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/seccomp.h>
  38 #include <linux/swap.h>
  39 #include <linux/syscalls.h>
  40 #include <linux/jiffies.h>
  41 #include <linux/futex.h>
  42 #include <linux/compat.h>
  43 #include <linux/kthread.h>
  44 #include <linux/task_io_accounting_ops.h>
  45 #include <linux/rcupdate.h>
  46 #include <linux/ptrace.h>
  47 #include <linux/mount.h>
  48 #include <linux/audit.h>
  49 #include <linux/memcontrol.h>
  50 #include <linux/ftrace.h>
  51 #include <linux/proc_fs.h>
  52 #include <linux/profile.h>
  53 #include <linux/rmap.h>
  54 #include <linux/ksm.h>
  55 #include <linux/acct.h>
  56 #include <linux/tsacct_kern.h>
  57 #include <linux/cn_proc.h>
  58 #include <linux/freezer.h>
  59 #include <linux/delayacct.h>
  60 #include <linux/taskstats_kern.h>
  61 #include <linux/random.h>
  62 #include <linux/tty.h>
  63 #include <linux/blkdev.h>
  64 #include <linux/fs_struct.h>
  65 #include <linux/magic.h>
  66 #include <linux/perf_event.h>
  67 #include <linux/posix-timers.h>
  68 #include <linux/user-return-notifier.h>
  69 #include <linux/oom.h>
  70 #include <linux/khugepaged.h>
  71 #include <linux/signalfd.h>
  72
  73 #include <asm/pgtable.h>
  74 #include <asm/pgalloc.h>
  75 #include <asm/uaccess.h>
  76 #include <asm/mmu_context.h>
  77 #include <asm/cacheflush.h>
  78 #include <asm/tlbflush.h>
  79
  80 #include <trace/events/sched.h>
  81
  82 #define CREATE_TRACE_POINTS
  83 #include <trace/events/task.h>
  84
  85 /*
  86  * Protected counters by write_lock_irq(&tasklist_lock)
  87  */
  88 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  89 int nr_threads;                 /* The idle threads do not count.. */
  90
  91 int max_threads;                /* tunable limit on nr_threads */
  92
  93 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  94
  95 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  96
  97 #ifdef CONFIG_PROVE_RCU
  98 int lockdep_tasklist_lock_is_held(void)
  99 {
 100         return lockdep_is_held(&tasklist_lock);
 101 }
 102 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 103 #endif /* #ifdef CONFIG_PROVE_RCU */
 104
 105 int nr_processes(void)
 106 {
 107         int cpu;
 108         int total = 0;
 109
 110         for_each_possible_cpu(cpu)
 111                 total += per_cpu(process_counts, cpu);
 112
 113         return total;
 114 }
 115
 116 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 117 static struct kmem_cache *task_struct_cachep;
 118
 119 static inline struct task_struct *alloc_task_struct_node(int node)
 120 {
 121         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 122 }
 123
 124 void __weak arch_release_task_struct(struct task_struct *tsk) { }
 125
 126 static inline void free_task_struct(struct task_struct *tsk)
 127 {
 128         arch_release_task_struct(tsk);
 129         kmem_cache_free(task_struct_cachep, tsk);
 130 }
 131 #endif
 132
 133 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
 134 void __weak arch_release_thread_info(struct thread_info *ti) { }
 135
 136 /*
 137  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 138  * kmemcache based allocator.
 139  */
 140 # if THREAD_SIZE >= PAGE_SIZE
 141 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 142                                                   int node)
 143 {
 144         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 145                                              THREAD_SIZE_ORDER);
 146
 147         return page ? page_address(page) : NULL;
 148 }
 149
 150 static inline void free_thread_info(struct thread_info *ti)
 151 {
 152         arch_release_thread_info(ti);
 153         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 154 }
 155 # else
 156 static struct kmem_cache *thread_info_cache;
 157
 158 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 159                                                   int node)
 160 {
 161         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
 162 }
 163
 164 static void free_thread_info(struct thread_info *ti)
 165 {
 166         arch_release_thread_info(ti);
 167         kmem_cache_free(thread_info_cache, ti);
 168 }
 169
 170 void thread_info_cache_init(void)
 171 {
 172         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
 173                                               THREAD_SIZE, 0, NULL);
 174         BUG_ON(thread_info_cache == NULL);
 175 }
 176 # endif
 177 #endif
 178
 179 /* SLAB cache for signal_struct structures (tsk->signal) */
 180 static struct kmem_cache *signal_cachep;
 181
 182 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 183 struct kmem_cache *sighand_cachep;
 184
 185 /* SLAB cache for files_struct structures (tsk->files) */
 186 struct kmem_cache *files_cachep;
 187
 188 /* SLAB cache for fs_struct structures (tsk->fs) */
 189 struct kmem_cache *fs_cachep;
 190
 191 /* SLAB cache for vm_area_struct structures */
 192 struct kmem_cache *vm_area_cachep;
 193
 194 /* SLAB cache for mm_struct structures (tsk->mm) */
 195 static struct kmem_cache *mm_cachep;
 196
 197 static void account_kernel_stack(struct thread_info *ti, int account)
 198 {
 199         struct zone *zone = page_zone(virt_to_page(ti));
 200
 201         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 202 }
 203
 204 void free_task(struct task_struct *tsk)
 205 {
 206         account_kernel_stack(tsk->stack, -1);
 207         free_thread_info(tsk->stack);
 208         rt_mutex_debug_task_free(tsk);
 209         ftrace_graph_exit_task(tsk);
 210         put_seccomp_filter(tsk);
 211         free_task_struct(tsk);
 212 }
 213 EXPORT_SYMBOL(free_task);
 214
 215 static inline void free_signal_struct(struct signal_struct *sig)
 216 {
 217         taskstats_tgid_free(sig);
 218         sched_autogroup_exit(sig);
 219         kmem_cache_free(signal_cachep, sig);
 220 }
 221
 222 static inline void put_signal_struct(struct signal_struct *sig)
 223 {
 224         if (atomic_dec_and_test(&sig->sigcnt))
 225                 free_signal_struct(sig);
 226 }
 227
 228 void __put_task_struct(struct task_struct *tsk)
 229 {
 230         WARN_ON(!tsk->exit_state);
 231         WARN_ON(atomic_read(&tsk->usage));
 232         WARN_ON(tsk == current);
 233
 234         security_task_free(tsk);
 235         exit_creds(tsk);
 236         delayacct_tsk_free(tsk);
 237         put_signal_struct(tsk->signal);
 238
 239         if (!profile_handoff_task(tsk))
 240                 free_task(tsk);
 241 }
 242 EXPORT_SYMBOL_GPL(__put_task_struct);
 243
 244 void __init __weak arch_task_cache_init(void) { }
 245
 246 void __init fork_init(unsigned long mempages)
 247 {
 248 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 249 #ifndef ARCH_MIN_TASKALIGN
 250 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 251 #endif
 252         /* create a slab on which task_structs can be allocated */
 253         task_struct_cachep =
 254                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 255                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 256 #endif
 257
 258         /* do the arch specific task caches init */
 259         arch_task_cache_init();
 260
 261         /*
 262          * The default maximum number of threads is set to a safe
 263          * value: the thread structures can take up at most half
 264          * of memory.
 265          */
 266         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 267
 268         /*
 269          * we need to allow at least 20 threads to boot a system
 270          */
 271         if (max_threads < 20)
 272                 max_threads = 20;
 273
 274         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 275         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 276         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 277                 init_task.signal->rlim[RLIMIT_NPROC];
 278 }
 279
 280 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 281                                                struct task_struct *src)
 282 {
 283         *dst = *src;
 284         return 0;
 285 }
 286
 287 static struct task_struct *dup_task_struct(struct task_struct *orig)
 288 {
 289         struct task_struct *tsk;
 290         struct thread_info *ti;
 291         unsigned long *stackend;
 292         int node = tsk_fork_get_node(orig);
 293         int err;
 294
 295         prepare_to_copy(orig);
 296
 297         tsk = alloc_task_struct_node(node);
 298         if (!tsk)
 299                 return NULL;
 300
 301         ti = alloc_thread_info_node(tsk, node);
 302         if (!ti) {
 303                 free_task_struct(tsk);
 304                 return NULL;
 305         }
 306
 307         err = arch_dup_task_struct(tsk, orig);
 308         if (err)
 309                 goto out;
 310
 311         tsk->stack = ti;
 312
 313         setup_thread_stack(tsk, orig);
 314         clear_user_return_notifier(tsk);
 315         clear_tsk_need_resched(tsk);
 316         stackend = end_of_stack(tsk);
 317         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 318
 319 #ifdef CONFIG_CC_STACKPROTECTOR
 320         tsk->stack_canary = get_random_int();
 321 #endif
 322
 323         /*
 324          * One for us, one for whoever does the "release_task()" (usually
 325          * parent)
 326          */
 327         atomic_set(&tsk->usage, 2);
 328 #ifdef CONFIG_BLK_DEV_IO_TRACE
 329         tsk->btrace_seq = 0;
 330 #endif
 331         tsk->splice_pipe = NULL;
 332
 333         account_kernel_stack(ti, 1);
 334
 335         return tsk;
 336
 337 out:
 338         free_thread_info(ti);
 339         free_task_struct(tsk);
 340         return NULL;
 341 }
 342
 343 #ifdef CONFIG_MMU
 344 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 345 {
 346         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 347         struct rb_node **rb_link, *rb_parent;
 348         int retval;
 349         unsigned long charge;
 350         struct mempolicy *pol;
 351
 352         down_write(&oldmm->mmap_sem);
 353         flush_cache_dup_mm(oldmm);
 354         /*
 355          * Not linked in yet - no deadlock potential:
 356          */
 357         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 358
 359         mm->locked_vm = 0;
 360         mm->mmap = NULL;
 361         mm->mmap_cache = NULL;
 362         mm->free_area_cache = oldmm->mmap_base;
 363         mm->cached_hole_size = ~0UL;
 364         mm->map_count = 0;
 365         cpumask_clear(mm_cpumask(mm));
 366         mm->mm_rb = RB_ROOT;
 367         rb_link = &mm->mm_rb.rb_node;
 368         rb_parent = NULL;
 369         pprev = &mm->mmap;
 370         retval = ksm_fork(mm, oldmm);
 371         if (retval)
 372                 goto out;
 373         retval = khugepaged_fork(mm, oldmm);
 374         if (retval)
 375                 goto out;
 376
 377         prev = NULL;
 378         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 379                 struct file *file;
 380
 381                 if (mpnt->vm_flags & VM_DONTCOPY) {
 382                         long pages = vma_pages(mpnt);
 383                         mm->total_vm -= pages;
 384                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 385                                                                 -pages);
 386                         continue;
 387                 }
 388                 charge = 0;
 389                 if (mpnt->vm_flags & VM_ACCOUNT) {
 390                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 391                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 392                                 goto fail_nomem;
 393                         charge = len;
 394                 }
 395                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 396                 if (!tmp)
 397                         goto fail_nomem;
 398                 *tmp = *mpnt;
 399                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 400                 pol = mpol_dup(vma_policy(mpnt));
 401                 retval = PTR_ERR(pol);
 402                 if (IS_ERR(pol))
 403                         goto fail_nomem_policy;
 404                 vma_set_policy(tmp, pol);
 405                 tmp->vm_mm = mm;
 406                 if (anon_vma_fork(tmp, mpnt))
 407                         goto fail_nomem_anon_vma_fork;
 408                 tmp->vm_flags &= ~VM_LOCKED;
 409                 tmp->vm_next = tmp->vm_prev = NULL;
 410                 file = tmp->vm_file;
 411                 if (file) {
 412                         struct inode *inode = file->f_path.dentry->d_inode;
 413                         struct address_space *mapping = file->f_mapping;
 414
 415                         get_file(file);
 416                         if (tmp->vm_flags & VM_DENYWRITE)
 417                                 atomic_dec(&inode->i_writecount);
 418                         mutex_lock(&mapping->i_mmap_mutex);
 419                         if (tmp->vm_flags & VM_SHARED)
 420                                 mapping->i_mmap_writable++;
 421                         flush_dcache_mmap_lock(mapping);
 422                         /* insert tmp into the share list, just after mpnt */
 423                         vma_prio_tree_add(tmp, mpnt);
 424                         flush_dcache_mmap_unlock(mapping);
 425                         mutex_unlock(&mapping->i_mmap_mutex);
 426                 }
 427
 428                 /*
 429                  * Clear hugetlb-related page reserves for children. This only
 430                  * affects MAP_PRIVATE mappings. Faults generated by the child
 431                  * are not guaranteed to succeed, even if read-only
 432                  */
 433                 if (is_vm_hugetlb_page(tmp))
 434                         reset_vma_resv_huge_pages(tmp);
 435
 436                 /*
 437                  * Link in the new vma and copy the page table entries.
 438                  */
 439                 *pprev = tmp;
 440                 pprev = &tmp->vm_next;
 441                 tmp->vm_prev = prev;
 442                 prev = tmp;
 443
 444                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 445                 rb_link = &tmp->vm_rb.rb_right;
 446                 rb_parent = &tmp->vm_rb;
 447
 448                 mm->map_count++;
 449                 retval = copy_page_range(mm, oldmm, mpnt);
 450
 451                 if (tmp->vm_ops && tmp->vm_ops->open)
 452                         tmp->vm_ops->open(tmp);
 453
 454                 if (retval)
 455                         goto out;
 456         }
 457         /* a new mm has just been created */
 458         arch_dup_mmap(oldmm, mm);
 459         retval = 0;
 460 out:
 461         up_write(&mm->mmap_sem);
 462         flush_tlb_mm(oldmm);
 463         up_write(&oldmm->mmap_sem);
 464         return retval;
 465 fail_nomem_anon_vma_fork:
 466         mpol_put(pol);
 467 fail_nomem_policy:
 468         kmem_cache_free(vm_area_cachep, tmp);
 469 fail_nomem:
 470         retval = -ENOMEM;
 471         vm_unacct_memory(charge);
 472         goto out;
 473 }
 474
 475 static inline int mm_alloc_pgd(struct mm_struct *mm)
 476 {
 477         mm->pgd = pgd_alloc(mm);
 478         if (unlikely(!mm->pgd))
 479                 return -ENOMEM;
 480         return 0;
 481 }
 482
 483 static inline void mm_free_pgd(struct mm_struct *mm)
 484 {
 485         pgd_free(mm, mm->pgd);
 486 }
 487 #else
 488 #define dup_mmap(mm, oldmm)     (0)
 489 #define mm_alloc_pgd(mm)        (0)
 490 #define mm_free_pgd(mm)
 491 #endif /* CONFIG_MMU */
 492
 493 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 494
 495 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 496 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 497
 498 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 499
 500 static int __init coredump_filter_setup(char *s)
 501 {
 502         default_dump_filter =
 503                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 504                 MMF_DUMP_FILTER_MASK;
 505         return 1;
 506 }
 507
 508 __setup("coredump_filter=", coredump_filter_setup);
 509
 510 #include <linux/init_task.h>
 511
 512 static void mm_init_aio(struct mm_struct *mm)
 513 {
 514 #ifdef CONFIG_AIO
 515         spin_lock_init(&mm->ioctx_lock);
 516         INIT_HLIST_HEAD(&mm->ioctx_list);
 517 #endif
 518 }
 519
 520 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 521 {
 522         atomic_set(&mm->mm_users, 1);
 523         atomic_set(&mm->mm_count, 1);
 524         init_rwsem(&mm->mmap_sem);
 525         INIT_LIST_HEAD(&mm->mmlist);
 526         mm->flags = (current->mm) ?
 527                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 528         mm->core_state = NULL;
 529         mm->nr_ptes = 0;
 530         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 531         spin_lock_init(&mm->page_table_lock);
 532         mm->free_area_cache = TASK_UNMAPPED_BASE;
 533         mm->cached_hole_size = ~0UL;
 534         mm_init_aio(mm);
 535         mm_init_owner(mm, p);
 536
 537         if (likely(!mm_alloc_pgd(mm))) {
 538                 mm->def_flags = 0;
 539                 mmu_notifier_mm_init(mm);
 540                 return mm;
 541         }
 542
 543         free_mm(mm);
 544         return NULL;
 545 }
 546
 547 static void check_mm(struct mm_struct *mm)
 548 {
 549         int i;
 550
 551         for (i = 0; i < NR_MM_COUNTERS; i++) {
 552                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 553
 554                 if (unlikely(x))
 555                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 556                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 557         }
 558
 559 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 560         VM_BUG_ON(mm->pmd_huge_pte);
 561 #endif
 562 }
 563
 564 /*
 565  * Allocate and initialize an mm_struct.
 566  */
 567 struct mm_struct *mm_alloc(void)
 568 {
 569         struct mm_struct *mm;
 570
 571         mm = allocate_mm();
 572         if (!mm)
 573                 return NULL;
 574
 575         memset(mm, 0, sizeof(*mm));
 576         mm_init_cpumask(mm);
 577         return mm_init(mm, current);
 578 }
 579
 580 /*
 581  * Called when the last reference to the mm
 582  * is dropped: either by a lazy thread or by
 583  * mmput. Free the page directory and the mm.
 584  */
 585 void __mmdrop(struct mm_struct *mm)
 586 {
 587         BUG_ON(mm == &init_mm);
 588         mm_free_pgd(mm);
 589         destroy_context(mm);
 590         mmu_notifier_mm_destroy(mm);
 591         check_mm(mm);
 592         free_mm(mm);
 593 }
 594 EXPORT_SYMBOL_GPL(__mmdrop);
 595
 596 /*
 597  * Decrement the use count and release all resources for an mm.
 598  */
 599 void mmput(struct mm_struct *mm)
 600 {
 601         might_sleep();
 602
 603         if (atomic_dec_and_test(&mm->mm_users)) {
 604                 exit_aio(mm);
 605                 ksm_exit(mm);
 606                 khugepaged_exit(mm); /* must run before exit_mmap */
 607                 exit_mmap(mm);
 608                 set_mm_exe_file(mm, NULL);
 609                 if (!list_empty(&mm->mmlist)) {
 610                         spin_lock(&mmlist_lock);
 611                         list_del(&mm->mmlist);
 612                         spin_unlock(&mmlist_lock);
 613                 }
 614                 put_swap_token(mm);
 615                 if (mm->binfmt)
 616                         module_put(mm->binfmt->module);
 617                 mmdrop(mm);
 618         }
 619 }
 620 EXPORT_SYMBOL_GPL(mmput);
 621
 622 /*
 623  * We added or removed a vma mapping the executable. The vmas are only mapped
 624  * during exec and are not mapped with the mmap system call.
 625  * Callers must hold down_write() on the mm's mmap_sem for these
 626  */
 627 void added_exe_file_vma(struct mm_struct *mm)
 628 {
 629         mm->num_exe_file_vmas++;
 630 }
 631
 632 void removed_exe_file_vma(struct mm_struct *mm)
 633 {
 634         mm->num_exe_file_vmas--;
 635         if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
 636                 fput(mm->exe_file);
 637                 mm->exe_file = NULL;
 638         }
 639
 640 }
 641
 642 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 643 {
 644         if (new_exe_file)
 645                 get_file(new_exe_file);
 646         if (mm->exe_file)
 647                 fput(mm->exe_file);
 648         mm->exe_file = new_exe_file;
 649         mm->num_exe_file_vmas = 0;
 650 }
 651
 652 struct file *get_mm_exe_file(struct mm_struct *mm)
 653 {
 654         struct file *exe_file;
 655
 656         /* We need mmap_sem to protect against races with removal of
 657          * VM_EXECUTABLE vmas */
 658         down_read(&mm->mmap_sem);
 659         exe_file = mm->exe_file;
 660         if (exe_file)
 661                 get_file(exe_file);
 662         up_read(&mm->mmap_sem);
 663         return exe_file;
 664 }
 665
 666 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 667 {
 668         /* It's safe to write the exe_file pointer without exe_file_lock because
 669          * this is called during fork when the task is not yet in /proc */
 670         newmm->exe_file = get_mm_exe_file(oldmm);
 671 }
 672
 673 /**
 674  * get_task_mm - acquire a reference to the task's mm
 675  *
 676  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 677  * this kernel workthread has transiently adopted a user mm with use_mm,
 678  * to do its AIO) is not set and if so returns a reference to it, after
 679  * bumping up the use count.  User must release the mm via mmput()
 680  * after use.  Typically used by /proc and ptrace.
 681  */
 682 struct mm_struct *get_task_mm(struct task_struct *task)
 683 {
 684         struct mm_struct *mm;
 685
 686         task_lock(task);
 687         mm = task->mm;
 688         if (mm) {
 689                 if (task->flags & PF_KTHREAD)
 690                         mm = NULL;
 691                 else
 692                         atomic_inc(&mm->mm_users);
 693         }
 694         task_unlock(task);
 695         return mm;
 696 }
 697 EXPORT_SYMBOL_GPL(get_task_mm);
 698
 699 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 700 {
 701         struct mm_struct *mm;
 702         int err;
 703
 704         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 705         if (err)
 706                 return ERR_PTR(err);
 707
 708         mm = get_task_mm(task);
 709         if (mm && mm != current->mm &&
 710                         !ptrace_may_access(task, mode)) {
 711                 mmput(mm);
 712                 mm = ERR_PTR(-EACCES);
 713         }
 714         mutex_unlock(&task->signal->cred_guard_mutex);
 715
 716         return mm;
 717 }
 718
 719 static void complete_vfork_done(struct task_struct *tsk)
 720 {
 721         struct completion *vfork;
 722
 723         task_lock(tsk);
 724         vfork = tsk->vfork_done;
 725         if (likely(vfork)) {
 726                 tsk->vfork_done = NULL;
 727                 complete(vfork);
 728         }
 729         task_unlock(tsk);
 730 }
 731
 732 static int wait_for_vfork_done(struct task_struct *child,
 733                                 struct completion *vfork)
 734 {
 735         int killed;
 736
 737         freezer_do_not_count();
 738         killed = wait_for_completion_killable(vfork);
 739         freezer_count();
 740
 741         if (killed) {
 742                 task_lock(child);
 743                 child->vfork_done = NULL;
 744                 task_unlock(child);
 745         }
 746
 747         put_task_struct(child);
 748         return killed;
 749 }
 750
 751 /* Please note the differences between mmput and mm_release.
 752  * mmput is called whenever we stop holding onto a mm_struct,
 753  * error success whatever.
 754  *
 755  * mm_release is called after a mm_struct has been removed
 756  * from the current process.
 757  *
 758  * This difference is important for error handling, when we
 759  * only half set up a mm_struct for a new process and need to restore
 760  * the old one.  Because we mmput the new mm_struct before
 761  * restoring the old one. . .
 762  * Eric Biederman 10 January 1998
 763  */
 764 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 765 {
 766         /* Get rid of any futexes when releasing the mm */
 767 #ifdef CONFIG_FUTEX
 768         if (unlikely(tsk->robust_list)) {
 769                 exit_robust_list(tsk);
 770                 tsk->robust_list = NULL;
 771         }
 772 #ifdef CONFIG_COMPAT
 773         if (unlikely(tsk->compat_robust_list)) {
 774                 compat_exit_robust_list(tsk);
 775                 tsk->compat_robust_list = NULL;
 776         }
 777 #endif
 778         if (unlikely(!list_empty(&tsk->pi_state_list)))
 779                 exit_pi_state_list(tsk);
 780 #endif
 781
 782         /* Get rid of any cached register state */
 783         deactivate_mm(tsk, mm);
 784
 785         if (tsk->vfork_done)
 786                 complete_vfork_done(tsk);
 787
 788         /*
 789          * If we're exiting normally, clear a user-space tid field if
 790          * requested.  We leave this alone when dying by signal, to leave
 791          * the value intact in a core dump, and to save the unnecessary
 792          * trouble, say, a killed vfork parent shouldn't touch this mm.
 793          * Userland only wants this done for a sys_exit.
 794          */
 795         if (tsk->clear_child_tid) {
 796                 if (!(tsk->flags & PF_SIGNALED) &&
 797                     atomic_read(&mm->mm_users) > 1) {
 798                         /*
 799                          * We don't check the error code - if userspace has
 800                          * not set up a proper pointer then tough luck.
 801                          */
 802                         put_user(0, tsk->clear_child_tid);
 803                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 804                                         1, NULL, NULL, 0);
 805                 }
 806                 tsk->clear_child_tid = NULL;
 807         }
 808 }
 809
 810 /*
 811  * Allocate a new mm structure and copy contents from the
 812  * mm structure of the passed in task structure.
 813  */
 814 struct mm_struct *dup_mm(struct task_struct *tsk)
 815 {
 816         struct mm_struct *mm, *oldmm = current->mm;
 817         int err;
 818
 819         if (!oldmm)
 820                 return NULL;
 821
 822         mm = allocate_mm();
 823         if (!mm)
 824                 goto fail_nomem;
 825
 826         memcpy(mm, oldmm, sizeof(*mm));
 827         mm_init_cpumask(mm);
 828
 829         /* Initializing for Swap token stuff */
 830         mm->token_priority = 0;
 831         mm->last_interval = 0;
 832
 833 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 834         mm->pmd_huge_pte = NULL;
 835 #endif
 836
 837         if (!mm_init(mm, tsk))
 838                 goto fail_nomem;
 839
 840         if (init_new_context(tsk, mm))
 841                 goto fail_nocontext;
 842
 843         dup_mm_exe_file(oldmm, mm);
 844
 845         err = dup_mmap(mm, oldmm);
 846         if (err)
 847                 goto free_pt;
 848
 849         mm->hiwater_rss = get_mm_rss(mm);
 850         mm->hiwater_vm = mm->total_vm;
 851
 852         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 853                 goto free_pt;
 854
 855         return mm;
 856
 857 free_pt:
 858         /* don't put binfmt in mmput, we haven't got module yet */
 859         mm->binfmt = NULL;
 860         mmput(mm);
 861
 862 fail_nomem:
 863         return NULL;
 864
 865 fail_nocontext:
 866         /*
 867          * If init_new_context() failed, we cannot use mmput() to free the mm
 868          * because it calls destroy_context()
 869          */
 870         mm_free_pgd(mm);
 871         free_mm(mm);
 872         return NULL;
 873 }
 874
 875 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 876 {
 877         struct mm_struct *mm, *oldmm;
 878         int retval;
 879
 880         tsk->min_flt = tsk->maj_flt = 0;
 881         tsk->nvcsw = tsk->nivcsw = 0;
 882 #ifdef CONFIG_DETECT_HUNG_TASK
 883         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 884 #endif
 885
 886         tsk->mm = NULL;
 887         tsk->active_mm = NULL;
 888
 889         /*
 890          * Are we cloning a kernel thread?
 891          *
 892          * We need to steal a active VM for that..
 893          */
 894         oldmm = current->mm;
 895         if (!oldmm)
 896                 return 0;
 897
 898         if (clone_flags & CLONE_VM) {
 899                 atomic_inc(&oldmm->mm_users);
 900                 mm = oldmm;
 901                 goto good_mm;
 902         }
 903
 904         retval = -ENOMEM;
 905         mm = dup_mm(tsk);
 906         if (!mm)
 907                 goto fail_nomem;
 908
 909 good_mm:
 910         /* Initializing for Swap token stuff */
 911         mm->token_priority = 0;
 912         mm->last_interval = 0;
 913
 914         tsk->mm = mm;
 915         tsk->active_mm = mm;
 916         return 0;
 917
 918 fail_nomem:
 919         return retval;
 920 }
 921
 922 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 923 {
 924         struct fs_struct *fs = current->fs;
 925         if (clone_flags & CLONE_FS) {
 926                 /* tsk->fs is already what we want */
 927                 spin_lock(&fs->lock);
 928                 if (fs->in_exec) {
 929                         spin_unlock(&fs->lock);
 930                         return -EAGAIN;
 931                 }
 932                 fs->users++;
 933                 spin_unlock(&fs->lock);
 934                 return 0;
 935         }
 936         tsk->fs = copy_fs_struct(fs);
 937         if (!tsk->fs)
 938                 return -ENOMEM;
 939         return 0;
 940 }
 941
 942 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 943 {
 944         struct files_struct *oldf, *newf;
 945         int error = 0;
 946
 947         /*
 948          * A background process may not have any files ...
 949          */
 950         oldf = current->files;
 951         if (!oldf)
 952                 goto out;
 953
 954         if (clone_flags & CLONE_FILES) {
 955                 atomic_inc(&oldf->count);
 956                 goto out;
 957         }
 958
 959         newf = dup_fd(oldf, &error);
 960         if (!newf)
 961                 goto out;
 962
 963         tsk->files = newf;
 964         error = 0;
 965 out:
 966         return error;
 967 }
 968
 969 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 970 {
 971 #ifdef CONFIG_BLOCK
 972         struct io_context *ioc = current->io_context;
 973         struct io_context *new_ioc;
 974
 975         if (!ioc)
 976                 return 0;
 977         /*
 978          * Share io context with parent, if CLONE_IO is set
 979          */
 980         if (clone_flags & CLONE_IO) {
 981                 tsk->io_context = ioc_task_link(ioc);
 982                 if (unlikely(!tsk->io_context))
 983                         return -ENOMEM;
 984         } else if (ioprio_valid(ioc->ioprio)) {
 985                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
 986                 if (unlikely(!new_ioc))
 987                         return -ENOMEM;
 988
 989                 new_ioc->ioprio = ioc->ioprio;
 990                 put_io_context(new_ioc);
 991         }
 992 #endif
 993         return 0;
 994 }
 995
 996 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 997 {
 998         struct sighand_struct *sig;
 999
1000         if (clone_flags & CLONE_SIGHAND) {
1001                 atomic_inc(&current->sighand->count);
1002                 return 0;
1003         }
1004         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1005         rcu_assign_pointer(tsk->sighand, sig);
1006         if (!sig)
1007                 return -ENOMEM;
1008         atomic_set(&sig->count, 1);
1009         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1010         return 0;
1011 }
1012
1013 void __cleanup_sighand(struct sighand_struct *sighand)
1014 {
1015         if (atomic_dec_and_test(&sighand->count)) {
1016                 signalfd_cleanup(sighand);
1017                 kmem_cache_free(sighand_cachep, sighand);
1018         }
1019 }
1020
1021
1022 /*
1023  * Initialize POSIX timer handling for a thread group.
1024  */
1025 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1026 {
1027         unsigned long cpu_limit;
1028
1029         /* Thread group counters. */
1030         thread_group_cputime_init(sig);
1031
1032         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1033         if (cpu_limit != RLIM_INFINITY) {
1034                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1035                 sig->cputimer.running = 1;
1036         }
1037
1038         /* The timer lists. */
1039         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1040         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1041         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1042 }
1043
1044 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1045 {
1046         struct signal_struct *sig;
1047
1048         if (clone_flags & CLONE_THREAD)
1049                 return 0;
1050
1051         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1052         tsk->signal = sig;
1053         if (!sig)
1054                 return -ENOMEM;
1055
1056         sig->nr_threads = 1;
1057         atomic_set(&sig->live, 1);
1058         atomic_set(&sig->sigcnt, 1);
1059         init_waitqueue_head(&sig->wait_chldexit);
1060         if (clone_flags & CLONE_NEWPID)
1061                 sig->flags |= SIGNAL_UNKILLABLE;
1062         sig->curr_target = tsk;
1063         init_sigpending(&sig->shared_pending);
1064         INIT_LIST_HEAD(&sig->posix_timers);
1065
1066         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1067         sig->real_timer.function = it_real_fn;
1068
1069         task_lock(current->group_leader);
1070         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1071         task_unlock(current->group_leader);
1072
1073         posix_cpu_timers_init_group(sig);
1074
1075         tty_audit_fork(sig);
1076         sched_autogroup_fork(sig);
1077
1078 #ifdef CONFIG_CGROUPS
1079         init_rwsem(&sig->group_rwsem);
1080 #endif
1081
1082         sig->oom_adj = current->signal->oom_adj;
1083         sig->oom_score_adj = current->signal->oom_score_adj;
1084         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1085
1086         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1087                                    current->signal->is_child_subreaper;
1088
1089         mutex_init(&sig->cred_guard_mutex);
1090
1091         return 0;
1092 }
1093
1094 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1095 {
1096         unsigned long new_flags = p->flags;
1097
1098         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1099         new_flags |= PF_FORKNOEXEC;
1100         p->flags = new_flags;
1101 }
1102
1103 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1104 {
1105         current->clear_child_tid = tidptr;
1106
1107         return task_pid_vnr(current);
1108 }
1109
1110 static void rt_mutex_init_task(struct task_struct *p)
1111 {
1112         raw_spin_lock_init(&p->pi_lock);
1113 #ifdef CONFIG_RT_MUTEXES
1114         plist_head_init(&p->pi_waiters);
1115         p->pi_blocked_on = NULL;
1116 #endif
1117 }
1118
1119 #ifdef CONFIG_MM_OWNER
1120 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1121 {
1122         mm->owner = p;
1123 }
1124 #endif /* CONFIG_MM_OWNER */
1125
1126 /*
1127  * Initialize POSIX timer handling for a single task.
1128  */
1129 static void posix_cpu_timers_init(struct task_struct *tsk)
1130 {
1131         tsk->cputime_expires.prof_exp = 0;
1132         tsk->cputime_expires.virt_exp = 0;
1133         tsk->cputime_expires.sched_exp = 0;
1134         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1135         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1136         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1137 }
1138
1139 /*
1140  * This creates a new process as a copy of the old one,
1141  * but does not actually start it yet.
1142  *
1143  * It copies the registers, and all the appropriate
1144  * parts of the process environment (as per the clone
1145  * flags). The actual kick-off is left to the caller.
1146  */
1147 static struct task_struct *copy_process(unsigned long clone_flags,
1148                                         unsigned long stack_start,
1149                                         struct pt_regs *regs,
1150                                         unsigned long stack_size,
1151                                         int __user *child_tidptr,
1152                                         struct pid *pid,
1153                                         int trace)
1154 {
1155         int retval;
1156         struct task_struct *p;
1157         int cgroup_callbacks_done = 0;
1158
1159         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1160                 return ERR_PTR(-EINVAL);
1161
1162         /*
1163          * Thread groups must share signals as well, and detached threads
1164          * can only be started up within the thread group.
1165          */
1166         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1167                 return ERR_PTR(-EINVAL);
1168
1169         /*
1170          * Shared signal handlers imply shared VM. By way of the above,
1171          * thread groups also imply shared VM. Blocking this case allows
1172          * for various simplifications in other code.
1173          */
1174         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1175                 return ERR_PTR(-EINVAL);
1176
1177         /*
1178          * Siblings of global init remain as zombies on exit since they are
1179          * not reaped by their parent (swapper). To solve this and to avoid
1180          * multi-rooted process trees, prevent global and container-inits
1181          * from creating siblings.
1182          */
1183         if ((clone_flags & CLONE_PARENT) &&
1184                                 current->signal->flags & SIGNAL_UNKILLABLE)
1185                 return ERR_PTR(-EINVAL);
1186
1187         retval = security_task_create(clone_flags);
1188         if (retval)
1189                 goto fork_out;
1190
1191         retval = -ENOMEM;
1192         p = dup_task_struct(current);
1193         if (!p)
1194                 goto fork_out;
1195
1196         ftrace_graph_init_task(p);
1197         get_seccomp_filter(p);
1198
1199         rt_mutex_init_task(p);
1200
1201 #ifdef CONFIG_PROVE_LOCKING
1202         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1203         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1204 #endif
1205         retval = -EAGAIN;
1206         if (atomic_read(&p->real_cred->user->processes) >=
1207                         task_rlimit(p, RLIMIT_NPROC)) {
1208                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1209                     p->real_cred->user != INIT_USER)
1210                         goto bad_fork_free;
1211         }
1212         current->flags &= ~PF_NPROC_EXCEEDED;
1213
1214         retval = copy_creds(p, clone_flags);
1215         if (retval < 0)
1216                 goto bad_fork_free;
1217
1218         /*
1219          * If multiple threads are within copy_process(), then this check
1220          * triggers too late. This doesn't hurt, the check is only there
1221          * to stop root fork bombs.
1222          */
1223         retval = -EAGAIN;
1224         if (nr_threads >= max_threads)
1225                 goto bad_fork_cleanup_count;
1226
1227         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1228                 goto bad_fork_cleanup_count;
1229
1230         p->did_exec = 0;
1231         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1232         copy_flags(clone_flags, p);
1233         INIT_LIST_HEAD(&p->children);
1234         INIT_LIST_HEAD(&p->sibling);
1235         rcu_copy_process(p);
1236         p->vfork_done = NULL;
1237         spin_lock_init(&p->alloc_lock);
1238
1239         init_sigpending(&p->pending);
1240
1241         p->utime = p->stime = p->gtime = 0;
1242         p->utimescaled = p->stimescaled = 0;
1243 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1244         p->prev_utime = p->prev_stime = 0;
1245 #endif
1246 #if defined(SPLIT_RSS_COUNTING)
1247         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1248 #endif
1249
1250         p->default_timer_slack_ns = current->timer_slack_ns;
1251
1252         task_io_accounting_init(&p->ioac);
1253         acct_clear_integrals(p);
1254
1255         posix_cpu_timers_init(p);
1256
1257         do_posix_clock_monotonic_gettime(&p->start_time);
1258         p->real_start_time = p->start_time;
1259         monotonic_to_bootbased(&p->real_start_time);
1260         p->io_context = NULL;
1261         p->audit_context = NULL;
1262         if (clone_flags & CLONE_THREAD)
1263                 threadgroup_change_begin(current);
1264         cgroup_fork(p);
1265 #ifdef CONFIG_NUMA
1266         p->mempolicy = mpol_dup(p->mempolicy);
1267         if (IS_ERR(p->mempolicy)) {
1268                 retval = PTR_ERR(p->mempolicy);
1269                 p->mempolicy = NULL;
1270                 goto bad_fork_cleanup_cgroup;
1271         }
1272         mpol_fix_fork_child_flag(p);
1273 #endif
1274 #ifdef CONFIG_CPUSETS
1275         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1276         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1277         seqcount_init(&p->mems_allowed_seq);
1278 #endif
1279 #ifdef CONFIG_TRACE_IRQFLAGS
1280         p->irq_events = 0;
1281 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1282         p->hardirqs_enabled = 1;
1283 #else
1284         p->hardirqs_enabled = 0;
1285 #endif
1286         p->hardirq_enable_ip = 0;
1287         p->hardirq_enable_event = 0;
1288         p->hardirq_disable_ip = _THIS_IP_;
1289         p->hardirq_disable_event = 0;
1290         p->softirqs_enabled = 1;
1291         p->softirq_enable_ip = _THIS_IP_;
1292         p->softirq_enable_event = 0;
1293         p->softirq_disable_ip = 0;
1294         p->softirq_disable_event = 0;
1295         p->hardirq_context = 0;
1296         p->softirq_context = 0;
1297 #endif
1298 #ifdef CONFIG_LOCKDEP
1299         p->lockdep_depth = 0; /* no locks held yet */
1300         p->curr_chain_key = 0;
1301         p->lockdep_recursion = 0;
1302 #endif
1303
1304 #ifdef CONFIG_DEBUG_MUTEXES
1305         p->blocked_on = NULL; /* not blocked yet */
1306 #endif
1307 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1308         p->memcg_batch.do_batch = 0;
1309         p->memcg_batch.memcg = NULL;
1310 #endif
1311
1312         /* Perform scheduler related setup. Assign this task to a CPU. */
1313         sched_fork(p);
1314
1315         retval = perf_event_init_task(p);
1316         if (retval)
1317                 goto bad_fork_cleanup_policy;
1318         retval = audit_alloc(p);
1319         if (retval)
1320                 goto bad_fork_cleanup_policy;
1321         /* copy all the process information */
1322         retval = copy_semundo(clone_flags, p);
1323         if (retval)
1324                 goto bad_fork_cleanup_audit;
1325         retval = copy_files(clone_flags, p);
1326         if (retval)
1327                 goto bad_fork_cleanup_semundo;
1328         retval = copy_fs(clone_flags, p);
1329         if (retval)
1330                 goto bad_fork_cleanup_files;
1331         retval = copy_sighand(clone_flags, p);
1332         if (retval)
1333                 goto bad_fork_cleanup_fs;
1334         retval = copy_signal(clone_flags, p);
1335         if (retval)
1336                 goto bad_fork_cleanup_sighand;
1337         retval = copy_mm(clone_flags, p);
1338         if (retval)
1339                 goto bad_fork_cleanup_signal;
1340         retval = copy_namespaces(clone_flags, p);
1341         if (retval)
1342                 goto bad_fork_cleanup_mm;
1343         retval = copy_io(clone_flags, p);
1344         if (retval)
1345                 goto bad_fork_cleanup_namespaces;
1346         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1347         if (retval)
1348                 goto bad_fork_cleanup_io;
1349
1350         if (pid != &init_struct_pid) {
1351                 retval = -ENOMEM;
1352                 pid = alloc_pid(p->nsproxy->pid_ns);
1353                 if (!pid)
1354                         goto bad_fork_cleanup_io;
1355         }
1356
1357         p->pid = pid_nr(pid);
1358         p->tgid = p->pid;
1359         if (clone_flags & CLONE_THREAD)
1360                 p->tgid = current->tgid;
1361
1362         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1363         /*
1364          * Clear TID on mm_release()?
1365          */
1366         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1367 #ifdef CONFIG_BLOCK
1368         p->plug = NULL;
1369 #endif
1370 #ifdef CONFIG_FUTEX
1371         p->robust_list = NULL;
1372 #ifdef CONFIG_COMPAT
1373         p->compat_robust_list = NULL;
1374 #endif
1375         INIT_LIST_HEAD(&p->pi_state_list);
1376         p->pi_state_cache = NULL;
1377 #endif
1378         /*
1379          * sigaltstack should be cleared when sharing the same VM
1380          */
1381         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1382                 p->sas_ss_sp = p->sas_ss_size = 0;
1383
1384         /*
1385          * Syscall tracing and stepping should be turned off in the
1386          * child regardless of CLONE_PTRACE.
1387          */
1388         user_disable_single_step(p);
1389         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1390 #ifdef TIF_SYSCALL_EMU
1391         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1392 #endif
1393         clear_all_latency_tracing(p);
1394
1395         /* ok, now we should be set up.. */
1396         if (clone_flags & CLONE_THREAD)
1397                 p->exit_signal = -1;
1398         else if (clone_flags & CLONE_PARENT)
1399                 p->exit_signal = current->group_leader->exit_signal;
1400         else
1401                 p->exit_signal = (clone_flags & CSIGNAL);
1402
1403         p->pdeath_signal = 0;
1404         p->exit_state = 0;
1405
1406         p->nr_dirtied = 0;
1407         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1408         p->dirty_paused_when = 0;
1409
1410         /*
1411          * Ok, make it visible to the rest of the system.
1412          * We dont wake it up yet.
1413          */
1414         p->group_leader = p;
1415         INIT_LIST_HEAD(&p->thread_group);
1416
1417         /* Now that the task is set up, run cgroup callbacks if
1418          * necessary. We need to run them before the task is visible
1419          * on the tasklist. */
1420         cgroup_fork_callbacks(p);
1421         cgroup_callbacks_done = 1;
1422
1423         /* Need tasklist lock for parent etc handling! */
1424         write_lock_irq(&tasklist_lock);
1425
1426         /* CLONE_PARENT re-uses the old parent */
1427         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1428                 p->real_parent = current->real_parent;
1429                 p->parent_exec_id = current->parent_exec_id;
1430         } else {
1431                 p->real_parent = current;
1432                 p->parent_exec_id = current->self_exec_id;
1433         }
1434
1435         spin_lock(&current->sighand->siglock);
1436
1437         /*
1438          * Process group and session signals need to be delivered to just the
1439          * parent before the fork or both the parent and the child after the
1440          * fork. Restart if a signal comes in before we add the new process to
1441          * it's process group.
1442          * A fatal signal pending means that current will exit, so the new
1443          * thread can't slip out of an OOM kill (or normal SIGKILL).
1444         */
1445         recalc_sigpending();
1446         if (signal_pending(current)) {
1447                 spin_unlock(&current->sighand->siglock);
1448                 write_unlock_irq(&tasklist_lock);
1449                 retval = -ERESTARTNOINTR;
1450                 goto bad_fork_free_pid;
1451         }
1452
1453         if (clone_flags & CLONE_THREAD) {
1454                 current->signal->nr_threads++;
1455                 atomic_inc(&current->signal->live);
1456                 atomic_inc(&current->signal->sigcnt);
1457                 p->group_leader = current->group_leader;
1458                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1459         }
1460
1461         if (likely(p->pid)) {
1462                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1463
1464                 if (thread_group_leader(p)) {
1465                         if (is_child_reaper(pid))
1466                                 p->nsproxy->pid_ns->child_reaper = p;
1467
1468                         p->signal->leader_pid = pid;
1469                         p->signal->tty = tty_kref_get(current->signal->tty);
1470                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1471                         attach_pid(p, PIDTYPE_SID, task_session(current));
1472                         list_add_tail(&p->sibling, &p->real_parent->children);
1473                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1474                         __this_cpu_inc(process_counts);
1475                 }
1476                 attach_pid(p, PIDTYPE_PID, pid);
1477                 nr_threads++;
1478         }
1479
1480         total_forks++;
1481         spin_unlock(&current->sighand->siglock);
1482         write_unlock_irq(&tasklist_lock);
1483         proc_fork_connector(p);
1484         cgroup_post_fork(p);
1485         if (clone_flags & CLONE_THREAD)
1486                 threadgroup_change_end(current);
1487         perf_event_fork(p);
1488
1489         trace_task_newtask(p, clone_flags);
1490
1491         return p;
1492
1493 bad_fork_free_pid:
1494         if (pid != &init_struct_pid)
1495                 free_pid(pid);
1496 bad_fork_cleanup_io:
1497         if (p->io_context)
1498                 exit_io_context(p);
1499 bad_fork_cleanup_namespaces:
1500         if (unlikely(clone_flags & CLONE_NEWPID))
1501                 pid_ns_release_proc(p->nsproxy->pid_ns);
1502         exit_task_namespaces(p);
1503 bad_fork_cleanup_mm:
1504         if (p->mm)
1505                 mmput(p->mm);
1506 bad_fork_cleanup_signal:
1507         if (!(clone_flags & CLONE_THREAD))
1508                 free_signal_struct(p->signal);
1509 bad_fork_cleanup_sighand:
1510         __cleanup_sighand(p->sighand);
1511 bad_fork_cleanup_fs:
1512         exit_fs(p); /* blocking */
1513 bad_fork_cleanup_files:
1514         exit_files(p); /* blocking */
1515 bad_fork_cleanup_semundo:
1516         exit_sem(p);
1517 bad_fork_cleanup_audit:
1518         audit_free(p);
1519 bad_fork_cleanup_policy:
1520         perf_event_free_task(p);
1521 #ifdef CONFIG_NUMA
1522         mpol_put(p->mempolicy);
1523 bad_fork_cleanup_cgroup:
1524 #endif
1525         if (clone_flags & CLONE_THREAD)
1526                 threadgroup_change_end(current);
1527         cgroup_exit(p, cgroup_callbacks_done);
1528         delayacct_tsk_free(p);
1529         module_put(task_thread_info(p)->exec_domain->module);
1530 bad_fork_cleanup_count:
1531         atomic_dec(&p->cred->user->processes);
1532         exit_creds(p);
1533 bad_fork_free:
1534         free_task(p);
1535 fork_out:
1536         return ERR_PTR(retval);
1537 }
1538
1539 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1540 {
1541         memset(regs, 0, sizeof(struct pt_regs));
1542         return regs;
1543 }
1544
1545 static inline void init_idle_pids(struct pid_link *links)
1546 {
1547         enum pid_type type;
1548
1549         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1550                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1551                 links[type].pid = &init_struct_pid;
1552         }
1553 }
1554
1555 struct task_struct * __cpuinit fork_idle(int cpu)
1556 {
1557         struct task_struct *task;
1558         struct pt_regs regs;
1559
1560         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1561                             &init_struct_pid, 0);
1562         if (!IS_ERR(task)) {
1563                 init_idle_pids(task->pids);
1564                 init_idle(task, cpu);
1565         }
1566
1567         return task;
1568 }
1569
1570 /*
1571  *  Ok, this is the main fork-routine.
1572  *
1573  * It copies the process, and if successful kick-starts
1574  * it and waits for it to finish using the VM if required.
1575  */
1576 long do_fork(unsigned long clone_flags,
1577               unsigned long stack_start,
1578               struct pt_regs *regs,
1579               unsigned long stack_size,
1580               int __user *parent_tidptr,
1581               int __user *child_tidptr)
1582 {
1583         struct task_struct *p;
1584         int trace = 0;
1585         long nr;
1586
1587         /*
1588          * Do some preliminary argument and permissions checking before we
1589          * actually start allocating stuff
1590          */
1591         if (clone_flags & CLONE_NEWUSER) {
1592                 if (clone_flags & CLONE_THREAD)
1593                         return -EINVAL;
1594                 /* hopefully this check will go away when userns support is
1595                  * complete
1596                  */
1597                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1598                                 !capable(CAP_SETGID))
1599                         return -EPERM;
1600         }
1601
1602         /*
1603          * Determine whether and which event to report to ptracer.  When
1604          * called from kernel_thread or CLONE_UNTRACED is explicitly
1605          * requested, no event is reported; otherwise, report if the event
1606          * for the type of forking is enabled.
1607          */
1608         if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1609                 if (clone_flags & CLONE_VFORK)
1610                         trace = PTRACE_EVENT_VFORK;
1611                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1612                         trace = PTRACE_EVENT_CLONE;
1613                 else
1614                         trace = PTRACE_EVENT_FORK;
1615
1616                 if (likely(!ptrace_event_enabled(current, trace)))
1617                         trace = 0;
1618         }
1619
1620         p = copy_process(clone_flags, stack_start, regs, stack_size,
1621                          child_tidptr, NULL, trace);
1622         /*
1623          * Do this prior waking up the new thread - the thread pointer
1624          * might get invalid after that point, if the thread exits quickly.
1625          */
1626         if (!IS_ERR(p)) {
1627                 struct completion vfork;
1628
1629                 trace_sched_process_fork(current, p);
1630
1631                 nr = task_pid_vnr(p);
1632
1633                 if (clone_flags & CLONE_PARENT_SETTID)
1634                         put_user(nr, parent_tidptr);
1635
1636                 if (clone_flags & CLONE_VFORK) {
1637                         p->vfork_done = &vfork;
1638                         init_completion(&vfork);
1639                         get_task_struct(p);
1640                 }
1641
1642                 wake_up_new_task(p);
1643
1644                 /* forking complete and child started to run, tell ptracer */
1645                 if (unlikely(trace))
1646                         ptrace_event(trace, nr);
1647
1648                 if (clone_flags & CLONE_VFORK) {
1649                         if (!wait_for_vfork_done(p, &vfork))
1650                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1651                 }
1652         } else {
1653                 nr = PTR_ERR(p);
1654         }
1655         return nr;
1656 }
1657
1658 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1659 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1660 #endif
1661
1662 static void sighand_ctor(void *data)
1663 {
1664         struct sighand_struct *sighand = data;
1665
1666         spin_lock_init(&sighand->siglock);
1667         init_waitqueue_head(&sighand->signalfd_wqh);
1668 }
1669
1670 void __init proc_caches_init(void)
1671 {
1672         sighand_cachep = kmem_cache_create("sighand_cache",
1673                         sizeof(struct sighand_struct), 0,
1674                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1675                         SLAB_NOTRACK, sighand_ctor);
1676         signal_cachep = kmem_cache_create("signal_cache",
1677                         sizeof(struct signal_struct), 0,
1678                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1679         files_cachep = kmem_cache_create("files_cache",
1680                         sizeof(struct files_struct), 0,
1681                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1682         fs_cachep = kmem_cache_create("fs_cache",
1683                         sizeof(struct fs_struct), 0,
1684                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1685         /*
1686          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1687          * whole struct cpumask for the OFFSTACK case. We could change
1688          * this to *only* allocate as much of it as required by the
1689          * maximum number of CPU's we can ever have.  The cpumask_allocation
1690          * is at the end of the structure, exactly for that reason.
1691          */
1692         mm_cachep = kmem_cache_create("mm_struct",
1693                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1694                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1695         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1696         mmap_init();
1697         nsproxy_cache_init();
1698 }
1699
1700 /*
1701  * Check constraints on flags passed to the unshare system call.
1702  */
1703 static int check_unshare_flags(unsigned long unshare_flags)
1704 {
1705         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1706                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1707                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1708                 return -EINVAL;
1709         /*
1710          * Not implemented, but pretend it works if there is nothing to
1711          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1712          * needs to unshare vm.
1713          */
1714         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1715                 /* FIXME: get_task_mm() increments ->mm_users */
1716                 if (atomic_read(&current->mm->mm_users) > 1)
1717                         return -EINVAL;
1718         }
1719
1720         return 0;
1721 }
1722
1723 /*
1724  * Unshare the filesystem structure if it is being shared
1725  */
1726 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1727 {
1728         struct fs_struct *fs = current->fs;
1729
1730         if (!(unshare_flags & CLONE_FS) || !fs)
1731                 return 0;
1732
1733         /* don't need lock here; in the worst case we'll do useless copy */
1734         if (fs->users == 1)
1735                 return 0;
1736
1737         *new_fsp = copy_fs_struct(fs);
1738         if (!*new_fsp)
1739                 return -ENOMEM;
1740
1741         return 0;
1742 }
1743
1744 /*
1745  * Unshare file descriptor table if it is being shared
1746  */
1747 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1748 {
1749         struct files_struct *fd = current->files;
1750         int error = 0;
1751
1752         if ((unshare_flags & CLONE_FILES) &&
1753             (fd && atomic_read(&fd->count) > 1)) {
1754                 *new_fdp = dup_fd(fd, &error);
1755                 if (!*new_fdp)
1756                         return error;
1757         }
1758
1759         return 0;
1760 }
1761
1762 /*
1763  * unshare allows a process to 'unshare' part of the process
1764  * context which was originally shared using clone.  copy_*
1765  * functions used by do_fork() cannot be used here directly
1766  * because they modify an inactive task_struct that is being
1767  * constructed. Here we are modifying the current, active,
1768  * task_struct.
1769  */
1770 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1771 {
1772         struct fs_struct *fs, *new_fs = NULL;
1773         struct files_struct *fd, *new_fd = NULL;
1774         struct nsproxy *new_nsproxy = NULL;
1775         int do_sysvsem = 0;
1776         int err;
1777
1778         err = check_unshare_flags(unshare_flags);
1779         if (err)
1780                 goto bad_unshare_out;
1781
1782         /*
1783          * If unsharing namespace, must also unshare filesystem information.
1784          */
1785         if (unshare_flags & CLONE_NEWNS)
1786                 unshare_flags |= CLONE_FS;
1787         /*
1788          * CLONE_NEWIPC must also detach from the undolist: after switching
1789          * to a new ipc namespace, the semaphore arrays from the old
1790          * namespace are unreachable.
1791          */
1792         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1793                 do_sysvsem = 1;
1794         err = unshare_fs(unshare_flags, &new_fs);
1795         if (err)
1796                 goto bad_unshare_out;
1797         err = unshare_fd(unshare_flags, &new_fd);
1798         if (err)
1799                 goto bad_unshare_cleanup_fs;
1800         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1801         if (err)
1802                 goto bad_unshare_cleanup_fd;
1803
1804         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1805                 if (do_sysvsem) {
1806                         /*
1807                          * CLONE_SYSVSEM is equivalent to sys_exit().
1808                          */
1809                         exit_sem(current);
1810                 }
1811
1812                 if (new_nsproxy) {
1813                         switch_task_namespaces(current, new_nsproxy);
1814                         new_nsproxy = NULL;
1815                 }
1816
1817                 task_lock(current);
1818
1819                 if (new_fs) {
1820                         fs = current->fs;
1821                         spin_lock(&fs->lock);
1822                         current->fs = new_fs;
1823                         if (--fs->users)
1824                                 new_fs = NULL;
1825                         else
1826                                 new_fs = fs;
1827                         spin_unlock(&fs->lock);
1828                 }
1829
1830                 if (new_fd) {
1831                         fd = current->files;
1832                         current->files = new_fd;
1833                         new_fd = fd;
1834                 }
1835
1836                 task_unlock(current);
1837         }
1838
1839         if (new_nsproxy)
1840                 put_nsproxy(new_nsproxy);
1841
1842 bad_unshare_cleanup_fd:
1843         if (new_fd)
1844                 put_files_struct(new_fd);
1845
1846 bad_unshare_cleanup_fs:
1847         if (new_fs)
1848                 free_fs_struct(new_fs);
1849
1850 bad_unshare_out:
1851         return err;
1852 }
1853
1854 /*
1855  *      Helper to unshare the files of the current task.
1856  *      We don't want to expose copy_files internals to
1857  *      the exec layer of the kernel.
1858  */
1859
1860 int unshare_files(struct files_struct **displaced)
1861 {
1862         struct task_struct *task = current;
1863         struct files_struct *copy = NULL;
1864         int error;
1865
1866         error = unshare_fd(CLONE_FILES, &copy);
1867         if (error || !copy) {
1868                 *displaced = NULL;
1869                 return error;
1870         }
1871         *displaced = task->files;
1872         task_lock(task);
1873         task->files = copy;
1874         task_unlock(task);
1875         return 0;
1876 }