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/futex.h>
  41 #include <linux/task_io_accounting_ops.h>
  42 #include <linux/rcupdate.h>
  43 #include <linux/ptrace.h>
  44 #include <linux/mount.h>
  45 #include <linux/audit.h>
  46 #include <linux/memcontrol.h>
  47 #include <linux/profile.h>
  48 #include <linux/rmap.h>
  49 #include <linux/acct.h>
  50 #include <linux/tsacct_kern.h>
  51 #include <linux/cn_proc.h>
  52 #include <linux/freezer.h>
  53 #include <linux/delayacct.h>
  54 #include <linux/taskstats_kern.h>
  55 #include <linux/random.h>
  56 #include <linux/tty.h>
  57 #include <linux/proc_fs.h>
  58 #include <linux/blkdev.h>
  59
  60 #include <asm/pgtable.h>
  61 #include <asm/pgalloc.h>
  62 #include <asm/uaccess.h>
  63 #include <asm/mmu_context.h>
  64 #include <asm/cacheflush.h>
  65 #include <asm/tlbflush.h>
  66
  67 /*
  68  * Protected counters by write_lock_irq(&tasklist_lock)
  69  */
  70 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  71 int nr_threads;                 /* The idle threads do not count.. */
  72
  73 int max_threads;                /* tunable limit on nr_threads */
  74
  75 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  76
  77 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  78
  79 int nr_processes(void)
  80 {
  81         int cpu;
  82         int total = 0;
  83
  84         for_each_online_cpu(cpu)
  85                 total += per_cpu(process_counts, cpu);
  86
  87         return total;
  88 }
  89
  90 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  91 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  92 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  93 static struct kmem_cache *task_struct_cachep;
  94 #endif
  95
  96 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
  97 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
  98 {
  99 #ifdef CONFIG_DEBUG_STACK_USAGE
 100         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 101 #else
 102         gfp_t mask = GFP_KERNEL;
 103 #endif
 104         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
 105 }
 106
 107 static inline void free_thread_info(struct thread_info *ti)
 108 {
 109         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 110 }
 111 #endif
 112
 113 /* SLAB cache for signal_struct structures (tsk->signal) */
 114 static struct kmem_cache *signal_cachep;
 115
 116 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 117 struct kmem_cache *sighand_cachep;
 118
 119 /* SLAB cache for files_struct structures (tsk->files) */
 120 struct kmem_cache *files_cachep;
 121
 122 /* SLAB cache for fs_struct structures (tsk->fs) */
 123 struct kmem_cache *fs_cachep;
 124
 125 /* SLAB cache for vm_area_struct structures */
 126 struct kmem_cache *vm_area_cachep;
 127
 128 /* SLAB cache for mm_struct structures (tsk->mm) */
 129 static struct kmem_cache *mm_cachep;
 130
 131 void free_task(struct task_struct *tsk)
 132 {
 133         prop_local_destroy_single(&tsk->dirties);
 134         free_thread_info(tsk->stack);
 135         rt_mutex_debug_task_free(tsk);
 136         free_task_struct(tsk);
 137 }
 138 EXPORT_SYMBOL(free_task);
 139
 140 void __put_task_struct(struct task_struct *tsk)
 141 {
 142         WARN_ON(!tsk->exit_state);
 143         WARN_ON(atomic_read(&tsk->usage));
 144         WARN_ON(tsk == current);
 145
 146         security_task_free(tsk);
 147         free_uid(tsk->user);
 148         put_group_info(tsk->group_info);
 149         delayacct_tsk_free(tsk);
 150
 151         if (!profile_handoff_task(tsk))
 152                 free_task(tsk);
 153 }
 154
 155 /*
 156  * macro override instead of weak attribute alias, to workaround
 157  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 158  */
 159 #ifndef arch_task_cache_init
 160 #define arch_task_cache_init()
 161 #endif
 162
 163 void __init fork_init(unsigned long mempages)
 164 {
 165 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 166 #ifndef ARCH_MIN_TASKALIGN
 167 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 168 #endif
 169         /* create a slab on which task_structs can be allocated */
 170         task_struct_cachep =
 171                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 172                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
 173 #endif
 174
 175         /* do the arch specific task caches init */
 176         arch_task_cache_init();
 177
 178         /*
 179          * The default maximum number of threads is set to a safe
 180          * value: the thread structures can take up at most half
 181          * of memory.
 182          */
 183         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 184
 185         /*
 186          * we need to allow at least 20 threads to boot a system
 187          */
 188         if(max_threads < 20)
 189                 max_threads = 20;
 190
 191         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 192         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 193         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 194                 init_task.signal->rlim[RLIMIT_NPROC];
 195 }
 196
 197 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 198                                                struct task_struct *src)
 199 {
 200         *dst = *src;
 201         return 0;
 202 }
 203
 204 static struct task_struct *dup_task_struct(struct task_struct *orig)
 205 {
 206         struct task_struct *tsk;
 207         struct thread_info *ti;
 208         int err;
 209
 210         prepare_to_copy(orig);
 211
 212         tsk = alloc_task_struct();
 213         if (!tsk)
 214                 return NULL;
 215
 216         ti = alloc_thread_info(tsk);
 217         if (!ti) {
 218                 free_task_struct(tsk);
 219                 return NULL;
 220         }
 221
 222         err = arch_dup_task_struct(tsk, orig);
 223         if (err)
 224                 goto out;
 225
 226         tsk->stack = ti;
 227
 228         err = prop_local_init_single(&tsk->dirties);
 229         if (err)
 230                 goto out;
 231
 232         setup_thread_stack(tsk, orig);
 233
 234 #ifdef CONFIG_CC_STACKPROTECTOR
 235         tsk->stack_canary = get_random_int();
 236 #endif
 237
 238         /* One for us, one for whoever does the "release_task()" (usually parent) */
 239         atomic_set(&tsk->usage,2);
 240         atomic_set(&tsk->fs_excl, 0);
 241 #ifdef CONFIG_BLK_DEV_IO_TRACE
 242         tsk->btrace_seq = 0;
 243 #endif
 244         tsk->splice_pipe = NULL;
 245         return tsk;
 246
 247 out:
 248         free_thread_info(ti);
 249         free_task_struct(tsk);
 250         return NULL;
 251 }
 252
 253 #ifdef CONFIG_MMU
 254 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 255 {
 256         struct vm_area_struct *mpnt, *tmp, **pprev;
 257         struct rb_node **rb_link, *rb_parent;
 258         int retval;
 259         unsigned long charge;
 260         struct mempolicy *pol;
 261
 262         down_write(&oldmm->mmap_sem);
 263         flush_cache_dup_mm(oldmm);
 264         /*
 265          * Not linked in yet - no deadlock potential:
 266          */
 267         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 268
 269         mm->locked_vm = 0;
 270         mm->mmap = NULL;
 271         mm->mmap_cache = NULL;
 272         mm->free_area_cache = oldmm->mmap_base;
 273         mm->cached_hole_size = ~0UL;
 274         mm->map_count = 0;
 275         cpus_clear(mm->cpu_vm_mask);
 276         mm->mm_rb = RB_ROOT;
 277         rb_link = &mm->mm_rb.rb_node;
 278         rb_parent = NULL;
 279         pprev = &mm->mmap;
 280
 281         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 282                 struct file *file;
 283
 284                 if (mpnt->vm_flags & VM_DONTCOPY) {
 285                         long pages = vma_pages(mpnt);
 286                         mm->total_vm -= pages;
 287                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 288                                                                 -pages);
 289                         continue;
 290                 }
 291                 charge = 0;
 292                 if (mpnt->vm_flags & VM_ACCOUNT) {
 293                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 294                         if (security_vm_enough_memory(len))
 295                                 goto fail_nomem;
 296                         charge = len;
 297                 }
 298                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 299                 if (!tmp)
 300                         goto fail_nomem;
 301                 *tmp = *mpnt;
 302                 pol = mpol_dup(vma_policy(mpnt));
 303                 retval = PTR_ERR(pol);
 304                 if (IS_ERR(pol))
 305                         goto fail_nomem_policy;
 306                 vma_set_policy(tmp, pol);
 307                 tmp->vm_flags &= ~VM_LOCKED;
 308                 tmp->vm_mm = mm;
 309                 tmp->vm_next = NULL;
 310                 anon_vma_link(tmp);
 311                 file = tmp->vm_file;
 312                 if (file) {
 313                         struct inode *inode = file->f_path.dentry->d_inode;
 314                         get_file(file);
 315                         if (tmp->vm_flags & VM_DENYWRITE)
 316                                 atomic_dec(&inode->i_writecount);
 317
 318                         /* insert tmp into the share list, just after mpnt */
 319                         spin_lock(&file->f_mapping->i_mmap_lock);
 320                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 321                         flush_dcache_mmap_lock(file->f_mapping);
 322                         vma_prio_tree_add(tmp, mpnt);
 323                         flush_dcache_mmap_unlock(file->f_mapping);
 324                         spin_unlock(&file->f_mapping->i_mmap_lock);
 325                 }
 326
 327                 /*
 328                  * Clear hugetlb-related page reserves for children. This only
 329                  * affects MAP_PRIVATE mappings. Faults generated by the child
 330                  * are not guaranteed to succeed, even if read-only
 331                  */
 332                 if (is_vm_hugetlb_page(tmp))
 333                         reset_vma_resv_huge_pages(tmp);
 334
 335                 /*
 336                  * Link in the new vma and copy the page table entries.
 337                  */
 338                 *pprev = tmp;
 339                 pprev = &tmp->vm_next;
 340
 341                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 342                 rb_link = &tmp->vm_rb.rb_right;
 343                 rb_parent = &tmp->vm_rb;
 344
 345                 mm->map_count++;
 346                 retval = copy_page_range(mm, oldmm, mpnt);
 347
 348                 if (tmp->vm_ops && tmp->vm_ops->open)
 349                         tmp->vm_ops->open(tmp);
 350
 351                 if (retval)
 352                         goto out;
 353         }
 354         /* a new mm has just been created */
 355         arch_dup_mmap(oldmm, mm);
 356         retval = 0;
 357 out:
 358         up_write(&mm->mmap_sem);
 359         flush_tlb_mm(oldmm);
 360         up_write(&oldmm->mmap_sem);
 361         return retval;
 362 fail_nomem_policy:
 363         kmem_cache_free(vm_area_cachep, tmp);
 364 fail_nomem:
 365         retval = -ENOMEM;
 366         vm_unacct_memory(charge);
 367         goto out;
 368 }
 369
 370 static inline int mm_alloc_pgd(struct mm_struct * mm)
 371 {
 372         mm->pgd = pgd_alloc(mm);
 373         if (unlikely(!mm->pgd))
 374                 return -ENOMEM;
 375         return 0;
 376 }
 377
 378 static inline void mm_free_pgd(struct mm_struct * mm)
 379 {
 380         pgd_free(mm, mm->pgd);
 381 }
 382 #else
 383 #define dup_mmap(mm, oldmm)     (0)
 384 #define mm_alloc_pgd(mm)        (0)
 385 #define mm_free_pgd(mm)
 386 #endif /* CONFIG_MMU */
 387
 388 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 389
 390 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 391 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 392
 393 #include <linux/init_task.h>
 394
 395 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
 396 {
 397         atomic_set(&mm->mm_users, 1);
 398         atomic_set(&mm->mm_count, 1);
 399         init_rwsem(&mm->mmap_sem);
 400         INIT_LIST_HEAD(&mm->mmlist);
 401         mm->flags = (current->mm) ? current->mm->flags
 402                                   : MMF_DUMP_FILTER_DEFAULT;
 403         mm->core_waiters = 0;
 404         mm->nr_ptes = 0;
 405         set_mm_counter(mm, file_rss, 0);
 406         set_mm_counter(mm, anon_rss, 0);
 407         spin_lock_init(&mm->page_table_lock);
 408         rwlock_init(&mm->ioctx_list_lock);
 409         mm->ioctx_list = NULL;
 410         mm->free_area_cache = TASK_UNMAPPED_BASE;
 411         mm->cached_hole_size = ~0UL;
 412         mm_init_owner(mm, p);
 413
 414         if (likely(!mm_alloc_pgd(mm))) {
 415                 mm->def_flags = 0;
 416                 return mm;
 417         }
 418
 419         free_mm(mm);
 420         return NULL;
 421 }
 422
 423 /*
 424  * Allocate and initialize an mm_struct.
 425  */
 426 struct mm_struct * mm_alloc(void)
 427 {
 428         struct mm_struct * mm;
 429
 430         mm = allocate_mm();
 431         if (mm) {
 432                 memset(mm, 0, sizeof(*mm));
 433                 mm = mm_init(mm, current);
 434         }
 435         return mm;
 436 }
 437
 438 /*
 439  * Called when the last reference to the mm
 440  * is dropped: either by a lazy thread or by
 441  * mmput. Free the page directory and the mm.
 442  */
 443 void __mmdrop(struct mm_struct *mm)
 444 {
 445         BUG_ON(mm == &init_mm);
 446         mm_free_pgd(mm);
 447         destroy_context(mm);
 448         free_mm(mm);
 449 }
 450 EXPORT_SYMBOL_GPL(__mmdrop);
 451
 452 /*
 453  * Decrement the use count and release all resources for an mm.
 454  */
 455 void mmput(struct mm_struct *mm)
 456 {
 457         might_sleep();
 458
 459         if (atomic_dec_and_test(&mm->mm_users)) {
 460                 exit_aio(mm);
 461                 exit_mmap(mm);
 462                 set_mm_exe_file(mm, NULL);
 463                 if (!list_empty(&mm->mmlist)) {
 464                         spin_lock(&mmlist_lock);
 465                         list_del(&mm->mmlist);
 466                         spin_unlock(&mmlist_lock);
 467                 }
 468                 put_swap_token(mm);
 469                 mmdrop(mm);
 470         }
 471 }
 472 EXPORT_SYMBOL_GPL(mmput);
 473
 474 /**
 475  * get_task_mm - acquire a reference to the task's mm
 476  *
 477  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
 478  * this kernel workthread has transiently adopted a user mm with use_mm,
 479  * to do its AIO) is not set and if so returns a reference to it, after
 480  * bumping up the use count.  User must release the mm via mmput()
 481  * after use.  Typically used by /proc and ptrace.
 482  */
 483 struct mm_struct *get_task_mm(struct task_struct *task)
 484 {
 485         struct mm_struct *mm;
 486
 487         task_lock(task);
 488         mm = task->mm;
 489         if (mm) {
 490                 if (task->flags & PF_BORROWED_MM)
 491                         mm = NULL;
 492                 else
 493                         atomic_inc(&mm->mm_users);
 494         }
 495         task_unlock(task);
 496         return mm;
 497 }
 498 EXPORT_SYMBOL_GPL(get_task_mm);
 499
 500 /* Please note the differences between mmput and mm_release.
 501  * mmput is called whenever we stop holding onto a mm_struct,
 502  * error success whatever.
 503  *
 504  * mm_release is called after a mm_struct has been removed
 505  * from the current process.
 506  *
 507  * This difference is important for error handling, when we
 508  * only half set up a mm_struct for a new process and need to restore
 509  * the old one.  Because we mmput the new mm_struct before
 510  * restoring the old one. . .
 511  * Eric Biederman 10 January 1998
 512  */
 513 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 514 {
 515         struct completion *vfork_done = tsk->vfork_done;
 516
 517         /* Get rid of any cached register state */
 518         deactivate_mm(tsk, mm);
 519
 520         /* notify parent sleeping on vfork() */
 521         if (vfork_done) {
 522                 tsk->vfork_done = NULL;
 523                 complete(vfork_done);
 524         }
 525
 526         /*
 527          * If we're exiting normally, clear a user-space tid field if
 528          * requested.  We leave this alone when dying by signal, to leave
 529          * the value intact in a core dump, and to save the unnecessary
 530          * trouble otherwise.  Userland only wants this done for a sys_exit.
 531          */
 532         if (tsk->clear_child_tid
 533             && !(tsk->flags & PF_SIGNALED)
 534             && atomic_read(&mm->mm_users) > 1) {
 535                 u32 __user * tidptr = tsk->clear_child_tid;
 536                 tsk->clear_child_tid = NULL;
 537
 538                 /*
 539                  * We don't check the error code - if userspace has
 540                  * not set up a proper pointer then tough luck.
 541                  */
 542                 put_user(0, tidptr);
 543                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 544         }
 545 }
 546
 547 /*
 548  * Allocate a new mm structure and copy contents from the
 549  * mm structure of the passed in task structure.
 550  */
 551 struct mm_struct *dup_mm(struct task_struct *tsk)
 552 {
 553         struct mm_struct *mm, *oldmm = current->mm;
 554         int err;
 555
 556         if (!oldmm)
 557                 return NULL;
 558
 559         mm = allocate_mm();
 560         if (!mm)
 561                 goto fail_nomem;
 562
 563         memcpy(mm, oldmm, sizeof(*mm));
 564
 565         /* Initializing for Swap token stuff */
 566         mm->token_priority = 0;
 567         mm->last_interval = 0;
 568
 569         if (!mm_init(mm, tsk))
 570                 goto fail_nomem;
 571
 572         if (init_new_context(tsk, mm))
 573                 goto fail_nocontext;
 574
 575         dup_mm_exe_file(oldmm, mm);
 576
 577         err = dup_mmap(mm, oldmm);
 578         if (err)
 579                 goto free_pt;
 580
 581         mm->hiwater_rss = get_mm_rss(mm);
 582         mm->hiwater_vm = mm->total_vm;
 583
 584         return mm;
 585
 586 free_pt:
 587         mmput(mm);
 588
 589 fail_nomem:
 590         return NULL;
 591
 592 fail_nocontext:
 593         /*
 594          * If init_new_context() failed, we cannot use mmput() to free the mm
 595          * because it calls destroy_context()
 596          */
 597         mm_free_pgd(mm);
 598         free_mm(mm);
 599         return NULL;
 600 }
 601
 602 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 603 {
 604         struct mm_struct * mm, *oldmm;
 605         int retval;
 606
 607         tsk->min_flt = tsk->maj_flt = 0;
 608         tsk->nvcsw = tsk->nivcsw = 0;
 609
 610         tsk->mm = NULL;
 611         tsk->active_mm = NULL;
 612
 613         /*
 614          * Are we cloning a kernel thread?
 615          *
 616          * We need to steal a active VM for that..
 617          */
 618         oldmm = current->mm;
 619         if (!oldmm)
 620                 return 0;
 621
 622         if (clone_flags & CLONE_VM) {
 623                 atomic_inc(&oldmm->mm_users);
 624                 mm = oldmm;
 625                 goto good_mm;
 626         }
 627
 628         retval = -ENOMEM;
 629         mm = dup_mm(tsk);
 630         if (!mm)
 631                 goto fail_nomem;
 632
 633 good_mm:
 634         /* Initializing for Swap token stuff */
 635         mm->token_priority = 0;
 636         mm->last_interval = 0;
 637
 638         tsk->mm = mm;
 639         tsk->active_mm = mm;
 640         return 0;
 641
 642 fail_nomem:
 643         return retval;
 644 }
 645
 646 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 647 {
 648         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 649         /* We don't need to lock fs - think why ;-) */
 650         if (fs) {
 651                 atomic_set(&fs->count, 1);
 652                 rwlock_init(&fs->lock);
 653                 fs->umask = old->umask;
 654                 read_lock(&old->lock);
 655                 fs->root = old->root;
 656                 path_get(&old->root);
 657                 fs->pwd = old->pwd;
 658                 path_get(&old->pwd);
 659                 if (old->altroot.dentry) {
 660                         fs->altroot = old->altroot;
 661                         path_get(&old->altroot);
 662                 } else {
 663                         fs->altroot.mnt = NULL;
 664                         fs->altroot.dentry = NULL;
 665                 }
 666                 read_unlock(&old->lock);
 667         }
 668         return fs;
 669 }
 670
 671 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 672 {
 673         return __copy_fs_struct(old);
 674 }
 675
 676 EXPORT_SYMBOL_GPL(copy_fs_struct);
 677
 678 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 679 {
 680         if (clone_flags & CLONE_FS) {
 681                 atomic_inc(&current->fs->count);
 682                 return 0;
 683         }
 684         tsk->fs = __copy_fs_struct(current->fs);
 685         if (!tsk->fs)
 686                 return -ENOMEM;
 687         return 0;
 688 }
 689
 690 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 691 {
 692         struct files_struct *oldf, *newf;
 693         int error = 0;
 694
 695         /*
 696          * A background process may not have any files ...
 697          */
 698         oldf = current->files;
 699         if (!oldf)
 700                 goto out;
 701
 702         if (clone_flags & CLONE_FILES) {
 703                 atomic_inc(&oldf->count);
 704                 goto out;
 705         }
 706
 707         newf = dup_fd(oldf, &error);
 708         if (!newf)
 709                 goto out;
 710
 711         tsk->files = newf;
 712         error = 0;
 713 out:
 714         return error;
 715 }
 716
 717 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 718 {
 719 #ifdef CONFIG_BLOCK
 720         struct io_context *ioc = current->io_context;
 721
 722         if (!ioc)
 723                 return 0;
 724         /*
 725          * Share io context with parent, if CLONE_IO is set
 726          */
 727         if (clone_flags & CLONE_IO) {
 728                 tsk->io_context = ioc_task_link(ioc);
 729                 if (unlikely(!tsk->io_context))
 730                         return -ENOMEM;
 731         } else if (ioprio_valid(ioc->ioprio)) {
 732                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 733                 if (unlikely(!tsk->io_context))
 734                         return -ENOMEM;
 735
 736                 tsk->io_context->ioprio = ioc->ioprio;
 737         }
 738 #endif
 739         return 0;
 740 }
 741
 742 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 743 {
 744         struct sighand_struct *sig;
 745
 746         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 747                 atomic_inc(&current->sighand->count);
 748                 return 0;
 749         }
 750         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 751         rcu_assign_pointer(tsk->sighand, sig);
 752         if (!sig)
 753                 return -ENOMEM;
 754         atomic_set(&sig->count, 1);
 755         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 756         return 0;
 757 }
 758
 759 void __cleanup_sighand(struct sighand_struct *sighand)
 760 {
 761         if (atomic_dec_and_test(&sighand->count))
 762                 kmem_cache_free(sighand_cachep, sighand);
 763 }
 764
 765 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 766 {
 767         struct signal_struct *sig;
 768         int ret;
 769
 770         if (clone_flags & CLONE_THREAD) {
 771                 atomic_inc(&current->signal->count);
 772                 atomic_inc(&current->signal->live);
 773                 return 0;
 774         }
 775         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 776         tsk->signal = sig;
 777         if (!sig)
 778                 return -ENOMEM;
 779
 780         ret = copy_thread_group_keys(tsk);
 781         if (ret < 0) {
 782                 kmem_cache_free(signal_cachep, sig);
 783                 return ret;
 784         }
 785
 786         atomic_set(&sig->count, 1);
 787         atomic_set(&sig->live, 1);
 788         init_waitqueue_head(&sig->wait_chldexit);
 789         sig->flags = 0;
 790         sig->group_exit_code = 0;
 791         sig->group_exit_task = NULL;
 792         sig->group_stop_count = 0;
 793         sig->curr_target = tsk;
 794         init_sigpending(&sig->shared_pending);
 795         INIT_LIST_HEAD(&sig->posix_timers);
 796
 797         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 798         sig->it_real_incr.tv64 = 0;
 799         sig->real_timer.function = it_real_fn;
 800
 801         sig->it_virt_expires = cputime_zero;
 802         sig->it_virt_incr = cputime_zero;
 803         sig->it_prof_expires = cputime_zero;
 804         sig->it_prof_incr = cputime_zero;
 805
 806         sig->leader = 0;        /* session leadership doesn't inherit */
 807         sig->tty_old_pgrp = NULL;
 808
 809         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 810         sig->gtime = cputime_zero;
 811         sig->cgtime = cputime_zero;
 812         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 813         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 814         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
 815         sig->sum_sched_runtime = 0;
 816         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 817         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 818         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 819         taskstats_tgid_init(sig);
 820
 821         task_lock(current->group_leader);
 822         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 823         task_unlock(current->group_leader);
 824
 825         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 826                 /*
 827                  * New sole thread in the process gets an expiry time
 828                  * of the whole CPU time limit.
 829                  */
 830                 tsk->it_prof_expires =
 831                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 832         }
 833         acct_init_pacct(&sig->pacct);
 834
 835         tty_audit_fork(sig);
 836
 837         return 0;
 838 }
 839
 840 void __cleanup_signal(struct signal_struct *sig)
 841 {
 842         exit_thread_group_keys(sig);
 843         kmem_cache_free(signal_cachep, sig);
 844 }
 845
 846 static void cleanup_signal(struct task_struct *tsk)
 847 {
 848         struct signal_struct *sig = tsk->signal;
 849
 850         atomic_dec(&sig->live);
 851
 852         if (atomic_dec_and_test(&sig->count))
 853                 __cleanup_signal(sig);
 854 }
 855
 856 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 857 {
 858         unsigned long new_flags = p->flags;
 859
 860         new_flags &= ~PF_SUPERPRIV;
 861         new_flags |= PF_FORKNOEXEC;
 862         if (!(clone_flags & CLONE_PTRACE))
 863                 p->ptrace = 0;
 864         p->flags = new_flags;
 865         clear_freeze_flag(p);
 866 }
 867
 868 asmlinkage long sys_set_tid_address(int __user *tidptr)
 869 {
 870         current->clear_child_tid = tidptr;
 871
 872         return task_pid_vnr(current);
 873 }
 874
 875 static void rt_mutex_init_task(struct task_struct *p)
 876 {
 877         spin_lock_init(&p->pi_lock);
 878 #ifdef CONFIG_RT_MUTEXES
 879         plist_head_init(&p->pi_waiters, &p->pi_lock);
 880         p->pi_blocked_on = NULL;
 881 #endif
 882 }
 883
 884 #ifdef CONFIG_MM_OWNER
 885 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 886 {
 887         mm->owner = p;
 888 }
 889 #endif /* CONFIG_MM_OWNER */
 890
 891 /*
 892  * This creates a new process as a copy of the old one,
 893  * but does not actually start it yet.
 894  *
 895  * It copies the registers, and all the appropriate
 896  * parts of the process environment (as per the clone
 897  * flags). The actual kick-off is left to the caller.
 898  */
 899 static struct task_struct *copy_process(unsigned long clone_flags,
 900                                         unsigned long stack_start,
 901                                         struct pt_regs *regs,
 902                                         unsigned long stack_size,
 903                                         int __user *child_tidptr,
 904                                         struct pid *pid)
 905 {
 906         int retval;
 907         struct task_struct *p;
 908         int cgroup_callbacks_done = 0;
 909
 910         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 911                 return ERR_PTR(-EINVAL);
 912
 913         /*
 914          * Thread groups must share signals as well, and detached threads
 915          * can only be started up within the thread group.
 916          */
 917         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 918                 return ERR_PTR(-EINVAL);
 919
 920         /*
 921          * Shared signal handlers imply shared VM. By way of the above,
 922          * thread groups also imply shared VM. Blocking this case allows
 923          * for various simplifications in other code.
 924          */
 925         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 926                 return ERR_PTR(-EINVAL);
 927
 928         retval = security_task_create(clone_flags);
 929         if (retval)
 930                 goto fork_out;
 931
 932         retval = -ENOMEM;
 933         p = dup_task_struct(current);
 934         if (!p)
 935                 goto fork_out;
 936
 937         rt_mutex_init_task(p);
 938
 939 #ifdef CONFIG_PROVE_LOCKING
 940         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
 941         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
 942 #endif
 943         retval = -EAGAIN;
 944         if (atomic_read(&p->user->processes) >=
 945                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
 946                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 947                     p->user != current->nsproxy->user_ns->root_user)
 948                         goto bad_fork_free;
 949         }
 950
 951         atomic_inc(&p->user->__count);
 952         atomic_inc(&p->user->processes);
 953         get_group_info(p->group_info);
 954
 955         /*
 956          * If multiple threads are within copy_process(), then this check
 957          * triggers too late. This doesn't hurt, the check is only there
 958          * to stop root fork bombs.
 959          */
 960         if (nr_threads >= max_threads)
 961                 goto bad_fork_cleanup_count;
 962
 963         if (!try_module_get(task_thread_info(p)->exec_domain->module))
 964                 goto bad_fork_cleanup_count;
 965
 966         if (p->binfmt && !try_module_get(p->binfmt->module))
 967                 goto bad_fork_cleanup_put_domain;
 968
 969         p->did_exec = 0;
 970         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
 971         copy_flags(clone_flags, p);
 972         INIT_LIST_HEAD(&p->children);
 973         INIT_LIST_HEAD(&p->sibling);
 974 #ifdef CONFIG_PREEMPT_RCU
 975         p->rcu_read_lock_nesting = 0;
 976         p->rcu_flipctr_idx = 0;
 977 #endif /* #ifdef CONFIG_PREEMPT_RCU */
 978         p->vfork_done = NULL;
 979         spin_lock_init(&p->alloc_lock);
 980
 981         clear_tsk_thread_flag(p, TIF_SIGPENDING);
 982         init_sigpending(&p->pending);
 983
 984         p->utime = cputime_zero;
 985         p->stime = cputime_zero;
 986         p->gtime = cputime_zero;
 987         p->utimescaled = cputime_zero;
 988         p->stimescaled = cputime_zero;
 989         p->prev_utime = cputime_zero;
 990         p->prev_stime = cputime_zero;
 991
 992 #ifdef CONFIG_DETECT_SOFTLOCKUP
 993         p->last_switch_count = 0;
 994         p->last_switch_timestamp = 0;
 995 #endif
 996
 997 #ifdef CONFIG_TASK_XACCT
 998         p->rchar = 0;           /* I/O counter: bytes read */
 999         p->wchar = 0;           /* I/O counter: bytes written */
1000         p->syscr = 0;           /* I/O counter: read syscalls */
1001         p->syscw = 0;           /* I/O counter: write syscalls */
1002 #endif
1003         task_io_accounting_init(p);
1004         acct_clear_integrals(p);
1005
1006         p->it_virt_expires = cputime_zero;
1007         p->it_prof_expires = cputime_zero;
1008         p->it_sched_expires = 0;
1009         INIT_LIST_HEAD(&p->cpu_timers[0]);
1010         INIT_LIST_HEAD(&p->cpu_timers[1]);
1011         INIT_LIST_HEAD(&p->cpu_timers[2]);
1012
1013         p->lock_depth = -1;             /* -1 = no lock */
1014         do_posix_clock_monotonic_gettime(&p->start_time);
1015         p->real_start_time = p->start_time;
1016         monotonic_to_bootbased(&p->real_start_time);
1017 #ifdef CONFIG_SECURITY
1018         p->security = NULL;
1019 #endif
1020         p->cap_bset = current->cap_bset;
1021         p->io_context = NULL;
1022         p->audit_context = NULL;
1023         cgroup_fork(p);
1024 #ifdef CONFIG_NUMA
1025         p->mempolicy = mpol_dup(p->mempolicy);
1026         if (IS_ERR(p->mempolicy)) {
1027                 retval = PTR_ERR(p->mempolicy);
1028                 p->mempolicy = NULL;
1029                 goto bad_fork_cleanup_cgroup;
1030         }
1031         mpol_fix_fork_child_flag(p);
1032 #endif
1033 #ifdef CONFIG_TRACE_IRQFLAGS
1034         p->irq_events = 0;
1035 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1036         p->hardirqs_enabled = 1;
1037 #else
1038         p->hardirqs_enabled = 0;
1039 #endif
1040         p->hardirq_enable_ip = 0;
1041         p->hardirq_enable_event = 0;
1042         p->hardirq_disable_ip = _THIS_IP_;
1043         p->hardirq_disable_event = 0;
1044         p->softirqs_enabled = 1;
1045         p->softirq_enable_ip = _THIS_IP_;
1046         p->softirq_enable_event = 0;
1047         p->softirq_disable_ip = 0;
1048         p->softirq_disable_event = 0;
1049         p->hardirq_context = 0;
1050         p->softirq_context = 0;
1051 #endif
1052 #ifdef CONFIG_LOCKDEP
1053         p->lockdep_depth = 0; /* no locks held yet */
1054         p->curr_chain_key = 0;
1055         p->lockdep_recursion = 0;
1056 #endif
1057
1058 #ifdef CONFIG_DEBUG_MUTEXES
1059         p->blocked_on = NULL; /* not blocked yet */
1060 #endif
1061
1062         /* Perform scheduler related setup. Assign this task to a CPU. */
1063         sched_fork(p, clone_flags);
1064
1065         if ((retval = security_task_alloc(p)))
1066                 goto bad_fork_cleanup_policy;
1067         if ((retval = audit_alloc(p)))
1068                 goto bad_fork_cleanup_security;
1069         /* copy all the process information */
1070         if ((retval = copy_semundo(clone_flags, p)))
1071                 goto bad_fork_cleanup_audit;
1072         if ((retval = copy_files(clone_flags, p)))
1073                 goto bad_fork_cleanup_semundo;
1074         if ((retval = copy_fs(clone_flags, p)))
1075                 goto bad_fork_cleanup_files;
1076         if ((retval = copy_sighand(clone_flags, p)))
1077                 goto bad_fork_cleanup_fs;
1078         if ((retval = copy_signal(clone_flags, p)))
1079                 goto bad_fork_cleanup_sighand;
1080         if ((retval = copy_mm(clone_flags, p)))
1081                 goto bad_fork_cleanup_signal;
1082         if ((retval = copy_keys(clone_flags, p)))
1083                 goto bad_fork_cleanup_mm;
1084         if ((retval = copy_namespaces(clone_flags, p)))
1085                 goto bad_fork_cleanup_keys;
1086         if ((retval = copy_io(clone_flags, p)))
1087                 goto bad_fork_cleanup_namespaces;
1088         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1089         if (retval)
1090                 goto bad_fork_cleanup_io;
1091
1092         if (pid != &init_struct_pid) {
1093                 retval = -ENOMEM;
1094                 pid = alloc_pid(task_active_pid_ns(p));
1095                 if (!pid)
1096                         goto bad_fork_cleanup_io;
1097
1098                 if (clone_flags & CLONE_NEWPID) {
1099                         retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1100                         if (retval < 0)
1101                                 goto bad_fork_free_pid;
1102                 }
1103         }
1104
1105         p->pid = pid_nr(pid);
1106         p->tgid = p->pid;
1107         if (clone_flags & CLONE_THREAD)
1108                 p->tgid = current->tgid;
1109
1110         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1111         /*
1112          * Clear TID on mm_release()?
1113          */
1114         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1115 #ifdef CONFIG_FUTEX
1116         p->robust_list = NULL;
1117 #ifdef CONFIG_COMPAT
1118         p->compat_robust_list = NULL;
1119 #endif
1120         INIT_LIST_HEAD(&p->pi_state_list);
1121         p->pi_state_cache = NULL;
1122 #endif
1123         /*
1124          * sigaltstack should be cleared when sharing the same VM
1125          */
1126         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1127                 p->sas_ss_sp = p->sas_ss_size = 0;
1128
1129         /*
1130          * Syscall tracing should be turned off in the child regardless
1131          * of CLONE_PTRACE.
1132          */
1133         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1134 #ifdef TIF_SYSCALL_EMU
1135         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1136 #endif
1137         clear_all_latency_tracing(p);
1138
1139         /* Our parent execution domain becomes current domain
1140            These must match for thread signalling to apply */
1141         p->parent_exec_id = p->self_exec_id;
1142
1143         /* ok, now we should be set up.. */
1144         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1145         p->pdeath_signal = 0;
1146         p->exit_state = 0;
1147
1148         /*
1149          * Ok, make it visible to the rest of the system.
1150          * We dont wake it up yet.
1151          */
1152         p->group_leader = p;
1153         INIT_LIST_HEAD(&p->thread_group);
1154         INIT_LIST_HEAD(&p->ptrace_entry);
1155         INIT_LIST_HEAD(&p->ptraced);
1156
1157         /* Now that the task is set up, run cgroup callbacks if
1158          * necessary. We need to run them before the task is visible
1159          * on the tasklist. */
1160         cgroup_fork_callbacks(p);
1161         cgroup_callbacks_done = 1;
1162
1163         /* Need tasklist lock for parent etc handling! */
1164         write_lock_irq(&tasklist_lock);
1165
1166         /*
1167          * The task hasn't been attached yet, so its cpus_allowed mask will
1168          * not be changed, nor will its assigned CPU.
1169          *
1170          * The cpus_allowed mask of the parent may have changed after it was
1171          * copied first time - so re-copy it here, then check the child's CPU
1172          * to ensure it is on a valid CPU (and if not, just force it back to
1173          * parent's CPU). This avoids alot of nasty races.
1174          */
1175         p->cpus_allowed = current->cpus_allowed;
1176         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1177         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1178                         !cpu_online(task_cpu(p))))
1179                 set_task_cpu(p, smp_processor_id());
1180
1181         /* CLONE_PARENT re-uses the old parent */
1182         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1183                 p->real_parent = current->real_parent;
1184         else
1185                 p->real_parent = current;
1186         p->parent = p->real_parent;
1187
1188         spin_lock(&current->sighand->siglock);
1189
1190         /*
1191          * Process group and session signals need to be delivered to just the
1192          * parent before the fork or both the parent and the child after the
1193          * fork. Restart if a signal comes in before we add the new process to
1194          * it's process group.
1195          * A fatal signal pending means that current will exit, so the new
1196          * thread can't slip out of an OOM kill (or normal SIGKILL).
1197          */
1198         recalc_sigpending();
1199         if (signal_pending(current)) {
1200                 spin_unlock(&current->sighand->siglock);
1201                 write_unlock_irq(&tasklist_lock);
1202                 retval = -ERESTARTNOINTR;
1203                 goto bad_fork_free_pid;
1204         }
1205
1206         if (clone_flags & CLONE_THREAD) {
1207                 p->group_leader = current->group_leader;
1208                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1209
1210                 if (!cputime_eq(current->signal->it_virt_expires,
1211                                 cputime_zero) ||
1212                     !cputime_eq(current->signal->it_prof_expires,
1213                                 cputime_zero) ||
1214                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1215                     !list_empty(&current->signal->cpu_timers[0]) ||
1216                     !list_empty(&current->signal->cpu_timers[1]) ||
1217                     !list_empty(&current->signal->cpu_timers[2])) {
1218                         /*
1219                          * Have child wake up on its first tick to check
1220                          * for process CPU timers.
1221                          */
1222                         p->it_prof_expires = jiffies_to_cputime(1);
1223                 }
1224         }
1225
1226         if (likely(p->pid)) {
1227                 list_add_tail(&p->sibling, &p->real_parent->children);
1228                 if (unlikely(p->ptrace & PT_PTRACED))
1229                         __ptrace_link(p, current->parent);
1230
1231                 if (thread_group_leader(p)) {
1232                         if (clone_flags & CLONE_NEWPID)
1233                                 p->nsproxy->pid_ns->child_reaper = p;
1234
1235                         p->signal->leader_pid = pid;
1236                         p->signal->tty = current->signal->tty;
1237                         set_task_pgrp(p, task_pgrp_nr(current));
1238                         set_task_session(p, task_session_nr(current));
1239                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1240                         attach_pid(p, PIDTYPE_SID, task_session(current));
1241                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1242                         __get_cpu_var(process_counts)++;
1243                 }
1244                 attach_pid(p, PIDTYPE_PID, pid);
1245                 nr_threads++;
1246         }
1247
1248         total_forks++;
1249         spin_unlock(&current->sighand->siglock);
1250         write_unlock_irq(&tasklist_lock);
1251         proc_fork_connector(p);
1252         cgroup_post_fork(p);
1253         return p;
1254
1255 bad_fork_free_pid:
1256         if (pid != &init_struct_pid)
1257                 free_pid(pid);
1258 bad_fork_cleanup_io:
1259         put_io_context(p->io_context);
1260 bad_fork_cleanup_namespaces:
1261         exit_task_namespaces(p);
1262 bad_fork_cleanup_keys:
1263         exit_keys(p);
1264 bad_fork_cleanup_mm:
1265         if (p->mm)
1266                 mmput(p->mm);
1267 bad_fork_cleanup_signal:
1268         cleanup_signal(p);
1269 bad_fork_cleanup_sighand:
1270         __cleanup_sighand(p->sighand);
1271 bad_fork_cleanup_fs:
1272         exit_fs(p); /* blocking */
1273 bad_fork_cleanup_files:
1274         exit_files(p); /* blocking */
1275 bad_fork_cleanup_semundo:
1276         exit_sem(p);
1277 bad_fork_cleanup_audit:
1278         audit_free(p);
1279 bad_fork_cleanup_security:
1280         security_task_free(p);
1281 bad_fork_cleanup_policy:
1282 #ifdef CONFIG_NUMA
1283         mpol_put(p->mempolicy);
1284 bad_fork_cleanup_cgroup:
1285 #endif
1286         cgroup_exit(p, cgroup_callbacks_done);
1287         delayacct_tsk_free(p);
1288         if (p->binfmt)
1289                 module_put(p->binfmt->module);
1290 bad_fork_cleanup_put_domain:
1291         module_put(task_thread_info(p)->exec_domain->module);
1292 bad_fork_cleanup_count:
1293         put_group_info(p->group_info);
1294         atomic_dec(&p->user->processes);
1295         free_uid(p->user);
1296 bad_fork_free:
1297         free_task(p);
1298 fork_out:
1299         return ERR_PTR(retval);
1300 }
1301
1302 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1303 {
1304         memset(regs, 0, sizeof(struct pt_regs));
1305         return regs;
1306 }
1307
1308 struct task_struct * __cpuinit fork_idle(int cpu)
1309 {
1310         struct task_struct *task;
1311         struct pt_regs regs;
1312
1313         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1314                                 &init_struct_pid);
1315         if (!IS_ERR(task))
1316                 init_idle(task, cpu);
1317
1318         return task;
1319 }
1320
1321 static int fork_traceflag(unsigned clone_flags)
1322 {
1323         if (clone_flags & CLONE_UNTRACED)
1324                 return 0;
1325         else if (clone_flags & CLONE_VFORK) {
1326                 if (current->ptrace & PT_TRACE_VFORK)
1327                         return PTRACE_EVENT_VFORK;
1328         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1329                 if (current->ptrace & PT_TRACE_CLONE)
1330                         return PTRACE_EVENT_CLONE;
1331         } else if (current->ptrace & PT_TRACE_FORK)
1332                 return PTRACE_EVENT_FORK;
1333
1334         return 0;
1335 }
1336
1337 /*
1338  *  Ok, this is the main fork-routine.
1339  *
1340  * It copies the process, and if successful kick-starts
1341  * it and waits for it to finish using the VM if required.
1342  */
1343 long do_fork(unsigned long clone_flags,
1344               unsigned long stack_start,
1345               struct pt_regs *regs,
1346               unsigned long stack_size,
1347               int __user *parent_tidptr,
1348               int __user *child_tidptr)
1349 {
1350         struct task_struct *p;
1351         int trace = 0;
1352         long nr;
1353
1354         /*
1355          * We hope to recycle these flags after 2.6.26
1356          */
1357         if (unlikely(clone_flags & CLONE_STOPPED)) {
1358                 static int __read_mostly count = 100;
1359
1360                 if (count > 0 && printk_ratelimit()) {
1361                         char comm[TASK_COMM_LEN];
1362
1363                         count--;
1364                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1365                                         "clone flags 0x%lx\n",
1366                                 get_task_comm(comm, current),
1367                                 clone_flags & CLONE_STOPPED);
1368                 }
1369         }
1370
1371         if (unlikely(current->ptrace)) {
1372                 trace = fork_traceflag (clone_flags);
1373                 if (trace)
1374                         clone_flags |= CLONE_PTRACE;
1375         }
1376
1377         p = copy_process(clone_flags, stack_start, regs, stack_size,
1378                         child_tidptr, NULL);
1379         /*
1380          * Do this prior waking up the new thread - the thread pointer
1381          * might get invalid after that point, if the thread exits quickly.
1382          */
1383         if (!IS_ERR(p)) {
1384                 struct completion vfork;
1385
1386                 nr = task_pid_vnr(p);
1387
1388                 if (clone_flags & CLONE_PARENT_SETTID)
1389                         put_user(nr, parent_tidptr);
1390
1391                 if (clone_flags & CLONE_VFORK) {
1392                         p->vfork_done = &vfork;
1393                         init_completion(&vfork);
1394                 }
1395
1396                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1397                         /*
1398                          * We'll start up with an immediate SIGSTOP.
1399                          */
1400                         sigaddset(&p->pending.signal, SIGSTOP);
1401                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1402                 }
1403
1404                 if (!(clone_flags & CLONE_STOPPED))
1405                         wake_up_new_task(p, clone_flags);
1406                 else
1407                         __set_task_state(p, TASK_STOPPED);
1408
1409                 if (unlikely (trace)) {
1410                         current->ptrace_message = nr;
1411                         ptrace_notify ((trace << 8) | SIGTRAP);
1412                 }
1413
1414                 if (clone_flags & CLONE_VFORK) {
1415                         freezer_do_not_count();
1416                         wait_for_completion(&vfork);
1417                         freezer_count();
1418                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1419                                 current->ptrace_message = nr;
1420                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1421                         }
1422                 }
1423         } else {
1424                 nr = PTR_ERR(p);
1425         }
1426         return nr;
1427 }
1428
1429 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1430 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1431 #endif
1432
1433 static void sighand_ctor(struct kmem_cache *cachep, void *data)
1434 {
1435         struct sighand_struct *sighand = data;
1436
1437         spin_lock_init(&sighand->siglock);
1438         init_waitqueue_head(&sighand->signalfd_wqh);
1439 }
1440
1441 void __init proc_caches_init(void)
1442 {
1443         sighand_cachep = kmem_cache_create("sighand_cache",
1444                         sizeof(struct sighand_struct), 0,
1445                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1446                         sighand_ctor);
1447         signal_cachep = kmem_cache_create("signal_cache",
1448                         sizeof(struct signal_struct), 0,
1449                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1450         files_cachep = kmem_cache_create("files_cache",
1451                         sizeof(struct files_struct), 0,
1452                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1453         fs_cachep = kmem_cache_create("fs_cache",
1454                         sizeof(struct fs_struct), 0,
1455                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1456         vm_area_cachep = kmem_cache_create("vm_area_struct",
1457                         sizeof(struct vm_area_struct), 0,
1458                         SLAB_PANIC, NULL);
1459         mm_cachep = kmem_cache_create("mm_struct",
1460                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1461                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1462 }
1463
1464 /*
1465  * Check constraints on flags passed to the unshare system call and
1466  * force unsharing of additional process context as appropriate.
1467  */
1468 static void check_unshare_flags(unsigned long *flags_ptr)
1469 {
1470         /*
1471          * If unsharing a thread from a thread group, must also
1472          * unshare vm.
1473          */
1474         if (*flags_ptr & CLONE_THREAD)
1475                 *flags_ptr |= CLONE_VM;
1476
1477         /*
1478          * If unsharing vm, must also unshare signal handlers.
1479          */
1480         if (*flags_ptr & CLONE_VM)
1481                 *flags_ptr |= CLONE_SIGHAND;
1482
1483         /*
1484          * If unsharing signal handlers and the task was created
1485          * using CLONE_THREAD, then must unshare the thread
1486          */
1487         if ((*flags_ptr & CLONE_SIGHAND) &&
1488             (atomic_read(&current->signal->count) > 1))
1489                 *flags_ptr |= CLONE_THREAD;
1490
1491         /*
1492          * If unsharing namespace, must also unshare filesystem information.
1493          */
1494         if (*flags_ptr & CLONE_NEWNS)
1495                 *flags_ptr |= CLONE_FS;
1496 }
1497
1498 /*
1499  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1500  */
1501 static int unshare_thread(unsigned long unshare_flags)
1502 {
1503         if (unshare_flags & CLONE_THREAD)
1504                 return -EINVAL;
1505
1506         return 0;
1507 }
1508
1509 /*
1510  * Unshare the filesystem structure if it is being shared
1511  */
1512 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1513 {
1514         struct fs_struct *fs = current->fs;
1515
1516         if ((unshare_flags & CLONE_FS) &&
1517             (fs && atomic_read(&fs->count) > 1)) {
1518                 *new_fsp = __copy_fs_struct(current->fs);
1519                 if (!*new_fsp)
1520                         return -ENOMEM;
1521         }
1522
1523         return 0;
1524 }
1525
1526 /*
1527  * Unsharing of sighand is not supported yet
1528  */
1529 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1530 {
1531         struct sighand_struct *sigh = current->sighand;
1532
1533         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1534                 return -EINVAL;
1535         else
1536                 return 0;
1537 }
1538
1539 /*
1540  * Unshare vm if it is being shared
1541  */
1542 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1543 {
1544         struct mm_struct *mm = current->mm;
1545
1546         if ((unshare_flags & CLONE_VM) &&
1547             (mm && atomic_read(&mm->mm_users) > 1)) {
1548                 return -EINVAL;
1549         }
1550
1551         return 0;
1552 }
1553
1554 /*
1555  * Unshare file descriptor table if it is being shared
1556  */
1557 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1558 {
1559         struct files_struct *fd = current->files;
1560         int error = 0;
1561
1562         if ((unshare_flags & CLONE_FILES) &&
1563             (fd && atomic_read(&fd->count) > 1)) {
1564                 *new_fdp = dup_fd(fd, &error);
1565                 if (!*new_fdp)
1566                         return error;
1567         }
1568
1569         return 0;
1570 }
1571
1572 /*
1573  * unshare allows a process to 'unshare' part of the process
1574  * context which was originally shared using clone.  copy_*
1575  * functions used by do_fork() cannot be used here directly
1576  * because they modify an inactive task_struct that is being
1577  * constructed. Here we are modifying the current, active,
1578  * task_struct.
1579  */
1580 asmlinkage long sys_unshare(unsigned long unshare_flags)
1581 {
1582         int err = 0;
1583         struct fs_struct *fs, *new_fs = NULL;
1584         struct sighand_struct *new_sigh = NULL;
1585         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1586         struct files_struct *fd, *new_fd = NULL;
1587         struct nsproxy *new_nsproxy = NULL;
1588         int do_sysvsem = 0;
1589
1590         check_unshare_flags(&unshare_flags);
1591
1592         /* Return -EINVAL for all unsupported flags */
1593         err = -EINVAL;
1594         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1595                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1596                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
1597                                 CLONE_NEWNET))
1598                 goto bad_unshare_out;
1599
1600         /*
1601          * CLONE_NEWIPC must also detach from the undolist: after switching
1602          * to a new ipc namespace, the semaphore arrays from the old
1603          * namespace are unreachable.
1604          */
1605         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1606                 do_sysvsem = 1;
1607         if ((err = unshare_thread(unshare_flags)))
1608                 goto bad_unshare_out;
1609         if ((err = unshare_fs(unshare_flags, &new_fs)))
1610                 goto bad_unshare_cleanup_thread;
1611         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1612                 goto bad_unshare_cleanup_fs;
1613         if ((err = unshare_vm(unshare_flags, &new_mm)))
1614                 goto bad_unshare_cleanup_sigh;
1615         if ((err = unshare_fd(unshare_flags, &new_fd)))
1616                 goto bad_unshare_cleanup_vm;
1617         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1618                         new_fs)))
1619                 goto bad_unshare_cleanup_fd;
1620
1621         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1622                 if (do_sysvsem) {
1623                         /*
1624                          * CLONE_SYSVSEM is equivalent to sys_exit().
1625                          */
1626                         exit_sem(current);
1627                 }
1628
1629                 if (new_nsproxy) {
1630                         switch_task_namespaces(current, new_nsproxy);
1631                         new_nsproxy = NULL;
1632                 }
1633
1634                 task_lock(current);
1635
1636                 if (new_fs) {
1637                         fs = current->fs;
1638                         current->fs = new_fs;
1639                         new_fs = fs;
1640                 }
1641
1642                 if (new_mm) {
1643                         mm = current->mm;
1644                         active_mm = current->active_mm;
1645                         current->mm = new_mm;
1646                         current->active_mm = new_mm;
1647                         activate_mm(active_mm, new_mm);
1648                         new_mm = mm;
1649                 }
1650
1651                 if (new_fd) {
1652                         fd = current->files;
1653                         current->files = new_fd;
1654                         new_fd = fd;
1655                 }
1656
1657                 task_unlock(current);
1658         }
1659
1660         if (new_nsproxy)
1661                 put_nsproxy(new_nsproxy);
1662
1663 bad_unshare_cleanup_fd:
1664         if (new_fd)
1665                 put_files_struct(new_fd);
1666
1667 bad_unshare_cleanup_vm:
1668         if (new_mm)
1669                 mmput(new_mm);
1670
1671 bad_unshare_cleanup_sigh:
1672         if (new_sigh)
1673                 if (atomic_dec_and_test(&new_sigh->count))
1674                         kmem_cache_free(sighand_cachep, new_sigh);
1675
1676 bad_unshare_cleanup_fs:
1677         if (new_fs)
1678                 put_fs_struct(new_fs);
1679
1680 bad_unshare_cleanup_thread:
1681 bad_unshare_out:
1682         return err;
1683 }
1684
1685 /*
1686  *      Helper to unshare the files of the current task.
1687  *      We don't want to expose copy_files internals to
1688  *      the exec layer of the kernel.
1689  */
1690
1691 int unshare_files(struct files_struct **displaced)
1692 {
1693         struct task_struct *task = current;
1694         struct files_struct *copy = NULL;
1695         int error;
1696
1697         error = unshare_fd(CLONE_FILES, &copy);
1698         if (error || !copy) {
1699                 *displaced = NULL;
1700                 return error;
1701         }
1702         *displaced = task->files;
1703         task_lock(task);
1704         task->files = copy;
1705         task_unlock(task);
1706         return 0;
1707 }