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/fs.h>
  31 #include <linux/nsproxy.h>
  32 #include <linux/capability.h>
  33 #include <linux/cpu.h>
  34 #include <linux/cgroup.h>
  35 #include <linux/security.h>
  36 #include <linux/hugetlb.h>
  37 #include <linux/swap.h>
  38 #include <linux/syscalls.h>
  39 #include <linux/jiffies.h>
  40 #include <linux/tracehook.h>
  41 #include <linux/futex.h>
  42 #include <linux/task_io_accounting_ops.h>
  43 #include <linux/rcupdate.h>
  44 #include <linux/ptrace.h>
  45 #include <linux/mount.h>
  46 #include <linux/audit.h>
  47 #include <linux/memcontrol.h>
  48 #include <linux/profile.h>
  49 #include <linux/rmap.h>
  50 #include <linux/acct.h>
  51 #include <linux/tsacct_kern.h>
  52 #include <linux/cn_proc.h>
  53 #include <linux/freezer.h>
  54 #include <linux/delayacct.h>
  55 #include <linux/taskstats_kern.h>
  56 #include <linux/random.h>
  57 #include <linux/tty.h>
  58 #include <linux/proc_fs.h>
  59 #include <linux/blkdev.h>
  60
  61 #include <asm/pgtable.h>
  62 #include <asm/pgalloc.h>
  63 #include <asm/uaccess.h>
  64 #include <asm/mmu_context.h>
  65 #include <asm/cacheflush.h>
  66 #include <asm/tlbflush.h>
  67
  68 /*
  69  * Protected counters by write_lock_irq(&tasklist_lock)
  70  */
  71 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  72 int nr_threads;                 /* The idle threads do not count.. */
  73
  74 int max_threads;                /* tunable limit on nr_threads */
  75
  76 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  77
  78 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  79
  80 int nr_processes(void)
  81 {
  82         int cpu;
  83         int total = 0;
  84
  85         for_each_online_cpu(cpu)
  86                 total += per_cpu(process_counts, cpu);
  87
  88         return total;
  89 }
  90
  91 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  92 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  93 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  94 static struct kmem_cache *task_struct_cachep;
  95 #endif
  96
  97 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
  98 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
  99 {
 100 #ifdef CONFIG_DEBUG_STACK_USAGE
 101         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 102 #else
 103         gfp_t mask = GFP_KERNEL;
 104 #endif
 105         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
 106 }
 107
 108 static inline void free_thread_info(struct thread_info *ti)
 109 {
 110         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 111 }
 112 #endif
 113
 114 /* SLAB cache for signal_struct structures (tsk->signal) */
 115 static struct kmem_cache *signal_cachep;
 116
 117 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 118 struct kmem_cache *sighand_cachep;
 119
 120 /* SLAB cache for files_struct structures (tsk->files) */
 121 struct kmem_cache *files_cachep;
 122
 123 /* SLAB cache for fs_struct structures (tsk->fs) */
 124 struct kmem_cache *fs_cachep;
 125
 126 /* SLAB cache for vm_area_struct structures */
 127 struct kmem_cache *vm_area_cachep;
 128
 129 /* SLAB cache for mm_struct structures (tsk->mm) */
 130 static struct kmem_cache *mm_cachep;
 131
 132 void free_task(struct task_struct *tsk)
 133 {
 134         prop_local_destroy_single(&tsk->dirties);
 135         free_thread_info(tsk->stack);
 136         rt_mutex_debug_task_free(tsk);
 137         free_task_struct(tsk);
 138 }
 139 EXPORT_SYMBOL(free_task);
 140
 141 void __put_task_struct(struct task_struct *tsk)
 142 {
 143         WARN_ON(!tsk->exit_state);
 144         WARN_ON(atomic_read(&tsk->usage));
 145         WARN_ON(tsk == current);
 146
 147         security_task_free(tsk);
 148         free_uid(tsk->user);
 149         put_group_info(tsk->group_info);
 150         delayacct_tsk_free(tsk);
 151
 152         if (!profile_handoff_task(tsk))
 153                 free_task(tsk);
 154 }
 155
 156 /*
 157  * macro override instead of weak attribute alias, to workaround
 158  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 159  */
 160 #ifndef arch_task_cache_init
 161 #define arch_task_cache_init()
 162 #endif
 163
 164 void __init fork_init(unsigned long mempages)
 165 {
 166 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 167 #ifndef ARCH_MIN_TASKALIGN
 168 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 169 #endif
 170         /* create a slab on which task_structs can be allocated */
 171         task_struct_cachep =
 172                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 173                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
 174 #endif
 175
 176         /* do the arch specific task caches init */
 177         arch_task_cache_init();
 178
 179         /*
 180          * The default maximum number of threads is set to a safe
 181          * value: the thread structures can take up at most half
 182          * of memory.
 183          */
 184         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 185
 186         /*
 187          * we need to allow at least 20 threads to boot a system
 188          */
 189         if(max_threads < 20)
 190                 max_threads = 20;
 191
 192         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 193         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 194         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 195                 init_task.signal->rlim[RLIMIT_NPROC];
 196 }
 197
 198 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 199                                                struct task_struct *src)
 200 {
 201         *dst = *src;
 202         return 0;
 203 }
 204
 205 static struct task_struct *dup_task_struct(struct task_struct *orig)
 206 {
 207         struct task_struct *tsk;
 208         struct thread_info *ti;
 209         int err;
 210
 211         prepare_to_copy(orig);
 212
 213         tsk = alloc_task_struct();
 214         if (!tsk)
 215                 return NULL;
 216
 217         ti = alloc_thread_info(tsk);
 218         if (!ti) {
 219                 free_task_struct(tsk);
 220                 return NULL;
 221         }
 222
 223         err = arch_dup_task_struct(tsk, orig);
 224         if (err)
 225                 goto out;
 226
 227         tsk->stack = ti;
 228
 229         err = prop_local_init_single(&tsk->dirties);
 230         if (err)
 231                 goto out;
 232
 233         setup_thread_stack(tsk, orig);
 234
 235 #ifdef CONFIG_CC_STACKPROTECTOR
 236         tsk->stack_canary = get_random_int();
 237 #endif
 238
 239         /* One for us, one for whoever does the "release_task()" (usually parent) */
 240         atomic_set(&tsk->usage,2);
 241         atomic_set(&tsk->fs_excl, 0);
 242 #ifdef CONFIG_BLK_DEV_IO_TRACE
 243         tsk->btrace_seq = 0;
 244 #endif
 245         tsk->splice_pipe = NULL;
 246         return tsk;
 247
 248 out:
 249         free_thread_info(ti);
 250         free_task_struct(tsk);
 251         return NULL;
 252 }
 253
 254 #ifdef CONFIG_MMU
 255 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 256 {
 257         struct vm_area_struct *mpnt, *tmp, **pprev;
 258         struct rb_node **rb_link, *rb_parent;
 259         int retval;
 260         unsigned long charge;
 261         struct mempolicy *pol;
 262
 263         down_write(&oldmm->mmap_sem);
 264         flush_cache_dup_mm(oldmm);
 265         /*
 266          * Not linked in yet - no deadlock potential:
 267          */
 268         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 269
 270         mm->locked_vm = 0;
 271         mm->mmap = NULL;
 272         mm->mmap_cache = NULL;
 273         mm->free_area_cache = oldmm->mmap_base;
 274         mm->cached_hole_size = ~0UL;
 275         mm->map_count = 0;
 276         cpus_clear(mm->cpu_vm_mask);
 277         mm->mm_rb = RB_ROOT;
 278         rb_link = &mm->mm_rb.rb_node;
 279         rb_parent = NULL;
 280         pprev = &mm->mmap;
 281
 282         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 283                 struct file *file;
 284
 285                 if (mpnt->vm_flags & VM_DONTCOPY) {
 286                         long pages = vma_pages(mpnt);
 287                         mm->total_vm -= pages;
 288                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 289                                                                 -pages);
 290                         continue;
 291                 }
 292                 charge = 0;
 293                 if (mpnt->vm_flags & VM_ACCOUNT) {
 294                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 295                         if (security_vm_enough_memory(len))
 296                                 goto fail_nomem;
 297                         charge = len;
 298                 }
 299                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 300                 if (!tmp)
 301                         goto fail_nomem;
 302                 *tmp = *mpnt;
 303                 pol = mpol_dup(vma_policy(mpnt));
 304                 retval = PTR_ERR(pol);
 305                 if (IS_ERR(pol))
 306                         goto fail_nomem_policy;
 307                 vma_set_policy(tmp, pol);
 308                 tmp->vm_flags &= ~VM_LOCKED;
 309                 tmp->vm_mm = mm;
 310                 tmp->vm_next = NULL;
 311                 anon_vma_link(tmp);
 312                 file = tmp->vm_file;
 313                 if (file) {
 314                         struct inode *inode = file->f_path.dentry->d_inode;
 315                         get_file(file);
 316                         if (tmp->vm_flags & VM_DENYWRITE)
 317                                 atomic_dec(&inode->i_writecount);
 318
 319                         /* insert tmp into the share list, just after mpnt */
 320                         spin_lock(&file->f_mapping->i_mmap_lock);
 321                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 322                         flush_dcache_mmap_lock(file->f_mapping);
 323                         vma_prio_tree_add(tmp, mpnt);
 324                         flush_dcache_mmap_unlock(file->f_mapping);
 325                         spin_unlock(&file->f_mapping->i_mmap_lock);
 326                 }
 327
 328                 /*
 329                  * Clear hugetlb-related page reserves for children. This only
 330                  * affects MAP_PRIVATE mappings. Faults generated by the child
 331                  * are not guaranteed to succeed, even if read-only
 332                  */
 333                 if (is_vm_hugetlb_page(tmp))
 334                         reset_vma_resv_huge_pages(tmp);
 335
 336                 /*
 337                  * Link in the new vma and copy the page table entries.
 338                  */
 339                 *pprev = tmp;
 340                 pprev = &tmp->vm_next;
 341
 342                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 343                 rb_link = &tmp->vm_rb.rb_right;
 344                 rb_parent = &tmp->vm_rb;
 345
 346                 mm->map_count++;
 347                 retval = copy_page_range(mm, oldmm, mpnt);
 348
 349                 if (tmp->vm_ops && tmp->vm_ops->open)
 350                         tmp->vm_ops->open(tmp);
 351
 352                 if (retval)
 353                         goto out;
 354         }
 355         /* a new mm has just been created */
 356         arch_dup_mmap(oldmm, mm);
 357         retval = 0;
 358 out:
 359         up_write(&mm->mmap_sem);
 360         flush_tlb_mm(oldmm);
 361         up_write(&oldmm->mmap_sem);
 362         return retval;
 363 fail_nomem_policy:
 364         kmem_cache_free(vm_area_cachep, tmp);
 365 fail_nomem:
 366         retval = -ENOMEM;
 367         vm_unacct_memory(charge);
 368         goto out;
 369 }
 370
 371 static inline int mm_alloc_pgd(struct mm_struct * mm)
 372 {
 373         mm->pgd = pgd_alloc(mm);
 374         if (unlikely(!mm->pgd))
 375                 return -ENOMEM;
 376         return 0;
 377 }
 378
 379 static inline void mm_free_pgd(struct mm_struct * mm)
 380 {
 381         pgd_free(mm, mm->pgd);
 382 }
 383 #else
 384 #define dup_mmap(mm, oldmm)     (0)
 385 #define mm_alloc_pgd(mm)        (0)
 386 #define mm_free_pgd(mm)
 387 #endif /* CONFIG_MMU */
 388
 389 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 390
 391 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 392 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 393
 394 #include <linux/init_task.h>
 395
 396 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
 397 {
 398         atomic_set(&mm->mm_users, 1);
 399         atomic_set(&mm->mm_count, 1);
 400         init_rwsem(&mm->mmap_sem);
 401         INIT_LIST_HEAD(&mm->mmlist);
 402         mm->flags = (current->mm) ? current->mm->flags
 403                                   : MMF_DUMP_FILTER_DEFAULT;
 404         mm->core_state = NULL;
 405         mm->nr_ptes = 0;
 406         set_mm_counter(mm, file_rss, 0);
 407         set_mm_counter(mm, anon_rss, 0);
 408         spin_lock_init(&mm->page_table_lock);
 409         rwlock_init(&mm->ioctx_list_lock);
 410         mm->ioctx_list = NULL;
 411         mm->free_area_cache = TASK_UNMAPPED_BASE;
 412         mm->cached_hole_size = ~0UL;
 413         mm_init_owner(mm, p);
 414
 415         if (likely(!mm_alloc_pgd(mm))) {
 416                 mm->def_flags = 0;
 417                 return mm;
 418         }
 419
 420         free_mm(mm);
 421         return NULL;
 422 }
 423
 424 /*
 425  * Allocate and initialize an mm_struct.
 426  */
 427 struct mm_struct * mm_alloc(void)
 428 {
 429         struct mm_struct * mm;
 430
 431         mm = allocate_mm();
 432         if (mm) {
 433                 memset(mm, 0, sizeof(*mm));
 434                 mm = mm_init(mm, current);
 435         }
 436         return mm;
 437 }
 438
 439 /*
 440  * Called when the last reference to the mm
 441  * is dropped: either by a lazy thread or by
 442  * mmput. Free the page directory and the mm.
 443  */
 444 void __mmdrop(struct mm_struct *mm)
 445 {
 446         BUG_ON(mm == &init_mm);
 447         mm_free_pgd(mm);
 448         destroy_context(mm);
 449         free_mm(mm);
 450 }
 451 EXPORT_SYMBOL_GPL(__mmdrop);
 452
 453 /*
 454  * Decrement the use count and release all resources for an mm.
 455  */
 456 void mmput(struct mm_struct *mm)
 457 {
 458         might_sleep();
 459
 460         if (atomic_dec_and_test(&mm->mm_users)) {
 461                 exit_aio(mm);
 462                 exit_mmap(mm);
 463                 set_mm_exe_file(mm, NULL);
 464                 if (!list_empty(&mm->mmlist)) {
 465                         spin_lock(&mmlist_lock);
 466                         list_del(&mm->mmlist);
 467                         spin_unlock(&mmlist_lock);
 468                 }
 469                 put_swap_token(mm);
 470                 mmdrop(mm);
 471         }
 472 }
 473 EXPORT_SYMBOL_GPL(mmput);
 474
 475 /**
 476  * get_task_mm - acquire a reference to the task's mm
 477  *
 478  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 479  * this kernel workthread has transiently adopted a user mm with use_mm,
 480  * to do its AIO) is not set and if so returns a reference to it, after
 481  * bumping up the use count.  User must release the mm via mmput()
 482  * after use.  Typically used by /proc and ptrace.
 483  */
 484 struct mm_struct *get_task_mm(struct task_struct *task)
 485 {
 486         struct mm_struct *mm;
 487
 488         task_lock(task);
 489         mm = task->mm;
 490         if (mm) {
 491                 if (task->flags & PF_KTHREAD)
 492                         mm = NULL;
 493                 else
 494                         atomic_inc(&mm->mm_users);
 495         }
 496         task_unlock(task);
 497         return mm;
 498 }
 499 EXPORT_SYMBOL_GPL(get_task_mm);
 500
 501 /* Please note the differences between mmput and mm_release.
 502  * mmput is called whenever we stop holding onto a mm_struct,
 503  * error success whatever.
 504  *
 505  * mm_release is called after a mm_struct has been removed
 506  * from the current process.
 507  *
 508  * This difference is important for error handling, when we
 509  * only half set up a mm_struct for a new process and need to restore
 510  * the old one.  Because we mmput the new mm_struct before
 511  * restoring the old one. . .
 512  * Eric Biederman 10 January 1998
 513  */
 514 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 515 {
 516         struct completion *vfork_done = tsk->vfork_done;
 517
 518         /* Get rid of any cached register state */
 519         deactivate_mm(tsk, mm);
 520
 521         /* notify parent sleeping on vfork() */
 522         if (vfork_done) {
 523                 tsk->vfork_done = NULL;
 524                 complete(vfork_done);
 525         }
 526
 527         /*
 528          * If we're exiting normally, clear a user-space tid field if
 529          * requested.  We leave this alone when dying by signal, to leave
 530          * the value intact in a core dump, and to save the unnecessary
 531          * trouble otherwise.  Userland only wants this done for a sys_exit.
 532          */
 533         if (tsk->clear_child_tid
 534             && !(tsk->flags & PF_SIGNALED)
 535             && atomic_read(&mm->mm_users) > 1) {
 536                 u32 __user * tidptr = tsk->clear_child_tid;
 537                 tsk->clear_child_tid = NULL;
 538
 539                 /*
 540                  * We don't check the error code - if userspace has
 541                  * not set up a proper pointer then tough luck.
 542                  */
 543                 put_user(0, tidptr);
 544                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 545         }
 546 }
 547
 548 /*
 549  * Allocate a new mm structure and copy contents from the
 550  * mm structure of the passed in task structure.
 551  */
 552 struct mm_struct *dup_mm(struct task_struct *tsk)
 553 {
 554         struct mm_struct *mm, *oldmm = current->mm;
 555         int err;
 556
 557         if (!oldmm)
 558                 return NULL;
 559
 560         mm = allocate_mm();
 561         if (!mm)
 562                 goto fail_nomem;
 563
 564         memcpy(mm, oldmm, sizeof(*mm));
 565
 566         /* Initializing for Swap token stuff */
 567         mm->token_priority = 0;
 568         mm->last_interval = 0;
 569
 570         if (!mm_init(mm, tsk))
 571                 goto fail_nomem;
 572
 573         if (init_new_context(tsk, mm))
 574                 goto fail_nocontext;
 575
 576         dup_mm_exe_file(oldmm, mm);
 577
 578         err = dup_mmap(mm, oldmm);
 579         if (err)
 580                 goto free_pt;
 581
 582         mm->hiwater_rss = get_mm_rss(mm);
 583         mm->hiwater_vm = mm->total_vm;
 584
 585         return mm;
 586
 587 free_pt:
 588         mmput(mm);
 589
 590 fail_nomem:
 591         return NULL;
 592
 593 fail_nocontext:
 594         /*
 595          * If init_new_context() failed, we cannot use mmput() to free the mm
 596          * because it calls destroy_context()
 597          */
 598         mm_free_pgd(mm);
 599         free_mm(mm);
 600         return NULL;
 601 }
 602
 603 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 604 {
 605         struct mm_struct * mm, *oldmm;
 606         int retval;
 607
 608         tsk->min_flt = tsk->maj_flt = 0;
 609         tsk->nvcsw = tsk->nivcsw = 0;
 610
 611         tsk->mm = NULL;
 612         tsk->active_mm = NULL;
 613
 614         /*
 615          * Are we cloning a kernel thread?
 616          *
 617          * We need to steal a active VM for that..
 618          */
 619         oldmm = current->mm;
 620         if (!oldmm)
 621                 return 0;
 622
 623         if (clone_flags & CLONE_VM) {
 624                 atomic_inc(&oldmm->mm_users);
 625                 mm = oldmm;
 626                 goto good_mm;
 627         }
 628
 629         retval = -ENOMEM;
 630         mm = dup_mm(tsk);
 631         if (!mm)
 632                 goto fail_nomem;
 633
 634 good_mm:
 635         /* Initializing for Swap token stuff */
 636         mm->token_priority = 0;
 637         mm->last_interval = 0;
 638
 639         tsk->mm = mm;
 640         tsk->active_mm = mm;
 641         return 0;
 642
 643 fail_nomem:
 644         return retval;
 645 }
 646
 647 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 648 {
 649         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 650         /* We don't need to lock fs - think why ;-) */
 651         if (fs) {
 652                 atomic_set(&fs->count, 1);
 653                 rwlock_init(&fs->lock);
 654                 fs->umask = old->umask;
 655                 read_lock(&old->lock);
 656                 fs->root = old->root;
 657                 path_get(&old->root);
 658                 fs->pwd = old->pwd;
 659                 path_get(&old->pwd);
 660                 read_unlock(&old->lock);
 661         }
 662         return fs;
 663 }
 664
 665 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 666 {
 667         return __copy_fs_struct(old);
 668 }
 669
 670 EXPORT_SYMBOL_GPL(copy_fs_struct);
 671
 672 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 673 {
 674         if (clone_flags & CLONE_FS) {
 675                 atomic_inc(&current->fs->count);
 676                 return 0;
 677         }
 678         tsk->fs = __copy_fs_struct(current->fs);
 679         if (!tsk->fs)
 680                 return -ENOMEM;
 681         return 0;
 682 }
 683
 684 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 685 {
 686         struct files_struct *oldf, *newf;
 687         int error = 0;
 688
 689         /*
 690          * A background process may not have any files ...
 691          */
 692         oldf = current->files;
 693         if (!oldf)
 694                 goto out;
 695
 696         if (clone_flags & CLONE_FILES) {
 697                 atomic_inc(&oldf->count);
 698                 goto out;
 699         }
 700
 701         newf = dup_fd(oldf, &error);
 702         if (!newf)
 703                 goto out;
 704
 705         tsk->files = newf;
 706         error = 0;
 707 out:
 708         return error;
 709 }
 710
 711 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 712 {
 713 #ifdef CONFIG_BLOCK
 714         struct io_context *ioc = current->io_context;
 715
 716         if (!ioc)
 717                 return 0;
 718         /*
 719          * Share io context with parent, if CLONE_IO is set
 720          */
 721         if (clone_flags & CLONE_IO) {
 722                 tsk->io_context = ioc_task_link(ioc);
 723                 if (unlikely(!tsk->io_context))
 724                         return -ENOMEM;
 725         } else if (ioprio_valid(ioc->ioprio)) {
 726                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 727                 if (unlikely(!tsk->io_context))
 728                         return -ENOMEM;
 729
 730                 tsk->io_context->ioprio = ioc->ioprio;
 731         }
 732 #endif
 733         return 0;
 734 }
 735
 736 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 737 {
 738         struct sighand_struct *sig;
 739
 740         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 741                 atomic_inc(&current->sighand->count);
 742                 return 0;
 743         }
 744         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 745         rcu_assign_pointer(tsk->sighand, sig);
 746         if (!sig)
 747                 return -ENOMEM;
 748         atomic_set(&sig->count, 1);
 749         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 750         return 0;
 751 }
 752
 753 void __cleanup_sighand(struct sighand_struct *sighand)
 754 {
 755         if (atomic_dec_and_test(&sighand->count))
 756                 kmem_cache_free(sighand_cachep, sighand);
 757 }
 758
 759 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 760 {
 761         struct signal_struct *sig;
 762         int ret;
 763
 764         if (clone_flags & CLONE_THREAD) {
 765                 atomic_inc(&current->signal->count);
 766                 atomic_inc(&current->signal->live);
 767                 return 0;
 768         }
 769         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 770         tsk->signal = sig;
 771         if (!sig)
 772                 return -ENOMEM;
 773
 774         ret = copy_thread_group_keys(tsk);
 775         if (ret < 0) {
 776                 kmem_cache_free(signal_cachep, sig);
 777                 return ret;
 778         }
 779
 780         atomic_set(&sig->count, 1);
 781         atomic_set(&sig->live, 1);
 782         init_waitqueue_head(&sig->wait_chldexit);
 783         sig->flags = 0;
 784         sig->group_exit_code = 0;
 785         sig->group_exit_task = NULL;
 786         sig->group_stop_count = 0;
 787         sig->curr_target = tsk;
 788         init_sigpending(&sig->shared_pending);
 789         INIT_LIST_HEAD(&sig->posix_timers);
 790
 791         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 792         sig->it_real_incr.tv64 = 0;
 793         sig->real_timer.function = it_real_fn;
 794
 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         sig->leader = 0;        /* session leadership doesn't inherit */
 801         sig->tty_old_pgrp = NULL;
 802
 803         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 804         sig->gtime = cputime_zero;
 805         sig->cgtime = cputime_zero;
 806         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 807         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 808         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
 809         task_io_accounting_init(&sig->ioac);
 810         sig->sum_sched_runtime = 0;
 811         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 812         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 813         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 814         taskstats_tgid_init(sig);
 815
 816         task_lock(current->group_leader);
 817         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 818         task_unlock(current->group_leader);
 819
 820         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 821                 /*
 822                  * New sole thread in the process gets an expiry time
 823                  * of the whole CPU time limit.
 824                  */
 825                 tsk->it_prof_expires =
 826                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 827         }
 828         acct_init_pacct(&sig->pacct);
 829
 830         tty_audit_fork(sig);
 831
 832         return 0;
 833 }
 834
 835 void __cleanup_signal(struct signal_struct *sig)
 836 {
 837         exit_thread_group_keys(sig);
 838         kmem_cache_free(signal_cachep, sig);
 839 }
 840
 841 static void cleanup_signal(struct task_struct *tsk)
 842 {
 843         struct signal_struct *sig = tsk->signal;
 844
 845         atomic_dec(&sig->live);
 846
 847         if (atomic_dec_and_test(&sig->count))
 848                 __cleanup_signal(sig);
 849 }
 850
 851 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 852 {
 853         unsigned long new_flags = p->flags;
 854
 855         new_flags &= ~PF_SUPERPRIV;
 856         new_flags |= PF_FORKNOEXEC;
 857         new_flags |= PF_STARTING;
 858         p->flags = new_flags;
 859         clear_freeze_flag(p);
 860 }
 861
 862 asmlinkage long sys_set_tid_address(int __user *tidptr)
 863 {
 864         current->clear_child_tid = tidptr;
 865
 866         return task_pid_vnr(current);
 867 }
 868
 869 static void rt_mutex_init_task(struct task_struct *p)
 870 {
 871         spin_lock_init(&p->pi_lock);
 872 #ifdef CONFIG_RT_MUTEXES
 873         plist_head_init(&p->pi_waiters, &p->pi_lock);
 874         p->pi_blocked_on = NULL;
 875 #endif
 876 }
 877
 878 #ifdef CONFIG_MM_OWNER
 879 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 880 {
 881         mm->owner = p;
 882 }
 883 #endif /* CONFIG_MM_OWNER */
 884
 885 /*
 886  * This creates a new process as a copy of the old one,
 887  * but does not actually start it yet.
 888  *
 889  * It copies the registers, and all the appropriate
 890  * parts of the process environment (as per the clone
 891  * flags). The actual kick-off is left to the caller.
 892  */
 893 static struct task_struct *copy_process(unsigned long clone_flags,
 894                                         unsigned long stack_start,
 895                                         struct pt_regs *regs,
 896                                         unsigned long stack_size,
 897                                         int __user *child_tidptr,
 898                                         struct pid *pid,
 899                                         int trace)
 900 {
 901         int retval;
 902         struct task_struct *p;
 903         int cgroup_callbacks_done = 0;
 904
 905         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 906                 return ERR_PTR(-EINVAL);
 907
 908         /*
 909          * Thread groups must share signals as well, and detached threads
 910          * can only be started up within the thread group.
 911          */
 912         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 913                 return ERR_PTR(-EINVAL);
 914
 915         /*
 916          * Shared signal handlers imply shared VM. By way of the above,
 917          * thread groups also imply shared VM. Blocking this case allows
 918          * for various simplifications in other code.
 919          */
 920         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 921                 return ERR_PTR(-EINVAL);
 922
 923         retval = security_task_create(clone_flags);
 924         if (retval)
 925                 goto fork_out;
 926
 927         retval = -ENOMEM;
 928         p = dup_task_struct(current);
 929         if (!p)
 930                 goto fork_out;
 931
 932         rt_mutex_init_task(p);
 933
 934 #ifdef CONFIG_PROVE_LOCKING
 935         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
 936         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
 937 #endif
 938         retval = -EAGAIN;
 939         if (atomic_read(&p->user->processes) >=
 940                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
 941                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 942                     p->user != current->nsproxy->user_ns->root_user)
 943                         goto bad_fork_free;
 944         }
 945
 946         atomic_inc(&p->user->__count);
 947         atomic_inc(&p->user->processes);
 948         get_group_info(p->group_info);
 949
 950         /*
 951          * If multiple threads are within copy_process(), then this check
 952          * triggers too late. This doesn't hurt, the check is only there
 953          * to stop root fork bombs.
 954          */
 955         if (nr_threads >= max_threads)
 956                 goto bad_fork_cleanup_count;
 957
 958         if (!try_module_get(task_thread_info(p)->exec_domain->module))
 959                 goto bad_fork_cleanup_count;
 960
 961         if (p->binfmt && !try_module_get(p->binfmt->module))
 962                 goto bad_fork_cleanup_put_domain;
 963
 964         p->did_exec = 0;
 965         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
 966         copy_flags(clone_flags, p);
 967         INIT_LIST_HEAD(&p->children);
 968         INIT_LIST_HEAD(&p->sibling);
 969 #ifdef CONFIG_PREEMPT_RCU
 970         p->rcu_read_lock_nesting = 0;
 971         p->rcu_flipctr_idx = 0;
 972 #endif /* #ifdef CONFIG_PREEMPT_RCU */
 973         p->vfork_done = NULL;
 974         spin_lock_init(&p->alloc_lock);
 975
 976         clear_tsk_thread_flag(p, TIF_SIGPENDING);
 977         init_sigpending(&p->pending);
 978
 979         p->utime = cputime_zero;
 980         p->stime = cputime_zero;
 981         p->gtime = cputime_zero;
 982         p->utimescaled = cputime_zero;
 983         p->stimescaled = cputime_zero;
 984         p->prev_utime = cputime_zero;
 985         p->prev_stime = cputime_zero;
 986
 987 #ifdef CONFIG_DETECT_SOFTLOCKUP
 988         p->last_switch_count = 0;
 989         p->last_switch_timestamp = 0;
 990 #endif
 991
 992         task_io_accounting_init(&p->ioac);
 993         acct_clear_integrals(p);
 994
 995         p->it_virt_expires = cputime_zero;
 996         p->it_prof_expires = cputime_zero;
 997         p->it_sched_expires = 0;
 998         INIT_LIST_HEAD(&p->cpu_timers[0]);
 999         INIT_LIST_HEAD(&p->cpu_timers[1]);
1000         INIT_LIST_HEAD(&p->cpu_timers[2]);
1001
1002         p->lock_depth = -1;             /* -1 = no lock */
1003         do_posix_clock_monotonic_gettime(&p->start_time);
1004         p->real_start_time = p->start_time;
1005         monotonic_to_bootbased(&p->real_start_time);
1006 #ifdef CONFIG_SECURITY
1007         p->security = NULL;
1008 #endif
1009         p->cap_bset = current->cap_bset;
1010         p->io_context = NULL;
1011         p->audit_context = NULL;
1012         cgroup_fork(p);
1013 #ifdef CONFIG_NUMA
1014         p->mempolicy = mpol_dup(p->mempolicy);
1015         if (IS_ERR(p->mempolicy)) {
1016                 retval = PTR_ERR(p->mempolicy);
1017                 p->mempolicy = NULL;
1018                 goto bad_fork_cleanup_cgroup;
1019         }
1020         mpol_fix_fork_child_flag(p);
1021 #endif
1022 #ifdef CONFIG_TRACE_IRQFLAGS
1023         p->irq_events = 0;
1024 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1025         p->hardirqs_enabled = 1;
1026 #else
1027         p->hardirqs_enabled = 0;
1028 #endif
1029         p->hardirq_enable_ip = 0;
1030         p->hardirq_enable_event = 0;
1031         p->hardirq_disable_ip = _THIS_IP_;
1032         p->hardirq_disable_event = 0;
1033         p->softirqs_enabled = 1;
1034         p->softirq_enable_ip = _THIS_IP_;
1035         p->softirq_enable_event = 0;
1036         p->softirq_disable_ip = 0;
1037         p->softirq_disable_event = 0;
1038         p->hardirq_context = 0;
1039         p->softirq_context = 0;
1040 #endif
1041 #ifdef CONFIG_LOCKDEP
1042         p->lockdep_depth = 0; /* no locks held yet */
1043         p->curr_chain_key = 0;
1044         p->lockdep_recursion = 0;
1045 #endif
1046
1047 #ifdef CONFIG_DEBUG_MUTEXES
1048         p->blocked_on = NULL; /* not blocked yet */
1049 #endif
1050
1051         /* Perform scheduler related setup. Assign this task to a CPU. */
1052         sched_fork(p, clone_flags);
1053
1054         if ((retval = security_task_alloc(p)))
1055                 goto bad_fork_cleanup_policy;
1056         if ((retval = audit_alloc(p)))
1057                 goto bad_fork_cleanup_security;
1058         /* copy all the process information */
1059         if ((retval = copy_semundo(clone_flags, p)))
1060                 goto bad_fork_cleanup_audit;
1061         if ((retval = copy_files(clone_flags, p)))
1062                 goto bad_fork_cleanup_semundo;
1063         if ((retval = copy_fs(clone_flags, p)))
1064                 goto bad_fork_cleanup_files;
1065         if ((retval = copy_sighand(clone_flags, p)))
1066                 goto bad_fork_cleanup_fs;
1067         if ((retval = copy_signal(clone_flags, p)))
1068                 goto bad_fork_cleanup_sighand;
1069         if ((retval = copy_mm(clone_flags, p)))
1070                 goto bad_fork_cleanup_signal;
1071         if ((retval = copy_keys(clone_flags, p)))
1072                 goto bad_fork_cleanup_mm;
1073         if ((retval = copy_namespaces(clone_flags, p)))
1074                 goto bad_fork_cleanup_keys;
1075         if ((retval = copy_io(clone_flags, p)))
1076                 goto bad_fork_cleanup_namespaces;
1077         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1078         if (retval)
1079                 goto bad_fork_cleanup_io;
1080
1081         if (pid != &init_struct_pid) {
1082                 retval = -ENOMEM;
1083                 pid = alloc_pid(task_active_pid_ns(p));
1084                 if (!pid)
1085                         goto bad_fork_cleanup_io;
1086
1087                 if (clone_flags & CLONE_NEWPID) {
1088                         retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1089                         if (retval < 0)
1090                                 goto bad_fork_free_pid;
1091                 }
1092         }
1093
1094         p->pid = pid_nr(pid);
1095         p->tgid = p->pid;
1096         if (clone_flags & CLONE_THREAD)
1097                 p->tgid = current->tgid;
1098
1099         if (current->nsproxy != p->nsproxy) {
1100                 retval = ns_cgroup_clone(p, pid);
1101                 if (retval)
1102                         goto bad_fork_free_pid;
1103         }
1104
1105         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1106         /*
1107          * Clear TID on mm_release()?
1108          */
1109         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1110 #ifdef CONFIG_FUTEX
1111         p->robust_list = NULL;
1112 #ifdef CONFIG_COMPAT
1113         p->compat_robust_list = NULL;
1114 #endif
1115         INIT_LIST_HEAD(&p->pi_state_list);
1116         p->pi_state_cache = NULL;
1117 #endif
1118         /*
1119          * sigaltstack should be cleared when sharing the same VM
1120          */
1121         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1122                 p->sas_ss_sp = p->sas_ss_size = 0;
1123
1124         /*
1125          * Syscall tracing should be turned off in the child regardless
1126          * of CLONE_PTRACE.
1127          */
1128         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1129 #ifdef TIF_SYSCALL_EMU
1130         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1131 #endif
1132         clear_all_latency_tracing(p);
1133
1134         /* Our parent execution domain becomes current domain
1135            These must match for thread signalling to apply */
1136         p->parent_exec_id = p->self_exec_id;
1137
1138         /* ok, now we should be set up.. */
1139         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1140         p->pdeath_signal = 0;
1141         p->exit_state = 0;
1142
1143         /*
1144          * Ok, make it visible to the rest of the system.
1145          * We dont wake it up yet.
1146          */
1147         p->group_leader = p;
1148         INIT_LIST_HEAD(&p->thread_group);
1149
1150         /* Now that the task is set up, run cgroup callbacks if
1151          * necessary. We need to run them before the task is visible
1152          * on the tasklist. */
1153         cgroup_fork_callbacks(p);
1154         cgroup_callbacks_done = 1;
1155
1156         /* Need tasklist lock for parent etc handling! */
1157         write_lock_irq(&tasklist_lock);
1158
1159         /*
1160          * The task hasn't been attached yet, so its cpus_allowed mask will
1161          * not be changed, nor will its assigned CPU.
1162          *
1163          * The cpus_allowed mask of the parent may have changed after it was
1164          * copied first time - so re-copy it here, then check the child's CPU
1165          * to ensure it is on a valid CPU (and if not, just force it back to
1166          * parent's CPU). This avoids alot of nasty races.
1167          */
1168         p->cpus_allowed = current->cpus_allowed;
1169         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1170         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1171                         !cpu_online(task_cpu(p))))
1172                 set_task_cpu(p, smp_processor_id());
1173
1174         /* CLONE_PARENT re-uses the old parent */
1175         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1176                 p->real_parent = current->real_parent;
1177         else
1178                 p->real_parent = current;
1179
1180         spin_lock(&current->sighand->siglock);
1181
1182         /*
1183          * Process group and session signals need to be delivered to just the
1184          * parent before the fork or both the parent and the child after the
1185          * fork. Restart if a signal comes in before we add the new process to
1186          * it's process group.
1187          * A fatal signal pending means that current will exit, so the new
1188          * thread can't slip out of an OOM kill (or normal SIGKILL).
1189          */
1190         recalc_sigpending();
1191         if (signal_pending(current)) {
1192                 spin_unlock(&current->sighand->siglock);
1193                 write_unlock_irq(&tasklist_lock);
1194                 retval = -ERESTARTNOINTR;
1195                 goto bad_fork_free_pid;
1196         }
1197
1198         if (clone_flags & CLONE_THREAD) {
1199                 p->group_leader = current->group_leader;
1200                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1201
1202                 if (!cputime_eq(current->signal->it_virt_expires,
1203                                 cputime_zero) ||
1204                     !cputime_eq(current->signal->it_prof_expires,
1205                                 cputime_zero) ||
1206                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1207                     !list_empty(&current->signal->cpu_timers[0]) ||
1208                     !list_empty(&current->signal->cpu_timers[1]) ||
1209                     !list_empty(&current->signal->cpu_timers[2])) {
1210                         /*
1211                          * Have child wake up on its first tick to check
1212                          * for process CPU timers.
1213                          */
1214                         p->it_prof_expires = jiffies_to_cputime(1);
1215                 }
1216         }
1217
1218         if (likely(p->pid)) {
1219                 list_add_tail(&p->sibling, &p->real_parent->children);
1220                 tracehook_finish_clone(p, clone_flags, trace);
1221
1222                 if (thread_group_leader(p)) {
1223                         if (clone_flags & CLONE_NEWPID)
1224                                 p->nsproxy->pid_ns->child_reaper = p;
1225
1226                         p->signal->leader_pid = pid;
1227                         p->signal->tty = current->signal->tty;
1228                         set_task_pgrp(p, task_pgrp_nr(current));
1229                         set_task_session(p, task_session_nr(current));
1230                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1231                         attach_pid(p, PIDTYPE_SID, task_session(current));
1232                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1233                         __get_cpu_var(process_counts)++;
1234                 }
1235                 attach_pid(p, PIDTYPE_PID, pid);
1236                 nr_threads++;
1237         }
1238
1239         total_forks++;
1240         spin_unlock(&current->sighand->siglock);
1241         write_unlock_irq(&tasklist_lock);
1242         proc_fork_connector(p);
1243         cgroup_post_fork(p);
1244         return p;
1245
1246 bad_fork_free_pid:
1247         if (pid != &init_struct_pid)
1248                 free_pid(pid);
1249 bad_fork_cleanup_io:
1250         put_io_context(p->io_context);
1251 bad_fork_cleanup_namespaces:
1252         exit_task_namespaces(p);
1253 bad_fork_cleanup_keys:
1254         exit_keys(p);
1255 bad_fork_cleanup_mm:
1256         if (p->mm)
1257                 mmput(p->mm);
1258 bad_fork_cleanup_signal:
1259         cleanup_signal(p);
1260 bad_fork_cleanup_sighand:
1261         __cleanup_sighand(p->sighand);
1262 bad_fork_cleanup_fs:
1263         exit_fs(p); /* blocking */
1264 bad_fork_cleanup_files:
1265         exit_files(p); /* blocking */
1266 bad_fork_cleanup_semundo:
1267         exit_sem(p);
1268 bad_fork_cleanup_audit:
1269         audit_free(p);
1270 bad_fork_cleanup_security:
1271         security_task_free(p);
1272 bad_fork_cleanup_policy:
1273 #ifdef CONFIG_NUMA
1274         mpol_put(p->mempolicy);
1275 bad_fork_cleanup_cgroup:
1276 #endif
1277         cgroup_exit(p, cgroup_callbacks_done);
1278         delayacct_tsk_free(p);
1279         if (p->binfmt)
1280                 module_put(p->binfmt->module);
1281 bad_fork_cleanup_put_domain:
1282         module_put(task_thread_info(p)->exec_domain->module);
1283 bad_fork_cleanup_count:
1284         put_group_info(p->group_info);
1285         atomic_dec(&p->user->processes);
1286         free_uid(p->user);
1287 bad_fork_free:
1288         free_task(p);
1289 fork_out:
1290         return ERR_PTR(retval);
1291 }
1292
1293 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1294 {
1295         memset(regs, 0, sizeof(struct pt_regs));
1296         return regs;
1297 }
1298
1299 struct task_struct * __cpuinit fork_idle(int cpu)
1300 {
1301         struct task_struct *task;
1302         struct pt_regs regs;
1303
1304         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1305                             &init_struct_pid, 0);
1306         if (!IS_ERR(task))
1307                 init_idle(task, cpu);
1308
1309         return task;
1310 }
1311
1312 /*
1313  *  Ok, this is the main fork-routine.
1314  *
1315  * It copies the process, and if successful kick-starts
1316  * it and waits for it to finish using the VM if required.
1317  */
1318 long do_fork(unsigned long clone_flags,
1319               unsigned long stack_start,
1320               struct pt_regs *regs,
1321               unsigned long stack_size,
1322               int __user *parent_tidptr,
1323               int __user *child_tidptr)
1324 {
1325         struct task_struct *p;
1326         int trace = 0;
1327         long nr;
1328
1329         /*
1330          * We hope to recycle these flags after 2.6.26
1331          */
1332         if (unlikely(clone_flags & CLONE_STOPPED)) {
1333                 static int __read_mostly count = 100;
1334
1335                 if (count > 0 && printk_ratelimit()) {
1336                         char comm[TASK_COMM_LEN];
1337
1338                         count--;
1339                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1340                                         "clone flags 0x%lx\n",
1341                                 get_task_comm(comm, current),
1342                                 clone_flags & CLONE_STOPPED);
1343                 }
1344         }
1345
1346         /*
1347          * When called from kernel_thread, don't do user tracing stuff.
1348          */
1349         if (likely(user_mode(regs)))
1350                 trace = tracehook_prepare_clone(clone_flags);
1351
1352         p = copy_process(clone_flags, stack_start, regs, stack_size,
1353                          child_tidptr, NULL, trace);
1354         /*
1355          * Do this prior waking up the new thread - the thread pointer
1356          * might get invalid after that point, if the thread exits quickly.
1357          */
1358         if (!IS_ERR(p)) {
1359                 struct completion vfork;
1360
1361                 nr = task_pid_vnr(p);
1362
1363                 if (clone_flags & CLONE_PARENT_SETTID)
1364                         put_user(nr, parent_tidptr);
1365
1366                 if (clone_flags & CLONE_VFORK) {
1367                         p->vfork_done = &vfork;
1368                         init_completion(&vfork);
1369                 }
1370
1371                 tracehook_report_clone(trace, regs, clone_flags, nr, p);
1372
1373                 /*
1374                  * We set PF_STARTING at creation in case tracing wants to
1375                  * use this to distinguish a fully live task from one that
1376                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1377                  * clear it and set the child going.
1378                  */
1379                 p->flags &= ~PF_STARTING;
1380
1381                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1382                         /*
1383                          * We'll start up with an immediate SIGSTOP.
1384                          */
1385                         sigaddset(&p->pending.signal, SIGSTOP);
1386                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1387                         __set_task_state(p, TASK_STOPPED);
1388                 } else {
1389                         wake_up_new_task(p, clone_flags);
1390                 }
1391
1392                 tracehook_report_clone_complete(trace, regs,
1393                                                 clone_flags, nr, p);
1394
1395                 if (clone_flags & CLONE_VFORK) {
1396                         freezer_do_not_count();
1397                         wait_for_completion(&vfork);
1398                         freezer_count();
1399                         tracehook_report_vfork_done(p, nr);
1400                 }
1401         } else {
1402                 nr = PTR_ERR(p);
1403         }
1404         return nr;
1405 }
1406
1407 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1408 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1409 #endif
1410
1411 static void sighand_ctor(void *data)
1412 {
1413         struct sighand_struct *sighand = data;
1414
1415         spin_lock_init(&sighand->siglock);
1416         init_waitqueue_head(&sighand->signalfd_wqh);
1417 }
1418
1419 void __init proc_caches_init(void)
1420 {
1421         sighand_cachep = kmem_cache_create("sighand_cache",
1422                         sizeof(struct sighand_struct), 0,
1423                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1424                         sighand_ctor);
1425         signal_cachep = kmem_cache_create("signal_cache",
1426                         sizeof(struct signal_struct), 0,
1427                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1428         files_cachep = kmem_cache_create("files_cache",
1429                         sizeof(struct files_struct), 0,
1430                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1431         fs_cachep = kmem_cache_create("fs_cache",
1432                         sizeof(struct fs_struct), 0,
1433                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1434         vm_area_cachep = kmem_cache_create("vm_area_struct",
1435                         sizeof(struct vm_area_struct), 0,
1436                         SLAB_PANIC, NULL);
1437         mm_cachep = kmem_cache_create("mm_struct",
1438                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1439                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1440 }
1441
1442 /*
1443  * Check constraints on flags passed to the unshare system call and
1444  * force unsharing of additional process context as appropriate.
1445  */
1446 static void check_unshare_flags(unsigned long *flags_ptr)
1447 {
1448         /*
1449          * If unsharing a thread from a thread group, must also
1450          * unshare vm.
1451          */
1452         if (*flags_ptr & CLONE_THREAD)
1453                 *flags_ptr |= CLONE_VM;
1454
1455         /*
1456          * If unsharing vm, must also unshare signal handlers.
1457          */
1458         if (*flags_ptr & CLONE_VM)
1459                 *flags_ptr |= CLONE_SIGHAND;
1460
1461         /*
1462          * If unsharing signal handlers and the task was created
1463          * using CLONE_THREAD, then must unshare the thread
1464          */
1465         if ((*flags_ptr & CLONE_SIGHAND) &&
1466             (atomic_read(&current->signal->count) > 1))
1467                 *flags_ptr |= CLONE_THREAD;
1468
1469         /*
1470          * If unsharing namespace, must also unshare filesystem information.
1471          */
1472         if (*flags_ptr & CLONE_NEWNS)
1473                 *flags_ptr |= CLONE_FS;
1474 }
1475
1476 /*
1477  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1478  */
1479 static int unshare_thread(unsigned long unshare_flags)
1480 {
1481         if (unshare_flags & CLONE_THREAD)
1482                 return -EINVAL;
1483
1484         return 0;
1485 }
1486
1487 /*
1488  * Unshare the filesystem structure if it is being shared
1489  */
1490 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1491 {
1492         struct fs_struct *fs = current->fs;
1493
1494         if ((unshare_flags & CLONE_FS) &&
1495             (fs && atomic_read(&fs->count) > 1)) {
1496                 *new_fsp = __copy_fs_struct(current->fs);
1497                 if (!*new_fsp)
1498                         return -ENOMEM;
1499         }
1500
1501         return 0;
1502 }
1503
1504 /*
1505  * Unsharing of sighand is not supported yet
1506  */
1507 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1508 {
1509         struct sighand_struct *sigh = current->sighand;
1510
1511         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1512                 return -EINVAL;
1513         else
1514                 return 0;
1515 }
1516
1517 /*
1518  * Unshare vm if it is being shared
1519  */
1520 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1521 {
1522         struct mm_struct *mm = current->mm;
1523
1524         if ((unshare_flags & CLONE_VM) &&
1525             (mm && atomic_read(&mm->mm_users) > 1)) {
1526                 return -EINVAL;
1527         }
1528
1529         return 0;
1530 }
1531
1532 /*
1533  * Unshare file descriptor table if it is being shared
1534  */
1535 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1536 {
1537         struct files_struct *fd = current->files;
1538         int error = 0;
1539
1540         if ((unshare_flags & CLONE_FILES) &&
1541             (fd && atomic_read(&fd->count) > 1)) {
1542                 *new_fdp = dup_fd(fd, &error);
1543                 if (!*new_fdp)
1544                         return error;
1545         }
1546
1547         return 0;
1548 }
1549
1550 /*
1551  * unshare allows a process to 'unshare' part of the process
1552  * context which was originally shared using clone.  copy_*
1553  * functions used by do_fork() cannot be used here directly
1554  * because they modify an inactive task_struct that is being
1555  * constructed. Here we are modifying the current, active,
1556  * task_struct.
1557  */
1558 asmlinkage long sys_unshare(unsigned long unshare_flags)
1559 {
1560         int err = 0;
1561         struct fs_struct *fs, *new_fs = NULL;
1562         struct sighand_struct *new_sigh = NULL;
1563         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1564         struct files_struct *fd, *new_fd = NULL;
1565         struct nsproxy *new_nsproxy = NULL;
1566         int do_sysvsem = 0;
1567
1568         check_unshare_flags(&unshare_flags);
1569
1570         /* Return -EINVAL for all unsupported flags */
1571         err = -EINVAL;
1572         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1573                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1574                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
1575                                 CLONE_NEWNET))
1576                 goto bad_unshare_out;
1577
1578         /*
1579          * CLONE_NEWIPC must also detach from the undolist: after switching
1580          * to a new ipc namespace, the semaphore arrays from the old
1581          * namespace are unreachable.
1582          */
1583         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1584                 do_sysvsem = 1;
1585         if ((err = unshare_thread(unshare_flags)))
1586                 goto bad_unshare_out;
1587         if ((err = unshare_fs(unshare_flags, &new_fs)))
1588                 goto bad_unshare_cleanup_thread;
1589         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1590                 goto bad_unshare_cleanup_fs;
1591         if ((err = unshare_vm(unshare_flags, &new_mm)))
1592                 goto bad_unshare_cleanup_sigh;
1593         if ((err = unshare_fd(unshare_flags, &new_fd)))
1594                 goto bad_unshare_cleanup_vm;
1595         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1596                         new_fs)))
1597                 goto bad_unshare_cleanup_fd;
1598
1599         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1600                 if (do_sysvsem) {
1601                         /*
1602                          * CLONE_SYSVSEM is equivalent to sys_exit().
1603                          */
1604                         exit_sem(current);
1605                 }
1606
1607                 if (new_nsproxy) {
1608                         switch_task_namespaces(current, new_nsproxy);
1609                         new_nsproxy = NULL;
1610                 }
1611
1612                 task_lock(current);
1613
1614                 if (new_fs) {
1615                         fs = current->fs;
1616                         current->fs = new_fs;
1617                         new_fs = fs;
1618                 }
1619
1620                 if (new_mm) {
1621                         mm = current->mm;
1622                         active_mm = current->active_mm;
1623                         current->mm = new_mm;
1624                         current->active_mm = new_mm;
1625                         activate_mm(active_mm, new_mm);
1626                         new_mm = mm;
1627                 }
1628
1629                 if (new_fd) {
1630                         fd = current->files;
1631                         current->files = new_fd;
1632                         new_fd = fd;
1633                 }
1634
1635                 task_unlock(current);
1636         }
1637
1638         if (new_nsproxy)
1639                 put_nsproxy(new_nsproxy);
1640
1641 bad_unshare_cleanup_fd:
1642         if (new_fd)
1643                 put_files_struct(new_fd);
1644
1645 bad_unshare_cleanup_vm:
1646         if (new_mm)
1647                 mmput(new_mm);
1648
1649 bad_unshare_cleanup_sigh:
1650         if (new_sigh)
1651                 if (atomic_dec_and_test(&new_sigh->count))
1652                         kmem_cache_free(sighand_cachep, new_sigh);
1653
1654 bad_unshare_cleanup_fs:
1655         if (new_fs)
1656                 put_fs_struct(new_fs);
1657
1658 bad_unshare_cleanup_thread:
1659 bad_unshare_out:
1660         return err;
1661 }
1662
1663 /*
1664  *      Helper to unshare the files of the current task.
1665  *      We don't want to expose copy_files internals to
1666  *      the exec layer of the kernel.
1667  */
1668
1669 int unshare_files(struct files_struct **displaced)
1670 {
1671         struct task_struct *task = current;
1672         struct files_struct *copy = NULL;
1673         int error;
1674
1675         error = unshare_fd(CLONE_FILES, &copy);
1676         if (error || !copy) {
1677                 *displaced = NULL;
1678                 return error;
1679         }
1680         *displaced = task->files;
1681         task_lock(task);
1682         task->files = copy;
1683         task_unlock(task);
1684         return 0;
1685 }