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