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