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/key.h>
  26 #include <linux/binfmts.h>
  27 #include <linux/mman.h>
  28 #include <linux/fs.h>
  29 #include <linux/nsproxy.h>
  30 #include <linux/capability.h>
  31 #include <linux/cpu.h>
  32 #include <linux/cgroup.h>
  33 #include <linux/security.h>
  34 #include <linux/swap.h>
  35 #include <linux/syscalls.h>
  36 #include <linux/jiffies.h>
  37 #include <linux/futex.h>
  38 #include <linux/task_io_accounting_ops.h>
  39 #include <linux/rcupdate.h>
  40 #include <linux/ptrace.h>
  41 #include <linux/mount.h>
  42 #include <linux/audit.h>
  43 #include <linux/memcontrol.h>
  44 #include <linux/profile.h>
  45 #include <linux/rmap.h>
  46 #include <linux/acct.h>
  47 #include <linux/tsacct_kern.h>
  48 #include <linux/cn_proc.h>
  49 #include <linux/freezer.h>
  50 #include <linux/delayacct.h>
  51 #include <linux/taskstats_kern.h>
  52 #include <linux/random.h>
  53 #include <linux/tty.h>
  54 #include <linux/proc_fs.h>
  55 #include <linux/blkdev.h>
  56
  57 #include <asm/pgtable.h>
  58 #include <asm/pgalloc.h>
  59 #include <asm/uaccess.h>
  60 #include <asm/mmu_context.h>
  61 #include <asm/cacheflush.h>
  62 #include <asm/tlbflush.h>
  63
  64 /*
  65  * Protected counters by write_lock_irq(&tasklist_lock)
  66  */
  67 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  68 int nr_threads;                 /* The idle threads do not count.. */
  69
  70 int max_threads;                /* tunable limit on nr_threads */
  71
  72 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  73
  74 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  75
  76 int nr_processes(void)
  77 {
  78         int cpu;
  79         int total = 0;
  80
  81         for_each_online_cpu(cpu)
  82                 total += per_cpu(process_counts, cpu);
  83
  84         return total;
  85 }
  86
  87 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
  88 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
  89 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
  90 static struct kmem_cache *task_struct_cachep;
  91 #endif
  92
  93 /* SLAB cache for signal_struct structures (tsk->signal) */
  94 static struct kmem_cache *signal_cachep;
  95
  96 /* SLAB cache for sighand_struct structures (tsk->sighand) */
  97 struct kmem_cache *sighand_cachep;
  98
  99 /* SLAB cache for files_struct structures (tsk->files) */
 100 struct kmem_cache *files_cachep;
 101
 102 /* SLAB cache for fs_struct structures (tsk->fs) */
 103 struct kmem_cache *fs_cachep;
 104
 105 /* SLAB cache for vm_area_struct structures */
 106 struct kmem_cache *vm_area_cachep;
 107
 108 /* SLAB cache for mm_struct structures (tsk->mm) */
 109 static struct kmem_cache *mm_cachep;
 110
 111 void free_task(struct task_struct *tsk)
 112 {
 113         prop_local_destroy_single(&tsk->dirties);
 114         free_thread_info(tsk->stack);
 115         rt_mutex_debug_task_free(tsk);
 116         free_task_struct(tsk);
 117 }
 118 EXPORT_SYMBOL(free_task);
 119
 120 void __put_task_struct(struct task_struct *tsk)
 121 {
 122         WARN_ON(!tsk->exit_state);
 123         WARN_ON(atomic_read(&tsk->usage));
 124         WARN_ON(tsk == current);
 125
 126         security_task_free(tsk);
 127         free_uid(tsk->user);
 128         put_group_info(tsk->group_info);
 129         delayacct_tsk_free(tsk);
 130
 131         if (!profile_handoff_task(tsk))
 132                 free_task(tsk);
 133 }
 134
 135 /*
 136  * macro override instead of weak attribute alias, to workaround
 137  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 138  */
 139 #ifndef arch_task_cache_init
 140 #define arch_task_cache_init()
 141 #endif
 142
 143 void __init fork_init(unsigned long mempages)
 144 {
 145 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 146 #ifndef ARCH_MIN_TASKALIGN
 147 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 148 #endif
 149         /* create a slab on which task_structs can be allocated */
 150         task_struct_cachep =
 151                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 152                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
 153 #endif
 154
 155         /* do the arch specific task caches init */
 156         arch_task_cache_init();
 157
 158         /*
 159          * The default maximum number of threads is set to a safe
 160          * value: the thread structures can take up at most half
 161          * of memory.
 162          */
 163         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 164
 165         /*
 166          * we need to allow at least 20 threads to boot a system
 167          */
 168         if(max_threads < 20)
 169                 max_threads = 20;
 170
 171         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 172         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 173         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 174                 init_task.signal->rlim[RLIMIT_NPROC];
 175 }
 176
 177 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 178                                                struct task_struct *src)
 179 {
 180         *dst = *src;
 181         return 0;
 182 }
 183
 184 static struct task_struct *dup_task_struct(struct task_struct *orig)
 185 {
 186         struct task_struct *tsk;
 187         struct thread_info *ti;
 188         int err;
 189
 190         prepare_to_copy(orig);
 191
 192         tsk = alloc_task_struct();
 193         if (!tsk)
 194                 return NULL;
 195
 196         ti = alloc_thread_info(tsk);
 197         if (!ti) {
 198                 free_task_struct(tsk);
 199                 return NULL;
 200         }
 201
 202         err = arch_dup_task_struct(tsk, orig);
 203         if (err)
 204                 goto out;
 205
 206         tsk->stack = ti;
 207
 208         err = prop_local_init_single(&tsk->dirties);
 209         if (err)
 210                 goto out;
 211
 212         setup_thread_stack(tsk, orig);
 213
 214 #ifdef CONFIG_CC_STACKPROTECTOR
 215         tsk->stack_canary = get_random_int();
 216 #endif
 217
 218         /* One for us, one for whoever does the "release_task()" (usually parent) */
 219         atomic_set(&tsk->usage,2);
 220         atomic_set(&tsk->fs_excl, 0);
 221 #ifdef CONFIG_BLK_DEV_IO_TRACE
 222         tsk->btrace_seq = 0;
 223 #endif
 224         tsk->splice_pipe = NULL;
 225         return tsk;
 226
 227 out:
 228         free_thread_info(ti);
 229         free_task_struct(tsk);
 230         return NULL;
 231 }
 232
 233 #ifdef CONFIG_MMU
 234 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 235 {
 236         struct vm_area_struct *mpnt, *tmp, **pprev;
 237         struct rb_node **rb_link, *rb_parent;
 238         int retval;
 239         unsigned long charge;
 240         struct mempolicy *pol;
 241
 242         down_write(&oldmm->mmap_sem);
 243         flush_cache_dup_mm(oldmm);
 244         /*
 245          * Not linked in yet - no deadlock potential:
 246          */
 247         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 248
 249         mm->locked_vm = 0;
 250         mm->mmap = NULL;
 251         mm->mmap_cache = NULL;
 252         mm->free_area_cache = oldmm->mmap_base;
 253         mm->cached_hole_size = ~0UL;
 254         mm->map_count = 0;
 255         cpus_clear(mm->cpu_vm_mask);
 256         mm->mm_rb = RB_ROOT;
 257         rb_link = &mm->mm_rb.rb_node;
 258         rb_parent = NULL;
 259         pprev = &mm->mmap;
 260
 261         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 262                 struct file *file;
 263
 264                 if (mpnt->vm_flags & VM_DONTCOPY) {
 265                         long pages = vma_pages(mpnt);
 266                         mm->total_vm -= pages;
 267                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 268                                                                 -pages);
 269                         continue;
 270                 }
 271                 charge = 0;
 272                 if (mpnt->vm_flags & VM_ACCOUNT) {
 273                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 274                         if (security_vm_enough_memory(len))
 275                                 goto fail_nomem;
 276                         charge = len;
 277                 }
 278                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 279                 if (!tmp)
 280                         goto fail_nomem;
 281                 *tmp = *mpnt;
 282                 pol = mpol_dup(vma_policy(mpnt));
 283                 retval = PTR_ERR(pol);
 284                 if (IS_ERR(pol))
 285                         goto fail_nomem_policy;
 286                 vma_set_policy(tmp, pol);
 287                 tmp->vm_flags &= ~VM_LOCKED;
 288                 tmp->vm_mm = mm;
 289                 tmp->vm_next = NULL;
 290                 anon_vma_link(tmp);
 291                 file = tmp->vm_file;
 292                 if (file) {
 293                         struct inode *inode = file->f_path.dentry->d_inode;
 294                         get_file(file);
 295                         if (tmp->vm_flags & VM_DENYWRITE)
 296                                 atomic_dec(&inode->i_writecount);
 297
 298                         /* insert tmp into the share list, just after mpnt */
 299                         spin_lock(&file->f_mapping->i_mmap_lock);
 300                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 301                         flush_dcache_mmap_lock(file->f_mapping);
 302                         vma_prio_tree_add(tmp, mpnt);
 303                         flush_dcache_mmap_unlock(file->f_mapping);
 304                         spin_unlock(&file->f_mapping->i_mmap_lock);
 305                 }
 306
 307                 /*
 308                  * Link in the new vma and copy the page table entries.
 309                  */
 310                 *pprev = tmp;
 311                 pprev = &tmp->vm_next;
 312
 313                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 314                 rb_link = &tmp->vm_rb.rb_right;
 315                 rb_parent = &tmp->vm_rb;
 316
 317                 mm->map_count++;
 318                 retval = copy_page_range(mm, oldmm, mpnt);
 319
 320                 if (tmp->vm_ops && tmp->vm_ops->open)
 321                         tmp->vm_ops->open(tmp);
 322
 323                 if (retval)
 324                         goto out;
 325         }
 326         /* a new mm has just been created */
 327         arch_dup_mmap(oldmm, mm);
 328         retval = 0;
 329 out:
 330         up_write(&mm->mmap_sem);
 331         flush_tlb_mm(oldmm);
 332         up_write(&oldmm->mmap_sem);
 333         return retval;
 334 fail_nomem_policy:
 335         kmem_cache_free(vm_area_cachep, tmp);
 336 fail_nomem:
 337         retval = -ENOMEM;
 338         vm_unacct_memory(charge);
 339         goto out;
 340 }
 341
 342 static inline int mm_alloc_pgd(struct mm_struct * mm)
 343 {
 344         mm->pgd = pgd_alloc(mm);
 345         if (unlikely(!mm->pgd))
 346                 return -ENOMEM;
 347         return 0;
 348 }
 349
 350 static inline void mm_free_pgd(struct mm_struct * mm)
 351 {
 352         pgd_free(mm, mm->pgd);
 353 }
 354 #else
 355 #define dup_mmap(mm, oldmm)     (0)
 356 #define mm_alloc_pgd(mm)        (0)
 357 #define mm_free_pgd(mm)
 358 #endif /* CONFIG_MMU */
 359
 360 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 361
 362 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 363 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 364
 365 #include <linux/init_task.h>
 366
 367 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
 368 {
 369         atomic_set(&mm->mm_users, 1);
 370         atomic_set(&mm->mm_count, 1);
 371         init_rwsem(&mm->mmap_sem);
 372         INIT_LIST_HEAD(&mm->mmlist);
 373         mm->flags = (current->mm) ? current->mm->flags
 374                                   : MMF_DUMP_FILTER_DEFAULT;
 375         mm->core_waiters = 0;
 376         mm->nr_ptes = 0;
 377         set_mm_counter(mm, file_rss, 0);
 378         set_mm_counter(mm, anon_rss, 0);
 379         spin_lock_init(&mm->page_table_lock);
 380         rwlock_init(&mm->ioctx_list_lock);
 381         mm->ioctx_list = NULL;
 382         mm->free_area_cache = TASK_UNMAPPED_BASE;
 383         mm->cached_hole_size = ~0UL;
 384         mm_init_owner(mm, p);
 385
 386         if (likely(!mm_alloc_pgd(mm))) {
 387                 mm->def_flags = 0;
 388                 return mm;
 389         }
 390
 391         free_mm(mm);
 392         return NULL;
 393 }
 394
 395 /*
 396  * Allocate and initialize an mm_struct.
 397  */
 398 struct mm_struct * mm_alloc(void)
 399 {
 400         struct mm_struct * mm;
 401
 402         mm = allocate_mm();
 403         if (mm) {
 404                 memset(mm, 0, sizeof(*mm));
 405                 mm = mm_init(mm, current);
 406         }
 407         return mm;
 408 }
 409
 410 /*
 411  * Called when the last reference to the mm
 412  * is dropped: either by a lazy thread or by
 413  * mmput. Free the page directory and the mm.
 414  */
 415 void __mmdrop(struct mm_struct *mm)
 416 {
 417         BUG_ON(mm == &init_mm);
 418         mm_free_pgd(mm);
 419         destroy_context(mm);
 420         free_mm(mm);
 421 }
 422 EXPORT_SYMBOL_GPL(__mmdrop);
 423
 424 /*
 425  * Decrement the use count and release all resources for an mm.
 426  */
 427 void mmput(struct mm_struct *mm)
 428 {
 429         might_sleep();
 430
 431         if (atomic_dec_and_test(&mm->mm_users)) {
 432                 exit_aio(mm);
 433                 exit_mmap(mm);
 434                 set_mm_exe_file(mm, NULL);
 435                 if (!list_empty(&mm->mmlist)) {
 436                         spin_lock(&mmlist_lock);
 437                         list_del(&mm->mmlist);
 438                         spin_unlock(&mmlist_lock);
 439                 }
 440                 put_swap_token(mm);
 441                 mmdrop(mm);
 442         }
 443 }
 444 EXPORT_SYMBOL_GPL(mmput);
 445
 446 /**
 447  * get_task_mm - acquire a reference to the task's mm
 448  *
 449  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
 450  * this kernel workthread has transiently adopted a user mm with use_mm,
 451  * to do its AIO) is not set and if so returns a reference to it, after
 452  * bumping up the use count.  User must release the mm via mmput()
 453  * after use.  Typically used by /proc and ptrace.
 454  */
 455 struct mm_struct *get_task_mm(struct task_struct *task)
 456 {
 457         struct mm_struct *mm;
 458
 459         task_lock(task);
 460         mm = task->mm;
 461         if (mm) {
 462                 if (task->flags & PF_BORROWED_MM)
 463                         mm = NULL;
 464                 else
 465                         atomic_inc(&mm->mm_users);
 466         }
 467         task_unlock(task);
 468         return mm;
 469 }
 470 EXPORT_SYMBOL_GPL(get_task_mm);
 471
 472 /* Please note the differences between mmput and mm_release.
 473  * mmput is called whenever we stop holding onto a mm_struct,
 474  * error success whatever.
 475  *
 476  * mm_release is called after a mm_struct has been removed
 477  * from the current process.
 478  *
 479  * This difference is important for error handling, when we
 480  * only half set up a mm_struct for a new process and need to restore
 481  * the old one.  Because we mmput the new mm_struct before
 482  * restoring the old one. . .
 483  * Eric Biederman 10 January 1998
 484  */
 485 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 486 {
 487         struct completion *vfork_done = tsk->vfork_done;
 488
 489         /* Get rid of any cached register state */
 490         deactivate_mm(tsk, mm);
 491
 492         /* notify parent sleeping on vfork() */
 493         if (vfork_done) {
 494                 tsk->vfork_done = NULL;
 495                 complete(vfork_done);
 496         }
 497
 498         /*
 499          * If we're exiting normally, clear a user-space tid field if
 500          * requested.  We leave this alone when dying by signal, to leave
 501          * the value intact in a core dump, and to save the unnecessary
 502          * trouble otherwise.  Userland only wants this done for a sys_exit.
 503          */
 504         if (tsk->clear_child_tid
 505             && !(tsk->flags & PF_SIGNALED)
 506             && atomic_read(&mm->mm_users) > 1) {
 507                 u32 __user * tidptr = tsk->clear_child_tid;
 508                 tsk->clear_child_tid = NULL;
 509
 510                 /*
 511                  * We don't check the error code - if userspace has
 512                  * not set up a proper pointer then tough luck.
 513                  */
 514                 put_user(0, tidptr);
 515                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
 516         }
 517 }
 518
 519 /*
 520  * Allocate a new mm structure and copy contents from the
 521  * mm structure of the passed in task structure.
 522  */
 523 struct mm_struct *dup_mm(struct task_struct *tsk)
 524 {
 525         struct mm_struct *mm, *oldmm = current->mm;
 526         int err;
 527
 528         if (!oldmm)
 529                 return NULL;
 530
 531         mm = allocate_mm();
 532         if (!mm)
 533                 goto fail_nomem;
 534
 535         memcpy(mm, oldmm, sizeof(*mm));
 536
 537         /* Initializing for Swap token stuff */
 538         mm->token_priority = 0;
 539         mm->last_interval = 0;
 540
 541         if (!mm_init(mm, tsk))
 542                 goto fail_nomem;
 543
 544         if (init_new_context(tsk, mm))
 545                 goto fail_nocontext;
 546
 547         dup_mm_exe_file(oldmm, mm);
 548
 549         err = dup_mmap(mm, oldmm);
 550         if (err)
 551                 goto free_pt;
 552
 553         mm->hiwater_rss = get_mm_rss(mm);
 554         mm->hiwater_vm = mm->total_vm;
 555
 556         return mm;
 557
 558 free_pt:
 559         mmput(mm);
 560
 561 fail_nomem:
 562         return NULL;
 563
 564 fail_nocontext:
 565         /*
 566          * If init_new_context() failed, we cannot use mmput() to free the mm
 567          * because it calls destroy_context()
 568          */
 569         mm_free_pgd(mm);
 570         free_mm(mm);
 571         return NULL;
 572 }
 573
 574 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 575 {
 576         struct mm_struct * mm, *oldmm;
 577         int retval;
 578
 579         tsk->min_flt = tsk->maj_flt = 0;
 580         tsk->nvcsw = tsk->nivcsw = 0;
 581
 582         tsk->mm = NULL;
 583         tsk->active_mm = NULL;
 584
 585         /*
 586          * Are we cloning a kernel thread?
 587          *
 588          * We need to steal a active VM for that..
 589          */
 590         oldmm = current->mm;
 591         if (!oldmm)
 592                 return 0;
 593
 594         if (clone_flags & CLONE_VM) {
 595                 atomic_inc(&oldmm->mm_users);
 596                 mm = oldmm;
 597                 goto good_mm;
 598         }
 599
 600         retval = -ENOMEM;
 601         mm = dup_mm(tsk);
 602         if (!mm)
 603                 goto fail_nomem;
 604
 605 good_mm:
 606         /* Initializing for Swap token stuff */
 607         mm->token_priority = 0;
 608         mm->last_interval = 0;
 609
 610         tsk->mm = mm;
 611         tsk->active_mm = mm;
 612         return 0;
 613
 614 fail_nomem:
 615         return retval;
 616 }
 617
 618 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
 619 {
 620         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
 621         /* We don't need to lock fs - think why ;-) */
 622         if (fs) {
 623                 atomic_set(&fs->count, 1);
 624                 rwlock_init(&fs->lock);
 625                 fs->umask = old->umask;
 626                 read_lock(&old->lock);
 627                 fs->root = old->root;
 628                 path_get(&old->root);
 629                 fs->pwd = old->pwd;
 630                 path_get(&old->pwd);
 631                 if (old->altroot.dentry) {
 632                         fs->altroot = old->altroot;
 633                         path_get(&old->altroot);
 634                 } else {
 635                         fs->altroot.mnt = NULL;
 636                         fs->altroot.dentry = NULL;
 637                 }
 638                 read_unlock(&old->lock);
 639         }
 640         return fs;
 641 }
 642
 643 struct fs_struct *copy_fs_struct(struct fs_struct *old)
 644 {
 645         return __copy_fs_struct(old);
 646 }
 647
 648 EXPORT_SYMBOL_GPL(copy_fs_struct);
 649
 650 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 651 {
 652         if (clone_flags & CLONE_FS) {
 653                 atomic_inc(&current->fs->count);
 654                 return 0;
 655         }
 656         tsk->fs = __copy_fs_struct(current->fs);
 657         if (!tsk->fs)
 658                 return -ENOMEM;
 659         return 0;
 660 }
 661
 662 static int count_open_files(struct fdtable *fdt)
 663 {
 664         int size = fdt->max_fds;
 665         int i;
 666
 667         /* Find the last open fd */
 668         for (i = size/(8*sizeof(long)); i > 0; ) {
 669                 if (fdt->open_fds->fds_bits[--i])
 670                         break;
 671         }
 672         i = (i+1) * 8 * sizeof(long);
 673         return i;
 674 }
 675
 676 static struct files_struct *alloc_files(void)
 677 {
 678         struct files_struct *newf;
 679         struct fdtable *fdt;
 680
 681         newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
 682         if (!newf)
 683                 goto out;
 684
 685         atomic_set(&newf->count, 1);
 686
 687         spin_lock_init(&newf->file_lock);
 688         newf->next_fd = 0;
 689         fdt = &newf->fdtab;
 690         fdt->max_fds = NR_OPEN_DEFAULT;
 691         fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
 692         fdt->open_fds = (fd_set *)&newf->open_fds_init;
 693         fdt->fd = &newf->fd_array[0];
 694         INIT_RCU_HEAD(&fdt->rcu);
 695         fdt->next = NULL;
 696         rcu_assign_pointer(newf->fdt, fdt);
 697 out:
 698         return newf;
 699 }
 700
 701 /*
 702  * Allocate a new files structure and copy contents from the
 703  * passed in files structure.
 704  * errorp will be valid only when the returned files_struct is NULL.
 705  */
 706 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
 707 {
 708         struct files_struct *newf;
 709         struct file **old_fds, **new_fds;
 710         int open_files, size, i;
 711         struct fdtable *old_fdt, *new_fdt;
 712
 713         *errorp = -ENOMEM;
 714         newf = alloc_files();
 715         if (!newf)
 716                 goto out;
 717
 718         spin_lock(&oldf->file_lock);
 719         old_fdt = files_fdtable(oldf);
 720         new_fdt = files_fdtable(newf);
 721         open_files = count_open_files(old_fdt);
 722
 723         /*
 724          * Check whether we need to allocate a larger fd array and fd set.
 725          * Note: we're not a clone task, so the open count won't change.
 726          */
 727         if (open_files > new_fdt->max_fds) {
 728                 new_fdt->max_fds = 0;
 729                 spin_unlock(&oldf->file_lock);
 730                 spin_lock(&newf->file_lock);
 731                 *errorp = expand_files(newf, open_files-1);
 732                 spin_unlock(&newf->file_lock);
 733                 if (*errorp < 0)
 734                         goto out_release;
 735                 new_fdt = files_fdtable(newf);
 736                 /*
 737                  * Reacquire the oldf lock and a pointer to its fd table
 738                  * who knows it may have a new bigger fd table. We need
 739                  * the latest pointer.
 740                  */
 741                 spin_lock(&oldf->file_lock);
 742                 old_fdt = files_fdtable(oldf);
 743         }
 744
 745         old_fds = old_fdt->fd;
 746         new_fds = new_fdt->fd;
 747
 748         memcpy(new_fdt->open_fds->fds_bits,
 749                 old_fdt->open_fds->fds_bits, open_files/8);
 750         memcpy(new_fdt->close_on_exec->fds_bits,
 751                 old_fdt->close_on_exec->fds_bits, open_files/8);
 752
 753         for (i = open_files; i != 0; i--) {
 754                 struct file *f = *old_fds++;
 755                 if (f) {
 756                         get_file(f);
 757                 } else {
 758                         /*
 759                          * The fd may be claimed in the fd bitmap but not yet
 760                          * instantiated in the files array if a sibling thread
 761                          * is partway through open().  So make sure that this
 762                          * fd is available to the new process.
 763                          */
 764                         FD_CLR(open_files - i, new_fdt->open_fds);
 765                 }
 766                 rcu_assign_pointer(*new_fds++, f);
 767         }
 768         spin_unlock(&oldf->file_lock);
 769
 770         /* compute the remainder to be cleared */
 771         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
 772
 773         /* This is long word aligned thus could use a optimized version */
 774         memset(new_fds, 0, size);
 775
 776         if (new_fdt->max_fds > open_files) {
 777                 int left = (new_fdt->max_fds-open_files)/8;
 778                 int start = open_files / (8 * sizeof(unsigned long));
 779
 780                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
 781                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
 782         }
 783
 784         return newf;
 785
 786 out_release:
 787         kmem_cache_free(files_cachep, newf);
 788 out:
 789         return NULL;
 790 }
 791
 792 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 793 {
 794         struct files_struct *oldf, *newf;
 795         int error = 0;
 796
 797         /*
 798          * A background process may not have any files ...
 799          */
 800         oldf = current->files;
 801         if (!oldf)
 802                 goto out;
 803
 804         if (clone_flags & CLONE_FILES) {
 805                 atomic_inc(&oldf->count);
 806                 goto out;
 807         }
 808
 809         newf = dup_fd(oldf, &error);
 810         if (!newf)
 811                 goto out;
 812
 813         tsk->files = newf;
 814         error = 0;
 815 out:
 816         return error;
 817 }
 818
 819 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 820 {
 821 #ifdef CONFIG_BLOCK
 822         struct io_context *ioc = current->io_context;
 823
 824         if (!ioc)
 825                 return 0;
 826         /*
 827          * Share io context with parent, if CLONE_IO is set
 828          */
 829         if (clone_flags & CLONE_IO) {
 830                 tsk->io_context = ioc_task_link(ioc);
 831                 if (unlikely(!tsk->io_context))
 832                         return -ENOMEM;
 833         } else if (ioprio_valid(ioc->ioprio)) {
 834                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 835                 if (unlikely(!tsk->io_context))
 836                         return -ENOMEM;
 837
 838                 tsk->io_context->ioprio = ioc->ioprio;
 839         }
 840 #endif
 841         return 0;
 842 }
 843
 844 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 845 {
 846         struct sighand_struct *sig;
 847
 848         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
 849                 atomic_inc(&current->sighand->count);
 850                 return 0;
 851         }
 852         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 853         rcu_assign_pointer(tsk->sighand, sig);
 854         if (!sig)
 855                 return -ENOMEM;
 856         atomic_set(&sig->count, 1);
 857         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 858         return 0;
 859 }
 860
 861 void __cleanup_sighand(struct sighand_struct *sighand)
 862 {
 863         if (atomic_dec_and_test(&sighand->count))
 864                 kmem_cache_free(sighand_cachep, sighand);
 865 }
 866
 867 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 868 {
 869         struct signal_struct *sig;
 870         int ret;
 871
 872         if (clone_flags & CLONE_THREAD) {
 873                 atomic_inc(&current->signal->count);
 874                 atomic_inc(&current->signal->live);
 875                 return 0;
 876         }
 877         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 878         tsk->signal = sig;
 879         if (!sig)
 880                 return -ENOMEM;
 881
 882         ret = copy_thread_group_keys(tsk);
 883         if (ret < 0) {
 884                 kmem_cache_free(signal_cachep, sig);
 885                 return ret;
 886         }
 887
 888         atomic_set(&sig->count, 1);
 889         atomic_set(&sig->live, 1);
 890         init_waitqueue_head(&sig->wait_chldexit);
 891         sig->flags = 0;
 892         sig->group_exit_code = 0;
 893         sig->group_exit_task = NULL;
 894         sig->group_stop_count = 0;
 895         sig->curr_target = tsk;
 896         init_sigpending(&sig->shared_pending);
 897         INIT_LIST_HEAD(&sig->posix_timers);
 898
 899         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 900         sig->it_real_incr.tv64 = 0;
 901         sig->real_timer.function = it_real_fn;
 902
 903         sig->it_virt_expires = cputime_zero;
 904         sig->it_virt_incr = cputime_zero;
 905         sig->it_prof_expires = cputime_zero;
 906         sig->it_prof_incr = cputime_zero;
 907
 908         sig->leader = 0;        /* session leadership doesn't inherit */
 909         sig->tty_old_pgrp = NULL;
 910
 911         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 912         sig->gtime = cputime_zero;
 913         sig->cgtime = cputime_zero;
 914         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 915         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 916         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
 917         sig->sum_sched_runtime = 0;
 918         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 919         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 920         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 921         taskstats_tgid_init(sig);
 922
 923         task_lock(current->group_leader);
 924         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 925         task_unlock(current->group_leader);
 926
 927         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 928                 /*
 929                  * New sole thread in the process gets an expiry time
 930                  * of the whole CPU time limit.
 931                  */
 932                 tsk->it_prof_expires =
 933                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 934         }
 935         acct_init_pacct(&sig->pacct);
 936
 937         tty_audit_fork(sig);
 938
 939         return 0;
 940 }
 941
 942 void __cleanup_signal(struct signal_struct *sig)
 943 {
 944         exit_thread_group_keys(sig);
 945         kmem_cache_free(signal_cachep, sig);
 946 }
 947
 948 static void cleanup_signal(struct task_struct *tsk)
 949 {
 950         struct signal_struct *sig = tsk->signal;
 951
 952         atomic_dec(&sig->live);
 953
 954         if (atomic_dec_and_test(&sig->count))
 955                 __cleanup_signal(sig);
 956 }
 957
 958 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 959 {
 960         unsigned long new_flags = p->flags;
 961
 962         new_flags &= ~PF_SUPERPRIV;
 963         new_flags |= PF_FORKNOEXEC;
 964         if (!(clone_flags & CLONE_PTRACE))
 965                 p->ptrace = 0;
 966         p->flags = new_flags;
 967         clear_freeze_flag(p);
 968 }
 969
 970 asmlinkage long sys_set_tid_address(int __user *tidptr)
 971 {
 972         current->clear_child_tid = tidptr;
 973
 974         return task_pid_vnr(current);
 975 }
 976
 977 static void rt_mutex_init_task(struct task_struct *p)
 978 {
 979         spin_lock_init(&p->pi_lock);
 980 #ifdef CONFIG_RT_MUTEXES
 981         plist_head_init(&p->pi_waiters, &p->pi_lock);
 982         p->pi_blocked_on = NULL;
 983 #endif
 984 }
 985
 986 #ifdef CONFIG_MM_OWNER
 987 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 988 {
 989         mm->owner = p;
 990 }
 991 #endif /* CONFIG_MM_OWNER */
 992
 993 /*
 994  * This creates a new process as a copy of the old one,
 995  * but does not actually start it yet.
 996  *
 997  * It copies the registers, and all the appropriate
 998  * parts of the process environment (as per the clone
 999  * flags). The actual kick-off is left to the caller.
1000  */
1001 static struct task_struct *copy_process(unsigned long clone_flags,
1002                                         unsigned long stack_start,
1003                                         struct pt_regs *regs,
1004                                         unsigned long stack_size,
1005                                         int __user *child_tidptr,
1006                                         struct pid *pid)
1007 {
1008         int retval;
1009         struct task_struct *p;
1010         int cgroup_callbacks_done = 0;
1011
1012         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1013                 return ERR_PTR(-EINVAL);
1014
1015         /*
1016          * Thread groups must share signals as well, and detached threads
1017          * can only be started up within the thread group.
1018          */
1019         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1020                 return ERR_PTR(-EINVAL);
1021
1022         /*
1023          * Shared signal handlers imply shared VM. By way of the above,
1024          * thread groups also imply shared VM. Blocking this case allows
1025          * for various simplifications in other code.
1026          */
1027         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1028                 return ERR_PTR(-EINVAL);
1029
1030         retval = security_task_create(clone_flags);
1031         if (retval)
1032                 goto fork_out;
1033
1034         retval = -ENOMEM;
1035         p = dup_task_struct(current);
1036         if (!p)
1037                 goto fork_out;
1038
1039         rt_mutex_init_task(p);
1040
1041 #ifdef CONFIG_TRACE_IRQFLAGS
1042         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1043         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1044 #endif
1045         retval = -EAGAIN;
1046         if (atomic_read(&p->user->processes) >=
1047                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1048                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1049                     p->user != current->nsproxy->user_ns->root_user)
1050                         goto bad_fork_free;
1051         }
1052
1053         atomic_inc(&p->user->__count);
1054         atomic_inc(&p->user->processes);
1055         get_group_info(p->group_info);
1056
1057         /*
1058          * If multiple threads are within copy_process(), then this check
1059          * triggers too late. This doesn't hurt, the check is only there
1060          * to stop root fork bombs.
1061          */
1062         if (nr_threads >= max_threads)
1063                 goto bad_fork_cleanup_count;
1064
1065         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1066                 goto bad_fork_cleanup_count;
1067
1068         if (p->binfmt && !try_module_get(p->binfmt->module))
1069                 goto bad_fork_cleanup_put_domain;
1070
1071         p->did_exec = 0;
1072         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1073         copy_flags(clone_flags, p);
1074         INIT_LIST_HEAD(&p->children);
1075         INIT_LIST_HEAD(&p->sibling);
1076 #ifdef CONFIG_PREEMPT_RCU
1077         p->rcu_read_lock_nesting = 0;
1078         p->rcu_flipctr_idx = 0;
1079 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1080         p->vfork_done = NULL;
1081         spin_lock_init(&p->alloc_lock);
1082
1083         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1084         init_sigpending(&p->pending);
1085
1086         p->utime = cputime_zero;
1087         p->stime = cputime_zero;
1088         p->gtime = cputime_zero;
1089         p->utimescaled = cputime_zero;
1090         p->stimescaled = cputime_zero;
1091         p->prev_utime = cputime_zero;
1092         p->prev_stime = cputime_zero;
1093
1094 #ifdef CONFIG_DETECT_SOFTLOCKUP
1095         p->last_switch_count = 0;
1096         p->last_switch_timestamp = 0;
1097 #endif
1098
1099 #ifdef CONFIG_TASK_XACCT
1100         p->rchar = 0;           /* I/O counter: bytes read */
1101         p->wchar = 0;           /* I/O counter: bytes written */
1102         p->syscr = 0;           /* I/O counter: read syscalls */
1103         p->syscw = 0;           /* I/O counter: write syscalls */
1104 #endif
1105         task_io_accounting_init(p);
1106         acct_clear_integrals(p);
1107
1108         p->it_virt_expires = cputime_zero;
1109         p->it_prof_expires = cputime_zero;
1110         p->it_sched_expires = 0;
1111         INIT_LIST_HEAD(&p->cpu_timers[0]);
1112         INIT_LIST_HEAD(&p->cpu_timers[1]);
1113         INIT_LIST_HEAD(&p->cpu_timers[2]);
1114
1115         p->lock_depth = -1;             /* -1 = no lock */
1116         do_posix_clock_monotonic_gettime(&p->start_time);
1117         p->real_start_time = p->start_time;
1118         monotonic_to_bootbased(&p->real_start_time);
1119 #ifdef CONFIG_SECURITY
1120         p->security = NULL;
1121 #endif
1122         p->cap_bset = current->cap_bset;
1123         p->io_context = NULL;
1124         p->audit_context = NULL;
1125         cgroup_fork(p);
1126 #ifdef CONFIG_NUMA
1127         p->mempolicy = mpol_dup(p->mempolicy);
1128         if (IS_ERR(p->mempolicy)) {
1129                 retval = PTR_ERR(p->mempolicy);
1130                 p->mempolicy = NULL;
1131                 goto bad_fork_cleanup_cgroup;
1132         }
1133         mpol_fix_fork_child_flag(p);
1134 #endif
1135 #ifdef CONFIG_TRACE_IRQFLAGS
1136         p->irq_events = 0;
1137 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1138         p->hardirqs_enabled = 1;
1139 #else
1140         p->hardirqs_enabled = 0;
1141 #endif
1142         p->hardirq_enable_ip = 0;
1143         p->hardirq_enable_event = 0;
1144         p->hardirq_disable_ip = _THIS_IP_;
1145         p->hardirq_disable_event = 0;
1146         p->softirqs_enabled = 1;
1147         p->softirq_enable_ip = _THIS_IP_;
1148         p->softirq_enable_event = 0;
1149         p->softirq_disable_ip = 0;
1150         p->softirq_disable_event = 0;
1151         p->hardirq_context = 0;
1152         p->softirq_context = 0;
1153 #endif
1154 #ifdef CONFIG_LOCKDEP
1155         p->lockdep_depth = 0; /* no locks held yet */
1156         p->curr_chain_key = 0;
1157         p->lockdep_recursion = 0;
1158 #endif
1159
1160 #ifdef CONFIG_DEBUG_MUTEXES
1161         p->blocked_on = NULL; /* not blocked yet */
1162 #endif
1163
1164         /* Perform scheduler related setup. Assign this task to a CPU. */
1165         sched_fork(p, clone_flags);
1166
1167         if ((retval = security_task_alloc(p)))
1168                 goto bad_fork_cleanup_policy;
1169         if ((retval = audit_alloc(p)))
1170                 goto bad_fork_cleanup_security;
1171         /* copy all the process information */
1172         if ((retval = copy_semundo(clone_flags, p)))
1173                 goto bad_fork_cleanup_audit;
1174         if ((retval = copy_files(clone_flags, p)))
1175                 goto bad_fork_cleanup_semundo;
1176         if ((retval = copy_fs(clone_flags, p)))
1177                 goto bad_fork_cleanup_files;
1178         if ((retval = copy_sighand(clone_flags, p)))
1179                 goto bad_fork_cleanup_fs;
1180         if ((retval = copy_signal(clone_flags, p)))
1181                 goto bad_fork_cleanup_sighand;
1182         if ((retval = copy_mm(clone_flags, p)))
1183                 goto bad_fork_cleanup_signal;
1184         if ((retval = copy_keys(clone_flags, p)))
1185                 goto bad_fork_cleanup_mm;
1186         if ((retval = copy_namespaces(clone_flags, p)))
1187                 goto bad_fork_cleanup_keys;
1188         if ((retval = copy_io(clone_flags, p)))
1189                 goto bad_fork_cleanup_namespaces;
1190         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1191         if (retval)
1192                 goto bad_fork_cleanup_io;
1193
1194         if (pid != &init_struct_pid) {
1195                 retval = -ENOMEM;
1196                 pid = alloc_pid(task_active_pid_ns(p));
1197                 if (!pid)
1198                         goto bad_fork_cleanup_io;
1199
1200                 if (clone_flags & CLONE_NEWPID) {
1201                         retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1202                         if (retval < 0)
1203                                 goto bad_fork_free_pid;
1204                 }
1205         }
1206
1207         p->pid = pid_nr(pid);
1208         p->tgid = p->pid;
1209         if (clone_flags & CLONE_THREAD)
1210                 p->tgid = current->tgid;
1211
1212         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1213         /*
1214          * Clear TID on mm_release()?
1215          */
1216         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1217 #ifdef CONFIG_FUTEX
1218         p->robust_list = NULL;
1219 #ifdef CONFIG_COMPAT
1220         p->compat_robust_list = NULL;
1221 #endif
1222         INIT_LIST_HEAD(&p->pi_state_list);
1223         p->pi_state_cache = NULL;
1224 #endif
1225         /*
1226          * sigaltstack should be cleared when sharing the same VM
1227          */
1228         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1229                 p->sas_ss_sp = p->sas_ss_size = 0;
1230
1231         /*
1232          * Syscall tracing should be turned off in the child regardless
1233          * of CLONE_PTRACE.
1234          */
1235         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1236 #ifdef TIF_SYSCALL_EMU
1237         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1238 #endif
1239         clear_all_latency_tracing(p);
1240
1241         /* Our parent execution domain becomes current domain
1242            These must match for thread signalling to apply */
1243         p->parent_exec_id = p->self_exec_id;
1244
1245         /* ok, now we should be set up.. */
1246         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1247         p->pdeath_signal = 0;
1248         p->exit_state = 0;
1249
1250         /*
1251          * Ok, make it visible to the rest of the system.
1252          * We dont wake it up yet.
1253          */
1254         p->group_leader = p;
1255         INIT_LIST_HEAD(&p->thread_group);
1256         INIT_LIST_HEAD(&p->ptrace_children);
1257         INIT_LIST_HEAD(&p->ptrace_list);
1258
1259         /* Now that the task is set up, run cgroup callbacks if
1260          * necessary. We need to run them before the task is visible
1261          * on the tasklist. */
1262         cgroup_fork_callbacks(p);
1263         cgroup_callbacks_done = 1;
1264
1265         /* Need tasklist lock for parent etc handling! */
1266         write_lock_irq(&tasklist_lock);
1267
1268         /*
1269          * The task hasn't been attached yet, so its cpus_allowed mask will
1270          * not be changed, nor will its assigned CPU.
1271          *
1272          * The cpus_allowed mask of the parent may have changed after it was
1273          * copied first time - so re-copy it here, then check the child's CPU
1274          * to ensure it is on a valid CPU (and if not, just force it back to
1275          * parent's CPU). This avoids alot of nasty races.
1276          */
1277         p->cpus_allowed = current->cpus_allowed;
1278         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1279         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1280                         !cpu_online(task_cpu(p))))
1281                 set_task_cpu(p, smp_processor_id());
1282
1283         /* CLONE_PARENT re-uses the old parent */
1284         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1285                 p->real_parent = current->real_parent;
1286         else
1287                 p->real_parent = current;
1288         p->parent = p->real_parent;
1289
1290         spin_lock(&current->sighand->siglock);
1291
1292         /*
1293          * Process group and session signals need to be delivered to just the
1294          * parent before the fork or both the parent and the child after the
1295          * fork. Restart if a signal comes in before we add the new process to
1296          * it's process group.
1297          * A fatal signal pending means that current will exit, so the new
1298          * thread can't slip out of an OOM kill (or normal SIGKILL).
1299          */
1300         recalc_sigpending();
1301         if (signal_pending(current)) {
1302                 spin_unlock(&current->sighand->siglock);
1303                 write_unlock_irq(&tasklist_lock);
1304                 retval = -ERESTARTNOINTR;
1305                 goto bad_fork_free_pid;
1306         }
1307
1308         if (clone_flags & CLONE_THREAD) {
1309                 p->group_leader = current->group_leader;
1310                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1311
1312                 if (!cputime_eq(current->signal->it_virt_expires,
1313                                 cputime_zero) ||
1314                     !cputime_eq(current->signal->it_prof_expires,
1315                                 cputime_zero) ||
1316                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1317                     !list_empty(&current->signal->cpu_timers[0]) ||
1318                     !list_empty(&current->signal->cpu_timers[1]) ||
1319                     !list_empty(&current->signal->cpu_timers[2])) {
1320                         /*
1321                          * Have child wake up on its first tick to check
1322                          * for process CPU timers.
1323                          */
1324                         p->it_prof_expires = jiffies_to_cputime(1);
1325                 }
1326         }
1327
1328         if (likely(p->pid)) {
1329                 add_parent(p);
1330                 if (unlikely(p->ptrace & PT_PTRACED))
1331                         __ptrace_link(p, current->parent);
1332
1333                 if (thread_group_leader(p)) {
1334                         if (clone_flags & CLONE_NEWPID)
1335                                 p->nsproxy->pid_ns->child_reaper = p;
1336
1337                         p->signal->leader_pid = pid;
1338                         p->signal->tty = current->signal->tty;
1339                         set_task_pgrp(p, task_pgrp_nr(current));
1340                         set_task_session(p, task_session_nr(current));
1341                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1342                         attach_pid(p, PIDTYPE_SID, task_session(current));
1343                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1344                         __get_cpu_var(process_counts)++;
1345                 }
1346                 attach_pid(p, PIDTYPE_PID, pid);
1347                 nr_threads++;
1348         }
1349
1350         total_forks++;
1351         spin_unlock(&current->sighand->siglock);
1352         write_unlock_irq(&tasklist_lock);
1353         proc_fork_connector(p);
1354         cgroup_post_fork(p);
1355         return p;
1356
1357 bad_fork_free_pid:
1358         if (pid != &init_struct_pid)
1359                 free_pid(pid);
1360 bad_fork_cleanup_io:
1361         put_io_context(p->io_context);
1362 bad_fork_cleanup_namespaces:
1363         exit_task_namespaces(p);
1364 bad_fork_cleanup_keys:
1365         exit_keys(p);
1366 bad_fork_cleanup_mm:
1367         if (p->mm)
1368                 mmput(p->mm);
1369 bad_fork_cleanup_signal:
1370         cleanup_signal(p);
1371 bad_fork_cleanup_sighand:
1372         __cleanup_sighand(p->sighand);
1373 bad_fork_cleanup_fs:
1374         exit_fs(p); /* blocking */
1375 bad_fork_cleanup_files:
1376         exit_files(p); /* blocking */
1377 bad_fork_cleanup_semundo:
1378         exit_sem(p);
1379 bad_fork_cleanup_audit:
1380         audit_free(p);
1381 bad_fork_cleanup_security:
1382         security_task_free(p);
1383 bad_fork_cleanup_policy:
1384 #ifdef CONFIG_NUMA
1385         mpol_put(p->mempolicy);
1386 bad_fork_cleanup_cgroup:
1387 #endif
1388         cgroup_exit(p, cgroup_callbacks_done);
1389         delayacct_tsk_free(p);
1390         if (p->binfmt)
1391                 module_put(p->binfmt->module);
1392 bad_fork_cleanup_put_domain:
1393         module_put(task_thread_info(p)->exec_domain->module);
1394 bad_fork_cleanup_count:
1395         put_group_info(p->group_info);
1396         atomic_dec(&p->user->processes);
1397         free_uid(p->user);
1398 bad_fork_free:
1399         free_task(p);
1400 fork_out:
1401         return ERR_PTR(retval);
1402 }
1403
1404 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1405 {
1406         memset(regs, 0, sizeof(struct pt_regs));
1407         return regs;
1408 }
1409
1410 struct task_struct * __cpuinit fork_idle(int cpu)
1411 {
1412         struct task_struct *task;
1413         struct pt_regs regs;
1414
1415         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1416                                 &init_struct_pid);
1417         if (!IS_ERR(task))
1418                 init_idle(task, cpu);
1419
1420         return task;
1421 }
1422
1423 static int fork_traceflag(unsigned clone_flags)
1424 {
1425         if (clone_flags & CLONE_UNTRACED)
1426                 return 0;
1427         else if (clone_flags & CLONE_VFORK) {
1428                 if (current->ptrace & PT_TRACE_VFORK)
1429                         return PTRACE_EVENT_VFORK;
1430         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1431                 if (current->ptrace & PT_TRACE_CLONE)
1432                         return PTRACE_EVENT_CLONE;
1433         } else if (current->ptrace & PT_TRACE_FORK)
1434                 return PTRACE_EVENT_FORK;
1435
1436         return 0;
1437 }
1438
1439 /*
1440  *  Ok, this is the main fork-routine.
1441  *
1442  * It copies the process, and if successful kick-starts
1443  * it and waits for it to finish using the VM if required.
1444  */
1445 long do_fork(unsigned long clone_flags,
1446               unsigned long stack_start,
1447               struct pt_regs *regs,
1448               unsigned long stack_size,
1449               int __user *parent_tidptr,
1450               int __user *child_tidptr)
1451 {
1452         struct task_struct *p;
1453         int trace = 0;
1454         long nr;
1455
1456         /*
1457          * We hope to recycle these flags after 2.6.26
1458          */
1459         if (unlikely(clone_flags & CLONE_STOPPED)) {
1460                 static int __read_mostly count = 100;
1461
1462                 if (count > 0 && printk_ratelimit()) {
1463                         char comm[TASK_COMM_LEN];
1464
1465                         count--;
1466                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1467                                         "clone flags 0x%lx\n",
1468                                 get_task_comm(comm, current),
1469                                 clone_flags & CLONE_STOPPED);
1470                 }
1471         }
1472
1473         if (unlikely(current->ptrace)) {
1474                 trace = fork_traceflag (clone_flags);
1475                 if (trace)
1476                         clone_flags |= CLONE_PTRACE;
1477         }
1478
1479         p = copy_process(clone_flags, stack_start, regs, stack_size,
1480                         child_tidptr, NULL);
1481         /*
1482          * Do this prior waking up the new thread - the thread pointer
1483          * might get invalid after that point, if the thread exits quickly.
1484          */
1485         if (!IS_ERR(p)) {
1486                 struct completion vfork;
1487
1488                 nr = task_pid_vnr(p);
1489
1490                 if (clone_flags & CLONE_PARENT_SETTID)
1491                         put_user(nr, parent_tidptr);
1492
1493                 if (clone_flags & CLONE_VFORK) {
1494                         p->vfork_done = &vfork;
1495                         init_completion(&vfork);
1496                 }
1497
1498                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1499                         /*
1500                          * We'll start up with an immediate SIGSTOP.
1501                          */
1502                         sigaddset(&p->pending.signal, SIGSTOP);
1503                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1504                 }
1505
1506                 if (!(clone_flags & CLONE_STOPPED))
1507                         wake_up_new_task(p, clone_flags);
1508                 else
1509                         __set_task_state(p, TASK_STOPPED);
1510
1511                 if (unlikely (trace)) {
1512                         current->ptrace_message = nr;
1513                         ptrace_notify ((trace << 8) | SIGTRAP);
1514                 }
1515
1516                 if (clone_flags & CLONE_VFORK) {
1517                         freezer_do_not_count();
1518                         wait_for_completion(&vfork);
1519                         freezer_count();
1520                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1521                                 current->ptrace_message = nr;
1522                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1523                         }
1524                 }
1525         } else {
1526                 nr = PTR_ERR(p);
1527         }
1528         return nr;
1529 }
1530
1531 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1532 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1533 #endif
1534
1535 static void sighand_ctor(struct kmem_cache *cachep, void *data)
1536 {
1537         struct sighand_struct *sighand = data;
1538
1539         spin_lock_init(&sighand->siglock);
1540         init_waitqueue_head(&sighand->signalfd_wqh);
1541 }
1542
1543 void __init proc_caches_init(void)
1544 {
1545         sighand_cachep = kmem_cache_create("sighand_cache",
1546                         sizeof(struct sighand_struct), 0,
1547                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1548                         sighand_ctor);
1549         signal_cachep = kmem_cache_create("signal_cache",
1550                         sizeof(struct signal_struct), 0,
1551                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1552         files_cachep = kmem_cache_create("files_cache",
1553                         sizeof(struct files_struct), 0,
1554                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1555         fs_cachep = kmem_cache_create("fs_cache",
1556                         sizeof(struct fs_struct), 0,
1557                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1558         vm_area_cachep = kmem_cache_create("vm_area_struct",
1559                         sizeof(struct vm_area_struct), 0,
1560                         SLAB_PANIC, NULL);
1561         mm_cachep = kmem_cache_create("mm_struct",
1562                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1563                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1564 }
1565
1566 /*
1567  * Check constraints on flags passed to the unshare system call and
1568  * force unsharing of additional process context as appropriate.
1569  */
1570 static void check_unshare_flags(unsigned long *flags_ptr)
1571 {
1572         /*
1573          * If unsharing a thread from a thread group, must also
1574          * unshare vm.
1575          */
1576         if (*flags_ptr & CLONE_THREAD)
1577                 *flags_ptr |= CLONE_VM;
1578
1579         /*
1580          * If unsharing vm, must also unshare signal handlers.
1581          */
1582         if (*flags_ptr & CLONE_VM)
1583                 *flags_ptr |= CLONE_SIGHAND;
1584
1585         /*
1586          * If unsharing signal handlers and the task was created
1587          * using CLONE_THREAD, then must unshare the thread
1588          */
1589         if ((*flags_ptr & CLONE_SIGHAND) &&
1590             (atomic_read(&current->signal->count) > 1))
1591                 *flags_ptr |= CLONE_THREAD;
1592
1593         /*
1594          * If unsharing namespace, must also unshare filesystem information.
1595          */
1596         if (*flags_ptr & CLONE_NEWNS)
1597                 *flags_ptr |= CLONE_FS;
1598 }
1599
1600 /*
1601  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1602  */
1603 static int unshare_thread(unsigned long unshare_flags)
1604 {
1605         if (unshare_flags & CLONE_THREAD)
1606                 return -EINVAL;
1607
1608         return 0;
1609 }
1610
1611 /*
1612  * Unshare the filesystem structure if it is being shared
1613  */
1614 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1615 {
1616         struct fs_struct *fs = current->fs;
1617
1618         if ((unshare_flags & CLONE_FS) &&
1619             (fs && atomic_read(&fs->count) > 1)) {
1620                 *new_fsp = __copy_fs_struct(current->fs);
1621                 if (!*new_fsp)
1622                         return -ENOMEM;
1623         }
1624
1625         return 0;
1626 }
1627
1628 /*
1629  * Unsharing of sighand is not supported yet
1630  */
1631 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1632 {
1633         struct sighand_struct *sigh = current->sighand;
1634
1635         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1636                 return -EINVAL;
1637         else
1638                 return 0;
1639 }
1640
1641 /*
1642  * Unshare vm if it is being shared
1643  */
1644 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1645 {
1646         struct mm_struct *mm = current->mm;
1647
1648         if ((unshare_flags & CLONE_VM) &&
1649             (mm && atomic_read(&mm->mm_users) > 1)) {
1650                 return -EINVAL;
1651         }
1652
1653         return 0;
1654 }
1655
1656 /*
1657  * Unshare file descriptor table if it is being shared
1658  */
1659 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1660 {
1661         struct files_struct *fd = current->files;
1662         int error = 0;
1663
1664         if ((unshare_flags & CLONE_FILES) &&
1665             (fd && atomic_read(&fd->count) > 1)) {
1666                 *new_fdp = dup_fd(fd, &error);
1667                 if (!*new_fdp)
1668                         return error;
1669         }
1670
1671         return 0;
1672 }
1673
1674 /*
1675  * unshare allows a process to 'unshare' part of the process
1676  * context which was originally shared using clone.  copy_*
1677  * functions used by do_fork() cannot be used here directly
1678  * because they modify an inactive task_struct that is being
1679  * constructed. Here we are modifying the current, active,
1680  * task_struct.
1681  */
1682 asmlinkage long sys_unshare(unsigned long unshare_flags)
1683 {
1684         int err = 0;
1685         struct fs_struct *fs, *new_fs = NULL;
1686         struct sighand_struct *new_sigh = NULL;
1687         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1688         struct files_struct *fd, *new_fd = NULL;
1689         struct nsproxy *new_nsproxy = NULL;
1690         int do_sysvsem = 0;
1691
1692         check_unshare_flags(&unshare_flags);
1693
1694         /* Return -EINVAL for all unsupported flags */
1695         err = -EINVAL;
1696         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1697                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1698                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
1699                                 CLONE_NEWNET))
1700                 goto bad_unshare_out;
1701
1702         /*
1703          * CLONE_NEWIPC must also detach from the undolist: after switching
1704          * to a new ipc namespace, the semaphore arrays from the old
1705          * namespace are unreachable.
1706          */
1707         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1708                 do_sysvsem = 1;
1709         if ((err = unshare_thread(unshare_flags)))
1710                 goto bad_unshare_out;
1711         if ((err = unshare_fs(unshare_flags, &new_fs)))
1712                 goto bad_unshare_cleanup_thread;
1713         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1714                 goto bad_unshare_cleanup_fs;
1715         if ((err = unshare_vm(unshare_flags, &new_mm)))
1716                 goto bad_unshare_cleanup_sigh;
1717         if ((err = unshare_fd(unshare_flags, &new_fd)))
1718                 goto bad_unshare_cleanup_vm;
1719         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1720                         new_fs)))
1721                 goto bad_unshare_cleanup_fd;
1722
1723         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1724                 if (do_sysvsem) {
1725                         /*
1726                          * CLONE_SYSVSEM is equivalent to sys_exit().
1727                          */
1728                         exit_sem(current);
1729                 }
1730
1731                 if (new_nsproxy) {
1732                         switch_task_namespaces(current, new_nsproxy);
1733                         new_nsproxy = NULL;
1734                 }
1735
1736                 task_lock(current);
1737
1738                 if (new_fs) {
1739                         fs = current->fs;
1740                         current->fs = new_fs;
1741                         new_fs = fs;
1742                 }
1743
1744                 if (new_mm) {
1745                         mm = current->mm;
1746                         active_mm = current->active_mm;
1747                         current->mm = new_mm;
1748                         current->active_mm = new_mm;
1749                         activate_mm(active_mm, new_mm);
1750                         new_mm = mm;
1751                 }
1752
1753                 if (new_fd) {
1754                         fd = current->files;
1755                         current->files = new_fd;
1756                         new_fd = fd;
1757                 }
1758
1759                 task_unlock(current);
1760         }
1761
1762         if (new_nsproxy)
1763                 put_nsproxy(new_nsproxy);
1764
1765 bad_unshare_cleanup_fd:
1766         if (new_fd)
1767                 put_files_struct(new_fd);
1768
1769 bad_unshare_cleanup_vm:
1770         if (new_mm)
1771                 mmput(new_mm);
1772
1773 bad_unshare_cleanup_sigh:
1774         if (new_sigh)
1775                 if (atomic_dec_and_test(&new_sigh->count))
1776                         kmem_cache_free(sighand_cachep, new_sigh);
1777
1778 bad_unshare_cleanup_fs:
1779         if (new_fs)
1780                 put_fs_struct(new_fs);
1781
1782 bad_unshare_cleanup_thread:
1783 bad_unshare_out:
1784         return err;
1785 }
1786
1787 /*
1788  *      Helper to unshare the files of the current task.
1789  *      We don't want to expose copy_files internals to
1790  *      the exec layer of the kernel.
1791  */
1792
1793 int unshare_files(struct files_struct **displaced)
1794 {
1795         struct task_struct *task = current;
1796         struct files_struct *copy = NULL;
1797         int error;
1798
1799         error = unshare_fd(CLONE_FILES, &copy);
1800         if (error || !copy) {
1801                 *displaced = NULL;
1802                 return error;
1803         }
1804         *displaced = task->files;
1805         task_lock(task);
1806         task->files = copy;
1807         task_unlock(task);
1808         return 0;
1809 }