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