copy_*
1564  * functions used by do_fork() cannot be used here directly
1565  * because they modify an inactive task_struct that is being
1566  * constructed. Here we are modifying the current, active,
1567  * task_struct.
1568  */
1569 asmlinkage long sys_unshare(unsigned long unshare_flags)
1570 {
1571         int err = 0;
1572         struct fs_struct *fs, *new_fs = NULL;
1573         struct sighand_struct *new_sigh = NULL;
1574         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1575         struct files_struct *fd, *new_fd = NULL;
1576         struct nsproxy *new_nsproxy = NULL;
1577         int do_sysvsem = 0;
1578
1579         check_unshare_flags(&unshare_flags);
1580
1581         /* Return -EINVAL for all unsupported flags */
1582         err = -EINVAL;
1583         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1584                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1585                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
1586                                 CLONE_NEWNET))
1587                 goto bad_unshare_out;
1588
1589         /*
1590          * CLONE_NEWIPC must also detach from the undolist: after switching
1591          * to a new ipc namespace, the semaphore arrays from the old
1592          * namespace are unreachable.
1593          */
1594         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1595                 do_sysvsem = 1;
1596         if ((err = unshare_thread(unshare_flags)))
1597                 goto bad_unshare_out;
1598         if ((err = unshare_fs(unshare_flags, &new_fs)))
1599                 goto bad_unshare_cleanup_thread;
1600         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1601                 goto bad_unshare_cleanup_fs;
1602         if ((err = unshare_vm(unshare_flags, &new_mm)))
1603                 goto bad_unshare_cleanup_sigh;
1604         if ((err = unshare_fd(unshare_flags, &new_fd)))
1605                 goto bad_unshare_cleanup_vm;
1606         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1607                         new_fs)))
1608                 goto bad_unshare_cleanup_fd;
1609
1610         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1611                 if (do_sysvsem) {
1612                         /*
1613                          * CLONE_SYSVSEM is equivalent to sys_exit().
1614                          */
1615                         exit_sem(current);
1616                 }
1617
1618                 if (new_nsproxy) {
1619                         switch_task_namespaces(current, new_nsproxy);
1620                         new_nsproxy = NULL;
1621                 }
1622
1623                 task_lock(current);
1624
1625                 if (new_fs) {
1626                         fs = current->fs;
1627                         current->fs = new_fs;
1628                         new_fs = fs;
1629                 }
1630
1631                 if (new_mm) {
1632                         mm = current->mm;
1633                         active_mm = current->active_mm;
1634                         current->mm = new_mm;
1635                         current->active_mm = new_mm;
1636                         activate_mm(active_mm, new_mm);
1637                         new_mm = mm;
1638                 }
1639
1640                 if (new_fd) {
1641                         fd = current->files;
1642                         current->files = new_fd;
1643                         new_fd = fd;
1644                 }
1645
1646                 task_unlock(current);
1647         }
1648
1649         if (new_nsproxy)
1650                 put_nsproxy(new_nsproxy);
1651
1652 bad_unshare_cleanup_fd:
1653         if (new_fd)
1654                 put_files_struct(new_fd);
1655
1656 bad_unshare_cleanup_vm:
1657         if (new_mm)
1658                 mmput(new_mm);
1659
1660 bad_unshare_cleanup_sigh:
1661         if (new_sigh)
1662                 if (atomic_dec_and_test(&new_sigh->count))
1663                         kmem_cache_free(sighand_cachep, new_sigh);
1664
1665 bad_unshare_cleanup_fs:
1666         if (new_fs)
1667                 put_fs_struct(new_fs);
1668
1669 bad_unshare_cleanup_thread:
1670 bad_unshare_out:
1671         return err;
1672 }
1673
1674 /*
1675  *      Helper to unshare the files of the current task.
1676  *      We don't want to expose copy_files internals to
1677  *      the exec layer of the kernel.
1678  */
1679
1680 int unshare_files(struct files_struct **displaced)
1681 {
1682         struct task_struct *task = current;
1683         struct files_struct *copy = NULL;
1684         int error;
1685
1686         error = unshare_fd(CLONE_FILES, &copy);
1687         if (error || !copy) {
1688                 *displaced = NULL;
1689                 return error;
1690         }
1691         *displaced = task->files;
1692         task_lock(task);
1693         task->files = copy;
1694         task_unlock(task);
1695         return 0;
1696 }