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