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/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/nsproxy.h>
  32 #include <linux/capability.h>
  33 #include <linux/cpu.h>
  34 #include <linux/cgroup.h>
  35 #include <linux/security.h>
  36 #include <linux/hugetlb.h>
  37 #include <linux/swap.h>
  38 #include <linux/syscalls.h>
  39 #include <linux/jiffies.h>
  40 #include <linux/tracehook.h>
  41 #include <linux/futex.h>
  42 #include <linux/compat.h>
  43 #include <linux/task_io_accounting_ops.h>
  44 #include <linux/rcupdate.h>
  45 #include <linux/ptrace.h>
  46 #include <linux/mount.h>
  47 #include <linux/audit.h>
  48 #include <linux/memcontrol.h>
  49 #include <linux/ftrace.h>
  50 #include <linux/profile.h>
  51 #include <linux/rmap.h>
  52 #include <linux/ksm.h>
  53 #include <linux/acct.h>
  54 #include <linux/tsacct_kern.h>
  55 #include <linux/cn_proc.h>
  56 #include <linux/freezer.h>
  57 #include <linux/delayacct.h>
  58 #include <linux/taskstats_kern.h>
  59 #include <linux/random.h>
  60 #include <linux/tty.h>
  61 #include <linux/proc_fs.h>
  62 #include <linux/blkdev.h>
  63 #include <linux/fs_struct.h>
  64 #include <linux/magic.h>
  65 #include <linux/perf_event.h>
  66 #include <linux/posix-timers.h>
  67
  68 #include <asm/pgtable.h>
  69 #include <asm/pgalloc.h>
  70 #include <asm/uaccess.h>
  71 #include <asm/mmu_context.h>
  72 #include <asm/cacheflush.h>
  73 #include <asm/tlbflush.h>
  74
  75 #include <trace/events/sched.h>
  76
  77 /*
  78  * Protected counters by write_lock_irq(&tasklist_lock)
  79  */
  80 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  81 int nr_threads;                 /* The idle threads do not count.. */
  82
  83 int max_threads;                /* tunable limit on nr_threads */
  84
  85 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  86
  87 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  88
  89 int nr_processes(void)
  90 {
  91         int cpu;
  92         int total = 0;
  93
  94         for_each_possible_cpu(cpu)
  95                 total += per_cpu(process_counts, cpu);
  96
  97         return total;
  98 }
  99
 100 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 101 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
 102 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
 103 static struct kmem_cache *task_struct_cachep;
 104 #endif
 105
 106 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 107 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
 108 {
 109 #ifdef CONFIG_DEBUG_STACK_USAGE
 110         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 111 #else
 112         gfp_t mask = GFP_KERNEL;
 113 #endif
 114         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
 115 }
 116
 117 static inline void free_thread_info(struct thread_info *ti)
 118 {
 119         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 120 }
 121 #endif
 122
 123 /* SLAB cache for signal_struct structures (tsk->signal) */
 124 static struct kmem_cache *signal_cachep;
 125
 126 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 127 struct kmem_cache *sighand_cachep;
 128
 129 /* SLAB cache for files_struct structures (tsk->files) */
 130 struct kmem_cache *files_cachep;
 131
 132 /* SLAB cache for fs_struct structures (tsk->fs) */
 133 struct kmem_cache *fs_cachep;
 134
 135 /* SLAB cache for vm_area_struct structures */
 136 struct kmem_cache *vm_area_cachep;
 137
 138 /* SLAB cache for mm_struct structures (tsk->mm) */
 139 static struct kmem_cache *mm_cachep;
 140
 141 /* Notifier list called when a task struct is freed */
 142 static ATOMIC_NOTIFIER_HEAD(task_free_notifier);
 143
 144 static void account_kernel_stack(struct thread_info *ti, int account)
 145 {
 146         struct zone *zone = page_zone(virt_to_page(ti));
 147
 148         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 149 }
 150
 151 void free_task(struct task_struct *tsk)
 152 {
 153         prop_local_destroy_single(&tsk->dirties);
 154         account_kernel_stack(tsk->stack, -1);
 155         free_thread_info(tsk->stack);
 156         rt_mutex_debug_task_free(tsk);
 157         ftrace_graph_exit_task(tsk);
 158         free_task_struct(tsk);
 159 }
 160 EXPORT_SYMBOL(free_task);
 161
 162 int task_free_register(struct notifier_block *n)
 163 {
 164         return atomic_notifier_chain_register(&task_free_notifier, n);
 165 }
 166 EXPORT_SYMBOL(task_free_register);
 167
 168 int task_free_unregister(struct notifier_block *n)
 169 {
 170         return atomic_notifier_chain_unregister(&task_free_notifier, n);
 171 }
 172 EXPORT_SYMBOL(task_free_unregister);
 173
 174 void __put_task_struct(struct task_struct *tsk)
 175 {
 176         WARN_ON(!tsk->exit_state);
 177         WARN_ON(atomic_read(&tsk->usage));
 178         WARN_ON(tsk == current);
 179
 180         exit_creds(tsk);
 181         delayacct_tsk_free(tsk);
 182
 183         atomic_notifier_call_chain(&task_free_notifier, 0, tsk);
 184         if (!profile_handoff_task(tsk))
 185                 free_task(tsk);
 186 }
 187
 188 /*
 189  * macro override instead of weak attribute alias, to workaround
 190  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 191  */
 192 #ifndef arch_task_cache_init
 193 #define arch_task_cache_init()
 194 #endif
 195
 196 void __init fork_init(unsigned long mempages)
 197 {
 198 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 199 #ifndef ARCH_MIN_TASKALIGN
 200 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 201 #endif
 202         /* create a slab on which task_structs can be allocated */
 203         task_struct_cachep =
 204                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 205                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 206 #endif
 207
 208         /* do the arch specific task caches init */
 209         arch_task_cache_init();
 210
 211         /*
 212          * The default maximum number of threads is set to a safe
 213          * value: the thread structures can take up at most half
 214          * of memory.
 215          */
 216         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 217
 218         /*
 219          * we need to allow at least 20 threads to boot a system
 220          */
 221         if(max_threads < 20)
 222                 max_threads = 20;
 223
 224         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 225         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 226         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 227                 init_task.signal->rlim[RLIMIT_NPROC];
 228 }
 229
 230 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 231                                                struct task_struct *src)
 232 {
 233         *dst = *src;
 234         return 0;
 235 }
 236
 237 static struct task_struct *dup_task_struct(struct task_struct *orig)
 238 {
 239         struct task_struct *tsk;
 240         struct thread_info *ti;
 241         unsigned long *stackend;
 242
 243         int err;
 244
 245         prepare_to_copy(orig);
 246
 247         tsk = alloc_task_struct();
 248         if (!tsk)
 249                 return NULL;
 250
 251         ti = alloc_thread_info(tsk);
 252         if (!ti) {
 253                 free_task_struct(tsk);
 254                 return NULL;
 255         }
 256
 257         err = arch_dup_task_struct(tsk, orig);
 258         if (err)
 259                 goto out;
 260
 261         tsk->stack = ti;
 262
 263         err = prop_local_init_single(&tsk->dirties);
 264         if (err)
 265                 goto out;
 266
 267         setup_thread_stack(tsk, orig);
 268         stackend = end_of_stack(tsk);
 269         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 270
 271 #ifdef CONFIG_CC_STACKPROTECTOR
 272         tsk->stack_canary = get_random_int();
 273 #endif
 274
 275         /* One for us, one for whoever does the "release_task()" (usually parent) */
 276         atomic_set(&tsk->usage,2);
 277         atomic_set(&tsk->fs_excl, 0);
 278 #ifdef CONFIG_BLK_DEV_IO_TRACE
 279         tsk->btrace_seq = 0;
 280 #endif
 281         tsk->splice_pipe = NULL;
 282
 283         account_kernel_stack(ti, 1);
 284
 285         return tsk;
 286
 287 out:
 288         free_thread_info(ti);
 289         free_task_struct(tsk);
 290         return NULL;
 291 }
 292
 293 #ifdef CONFIG_MMU
 294 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 295 {
 296         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 297         struct rb_node **rb_link, *rb_parent;
 298         int retval;
 299         unsigned long charge;
 300         struct mempolicy *pol;
 301
 302         down_write(&oldmm->mmap_sem);
 303         flush_cache_dup_mm(oldmm);
 304         /*
 305          * Not linked in yet - no deadlock potential:
 306          */
 307         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 308
 309         mm->locked_vm = 0;
 310         mm->mmap = NULL;
 311         mm->mmap_cache = NULL;
 312         mm->free_area_cache = oldmm->mmap_base;
 313         mm->cached_hole_size = ~0UL;
 314         mm->map_count = 0;
 315         cpumask_clear(mm_cpumask(mm));
 316         mm->mm_rb = RB_ROOT;
 317         rb_link = &mm->mm_rb.rb_node;
 318         rb_parent = NULL;
 319         pprev = &mm->mmap;
 320         retval = ksm_fork(mm, oldmm);
 321         if (retval)
 322                 goto out;
 323
 324         prev = NULL;
 325         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 326                 struct file *file;
 327
 328                 if (mpnt->vm_flags & VM_DONTCOPY) {
 329                         long pages = vma_pages(mpnt);
 330                         mm->total_vm -= pages;
 331                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 332                                                                 -pages);
 333                         continue;
 334                 }
 335                 charge = 0;
 336                 if (mpnt->vm_flags & VM_ACCOUNT) {
 337                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 338                         if (security_vm_enough_memory(len))
 339                                 goto fail_nomem;
 340                         charge = len;
 341                 }
 342                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 343                 if (!tmp)
 344                         goto fail_nomem;
 345                 *tmp = *mpnt;
 346                 pol = mpol_dup(vma_policy(mpnt));
 347                 retval = PTR_ERR(pol);
 348                 if (IS_ERR(pol))
 349                         goto fail_nomem_policy;
 350                 vma_set_policy(tmp, pol);
 351                 tmp->vm_flags &= ~VM_LOCKED;
 352                 tmp->vm_mm = mm;
 353                 tmp->vm_next = tmp->vm_prev = NULL;
 354                 anon_vma_link(tmp);
 355                 file = tmp->vm_file;
 356                 if (file) {
 357                         struct inode *inode = file->f_path.dentry->d_inode;
 358                         struct address_space *mapping = file->f_mapping;
 359
 360                         get_file(file);
 361                         if (tmp->vm_flags & VM_DENYWRITE)
 362                                 atomic_dec(&inode->i_writecount);
 363                         spin_lock(&mapping->i_mmap_lock);
 364                         if (tmp->vm_flags & VM_SHARED)
 365                                 mapping->i_mmap_writable++;
 366                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 367                         flush_dcache_mmap_lock(mapping);
 368                         /* insert tmp into the share list, just after mpnt */
 369                         vma_prio_tree_add(tmp, mpnt);
 370                         flush_dcache_mmap_unlock(mapping);
 371                         spin_unlock(&mapping->i_mmap_lock);
 372                 }
 373
 374                 /*
 375                  * Clear hugetlb-related page reserves for children. This only
 376                  * affects MAP_PRIVATE mappings. Faults generated by the child
 377                  * are not guaranteed to succeed, even if read-only
 378                  */
 379                 if (is_vm_hugetlb_page(tmp))
 380                         reset_vma_resv_huge_pages(tmp);
 381
 382                 /*
 383                  * Link in the new vma and copy the page table entries.
 384                  */
 385                 *pprev = tmp;
 386                 pprev = &tmp->vm_next;
 387                 tmp->vm_prev = prev;
 388                 prev = tmp;
 389
 390                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 391                 rb_link = &tmp->vm_rb.rb_right;
 392                 rb_parent = &tmp->vm_rb;
 393
 394                 mm->map_count++;
 395                 retval = copy_page_range(mm, oldmm, mpnt);
 396
 397                 if (tmp->vm_ops && tmp->vm_ops->open)
 398                         tmp->vm_ops->open(tmp);
 399
 400                 if (retval)
 401                         goto out;
 402         }
 403         /* a new mm has just been created */
 404         arch_dup_mmap(oldmm, mm);
 405         retval = 0;
 406 out:
 407         up_write(&mm->mmap_sem);
 408         flush_tlb_mm(oldmm);
 409         up_write(&oldmm->mmap_sem);
 410         return retval;
 411 fail_nomem_policy:
 412         kmem_cache_free(vm_area_cachep, tmp);
 413 fail_nomem:
 414         retval = -ENOMEM;
 415         vm_unacct_memory(charge);
 416         goto out;
 417 }
 418
 419 static inline int mm_alloc_pgd(struct mm_struct * mm)
 420 {
 421         mm->pgd = pgd_alloc(mm);
 422         if (unlikely(!mm->pgd))
 423                 return -ENOMEM;
 424         return 0;
 425 }
 426
 427 static inline void mm_free_pgd(struct mm_struct * mm)
 428 {
 429         pgd_free(mm, mm->pgd);
 430 }
 431 #else
 432 #define dup_mmap(mm, oldmm)     (0)
 433 #define mm_alloc_pgd(mm)        (0)
 434 #define mm_free_pgd(mm)
 435 #endif /* CONFIG_MMU */
 436
 437 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 438
 439 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 440 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 441
 442 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 443
 444 static int __init coredump_filter_setup(char *s)
 445 {
 446         default_dump_filter =
 447                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 448                 MMF_DUMP_FILTER_MASK;
 449         return 1;
 450 }
 451
 452 __setup("coredump_filter=", coredump_filter_setup);
 453
 454 #include <linux/init_task.h>
 455
 456 static void mm_init_aio(struct mm_struct *mm)
 457 {
 458 #ifdef CONFIG_AIO
 459         spin_lock_init(&mm->ioctx_lock);
 460         INIT_HLIST_HEAD(&mm->ioctx_list);
 461 #endif
 462 }
 463
 464 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
 465 {
 466         atomic_set(&mm->mm_users, 1);
 467         atomic_set(&mm->mm_count, 1);
 468         init_rwsem(&mm->mmap_sem);
 469         INIT_LIST_HEAD(&mm->mmlist);
 470         mm->flags = (current->mm) ?
 471                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 472         mm->core_state = NULL;
 473         mm->nr_ptes = 0;
 474         set_mm_counter(mm, file_rss, 0);
 475         set_mm_counter(mm, anon_rss, 0);
 476         spin_lock_init(&mm->page_table_lock);
 477         mm->free_area_cache = TASK_UNMAPPED_BASE;
 478         mm->cached_hole_size = ~0UL;
 479         mm_init_aio(mm);
 480         mm_init_owner(mm, p);
 481
 482         if (likely(!mm_alloc_pgd(mm))) {
 483                 mm->def_flags = 0;
 484                 mmu_notifier_mm_init(mm);
 485                 return mm;
 486         }
 487
 488         free_mm(mm);
 489         return NULL;
 490 }
 491
 492 /*
 493  * Allocate and initialize an mm_struct.
 494  */
 495 struct mm_struct * mm_alloc(void)
 496 {
 497         struct mm_struct * mm;
 498
 499         mm = allocate_mm();
 500         if (mm) {
 501                 memset(mm, 0, sizeof(*mm));
 502                 mm = mm_init(mm, current);
 503         }
 504         return mm;
 505 }
 506
 507 /*
 508  * Called when the last reference to the mm
 509  * is dropped: either by a lazy thread or by
 510  * mmput. Free the page directory and the mm.
 511  */
 512 void __mmdrop(struct mm_struct *mm)
 513 {
 514         BUG_ON(mm == &init_mm);
 515         mm_free_pgd(mm);
 516         destroy_context(mm);
 517         mmu_notifier_mm_destroy(mm);
 518         free_mm(mm);
 519 }
 520 EXPORT_SYMBOL_GPL(__mmdrop);
 521
 522 /*
 523  * Decrement the use count and release all resources for an mm.
 524  */
 525 void mmput(struct mm_struct *mm)
 526 {
 527         might_sleep();
 528
 529         if (atomic_dec_and_test(&mm->mm_users)) {
 530                 exit_aio(mm);
 531                 ksm_exit(mm);
 532                 exit_mmap(mm);
 533                 set_mm_exe_file(mm, NULL);
 534                 if (!list_empty(&mm->mmlist)) {
 535                         spin_lock(&mmlist_lock);
 536                         list_del(&mm->mmlist);
 537                         spin_unlock(&mmlist_lock);
 538                 }
 539                 put_swap_token(mm);
 540                 if (mm->binfmt)
 541                         module_put(mm->binfmt->module);
 542                 mmdrop(mm);
 543         }
 544 }
 545 EXPORT_SYMBOL_GPL(mmput);
 546
 547 /**
 548  * get_task_mm - acquire a reference to the task's mm
 549  *
 550  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 551  * this kernel workthread has transiently adopted a user mm with use_mm,
 552  * to do its AIO) is not set and if so returns a reference to it, after
 553  * bumping up the use count.  User must release the mm via mmput()
 554  * after use.  Typically used by /proc and ptrace.
 555  */
 556 struct mm_struct *get_task_mm(struct task_struct *task)
 557 {
 558         struct mm_struct *mm;
 559
 560         task_lock(task);
 561         mm = task->mm;
 562         if (mm) {
 563                 if (task->flags & PF_KTHREAD)
 564                         mm = NULL;
 565                 else
 566                         atomic_inc(&mm->mm_users);
 567         }
 568         task_unlock(task);
 569         return mm;
 570 }
 571 EXPORT_SYMBOL_GPL(get_task_mm);
 572
 573 /* Please note the differences between mmput and mm_release.
 574  * mmput is called whenever we stop holding onto a mm_struct,
 575  * error success whatever.
 576  *
 577  * mm_release is called after a mm_struct has been removed
 578  * from the current process.
 579  *
 580  * This difference is important for error handling, when we
 581  * only half set up a mm_struct for a new process and need to restore
 582  * the old one.  Because we mmput the new mm_struct before
 583  * restoring the old one. . .
 584  * Eric Biederman 10 January 1998
 585  */
 586 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 587 {
 588         struct completion *vfork_done = tsk->vfork_done;
 589
 590         /* Get rid of any futexes when releasing the mm */
 591 #ifdef CONFIG_FUTEX
 592         if (unlikely(tsk->robust_list)) {
 593                 exit_robust_list(tsk);
 594                 tsk->robust_list = NULL;
 595         }
 596 #ifdef CONFIG_COMPAT
 597         if (unlikely(tsk->compat_robust_list)) {
 598                 compat_exit_robust_list(tsk);
 599                 tsk->compat_robust_list = NULL;
 600         }
 601 #endif
 602         if (unlikely(!list_empty(&tsk->pi_state_list)))
 603                 exit_pi_state_list(tsk);
 604 #endif
 605
 606         /* Get rid of any cached register state */
 607         deactivate_mm(tsk, mm);
 608
 609         /* notify parent sleeping on vfork() */
 610         if (vfork_done) {
 611                 tsk->vfork_done = NULL;
 612                 complete(vfork_done);
 613         }
 614
 615         /*
 616          * If we're exiting normally, clear a user-space tid field if
 617          * requested.  We leave this alone when dying by signal, to leave
 618          * the value intact in a core dump, and to save the unnecessary
 619          * trouble otherwise.  Userland only wants this done for a sys_exit.
 620          */
 621         if (tsk->clear_child_tid) {
 622                 if (!(tsk->flags & PF_SIGNALED) &&
 623                     atomic_read(&mm->mm_users) > 1) {
 624                         /*
 625                          * We don't check the error code - if userspace has
 626                          * not set up a proper pointer then tough luck.
 627                          */
 628                         put_user(0, tsk->clear_child_tid);
 629                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 630                                         1, NULL, NULL, 0);
 631                 }
 632                 tsk->clear_child_tid = NULL;
 633         }
 634 }
 635
 636 /*
 637  * Allocate a new mm structure and copy contents from the
 638  * mm structure of the passed in task structure.
 639  */
 640 struct mm_struct *dup_mm(struct task_struct *tsk)
 641 {
 642         struct mm_struct *mm, *oldmm = current->mm;
 643         int err;
 644
 645         if (!oldmm)
 646                 return NULL;
 647
 648         mm = allocate_mm();
 649         if (!mm)
 650                 goto fail_nomem;
 651
 652         memcpy(mm, oldmm, sizeof(*mm));
 653
 654         /* Initializing for Swap token stuff */
 655         mm->token_priority = 0;
 656         mm->last_interval = 0;
 657
 658         if (!mm_init(mm, tsk))
 659                 goto fail_nomem;
 660
 661         if (init_new_context(tsk, mm))
 662                 goto fail_nocontext;
 663
 664         dup_mm_exe_file(oldmm, mm);
 665
 666         err = dup_mmap(mm, oldmm);
 667         if (err)
 668                 goto free_pt;
 669
 670         mm->hiwater_rss = get_mm_rss(mm);
 671         mm->hiwater_vm = mm->total_vm;
 672
 673         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 674                 goto free_pt;
 675
 676         return mm;
 677
 678 free_pt:
 679         /* don't put binfmt in mmput, we haven't got module yet */
 680         mm->binfmt = NULL;
 681         mmput(mm);
 682
 683 fail_nomem:
 684         return NULL;
 685
 686 fail_nocontext:
 687         /*
 688          * If init_new_context() failed, we cannot use mmput() to free the mm
 689          * because it calls destroy_context()
 690          */
 691         mm_free_pgd(mm);
 692         free_mm(mm);
 693         return NULL;
 694 }
 695
 696 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 697 {
 698         struct mm_struct * mm, *oldmm;
 699         int retval;
 700
 701         tsk->min_flt = tsk->maj_flt = 0;
 702         tsk->nvcsw = tsk->nivcsw = 0;
 703 #ifdef CONFIG_DETECT_HUNG_TASK
 704         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 705 #endif
 706
 707         tsk->mm = NULL;
 708         tsk->active_mm = NULL;
 709
 710         /*
 711          * Are we cloning a kernel thread?
 712          *
 713          * We need to steal a active VM for that..
 714          */
 715         oldmm = current->mm;
 716         if (!oldmm)
 717                 return 0;
 718
 719         if (clone_flags & CLONE_VM) {
 720                 atomic_inc(&oldmm->mm_users);
 721                 mm = oldmm;
 722                 goto good_mm;
 723         }
 724
 725         retval = -ENOMEM;
 726         mm = dup_mm(tsk);
 727         if (!mm)
 728                 goto fail_nomem;
 729
 730 good_mm:
 731         /* Initializing for Swap token stuff */
 732         mm->token_priority = 0;
 733         mm->last_interval = 0;
 734
 735         tsk->mm = mm;
 736         tsk->active_mm = mm;
 737         return 0;
 738
 739 fail_nomem:
 740         return retval;
 741 }
 742
 743 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 744 {
 745         struct fs_struct *fs = current->fs;
 746         if (clone_flags & CLONE_FS) {
 747                 /* tsk->fs is already what we want */
 748                 write_lock(&fs->lock);
 749                 if (fs->in_exec) {
 750                         write_unlock(&fs->lock);
 751                         return -EAGAIN;
 752                 }
 753                 fs->users++;
 754                 write_unlock(&fs->lock);
 755                 return 0;
 756         }
 757         tsk->fs = copy_fs_struct(fs);
 758         if (!tsk->fs)
 759                 return -ENOMEM;
 760         return 0;
 761 }
 762
 763 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 764 {
 765         struct files_struct *oldf, *newf;
 766         int error = 0;
 767
 768         /*
 769          * A background process may not have any files ...
 770          */
 771         oldf = current->files;
 772         if (!oldf)
 773                 goto out;
 774
 775         if (clone_flags & CLONE_FILES) {
 776                 atomic_inc(&oldf->count);
 777                 goto out;
 778         }
 779
 780         newf = dup_fd(oldf, &error);
 781         if (!newf)
 782                 goto out;
 783
 784         tsk->files = newf;
 785         error = 0;
 786 out:
 787         return error;
 788 }
 789
 790 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 791 {
 792 #ifdef CONFIG_BLOCK
 793         struct io_context *ioc = current->io_context;
 794
 795         if (!ioc)
 796                 return 0;
 797         /*
 798          * Share io context with parent, if CLONE_IO is set
 799          */
 800         if (clone_flags & CLONE_IO) {
 801                 tsk->io_context = ioc_task_link(ioc);
 802                 if (unlikely(!tsk->io_context))
 803                         return -ENOMEM;
 804         } else if (ioprio_valid(ioc->ioprio)) {
 805                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 806                 if (unlikely(!tsk->io_context))
 807                         return -ENOMEM;
 808
 809                 tsk->io_context->ioprio = ioc->ioprio;
 810         }
 811 #endif
 812         return 0;
 813 }
 814
 815 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 816 {
 817         struct sighand_struct *sig;
 818
 819         if (clone_flags & CLONE_SIGHAND) {
 820                 atomic_inc(&current->sighand->count);
 821                 return 0;
 822         }
 823         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 824         rcu_assign_pointer(tsk->sighand, sig);
 825         if (!sig)
 826                 return -ENOMEM;
 827         atomic_set(&sig->count, 1);
 828         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 829         return 0;
 830 }
 831
 832 void __cleanup_sighand(struct sighand_struct *sighand)
 833 {
 834         if (atomic_dec_and_test(&sighand->count))
 835                 kmem_cache_free(sighand_cachep, sighand);
 836 }
 837
 838
 839 /*
 840  * Initialize POSIX timer handling for a thread group.
 841  */
 842 static void posix_cpu_timers_init_group(struct signal_struct *sig)
 843 {
 844         /* Thread group counters. */
 845         thread_group_cputime_init(sig);
 846
 847         /* Expiration times and increments. */
 848         sig->it[CPUCLOCK_PROF].expires = cputime_zero;
 849         sig->it[CPUCLOCK_PROF].incr = cputime_zero;
 850         sig->it[CPUCLOCK_VIRT].expires = cputime_zero;
 851         sig->it[CPUCLOCK_VIRT].incr = cputime_zero;
 852
 853         /* Cached expiration times. */
 854         sig->cputime_expires.prof_exp = cputime_zero;
 855         sig->cputime_expires.virt_exp = cputime_zero;
 856         sig->cputime_expires.sched_exp = 0;
 857
 858         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
 859                 sig->cputime_expires.prof_exp =
 860                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
 861                 sig->cputimer.running = 1;
 862         }
 863
 864         /* The timer lists. */
 865         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 866         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 867         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 868 }
 869
 870 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 871 {
 872         struct signal_struct *sig;
 873
 874         if (clone_flags & CLONE_THREAD)
 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         atomic_set(&sig->count, 1);
 883         atomic_set(&sig->live, 1);
 884         init_waitqueue_head(&sig->wait_chldexit);
 885         sig->flags = 0;
 886         if (clone_flags & CLONE_NEWPID)
 887                 sig->flags |= SIGNAL_UNKILLABLE;
 888         sig->group_exit_code = 0;
 889         sig->group_exit_task = NULL;
 890         sig->group_stop_count = 0;
 891         sig->curr_target = tsk;
 892         init_sigpending(&sig->shared_pending);
 893         INIT_LIST_HEAD(&sig->posix_timers);
 894
 895         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 896         sig->it_real_incr.tv64 = 0;
 897         sig->real_timer.function = it_real_fn;
 898
 899         sig->leader = 0;        /* session leadership doesn't inherit */
 900         sig->tty_old_pgrp = NULL;
 901         sig->tty = NULL;
 902
 903         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
 904         sig->gtime = cputime_zero;
 905         sig->cgtime = cputime_zero;
 906 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
 907         sig->prev_utime = sig->prev_stime = cputime_zero;
 908 #endif
 909         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
 910         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
 911         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
 912         sig->maxrss = sig->cmaxrss = 0;
 913         task_io_accounting_init(&sig->ioac);
 914         sig->sum_sched_runtime = 0;
 915         taskstats_tgid_init(sig);
 916
 917         task_lock(current->group_leader);
 918         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 919         task_unlock(current->group_leader);
 920
 921         posix_cpu_timers_init_group(sig);
 922
 923         acct_init_pacct(&sig->pacct);
 924
 925         tty_audit_fork(sig);
 926
 927         sig->oom_adj = current->signal->oom_adj;
 928
 929         return 0;
 930 }
 931
 932 void __cleanup_signal(struct signal_struct *sig)
 933 {
 934         thread_group_cputime_free(sig);
 935         tty_kref_put(sig->tty);
 936         kmem_cache_free(signal_cachep, sig);
 937 }
 938
 939 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 940 {
 941         unsigned long new_flags = p->flags;
 942
 943         new_flags &= ~PF_SUPERPRIV;
 944         new_flags |= PF_FORKNOEXEC;
 945         new_flags |= PF_STARTING;
 946         p->flags = new_flags;
 947         clear_freeze_flag(p);
 948 }
 949
 950 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
 951 {
 952         current->clear_child_tid = tidptr;
 953
 954         return task_pid_vnr(current);
 955 }
 956
 957 static void rt_mutex_init_task(struct task_struct *p)
 958 {
 959         spin_lock_init(&p->pi_lock);
 960 #ifdef CONFIG_RT_MUTEXES
 961         plist_head_init(&p->pi_waiters, &p->pi_lock);
 962         p->pi_blocked_on = NULL;
 963 #endif
 964 }
 965
 966 #ifdef CONFIG_MM_OWNER
 967 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 968 {
 969         mm->owner = p;
 970 }
 971 #endif /* CONFIG_MM_OWNER */
 972
 973 /*
 974  * Initialize POSIX timer handling for a single task.
 975  */
 976 static void posix_cpu_timers_init(struct task_struct *tsk)
 977 {
 978         tsk->cputime_expires.prof_exp = cputime_zero;
 979         tsk->cputime_expires.virt_exp = cputime_zero;
 980         tsk->cputime_expires.sched_exp = 0;
 981         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
 982         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
 983         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
 984 }
 985
 986 /*
 987  * This creates a new process as a copy of the old one,
 988  * but does not actually start it yet.
 989  *
 990  * It copies the registers, and all the appropriate
 991  * parts of the process environment (as per the clone
 992  * flags). The actual kick-off is left to the caller.
 993  */
 994 static struct task_struct *copy_process(unsigned long clone_flags,
 995                                         unsigned long stack_start,
 996                                         struct pt_regs *regs,
 997                                         unsigned long stack_size,
 998                                         int __user *child_tidptr,
 999                                         struct pid *pid,
1000                                         int trace)
1001 {
1002         int retval;
1003         struct task_struct *p;
1004         int cgroup_callbacks_done = 0;
1005
1006         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1007                 return ERR_PTR(-EINVAL);
1008
1009         /*
1010          * Thread groups must share signals as well, and detached threads
1011          * can only be started up within the thread group.
1012          */
1013         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1014                 return ERR_PTR(-EINVAL);
1015
1016         /*
1017          * Shared signal handlers imply shared VM. By way of the above,
1018          * thread groups also imply shared VM. Blocking this case allows
1019          * for various simplifications in other code.
1020          */
1021         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1022                 return ERR_PTR(-EINVAL);
1023
1024         /*
1025          * Siblings of global init remain as zombies on exit since they are
1026          * not reaped by their parent (swapper). To solve this and to avoid
1027          * multi-rooted process trees, prevent global and container-inits
1028          * from creating siblings.
1029          */
1030         if ((clone_flags & CLONE_PARENT) &&
1031                                 current->signal->flags & SIGNAL_UNKILLABLE)
1032                 return ERR_PTR(-EINVAL);
1033
1034         retval = security_task_create(clone_flags);
1035         if (retval)
1036                 goto fork_out;
1037
1038         retval = -ENOMEM;
1039         p = dup_task_struct(current);
1040         if (!p)
1041                 goto fork_out;
1042
1043         ftrace_graph_init_task(p);
1044
1045         rt_mutex_init_task(p);
1046
1047 #ifdef CONFIG_PROVE_LOCKING
1048         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1049         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1050 #endif
1051         retval = -EAGAIN;
1052         if (atomic_read(&p->real_cred->user->processes) >=
1053                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1054                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1055                     p->real_cred->user != INIT_USER)
1056                         goto bad_fork_free;
1057         }
1058
1059         retval = copy_creds(p, clone_flags);
1060         if (retval < 0)
1061                 goto bad_fork_free;
1062
1063         /*
1064          * If multiple threads are within copy_process(), then this check
1065          * triggers too late. This doesn't hurt, the check is only there
1066          * to stop root fork bombs.
1067          */
1068         retval = -EAGAIN;
1069         if (nr_threads >= max_threads)
1070                 goto bad_fork_cleanup_count;
1071
1072         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1073                 goto bad_fork_cleanup_count;
1074
1075         p->did_exec = 0;
1076         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1077         copy_flags(clone_flags, p);
1078         INIT_LIST_HEAD(&p->children);
1079         INIT_LIST_HEAD(&p->sibling);
1080         rcu_copy_process(p);
1081         p->vfork_done = NULL;
1082         spin_lock_init(&p->alloc_lock);
1083
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         p->default_timer_slack_ns = current->timer_slack_ns;
1095
1096         task_io_accounting_init(&p->ioac);
1097         acct_clear_integrals(p);
1098
1099         posix_cpu_timers_init(p);
1100
1101         p->lock_depth = -1;             /* -1 = no lock */
1102         do_posix_clock_monotonic_gettime(&p->start_time);
1103         p->real_start_time = p->start_time;
1104         monotonic_to_bootbased(&p->real_start_time);
1105         p->io_context = NULL;
1106         p->audit_context = NULL;
1107         cgroup_fork(p);
1108 #ifdef CONFIG_NUMA
1109         p->mempolicy = mpol_dup(p->mempolicy);
1110         if (IS_ERR(p->mempolicy)) {
1111                 retval = PTR_ERR(p->mempolicy);
1112                 p->mempolicy = NULL;
1113                 goto bad_fork_cleanup_cgroup;
1114         }
1115         mpol_fix_fork_child_flag(p);
1116 #endif
1117 #ifdef CONFIG_TRACE_IRQFLAGS
1118         p->irq_events = 0;
1119 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1120         p->hardirqs_enabled = 1;
1121 #else
1122         p->hardirqs_enabled = 0;
1123 #endif
1124         p->hardirq_enable_ip = 0;
1125         p->hardirq_enable_event = 0;
1126         p->hardirq_disable_ip = _THIS_IP_;
1127         p->hardirq_disable_event = 0;
1128         p->softirqs_enabled = 1;
1129         p->softirq_enable_ip = _THIS_IP_;
1130         p->softirq_enable_event = 0;
1131         p->softirq_disable_ip = 0;
1132         p->softirq_disable_event = 0;
1133         p->hardirq_context = 0;
1134         p->softirq_context = 0;
1135 #endif
1136 #ifdef CONFIG_LOCKDEP
1137         p->lockdep_depth = 0; /* no locks held yet */
1138         p->curr_chain_key = 0;
1139         p->lockdep_recursion = 0;
1140 #endif
1141
1142 #ifdef CONFIG_DEBUG_MUTEXES
1143         p->blocked_on = NULL; /* not blocked yet */
1144 #endif
1145
1146         p->bts = NULL;
1147
1148         /* Perform scheduler related setup. Assign this task to a CPU. */
1149         sched_fork(p, clone_flags);
1150
1151         retval = perf_event_init_task(p);
1152         if (retval)
1153                 goto bad_fork_cleanup_policy;
1154
1155         if ((retval = audit_alloc(p)))
1156                 goto bad_fork_cleanup_policy;
1157         /* copy all the process information */
1158         if ((retval = copy_semundo(clone_flags, p)))
1159                 goto bad_fork_cleanup_audit;
1160         if ((retval = copy_files(clone_flags, p)))
1161                 goto bad_fork_cleanup_semundo;
1162         if ((retval = copy_fs(clone_flags, p)))
1163                 goto bad_fork_cleanup_files;
1164         if ((retval = copy_sighand(clone_flags, p)))
1165                 goto bad_fork_cleanup_fs;
1166         if ((retval = copy_signal(clone_flags, p)))
1167                 goto bad_fork_cleanup_sighand;
1168         if ((retval = copy_mm(clone_flags, p)))
1169                 goto bad_fork_cleanup_signal;
1170         if ((retval = copy_namespaces(clone_flags, p)))
1171                 goto bad_fork_cleanup_mm;
1172         if ((retval = copy_io(clone_flags, p)))
1173                 goto bad_fork_cleanup_namespaces;
1174         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1175         if (retval)
1176                 goto bad_fork_cleanup_io;
1177
1178         if (pid != &init_struct_pid) {
1179                 retval = -ENOMEM;
1180                 pid = alloc_pid(p->nsproxy->pid_ns);
1181                 if (!pid)
1182                         goto bad_fork_cleanup_io;
1183
1184                 if (clone_flags & CLONE_NEWPID) {
1185                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1186                         if (retval < 0)
1187                                 goto bad_fork_free_pid;
1188                 }
1189         }
1190
1191         p->pid = pid_nr(pid);
1192         p->tgid = p->pid;
1193         if (clone_flags & CLONE_THREAD)
1194                 p->tgid = current->tgid;
1195
1196         if (current->nsproxy != p->nsproxy) {
1197                 retval = ns_cgroup_clone(p, pid);
1198                 if (retval)
1199                         goto bad_fork_free_pid;
1200         }
1201
1202         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1203         /*
1204          * Clear TID on mm_release()?
1205          */
1206         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1207 #ifdef CONFIG_FUTEX
1208         p->robust_list = NULL;
1209 #ifdef CONFIG_COMPAT
1210         p->compat_robust_list = NULL;
1211 #endif
1212         INIT_LIST_HEAD(&p->pi_state_list);
1213         p->pi_state_cache = NULL;
1214 #endif
1215         /*
1216          * sigaltstack should be cleared when sharing the same VM
1217          */
1218         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1219                 p->sas_ss_sp = p->sas_ss_size = 0;
1220
1221         /*
1222          * Syscall tracing should be turned off in the child regardless
1223          * of CLONE_PTRACE.
1224          */
1225         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1226 #ifdef TIF_SYSCALL_EMU
1227         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1228 #endif
1229         clear_all_latency_tracing(p);
1230
1231         /* ok, now we should be set up.. */
1232         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1233         p->pdeath_signal = 0;
1234         p->exit_state = 0;
1235
1236         /*
1237          * Ok, make it visible to the rest of the system.
1238          * We dont wake it up yet.
1239          */
1240         p->group_leader = p;
1241         INIT_LIST_HEAD(&p->thread_group);
1242
1243         /* Now that the task is set up, run cgroup callbacks if
1244          * necessary. We need to run them before the task is visible
1245          * on the tasklist. */
1246         cgroup_fork_callbacks(p);
1247         cgroup_callbacks_done = 1;
1248
1249         /* Need tasklist lock for parent etc handling! */
1250         write_lock_irq(&tasklist_lock);
1251
1252         /* CLONE_PARENT re-uses the old parent */
1253         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1254                 p->real_parent = current->real_parent;
1255                 p->parent_exec_id = current->parent_exec_id;
1256         } else {
1257                 p->real_parent = current;
1258                 p->parent_exec_id = current->self_exec_id;
1259         }
1260
1261         spin_lock(&current->sighand->siglock);
1262
1263         /*
1264          * Process group and session signals need to be delivered to just the
1265          * parent before the fork or both the parent and the child after the
1266          * fork. Restart if a signal comes in before we add the new process to
1267          * it's process group.
1268          * A fatal signal pending means that current will exit, so the new
1269          * thread can't slip out of an OOM kill (or normal SIGKILL).
1270          */
1271         recalc_sigpending();
1272         if (signal_pending(current)) {
1273                 spin_unlock(&current->sighand->siglock);
1274                 write_unlock_irq(&tasklist_lock);
1275                 retval = -ERESTARTNOINTR;
1276                 goto bad_fork_free_pid;
1277         }
1278
1279         if (clone_flags & CLONE_THREAD) {
1280                 atomic_inc(&current->signal->count);
1281                 atomic_inc(&current->signal->live);
1282                 p->group_leader = current->group_leader;
1283                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1284         }
1285
1286         if (likely(p->pid)) {
1287                 list_add_tail(&p->sibling, &p->real_parent->children);
1288                 tracehook_finish_clone(p, clone_flags, trace);
1289
1290                 if (thread_group_leader(p)) {
1291                         if (clone_flags & CLONE_NEWPID)
1292                                 p->nsproxy->pid_ns->child_reaper = p;
1293
1294                         p->signal->leader_pid = pid;
1295                         tty_kref_put(p->signal->tty);
1296                         p->signal->tty = tty_kref_get(current->signal->tty);
1297                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1298                         attach_pid(p, PIDTYPE_SID, task_session(current));
1299                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1300                         __get_cpu_var(process_counts)++;
1301                 }
1302                 attach_pid(p, PIDTYPE_PID, pid);
1303                 nr_threads++;
1304         }
1305
1306         total_forks++;
1307         spin_unlock(&current->sighand->siglock);
1308         write_unlock_irq(&tasklist_lock);
1309         proc_fork_connector(p);
1310         cgroup_post_fork(p);
1311         perf_event_fork(p);
1312         return p;
1313
1314 bad_fork_free_pid:
1315         if (pid != &init_struct_pid)
1316                 free_pid(pid);
1317 bad_fork_cleanup_io:
1318         put_io_context(p->io_context);
1319 bad_fork_cleanup_namespaces:
1320         exit_task_namespaces(p);
1321 bad_fork_cleanup_mm:
1322         if (p->mm)
1323                 mmput(p->mm);
1324 bad_fork_cleanup_signal:
1325         if (!(clone_flags & CLONE_THREAD))
1326                 __cleanup_signal(p->signal);
1327 bad_fork_cleanup_sighand:
1328         __cleanup_sighand(p->sighand);
1329 bad_fork_cleanup_fs:
1330         exit_fs(p); /* blocking */
1331 bad_fork_cleanup_files:
1332         exit_files(p); /* blocking */
1333 bad_fork_cleanup_semundo:
1334         exit_sem(p);
1335 bad_fork_cleanup_audit:
1336         audit_free(p);
1337 bad_fork_cleanup_policy:
1338         perf_event_free_task(p);
1339 #ifdef CONFIG_NUMA
1340         mpol_put(p->mempolicy);
1341 bad_fork_cleanup_cgroup:
1342 #endif
1343         cgroup_exit(p, cgroup_callbacks_done);
1344         delayacct_tsk_free(p);
1345         module_put(task_thread_info(p)->exec_domain->module);
1346 bad_fork_cleanup_count:
1347         atomic_dec(&p->cred->user->processes);
1348         exit_creds(p);
1349 bad_fork_free:
1350         free_task(p);
1351 fork_out:
1352         return ERR_PTR(retval);
1353 }
1354
1355 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1356 {
1357         memset(regs, 0, sizeof(struct pt_regs));
1358         return regs;
1359 }
1360
1361 struct task_struct * __cpuinit fork_idle(int cpu)
1362 {
1363         struct task_struct *task;
1364         struct pt_regs regs;
1365
1366         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1367                             &init_struct_pid, 0);
1368         if (!IS_ERR(task))
1369                 init_idle(task, cpu);
1370
1371         return task;
1372 }
1373
1374 /*
1375  *  Ok, this is the main fork-routine.
1376  *
1377  * It copies the process, and if successful kick-starts
1378  * it and waits for it to finish using the VM if required.
1379  */
1380 long do_fork(unsigned long clone_flags,
1381               unsigned long stack_start,
1382               struct pt_regs *regs,
1383               unsigned long stack_size,
1384               int __user *parent_tidptr,
1385               int __user *child_tidptr)
1386 {
1387         struct task_struct *p;
1388         int trace = 0;
1389         long nr;
1390
1391         /*
1392          * Do some preliminary argument and permissions checking before we
1393          * actually start allocating stuff
1394          */
1395         if (clone_flags & CLONE_NEWUSER) {
1396                 if (clone_flags & CLONE_THREAD)
1397                         return -EINVAL;
1398                 /* hopefully this check will go away when userns support is
1399                  * complete
1400                  */
1401                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1402                                 !capable(CAP_SETGID))
1403                         return -EPERM;
1404         }
1405
1406         /*
1407          * We hope to recycle these flags after 2.6.26
1408          */
1409         if (unlikely(clone_flags & CLONE_STOPPED)) {
1410                 static int __read_mostly count = 100;
1411
1412                 if (count > 0 && printk_ratelimit()) {
1413                         char comm[TASK_COMM_LEN];
1414
1415                         count--;
1416                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1417                                         "clone flags 0x%lx\n",
1418                                 get_task_comm(comm, current),
1419                                 clone_flags & CLONE_STOPPED);
1420                 }
1421         }
1422
1423         /*
1424          * When called from kernel_thread, don't do user tracing stuff.
1425          */
1426         if (likely(user_mode(regs)))
1427                 trace = tracehook_prepare_clone(clone_flags);
1428
1429         p = copy_process(clone_flags, stack_start, regs, stack_size,
1430                          child_tidptr, NULL, trace);
1431         /*
1432          * Do this prior waking up the new thread - the thread pointer
1433          * might get invalid after that point, if the thread exits quickly.
1434          */
1435         if (!IS_ERR(p)) {
1436                 struct completion vfork;
1437
1438                 trace_sched_process_fork(current, p);
1439
1440                 nr = task_pid_vnr(p);
1441
1442                 if (clone_flags & CLONE_PARENT_SETTID)
1443                         put_user(nr, parent_tidptr);
1444
1445                 if (clone_flags & CLONE_VFORK) {
1446                         p->vfork_done = &vfork;
1447                         init_completion(&vfork);
1448                 }
1449
1450                 audit_finish_fork(p);
1451                 tracehook_report_clone(regs, clone_flags, nr, p);
1452
1453                 /*
1454                  * We set PF_STARTING at creation in case tracing wants to
1455                  * use this to distinguish a fully live task from one that
1456                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1457                  * clear it and set the child going.
1458                  */
1459                 p->flags &= ~PF_STARTING;
1460
1461                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1462                         /*
1463                          * We'll start up with an immediate SIGSTOP.
1464                          */
1465                         sigaddset(&p->pending.signal, SIGSTOP);
1466                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1467                         __set_task_state(p, TASK_STOPPED);
1468                 } else {
1469                         wake_up_new_task(p, clone_flags);
1470                 }
1471
1472                 tracehook_report_clone_complete(trace, regs,
1473                                                 clone_flags, nr, p);
1474
1475                 if (clone_flags & CLONE_VFORK) {
1476                         freezer_do_not_count();
1477                         wait_for_completion(&vfork);
1478                         freezer_count();
1479                         tracehook_report_vfork_done(p, nr);
1480                 }
1481         } else {
1482                 nr = PTR_ERR(p);
1483         }
1484         return nr;
1485 }
1486
1487 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1488 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1489 #endif
1490
1491 static void sighand_ctor(void *data)
1492 {
1493         struct sighand_struct *sighand = data;
1494
1495         spin_lock_init(&sighand->siglock);
1496         init_waitqueue_head(&sighand->signalfd_wqh);
1497 }
1498
1499 void __init proc_caches_init(void)
1500 {
1501         sighand_cachep = kmem_cache_create("sighand_cache",
1502                         sizeof(struct sighand_struct), 0,
1503                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1504                         SLAB_NOTRACK, sighand_ctor);
1505         signal_cachep = kmem_cache_create("signal_cache",
1506                         sizeof(struct signal_struct), 0,
1507                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1508         files_cachep = kmem_cache_create("files_cache",
1509                         sizeof(struct files_struct), 0,
1510                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1511         fs_cachep = kmem_cache_create("fs_cache",
1512                         sizeof(struct fs_struct), 0,
1513                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1514         mm_cachep = kmem_cache_create("mm_struct",
1515                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1516                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1517         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1518         mmap_init();
1519 }
1520
1521 /*
1522  * Check constraints on flags passed to the unshare system call and
1523  * force unsharing of additional process context as appropriate.
1524  */
1525 static void check_unshare_flags(unsigned long *flags_ptr)
1526 {
1527         /*
1528          * If unsharing a thread from a thread group, must also
1529          * unshare vm.
1530          */
1531         if (*flags_ptr & CLONE_THREAD)
1532                 *flags_ptr |= CLONE_VM;
1533
1534         /*
1535          * If unsharing vm, must also unshare signal handlers.
1536          */
1537         if (*flags_ptr & CLONE_VM)
1538                 *flags_ptr |= CLONE_SIGHAND;
1539
1540         /*
1541          * If unsharing signal handlers and the task was created
1542          * using CLONE_THREAD, then must unshare the thread
1543          */
1544         if ((*flags_ptr & CLONE_SIGHAND) &&
1545             (atomic_read(&current->signal->count) > 1))
1546                 *flags_ptr |= CLONE_THREAD;
1547
1548         /*
1549          * If unsharing namespace, must also unshare filesystem information.
1550          */
1551         if (*flags_ptr & CLONE_NEWNS)
1552                 *flags_ptr |= CLONE_FS;
1553 }
1554
1555 /*
1556  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1557  */
1558 static int unshare_thread(unsigned long unshare_flags)
1559 {
1560         if (unshare_flags & CLONE_THREAD)
1561                 return -EINVAL;
1562
1563         return 0;
1564 }
1565
1566 /*
1567  * Unshare the filesystem structure if it is being shared
1568  */
1569 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1570 {
1571         struct fs_struct *fs = current->fs;
1572
1573         if (!(unshare_flags & CLONE_FS) || !fs)
1574                 return 0;
1575
1576         /* don't need lock here; in the worst case we'll do useless copy */
1577         if (fs->users == 1)
1578                 return 0;
1579
1580         *new_fsp = copy_fs_struct(fs);
1581         if (!*new_fsp)
1582                 return -ENOMEM;
1583
1584         return 0;
1585 }
1586
1587 /*
1588  * Unsharing of sighand is not supported yet
1589  */
1590 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1591 {
1592         struct sighand_struct *sigh = current->sighand;
1593
1594         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1595                 return -EINVAL;
1596         else
1597                 return 0;
1598 }
1599
1600 /*
1601  * Unshare vm if it is being shared
1602  */
1603 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1604 {
1605         struct mm_struct *mm = current->mm;
1606
1607         if ((unshare_flags & CLONE_VM) &&
1608             (mm && atomic_read(&mm->mm_users) > 1)) {
1609                 return -EINVAL;
1610         }
1611
1612         return 0;
1613 }
1614
1615 /*
1616  * Unshare file descriptor table if it is being shared
1617  */
1618 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1619 {
1620         struct files_struct *fd = current->files;
1621         int error = 0;
1622
1623         if ((unshare_flags & CLONE_FILES) &&
1624             (fd && atomic_read(&fd->count) > 1)) {
1625                 *new_fdp = dup_fd(fd, &error);
1626                 if (!*new_fdp)
1627                         return error;
1628         }
1629
1630         return 0;
1631 }
1632
1633 /*
1634  * unshare allows a process to 'unshare' part of the process
1635  * context which was originally shared using clone.  copy_*
1636  * functions used by do_fork() cannot be used here directly
1637  * because they modify an inactive task_struct that is being
1638  * constructed. Here we are modifying the current, active,
1639  * task_struct.
1640  */
1641 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1642 {
1643         int err = 0;
1644         struct fs_struct *fs, *new_fs = NULL;
1645         struct sighand_struct *new_sigh = NULL;
1646         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1647         struct files_struct *fd, *new_fd = NULL;
1648         struct nsproxy *new_nsproxy = NULL;
1649         int do_sysvsem = 0;
1650
1651         check_unshare_flags(&unshare_flags);
1652
1653         /* Return -EINVAL for all unsupported flags */
1654         err = -EINVAL;
1655         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1656                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1657                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1658                 goto bad_unshare_out;
1659
1660         /*
1661          * CLONE_NEWIPC must also detach from the undolist: after switching
1662          * to a new ipc namespace, the semaphore arrays from the old
1663          * namespace are unreachable.
1664          */
1665         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1666                 do_sysvsem = 1;
1667         if ((err = unshare_thread(unshare_flags)))
1668                 goto bad_unshare_out;
1669         if ((err = unshare_fs(unshare_flags, &new_fs)))
1670                 goto bad_unshare_cleanup_thread;
1671         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1672                 goto bad_unshare_cleanup_fs;
1673         if ((err = unshare_vm(unshare_flags, &new_mm)))
1674                 goto bad_unshare_cleanup_sigh;
1675         if ((err = unshare_fd(unshare_flags, &new_fd)))
1676                 goto bad_unshare_cleanup_vm;
1677         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1678                         new_fs)))
1679                 goto bad_unshare_cleanup_fd;
1680
1681         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1682                 if (do_sysvsem) {
1683                         /*
1684                          * CLONE_SYSVSEM is equivalent to sys_exit().
1685                          */
1686                         exit_sem(current);
1687                 }
1688
1689                 if (new_nsproxy) {
1690                         switch_task_namespaces(current, new_nsproxy);
1691                         new_nsproxy = NULL;
1692                 }
1693
1694                 task_lock(current);
1695
1696                 if (new_fs) {
1697                         fs = current->fs;
1698                         write_lock(&fs->lock);
1699                         current->fs = new_fs;
1700                         if (--fs->users)
1701                                 new_fs = NULL;
1702                         else
1703                                 new_fs = fs;
1704                         write_unlock(&fs->lock);
1705                 }
1706
1707                 if (new_mm) {
1708                         mm = current->mm;
1709                         active_mm = current->active_mm;
1710                         current->mm = new_mm;
1711                         current->active_mm = new_mm;
1712                         activate_mm(active_mm, new_mm);
1713                         new_mm = mm;
1714                 }
1715
1716                 if (new_fd) {
1717                         fd = current->files;
1718                         current->files = new_fd;
1719                         new_fd = fd;
1720                 }
1721
1722                 task_unlock(current);
1723         }
1724
1725         if (new_nsproxy)
1726                 put_nsproxy(new_nsproxy);
1727
1728 bad_unshare_cleanup_fd:
1729         if (new_fd)
1730                 put_files_struct(new_fd);
1731
1732 bad_unshare_cleanup_vm:
1733         if (new_mm)
1734                 mmput(new_mm);
1735
1736 bad_unshare_cleanup_sigh:
1737         if (new_sigh)
1738                 if (atomic_dec_and_test(&new_sigh->count))
1739                         kmem_cache_free(sighand_cachep, new_sigh);
1740
1741 bad_unshare_cleanup_fs:
1742         if (new_fs)
1743                 free_fs_struct(new_fs);
1744
1745 bad_unshare_cleanup_thread:
1746 bad_unshare_out:
1747         return err;
1748 }
1749
1750 /*
1751  *      Helper to unshare the files of the current task.
1752  *      We don't want to expose copy_files internals to
1753  *      the exec layer of the kernel.
1754  */
1755
1756 int unshare_files(struct files_struct **displaced)
1757 {
1758         struct task_struct *task = current;
1759         struct files_struct *copy = NULL;
1760         int error;
1761
1762         error = unshare_fd(CLONE_FILES, &copy);
1763         if (error || !copy) {
1764                 *displaced = NULL;
1765                 return error;
1766         }
1767         *displaced = task->files;
1768         task_lock(task);
1769         task->files = copy;
1770         task_unlock(task);
1771         return 0;
1772 }