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/compat.h>
  44 #include <linux/task_io_accounting_ops.h>
  45 #include <linux/rcupdate.h>
  46 #include <linux/ptrace.h>
  47 #include <linux/mount.h>
  48 #include <linux/audit.h>
  49 #include <linux/memcontrol.h>
  50 #include <linux/ftrace.h>
  51 #include <linux/profile.h>
  52 #include <linux/rmap.h>
  53 #include <linux/acct.h>
  54 #include <linux/tsacct_kern.h>
  55 #include <linux/cn_proc.h>
  56 #include <linux/freezer.h>
  57 #include <linux/delayacct.h>
  58 #include <linux/taskstats_kern.h>
  59 #include <linux/random.h>
  60 #include <linux/tty.h>
  61 #include <linux/proc_fs.h>
  62 #include <linux/blkdev.h>
  63 #include <trace/sched.h>
  64 #include <linux/magic.h>
  65
  66 #include <asm/pgtable.h>
  67 #include <asm/pgalloc.h>
  68 #include <asm/uaccess.h>
  69 #include <asm/mmu_context.h>
  70 #include <asm/cacheflush.h>
  71 #include <asm/tlbflush.h>
  72
  73 /*
  74  * Protected counters by write_lock_irq(&tasklist_lock)
  75  */
  76 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  77 int nr_threads;                 /* The idle threads do not count.. */
  78
  79 int max_threads;                /* tunable limit on nr_threads */
  80
  81 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  82
  83 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  84
  85 DEFINE_TRACE(sched_process_fork);
  86
  87 int nr_processes(void)
  88 {
  89         int cpu;
  90         int total = 0;
  91
  92         for_each_online_cpu(cpu)
  93                 total += per_cpu(process_counts, cpu);
  94
  95         return total;
  96 }
  97
  98 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  99 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
 100 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
 101 static struct kmem_cache *task_struct_cachep;
 102 #endif
 103
 104 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 105 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
 106 {
 107 #ifdef CONFIG_DEBUG_STACK_USAGE
 108         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 109 #else
 110         gfp_t mask = GFP_KERNEL;
 111 #endif
 112         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
 113 }
 114
 115 static inline void free_thread_info(struct thread_info *ti)
 116 {
 117         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 118 }
 119 #endif
 120
 121 /* SLAB cache for signal_struct structures (tsk->signal) */
 122 static struct kmem_cache *signal_cachep;
 123
 124 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 125 struct kmem_cache *sighand_cachep;
 126
 127 /* SLAB cache for files_struct structures (tsk->files) */
 128 struct kmem_cache *files_cachep;
 129
 130 /* SLAB cache for fs_struct structures (tsk->fs) */
 131 struct kmem_cache *fs_cachep;
 132
 133 /* SLAB cache for vm_area_struct structures */
 134 struct kmem_cache *vm_area_cachep;
 135
 136 /* SLAB cache for mm_struct structures (tsk->mm) */
 137 static struct kmem_cache *mm_cachep;
 138
 139 void free_task(struct task_struct *tsk)
 140 {
 141         prop_local_destroy_single(&tsk->dirties);
 142         free_thread_info(tsk->stack);
 143         rt_mutex_debug_task_free(tsk);
 144         ftrace_graph_exit_task(tsk);
 145         free_task_struct(tsk);
 146 }
 147 EXPORT_SYMBOL(free_task);
 148
 149 void __put_task_struct(struct task_struct *tsk)
 150 {
 151         WARN_ON(!tsk->exit_state);
 152         WARN_ON(atomic_read(&tsk->usage));
 153         WARN_ON(tsk == current);
 154
 155         put_cred(tsk->real_cred);
 156         put_cred(tsk->cred);
 157         delayacct_tsk_free(tsk);
 158
 159         if (!profile_handoff_task(tsk))
 160                 free_task(tsk);
 161 }
 162
 163 /*
 164  * macro override instead of weak attribute alias, to workaround
 165  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 166  */
 167 #ifndef arch_task_cache_init
 168 #define arch_task_cache_init()
 169 #endif
 170
 171 void __init fork_init(unsigned long mempages)
 172 {
 173 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 174 #ifndef ARCH_MIN_TASKALIGN
 175 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 176 #endif
 177         /* create a slab on which task_structs can be allocated */
 178         task_struct_cachep =
 179                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 180                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
 181 #endif
 182
 183         /* do the arch specific task caches init */
 184         arch_task_cache_init();
 185
 186         /*
 187          * The default maximum number of threads is set to a safe
 188          * value: the thread structures can take up at most half
 189          * of memory.
 190          */
 191         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 192
 193         /*
 194          * we need to allow at least 20 threads to boot a system
 195          */
 196         if(max_threads < 20)
 197                 max_threads = 20;
 198
 199         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 200         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 201         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 202                 init_task.signal->rlim[RLIMIT_NPROC];
 203 }
 204
 205 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 206                                                struct task_struct *src)
 207 {
 208         *dst = *src;
 209         return 0;
 210 }
 211
 212 static struct task_struct *dup_task_struct(struct task_struct *orig)
 213 {
 214         struct task_struct *tsk;
 215         struct thread_info *ti;
 216         unsigned long *stackend;
 217
 218         int err;
 219
 220         prepare_to_copy(orig);
 221
 222         tsk = alloc_task_struct();
 223         if (!tsk)
 224                 return NULL;
 225
 226         ti = alloc_thread_info(tsk);
 227         if (!ti) {
 228                 free_task_struct(tsk);
 229                 return NULL;
 230         }
 231
 232         err = arch_dup_task_struct(tsk, orig);
 233         if (err)
 234                 goto out;
 235
 236         tsk->stack = ti;
 237
 238         err = prop_local_init_single(&tsk->dirties);
 239         if (err)
 240                 goto out;
 241
 242         setup_thread_stack(tsk, orig);
 243         stackend = end_of_stack(tsk);
 244         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 245
 246 #ifdef CONFIG_CC_STACKPROTECTOR
 247         tsk->stack_canary = get_random_int();
 248 #endif
 249
 250         /* One for us, one for whoever does the "release_task()" (usually parent) */
 251         atomic_set(&tsk->usage,2);
 252         atomic_set(&tsk->fs_excl, 0);
 253 #ifdef CONFIG_BLK_DEV_IO_TRACE
 254         tsk->btrace_seq = 0;
 255 #endif
 256         tsk->splice_pipe = NULL;
 257         return tsk;
 258
 259 out:
 260         free_thread_info(ti);
 261         free_task_struct(tsk);
 262         return NULL;
 263 }
 264
 265 #ifdef CONFIG_MMU
 266 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 267 {
 268         struct vm_area_struct *mpnt, *tmp, **pprev;
 269         struct rb_node **rb_link, *rb_parent;
 270         int retval;
 271         unsigned long charge;
 272         struct mempolicy *pol;
 273
 274         down_write(&oldmm->mmap_sem);
 275         flush_cache_dup_mm(oldmm);
 276         /*
 277          * Not linked in yet - no deadlock potential:
 278          */
 279         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 280
 281         mm->locked_vm = 0;
 282         mm->mmap = NULL;
 283         mm->mmap_cache = NULL;
 284         mm->free_area_cache = oldmm->mmap_base;
 285         mm->cached_hole_size = ~0UL;
 286         mm->map_count = 0;
 287         cpus_clear(mm->cpu_vm_mask);
 288         mm->mm_rb = RB_ROOT;
 289         rb_link = &mm->mm_rb.rb_node;
 290         rb_parent = NULL;
 291         pprev = &mm->mmap;
 292
 293         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 294                 struct file *file;
 295
 296                 if (mpnt->vm_flags & VM_DONTCOPY) {
 297                         long pages = vma_pages(mpnt);
 298                         mm->total_vm -= pages;
 299                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 300                                                                 -pages);
 301                         continue;
 302                 }
 303                 charge = 0;
 304                 if (mpnt->vm_flags & VM_ACCOUNT) {
 305                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 306                         if (security_vm_enough_memory(len))
 307                                 goto fail_nomem;
 308                         charge = len;
 309                 }
 310                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 311                 if (!tmp)
 312                         goto fail_nomem;
 313                 *tmp = *mpnt;
 314                 pol = mpol_dup(vma_policy(mpnt));
 315                 retval = PTR_ERR(pol);
 316                 if (IS_ERR(pol))
 317                         goto fail_nomem_policy;
 318                 vma_set_policy(tmp, pol);
 319                 tmp->vm_flags &= ~VM_LOCKED;
 320                 tmp->vm_mm = mm;
 321                 tmp->vm_next = NULL;
 322                 anon_vma_link(tmp);
 323                 file = tmp->vm_file;
 324                 if (file) {
 325                         struct inode *inode = file->f_path.dentry->d_inode;
 326                         struct address_space *mapping = file->f_mapping;
 327
 328                         get_file(file);
 329                         if (tmp->vm_flags & VM_DENYWRITE)
 330                                 atomic_dec(&inode->i_writecount);
 331                         spin_lock(&mapping->i_mmap_lock);
 332                         if (tmp->vm_flags & VM_SHARED)
 333                                 mapping->i_mmap_writable++;
 334                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 335                         flush_dcache_mmap_lock(mapping);
 336                         /* insert tmp into the share list, just after mpnt */
 337                         vma_prio_tree_add(tmp, mpnt);
 338                         flush_dcache_mmap_unlock(mapping);
 339                         spin_unlock(&mapping->i_mmap_lock);
 340                 }
 341
 342                 /*
 343                  * Clear hugetlb-related page reserves for children. This only
 344                  * affects MAP_PRIVATE mappings. Faults generated by the child
 345                  * are not guaranteed to succeed, even if read-only
 346                  */
 347                 if (is_vm_hugetlb_page(tmp))
 348                         reset_vma_resv_huge_pages(tmp);
 349
 350                 /*
 351                  * Link in the new vma and copy the page table entries.
 352                  */
 353                 *pprev = tmp;
 354                 pprev = &tmp->vm_next;
 355
 356                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 357                 rb_link = &tmp->vm_rb.rb_right;
 358                 rb_parent = &tmp->vm_rb;
 359
 360                 mm->map_count++;
 361                 retval = copy_page_range(mm, oldmm, mpnt);
 362
 363                 if (tmp->vm_ops && tmp->vm_ops->open)
 364                         tmp->vm_ops->open(tmp);
 365
 366                 if (retval)
 367                         goto out;
 368         }
 369         /* a new mm has just been created */
 370         arch_dup_mmap(oldmm, mm);
 371         retval = 0;
 372 out:
 373         up_write(&mm->mmap_sem);
 374         flush_tlb_mm(oldmm);
 375         up_write(&oldmm->mmap_sem);
 376         return retval;
 377 fail_nomem_policy:
 378         kmem_cache_free(vm_area_cachep, tmp);
 379 fail_nomem:
 380         retval = -ENOMEM;
 381         vm_unacct_memory(charge);
 382         goto out;
 383 }
 384
 385 static inline int mm_alloc_pgd(struct mm_struct * mm)
 386 {
 387         mm->pgd = pgd_alloc(mm);
 388         if (unlikely(!mm->pgd))
 389                 return -ENOMEM;
 390         return 0;
 391 }
 392
 393 static inline void mm_free_pgd(struct mm_struct * mm)
 394 {
 395         pgd_free(mm, mm->pgd);
 396 }
 397 #else
 398 #define dup_mmap(mm, oldmm)     (0)
 399 #define mm_alloc_pgd(mm)        (0)
 400 #define mm_free_pgd(mm)
 401 #endif /* CONFIG_MMU */
 402
 403 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 404
 405 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 406 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 407
 408 #include <linux/init_task.h>
 409
 410 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
 411 {
 412         atomic_set(&mm->mm_users, 1);
 413         atomic_set(&mm->mm_count, 1);
 414         init_rwsem(&mm->mmap_sem);
 415         INIT_LIST_HEAD(&mm->mmlist);
 416         mm->flags = (current->mm) ? current->mm->flags
 417                                   : MMF_DUMP_FILTER_DEFAULT;
 418         mm->core_state = NULL;
 419         mm->nr_ptes = 0;
 420         set_mm_counter(mm, file_rss, 0);
 421         set_mm_counter(mm, anon_rss, 0);
 422         spin_lock_init(&mm->page_table_lock);
 423         spin_lock_init(&mm->ioctx_lock);
 424         INIT_HLIST_HEAD(&mm->ioctx_list);
 425         mm->free_area_cache = TASK_UNMAPPED_BASE;
 426         mm->cached_hole_size = ~0UL;
 427         mm_init_owner(mm, p);
 428
 429         if (likely(!mm_alloc_pgd(mm))) {
 430                 mm->def_flags = 0;
 431                 mmu_notifier_mm_init(mm);
 432                 return mm;
 433         }
 434
 435         free_mm(mm);
 436         return NULL;
 437 }
 438
 439 /*
 440  * Allocate and initialize an mm_struct.
 441  */
 442 struct mm_struct * mm_alloc(void)
 443 {
 444         struct mm_struct * mm;
 445
 446         mm = allocate_mm();
 447         if (mm) {
 448                 memset(mm, 0, sizeof(*mm));
 449                 mm = mm_init(mm, current);
 450         }
 451         return mm;
 452 }
 453
 454 /*
 455  * Called when the last reference to the mm
 456  * is dropped: either by a lazy thread or by
 457  * mmput. Free the page directory and the mm.
 458  */
 459 void __mmdrop(struct mm_struct *mm)
 460 {
 461         BUG_ON(mm == &init_mm);
 462         mm_free_pgd(mm);
 463         destroy_context(mm);
 464         mmu_notifier_mm_destroy(mm);
 465         free_mm(mm);
 466 }
 467 EXPORT_SYMBOL_GPL(__mmdrop);
 468
 469 /*
 470  * Decrement the use count and release all resources for an mm.
 471  */
 472 void mmput(struct mm_struct *mm)
 473 {
 474         might_sleep();
 475
 476         if (atomic_dec_and_test(&mm->mm_users)) {
 477                 exit_aio(mm);
 478                 exit_mmap(mm);
 479                 set_mm_exe_file(mm, NULL);
 480                 if (!list_empty(&mm->mmlist)) {
 481                         spin_lock(&mmlist_lock);
 482                         list_del(&mm->mmlist);
 483                         spin_unlock(&mmlist_lock);
 484                 }
 485                 put_swap_token(mm);
 486                 mmdrop(mm);
 487         }
 488 }
 489 EXPORT_SYMBOL_GPL(mmput);
 490
 491 /**
 492  * get_task_mm - acquire a reference to the task's mm
 493  *
 494  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 495  * this kernel workthread has transiently adopted a user mm with use_mm,
 496  * to do its AIO) is not set and if so returns a reference to it, after
 497  * bumping up the use count.  User must release the mm via mmput()
 498  * after use.  Typically used by /proc and ptrace.
 499  */
 500 struct mm_struct *get_task_mm(struct task_struct *task)
 501 {
 502         struct mm_struct *mm;
 503
 504         task_lock(task);
 505         mm = task->mm;
 506         if (mm) {
 507                 if (task->flags & PF_KTHREAD)
 508                         mm = NULL;
 509                 else
 510                         atomic_inc(&mm->mm_users);
 511         }
 512         task_unlock(task);
 513         return mm;
 514 }
 515 EXPORT_SYMBOL_GPL(get_task_mm);
 516
 517 /* Please note the differences between mmput and mm_release.
 518  * mmput is called whenever we stop holding onto a mm_struct,
 519  * error success whatever.
 520  *
 521  * mm_release is called after a mm_struct has been removed
 522  * from the current process.
 523  *
 524  * This difference is important for error handling, when we
 525  * only half set up a mm_struct for a new process and need to restore
 526  * the old one.  Because we mmput the new mm_struct before
 527  * restoring the old one. . .
 528  * Eric Biederman 10 January 1998
 529  */
 530 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 531 {
 532         struct completion *vfork_done = tsk->vfork_done;
 533
 534         /* Get rid of any futexes when releasing the mm */
 535 #ifdef CONFIG_FUTEX
 536         if (unlikely(tsk->robust_list))
 537                 exit_robust_list(tsk);
 538 #ifdef CONFIG_COMPAT
 539         if (unlikely(tsk->compat_robust_list))
 540                 compat_exit_robust_list(tsk);
 541 #endif
 542 #endif
 543
 544         /* Get rid of any cached register state */
 545         deactivate_mm(tsk, mm);
 546
 547         /* notify parent sleeping on vfork() */
 548         if (vfork_done) {
 549                 tsk->vfork_done = NULL;
 550                 complete(vfork_done);
 551         }
 552
 553         /*
 554          * If we're exiting normally, clear a user-space tid field if
 555          * requested.  We leave this alone when dying by signal, to leave
 556          * the value intact in a core dump, and to save the unnecessary
 557          * trouble otherwise.  Userland only wants this done for a sys_exit.
 558          */
 559         if (tsk->clear_child_tid
 560             && !(tsk->flags & PF_SIGNALED)
 561             && atomic_read(&mm->mm_users) > 1) {
 562                 u32 __user * tidptr = tsk->clear_child_tid;
 563                 tsk->clear_child_tid = NULL;
 564
 565                 /*
 566                  * We don't check the error code - if userspace has
 567                  * not set up a proper pointer then tough luck.
 568                  */
 569                 put_user(0, tidptr);
 570                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 571         }
 572 }
 573
 574 /*
 575  * Allocate a new mm structure and copy contents from the
 576  * mm structure of the passed in task structure.
 577  */
 578 struct mm_struct *dup_mm(struct task_struct *tsk)
 579 {
 580         struct mm_struct *mm, *oldmm = current->mm;
 581         int err;
 582
 583         if (!oldmm)
 584                 return NULL;
 585
 586         mm = allocate_mm();
 587         if (!mm)
 588                 goto fail_nomem;
 589
 590         memcpy(mm, oldmm, sizeof(*mm));
 591
 592         /* Initializing for Swap token stuff */
 593         mm->token_priority = 0;
 594         mm->last_interval = 0;
 595
 596         if (!mm_init(mm, tsk))
 597                 goto fail_nomem;
 598
 599         if (init_new_context(tsk, mm))
 600                 goto fail_nocontext;
 601
 602         dup_mm_exe_file(oldmm, mm);
 603
 604         err = dup_mmap(mm, oldmm);
 605         if (err)
 606                 goto free_pt;
 607
 608         mm->hiwater_rss = get_mm_rss(mm);
 609         mm->hiwater_vm = mm->total_vm;
 610
 611         return mm;
 612
 613 free_pt:
 614         mmput(mm);
 615
 616 fail_nomem:
 617         return NULL;
 618
 619 fail_nocontext:
 620         /*
 621          * If init_new_context() failed, we cannot use mmput() to free the mm
 622          * because it calls destroy_context()
 623          */
 624         mm_free_pgd(mm);
 625         free_mm(mm);
 626         return NULL;
 627 }
 628
 629 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 630 {
 631         struct mm_struct * mm, *oldmm;
 632         int retval;
 633
 634         tsk->min_flt = tsk->maj_flt = 0;
 635         tsk->nvcsw = tsk->nivcsw = 0;
 636
 637         tsk->mm = NULL;
 638         tsk->active_mm = NULL;
 639
 640         /*
 641          * Are we cloning a kernel thread?
 642          *
 643          * We need to steal a active VM for that..
 644          */
 645         oldmm = current->mm;
 646         if (!oldmm)
 647                 return 0;
 648
 649         if (clone_flags & CLONE_VM) {
 650                 atomic_inc(&oldmm->mm_users);
 651                 mm = oldmm;
 652                 goto good_mm;
 653         }
 654
 655         retval = -ENOMEM;
 656         mm = dup_mm(tsk);
 657         if (!mm)
 658                 goto fail_nomem;
 659
 660 good_mm:
 661         /* Initializing for Swap token stuff */
 662         mm->token_priority = 0;
 663         mm->last_interval = 0;
 664
 665         tsk->mm = mm;
 666         tsk->active_mm = mm;
 667         return 0;
 668
 669 fail_nomem:
 670         return retval;
 671 }
 672
 673 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 674 {
 675         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 676         /* We don't need to lock fs - think why ;-) */
 677         if (fs) {
 678                 atomic_set(&fs->count, 1);
 679                 rwlock_init(&fs->lock);
 680                 fs->umask = old->umask;
 681                 read_lock(&old->lock);
 682                 fs->root = old->root;
 683                 path_get(&old->root);
 684                 fs->pwd = old->pwd;
 685                 path_get(&old->pwd);
 686                 read_unlock(&old->lock);
 687         }
 688         return fs;
 689 }
 690
 691 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 692 {
 693         return __copy_fs_struct(old);
 694 }
 695
 696 EXPORT_SYMBOL_GPL(copy_fs_struct);
 697
 698 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 699 {
 700         if (clone_flags & CLONE_FS) {
 701                 atomic_inc(&current->fs->count);
 702                 return 0;
 703         }
 704         tsk->fs = __copy_fs_struct(current->fs);
 705         if (!tsk->fs)
 706                 return -ENOMEM;
 707         return 0;
 708 }
 709
 710 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 711 {
 712         struct files_struct *oldf, *newf;
 713         int error = 0;
 714
 715         /*
 716          * A background process may not have any files ...
 717          */
 718         oldf = current->files;
 719         if (!oldf)
 720                 goto out;
 721
 722         if (clone_flags & CLONE_FILES) {
 723                 atomic_inc(&oldf->count);
 724                 goto out;
 725         }
 726
 727         newf = dup_fd(oldf, &error);
 728         if (!newf)
 729                 goto out;
 730
 731         tsk->files = newf;
 732         error = 0;
 733 out:
 734         return error;
 735 }
 736
 737 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 738 {
 739 #ifdef CONFIG_BLOCK
 740         struct io_context *ioc = current->io_context;
 741
 742         if (!ioc)
 743                 return 0;
 744         /*
 745          * Share io context with parent, if CLONE_IO is set
 746          */
 747         if (clone_flags & CLONE_IO) {
 748                 tsk->io_context = ioc_task_link(ioc);
 749                 if (unlikely(!tsk->io_context))
 750                         return -ENOMEM;
 751         } else if (ioprio_valid(ioc->ioprio)) {
 752                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 753                 if (unlikely(!tsk->io_context))
 754                         return -ENOMEM;
 755
 756                 tsk->io_context->ioprio = ioc->ioprio;
 757         }
 758 #endif
 759         return 0;
 760 }
 761
 762 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 763 {
 764         struct sighand_struct *sig;
 765
 766         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 767                 atomic_inc(&current->sighand->count);
 768                 return 0;
 769         }
 770         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 771         rcu_assign_pointer(tsk->sighand, sig);
 772         if (!sig)
 773                 return -ENOMEM;
 774         atomic_set(&sig->count, 1);
 775         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 776         return 0;
 777 }
 778
 779 void __cleanup_sighand(struct sighand_struct *sighand)
 780 {
 781         if (atomic_dec_and_test(&sighand->count))
 782                 kmem_cache_free(sighand_cachep, sighand);
 783 }
 784
 785
 786 /*
 787  * Initialize POSIX timer handling for a thread group.
 788  */
 789 static void posix_cpu_timers_init_group(struct signal_struct *sig)
 790 {
 791         /* Thread group counters. */
 792         thread_group_cputime_init(sig);
 793
 794         /* Expiration times and increments. */
 795         sig->it_virt_expires = cputime_zero;
 796         sig->it_virt_incr = cputime_zero;
 797         sig->it_prof_expires = cputime_zero;
 798         sig->it_prof_incr = cputime_zero;
 799
 800         /* Cached expiration times. */
 801         sig->cputime_expires.prof_exp = cputime_zero;
 802         sig->cputime_expires.virt_exp = cputime_zero;
 803         sig->cputime_expires.sched_exp = 0;
 804
 805         /* The timer lists. */
 806         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 807         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 808         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 809 }
 810
 811 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 812 {
 813         struct signal_struct *sig;
 814         int ret;
 815
 816         if (clone_flags & CLONE_THREAD) {
 817                 ret = thread_group_cputime_clone_thread(current);
 818                 if (likely(!ret)) {
 819                         atomic_inc(&current->signal->count);
 820                         atomic_inc(&current->signal->live);
 821                 }
 822                 return ret;
 823         }
 824         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 825         tsk->signal = sig;
 826         if (!sig)
 827                 return -ENOMEM;
 828
 829         atomic_set(&sig->count, 1);
 830         atomic_set(&sig->live, 1);
 831         init_waitqueue_head(&sig->wait_chldexit);
 832         sig->flags = 0;
 833         sig->group_exit_code = 0;
 834         sig->group_exit_task = NULL;
 835         sig->group_stop_count = 0;
 836         sig->curr_target = tsk;
 837         init_sigpending(&sig->shared_pending);
 838         INIT_LIST_HEAD(&sig->posix_timers);
 839
 840         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 841         sig->it_real_incr.tv64 = 0;
 842         sig->real_timer.function = it_real_fn;
 843
 844         sig->leader = 0;        /* session leadership doesn't inherit */
 845         sig->tty_old_pgrp = NULL;
 846         sig->tty = NULL;
 847
 848         sig->cutime = sig->cstime = cputime_zero;
 849         sig->gtime = cputime_zero;
 850         sig->cgtime = cputime_zero;
 851         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 852         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 853         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
 854         task_io_accounting_init(&sig->ioac);
 855         taskstats_tgid_init(sig);
 856
 857         task_lock(current->group_leader);
 858         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 859         task_unlock(current->group_leader);
 860
 861         posix_cpu_timers_init_group(sig);
 862
 863         acct_init_pacct(&sig->pacct);
 864
 865         tty_audit_fork(sig);
 866
 867         return 0;
 868 }
 869
 870 void __cleanup_signal(struct signal_struct *sig)
 871 {
 872         thread_group_cputime_free(sig);
 873         tty_kref_put(sig->tty);
 874         kmem_cache_free(signal_cachep, sig);
 875 }
 876
 877 static void cleanup_signal(struct task_struct *tsk)
 878 {
 879         struct signal_struct *sig = tsk->signal;
 880
 881         atomic_dec(&sig->live);
 882
 883         if (atomic_dec_and_test(&sig->count))
 884                 __cleanup_signal(sig);
 885 }
 886
 887 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 888 {
 889         unsigned long new_flags = p->flags;
 890
 891         new_flags &= ~PF_SUPERPRIV;
 892         new_flags |= PF_FORKNOEXEC;
 893         new_flags |= PF_STARTING;
 894         p->flags = new_flags;
 895         clear_freeze_flag(p);
 896 }
 897
 898 asmlinkage long sys_set_tid_address(int __user *tidptr)
 899 {
 900         current->clear_child_tid = tidptr;
 901
 902         return task_pid_vnr(current);
 903 }
 904
 905 static void rt_mutex_init_task(struct task_struct *p)
 906 {
 907         spin_lock_init(&p->pi_lock);
 908 #ifdef CONFIG_RT_MUTEXES
 909         plist_head_init(&p->pi_waiters, &p->pi_lock);
 910         p->pi_blocked_on = NULL;
 911 #endif
 912 }
 913
 914 #ifdef CONFIG_MM_OWNER
 915 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 916 {
 917         mm->owner = p;
 918 }
 919 #endif /* CONFIG_MM_OWNER */
 920
 921 /*
 922  * Initialize POSIX timer handling for a single task.
 923  */
 924 static void posix_cpu_timers_init(struct task_struct *tsk)
 925 {
 926         tsk->cputime_expires.prof_exp = cputime_zero;
 927         tsk->cputime_expires.virt_exp = cputime_zero;
 928         tsk->cputime_expires.sched_exp = 0;
 929         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
 930         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
 931         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
 932 }
 933
 934 /*
 935  * This creates a new process as a copy of the old one,
 936  * but does not actually start it yet.
 937  *
 938  * It copies the registers, and all the appropriate
 939  * parts of the process environment (as per the clone
 940  * flags). The actual kick-off is left to the caller.
 941  */
 942 static struct task_struct *copy_process(unsigned long clone_flags,
 943                                         unsigned long stack_start,
 944                                         struct pt_regs *regs,
 945                                         unsigned long stack_size,
 946                                         int __user *child_tidptr,
 947                                         struct pid *pid,
 948                                         int trace)
 949 {
 950         int retval;
 951         struct task_struct *p;
 952         int cgroup_callbacks_done = 0;
 953
 954         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 955                 return ERR_PTR(-EINVAL);
 956
 957         /*
 958          * Thread groups must share signals as well, and detached threads
 959          * can only be started up within the thread group.
 960          */
 961         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 962                 return ERR_PTR(-EINVAL);
 963
 964         /*
 965          * Shared signal handlers imply shared VM. By way of the above,
 966          * thread groups also imply shared VM. Blocking this case allows
 967          * for various simplifications in other code.
 968          */
 969         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 970                 return ERR_PTR(-EINVAL);
 971
 972         retval = security_task_create(clone_flags);
 973         if (retval)
 974                 goto fork_out;
 975
 976         retval = -ENOMEM;
 977         p = dup_task_struct(current);
 978         if (!p)
 979                 goto fork_out;
 980
 981         rt_mutex_init_task(p);
 982
 983 #ifdef CONFIG_PROVE_LOCKING
 984         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
 985         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
 986 #endif
 987         retval = -EAGAIN;
 988         if (atomic_read(&p->real_cred->user->processes) >=
 989                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
 990                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 991                     p->real_cred->user != INIT_USER)
 992                         goto bad_fork_free;
 993         }
 994
 995         retval = copy_creds(p, clone_flags);
 996         if (retval < 0)
 997                 goto bad_fork_free;
 998
 999         /*
1000          * If multiple threads are within copy_process(), then this check
1001          * triggers too late. This doesn't hurt, the check is only there
1002          * to stop root fork bombs.
1003          */
1004         if (nr_threads >= max_threads)
1005                 goto bad_fork_cleanup_count;
1006
1007         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1008                 goto bad_fork_cleanup_count;
1009
1010         if (p->binfmt && !try_module_get(p->binfmt->module))
1011                 goto bad_fork_cleanup_put_domain;
1012
1013         p->did_exec = 0;
1014         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1015         copy_flags(clone_flags, p);
1016         INIT_LIST_HEAD(&p->children);
1017         INIT_LIST_HEAD(&p->sibling);
1018 #ifdef CONFIG_PREEMPT_RCU
1019         p->rcu_read_lock_nesting = 0;
1020         p->rcu_flipctr_idx = 0;
1021 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1022         p->vfork_done = NULL;
1023         spin_lock_init(&p->alloc_lock);
1024
1025         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1026         init_sigpending(&p->pending);
1027
1028         p->utime = cputime_zero;
1029         p->stime = cputime_zero;
1030         p->gtime = cputime_zero;
1031         p->utimescaled = cputime_zero;
1032         p->stimescaled = cputime_zero;
1033         p->prev_utime = cputime_zero;
1034         p->prev_stime = cputime_zero;
1035
1036         p->default_timer_slack_ns = current->timer_slack_ns;
1037
1038 #ifdef CONFIG_DETECT_SOFTLOCKUP
1039         p->last_switch_count = 0;
1040         p->last_switch_timestamp = 0;
1041 #endif
1042
1043         task_io_accounting_init(&p->ioac);
1044         acct_clear_integrals(p);
1045
1046         posix_cpu_timers_init(p);
1047
1048         p->lock_depth = -1;             /* -1 = no lock */
1049         do_posix_clock_monotonic_gettime(&p->start_time);
1050         p->real_start_time = p->start_time;
1051         monotonic_to_bootbased(&p->real_start_time);
1052         p->io_context = NULL;
1053         p->audit_context = NULL;
1054         cgroup_fork(p);
1055 #ifdef CONFIG_NUMA
1056         p->mempolicy = mpol_dup(p->mempolicy);
1057         if (IS_ERR(p->mempolicy)) {
1058                 retval = PTR_ERR(p->mempolicy);
1059                 p->mempolicy = NULL;
1060                 goto bad_fork_cleanup_cgroup;
1061         }
1062         mpol_fix_fork_child_flag(p);
1063 #endif
1064 #ifdef CONFIG_TRACE_IRQFLAGS
1065         p->irq_events = 0;
1066 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1067         p->hardirqs_enabled = 1;
1068 #else
1069         p->hardirqs_enabled = 0;
1070 #endif
1071         p->hardirq_enable_ip = 0;
1072         p->hardirq_enable_event = 0;
1073         p->hardirq_disable_ip = _THIS_IP_;
1074         p->hardirq_disable_event = 0;
1075         p->softirqs_enabled = 1;
1076         p->softirq_enable_ip = _THIS_IP_;
1077         p->softirq_enable_event = 0;
1078         p->softirq_disable_ip = 0;
1079         p->softirq_disable_event = 0;
1080         p->hardirq_context = 0;
1081         p->softirq_context = 0;
1082 #endif
1083 #ifdef CONFIG_LOCKDEP
1084         p->lockdep_depth = 0; /* no locks held yet */
1085         p->curr_chain_key = 0;
1086         p->lockdep_recursion = 0;
1087 #endif
1088
1089 #ifdef CONFIG_DEBUG_MUTEXES
1090         p->blocked_on = NULL; /* not blocked yet */
1091 #endif
1092         if (unlikely(ptrace_reparented(current)))
1093                 ptrace_fork(p, clone_flags);
1094
1095         /* Perform scheduler related setup. Assign this task to a CPU. */
1096         sched_fork(p, clone_flags);
1097
1098         if ((retval = audit_alloc(p)))
1099                 goto bad_fork_cleanup_policy;
1100         /* copy all the process information */
1101         if ((retval = copy_semundo(clone_flags, p)))
1102                 goto bad_fork_cleanup_audit;
1103         if ((retval = copy_files(clone_flags, p)))
1104                 goto bad_fork_cleanup_semundo;
1105         if ((retval = copy_fs(clone_flags, p)))
1106                 goto bad_fork_cleanup_files;
1107         if ((retval = copy_sighand(clone_flags, p)))
1108                 goto bad_fork_cleanup_fs;
1109         if ((retval = copy_signal(clone_flags, p)))
1110                 goto bad_fork_cleanup_sighand;
1111         if ((retval = copy_mm(clone_flags, p)))
1112                 goto bad_fork_cleanup_signal;
1113         if ((retval = copy_namespaces(clone_flags, p)))
1114                 goto bad_fork_cleanup_mm;
1115         if ((retval = copy_io(clone_flags, p)))
1116                 goto bad_fork_cleanup_namespaces;
1117         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1118         if (retval)
1119                 goto bad_fork_cleanup_io;
1120
1121         if (pid != &init_struct_pid) {
1122                 retval = -ENOMEM;
1123                 pid = alloc_pid(task_active_pid_ns(p));
1124                 if (!pid)
1125                         goto bad_fork_cleanup_io;
1126
1127                 if (clone_flags & CLONE_NEWPID) {
1128                         retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1129                         if (retval < 0)
1130                                 goto bad_fork_free_pid;
1131                 }
1132         }
1133
1134         ftrace_graph_init_task(p);
1135
1136         p->pid = pid_nr(pid);
1137         p->tgid = p->pid;
1138         if (clone_flags & CLONE_THREAD)
1139                 p->tgid = current->tgid;
1140
1141         if (current->nsproxy != p->nsproxy) {
1142                 retval = ns_cgroup_clone(p, pid);
1143                 if (retval)
1144                         goto bad_fork_free_graph;
1145         }
1146
1147         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1148         /*
1149          * Clear TID on mm_release()?
1150          */
1151         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1152 #ifdef CONFIG_FUTEX
1153         p->robust_list = NULL;
1154 #ifdef CONFIG_COMPAT
1155         p->compat_robust_list = NULL;
1156 #endif
1157         INIT_LIST_HEAD(&p->pi_state_list);
1158         p->pi_state_cache = NULL;
1159 #endif
1160         /*
1161          * sigaltstack should be cleared when sharing the same VM
1162          */
1163         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1164                 p->sas_ss_sp = p->sas_ss_size = 0;
1165
1166         /*
1167          * Syscall tracing should be turned off in the child regardless
1168          * of CLONE_PTRACE.
1169          */
1170         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1171 #ifdef TIF_SYSCALL_EMU
1172         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1173 #endif
1174         clear_all_latency_tracing(p);
1175
1176         /* Our parent execution domain becomes current domain
1177            These must match for thread signalling to apply */
1178         p->parent_exec_id = p->self_exec_id;
1179
1180         /* ok, now we should be set up.. */
1181         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1182         p->pdeath_signal = 0;
1183         p->exit_state = 0;
1184
1185         /*
1186          * Ok, make it visible to the rest of the system.
1187          * We dont wake it up yet.
1188          */
1189         p->group_leader = p;
1190         INIT_LIST_HEAD(&p->thread_group);
1191
1192         /* Now that the task is set up, run cgroup callbacks if
1193          * necessary. We need to run them before the task is visible
1194          * on the tasklist. */
1195         cgroup_fork_callbacks(p);
1196         cgroup_callbacks_done = 1;
1197
1198         /* Need tasklist lock for parent etc handling! */
1199         write_lock_irq(&tasklist_lock);
1200
1201         /*
1202          * The task hasn't been attached yet, so its cpus_allowed mask will
1203          * not be changed, nor will its assigned CPU.
1204          *
1205          * The cpus_allowed mask of the parent may have changed after it was
1206          * copied first time - so re-copy it here, then check the child's CPU
1207          * to ensure it is on a valid CPU (and if not, just force it back to
1208          * parent's CPU). This avoids alot of nasty races.
1209          */
1210         p->cpus_allowed = current->cpus_allowed;
1211         p->rt.nr_cpus_allowed = current->rt.nr_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
1222         spin_lock(&current->sighand->siglock);
1223
1224         /*
1225          * Process group and session signals need to be delivered to just the
1226          * parent before the fork or both the parent and the child after the
1227          * fork. Restart if a signal comes in before we add the new process to
1228          * it's process group.
1229          * A fatal signal pending means that current will exit, so the new
1230          * thread can't slip out of an OOM kill (or normal SIGKILL).
1231          */
1232         recalc_sigpending();
1233         if (signal_pending(current)) {
1234                 spin_unlock(&current->sighand->siglock);
1235                 write_unlock_irq(&tasklist_lock);
1236                 retval = -ERESTARTNOINTR;
1237                 goto bad_fork_free_graph;
1238         }
1239
1240         if (clone_flags & CLONE_THREAD) {
1241                 p->group_leader = current->group_leader;
1242                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1243         }
1244
1245         if (likely(p->pid)) {
1246                 list_add_tail(&p->sibling, &p->real_parent->children);
1247                 tracehook_finish_clone(p, clone_flags, trace);
1248
1249                 if (thread_group_leader(p)) {
1250                         if (clone_flags & CLONE_NEWPID)
1251                                 p->nsproxy->pid_ns->child_reaper = p;
1252
1253                         p->signal->leader_pid = pid;
1254                         tty_kref_put(p->signal->tty);
1255                         p->signal->tty = tty_kref_get(current->signal->tty);
1256                         set_task_pgrp(p, task_pgrp_nr(current));
1257                         set_task_session(p, task_session_nr(current));
1258                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1259                         attach_pid(p, PIDTYPE_SID, task_session(current));
1260                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1261                         __get_cpu_var(process_counts)++;
1262                 }
1263                 attach_pid(p, PIDTYPE_PID, pid);
1264                 nr_threads++;
1265         }
1266
1267         total_forks++;
1268         spin_unlock(&current->sighand->siglock);
1269         write_unlock_irq(&tasklist_lock);
1270         proc_fork_connector(p);
1271         cgroup_post_fork(p);
1272         return p;
1273
1274 bad_fork_free_graph:
1275         ftrace_graph_exit_task(p);
1276 bad_fork_free_pid:
1277         if (pid != &init_struct_pid)
1278                 free_pid(pid);
1279 bad_fork_cleanup_io:
1280         put_io_context(p->io_context);
1281 bad_fork_cleanup_namespaces:
1282         exit_task_namespaces(p);
1283 bad_fork_cleanup_mm:
1284         if (p->mm)
1285                 mmput(p->mm);
1286 bad_fork_cleanup_signal:
1287         cleanup_signal(p);
1288 bad_fork_cleanup_sighand:
1289         __cleanup_sighand(p->sighand);
1290 bad_fork_cleanup_fs:
1291         exit_fs(p); /* blocking */
1292 bad_fork_cleanup_files:
1293         exit_files(p); /* blocking */
1294 bad_fork_cleanup_semundo:
1295         exit_sem(p);
1296 bad_fork_cleanup_audit:
1297         audit_free(p);
1298 bad_fork_cleanup_policy:
1299 #ifdef CONFIG_NUMA
1300         mpol_put(p->mempolicy);
1301 bad_fork_cleanup_cgroup:
1302 #endif
1303         cgroup_exit(p, cgroup_callbacks_done);
1304         delayacct_tsk_free(p);
1305         if (p->binfmt)
1306                 module_put(p->binfmt->module);
1307 bad_fork_cleanup_put_domain:
1308         module_put(task_thread_info(p)->exec_domain->module);
1309 bad_fork_cleanup_count:
1310         atomic_dec(&p->cred->user->processes);
1311         put_cred(p->real_cred);
1312         put_cred(p->cred);
1313 bad_fork_free:
1314         free_task(p);
1315 fork_out:
1316         return ERR_PTR(retval);
1317 }
1318
1319 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1320 {
1321         memset(regs, 0, sizeof(struct pt_regs));
1322         return regs;
1323 }
1324
1325 struct task_struct * __cpuinit fork_idle(int cpu)
1326 {
1327         struct task_struct *task;
1328         struct pt_regs regs;
1329
1330         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1331                             &init_struct_pid, 0);
1332         if (!IS_ERR(task))
1333                 init_idle(task, cpu);
1334
1335         return task;
1336 }
1337
1338 /*
1339  *  Ok, this is the main fork-routine.
1340  *
1341  * It copies the process, and if successful kick-starts
1342  * it and waits for it to finish using the VM if required.
1343  */
1344 long do_fork(unsigned long clone_flags,
1345               unsigned long stack_start,
1346               struct pt_regs *regs,
1347               unsigned long stack_size,
1348               int __user *parent_tidptr,
1349               int __user *child_tidptr)
1350 {
1351         struct task_struct *p;
1352         int trace = 0;
1353         long nr;
1354
1355         /*
1356          * Do some preliminary argument and permissions checking before we
1357          * actually start allocating stuff
1358          */
1359         if (clone_flags & CLONE_NEWUSER) {
1360                 if (clone_flags & CLONE_THREAD)
1361                         return -EINVAL;
1362                 /* hopefully this check will go away when userns support is
1363                  * complete
1364                  */
1365                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1366                                 !capable(CAP_SETGID))
1367                         return -EPERM;
1368         }
1369
1370         /*
1371          * We hope to recycle these flags after 2.6.26
1372          */
1373         if (unlikely(clone_flags & CLONE_STOPPED)) {
1374                 static int __read_mostly count = 100;
1375
1376                 if (count > 0 && printk_ratelimit()) {
1377                         char comm[TASK_COMM_LEN];
1378
1379                         count--;
1380                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1381                                         "clone flags 0x%lx\n",
1382                                 get_task_comm(comm, current),
1383                                 clone_flags & CLONE_STOPPED);
1384                 }
1385         }
1386
1387         /*
1388          * When called from kernel_thread, don't do user tracing stuff.
1389          */
1390         if (likely(user_mode(regs)))
1391                 trace = tracehook_prepare_clone(clone_flags);
1392
1393         p = copy_process(clone_flags, stack_start, regs, stack_size,
1394                          child_tidptr, NULL, trace);
1395         /*
1396          * Do this prior waking up the new thread - the thread pointer
1397          * might get invalid after that point, if the thread exits quickly.
1398          */
1399         if (!IS_ERR(p)) {
1400                 struct completion vfork;
1401
1402                 trace_sched_process_fork(current, p);
1403
1404                 nr = task_pid_vnr(p);
1405
1406                 if (clone_flags & CLONE_PARENT_SETTID)
1407                         put_user(nr, parent_tidptr);
1408
1409                 if (clone_flags & CLONE_VFORK) {
1410                         p->vfork_done = &vfork;
1411                         init_completion(&vfork);
1412                 }
1413
1414                 audit_finish_fork(p);
1415                 tracehook_report_clone(trace, regs, clone_flags, nr, p);
1416
1417                 /*
1418                  * We set PF_STARTING at creation in case tracing wants to
1419                  * use this to distinguish a fully live task from one that
1420                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1421                  * clear it and set the child going.
1422                  */
1423                 p->flags &= ~PF_STARTING;
1424
1425                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1426                         /*
1427                          * We'll start up with an immediate SIGSTOP.
1428                          */
1429                         sigaddset(&p->pending.signal, SIGSTOP);
1430                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1431                         __set_task_state(p, TASK_STOPPED);
1432                 } else {
1433                         wake_up_new_task(p, clone_flags);
1434                 }
1435
1436                 tracehook_report_clone_complete(trace, regs,
1437                                                 clone_flags, nr, p);
1438
1439                 if (clone_flags & CLONE_VFORK) {
1440                         freezer_do_not_count();
1441                         wait_for_completion(&vfork);
1442                         freezer_count();
1443                         tracehook_report_vfork_done(p, nr);
1444                 }
1445         } else {
1446                 nr = PTR_ERR(p);
1447         }
1448         return nr;
1449 }
1450
1451 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1452 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1453 #endif
1454
1455 static void sighand_ctor(void *data)
1456 {
1457         struct sighand_struct *sighand = data;
1458
1459         spin_lock_init(&sighand->siglock);
1460         init_waitqueue_head(&sighand->signalfd_wqh);
1461 }
1462
1463 void __init proc_caches_init(void)
1464 {
1465         sighand_cachep = kmem_cache_create("sighand_cache",
1466                         sizeof(struct sighand_struct), 0,
1467                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1468                         sighand_ctor);
1469         signal_cachep = kmem_cache_create("signal_cache",
1470                         sizeof(struct signal_struct), 0,
1471                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1472         files_cachep = kmem_cache_create("files_cache",
1473                         sizeof(struct files_struct), 0,
1474                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1475         fs_cachep = kmem_cache_create("fs_cache",
1476                         sizeof(struct fs_struct), 0,
1477                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1478         vm_area_cachep = kmem_cache_create("vm_area_struct",
1479                         sizeof(struct vm_area_struct), 0,
1480                         SLAB_PANIC, NULL);
1481         mm_cachep = kmem_cache_create("mm_struct",
1482                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1483                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1484 }
1485
1486 /*
1487  * Check constraints on flags passed to the unshare system call and
1488  * force unsharing of additional process context as appropriate.
1489  */
1490 static void check_unshare_flags(unsigned long *flags_ptr)
1491 {
1492         /*
1493          * If unsharing a thread from a thread group, must also
1494          * unshare vm.
1495          */
1496         if (*flags_ptr & CLONE_THREAD)
1497                 *flags_ptr |= CLONE_VM;
1498
1499         /*
1500          * If unsharing vm, must also unshare signal handlers.
1501          */
1502         if (*flags_ptr & CLONE_VM)
1503                 *flags_ptr |= CLONE_SIGHAND;
1504
1505         /*
1506          * If unsharing signal handlers and the task was created
1507          * using CLONE_THREAD, then must unshare the thread
1508          */
1509         if ((*flags_ptr & CLONE_SIGHAND) &&
1510             (atomic_read(&current->signal->count) > 1))
1511                 *flags_ptr |= CLONE_THREAD;
1512
1513         /*
1514          * If unsharing namespace, must also unshare filesystem information.
1515          */
1516         if (*flags_ptr & CLONE_NEWNS)
1517                 *flags_ptr |= CLONE_FS;
1518 }
1519
1520 /*
1521  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1522  */
1523 static int unshare_thread(unsigned long unshare_flags)
1524 {
1525         if (unshare_flags & CLONE_THREAD)
1526                 return -EINVAL;
1527
1528         return 0;
1529 }
1530
1531 /*
1532  * Unshare the filesystem structure if it is being shared
1533  */
1534 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1535 {
1536         struct fs_struct *fs = current->fs;
1537
1538         if ((unshare_flags & CLONE_FS) &&
1539             (fs && atomic_read(&fs->count) > 1)) {
1540                 *new_fsp = __copy_fs_struct(current->fs);
1541                 if (!*new_fsp)
1542                         return -ENOMEM;
1543         }
1544
1545         return 0;
1546 }
1547
1548 /*
1549  * Unsharing of sighand is not supported yet
1550  */
1551 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1552 {
1553         struct sighand_struct *sigh = current->sighand;
1554
1555         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1556                 return -EINVAL;
1557         else
1558                 return 0;
1559 }
1560
1561 /*
1562  * Unshare vm if it is being shared
1563  */
1564 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1565 {
1566         struct mm_struct *mm = current->mm;
1567
1568         if ((unshare_flags & CLONE_VM) &&
1569             (mm && atomic_read(&mm->mm_users) > 1)) {
1570                 return -EINVAL;
1571         }
1572
1573         return 0;
1574 }
1575
1576 /*
1577  * Unshare file descriptor table if it is being shared
1578  */
1579 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1580 {
1581         struct files_struct *fd = current->files;
1582         int error = 0;
1583
1584         if ((unshare_flags & CLONE_FILES) &&
1585             (fd && atomic_read(&fd->count) > 1)) {
1586                 *new_fdp = dup_fd(fd, &error);
1587                 if (!*new_fdp)
1588                         return error;
1589         }
1590
1591         return 0;
1592 }
1593
1594 /*
1595  * unshare allows a process to 'unshare' part of the process
1596  * context which was originally shared using clone.  copy_*
1597  * functions used by do_fork() cannot be used here directly
1598  * because they modify an inactive task_struct that is being
1599  * constructed. Here we are modifying the current, active,
1600  * task_struct.
1601  */
1602 asmlinkage long sys_unshare(unsigned long unshare_flags)
1603 {
1604         int err = 0;
1605         struct fs_struct *fs, *new_fs = NULL;
1606         struct sighand_struct *new_sigh = NULL;
1607         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1608         struct files_struct *fd, *new_fd = NULL;
1609         struct nsproxy *new_nsproxy = NULL;
1610         int do_sysvsem = 0;
1611
1612         check_unshare_flags(&unshare_flags);
1613
1614         /* Return -EINVAL for all unsupported flags */
1615         err = -EINVAL;
1616         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1617                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1618                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1619                 goto bad_unshare_out;
1620
1621         /*
1622          * CLONE_NEWIPC must also detach from the undolist: after switching
1623          * to a new ipc namespace, the semaphore arrays from the old
1624          * namespace are unreachable.
1625          */
1626         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1627                 do_sysvsem = 1;
1628         if ((err = unshare_thread(unshare_flags)))
1629                 goto bad_unshare_out;
1630         if ((err = unshare_fs(unshare_flags, &new_fs)))
1631                 goto bad_unshare_cleanup_thread;
1632         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1633                 goto bad_unshare_cleanup_fs;
1634         if ((err = unshare_vm(unshare_flags, &new_mm)))
1635                 goto bad_unshare_cleanup_sigh;
1636         if ((err = unshare_fd(unshare_flags, &new_fd)))
1637                 goto bad_unshare_cleanup_vm;
1638         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1639                         new_fs)))
1640                 goto bad_unshare_cleanup_fd;
1641
1642         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1643                 if (do_sysvsem) {
1644                         /*
1645                          * CLONE_SYSVSEM is equivalent to sys_exit().
1646                          */
1647                         exit_sem(current);
1648                 }
1649
1650                 if (new_nsproxy) {
1651                         switch_task_namespaces(current, new_nsproxy);
1652                         new_nsproxy = NULL;
1653                 }
1654
1655                 task_lock(current);
1656
1657                 if (new_fs) {
1658                         fs = current->fs;
1659                         current->fs = new_fs;
1660                         new_fs = fs;
1661                 }
1662
1663                 if (new_mm) {
1664                         mm = current->mm;
1665                         active_mm = current->active_mm;
1666                         current->mm = new_mm;
1667                         current->active_mm = new_mm;
1668                         activate_mm(active_mm, new_mm);
1669                         new_mm = mm;
1670                 }
1671
1672                 if (new_fd) {
1673                         fd = current->files;
1674                         current->files = new_fd;
1675                         new_fd = fd;
1676                 }
1677
1678                 task_unlock(current);
1679         }
1680
1681         if (new_nsproxy)
1682                 put_nsproxy(new_nsproxy);
1683
1684 bad_unshare_cleanup_fd:
1685         if (new_fd)
1686                 put_files_struct(new_fd);
1687
1688 bad_unshare_cleanup_vm:
1689         if (new_mm)
1690                 mmput(new_mm);
1691
1692 bad_unshare_cleanup_sigh:
1693         if (new_sigh)
1694                 if (atomic_dec_and_test(&new_sigh->count))
1695                         kmem_cache_free(sighand_cachep, new_sigh);
1696
1697 bad_unshare_cleanup_fs:
1698         if (new_fs)
1699                 put_fs_struct(new_fs);
1700
1701 bad_unshare_cleanup_thread:
1702 bad_unshare_out:
1703         return err;
1704 }
1705
1706 /*
1707  *      Helper to unshare the files of the current task.
1708  *      We don't want to expose copy_files internals to
1709  *      the exec layer of the kernel.
1710  */
1711
1712 int unshare_files(struct files_struct **displaced)
1713 {
1714         struct task_struct *task = current;
1715         struct files_struct *copy = NULL;
1716         int error;
1717
1718         error = unshare_fd(CLONE_FILES, &copy);
1719         if (error || !copy) {
1720                 *displaced = NULL;
1721                 return error;
1722         }
1723         *displaced = task->files;
1724         task_lock(task);
1725         task->files = copy;
1726         task_unlock(task);
1727         return 0;
1728 }