err.no Git - linux-2.6/blob - 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/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/fs.h>
  30 #include <linux/nsproxy.h>
  31 #include <linux/capability.h>
  32 #include <linux/cpu.h>
  33 #include <linux/cgroup.h>
  34 #include <linux/security.h>
  35 #include <linux/swap.h>
  36 #include <linux/syscalls.h>
  37 #include <linux/jiffies.h>
  38 #include <linux/futex.h>
  39 #include <linux/task_io_accounting_ops.h>
  40 #include <linux/rcupdate.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/mount.h>
  43 #include <linux/audit.h>
  44 #include <linux/memcontrol.h>
  45 #include <linux/profile.h>
  46 #include <linux/rmap.h>
  47 #include <linux/acct.h>
  48 #include <linux/tsacct_kern.h>
  49 #include <linux/cn_proc.h>
  50 #include <linux/freezer.h>
  51 #include <linux/delayacct.h>
  52 #include <linux/taskstats_kern.h>
  53 #include <linux/random.h>
  54 #include <linux/tty.h>
  55 #include <linux/proc_fs.h>
  56 #include <linux/blkdev.h>
  57
  58 #include <asm/pgtable.h>
  59 #include <asm/pgalloc.h>
  60 #include <asm/uaccess.h>
  61 #include <asm/mmu_context.h>
  62 #include <asm/cacheflush.h>
  63 #include <asm/tlbflush.h>
  64
  65 /*
  66  * Protected counters by write_lock_irq(&tasklist_lock)
  67  */
  68 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  69 int nr_threads;                 /* The idle threads do not count.. */
  70
  71 int max_threads;                /* tunable limit on nr_threads */
  72
  73 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  74
  75 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  76
  77 int nr_processes(void)
  78 {
  79         int cpu;
  80         int total = 0;
  81
  82         for_each_online_cpu(cpu)
  83                 total += per_cpu(process_counts, cpu);
  84
  85         return total;
  86 }
  87
  88 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  89 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  90 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  91 static struct kmem_cache *task_struct_cachep;
  92 #endif
  93
  94 /* SLAB cache for signal_struct structures (tsk->signal) */
  95 static struct kmem_cache *signal_cachep;
  96
  97 /* SLAB cache for sighand_struct structures (tsk->sighand) */
  98 struct kmem_cache *sighand_cachep;
  99
 100 /* SLAB cache for files_struct structures (tsk->files) */
 101 struct kmem_cache *files_cachep;
 102
 103 /* SLAB cache for fs_struct structures (tsk->fs) */
 104 struct kmem_cache *fs_cachep;
 105
 106 /* SLAB cache for vm_area_struct structures */
 107 struct kmem_cache *vm_area_cachep;
 108
 109 /* SLAB cache for mm_struct structures (tsk->mm) */
 110 static struct kmem_cache *mm_cachep;
 111
 112 void free_task(struct task_struct *tsk)
 113 {
 114         prop_local_destroy_single(&tsk->dirties);
 115         free_thread_info(tsk->stack);
 116         rt_mutex_debug_task_free(tsk);
 117         free_task_struct(tsk);
 118 }
 119 EXPORT_SYMBOL(free_task);
 120
 121 void __put_task_struct(struct task_struct *tsk)
 122 {
 123         WARN_ON(!tsk->exit_state);
 124         WARN_ON(atomic_read(&tsk->usage));
 125         WARN_ON(tsk == current);
 126
 127         security_task_free(tsk);
 128         free_uid(tsk->user);
 129         put_group_info(tsk->group_info);
 130         delayacct_tsk_free(tsk);
 131
 132         if (!profile_handoff_task(tsk))
 133                 free_task(tsk);
 134 }
 135
 136 /*
 137  * macro override instead of weak attribute alias, to workaround
 138  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 139  */
 140 #ifndef arch_task_cache_init
 141 #define arch_task_cache_init()
 142 #endif
 143
 144 void __init fork_init(unsigned long mempages)
 145 {
 146 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 147 #ifndef ARCH_MIN_TASKALIGN
 148 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 149 #endif
 150         /* create a slab on which task_structs can be allocated */
 151         task_struct_cachep =
 152                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 153                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
 154 #endif
 155
 156         /* do the arch specific task caches init */
 157         arch_task_cache_init();
 158
 159         /*
 160          * The default maximum number of threads is set to a safe
 161          * value: the thread structures can take up at most half
 162          * of memory.
 163          */
 164         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 165
 166         /*
 167          * we need to allow at least 20 threads to boot a system
 168          */
 169         if(max_threads < 20)
 170                 max_threads = 20;
 171
 172         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 173         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 174         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 175                 init_task.signal->rlim[RLIMIT_NPROC];
 176 }
 177
 178 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 179                                                struct task_struct *src)
 180 {
 181         *dst = *src;
 182         return 0;
 183 }
 184
 185 static struct task_struct *dup_task_struct(struct task_struct *orig)
 186 {
 187         struct task_struct *tsk;
 188         struct thread_info *ti;
 189         int err;
 190
 191         prepare_to_copy(orig);
 192
 193         tsk = alloc_task_struct();
 194         if (!tsk)
 195                 return NULL;
 196
 197         ti = alloc_thread_info(tsk);
 198         if (!ti) {
 199                 free_task_struct(tsk);
 200                 return NULL;
 201         }
 202
 203         err = arch_dup_task_struct(tsk, orig);
 204         if (err)
 205                 goto out;
 206
 207         tsk->stack = ti;
 208
 209         err = prop_local_init_single(&tsk->dirties);
 210         if (err)
 211                 goto out;
 212
 213         setup_thread_stack(tsk, orig);
 214
 215 #ifdef CONFIG_CC_STACKPROTECTOR
 216         tsk->stack_canary = get_random_int();
 217 #endif
 218
 219         /* One for us, one for whoever does the "release_task()" (usually parent) */
 220         atomic_set(&tsk->usage,2);
 221         atomic_set(&tsk->fs_excl, 0);
 222 #ifdef CONFIG_BLK_DEV_IO_TRACE
 223         tsk->btrace_seq = 0;
 224 #endif
 225         tsk->splice_pipe = NULL;
 226         return tsk;
 227
 228 out:
 229         free_thread_info(ti);
 230         free_task_struct(tsk);
 231         return NULL;
 232 }
 233
 234 #ifdef CONFIG_MMU
 235 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 236 {
 237         struct vm_area_struct *mpnt, *tmp, **pprev;
 238         struct rb_node **rb_link, *rb_parent;
 239         int retval;
 240         unsigned long charge;
 241         struct mempolicy *pol;
 242
 243         down_write(&oldmm->mmap_sem);
 244         flush_cache_dup_mm(oldmm);
 245         /*
 246          * Not linked in yet - no deadlock potential:
 247          */
 248         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 249
 250         mm->locked_vm = 0;
 251         mm->mmap = NULL;
 252         mm->mmap_cache = NULL;
 253         mm->free_area_cache = oldmm->mmap_base;
 254         mm->cached_hole_size = ~0UL;
 255         mm->map_count = 0;
 256         cpus_clear(mm->cpu_vm_mask);
 257         mm->mm_rb = RB_ROOT;
 258         rb_link = &mm->mm_rb.rb_node;
 259         rb_parent = NULL;
 260         pprev = &mm->mmap;
 261
 262         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 263                 struct file *file;
 264
 265                 if (mpnt->vm_flags & VM_DONTCOPY) {
 266                         long pages = vma_pages(mpnt);
 267                         mm->total_vm -= pages;
 268                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 269                                                                 -pages);
 270                         continue;
 271                 }
 272                 charge = 0;
 273                 if (mpnt->vm_flags & VM_ACCOUNT) {
 274                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 275                         if (security_vm_enough_memory(len))
 276                                 goto fail_nomem;
 277                         charge = len;
 278                 }
 279                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 280                 if (!tmp)
 281                         goto fail_nomem;
 282                 *tmp = *mpnt;
 283                 pol = mpol_dup(vma_policy(mpnt));
 284                 retval = PTR_ERR(pol);
 285                 if (IS_ERR(pol))
 286                         goto fail_nomem_policy;
 287                 vma_set_policy(tmp, pol);
 288                 tmp->vm_flags &= ~VM_LOCKED;
 289                 tmp->vm_mm = mm;
 290                 tmp->vm_next = NULL;
 291                 anon_vma_link(tmp);
 292                 file = tmp->vm_file;
 293                 if (file) {
 294                         struct inode *inode = file->f_path.dentry->d_inode;
 295                         get_file(file);
 296                         if (tmp->vm_flags & VM_DENYWRITE)
 297                                 atomic_dec(&inode->i_writecount);
 298
 299                         /* insert tmp into the share list, just after mpnt */
 300                         spin_lock(&file->f_mapping->i_mmap_lock);
 301                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 302                         flush_dcache_mmap_lock(file->f_mapping);
 303                         vma_prio_tree_add(tmp, mpnt);
 304                         flush_dcache_mmap_unlock(file->f_mapping);
 305                         spin_unlock(&file->f_mapping->i_mmap_lock);
 306                 }
 307
 308                 /*
 309                  * Link in the new vma and copy the page table entries.
 310                  */
 311                 *pprev = tmp;
 312                 pprev = &tmp->vm_next;
 313
 314                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 315                 rb_link = &tmp->vm_rb.rb_right;
 316                 rb_parent = &tmp->vm_rb;
 317
 318                 mm->map_count++;
 319                 retval = copy_page_range(mm, oldmm, mpnt);
 320
 321                 if (tmp->vm_ops && tmp->vm_ops->open)
 322                         tmp->vm_ops->open(tmp);
 323
 324                 if (retval)
 325                         goto out;
 326         }
 327         /* a new mm has just been created */
 328         arch_dup_mmap(oldmm, mm);
 329         retval = 0;
 330 out:
 331         up_write(&mm->mmap_sem);
 332         flush_tlb_mm(oldmm);
 333         up_write(&oldmm->mmap_sem);
 334         return retval;
 335 fail_nomem_policy:
 336         kmem_cache_free(vm_area_cachep, tmp);
 337 fail_nomem:
 338         retval = -ENOMEM;
 339         vm_unacct_memory(charge);
 340         goto out;
 341 }
 342
 343 static inline int mm_alloc_pgd(struct mm_struct * mm)
 344 {
 345         mm->pgd = pgd_alloc(mm);
 346         if (unlikely(!mm->pgd))
 347                 return -ENOMEM;
 348         return 0;
 349 }
 350
 351 static inline void mm_free_pgd(struct mm_struct * mm)
 352 {
 353         pgd_free(mm, mm->pgd);
 354 }
 355 #else
 356 #define dup_mmap(mm, oldmm)     (0)
 357 #define mm_alloc_pgd(mm)        (0)
 358 #define mm_free_pgd(mm)
 359 #endif /* CONFIG_MMU */
 360
 361 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 362
 363 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 364 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 365
 366 #include <linux/init_task.h>
 367
 368 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
 369 {
 370         atomic_set(&mm->mm_users, 1);
 371         atomic_set(&mm->mm_count, 1);
 372         init_rwsem(&mm->mmap_sem);
 373         INIT_LIST_HEAD(&mm->mmlist);
 374         mm->flags = (current->mm) ? current->mm->flags
 375                                   : MMF_DUMP_FILTER_DEFAULT;
 376         mm->core_waiters = 0;
 377         mm->nr_ptes = 0;
 378         set_mm_counter(mm, file_rss, 0);
 379         set_mm_counter(mm, anon_rss, 0);
 380         spin_lock_init(&mm->page_table_lock);
 381         rwlock_init(&mm->ioctx_list_lock);
 382         mm->ioctx_list = NULL;
 383         mm->free_area_cache = TASK_UNMAPPED_BASE;
 384         mm->cached_hole_size = ~0UL;
 385         mm_init_owner(mm, p);
 386
 387         if (likely(!mm_alloc_pgd(mm))) {
 388                 mm->def_flags = 0;
 389                 return mm;
 390         }
 391
 392         free_mm(mm);
 393         return NULL;
 394 }
 395
 396 /*
 397  * Allocate and initialize an mm_struct.
 398  */
 399 struct mm_struct * mm_alloc(void)
 400 {
 401         struct mm_struct * mm;
 402
 403         mm = allocate_mm();
 404         if (mm) {
 405                 memset(mm, 0, sizeof(*mm));
 406                 mm = mm_init(mm, current);
 407         }
 408         return mm;
 409 }
 410
 411 /*
 412  * Called when the last reference to the mm
 413  * is dropped: either by a lazy thread or by
 414  * mmput. Free the page directory and the mm.
 415  */
 416 void __mmdrop(struct mm_struct *mm)
 417 {
 418         BUG_ON(mm == &init_mm);
 419         mm_free_pgd(mm);
 420         destroy_context(mm);
 421         free_mm(mm);
 422 }
 423 EXPORT_SYMBOL_GPL(__mmdrop);
 424
 425 /*
 426  * Decrement the use count and release all resources for an mm.
 427  */
 428 void mmput(struct mm_struct *mm)
 429 {
 430         might_sleep();
 431
 432         if (atomic_dec_and_test(&mm->mm_users)) {
 433                 exit_aio(mm);
 434                 exit_mmap(mm);
 435                 set_mm_exe_file(mm, NULL);
 436                 if (!list_empty(&mm->mmlist)) {
 437                         spin_lock(&mmlist_lock);
 438                         list_del(&mm->mmlist);
 439                         spin_unlock(&mmlist_lock);
 440                 }
 441                 put_swap_token(mm);
 442                 mmdrop(mm);
 443         }
 444 }
 445 EXPORT_SYMBOL_GPL(mmput);
 446
 447 /**
 448  * get_task_mm - acquire a reference to the task's mm
 449  *
 450  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
 451  * this kernel workthread has transiently adopted a user mm with use_mm,
 452  * to do its AIO) is not set and if so returns a reference to it, after
 453  * bumping up the use count.  User must release the mm via mmput()
 454  * after use.  Typically used by /proc and ptrace.
 455  */
 456 struct mm_struct *get_task_mm(struct task_struct *task)
 457 {
 458         struct mm_struct *mm;
 459
 460         task_lock(task);
 461         mm = task->mm;
 462         if (mm) {
 463                 if (task->flags & PF_BORROWED_MM)
 464                         mm = NULL;
 465                 else
 466                         atomic_inc(&mm->mm_users);
 467         }
 468         task_unlock(task);
 469         return mm;
 470 }
 471 EXPORT_SYMBOL_GPL(get_task_mm);
 472
 473 /* Please note the differences between mmput and mm_release.
 474  * mmput is called whenever we stop holding onto a mm_struct,
 475  * error success whatever.
 476  *
 477  * mm_release is called after a mm_struct has been removed
 478  * from the current process.
 479  *
 480  * This difference is important for error handling, when we
 481  * only half set up a mm_struct for a new process and need to restore
 482  * the old one.  Because we mmput the new mm_struct before
 483  * restoring the old one. . .
 484  * Eric Biederman 10 January 1998
 485  */
 486 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 487 {
 488         struct completion *vfork_done = tsk->vfork_done;
 489
 490         /* Get rid of any cached register state */
 491         deactivate_mm(tsk, mm);
 492
 493         /* notify parent sleeping on vfork() */
 494         if (vfork_done) {
 495                 tsk->vfork_done = NULL;
 496                 complete(vfork_done);
 497         }
 498
 499         /*
 500          * If we're exiting normally, clear a user-space tid field if
 501          * requested.  We leave this alone when dying by signal, to leave
 502          * the value intact in a core dump, and to save the unnecessary
 503          * trouble otherwise.  Userland only wants this done for a sys_exit.
 504          */
 505         if (tsk->clear_child_tid
 506             && !(tsk->flags & PF_SIGNALED)
 507             && atomic_read(&mm->mm_users) > 1) {
 508                 u32 __user * tidptr = tsk->clear_child_tid;
 509                 tsk->clear_child_tid = NULL;
 510
 511                 /*
 512                  * We don't check the error code - if userspace has
 513                  * not set up a proper pointer then tough luck.
 514                  */
 515                 put_user(0, tidptr);
 516                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 517         }
 518 }
 519
 520 /*
 521  * Allocate a new mm structure and copy contents from the
 522  * mm structure of the passed in task structure.
 523  */
 524 struct mm_struct *dup_mm(struct task_struct *tsk)
 525 {
 526         struct mm_struct *mm, *oldmm = current->mm;
 527         int err;
 528
 529         if (!oldmm)
 530                 return NULL;
 531
 532         mm = allocate_mm();
 533         if (!mm)
 534                 goto fail_nomem;
 535
 536         memcpy(mm, oldmm, sizeof(*mm));
 537
 538         /* Initializing for Swap token stuff */
 539         mm->token_priority = 0;
 540         mm->last_interval = 0;
 541
 542         if (!mm_init(mm, tsk))
 543                 goto fail_nomem;
 544
 545         if (init_new_context(tsk, mm))
 546                 goto fail_nocontext;
 547
 548         dup_mm_exe_file(oldmm, mm);
 549
 550         err = dup_mmap(mm, oldmm);
 551         if (err)
 552                 goto free_pt;
 553
 554         mm->hiwater_rss = get_mm_rss(mm);
 555         mm->hiwater_vm = mm->total_vm;
 556
 557         return mm;
 558
 559 free_pt:
 560         mmput(mm);
 561
 562 fail_nomem:
 563         return NULL;
 564
 565 fail_nocontext:
 566         /*
 567          * If init_new_context() failed, we cannot use mmput() to free the mm
 568          * because it calls destroy_context()
 569          */
 570         mm_free_pgd(mm);
 571         free_mm(mm);
 572         return NULL;
 573 }
 574
 575 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 576 {
 577         struct mm_struct * mm, *oldmm;
 578         int retval;
 579
 580         tsk->min_flt = tsk->maj_flt = 0;
 581         tsk->nvcsw = tsk->nivcsw = 0;
 582
 583         tsk->mm = NULL;
 584         tsk->active_mm = NULL;
 585
 586         /*
 587          * Are we cloning a kernel thread?
 588          *
 589          * We need to steal a active VM for that..
 590          */
 591         oldmm = current->mm;
 592         if (!oldmm)
 593                 return 0;
 594
 595         if (clone_flags & CLONE_VM) {
 596                 atomic_inc(&oldmm->mm_users);
 597                 mm = oldmm;
 598                 goto good_mm;
 599         }
 600
 601         retval = -ENOMEM;
 602         mm = dup_mm(tsk);
 603         if (!mm)
 604                 goto fail_nomem;
 605
 606 good_mm:
 607         /* Initializing for Swap token stuff */
 608         mm->token_priority = 0;
 609         mm->last_interval = 0;
 610
 611         tsk->mm = mm;
 612         tsk->active_mm = mm;
 613         return 0;
 614
 615 fail_nomem:
 616         return retval;
 617 }
 618
 619 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 620 {
 621         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 622         /* We don't need to lock fs - think why ;-) */
 623         if (fs) {
 624                 atomic_set(&fs->count, 1);
 625                 rwlock_init(&fs->lock);
 626                 fs->umask = old->umask;
 627                 read_lock(&old->lock);
 628                 fs->root = old->root;
 629                 path_get(&old->root);
 630                 fs->pwd = old->pwd;
 631                 path_get(&old->pwd);
 632                 if (old->altroot.dentry) {
 633                         fs->altroot = old->altroot;
 634                         path_get(&old->altroot);
 635                 } else {
 636                         fs->altroot.mnt = NULL;
 637                         fs->altroot.dentry = NULL;
 638                 }
 639                 read_unlock(&old->lock);
 640         }
 641         return fs;
 642 }
 643
 644 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 645 {
 646         return __copy_fs_struct(old);
 647 }
 648
 649 EXPORT_SYMBOL_GPL(copy_fs_struct);
 650
 651 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 652 {
 653         if (clone_flags & CLONE_FS) {
 654                 atomic_inc(&current->fs->count);
 655                 return 0;
 656         }
 657         tsk->fs = __copy_fs_struct(current->fs);
 658         if (!tsk->fs)
 659                 return -ENOMEM;
 660         return 0;
 661 }
 662
 663 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 664 {
 665         struct files_struct *oldf, *newf;
 666         int error = 0;
 667
 668         /*
 669          * A background process may not have any files ...
 670          */
 671         oldf = current->files;
 672         if (!oldf)
 673                 goto out;
 674
 675         if (clone_flags & CLONE_FILES) {
 676                 atomic_inc(&oldf->count);
 677                 goto out;
 678         }
 679
 680         newf = dup_fd(oldf, &error);
 681         if (!newf)
 682                 goto out;
 683
 684         tsk->files = newf;
 685         error = 0;
 686 out:
 687         return error;
 688 }
 689
 690 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 691 {
 692 #ifdef CONFIG_BLOCK
 693         struct io_context *ioc = current->io_context;
 694
 695         if (!ioc)
 696                 return 0;
 697         /*
 698          * Share io context with parent, if CLONE_IO is set
 699          */
 700         if (clone_flags & CLONE_IO) {
 701                 tsk->io_context = ioc_task_link(ioc);
 702                 if (unlikely(!tsk->io_context))
 703                         return -ENOMEM;
 704         } else if (ioprio_valid(ioc->ioprio)) {
 705                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 706                 if (unlikely(!tsk->io_context))
 707                         return -ENOMEM;
 708
 709                 tsk->io_context->ioprio = ioc->ioprio;
 710         }
 711 #endif
 712         return 0;
 713 }
 714
 715 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 716 {
 717         struct sighand_struct *sig;
 718
 719         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 720                 atomic_inc(&current->sighand->count);
 721                 return 0;
 722         }
 723         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 724         rcu_assign_pointer(tsk->sighand, sig);
 725         if (!sig)
 726                 return -ENOMEM;
 727         atomic_set(&sig->count, 1);
 728         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 729         return 0;
 730 }
 731
 732 void __cleanup_sighand(struct sighand_struct *sighand)
 733 {
 734         if (atomic_dec_and_test(&sighand->count))
 735                 kmem_cache_free(sighand_cachep, sighand);
 736 }
 737
 738 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 739 {
 740         struct signal_struct *sig;
 741         int ret;
 742
 743         if (clone_flags & CLONE_THREAD) {
 744                 atomic_inc(&current->signal->count);
 745                 atomic_inc(&current->signal->live);
 746                 return 0;
 747         }
 748         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 749         tsk->signal = sig;
 750         if (!sig)
 751                 return -ENOMEM;
 752
 753         ret = copy_thread_group_keys(tsk);
 754         if (ret < 0) {
 755                 kmem_cache_free(signal_cachep, sig);
 756                 return ret;
 757         }
 758
 759         atomic_set(&sig->count, 1);
 760         atomic_set(&sig->live, 1);
 761         init_waitqueue_head(&sig->wait_chldexit);
 762         sig->flags = 0;
 763         sig->group_exit_code = 0;
 764         sig->group_exit_task = NULL;
 765         sig->group_stop_count = 0;
 766         sig->curr_target = tsk;
 767         init_sigpending(&sig->shared_pending);
 768         INIT_LIST_HEAD(&sig->posix_timers);
 769
 770         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 771         sig->it_real_incr.tv64 = 0;
 772         sig->real_timer.function = it_real_fn;
 773
 774         sig->it_virt_expires = cputime_zero;
 775         sig->it_virt_incr = cputime_zero;
 776         sig->it_prof_expires = cputime_zero;
 777         sig->it_prof_incr = cputime_zero;
 778
 779         sig->leader = 0;        /* session leadership doesn't inherit */
 780         sig->tty_old_pgrp = NULL;
 781
 782         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 783         sig->gtime = cputime_zero;
 784         sig->cgtime = cputime_zero;
 785         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 786         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 787         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
 788         sig->sum_sched_runtime = 0;
 789         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 790         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 791         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 792         taskstats_tgid_init(sig);
 793
 794         task_lock(current->group_leader);
 795         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 796         task_unlock(current->group_leader);
 797
 798         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 799                 /*
 800                  * New sole thread in the process gets an expiry time
 801                  * of the whole CPU time limit.
 802                  */
 803                 tsk->it_prof_expires =
 804                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 805         }
 806         acct_init_pacct(&sig->pacct);
 807
 808         tty_audit_fork(sig);
 809
 810         return 0;
 811 }
 812
 813 void __cleanup_signal(struct signal_struct *sig)
 814 {
 815         exit_thread_group_keys(sig);
 816         kmem_cache_free(signal_cachep, sig);
 817 }
 818
 819 static void cleanup_signal(struct task_struct *tsk)
 820 {
 821         struct signal_struct *sig = tsk->signal;
 822
 823         atomic_dec(&sig->live);
 824
 825         if (atomic_dec_and_test(&sig->count))
 826                 __cleanup_signal(sig);
 827 }
 828
 829 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 830 {
 831         unsigned long new_flags = p->flags;
 832
 833         new_flags &= ~PF_SUPERPRIV;
 834         new_flags |= PF_FORKNOEXEC;
 835         if (!(clone_flags & CLONE_PTRACE))
 836                 p->ptrace = 0;
 837         p->flags = new_flags;
 838         clear_freeze_flag(p);
 839 }
 840
 841 asmlinkage long sys_set_tid_address(int __user *tidptr)
 842 {
 843         current->clear_child_tid = tidptr;
 844
 845         return task_pid_vnr(current);
 846 }
 847
 848 static void rt_mutex_init_task(struct task_struct *p)
 849 {
 850         spin_lock_init(&p->pi_lock);
 851 #ifdef CONFIG_RT_MUTEXES
 852         plist_head_init(&p->pi_waiters, &p->pi_lock);
 853         p->pi_blocked_on = NULL;
 854 #endif
 855 }
 856
 857 #ifdef CONFIG_MM_OWNER
 858 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 859 {
 860         mm->owner = p;
 861 }
 862 #endif /* CONFIG_MM_OWNER */
 863
 864 /*
 865  * This creates a new process as a copy of the old one,
 866  * but does not actually start it yet.
 867  *
 868  * It copies the registers, and all the appropriate
 869  * parts of the process environment (as per the clone
 870  * flags). The actual kick-off is left to the caller.
 871  */
 872 static struct task_struct *copy_process(unsigned long clone_flags,
 873                                         unsigned long stack_start,
 874                                         struct pt_regs *regs,
 875                                         unsigned long stack_size,
 876                                         int __user *child_tidptr,
 877                                         struct pid *pid)
 878 {
 879         int retval;
 880         struct task_struct *p;
 881         int cgroup_callbacks_done = 0;
 882
 883         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 884                 return ERR_PTR(-EINVAL);
 885
 886         /*
 887          * Thread groups must share signals as well, and detached threads
 888          * can only be started up within the thread group.
 889          */
 890         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
 891                 return ERR_PTR(-EINVAL);
 892
 893         /*
 894          * Shared signal handlers imply shared VM. By way of the above,
 895          * thread groups also imply shared VM. Blocking this case allows
 896          * for various simplifications in other code.
 897          */
 898         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
 899                 return ERR_PTR(-EINVAL);
 900
 901         retval = security_task_create(clone_flags);
 902         if (retval)
 903                 goto fork_out;
 904
 905         retval = -ENOMEM;
 906         p = dup_task_struct(current);
 907         if (!p)
 908                 goto fork_out;
 909
 910         rt_mutex_init_task(p);
 911
 912 #ifdef CONFIG_TRACE_IRQFLAGS
 913         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
 914         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
 915 #endif
 916         retval = -EAGAIN;
 917         if (atomic_read(&p->user->processes) >=
 918                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
 919                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
 920                     p->user != current->nsproxy->user_ns->root_user)
 921                         goto bad_fork_free;
 922         }
 923
 924         atomic_inc(&p->user->__count);
 925         atomic_inc(&p->user->processes);
 926         get_group_info(p->group_info);
 927
 928         /*
 929          * If multiple threads are within copy_process(), then this check
 930          * triggers too late. This doesn't hurt, the check is only there
 931          * to stop root fork bombs.
 932          */
 933         if (nr_threads >= max_threads)
 934                 goto bad_fork_cleanup_count;
 935
 936         if (!try_module_get(task_thread_info(p)->exec_domain->module))
 937                 goto bad_fork_cleanup_count;
 938
 939         if (p->binfmt && !try_module_get(p->binfmt->module))
 940                 goto bad_fork_cleanup_put_domain;
 941
 942         p->did_exec = 0;
 943         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
 944         copy_flags(clone_flags, p);
 945         INIT_LIST_HEAD(&p->children);
 946         INIT_LIST_HEAD(&p->sibling);
 947 #ifdef CONFIG_PREEMPT_RCU
 948         p->rcu_read_lock_nesting = 0;
 949         p->rcu_flipctr_idx = 0;
 950 #endif /* #ifdef CONFIG_PREEMPT_RCU */
 951         p->vfork_done = NULL;
 952         spin_lock_init(&p->alloc_lock);
 953
 954         clear_tsk_thread_flag(p, TIF_SIGPENDING);
 955         init_sigpending(&p->pending);
 956
 957         p->utime = cputime_zero;
 958         p->stime = cputime_zero;
 959         p->gtime = cputime_zero;
 960         p->utimescaled = cputime_zero;
 961         p->stimescaled = cputime_zero;
 962         p->prev_utime = cputime_zero;
 963         p->prev_stime = cputime_zero;
 964
 965 #ifdef CONFIG_DETECT_SOFTLOCKUP
 966         p->last_switch_count = 0;
 967         p->last_switch_timestamp = 0;
 968 #endif
 969
 970 #ifdef CONFIG_TASK_XACCT
 971         p->rchar = 0;           /* I/O counter: bytes read */
 972         p->wchar = 0;           /* I/O counter: bytes written */
 973         p->syscr = 0;           /* I/O counter: read syscalls */
 974         p->syscw = 0;           /* I/O counter: write syscalls */
 975 #endif
 976         task_io_accounting_init(p);
 977         acct_clear_integrals(p);
 978
 979         p->it_virt_expires = cputime_zero;
 980         p->it_prof_expires = cputime_zero;
 981         p->it_sched_expires = 0;
 982         INIT_LIST_HEAD(&p->cpu_timers[0]);
 983         INIT_LIST_HEAD(&p->cpu_timers[1]);
 984         INIT_LIST_HEAD(&p->cpu_timers[2]);
 985
 986         p->lock_depth = -1;             /* -1 = no lock */
 987         do_posix_clock_monotonic_gettime(&p->start_time);
 988         p->real_start_time = p->start_time;
 989         monotonic_to_bootbased(&p->real_start_time);
 990 #ifdef CONFIG_SECURITY
 991         p->security = NULL;
 992 #endif
 993         p->cap_bset = current->cap_bset;
 994         p->io_context = NULL;
 995         p->audit_context = NULL;
 996         cgroup_fork(p);
 997 #ifdef CONFIG_NUMA
 998         p->mempolicy = mpol_dup(p->mempolicy);
 999         if (IS_ERR(p->mempolicy)) {
1000                 retval = PTR_ERR(p->mempolicy);
1001                 p->mempolicy = NULL;
1002                 goto bad_fork_cleanup_cgroup;
1003         }
1004         mpol_fix_fork_child_flag(p);
1005 #endif
1006 #ifdef CONFIG_TRACE_IRQFLAGS
1007         p->irq_events = 0;
1008 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1009         p->hardirqs_enabled = 1;
1010 #else
1011         p->hardirqs_enabled = 0;
1012 #endif
1013         p->hardirq_enable_ip = 0;
1014         p->hardirq_enable_event = 0;
1015         p->hardirq_disable_ip = _THIS_IP_;
1016         p->hardirq_disable_event = 0;
1017         p->softirqs_enabled = 1;
1018         p->softirq_enable_ip = _THIS_IP_;
1019         p->softirq_enable_event = 0;
1020         p->softirq_disable_ip = 0;
1021         p->softirq_disable_event = 0;
1022         p->hardirq_context = 0;
1023         p->softirq_context = 0;
1024 #endif
1025 #ifdef CONFIG_LOCKDEP
1026         p->lockdep_depth = 0; /* no locks held yet */
1027         p->curr_chain_key = 0;
1028         p->lockdep_recursion = 0;
1029 #endif
1030
1031 #ifdef CONFIG_DEBUG_MUTEXES
1032         p->blocked_on = NULL; /* not blocked yet */
1033 #endif
1034
1035         /* Perform scheduler related setup. Assign this task to a CPU. */
1036         sched_fork(p, clone_flags);
1037
1038         if ((retval = security_task_alloc(p)))
1039                 goto bad_fork_cleanup_policy;
1040         if ((retval = audit_alloc(p)))
1041                 goto bad_fork_cleanup_security;
1042         /* copy all the process information */
1043         if ((retval = copy_semundo(clone_flags, p)))
1044                 goto bad_fork_cleanup_audit;
1045         if ((retval = copy_files(clone_flags, p)))
1046                 goto bad_fork_cleanup_semundo;
1047         if ((retval = copy_fs(clone_flags, p)))
1048                 goto bad_fork_cleanup_files;
1049         if ((retval = copy_sighand(clone_flags, p)))
1050                 goto bad_fork_cleanup_fs;
1051         if ((retval = copy_signal(clone_flags, p)))
1052                 goto bad_fork_cleanup_sighand;
1053         if ((retval = copy_mm(clone_flags, p)))
1054                 goto bad_fork_cleanup_signal;
1055         if ((retval = copy_keys(clone_flags, p)))
1056                 goto bad_fork_cleanup_mm;
1057         if ((retval = copy_namespaces(clone_flags, p)))
1058                 goto bad_fork_cleanup_keys;
1059         if ((retval = copy_io(clone_flags, p)))
1060                 goto bad_fork_cleanup_namespaces;
1061         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1062         if (retval)
1063                 goto bad_fork_cleanup_io;
1064
1065         if (pid != &init_struct_pid) {
1066                 retval = -ENOMEM;
1067                 pid = alloc_pid(task_active_pid_ns(p));
1068                 if (!pid)
1069                         goto bad_fork_cleanup_io;
1070
1071                 if (clone_flags & CLONE_NEWPID) {
1072                         retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1073                         if (retval < 0)
1074                                 goto bad_fork_free_pid;
1075                 }
1076         }
1077
1078         p->pid = pid_nr(pid);
1079         p->tgid = p->pid;
1080         if (clone_flags & CLONE_THREAD)
1081                 p->tgid = current->tgid;
1082
1083         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1084         /*
1085          * Clear TID on mm_release()?
1086          */
1087         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1088 #ifdef CONFIG_FUTEX
1089         p->robust_list = NULL;
1090 #ifdef CONFIG_COMPAT
1091         p->compat_robust_list = NULL;
1092 #endif
1093         INIT_LIST_HEAD(&p->pi_state_list);
1094         p->pi_state_cache = NULL;
1095 #endif
1096         /*
1097          * sigaltstack should be cleared when sharing the same VM
1098          */
1099         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1100                 p->sas_ss_sp = p->sas_ss_size = 0;
1101
1102         /*
1103          * Syscall tracing should be turned off in the child regardless
1104          * of CLONE_PTRACE.
1105          */
1106         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1107 #ifdef TIF_SYSCALL_EMU
1108         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1109 #endif
1110         clear_all_latency_tracing(p);
1111
1112         /* Our parent execution domain becomes current domain
1113            These must match for thread signalling to apply */
1114         p->parent_exec_id = p->self_exec_id;
1115
1116         /* ok, now we should be set up.. */
1117         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1118         p->pdeath_signal = 0;
1119         p->exit_state = 0;
1120
1121         /*
1122          * Ok, make it visible to the rest of the system.
1123          * We dont wake it up yet.
1124          */
1125         p->group_leader = p;
1126         INIT_LIST_HEAD(&p->thread_group);
1127         INIT_LIST_HEAD(&p->ptrace_children);
1128         INIT_LIST_HEAD(&p->ptrace_list);
1129
1130         /* Now that the task is set up, run cgroup callbacks if
1131          * necessary. We need to run them before the task is visible
1132          * on the tasklist. */
1133         cgroup_fork_callbacks(p);
1134         cgroup_callbacks_done = 1;
1135
1136         /* Need tasklist lock for parent etc handling! */
1137         write_lock_irq(&tasklist_lock);
1138
1139         /*
1140          * The task hasn't been attached yet, so its cpus_allowed mask will
1141          * not be changed, nor will its assigned CPU.
1142          *
1143          * The cpus_allowed mask of the parent may have changed after it was
1144          * copied first time - so re-copy it here, then check the child's CPU
1145          * to ensure it is on a valid CPU (and if not, just force it back to
1146          * parent's CPU). This avoids alot of nasty races.
1147          */
1148         p->cpus_allowed = current->cpus_allowed;
1149         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1150         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1151                         !cpu_online(task_cpu(p))))
1152                 set_task_cpu(p, smp_processor_id());
1153
1154         /* CLONE_PARENT re-uses the old parent */
1155         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1156                 p->real_parent = current->real_parent;
1157         else
1158                 p->real_parent = current;
1159         p->parent = p->real_parent;
1160
1161         spin_lock(&current->sighand->siglock);
1162
1163         /*
1164          * Process group and session signals need to be delivered to just the
1165          * parent before the fork or both the parent and the child after the
1166          * fork. Restart if a signal comes in before we add the new process to
1167          * it's process group.
1168          * A fatal signal pending means that current will exit, so the new
1169          * thread can't slip out of an OOM kill (or normal SIGKILL).
1170          */
1171         recalc_sigpending();
1172         if (signal_pending(current)) {
1173                 spin_unlock(&current->sighand->siglock);
1174                 write_unlock_irq(&tasklist_lock);
1175                 retval = -ERESTARTNOINTR;
1176                 goto bad_fork_free_pid;
1177         }
1178
1179         if (clone_flags & CLONE_THREAD) {
1180                 p->group_leader = current->group_leader;
1181                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1182
1183                 if (!cputime_eq(current->signal->it_virt_expires,
1184                                 cputime_zero) ||
1185                     !cputime_eq(current->signal->it_prof_expires,
1186                                 cputime_zero) ||
1187                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1188                     !list_empty(&current->signal->cpu_timers[0]) ||
1189                     !list_empty(&current->signal->cpu_timers[1]) ||
1190                     !list_empty(&current->signal->cpu_timers[2])) {
1191                         /*
1192                          * Have child wake up on its first tick to check
1193                          * for process CPU timers.
1194                          */
1195                         p->it_prof_expires = jiffies_to_cputime(1);
1196                 }
1197         }
1198
1199         if (likely(p->pid)) {
1200                 add_parent(p);
1201                 if (unlikely(p->ptrace & PT_PTRACED))
1202                         __ptrace_link(p, current->parent);
1203
1204                 if (thread_group_leader(p)) {
1205                         if (clone_flags & CLONE_NEWPID)
1206                                 p->nsproxy->pid_ns->child_reaper = p;
1207
1208                         p->signal->leader_pid = pid;
1209                         p->signal->tty = current->signal->tty;
1210                         set_task_pgrp(p, task_pgrp_nr(current));
1211                         set_task_session(p, task_session_nr(current));
1212                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1213                         attach_pid(p, PIDTYPE_SID, task_session(current));
1214                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1215                         __get_cpu_var(process_counts)++;
1216                 }
1217                 attach_pid(p, PIDTYPE_PID, pid);
1218                 nr_threads++;
1219         }
1220
1221         total_forks++;
1222         spin_unlock(&current->sighand->siglock);
1223         write_unlock_irq(&tasklist_lock);
1224         proc_fork_connector(p);
1225         cgroup_post_fork(p);
1226         return p;
1227
1228 bad_fork_free_pid:
1229         if (pid != &init_struct_pid)
1230                 free_pid(pid);
1231 bad_fork_cleanup_io:
1232         put_io_context(p->io_context);
1233 bad_fork_cleanup_namespaces:
1234         exit_task_namespaces(p);
1235 bad_fork_cleanup_keys:
1236         exit_keys(p);
1237 bad_fork_cleanup_mm:
1238         if (p->mm)
1239                 mmput(p->mm);
1240 bad_fork_cleanup_signal:
1241         cleanup_signal(p);
1242 bad_fork_cleanup_sighand:
1243         __cleanup_sighand(p->sighand);
1244 bad_fork_cleanup_fs:
1245         exit_fs(p); /* blocking */
1246 bad_fork_cleanup_files:
1247         exit_files(p); /* blocking */
1248 bad_fork_cleanup_semundo:
1249         exit_sem(p);
1250 bad_fork_cleanup_audit:
1251         audit_free(p);
1252 bad_fork_cleanup_security:
1253         security_task_free(p);
1254 bad_fork_cleanup_policy:
1255 #ifdef CONFIG_NUMA
1256         mpol_put(p->mempolicy);
1257 bad_fork_cleanup_cgroup:
1258 #endif
1259         cgroup_exit(p, cgroup_callbacks_done);
1260         delayacct_tsk_free(p);
1261         if (p->binfmt)
1262                 module_put(p->binfmt->module);
1263 bad_fork_cleanup_put_domain:
1264         module_put(task_thread_info(p)->exec_domain->module);
1265 bad_fork_cleanup_count:
1266         put_group_info(p->group_info);
1267         atomic_dec(&p->user->processes);
1268         free_uid(p->user);
1269 bad_fork_free:
1270         free_task(p);
1271 fork_out:
1272         return ERR_PTR(retval);
1273 }
1274
1275 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1276 {
1277         memset(regs, 0, sizeof(struct pt_regs));
1278         return regs;
1279 }
1280
1281 struct task_struct * __cpuinit fork_idle(int cpu)
1282 {
1283         struct task_struct *task;
1284         struct pt_regs regs;
1285
1286         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1287                                 &init_struct_pid);
1288         if (!IS_ERR(task))
1289                 init_idle(task, cpu);
1290
1291         return task;
1292 }
1293
1294 static int fork_traceflag(unsigned clone_flags)
1295 {
1296         if (clone_flags & CLONE_UNTRACED)
1297                 return 0;
1298         else if (clone_flags & CLONE_VFORK) {
1299                 if (current->ptrace & PT_TRACE_VFORK)
1300                         return PTRACE_EVENT_VFORK;
1301         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1302                 if (current->ptrace & PT_TRACE_CLONE)
1303                         return PTRACE_EVENT_CLONE;
1304         } else if (current->ptrace & PT_TRACE_FORK)
1305                 return PTRACE_EVENT_FORK;
1306
1307         return 0;
1308 }
1309
1310 /*
1311  *  Ok, this is the main fork-routine.
1312  *
1313  * It copies the process, and if successful kick-starts
1314  * it and waits for it to finish using the VM if required.
1315  */
1316 long do_fork(unsigned long clone_flags,
1317               unsigned long stack_start,
1318               struct pt_regs *regs,
1319               unsigned long stack_size,
1320               int __user *parent_tidptr,
1321               int __user *child_tidptr)
1322 {
1323         struct task_struct *p;
1324         int trace = 0;
1325         long nr;
1326
1327         /*
1328          * We hope to recycle these flags after 2.6.26
1329          */
1330         if (unlikely(clone_flags & CLONE_STOPPED)) {
1331                 static int __read_mostly count = 100;
1332
1333                 if (count > 0 && printk_ratelimit()) {
1334                         char comm[TASK_COMM_LEN];
1335
1336                         count--;
1337                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1338                                         "clone flags 0x%lx\n",
1339                                 get_task_comm(comm, current),
1340                                 clone_flags & CLONE_STOPPED);
1341                 }
1342         }
1343
1344         if (unlikely(current->ptrace)) {
1345                 trace = fork_traceflag (clone_flags);
1346                 if (trace)
1347                         clone_flags |= CLONE_PTRACE;
1348         }
1349
1350         p = copy_process(clone_flags, stack_start, regs, stack_size,
1351                         child_tidptr, NULL);
1352         /*
1353          * Do this prior waking up the new thread - the thread pointer
1354          * might get invalid after that point, if the thread exits quickly.
1355          */
1356         if (!IS_ERR(p)) {
1357                 struct completion vfork;
1358
1359                 nr = task_pid_vnr(p);
1360
1361                 if (clone_flags & CLONE_PARENT_SETTID)
1362                         put_user(nr, parent_tidptr);
1363
1364                 if (clone_flags & CLONE_VFORK) {
1365                         p->vfork_done = &vfork;
1366                         init_completion(&vfork);
1367                 }
1368
1369                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1370                         /*
1371                          * We'll start up with an immediate SIGSTOP.
1372                          */
1373                         sigaddset(&p->pending.signal, SIGSTOP);
1374                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1375                 }
1376
1377                 if (!(clone_flags & CLONE_STOPPED))
1378                         wake_up_new_task(p, clone_flags);
1379                 else
1380                         __set_task_state(p, TASK_STOPPED);
1381
1382                 if (unlikely (trace)) {
1383                         current->ptrace_message = nr;
1384                         ptrace_notify ((trace << 8) | SIGTRAP);
1385                 }
1386
1387                 if (clone_flags & CLONE_VFORK) {
1388                         freezer_do_not_count();
1389                         wait_for_completion(&vfork);
1390                         freezer_count();
1391                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1392                                 current->ptrace_message = nr;
1393                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1394                         }
1395                 }
1396         } else {
1397                 nr = PTR_ERR(p);
1398         }
1399         return nr;
1400 }
1401
1402 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1403 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1404 #endif
1405
1406 static void sighand_ctor(struct kmem_cache *cachep, void *data)
1407 {
1408         struct sighand_struct *sighand = data;
1409
1410         spin_lock_init(&sighand->siglock);
1411         init_waitqueue_head(&sighand->signalfd_wqh);
1412 }
1413
1414 void __init proc_caches_init(void)
1415 {
1416         sighand_cachep = kmem_cache_create("sighand_cache",
1417                         sizeof(struct sighand_struct), 0,
1418                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1419                         sighand_ctor);
1420         signal_cachep = kmem_cache_create("signal_cache",
1421                         sizeof(struct signal_struct), 0,
1422                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1423         files_cachep = kmem_cache_create("files_cache",
1424                         sizeof(struct files_struct), 0,
1425                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1426         fs_cachep = kmem_cache_create("fs_cache",
1427                         sizeof(struct fs_struct), 0,
1428                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1429         vm_area_cachep = kmem_cache_create("vm_area_struct",
1430                         sizeof(struct vm_area_struct), 0,
1431                         SLAB_PANIC, NULL);
1432         mm_cachep = kmem_cache_create("mm_struct",
1433                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1434                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1435 }
1436
1437 /*
1438  * Check constraints on flags passed to the unshare system call and
1439  * force unsharing of additional process context as appropriate.
1440  */
1441 static void check_unshare_flags(unsigned long *flags_ptr)
1442 {
1443         /*
1444          * If unsharing a thread from a thread group, must also
1445          * unshare vm.
1446          */
1447         if (*flags_ptr & CLONE_THREAD)
1448                 *flags_ptr |= CLONE_VM;
1449
1450         /*
1451          * If unsharing vm, must also unshare signal handlers.
1452          */
1453         if (*flags_ptr & CLONE_VM)
1454                 *flags_ptr |= CLONE_SIGHAND;
1455
1456         /*
1457          * If unsharing signal handlers and the task was created
1458          * using CLONE_THREAD, then must unshare the thread
1459          */
1460         if ((*flags_ptr & CLONE_SIGHAND) &&
1461             (atomic_read(&current->signal->count) > 1))
1462                 *flags_ptr |= CLONE_THREAD;
1463
1464         /*
1465          * If unsharing namespace, must also unshare filesystem information.
1466          */
1467         if (*flags_ptr & CLONE_NEWNS)
1468                 *flags_ptr |= CLONE_FS;
1469 }
1470
1471 /*
1472  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1473  */
1474 static int unshare_thread(unsigned long unshare_flags)
1475 {
1476         if (unshare_flags & CLONE_THREAD)
1477                 return -EINVAL;
1478
1479         return 0;
1480 }
1481
1482 /*
1483  * Unshare the filesystem structure if it is being shared
1484  */
1485 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1486 {
1487         struct fs_struct *fs = current->fs;
1488
1489         if ((unshare_flags & CLONE_FS) &&
1490             (fs && atomic_read(&fs->count) > 1)) {
1491                 *new_fsp = __copy_fs_struct(current->fs);
1492                 if (!*new_fsp)
1493                         return -ENOMEM;
1494         }
1495
1496         return 0;
1497 }
1498
1499 /*
1500  * Unsharing of sighand is not supported yet
1501  */
1502 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1503 {
1504         struct sighand_struct *sigh = current->sighand;
1505
1506         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1507                 return -EINVAL;
1508         else
1509                 return 0;
1510 }
1511
1512 /*
1513  * Unshare vm if it is being shared
1514  */
1515 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1516 {
1517         struct mm_struct *mm = current->mm;
1518
1519         if ((unshare_flags & CLONE_VM) &&
1520             (mm && atomic_read(&mm->mm_users) > 1)) {
1521                 return -EINVAL;
1522         }
1523
1524         return 0;
1525 }
1526
1527 /*
1528  * Unshare file descriptor table if it is being shared
1529  */
1530 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1531 {
1532         struct files_struct *fd = current->files;
1533         int error = 0;
1534
1535         if ((unshare_flags & CLONE_FILES) &&
1536             (fd && atomic_read(&fd->count) > 1)) {
1537                 *new_fdp = dup_fd(fd, &error);
1538                 if (!*new_fdp)
1539                         return error;
1540         }
1541
1542         return 0;
1543 }
1544
1545 /*
1546  * unshare allows a process to 'unshare' part of the process
1547  * context which was originally shared using clone.  copy_*
1548  * functions used by do_fork() cannot be used here directly
1549  * because they modify an inactive task_struct that is being
1550  * constructed. Here we are modifying the current, active,
1551  * task_struct.
1552  */
1553 asmlinkage long sys_unshare(unsigned long unshare_flags)
1554 {
1555         int err = 0;
1556         struct fs_struct *fs, *new_fs = NULL;
1557         struct sighand_struct *new_sigh = NULL;
1558         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1559         struct files_struct *fd, *new_fd = NULL;
1560         struct nsproxy *new_nsproxy = NULL;
1561         int do_sysvsem = 0;
1562
1563         check_unshare_flags(&unshare_flags);
1564
1565         /* Return -EINVAL for all unsupported flags */
1566         err = -EINVAL;
1567         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1568                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1569                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
1570                                 CLONE_NEWNET))
1571                 goto bad_unshare_out;
1572
1573         /*
1574          * CLONE_NEWIPC must also detach from the undolist: after switching
1575          * to a new ipc namespace, the semaphore arrays from the old
1576          * namespace are unreachable.
1577          */
1578         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1579                 do_sysvsem = 1;
1580         if ((err = unshare_thread(unshare_flags)))
1581                 goto bad_unshare_out;
1582         if ((err = unshare_fs(unshare_flags, &new_fs)))
1583                 goto bad_unshare_cleanup_thread;
1584         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1585                 goto bad_unshare_cleanup_fs;
1586         if ((err = unshare_vm(unshare_flags, &new_mm)))
1587                 goto bad_unshare_cleanup_sigh;
1588         if ((err = unshare_fd(unshare_flags, &new_fd)))
1589                 goto bad_unshare_cleanup_vm;
1590         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1591                         new_fs)))
1592                 goto bad_unshare_cleanup_fd;
1593
1594         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1595                 if (do_sysvsem) {
1596                         /*
1597                          * CLONE_SYSVSEM is equivalent to sys_exit().
1598                          */
1599                         exit_sem(current);
1600                 }
1601
1602                 if (new_nsproxy) {
1603                         switch_task_namespaces(current, new_nsproxy);
1604                         new_nsproxy = NULL;
1605                 }
1606
1607                 task_lock(current);
1608
1609                 if (new_fs) {
1610                         fs = current->fs;
1611                         current->fs = new_fs;
1612                         new_fs = fs;
1613                 }
1614
1615                 if (new_mm) {
1616                         mm = current->mm;
1617                         active_mm = current->active_mm;
1618                         current->mm = new_mm;
1619                         current->active_mm = new_mm;
1620                         activate_mm(active_mm, new_mm);
1621                         new_mm = mm;
1622                 }
1623
1624                 if (new_fd) {
1625                         fd = current->files;
1626                         current->files = new_fd;
1627                         new_fd = fd;
1628                 }
1629
1630                 task_unlock(current);
1631         }
1632
1633         if (new_nsproxy)
1634                 put_nsproxy(new_nsproxy);
1635
1636 bad_unshare_cleanup_fd:
1637         if (new_fd)
1638                 put_files_struct(new_fd);
1639
1640 bad_unshare_cleanup_vm:
1641         if (new_mm)
1642                 mmput(new_mm);
1643
1644 bad_unshare_cleanup_sigh:
1645         if (new_sigh)
1646                 if (atomic_dec_and_test(&new_sigh->count))
1647                         kmem_cache_free(sighand_cachep, new_sigh);
1648
1649 bad_unshare_cleanup_fs:
1650         if (new_fs)
1651                 put_fs_struct(new_fs);
1652
1653 bad_unshare_cleanup_thread:
1654 bad_unshare_out:
1655         return err;
1656 }
1657
1658 /*
1659  *      Helper to unshare the files of the current task.
1660  *      We don't want to expose copy_files internals to
1661  *      the exec layer of the kernel.
1662  */
1663
1664 int unshare_files(struct files_struct **displaced)
1665 {
1666         struct task_struct *task = current;
1667         struct files_struct *copy = NULL;
1668         int error;
1669
1670         error = unshare_fd(CLONE_FILES, &copy);
1671         if (error || !copy) {
1672                 *displaced = NULL;
1673                 return error;
1674         }
1675         *displaced = task->files;
1676         task_lock(task);
1677         task->files = copy;
1678         task_unlock(task);
1679         return 0;
1680 }