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 #include <linux/aio.h>
  74
  75 #include <asm/pgtable.h>
  76 #include <asm/pgalloc.h>
  77 #include <asm/uaccess.h>
  78 #include <asm/mmu_context.h>
  79 #include <asm/cacheflush.h>
  80 #include <asm/tlbflush.h>
  81
  82 #include <trace/events/sched.h>
  83
  84 #define CREATE_TRACE_POINTS
  85 #include <trace/events/task.h>
  86
  87 /*
  88  * Protected counters by write_lock_irq(&tasklist_lock)
  89  */
  90 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  91 int nr_threads;                 /* The idle threads do not count.. */
  92
  93 int max_threads;                /* tunable limit on nr_threads */
  94
  95 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  96
  97 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  98
  99 #ifdef CONFIG_PROVE_RCU
 100 int lockdep_tasklist_lock_is_held(void)
 101 {
 102         return lockdep_is_held(&tasklist_lock);
 103 }
 104 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 105 #endif /* #ifdef CONFIG_PROVE_RCU */
 106
 107 int nr_processes(void)
 108 {
 109         int cpu;
 110         int total = 0;
 111
 112         for_each_possible_cpu(cpu)
 113                 total += per_cpu(process_counts, cpu);
 114
 115         return total;
 116 }
 117
 118 void __weak arch_release_task_struct(struct task_struct *tsk)
 119 {
 120 }
 121
 122 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 123 static struct kmem_cache *task_struct_cachep;
 124
 125 static inline struct task_struct *alloc_task_struct_node(int node)
 126 {
 127         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 128 }
 129
 130 static inline void free_task_struct(struct task_struct *tsk)
 131 {
 132         kmem_cache_free(task_struct_cachep, tsk);
 133 }
 134 #endif
 135
 136 void __weak arch_release_thread_info(struct thread_info *ti)
 137 {
 138 }
 139
 140 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
 141
 142 /*
 143  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 144  * kmemcache based allocator.
 145  */
 146 # if THREAD_SIZE >= PAGE_SIZE
 147 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 148                                                   int node)
 149 {
 150         struct page *page = alloc_pages_node(node, THREADINFO_GFP_ACCOUNTED,
 151                                              THREAD_SIZE_ORDER);
 152
 153         return page ? page_address(page) : NULL;
 154 }
 155
 156 static inline void free_thread_info(struct thread_info *ti)
 157 {
 158         free_memcg_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 159 }
 160 # else
 161 static struct kmem_cache *thread_info_cache;
 162
 163 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 164                                                   int node)
 165 {
 166         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
 167 }
 168
 169 static void free_thread_info(struct thread_info *ti)
 170 {
 171         kmem_cache_free(thread_info_cache, ti);
 172 }
 173
 174 void thread_info_cache_init(void)
 175 {
 176         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
 177                                               THREAD_SIZE, 0, NULL);
 178         BUG_ON(thread_info_cache == NULL);
 179 }
 180 # endif
 181 #endif
 182
 183 /* SLAB cache for signal_struct structures (tsk->signal) */
 184 static struct kmem_cache *signal_cachep;
 185
 186 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 187 struct kmem_cache *sighand_cachep;
 188
 189 /* SLAB cache for files_struct structures (tsk->files) */
 190 struct kmem_cache *files_cachep;
 191
 192 /* SLAB cache for fs_struct structures (tsk->fs) */
 193 struct kmem_cache *fs_cachep;
 194
 195 /* SLAB cache for vm_area_struct structures */
 196 struct kmem_cache *vm_area_cachep;
 197
 198 /* SLAB cache for mm_struct structures (tsk->mm) */
 199 static struct kmem_cache *mm_cachep;
 200
 201 /* Notifier list called when a task struct is freed */
 202 static ATOMIC_NOTIFIER_HEAD(task_free_notifier);
 203
 204 static void account_kernel_stack(struct thread_info *ti, int account)
 205 {
 206         struct zone *zone = page_zone(virt_to_page(ti));
 207
 208         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 209 }
 210
 211 void free_task(struct task_struct *tsk)
 212 {
 213         account_kernel_stack(tsk->stack, -1);
 214         arch_release_thread_info(tsk->stack);
 215         free_thread_info(tsk->stack);
 216         rt_mutex_debug_task_free(tsk);
 217         ftrace_graph_exit_task(tsk);
 218         put_seccomp_filter(tsk);
 219         arch_release_task_struct(tsk);
 220         free_task_struct(tsk);
 221 }
 222 EXPORT_SYMBOL(free_task);
 223
 224 static inline void free_signal_struct(struct signal_struct *sig)
 225 {
 226         taskstats_tgid_free(sig);
 227         sched_autogroup_exit(sig);
 228         kmem_cache_free(signal_cachep, sig);
 229 }
 230
 231 static inline void put_signal_struct(struct signal_struct *sig)
 232 {
 233         if (atomic_dec_and_test(&sig->sigcnt))
 234                 free_signal_struct(sig);
 235 }
 236
 237 int task_free_register(struct notifier_block *n)
 238 {
 239         return atomic_notifier_chain_register(&task_free_notifier, n);
 240 }
 241 EXPORT_SYMBOL(task_free_register);
 242
 243 int task_free_unregister(struct notifier_block *n)
 244 {
 245         return atomic_notifier_chain_unregister(&task_free_notifier, n);
 246 }
 247 EXPORT_SYMBOL(task_free_unregister);
 248
 249 void __put_task_struct(struct task_struct *tsk)
 250 {
 251         WARN_ON(!tsk->exit_state);
 252         WARN_ON(atomic_read(&tsk->usage));
 253         WARN_ON(tsk == current);
 254
 255         security_task_free(tsk);
 256         exit_creds(tsk);
 257         delayacct_tsk_free(tsk);
 258         put_signal_struct(tsk->signal);
 259
 260         atomic_notifier_call_chain(&task_free_notifier, 0, tsk);
 261         if (!profile_handoff_task(tsk))
 262                 free_task(tsk);
 263 }
 264 EXPORT_SYMBOL_GPL(__put_task_struct);
 265
 266 void __init __weak arch_task_cache_init(void) { }
 267
 268 void __init fork_init(unsigned long mempages)
 269 {
 270 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 271 #ifndef ARCH_MIN_TASKALIGN
 272 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 273 #endif
 274         /* create a slab on which task_structs can be allocated */
 275         task_struct_cachep =
 276                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 277                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 278 #endif
 279
 280         /* do the arch specific task caches init */
 281         arch_task_cache_init();
 282
 283         /*
 284          * The default maximum number of threads is set to a safe
 285          * value: the thread structures can take up at most half
 286          * of memory.
 287          */
 288         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 289
 290         /*
 291          * we need to allow at least 20 threads to boot a system
 292          */
 293         if (max_threads < 20)
 294                 max_threads = 20;
 295
 296         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 297         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 298         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 299                 init_task.signal->rlim[RLIMIT_NPROC];
 300 }
 301
 302 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 303                                                struct task_struct *src)
 304 {
 305         *dst = *src;
 306         return 0;
 307 }
 308
 309 static struct task_struct *dup_task_struct(struct task_struct *orig)
 310 {
 311         struct task_struct *tsk;
 312         struct thread_info *ti;
 313         unsigned long *stackend;
 314         int node = tsk_fork_get_node(orig);
 315         int err;
 316
 317         tsk = alloc_task_struct_node(node);
 318         if (!tsk)
 319                 return NULL;
 320
 321         ti = alloc_thread_info_node(tsk, node);
 322         if (!ti)
 323                 goto free_tsk;
 324
 325         err = arch_dup_task_struct(tsk, orig);
 326         if (err)
 327                 goto free_ti;
 328
 329         tsk->stack = ti;
 330
 331         setup_thread_stack(tsk, orig);
 332         clear_user_return_notifier(tsk);
 333         clear_tsk_need_resched(tsk);
 334         stackend = end_of_stack(tsk);
 335         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 336
 337 #ifdef CONFIG_CC_STACKPROTECTOR
 338         tsk->stack_canary = get_random_int();
 339 #endif
 340
 341         /*
 342          * One for us, one for whoever does the "release_task()" (usually
 343          * parent)
 344          */
 345         atomic_set(&tsk->usage, 2);
 346 #ifdef CONFIG_BLK_DEV_IO_TRACE
 347         tsk->btrace_seq = 0;
 348 #endif
 349         tsk->splice_pipe = NULL;
 350         tsk->task_frag.page = NULL;
 351
 352         account_kernel_stack(ti, 1);
 353
 354         return tsk;
 355
 356 free_ti:
 357         free_thread_info(ti);
 358 free_tsk:
 359         free_task_struct(tsk);
 360         return NULL;
 361 }
 362
 363 #ifdef CONFIG_MMU
 364 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 365 {
 366         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 367         struct rb_node **rb_link, *rb_parent;
 368         int retval;
 369         unsigned long charge;
 370         struct mempolicy *pol;
 371
 372         uprobe_start_dup_mmap();
 373         down_write(&oldmm->mmap_sem);
 374         flush_cache_dup_mm(oldmm);
 375         uprobe_dup_mmap(oldmm, mm);
 376         /*
 377          * Not linked in yet - no deadlock potential:
 378          */
 379         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 380
 381         mm->locked_vm = 0;
 382         mm->mmap = NULL;
 383         mm->mmap_cache = NULL;
 384         mm->free_area_cache = oldmm->mmap_base;
 385         mm->cached_hole_size = ~0UL;
 386         mm->map_count = 0;
 387         cpumask_clear(mm_cpumask(mm));
 388         mm->mm_rb = RB_ROOT;
 389         rb_link = &mm->mm_rb.rb_node;
 390         rb_parent = NULL;
 391         pprev = &mm->mmap;
 392         retval = ksm_fork(mm, oldmm);
 393         if (retval)
 394                 goto out;
 395         retval = khugepaged_fork(mm, oldmm);
 396         if (retval)
 397                 goto out;
 398
 399         prev = NULL;
 400         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 401                 struct file *file;
 402
 403                 if (mpnt->vm_flags & VM_DONTCOPY) {
 404                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 405                                                         -vma_pages(mpnt));
 406                         continue;
 407                 }
 408                 charge = 0;
 409                 if (mpnt->vm_flags & VM_ACCOUNT) {
 410                         unsigned long len = vma_pages(mpnt);
 411
 412                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 413                                 goto fail_nomem;
 414                         charge = len;
 415                 }
 416                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 417                 if (!tmp)
 418                         goto fail_nomem;
 419                 *tmp = *mpnt;
 420                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 421                 pol = mpol_dup(vma_policy(mpnt));
 422                 retval = PTR_ERR(pol);
 423                 if (IS_ERR(pol))
 424                         goto fail_nomem_policy;
 425                 vma_set_policy(tmp, pol);
 426                 tmp->vm_mm = mm;
 427                 if (anon_vma_fork(tmp, mpnt))
 428                         goto fail_nomem_anon_vma_fork;
 429                 tmp->vm_flags &= ~VM_LOCKED;
 430                 tmp->vm_next = tmp->vm_prev = NULL;
 431                 file = tmp->vm_file;
 432                 if (file) {
 433                         struct inode *inode = file_inode(file);
 434                         struct address_space *mapping = file->f_mapping;
 435
 436                         get_file(file);
 437                         if (tmp->vm_flags & VM_DENYWRITE)
 438                                 atomic_dec(&inode->i_writecount);
 439                         mutex_lock(&mapping->i_mmap_mutex);
 440                         if (tmp->vm_flags & VM_SHARED)
 441                                 mapping->i_mmap_writable++;
 442                         flush_dcache_mmap_lock(mapping);
 443                         /* insert tmp into the share list, just after mpnt */
 444                         if (unlikely(tmp->vm_flags & VM_NONLINEAR))
 445                                 vma_nonlinear_insert(tmp,
 446                                                 &mapping->i_mmap_nonlinear);
 447                         else
 448                                 vma_interval_tree_insert_after(tmp, mpnt,
 449                                                         &mapping->i_mmap);
 450                         flush_dcache_mmap_unlock(mapping);
 451                         mutex_unlock(&mapping->i_mmap_mutex);
 452                 }
 453
 454                 /*
 455                  * Clear hugetlb-related page reserves for children. This only
 456                  * affects MAP_PRIVATE mappings. Faults generated by the child
 457                  * are not guaranteed to succeed, even if read-only
 458                  */
 459                 if (is_vm_hugetlb_page(tmp))
 460                         reset_vma_resv_huge_pages(tmp);
 461
 462                 /*
 463                  * Link in the new vma and copy the page table entries.
 464                  */
 465                 *pprev = tmp;
 466                 pprev = &tmp->vm_next;
 467                 tmp->vm_prev = prev;
 468                 prev = tmp;
 469
 470                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 471                 rb_link = &tmp->vm_rb.rb_right;
 472                 rb_parent = &tmp->vm_rb;
 473
 474                 mm->map_count++;
 475                 retval = copy_page_range(mm, oldmm, mpnt);
 476
 477                 if (tmp->vm_ops && tmp->vm_ops->open)
 478                         tmp->vm_ops->open(tmp);
 479
 480                 if (retval)
 481                         goto out;
 482         }
 483         /* a new mm has just been created */
 484         arch_dup_mmap(oldmm, mm);
 485         retval = 0;
 486 out:
 487         up_write(&mm->mmap_sem);
 488         flush_tlb_mm(oldmm);
 489         up_write(&oldmm->mmap_sem);
 490         uprobe_end_dup_mmap();
 491         return retval;
 492 fail_nomem_anon_vma_fork:
 493         mpol_put(pol);
 494 fail_nomem_policy:
 495         kmem_cache_free(vm_area_cachep, tmp);
 496 fail_nomem:
 497         retval = -ENOMEM;
 498         vm_unacct_memory(charge);
 499         goto out;
 500 }
 501
 502 static inline int mm_alloc_pgd(struct mm_struct *mm)
 503 {
 504         mm->pgd = pgd_alloc(mm);
 505         if (unlikely(!mm->pgd))
 506                 return -ENOMEM;
 507         return 0;
 508 }
 509
 510 static inline void mm_free_pgd(struct mm_struct *mm)
 511 {
 512         pgd_free(mm, mm->pgd);
 513 }
 514 #else
 515 #define dup_mmap(mm, oldmm)     (0)
 516 #define mm_alloc_pgd(mm)        (0)
 517 #define mm_free_pgd(mm)
 518 #endif /* CONFIG_MMU */
 519
 520 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 521
 522 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 523 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 524
 525 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 526
 527 static int __init coredump_filter_setup(char *s)
 528 {
 529         default_dump_filter =
 530                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 531                 MMF_DUMP_FILTER_MASK;
 532         return 1;
 533 }
 534
 535 __setup("coredump_filter=", coredump_filter_setup);
 536
 537 #include <linux/init_task.h>
 538
 539 static void mm_init_aio(struct mm_struct *mm)
 540 {
 541 #ifdef CONFIG_AIO
 542         spin_lock_init(&mm->ioctx_lock);
 543         INIT_HLIST_HEAD(&mm->ioctx_list);
 544 #endif
 545 }
 546
 547 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 548 {
 549         atomic_set(&mm->mm_users, 1);
 550         atomic_set(&mm->mm_count, 1);
 551         init_rwsem(&mm->mmap_sem);
 552         INIT_LIST_HEAD(&mm->mmlist);
 553         mm->flags = (current->mm) ?
 554                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 555         mm->core_state = NULL;
 556         mm->nr_ptes = 0;
 557         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 558         spin_lock_init(&mm->page_table_lock);
 559         mm->free_area_cache = TASK_UNMAPPED_BASE;
 560         mm->cached_hole_size = ~0UL;
 561         mm_init_aio(mm);
 562         mm_init_owner(mm, p);
 563
 564         if (likely(!mm_alloc_pgd(mm))) {
 565                 mm->def_flags = 0;
 566                 mmu_notifier_mm_init(mm);
 567                 return mm;
 568         }
 569
 570         free_mm(mm);
 571         return NULL;
 572 }
 573
 574 static void check_mm(struct mm_struct *mm)
 575 {
 576         int i;
 577
 578         for (i = 0; i < NR_MM_COUNTERS; i++) {
 579                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 580
 581                 if (unlikely(x))
 582                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 583                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 584         }
 585
 586 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 587         VM_BUG_ON(mm->pmd_huge_pte);
 588 #endif
 589 }
 590
 591 /*
 592  * Allocate and initialize an mm_struct.
 593  */
 594 struct mm_struct *mm_alloc(void)
 595 {
 596         struct mm_struct *mm;
 597
 598         mm = allocate_mm();
 599         if (!mm)
 600                 return NULL;
 601
 602         memset(mm, 0, sizeof(*mm));
 603         mm_init_cpumask(mm);
 604         return mm_init(mm, current);
 605 }
 606
 607 /*
 608  * Called when the last reference to the mm
 609  * is dropped: either by a lazy thread or by
 610  * mmput. Free the page directory and the mm.
 611  */
 612 void __mmdrop(struct mm_struct *mm)
 613 {
 614         BUG_ON(mm == &init_mm);
 615         mm_free_pgd(mm);
 616         destroy_context(mm);
 617         mmu_notifier_mm_destroy(mm);
 618         check_mm(mm);
 619         free_mm(mm);
 620 }
 621 EXPORT_SYMBOL_GPL(__mmdrop);
 622
 623 /*
 624  * Decrement the use count and release all resources for an mm.
 625  */
 626 void mmput(struct mm_struct *mm)
 627 {
 628         might_sleep();
 629
 630         if (atomic_dec_and_test(&mm->mm_users)) {
 631                 uprobe_clear_state(mm);
 632                 exit_aio(mm);
 633                 ksm_exit(mm);
 634                 khugepaged_exit(mm); /* must run before exit_mmap */
 635                 exit_mmap(mm);
 636                 set_mm_exe_file(mm, NULL);
 637                 if (!list_empty(&mm->mmlist)) {
 638                         spin_lock(&mmlist_lock);
 639                         list_del(&mm->mmlist);
 640                         spin_unlock(&mmlist_lock);
 641                 }
 642                 if (mm->binfmt)
 643                         module_put(mm->binfmt->module);
 644                 mmdrop(mm);
 645         }
 646 }
 647 EXPORT_SYMBOL_GPL(mmput);
 648
 649 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 650 {
 651         if (new_exe_file)
 652                 get_file(new_exe_file);
 653         if (mm->exe_file)
 654                 fput(mm->exe_file);
 655         mm->exe_file = new_exe_file;
 656 }
 657
 658 struct file *get_mm_exe_file(struct mm_struct *mm)
 659 {
 660         struct file *exe_file;
 661
 662         /* We need mmap_sem to protect against races with removal of exe_file */
 663         down_read(&mm->mmap_sem);
 664         exe_file = mm->exe_file;
 665         if (exe_file)
 666                 get_file(exe_file);
 667         up_read(&mm->mmap_sem);
 668         return exe_file;
 669 }
 670
 671 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 672 {
 673         /* It's safe to write the exe_file pointer without exe_file_lock because
 674          * this is called during fork when the task is not yet in /proc */
 675         newmm->exe_file = get_mm_exe_file(oldmm);
 676 }
 677
 678 /**
 679  * get_task_mm - acquire a reference to the task's mm
 680  *
 681  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 682  * this kernel workthread has transiently adopted a user mm with use_mm,
 683  * to do its AIO) is not set and if so returns a reference to it, after
 684  * bumping up the use count.  User must release the mm via mmput()
 685  * after use.  Typically used by /proc and ptrace.
 686  */
 687 struct mm_struct *get_task_mm(struct task_struct *task)
 688 {
 689         struct mm_struct *mm;
 690
 691         task_lock(task);
 692         mm = task->mm;
 693         if (mm) {
 694                 if (task->flags & PF_KTHREAD)
 695                         mm = NULL;
 696                 else
 697                         atomic_inc(&mm->mm_users);
 698         }
 699         task_unlock(task);
 700         return mm;
 701 }
 702 EXPORT_SYMBOL_GPL(get_task_mm);
 703
 704 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 705 {
 706         struct mm_struct *mm;
 707         int err;
 708
 709         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 710         if (err)
 711                 return ERR_PTR(err);
 712
 713         mm = get_task_mm(task);
 714         if (mm && mm != current->mm &&
 715                         !ptrace_may_access(task, mode) &&
 716                         !capable(CAP_SYS_RESOURCE)) {
 717                 mmput(mm);
 718                 mm = ERR_PTR(-EACCES);
 719         }
 720         mutex_unlock(&task->signal->cred_guard_mutex);
 721
 722         return mm;
 723 }
 724
 725 static void complete_vfork_done(struct task_struct *tsk)
 726 {
 727         struct completion *vfork;
 728
 729         task_lock(tsk);
 730         vfork = tsk->vfork_done;
 731         if (likely(vfork)) {
 732                 tsk->vfork_done = NULL;
 733                 complete(vfork);
 734         }
 735         task_unlock(tsk);
 736 }
 737
 738 static int wait_for_vfork_done(struct task_struct *child,
 739                                 struct completion *vfork)
 740 {
 741         int killed;
 742
 743         freezer_do_not_count();
 744         killed = wait_for_completion_killable(vfork);
 745         freezer_count();
 746
 747         if (killed) {
 748                 task_lock(child);
 749                 child->vfork_done = NULL;
 750                 task_unlock(child);
 751         }
 752
 753         put_task_struct(child);
 754         return killed;
 755 }
 756
 757 /* Please note the differences between mmput and mm_release.
 758  * mmput is called whenever we stop holding onto a mm_struct,
 759  * error success whatever.
 760  *
 761  * mm_release is called after a mm_struct has been removed
 762  * from the current process.
 763  *
 764  * This difference is important for error handling, when we
 765  * only half set up a mm_struct for a new process and need to restore
 766  * the old one.  Because we mmput the new mm_struct before
 767  * restoring the old one. . .
 768  * Eric Biederman 10 January 1998
 769  */
 770 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 771 {
 772         /* Get rid of any futexes when releasing the mm */
 773 #ifdef CONFIG_FUTEX
 774         if (unlikely(tsk->robust_list)) {
 775                 exit_robust_list(tsk);
 776                 tsk->robust_list = NULL;
 777         }
 778 #ifdef CONFIG_COMPAT
 779         if (unlikely(tsk->compat_robust_list)) {
 780                 compat_exit_robust_list(tsk);
 781                 tsk->compat_robust_list = NULL;
 782         }
 783 #endif
 784         if (unlikely(!list_empty(&tsk->pi_state_list)))
 785                 exit_pi_state_list(tsk);
 786 #endif
 787
 788         uprobe_free_utask(tsk);
 789
 790         /* Get rid of any cached register state */
 791         deactivate_mm(tsk, mm);
 792
 793         /*
 794          * If we're exiting normally, clear a user-space tid field if
 795          * requested.  We leave this alone when dying by signal, to leave
 796          * the value intact in a core dump, and to save the unnecessary
 797          * trouble, say, a killed vfork parent shouldn't touch this mm.
 798          * Userland only wants this done for a sys_exit.
 799          */
 800         if (tsk->clear_child_tid) {
 801                 if (!(tsk->flags & PF_SIGNALED) &&
 802                     atomic_read(&mm->mm_users) > 1) {
 803                         /*
 804                          * We don't check the error code - if userspace has
 805                          * not set up a proper pointer then tough luck.
 806                          */
 807                         put_user(0, tsk->clear_child_tid);
 808                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 809                                         1, NULL, NULL, 0);
 810                 }
 811                 tsk->clear_child_tid = NULL;
 812         }
 813
 814         /*
 815          * All done, finally we can wake up parent and return this mm to him.
 816          * Also kthread_stop() uses this completion for synchronization.
 817          */
 818         if (tsk->vfork_done)
 819                 complete_vfork_done(tsk);
 820 }
 821
 822 /*
 823  * Allocate a new mm structure and copy contents from the
 824  * mm structure of the passed in task structure.
 825  */
 826 struct mm_struct *dup_mm(struct task_struct *tsk)
 827 {
 828         struct mm_struct *mm, *oldmm = current->mm;
 829         int err;
 830
 831         if (!oldmm)
 832                 return NULL;
 833
 834         mm = allocate_mm();
 835         if (!mm)
 836                 goto fail_nomem;
 837
 838         memcpy(mm, oldmm, sizeof(*mm));
 839         mm_init_cpumask(mm);
 840
 841 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 842         mm->pmd_huge_pte = NULL;
 843 #endif
 844 #ifdef CONFIG_NUMA_BALANCING
 845         mm->first_nid = NUMA_PTE_SCAN_INIT;
 846 #endif
 847         if (!mm_init(mm, tsk))
 848                 goto fail_nomem;
 849
 850         if (init_new_context(tsk, mm))
 851                 goto fail_nocontext;
 852
 853         dup_mm_exe_file(oldmm, mm);
 854
 855         err = dup_mmap(mm, oldmm);
 856         if (err)
 857                 goto free_pt;
 858
 859         mm->hiwater_rss = get_mm_rss(mm);
 860         mm->hiwater_vm = mm->total_vm;
 861
 862         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 863                 goto free_pt;
 864
 865         return mm;
 866
 867 free_pt:
 868         /* don't put binfmt in mmput, we haven't got module yet */
 869         mm->binfmt = NULL;
 870         mmput(mm);
 871
 872 fail_nomem:
 873         return NULL;
 874
 875 fail_nocontext:
 876         /*
 877          * If init_new_context() failed, we cannot use mmput() to free the mm
 878          * because it calls destroy_context()
 879          */
 880         mm_free_pgd(mm);
 881         free_mm(mm);
 882         return NULL;
 883 }
 884
 885 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 886 {
 887         struct mm_struct *mm, *oldmm;
 888         int retval;
 889
 890         tsk->min_flt = tsk->maj_flt = 0;
 891         tsk->nvcsw = tsk->nivcsw = 0;
 892 #ifdef CONFIG_DETECT_HUNG_TASK
 893         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 894 #endif
 895
 896         tsk->mm = NULL;
 897         tsk->active_mm = NULL;
 898
 899         /*
 900          * Are we cloning a kernel thread?
 901          *
 902          * We need to steal a active VM for that..
 903          */
 904         oldmm = current->mm;
 905         if (!oldmm)
 906                 return 0;
 907
 908         if (clone_flags & CLONE_VM) {
 909                 atomic_inc(&oldmm->mm_users);
 910                 mm = oldmm;
 911                 goto good_mm;
 912         }
 913
 914         retval = -ENOMEM;
 915         mm = dup_mm(tsk);
 916         if (!mm)
 917                 goto fail_nomem;
 918
 919 good_mm:
 920         tsk->mm = mm;
 921         tsk->active_mm = mm;
 922         return 0;
 923
 924 fail_nomem:
 925         return retval;
 926 }
 927
 928 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 929 {
 930         struct fs_struct *fs = current->fs;
 931         if (clone_flags & CLONE_FS) {
 932                 /* tsk->fs is already what we want */
 933                 spin_lock(&fs->lock);
 934                 if (fs->in_exec) {
 935                         spin_unlock(&fs->lock);
 936                         return -EAGAIN;
 937                 }
 938                 fs->users++;
 939                 spin_unlock(&fs->lock);
 940                 return 0;
 941         }
 942         tsk->fs = copy_fs_struct(fs);
 943         if (!tsk->fs)
 944                 return -ENOMEM;
 945         return 0;
 946 }
 947
 948 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 949 {
 950         struct files_struct *oldf, *newf;
 951         int error = 0;
 952
 953         /*
 954          * A background process may not have any files ...
 955          */
 956         oldf = current->files;
 957         if (!oldf)
 958                 goto out;
 959
 960         if (clone_flags & CLONE_FILES) {
 961                 atomic_inc(&oldf->count);
 962                 goto out;
 963         }
 964
 965         newf = dup_fd(oldf, &error);
 966         if (!newf)
 967                 goto out;
 968
 969         tsk->files = newf;
 970         error = 0;
 971 out:
 972         return error;
 973 }
 974
 975 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 976 {
 977 #ifdef CONFIG_BLOCK
 978         struct io_context *ioc = current->io_context;
 979         struct io_context *new_ioc;
 980
 981         if (!ioc)
 982                 return 0;
 983         /*
 984          * Share io context with parent, if CLONE_IO is set
 985          */
 986         if (clone_flags & CLONE_IO) {
 987                 ioc_task_link(ioc);
 988                 tsk->io_context = ioc;
 989         } else if (ioprio_valid(ioc->ioprio)) {
 990                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
 991                 if (unlikely(!new_ioc))
 992                         return -ENOMEM;
 993
 994                 new_ioc->ioprio = ioc->ioprio;
 995                 put_io_context(new_ioc);
 996         }
 997 #endif
 998         return 0;
 999 }
1000
1001 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1002 {
1003         struct sighand_struct *sig;
1004
1005         if (clone_flags & CLONE_SIGHAND) {
1006                 atomic_inc(&current->sighand->count);
1007                 return 0;
1008         }
1009         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1010         rcu_assign_pointer(tsk->sighand, sig);
1011         if (!sig)
1012                 return -ENOMEM;
1013         atomic_set(&sig->count, 1);
1014         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1015         return 0;
1016 }
1017
1018 void __cleanup_sighand(struct sighand_struct *sighand)
1019 {
1020         if (atomic_dec_and_test(&sighand->count)) {
1021                 signalfd_cleanup(sighand);
1022                 kmem_cache_free(sighand_cachep, sighand);
1023         }
1024 }
1025
1026
1027 /*
1028  * Initialize POSIX timer handling for a thread group.
1029  */
1030 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1031 {
1032         unsigned long cpu_limit;
1033
1034         /* Thread group counters. */
1035         thread_group_cputime_init(sig);
1036
1037         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1038         if (cpu_limit != RLIM_INFINITY) {
1039                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1040                 sig->cputimer.running = 1;
1041         }
1042
1043         /* The timer lists. */
1044         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1045         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1046         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1047 }
1048
1049 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1050 {
1051         struct signal_struct *sig;
1052
1053         if (clone_flags & CLONE_THREAD)
1054                 return 0;
1055
1056         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1057         tsk->signal = sig;
1058         if (!sig)
1059                 return -ENOMEM;
1060
1061         sig->nr_threads = 1;
1062         atomic_set(&sig->live, 1);
1063         atomic_set(&sig->sigcnt, 1);
1064         init_waitqueue_head(&sig->wait_chldexit);
1065         sig->curr_target = tsk;
1066         init_sigpending(&sig->shared_pending);
1067         INIT_LIST_HEAD(&sig->posix_timers);
1068
1069         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1070         sig->real_timer.function = it_real_fn;
1071
1072         task_lock(current->group_leader);
1073         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1074         task_unlock(current->group_leader);
1075
1076         posix_cpu_timers_init_group(sig);
1077
1078         tty_audit_fork(sig);
1079         sched_autogroup_fork(sig);
1080
1081 #ifdef CONFIG_CGROUPS
1082         init_rwsem(&sig->group_rwsem);
1083 #endif
1084
1085         sig->oom_score_adj = current->signal->oom_score_adj;
1086         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1087
1088         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1089                                    current->signal->is_child_subreaper;
1090
1091         mutex_init(&sig->cred_guard_mutex);
1092
1093         return 0;
1094 }
1095
1096 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1097 {
1098         unsigned long new_flags = p->flags;
1099
1100         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1101         new_flags |= PF_FORKNOEXEC;
1102         p->flags = new_flags;
1103 }
1104
1105 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1106 {
1107         current->clear_child_tid = tidptr;
1108
1109         return task_pid_vnr(current);
1110 }
1111
1112 static void rt_mutex_init_task(struct task_struct *p)
1113 {
1114         raw_spin_lock_init(&p->pi_lock);
1115 #ifdef CONFIG_RT_MUTEXES
1116         plist_head_init(&p->pi_waiters);
1117         p->pi_blocked_on = NULL;
1118 #endif
1119 }
1120
1121 #ifdef CONFIG_MM_OWNER
1122 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1123 {
1124         mm->owner = p;
1125 }
1126 #endif /* CONFIG_MM_OWNER */
1127
1128 /*
1129  * Initialize POSIX timer handling for a single task.
1130  */
1131 static void posix_cpu_timers_init(struct task_struct *tsk)
1132 {
1133         tsk->cputime_expires.prof_exp = 0;
1134         tsk->cputime_expires.virt_exp = 0;
1135         tsk->cputime_expires.sched_exp = 0;
1136         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1137         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1138         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1139 }
1140
1141 /*
1142  * This creates a new process as a copy of the old one,
1143  * but does not actually start it yet.
1144  *
1145  * It copies the registers, and all the appropriate
1146  * parts of the process environment (as per the clone
1147  * flags). The actual kick-off is left to the caller.
1148  */
1149 static struct task_struct *copy_process(unsigned long clone_flags,
1150                                         unsigned long stack_start,
1151                                         unsigned long stack_size,
1152                                         int __user *child_tidptr,
1153                                         struct pid *pid,
1154                                         int trace)
1155 {
1156         int retval;
1157         struct task_struct *p;
1158
1159         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1160                 return ERR_PTR(-EINVAL);
1161
1162         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1163                 return ERR_PTR(-EINVAL);
1164
1165         /*
1166          * Thread groups must share signals as well, and detached threads
1167          * can only be started up within the thread group.
1168          */
1169         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1170                 return ERR_PTR(-EINVAL);
1171
1172         /*
1173          * Shared signal handlers imply shared VM. By way of the above,
1174          * thread groups also imply shared VM. Blocking this case allows
1175          * for various simplifications in other code.
1176          */
1177         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1178                 return ERR_PTR(-EINVAL);
1179
1180         /*
1181          * Siblings of global init remain as zombies on exit since they are
1182          * not reaped by their parent (swapper). To solve this and to avoid
1183          * multi-rooted process trees, prevent global and container-inits
1184          * from creating siblings.
1185          */
1186         if ((clone_flags & CLONE_PARENT) &&
1187                                 current->signal->flags & SIGNAL_UNKILLABLE)
1188                 return ERR_PTR(-EINVAL);
1189
1190         /*
1191          * If the new process will be in a different pid namespace don't
1192          * allow it to share a thread group or signal handlers with the
1193          * forking task.
1194          */
1195         if ((clone_flags & (CLONE_SIGHAND | CLONE_NEWPID)) &&
1196             (task_active_pid_ns(current) != current->nsproxy->pid_ns))
1197                 return ERR_PTR(-EINVAL);
1198
1199         retval = security_task_create(clone_flags);
1200         if (retval)
1201                 goto fork_out;
1202
1203         retval = -ENOMEM;
1204         p = dup_task_struct(current);
1205         if (!p)
1206                 goto fork_out;
1207
1208         ftrace_graph_init_task(p);
1209         get_seccomp_filter(p);
1210
1211         rt_mutex_init_task(p);
1212
1213 #ifdef CONFIG_PROVE_LOCKING
1214         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1215         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1216 #endif
1217         retval = -EAGAIN;
1218         if (atomic_read(&p->real_cred->user->processes) >=
1219                         task_rlimit(p, RLIMIT_NPROC)) {
1220                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1221                     p->real_cred->user != INIT_USER)
1222                         goto bad_fork_free;
1223         }
1224         current->flags &= ~PF_NPROC_EXCEEDED;
1225
1226         retval = copy_creds(p, clone_flags);
1227         if (retval < 0)
1228                 goto bad_fork_free;
1229
1230         /*
1231          * If multiple threads are within copy_process(), then this check
1232          * triggers too late. This doesn't hurt, the check is only there
1233          * to stop root fork bombs.
1234          */
1235         retval = -EAGAIN;
1236         if (nr_threads >= max_threads)
1237                 goto bad_fork_cleanup_count;
1238
1239         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1240                 goto bad_fork_cleanup_count;
1241
1242         p->did_exec = 0;
1243         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1244         copy_flags(clone_flags, p);
1245         INIT_LIST_HEAD(&p->children);
1246         INIT_LIST_HEAD(&p->sibling);
1247         rcu_copy_process(p);
1248         p->vfork_done = NULL;
1249         spin_lock_init(&p->alloc_lock);
1250
1251         init_sigpending(&p->pending);
1252
1253         p->utime = p->stime = p->gtime = 0;
1254         p->utimescaled = p->stimescaled = 0;
1255 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1256         p->prev_cputime.utime = p->prev_cputime.stime = 0;
1257 #endif
1258 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1259         seqlock_init(&p->vtime_seqlock);
1260         p->vtime_snap = 0;
1261         p->vtime_snap_whence = VTIME_SLEEPING;
1262 #endif
1263
1264 #if defined(SPLIT_RSS_COUNTING)
1265         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1266 #endif
1267
1268         p->default_timer_slack_ns = current->timer_slack_ns;
1269
1270         task_io_accounting_init(&p->ioac);
1271         acct_clear_integrals(p);
1272
1273         posix_cpu_timers_init(p);
1274
1275         do_posix_clock_monotonic_gettime(&p->start_time);
1276         p->real_start_time = p->start_time;
1277         monotonic_to_bootbased(&p->real_start_time);
1278         p->io_context = NULL;
1279         p->audit_context = NULL;
1280         if (clone_flags & CLONE_THREAD)
1281                 threadgroup_change_begin(current);
1282         cgroup_fork(p);
1283 #ifdef CONFIG_NUMA
1284         p->mempolicy = mpol_dup(p->mempolicy);
1285         if (IS_ERR(p->mempolicy)) {
1286                 retval = PTR_ERR(p->mempolicy);
1287                 p->mempolicy = NULL;
1288                 goto bad_fork_cleanup_cgroup;
1289         }
1290         mpol_fix_fork_child_flag(p);
1291 #endif
1292 #ifdef CONFIG_CPUSETS
1293         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1294         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1295         seqcount_init(&p->mems_allowed_seq);
1296 #endif
1297 #ifdef CONFIG_TRACE_IRQFLAGS
1298         p->irq_events = 0;
1299         p->hardirqs_enabled = 0;
1300         p->hardirq_enable_ip = 0;
1301         p->hardirq_enable_event = 0;
1302         p->hardirq_disable_ip = _THIS_IP_;
1303         p->hardirq_disable_event = 0;
1304         p->softirqs_enabled = 1;
1305         p->softirq_enable_ip = _THIS_IP_;
1306         p->softirq_enable_event = 0;
1307         p->softirq_disable_ip = 0;
1308         p->softirq_disable_event = 0;
1309         p->hardirq_context = 0;
1310         p->softirq_context = 0;
1311 #endif
1312 #ifdef CONFIG_LOCKDEP
1313         p->lockdep_depth = 0; /* no locks held yet */
1314         p->curr_chain_key = 0;
1315         p->lockdep_recursion = 0;
1316 #endif
1317
1318 #ifdef CONFIG_DEBUG_MUTEXES
1319         p->blocked_on = NULL; /* not blocked yet */
1320 #endif
1321 #ifdef CONFIG_MEMCG
1322         p->memcg_batch.do_batch = 0;
1323         p->memcg_batch.memcg = NULL;
1324 #endif
1325 #ifdef CONFIG_BCACHE
1326         p->sequential_io        = 0;
1327         p->sequential_io_avg    = 0;
1328 #endif
1329
1330         /* Perform scheduler related setup. Assign this task to a CPU. */
1331         sched_fork(p);
1332
1333         retval = perf_event_init_task(p);
1334         if (retval)
1335                 goto bad_fork_cleanup_policy;
1336         retval = audit_alloc(p);
1337         if (retval)
1338                 goto bad_fork_cleanup_policy;
1339         /* copy all the process information */
1340         retval = copy_semundo(clone_flags, p);
1341         if (retval)
1342                 goto bad_fork_cleanup_audit;
1343         retval = copy_files(clone_flags, p);
1344         if (retval)
1345                 goto bad_fork_cleanup_semundo;
1346         retval = copy_fs(clone_flags, p);
1347         if (retval)
1348                 goto bad_fork_cleanup_files;
1349         retval = copy_sighand(clone_flags, p);
1350         if (retval)
1351                 goto bad_fork_cleanup_fs;
1352         retval = copy_signal(clone_flags, p);
1353         if (retval)
1354                 goto bad_fork_cleanup_sighand;
1355         retval = copy_mm(clone_flags, p);
1356         if (retval)
1357                 goto bad_fork_cleanup_signal;
1358         retval = copy_namespaces(clone_flags, p);
1359         if (retval)
1360                 goto bad_fork_cleanup_mm;
1361         retval = copy_io(clone_flags, p);
1362         if (retval)
1363                 goto bad_fork_cleanup_namespaces;
1364         retval = copy_thread(clone_flags, stack_start, stack_size, p);
1365         if (retval)
1366                 goto bad_fork_cleanup_io;
1367
1368         if (pid != &init_struct_pid) {
1369                 retval = -ENOMEM;
1370                 pid = alloc_pid(p->nsproxy->pid_ns);
1371                 if (!pid)
1372                         goto bad_fork_cleanup_io;
1373         }
1374
1375         p->pid = pid_nr(pid);
1376         p->tgid = p->pid;
1377         if (clone_flags & CLONE_THREAD)
1378                 p->tgid = current->tgid;
1379
1380         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1381         /*
1382          * Clear TID on mm_release()?
1383          */
1384         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1385 #ifdef CONFIG_BLOCK
1386         p->plug = NULL;
1387 #endif
1388 #ifdef CONFIG_FUTEX
1389         p->robust_list = NULL;
1390 #ifdef CONFIG_COMPAT
1391         p->compat_robust_list = NULL;
1392 #endif
1393         INIT_LIST_HEAD(&p->pi_state_list);
1394         p->pi_state_cache = NULL;
1395 #endif
1396         uprobe_copy_process(p);
1397         /*
1398          * sigaltstack should be cleared when sharing the same VM
1399          */
1400         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1401                 p->sas_ss_sp = p->sas_ss_size = 0;
1402
1403         /*
1404          * Syscall tracing and stepping should be turned off in the
1405          * child regardless of CLONE_PTRACE.
1406          */
1407         user_disable_single_step(p);
1408         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1409 #ifdef TIF_SYSCALL_EMU
1410         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1411 #endif
1412         clear_all_latency_tracing(p);
1413
1414         /* ok, now we should be set up.. */
1415         if (clone_flags & CLONE_THREAD)
1416                 p->exit_signal = -1;
1417         else if (clone_flags & CLONE_PARENT)
1418                 p->exit_signal = current->group_leader->exit_signal;
1419         else
1420                 p->exit_signal = (clone_flags & CSIGNAL);
1421
1422         p->pdeath_signal = 0;
1423         p->exit_state = 0;
1424
1425         p->nr_dirtied = 0;
1426         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1427         p->dirty_paused_when = 0;
1428
1429         /*
1430          * Ok, make it visible to the rest of the system.
1431          * We dont wake it up yet.
1432          */
1433         p->group_leader = p;
1434         INIT_LIST_HEAD(&p->thread_group);
1435         p->task_works = NULL;
1436
1437         /* Need tasklist lock for parent etc handling! */
1438         write_lock_irq(&tasklist_lock);
1439
1440         /* CLONE_PARENT re-uses the old parent */
1441         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1442                 p->real_parent = current->real_parent;
1443                 p->parent_exec_id = current->parent_exec_id;
1444         } else {
1445                 p->real_parent = current;
1446                 p->parent_exec_id = current->self_exec_id;
1447         }
1448
1449         spin_lock(&current->sighand->siglock);
1450
1451         /*
1452          * Process group and session signals need to be delivered to just the
1453          * parent before the fork or both the parent and the child after the
1454          * fork. Restart if a signal comes in before we add the new process to
1455          * it's process group.
1456          * A fatal signal pending means that current will exit, so the new
1457          * thread can't slip out of an OOM kill (or normal SIGKILL).
1458         */
1459         recalc_sigpending();
1460         if (signal_pending(current)) {
1461                 spin_unlock(&current->sighand->siglock);
1462                 write_unlock_irq(&tasklist_lock);
1463                 retval = -ERESTARTNOINTR;
1464                 goto bad_fork_free_pid;
1465         }
1466
1467         if (clone_flags & CLONE_THREAD) {
1468                 current->signal->nr_threads++;
1469                 atomic_inc(&current->signal->live);
1470                 atomic_inc(&current->signal->sigcnt);
1471                 p->group_leader = current->group_leader;
1472                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1473         }
1474
1475         if (likely(p->pid)) {
1476                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1477
1478                 if (thread_group_leader(p)) {
1479                         if (is_child_reaper(pid)) {
1480                                 ns_of_pid(pid)->child_reaper = p;
1481                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1482                         }
1483
1484                         p->signal->leader_pid = pid;
1485                         p->signal->tty = tty_kref_get(current->signal->tty);
1486                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1487                         attach_pid(p, PIDTYPE_SID, task_session(current));
1488                         list_add_tail(&p->sibling, &p->real_parent->children);
1489                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1490                         __this_cpu_inc(process_counts);
1491                 }
1492                 attach_pid(p, PIDTYPE_PID, pid);
1493                 nr_threads++;
1494         }
1495
1496         total_forks++;
1497         spin_unlock(&current->sighand->siglock);
1498         write_unlock_irq(&tasklist_lock);
1499         proc_fork_connector(p);
1500         cgroup_post_fork(p);
1501         if (clone_flags & CLONE_THREAD)
1502                 threadgroup_change_end(current);
1503         perf_event_fork(p);
1504
1505         trace_task_newtask(p, clone_flags);
1506
1507         return p;
1508
1509 bad_fork_free_pid:
1510         if (pid != &init_struct_pid)
1511                 free_pid(pid);
1512 bad_fork_cleanup_io:
1513         if (p->io_context)
1514                 exit_io_context(p);
1515 bad_fork_cleanup_namespaces:
1516         exit_task_namespaces(p);
1517 bad_fork_cleanup_mm:
1518         if (p->mm)
1519                 mmput(p->mm);
1520 bad_fork_cleanup_signal:
1521         if (!(clone_flags & CLONE_THREAD))
1522                 free_signal_struct(p->signal);
1523 bad_fork_cleanup_sighand:
1524         __cleanup_sighand(p->sighand);
1525 bad_fork_cleanup_fs:
1526         exit_fs(p); /* blocking */
1527 bad_fork_cleanup_files:
1528         exit_files(p); /* blocking */
1529 bad_fork_cleanup_semundo:
1530         exit_sem(p);
1531 bad_fork_cleanup_audit:
1532         audit_free(p);
1533 bad_fork_cleanup_policy:
1534         perf_event_free_task(p);
1535 #ifdef CONFIG_NUMA
1536         mpol_put(p->mempolicy);
1537 bad_fork_cleanup_cgroup:
1538 #endif
1539         if (clone_flags & CLONE_THREAD)
1540                 threadgroup_change_end(current);
1541         cgroup_exit(p, 0);
1542         delayacct_tsk_free(p);
1543         module_put(task_thread_info(p)->exec_domain->module);
1544 bad_fork_cleanup_count:
1545         atomic_dec(&p->cred->user->processes);
1546         exit_creds(p);
1547 bad_fork_free:
1548         free_task(p);
1549 fork_out:
1550         return ERR_PTR(retval);
1551 }
1552
1553 static inline void init_idle_pids(struct pid_link *links)
1554 {
1555         enum pid_type type;
1556
1557         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1558                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1559                 links[type].pid = &init_struct_pid;
1560         }
1561 }
1562
1563 struct task_struct * __cpuinit fork_idle(int cpu)
1564 {
1565         struct task_struct *task;
1566         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1567         if (!IS_ERR(task)) {
1568                 init_idle_pids(task->pids);
1569                 init_idle(task, cpu);
1570         }
1571
1572         return task;
1573 }
1574
1575 /*
1576  *  Ok, this is the main fork-routine.
1577  *
1578  * It copies the process, and if successful kick-starts
1579  * it and waits for it to finish using the VM if required.
1580  */
1581 long do_fork(unsigned long clone_flags,
1582               unsigned long stack_start,
1583               unsigned long stack_size,
1584               int __user *parent_tidptr,
1585               int __user *child_tidptr)
1586 {
1587         struct task_struct *p;
1588         int trace = 0;
1589         long nr;
1590
1591         /*
1592          * Do some preliminary argument and permissions checking before we
1593          * actually start allocating stuff
1594          */
1595         if (clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) {
1596                 if (clone_flags & (CLONE_THREAD|CLONE_PARENT))
1597                         return -EINVAL;
1598         }
1599
1600         /*
1601          * Determine whether and which event to report to ptracer.  When
1602          * called from kernel_thread or CLONE_UNTRACED is explicitly
1603          * requested, no event is reported; otherwise, report if the event
1604          * for the type of forking is enabled.
1605          */
1606         if (!(clone_flags & CLONE_UNTRACED)) {
1607                 if (clone_flags & CLONE_VFORK)
1608                         trace = PTRACE_EVENT_VFORK;
1609                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1610                         trace = PTRACE_EVENT_CLONE;
1611                 else
1612                         trace = PTRACE_EVENT_FORK;
1613
1614                 if (likely(!ptrace_event_enabled(current, trace)))
1615                         trace = 0;
1616         }
1617
1618         p = copy_process(clone_flags, stack_start, stack_size,
1619                          child_tidptr, NULL, trace);
1620         /*
1621          * Do this prior waking up the new thread - the thread pointer
1622          * might get invalid after that point, if the thread exits quickly.
1623          */
1624         if (!IS_ERR(p)) {
1625                 struct completion vfork;
1626
1627                 trace_sched_process_fork(current, p);
1628
1629                 nr = task_pid_vnr(p);
1630
1631                 if (clone_flags & CLONE_PARENT_SETTID)
1632                         put_user(nr, parent_tidptr);
1633
1634                 if (clone_flags & CLONE_VFORK) {
1635                         p->vfork_done = &vfork;
1636                         init_completion(&vfork);
1637                         get_task_struct(p);
1638                 }
1639
1640                 wake_up_new_task(p);
1641
1642                 /* forking complete and child started to run, tell ptracer */
1643                 if (unlikely(trace))
1644                         ptrace_event(trace, nr);
1645
1646                 if (clone_flags & CLONE_VFORK) {
1647                         if (!wait_for_vfork_done(p, &vfork))
1648                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1649                 }
1650         } else {
1651                 nr = PTR_ERR(p);
1652         }
1653         return nr;
1654 }
1655
1656 /*
1657  * Create a kernel thread.
1658  */
1659 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1660 {
1661         return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1662                 (unsigned long)arg, NULL, NULL);
1663 }
1664
1665 #ifdef __ARCH_WANT_SYS_FORK
1666 SYSCALL_DEFINE0(fork)
1667 {
1668 #ifdef CONFIG_MMU
1669         return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1670 #else
1671         /* can not support in nommu mode */
1672         return(-EINVAL);
1673 #endif
1674 }
1675 #endif
1676
1677 #ifdef __ARCH_WANT_SYS_VFORK
1678 SYSCALL_DEFINE0(vfork)
1679 {
1680         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1681                         0, NULL, NULL);
1682 }
1683 #endif
1684
1685 #ifdef __ARCH_WANT_SYS_CLONE
1686 #ifdef CONFIG_CLONE_BACKWARDS
1687 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1688                  int __user *, parent_tidptr,
1689                  int, tls_val,
1690                  int __user *, child_tidptr)
1691 #elif defined(CONFIG_CLONE_BACKWARDS2)
1692 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1693                  int __user *, parent_tidptr,
1694                  int __user *, child_tidptr,
1695                  int, tls_val)
1696 #elif defined(CONFIG_CLONE_BACKWARDS3)
1697 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1698                 int, stack_size,
1699                 int __user *, parent_tidptr,
1700                 int __user *, child_tidptr,
1701                 int, tls_val)
1702 #else
1703 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1704                  int __user *, parent_tidptr,
1705                  int __user *, child_tidptr,
1706                  int, tls_val)
1707 #endif
1708 {
1709         return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1710 }
1711 #endif
1712
1713 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1714 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1715 #endif
1716
1717 static void sighand_ctor(void *data)
1718 {
1719         struct sighand_struct *sighand = data;
1720
1721         spin_lock_init(&sighand->siglock);
1722         init_waitqueue_head(&sighand->signalfd_wqh);
1723 }
1724
1725 void __init proc_caches_init(void)
1726 {
1727         sighand_cachep = kmem_cache_create("sighand_cache",
1728                         sizeof(struct sighand_struct), 0,
1729                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1730                         SLAB_NOTRACK, sighand_ctor);
1731         signal_cachep = kmem_cache_create("signal_cache",
1732                         sizeof(struct signal_struct), 0,
1733                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1734         files_cachep = kmem_cache_create("files_cache",
1735                         sizeof(struct files_struct), 0,
1736                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1737         fs_cachep = kmem_cache_create("fs_cache",
1738                         sizeof(struct fs_struct), 0,
1739                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1740         /*
1741          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1742          * whole struct cpumask for the OFFSTACK case. We could change
1743          * this to *only* allocate as much of it as required by the
1744          * maximum number of CPU's we can ever have.  The cpumask_allocation
1745          * is at the end of the structure, exactly for that reason.
1746          */
1747         mm_cachep = kmem_cache_create("mm_struct",
1748                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1749                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1750         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1751         mmap_init();
1752         nsproxy_cache_init();
1753 }
1754
1755 /*
1756  * Check constraints on flags passed to the unshare system call.
1757  */
1758 static int check_unshare_flags(unsigned long unshare_flags)
1759 {
1760         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1761                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1762                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1763                                 CLONE_NEWUSER|CLONE_NEWPID))
1764                 return -EINVAL;
1765         /*
1766          * Not implemented, but pretend it works if there is nothing to
1767          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1768          * needs to unshare vm.
1769          */
1770         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1771                 /* FIXME: get_task_mm() increments ->mm_users */
1772                 if (atomic_read(&current->mm->mm_users) > 1)
1773                         return -EINVAL;
1774         }
1775
1776         return 0;
1777 }
1778
1779 /*
1780  * Unshare the filesystem structure if it is being shared
1781  */
1782 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1783 {
1784         struct fs_struct *fs = current->fs;
1785
1786         if (!(unshare_flags & CLONE_FS) || !fs)
1787                 return 0;
1788
1789         /* don't need lock here; in the worst case we'll do useless copy */
1790         if (fs->users == 1)
1791                 return 0;
1792
1793         *new_fsp = copy_fs_struct(fs);
1794         if (!*new_fsp)
1795                 return -ENOMEM;
1796
1797         return 0;
1798 }
1799
1800 /*
1801  * Unshare file descriptor table if it is being shared
1802  */
1803 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1804 {
1805         struct files_struct *fd = current->files;
1806         int error = 0;
1807
1808         if ((unshare_flags & CLONE_FILES) &&
1809             (fd && atomic_read(&fd->count) > 1)) {
1810                 *new_fdp = dup_fd(fd, &error);
1811                 if (!*new_fdp)
1812                         return error;
1813         }
1814
1815         return 0;
1816 }
1817
1818 /*
1819  * unshare allows a process to 'unshare' part of the process
1820  * context which was originally shared using clone.  copy_*
1821  * functions used by do_fork() cannot be used here directly
1822  * because they modify an inactive task_struct that is being
1823  * constructed. Here we are modifying the current, active,
1824  * task_struct.
1825  */
1826 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1827 {
1828         struct fs_struct *fs, *new_fs = NULL;
1829         struct files_struct *fd, *new_fd = NULL;
1830         struct cred *new_cred = NULL;
1831         struct nsproxy *new_nsproxy = NULL;
1832         int do_sysvsem = 0;
1833         int err;
1834
1835         /*
1836          * If unsharing a user namespace must also unshare the thread.
1837          */
1838         if (unshare_flags & CLONE_NEWUSER)
1839                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1840         /*
1841          * If unsharing a pid namespace must also unshare the thread.
1842          */
1843         if (unshare_flags & CLONE_NEWPID)
1844                 unshare_flags |= CLONE_THREAD;
1845         /*
1846          * If unsharing a thread from a thread group, must also unshare vm.
1847          */
1848         if (unshare_flags & CLONE_THREAD)
1849                 unshare_flags |= CLONE_VM;
1850         /*
1851          * If unsharing vm, must also unshare signal handlers.
1852          */
1853         if (unshare_flags & CLONE_VM)
1854                 unshare_flags |= CLONE_SIGHAND;
1855         /*
1856          * If unsharing namespace, must also unshare filesystem information.
1857          */
1858         if (unshare_flags & CLONE_NEWNS)
1859                 unshare_flags |= CLONE_FS;
1860
1861         err = check_unshare_flags(unshare_flags);
1862         if (err)
1863                 goto bad_unshare_out;
1864         /*
1865          * CLONE_NEWIPC must also detach from the undolist: after switching
1866          * to a new ipc namespace, the semaphore arrays from the old
1867          * namespace are unreachable.
1868          */
1869         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1870                 do_sysvsem = 1;
1871         err = unshare_fs(unshare_flags, &new_fs);
1872         if (err)
1873                 goto bad_unshare_out;
1874         err = unshare_fd(unshare_flags, &new_fd);
1875         if (err)
1876                 goto bad_unshare_cleanup_fs;
1877         err = unshare_userns(unshare_flags, &new_cred);
1878         if (err)
1879                 goto bad_unshare_cleanup_fd;
1880         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1881                                          new_cred, new_fs);
1882         if (err)
1883                 goto bad_unshare_cleanup_cred;
1884
1885         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1886                 if (do_sysvsem) {
1887                         /*
1888                          * CLONE_SYSVSEM is equivalent to sys_exit().
1889                          */
1890                         exit_sem(current);
1891                 }
1892
1893                 if (new_nsproxy)
1894                         switch_task_namespaces(current, new_nsproxy);
1895
1896                 task_lock(current);
1897
1898                 if (new_fs) {
1899                         fs = current->fs;
1900                         spin_lock(&fs->lock);
1901                         current->fs = new_fs;
1902                         if (--fs->users)
1903                                 new_fs = NULL;
1904                         else
1905                                 new_fs = fs;
1906                         spin_unlock(&fs->lock);
1907                 }
1908
1909                 if (new_fd) {
1910                         fd = current->files;
1911                         current->files = new_fd;
1912                         new_fd = fd;
1913                 }
1914
1915                 task_unlock(current);
1916
1917                 if (new_cred) {
1918                         /* Install the new user namespace */
1919                         commit_creds(new_cred);
1920                         new_cred = NULL;
1921                 }
1922         }
1923
1924 bad_unshare_cleanup_cred:
1925         if (new_cred)
1926                 put_cred(new_cred);
1927 bad_unshare_cleanup_fd:
1928         if (new_fd)
1929                 put_files_struct(new_fd);
1930
1931 bad_unshare_cleanup_fs:
1932         if (new_fs)
1933                 free_fs_struct(new_fs);
1934
1935 bad_unshare_out:
1936         return err;
1937 }
1938
1939 /*
1940  *      Helper to unshare the files of the current task.
1941  *      We don't want to expose copy_files internals to
1942  *      the exec layer of the kernel.
1943  */
1944
1945 int unshare_files(struct files_struct **displaced)
1946 {
1947         struct task_struct *task = current;
1948         struct files_struct *copy = NULL;
1949         int error;
1950
1951         error = unshare_fd(CLONE_FILES, &copy);
1952         if (error || !copy) {
1953                 *displaced = NULL;
1954                 return error;
1955         }
1956         *displaced = task->files;
1957         task_lock(task);
1958         task->files = copy;
1959         task_unlock(task);
1960         return 0;
1961 }