kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/nsproxy.h>
  32 #include <linux/capability.h>
  33 #include <linux/cpu.h>
  34 #include <linux/cgroup.h>
  35 #include <linux/security.h>
  36 #include <linux/hugetlb.h>
  37 #include <linux/seccomp.h>
  38 #include <linux/swap.h>
  39 #include <linux/syscalls.h>
  40 #include <linux/jiffies.h>
  41 #include <linux/futex.h>
  42 #include <linux/compat.h>
  43 #include <linux/kthread.h>
  44 #include <linux/task_io_accounting_ops.h>
  45 #include <linux/rcupdate.h>
  46 #include <linux/ptrace.h>
  47 #include <linux/mount.h>
  48 #include <linux/audit.h>
  49 #include <linux/memcontrol.h>
  50 #include <linux/ftrace.h>
  51 #include <linux/proc_fs.h>
  52 #include <linux/profile.h>
  53 #include <linux/rmap.h>
  54 #include <linux/ksm.h>
  55 #include <linux/acct.h>
  56 #include <linux/tsacct_kern.h>
  57 #include <linux/cn_proc.h>
  58 #include <linux/freezer.h>
  59 #include <linux/delayacct.h>
  60 #include <linux/taskstats_kern.h>
  61 #include <linux/random.h>
  62 #include <linux/tty.h>
  63 #include <linux/blkdev.h>
  64 #include <linux/fs_struct.h>
  65 #include <linux/magic.h>
  66 #include <linux/perf_event.h>
  67 #include <linux/posix-timers.h>
  68 #include <linux/user-return-notifier.h>
  69 #include <linux/oom.h>
  70 #include <linux/khugepaged.h>
  71 #include <linux/signalfd.h>
  72 #include <linux/uprobes.h>
  73
  74 #include <asm/pgtable.h>
  75 #include <asm/pgalloc.h>
  76 #include <asm/uaccess.h>
  77 #include <asm/mmu_context.h>
  78 #include <asm/cacheflush.h>
  79 #include <asm/tlbflush.h>
  80
  81 #include <trace/events/sched.h>
  82
  83 #define CREATE_TRACE_POINTS
  84 #include <trace/events/task.h>
  85
  86 /*
  87  * Protected counters by write_lock_irq(&tasklist_lock)
  88  */
  89 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  90 int nr_threads;                 /* The idle threads do not count.. */
  91
  92 int max_threads;                /* tunable limit on nr_threads */
  93
  94 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  95
  96 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  97
  98 #ifdef CONFIG_PROVE_RCU
  99 int lockdep_tasklist_lock_is_held(void)
 100 {
 101         return lockdep_is_held(&tasklist_lock);
 102 }
 103 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 104 #endif /* #ifdef CONFIG_PROVE_RCU */
 105
 106 int nr_processes(void)
 107 {
 108         int cpu;
 109         int total = 0;
 110
 111         for_each_possible_cpu(cpu)
 112                 total += per_cpu(process_counts, cpu);
 113
 114         return total;
 115 }
 116
 117 void __weak arch_release_task_struct(struct task_struct *tsk)
 118 {
 119 }
 120
 121 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 122 static struct kmem_cache *task_struct_cachep;
 123
 124 static inline struct task_struct *alloc_task_struct_node(int node)
 125 {
 126         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 127 }
 128
 129 static inline void free_task_struct(struct task_struct *tsk)
 130 {
 131         kmem_cache_free(task_struct_cachep, tsk);
 132 }
 133 #endif
 134
 135 void __weak arch_release_thread_info(struct thread_info *ti)
 136 {
 137 }
 138
 139 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
 140
 141 /*
 142  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 143  * kmemcache based allocator.
 144  */
 145 # if THREAD_SIZE >= PAGE_SIZE
 146 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 147                                                   int node)
 148 {
 149         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 150                                              THREAD_SIZE_ORDER);
 151
 152         return page ? page_address(page) : NULL;
 153 }
 154
 155 static inline void free_thread_info(struct thread_info *ti)
 156 {
 157         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 158 }
 159 # else
 160 static struct kmem_cache *thread_info_cache;
 161
 162 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 163                                                   int node)
 164 {
 165         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
 166 }
 167
 168 static void free_thread_info(struct thread_info *ti)
 169 {
 170         kmem_cache_free(thread_info_cache, ti);
 171 }
 172
 173 void thread_info_cache_init(void)
 174 {
 175         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
 176                                               THREAD_SIZE, 0, NULL);
 177         BUG_ON(thread_info_cache == NULL);
 178 }
 179 # endif
 180 #endif
 181
 182 /* SLAB cache for signal_struct structures (tsk->signal) */
 183 static struct kmem_cache *signal_cachep;
 184
 185 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 186 struct kmem_cache *sighand_cachep;
 187
 188 /* SLAB cache for files_struct structures (tsk->files) */
 189 struct kmem_cache *files_cachep;
 190
 191 /* SLAB cache for fs_struct structures (tsk->fs) */
 192 struct kmem_cache *fs_cachep;
 193
 194 /* SLAB cache for vm_area_struct structures */
 195 struct kmem_cache *vm_area_cachep;
 196
 197 /* SLAB cache for mm_struct structures (tsk->mm) */
 198 static struct kmem_cache *mm_cachep;
 199
 200 static void account_kernel_stack(struct thread_info *ti, int account)
 201 {
 202         struct zone *zone = page_zone(virt_to_page(ti));
 203
 204         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 205 }
 206
 207 void free_task(struct task_struct *tsk)
 208 {
 209         account_kernel_stack(tsk->stack, -1);
 210         arch_release_thread_info(tsk->stack);
 211         free_thread_info(tsk->stack);
 212         rt_mutex_debug_task_free(tsk);
 213         ftrace_graph_exit_task(tsk);
 214         put_seccomp_filter(tsk);
 215         arch_release_task_struct(tsk);
 216         free_task_struct(tsk);
 217 }
 218 EXPORT_SYMBOL(free_task);
 219
 220 static inline void free_signal_struct(struct signal_struct *sig)
 221 {
 222         taskstats_tgid_free(sig);
 223         sched_autogroup_exit(sig);
 224         kmem_cache_free(signal_cachep, sig);
 225 }
 226
 227 static inline void put_signal_struct(struct signal_struct *sig)
 228 {
 229         if (atomic_dec_and_test(&sig->sigcnt))
 230                 free_signal_struct(sig);
 231 }
 232
 233 void __put_task_struct(struct task_struct *tsk)
 234 {
 235         WARN_ON(!tsk->exit_state);
 236         WARN_ON(atomic_read(&tsk->usage));
 237         WARN_ON(tsk == current);
 238
 239         security_task_free(tsk);
 240         exit_creds(tsk);
 241         delayacct_tsk_free(tsk);
 242         put_signal_struct(tsk->signal);
 243
 244         if (!profile_handoff_task(tsk))
 245                 free_task(tsk);
 246 }
 247 EXPORT_SYMBOL_GPL(__put_task_struct);
 248
 249 void __init __weak arch_task_cache_init(void) { }
 250
 251 void __init fork_init(unsigned long mempages)
 252 {
 253 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 254 #ifndef ARCH_MIN_TASKALIGN
 255 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 256 #endif
 257         /* create a slab on which task_structs can be allocated */
 258         task_struct_cachep =
 259                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 260                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 261 #endif
 262
 263         /* do the arch specific task caches init */
 264         arch_task_cache_init();
 265
 266         /*
 267          * The default maximum number of threads is set to a safe
 268          * value: the thread structures can take up at most half
 269          * of memory.
 270          */
 271         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 272
 273         /*
 274          * we need to allow at least 20 threads to boot a system
 275          */
 276         if (max_threads < 20)
 277                 max_threads = 20;
 278
 279         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 280         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 281         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 282                 init_task.signal->rlim[RLIMIT_NPROC];
 283 }
 284
 285 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 286                                                struct task_struct *src)
 287 {
 288         *dst = *src;
 289         return 0;
 290 }
 291
 292 static struct task_struct *dup_task_struct(struct task_struct *orig)
 293 {
 294         struct task_struct *tsk;
 295         struct thread_info *ti;
 296         unsigned long *stackend;
 297         int node = tsk_fork_get_node(orig);
 298         int err;
 299
 300         tsk = alloc_task_struct_node(node);
 301         if (!tsk)
 302                 return NULL;
 303
 304         ti = alloc_thread_info_node(tsk, node);
 305         if (!ti)
 306                 goto free_tsk;
 307
 308         err = arch_dup_task_struct(tsk, orig);
 309         if (err)
 310                 goto free_ti;
 311
 312         tsk->stack = ti;
 313
 314         setup_thread_stack(tsk, orig);
 315         clear_user_return_notifier(tsk);
 316         clear_tsk_need_resched(tsk);
 317         stackend = end_of_stack(tsk);
 318         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 319
 320 #ifdef CONFIG_CC_STACKPROTECTOR
 321         tsk->stack_canary = get_random_int();
 322 #endif
 323
 324         /*
 325          * One for us, one for whoever does the "release_task()" (usually
 326          * parent)
 327          */
 328         atomic_set(&tsk->usage, 2);
 329 #ifdef CONFIG_BLK_DEV_IO_TRACE
 330         tsk->btrace_seq = 0;
 331 #endif
 332         tsk->splice_pipe = NULL;
 333
 334         account_kernel_stack(ti, 1);
 335
 336         return tsk;
 337
 338 free_ti:
 339         free_thread_info(ti);
 340 free_tsk:
 341         free_task_struct(tsk);
 342         return NULL;
 343 }
 344
 345 #ifdef CONFIG_MMU
 346 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 347 {
 348         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 349         struct rb_node **rb_link, *rb_parent;
 350         int retval;
 351         unsigned long charge;
 352         struct mempolicy *pol;
 353
 354         down_write(&oldmm->mmap_sem);
 355         flush_cache_dup_mm(oldmm);
 356         /*
 357          * Not linked in yet - no deadlock potential:
 358          */
 359         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 360
 361         mm->locked_vm = 0;
 362         mm->mmap = NULL;
 363         mm->mmap_cache = NULL;
 364         mm->free_area_cache = oldmm->mmap_base;
 365         mm->cached_hole_size = ~0UL;
 366         mm->map_count = 0;
 367         cpumask_clear(mm_cpumask(mm));
 368         mm->mm_rb = RB_ROOT;
 369         rb_link = &mm->mm_rb.rb_node;
 370         rb_parent = NULL;
 371         pprev = &mm->mmap;
 372         retval = ksm_fork(mm, oldmm);
 373         if (retval)
 374                 goto out;
 375         retval = khugepaged_fork(mm, oldmm);
 376         if (retval)
 377                 goto out;
 378
 379         prev = NULL;
 380         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 381                 struct file *file;
 382
 383                 if (mpnt->vm_flags & VM_DONTCOPY) {
 384                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 385                                                         -vma_pages(mpnt));
 386                         continue;
 387                 }
 388                 charge = 0;
 389                 if (mpnt->vm_flags & VM_ACCOUNT) {
 390                         unsigned long len = vma_pages(mpnt);
 391
 392                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 393                                 goto fail_nomem;
 394                         charge = len;
 395                 }
 396                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 397                 if (!tmp)
 398                         goto fail_nomem;
 399                 *tmp = *mpnt;
 400                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 401                 pol = mpol_dup(vma_policy(mpnt));
 402                 retval = PTR_ERR(pol);
 403                 if (IS_ERR(pol))
 404                         goto fail_nomem_policy;
 405                 vma_set_policy(tmp, pol);
 406                 tmp->vm_mm = mm;
 407                 if (anon_vma_fork(tmp, mpnt))
 408                         goto fail_nomem_anon_vma_fork;
 409                 tmp->vm_flags &= ~VM_LOCKED;
 410                 tmp->vm_next = tmp->vm_prev = NULL;
 411                 file = tmp->vm_file;
 412                 if (file) {
 413                         struct inode *inode = file->f_path.dentry->d_inode;
 414                         struct address_space *mapping = file->f_mapping;
 415
 416                         get_file(file);
 417                         if (tmp->vm_flags & VM_DENYWRITE)
 418                                 atomic_dec(&inode->i_writecount);
 419                         mutex_lock(&mapping->i_mmap_mutex);
 420                         if (tmp->vm_flags & VM_SHARED)
 421                                 mapping->i_mmap_writable++;
 422                         flush_dcache_mmap_lock(mapping);
 423                         /* insert tmp into the share list, just after mpnt */
 424                         vma_prio_tree_add(tmp, mpnt);
 425                         flush_dcache_mmap_unlock(mapping);
 426                         mutex_unlock(&mapping->i_mmap_mutex);
 427                 }
 428
 429                 /*
 430                  * Clear hugetlb-related page reserves for children. This only
 431                  * affects MAP_PRIVATE mappings. Faults generated by the child
 432                  * are not guaranteed to succeed, even if read-only
 433                  */
 434                 if (is_vm_hugetlb_page(tmp))
 435                         reset_vma_resv_huge_pages(tmp);
 436
 437                 /*
 438                  * Link in the new vma and copy the page table entries.
 439                  */
 440                 *pprev = tmp;
 441                 pprev = &tmp->vm_next;
 442                 tmp->vm_prev = prev;
 443                 prev = tmp;
 444
 445                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 446                 rb_link = &tmp->vm_rb.rb_right;
 447                 rb_parent = &tmp->vm_rb;
 448
 449                 mm->map_count++;
 450                 retval = copy_page_range(mm, oldmm, mpnt);
 451
 452                 if (tmp->vm_ops && tmp->vm_ops->open)
 453                         tmp->vm_ops->open(tmp);
 454
 455                 if (retval)
 456                         goto out;
 457
 458                 if (file && uprobe_mmap(tmp))
 459                         goto out;
 460         }
 461         /* a new mm has just been created */
 462         arch_dup_mmap(oldmm, mm);
 463         retval = 0;
 464 out:
 465         up_write(&mm->mmap_sem);
 466         flush_tlb_mm(oldmm);
 467         up_write(&oldmm->mmap_sem);
 468         return retval;
 469 fail_nomem_anon_vma_fork:
 470         mpol_put(pol);
 471 fail_nomem_policy:
 472         kmem_cache_free(vm_area_cachep, tmp);
 473 fail_nomem:
 474         retval = -ENOMEM;
 475         vm_unacct_memory(charge);
 476         goto out;
 477 }
 478
 479 static inline int mm_alloc_pgd(struct mm_struct *mm)
 480 {
 481         mm->pgd = pgd_alloc(mm);
 482         if (unlikely(!mm->pgd))
 483                 return -ENOMEM;
 484         return 0;
 485 }
 486
 487 static inline void mm_free_pgd(struct mm_struct *mm)
 488 {
 489         pgd_free(mm, mm->pgd);
 490 }
 491 #else
 492 #define dup_mmap(mm, oldmm)     (0)
 493 #define mm_alloc_pgd(mm)        (0)
 494 #define mm_free_pgd(mm)
 495 #endif /* CONFIG_MMU */
 496
 497 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 498
 499 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 500 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 501
 502 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 503
 504 static int __init coredump_filter_setup(char *s)
 505 {
 506         default_dump_filter =
 507                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 508                 MMF_DUMP_FILTER_MASK;
 509         return 1;
 510 }
 511
 512 __setup("coredump_filter=", coredump_filter_setup);
 513
 514 #include <linux/init_task.h>
 515
 516 static void mm_init_aio(struct mm_struct *mm)
 517 {
 518 #ifdef CONFIG_AIO
 519         spin_lock_init(&mm->ioctx_lock);
 520         INIT_HLIST_HEAD(&mm->ioctx_list);
 521 #endif
 522 }
 523
 524 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 525 {
 526         atomic_set(&mm->mm_users, 1);
 527         atomic_set(&mm->mm_count, 1);
 528         init_rwsem(&mm->mmap_sem);
 529         INIT_LIST_HEAD(&mm->mmlist);
 530         mm->flags = (current->mm) ?
 531                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 532         mm->core_state = NULL;
 533         mm->nr_ptes = 0;
 534         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 535         spin_lock_init(&mm->page_table_lock);
 536         mm->free_area_cache = TASK_UNMAPPED_BASE;
 537         mm->cached_hole_size = ~0UL;
 538         mm_init_aio(mm);
 539         mm_init_owner(mm, p);
 540
 541         if (likely(!mm_alloc_pgd(mm))) {
 542                 mm->def_flags = 0;
 543                 mmu_notifier_mm_init(mm);
 544                 return mm;
 545         }
 546
 547         free_mm(mm);
 548         return NULL;
 549 }
 550
 551 static void check_mm(struct mm_struct *mm)
 552 {
 553         int i;
 554
 555         for (i = 0; i < NR_MM_COUNTERS; i++) {
 556                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 557
 558                 if (unlikely(x))
 559                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 560                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 561         }
 562
 563 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 564         VM_BUG_ON(mm->pmd_huge_pte);
 565 #endif
 566 }
 567
 568 /*
 569  * Allocate and initialize an mm_struct.
 570  */
 571 struct mm_struct *mm_alloc(void)
 572 {
 573         struct mm_struct *mm;
 574
 575         mm = allocate_mm();
 576         if (!mm)
 577                 return NULL;
 578
 579         memset(mm, 0, sizeof(*mm));
 580         mm_init_cpumask(mm);
 581         return mm_init(mm, current);
 582 }
 583
 584 /*
 585  * Called when the last reference to the mm
 586  * is dropped: either by a lazy thread or by
 587  * mmput. Free the page directory and the mm.
 588  */
 589 void __mmdrop(struct mm_struct *mm)
 590 {
 591         BUG_ON(mm == &init_mm);
 592         mm_free_pgd(mm);
 593         destroy_context(mm);
 594         mmu_notifier_mm_destroy(mm);
 595         check_mm(mm);
 596         free_mm(mm);
 597 }
 598 EXPORT_SYMBOL_GPL(__mmdrop);
 599
 600 /*
 601  * Decrement the use count and release all resources for an mm.
 602  */
 603 void mmput(struct mm_struct *mm)
 604 {
 605         might_sleep();
 606
 607         if (atomic_dec_and_test(&mm->mm_users)) {
 608                 uprobe_clear_state(mm);
 609                 exit_aio(mm);
 610                 ksm_exit(mm);
 611                 khugepaged_exit(mm); /* must run before exit_mmap */
 612                 exit_mmap(mm);
 613                 set_mm_exe_file(mm, NULL);
 614                 if (!list_empty(&mm->mmlist)) {
 615                         spin_lock(&mmlist_lock);
 616                         list_del(&mm->mmlist);
 617                         spin_unlock(&mmlist_lock);
 618                 }
 619                 if (mm->binfmt)
 620                         module_put(mm->binfmt->module);
 621                 mmdrop(mm);
 622         }
 623 }
 624 EXPORT_SYMBOL_GPL(mmput);
 625
 626 /*
 627  * We added or removed a vma mapping the executable. The vmas are only mapped
 628  * during exec and are not mapped with the mmap system call.
 629  * Callers must hold down_write() on the mm's mmap_sem for these
 630  */
 631 void added_exe_file_vma(struct mm_struct *mm)
 632 {
 633         mm->num_exe_file_vmas++;
 634 }
 635
 636 void removed_exe_file_vma(struct mm_struct *mm)
 637 {
 638         mm->num_exe_file_vmas--;
 639         if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
 640                 fput(mm->exe_file);
 641                 mm->exe_file = NULL;
 642         }
 643
 644 }
 645
 646 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 647 {
 648         if (new_exe_file)
 649                 get_file(new_exe_file);
 650         if (mm->exe_file)
 651                 fput(mm->exe_file);
 652         mm->exe_file = new_exe_file;
 653         mm->num_exe_file_vmas = 0;
 654 }
 655
 656 struct file *get_mm_exe_file(struct mm_struct *mm)
 657 {
 658         struct file *exe_file;
 659
 660         /* We need mmap_sem to protect against races with removal of
 661          * VM_EXECUTABLE vmas */
 662         down_read(&mm->mmap_sem);
 663         exe_file = mm->exe_file;
 664         if (exe_file)
 665                 get_file(exe_file);
 666         up_read(&mm->mmap_sem);
 667         return exe_file;
 668 }
 669
 670 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 671 {
 672         /* It's safe to write the exe_file pointer without exe_file_lock because
 673          * this is called during fork when the task is not yet in /proc */
 674         newmm->exe_file = get_mm_exe_file(oldmm);
 675 }
 676
 677 /**
 678  * get_task_mm - acquire a reference to the task's mm
 679  *
 680  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 681  * this kernel workthread has transiently adopted a user mm with use_mm,
 682  * to do its AIO) is not set and if so returns a reference to it, after
 683  * bumping up the use count.  User must release the mm via mmput()
 684  * after use.  Typically used by /proc and ptrace.
 685  */
 686 struct mm_struct *get_task_mm(struct task_struct *task)
 687 {
 688         struct mm_struct *mm;
 689
 690         task_lock(task);
 691         mm = task->mm;
 692         if (mm) {
 693                 if (task->flags & PF_KTHREAD)
 694                         mm = NULL;
 695                 else
 696                         atomic_inc(&mm->mm_users);
 697         }
 698         task_unlock(task);
 699         return mm;
 700 }
 701 EXPORT_SYMBOL_GPL(get_task_mm);
 702
 703 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 704 {
 705         struct mm_struct *mm;
 706         int err;
 707
 708         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 709         if (err)
 710                 return ERR_PTR(err);
 711
 712         mm = get_task_mm(task);
 713         if (mm && mm != current->mm &&
 714                         !ptrace_may_access(task, mode)) {
 715                 mmput(mm);
 716                 mm = ERR_PTR(-EACCES);
 717         }
 718         mutex_unlock(&task->signal->cred_guard_mutex);
 719
 720         return mm;
 721 }
 722
 723 static void complete_vfork_done(struct task_struct *tsk)
 724 {
 725         struct completion *vfork;
 726
 727         task_lock(tsk);
 728         vfork = tsk->vfork_done;
 729         if (likely(vfork)) {
 730                 tsk->vfork_done = NULL;
 731                 complete(vfork);
 732         }
 733         task_unlock(tsk);
 734 }
 735
 736 static int wait_for_vfork_done(struct task_struct *child,
 737                                 struct completion *vfork)
 738 {
 739         int killed;
 740
 741         freezer_do_not_count();
 742         killed = wait_for_completion_killable(vfork);
 743         freezer_count();
 744
 745         if (killed) {
 746                 task_lock(child);
 747                 child->vfork_done = NULL;
 748                 task_unlock(child);
 749         }
 750
 751         put_task_struct(child);
 752         return killed;
 753 }
 754
 755 /* Please note the differences between mmput and mm_release.
 756  * mmput is called whenever we stop holding onto a mm_struct,
 757  * error success whatever.
 758  *
 759  * mm_release is called after a mm_struct has been removed
 760  * from the current process.
 761  *
 762  * This difference is important for error handling, when we
 763  * only half set up a mm_struct for a new process and need to restore
 764  * the old one.  Because we mmput the new mm_struct before
 765  * restoring the old one. . .
 766  * Eric Biederman 10 January 1998
 767  */
 768 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 769 {
 770         /* Get rid of any futexes when releasing the mm */
 771 #ifdef CONFIG_FUTEX
 772         if (unlikely(tsk->robust_list)) {
 773                 exit_robust_list(tsk);
 774                 tsk->robust_list = NULL;
 775         }
 776 #ifdef CONFIG_COMPAT
 777         if (unlikely(tsk->compat_robust_list)) {
 778                 compat_exit_robust_list(tsk);
 779                 tsk->compat_robust_list = NULL;
 780         }
 781 #endif
 782         if (unlikely(!list_empty(&tsk->pi_state_list)))
 783                 exit_pi_state_list(tsk);
 784 #endif
 785
 786         uprobe_free_utask(tsk);
 787
 788         /* Get rid of any cached register state */
 789         deactivate_mm(tsk, mm);
 790
 791         /*
 792          * If we're exiting normally, clear a user-space tid field if
 793          * requested.  We leave this alone when dying by signal, to leave
 794          * the value intact in a core dump, and to save the unnecessary
 795          * trouble, say, a killed vfork parent shouldn't touch this mm.
 796          * Userland only wants this done for a sys_exit.
 797          */
 798         if (tsk->clear_child_tid) {
 799                 if (!(tsk->flags & PF_SIGNALED) &&
 800                     atomic_read(&mm->mm_users) > 1) {
 801                         /*
 802                          * We don't check the error code - if userspace has
 803                          * not set up a proper pointer then tough luck.
 804                          */
 805                         put_user(0, tsk->clear_child_tid);
 806                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 807                                         1, NULL, NULL, 0);
 808                 }
 809                 tsk->clear_child_tid = NULL;
 810         }
 811
 812         /*
 813          * All done, finally we can wake up parent and return this mm to him.
 814          * Also kthread_stop() uses this completion for synchronization.
 815          */
 816         if (tsk->vfork_done)
 817                 complete_vfork_done(tsk);
 818 }
 819
 820 /*
 821  * Allocate a new mm structure and copy contents from the
 822  * mm structure of the passed in task structure.
 823  */
 824 struct mm_struct *dup_mm(struct task_struct *tsk)
 825 {
 826         struct mm_struct *mm, *oldmm = current->mm;
 827         int err;
 828
 829         if (!oldmm)
 830                 return NULL;
 831
 832         mm = allocate_mm();
 833         if (!mm)
 834                 goto fail_nomem;
 835
 836         memcpy(mm, oldmm, sizeof(*mm));
 837         mm_init_cpumask(mm);
 838
 839 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 840         mm->pmd_huge_pte = NULL;
 841 #endif
 842         uprobe_reset_state(mm);
 843
 844         if (!mm_init(mm, tsk))
 845                 goto fail_nomem;
 846
 847         if (init_new_context(tsk, mm))
 848                 goto fail_nocontext;
 849
 850         dup_mm_exe_file(oldmm, mm);
 851
 852         err = dup_mmap(mm, oldmm);
 853         if (err)
 854                 goto free_pt;
 855
 856         mm->hiwater_rss = get_mm_rss(mm);
 857         mm->hiwater_vm = mm->total_vm;
 858
 859         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 860                 goto free_pt;
 861
 862         return mm;
 863
 864 free_pt:
 865         /* don't put binfmt in mmput, we haven't got module yet */
 866         mm->binfmt = NULL;
 867         mmput(mm);
 868
 869 fail_nomem:
 870         return NULL;
 871
 872 fail_nocontext:
 873         /*
 874          * If init_new_context() failed, we cannot use mmput() to free the mm
 875          * because it calls destroy_context()
 876          */
 877         mm_free_pgd(mm);
 878         free_mm(mm);
 879         return NULL;
 880 }
 881
 882 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 883 {
 884         struct mm_struct *mm, *oldmm;
 885         int retval;
 886
 887         tsk->min_flt = tsk->maj_flt = 0;
 888         tsk->nvcsw = tsk->nivcsw = 0;
 889 #ifdef CONFIG_DETECT_HUNG_TASK
 890         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 891 #endif
 892
 893         tsk->mm = NULL;
 894         tsk->active_mm = NULL;
 895
 896         /*
 897          * Are we cloning a kernel thread?
 898          *
 899          * We need to steal a active VM for that..
 900          */
 901         oldmm = current->mm;
 902         if (!oldmm)
 903                 return 0;
 904
 905         if (clone_flags & CLONE_VM) {
 906                 atomic_inc(&oldmm->mm_users);
 907                 mm = oldmm;
 908                 goto good_mm;
 909         }
 910
 911         retval = -ENOMEM;
 912         mm = dup_mm(tsk);
 913         if (!mm)
 914                 goto fail_nomem;
 915
 916 good_mm:
 917         tsk->mm = mm;
 918         tsk->active_mm = mm;
 919         return 0;
 920
 921 fail_nomem:
 922         return retval;
 923 }
 924
 925 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 926 {
 927         struct fs_struct *fs = current->fs;
 928         if (clone_flags & CLONE_FS) {
 929                 /* tsk->fs is already what we want */
 930                 spin_lock(&fs->lock);
 931                 if (fs->in_exec) {
 932                         spin_unlock(&fs->lock);
 933                         return -EAGAIN;
 934                 }
 935                 fs->users++;
 936                 spin_unlock(&fs->lock);
 937                 return 0;
 938         }
 939         tsk->fs = copy_fs_struct(fs);
 940         if (!tsk->fs)
 941                 return -ENOMEM;
 942         return 0;
 943 }
 944
 945 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 946 {
 947         struct files_struct *oldf, *newf;
 948         int error = 0;
 949
 950         /*
 951          * A background process may not have any files ...
 952          */
 953         oldf = current->files;
 954         if (!oldf)
 955                 goto out;
 956
 957         if (clone_flags & CLONE_FILES) {
 958                 atomic_inc(&oldf->count);
 959                 goto out;
 960         }
 961
 962         newf = dup_fd(oldf, &error);
 963         if (!newf)
 964                 goto out;
 965
 966         tsk->files = newf;
 967         error = 0;
 968 out:
 969         return error;
 970 }
 971
 972 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 973 {
 974 #ifdef CONFIG_BLOCK
 975         struct io_context *ioc = current->io_context;
 976         struct io_context *new_ioc;
 977
 978         if (!ioc)
 979                 return 0;
 980         /*
 981          * Share io context with parent, if CLONE_IO is set
 982          */
 983         if (clone_flags & CLONE_IO) {
 984                 ioc_task_link(ioc);
 985                 tsk->io_context = ioc;
 986         } else if (ioprio_valid(ioc->ioprio)) {
 987                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
 988                 if (unlikely(!new_ioc))
 989                         return -ENOMEM;
 990
 991                 new_ioc->ioprio = ioc->ioprio;
 992                 put_io_context(new_ioc);
 993         }
 994 #endif
 995         return 0;
 996 }
 997
 998 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 999 {
1000         struct sighand_struct *sig;
1001
1002         if (clone_flags & CLONE_SIGHAND) {
1003                 atomic_inc(&current->sighand->count);
1004                 return 0;
1005         }
1006         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1007         rcu_assign_pointer(tsk->sighand, sig);
1008         if (!sig)
1009                 return -ENOMEM;
1010         atomic_set(&sig->count, 1);
1011         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1012         return 0;
1013 }
1014
1015 void __cleanup_sighand(struct sighand_struct *sighand)
1016 {
1017         if (atomic_dec_and_test(&sighand->count)) {
1018                 signalfd_cleanup(sighand);
1019                 kmem_cache_free(sighand_cachep, sighand);
1020         }
1021 }
1022
1023
1024 /*
1025  * Initialize POSIX timer handling for a thread group.
1026  */
1027 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1028 {
1029         unsigned long cpu_limit;
1030
1031         /* Thread group counters. */
1032         thread_group_cputime_init(sig);
1033
1034         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1035         if (cpu_limit != RLIM_INFINITY) {
1036                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1037                 sig->cputimer.running = 1;
1038         }
1039
1040         /* The timer lists. */
1041         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1042         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1043         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1044 }
1045
1046 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1047 {
1048         struct signal_struct *sig;
1049
1050         if (clone_flags & CLONE_THREAD)
1051                 return 0;
1052
1053         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1054         tsk->signal = sig;
1055         if (!sig)
1056                 return -ENOMEM;
1057
1058         sig->nr_threads = 1;
1059         atomic_set(&sig->live, 1);
1060         atomic_set(&sig->sigcnt, 1);
1061         init_waitqueue_head(&sig->wait_chldexit);
1062         if (clone_flags & CLONE_NEWPID)
1063                 sig->flags |= SIGNAL_UNKILLABLE;
1064         sig->curr_target = tsk;
1065         init_sigpending(&sig->shared_pending);
1066         INIT_LIST_HEAD(&sig->posix_timers);
1067
1068         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1069         sig->real_timer.function = it_real_fn;
1070
1071         task_lock(current->group_leader);
1072         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1073         task_unlock(current->group_leader);
1074
1075         posix_cpu_timers_init_group(sig);
1076
1077         tty_audit_fork(sig);
1078         sched_autogroup_fork(sig);
1079
1080 #ifdef CONFIG_CGROUPS
1081         init_rwsem(&sig->group_rwsem);
1082 #endif
1083
1084         sig->oom_adj = current->signal->oom_adj;
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                                         struct pt_regs *regs,
1152                                         unsigned long stack_size,
1153                                         int __user *child_tidptr,
1154                                         struct pid *pid,
1155                                         int trace)
1156 {
1157         int retval;
1158         struct task_struct *p;
1159         int cgroup_callbacks_done = 0;
1160
1161         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1162                 return ERR_PTR(-EINVAL);
1163
1164         /*
1165          * Thread groups must share signals as well, and detached threads
1166          * can only be started up within the thread group.
1167          */
1168         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1169                 return ERR_PTR(-EINVAL);
1170
1171         /*
1172          * Shared signal handlers imply shared VM. By way of the above,
1173          * thread groups also imply shared VM. Blocking this case allows
1174          * for various simplifications in other code.
1175          */
1176         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1177                 return ERR_PTR(-EINVAL);
1178
1179         /*
1180          * Siblings of global init remain as zombies on exit since they are
1181          * not reaped by their parent (swapper). To solve this and to avoid
1182          * multi-rooted process trees, prevent global and container-inits
1183          * from creating siblings.
1184          */
1185         if ((clone_flags & CLONE_PARENT) &&
1186                                 current->signal->flags & SIGNAL_UNKILLABLE)
1187                 return ERR_PTR(-EINVAL);
1188
1189         retval = security_task_create(clone_flags);
1190         if (retval)
1191                 goto fork_out;
1192
1193         retval = -ENOMEM;
1194         p = dup_task_struct(current);
1195         if (!p)
1196                 goto fork_out;
1197
1198         ftrace_graph_init_task(p);
1199         get_seccomp_filter(p);
1200
1201         rt_mutex_init_task(p);
1202
1203 #ifdef CONFIG_PROVE_LOCKING
1204         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1205         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1206 #endif
1207         retval = -EAGAIN;
1208         if (atomic_read(&p->real_cred->user->processes) >=
1209                         task_rlimit(p, RLIMIT_NPROC)) {
1210                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1211                     p->real_cred->user != INIT_USER)
1212                         goto bad_fork_free;
1213         }
1214         current->flags &= ~PF_NPROC_EXCEEDED;
1215
1216         retval = copy_creds(p, clone_flags);
1217         if (retval < 0)
1218                 goto bad_fork_free;
1219
1220         /*
1221          * If multiple threads are within copy_process(), then this check
1222          * triggers too late. This doesn't hurt, the check is only there
1223          * to stop root fork bombs.
1224          */
1225         retval = -EAGAIN;
1226         if (nr_threads >= max_threads)
1227                 goto bad_fork_cleanup_count;
1228
1229         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1230                 goto bad_fork_cleanup_count;
1231
1232         p->did_exec = 0;
1233         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1234         copy_flags(clone_flags, p);
1235         INIT_LIST_HEAD(&p->children);
1236         INIT_LIST_HEAD(&p->sibling);
1237         rcu_copy_process(p);
1238         p->vfork_done = NULL;
1239         spin_lock_init(&p->alloc_lock);
1240
1241         init_sigpending(&p->pending);
1242
1243         p->utime = p->stime = p->gtime = 0;
1244         p->utimescaled = p->stimescaled = 0;
1245 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1246         p->prev_utime = p->prev_stime = 0;
1247 #endif
1248 #if defined(SPLIT_RSS_COUNTING)
1249         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1250 #endif
1251
1252         p->default_timer_slack_ns = current->timer_slack_ns;
1253
1254         task_io_accounting_init(&p->ioac);
1255         acct_clear_integrals(p);
1256
1257         posix_cpu_timers_init(p);
1258
1259         do_posix_clock_monotonic_gettime(&p->start_time);
1260         p->real_start_time = p->start_time;
1261         monotonic_to_bootbased(&p->real_start_time);
1262         p->io_context = NULL;
1263         p->audit_context = NULL;
1264         if (clone_flags & CLONE_THREAD)
1265                 threadgroup_change_begin(current);
1266         cgroup_fork(p);
1267 #ifdef CONFIG_NUMA
1268         p->mempolicy = mpol_dup(p->mempolicy);
1269         if (IS_ERR(p->mempolicy)) {
1270                 retval = PTR_ERR(p->mempolicy);
1271                 p->mempolicy = NULL;
1272                 goto bad_fork_cleanup_cgroup;
1273         }
1274         mpol_fix_fork_child_flag(p);
1275 #endif
1276 #ifdef CONFIG_CPUSETS
1277         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1278         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1279         seqcount_init(&p->mems_allowed_seq);
1280 #endif
1281 #ifdef CONFIG_TRACE_IRQFLAGS
1282         p->irq_events = 0;
1283 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1284         p->hardirqs_enabled = 1;
1285 #else
1286         p->hardirqs_enabled = 0;
1287 #endif
1288         p->hardirq_enable_ip = 0;
1289         p->hardirq_enable_event = 0;
1290         p->hardirq_disable_ip = _THIS_IP_;
1291         p->hardirq_disable_event = 0;
1292         p->softirqs_enabled = 1;
1293         p->softirq_enable_ip = _THIS_IP_;
1294         p->softirq_enable_event = 0;
1295         p->softirq_disable_ip = 0;
1296         p->softirq_disable_event = 0;
1297         p->hardirq_context = 0;
1298         p->softirq_context = 0;
1299 #endif
1300 #ifdef CONFIG_LOCKDEP
1301         p->lockdep_depth = 0; /* no locks held yet */
1302         p->curr_chain_key = 0;
1303         p->lockdep_recursion = 0;
1304 #endif
1305
1306 #ifdef CONFIG_DEBUG_MUTEXES
1307         p->blocked_on = NULL; /* not blocked yet */
1308 #endif
1309 #ifdef CONFIG_MEMCG
1310         p->memcg_batch.do_batch = 0;
1311         p->memcg_batch.memcg = NULL;
1312 #endif
1313
1314         /* Perform scheduler related setup. Assign this task to a CPU. */
1315         sched_fork(p);
1316
1317         retval = perf_event_init_task(p);
1318         if (retval)
1319                 goto bad_fork_cleanup_policy;
1320         retval = audit_alloc(p);
1321         if (retval)
1322                 goto bad_fork_cleanup_policy;
1323         /* copy all the process information */
1324         retval = copy_semundo(clone_flags, p);
1325         if (retval)
1326                 goto bad_fork_cleanup_audit;
1327         retval = copy_files(clone_flags, p);
1328         if (retval)
1329                 goto bad_fork_cleanup_semundo;
1330         retval = copy_fs(clone_flags, p);
1331         if (retval)
1332                 goto bad_fork_cleanup_files;
1333         retval = copy_sighand(clone_flags, p);
1334         if (retval)
1335                 goto bad_fork_cleanup_fs;
1336         retval = copy_signal(clone_flags, p);
1337         if (retval)
1338                 goto bad_fork_cleanup_sighand;
1339         retval = copy_mm(clone_flags, p);
1340         if (retval)
1341                 goto bad_fork_cleanup_signal;
1342         retval = copy_namespaces(clone_flags, p);
1343         if (retval)
1344                 goto bad_fork_cleanup_mm;
1345         retval = copy_io(clone_flags, p);
1346         if (retval)
1347                 goto bad_fork_cleanup_namespaces;
1348         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1349         if (retval)
1350                 goto bad_fork_cleanup_io;
1351
1352         if (pid != &init_struct_pid) {
1353                 retval = -ENOMEM;
1354                 pid = alloc_pid(p->nsproxy->pid_ns);
1355                 if (!pid)
1356                         goto bad_fork_cleanup_io;
1357         }
1358
1359         p->pid = pid_nr(pid);
1360         p->tgid = p->pid;
1361         if (clone_flags & CLONE_THREAD)
1362                 p->tgid = current->tgid;
1363
1364         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1365         /*
1366          * Clear TID on mm_release()?
1367          */
1368         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1369 #ifdef CONFIG_BLOCK
1370         p->plug = NULL;
1371 #endif
1372 #ifdef CONFIG_FUTEX
1373         p->robust_list = NULL;
1374 #ifdef CONFIG_COMPAT
1375         p->compat_robust_list = NULL;
1376 #endif
1377         INIT_LIST_HEAD(&p->pi_state_list);
1378         p->pi_state_cache = NULL;
1379 #endif
1380         uprobe_copy_process(p);
1381         /*
1382          * sigaltstack should be cleared when sharing the same VM
1383          */
1384         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1385                 p->sas_ss_sp = p->sas_ss_size = 0;
1386
1387         /*
1388          * Syscall tracing and stepping should be turned off in the
1389          * child regardless of CLONE_PTRACE.
1390          */
1391         user_disable_single_step(p);
1392         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1393 #ifdef TIF_SYSCALL_EMU
1394         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1395 #endif
1396         clear_all_latency_tracing(p);
1397
1398         /* ok, now we should be set up.. */
1399         if (clone_flags & CLONE_THREAD)
1400                 p->exit_signal = -1;
1401         else if (clone_flags & CLONE_PARENT)
1402                 p->exit_signal = current->group_leader->exit_signal;
1403         else
1404                 p->exit_signal = (clone_flags & CSIGNAL);
1405
1406         p->pdeath_signal = 0;
1407         p->exit_state = 0;
1408
1409         p->nr_dirtied = 0;
1410         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1411         p->dirty_paused_when = 0;
1412
1413         /*
1414          * Ok, make it visible to the rest of the system.
1415          * We dont wake it up yet.
1416          */
1417         p->group_leader = p;
1418         INIT_LIST_HEAD(&p->thread_group);
1419         p->task_works = NULL;
1420
1421         /* Now that the task is set up, run cgroup callbacks if
1422          * necessary. We need to run them before the task is visible
1423          * on the tasklist. */
1424         cgroup_fork_callbacks(p);
1425         cgroup_callbacks_done = 1;
1426
1427         /* Need tasklist lock for parent etc handling! */
1428         write_lock_irq(&tasklist_lock);
1429
1430         /* CLONE_PARENT re-uses the old parent */
1431         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1432                 p->real_parent = current->real_parent;
1433                 p->parent_exec_id = current->parent_exec_id;
1434         } else {
1435                 p->real_parent = current;
1436                 p->parent_exec_id = current->self_exec_id;
1437         }
1438
1439         spin_lock(&current->sighand->siglock);
1440
1441         /*
1442          * Process group and session signals need to be delivered to just the
1443          * parent before the fork or both the parent and the child after the
1444          * fork. Restart if a signal comes in before we add the new process to
1445          * it's process group.
1446          * A fatal signal pending means that current will exit, so the new
1447          * thread can't slip out of an OOM kill (or normal SIGKILL).
1448         */
1449         recalc_sigpending();
1450         if (signal_pending(current)) {
1451                 spin_unlock(&current->sighand->siglock);
1452                 write_unlock_irq(&tasklist_lock);
1453                 retval = -ERESTARTNOINTR;
1454                 goto bad_fork_free_pid;
1455         }
1456
1457         if (clone_flags & CLONE_THREAD) {
1458                 current->signal->nr_threads++;
1459                 atomic_inc(&current->signal->live);
1460                 atomic_inc(&current->signal->sigcnt);
1461                 p->group_leader = current->group_leader;
1462                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1463         }
1464
1465         if (likely(p->pid)) {
1466                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1467
1468                 if (thread_group_leader(p)) {
1469                         if (is_child_reaper(pid))
1470                                 p->nsproxy->pid_ns->child_reaper = p;
1471
1472                         p->signal->leader_pid = pid;
1473                         p->signal->tty = tty_kref_get(current->signal->tty);
1474                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1475                         attach_pid(p, PIDTYPE_SID, task_session(current));
1476                         list_add_tail(&p->sibling, &p->real_parent->children);
1477                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1478                         __this_cpu_inc(process_counts);
1479                 }
1480                 attach_pid(p, PIDTYPE_PID, pid);
1481                 nr_threads++;
1482         }
1483
1484         total_forks++;
1485         spin_unlock(&current->sighand->siglock);
1486         write_unlock_irq(&tasklist_lock);
1487         proc_fork_connector(p);
1488         cgroup_post_fork(p);
1489         if (clone_flags & CLONE_THREAD)
1490                 threadgroup_change_end(current);
1491         perf_event_fork(p);
1492
1493         trace_task_newtask(p, clone_flags);
1494
1495         return p;
1496
1497 bad_fork_free_pid:
1498         if (pid != &init_struct_pid)
1499                 free_pid(pid);
1500 bad_fork_cleanup_io:
1501         if (p->io_context)
1502                 exit_io_context(p);
1503 bad_fork_cleanup_namespaces:
1504         if (unlikely(clone_flags & CLONE_NEWPID))
1505                 pid_ns_release_proc(p->nsproxy->pid_ns);
1506         exit_task_namespaces(p);
1507 bad_fork_cleanup_mm:
1508         if (p->mm)
1509                 mmput(p->mm);
1510 bad_fork_cleanup_signal:
1511         if (!(clone_flags & CLONE_THREAD))
1512                 free_signal_struct(p->signal);
1513 bad_fork_cleanup_sighand:
1514         __cleanup_sighand(p->sighand);
1515 bad_fork_cleanup_fs:
1516         exit_fs(p); /* blocking */
1517 bad_fork_cleanup_files:
1518         exit_files(p); /* blocking */
1519 bad_fork_cleanup_semundo:
1520         exit_sem(p);
1521 bad_fork_cleanup_audit:
1522         audit_free(p);
1523 bad_fork_cleanup_policy:
1524         perf_event_free_task(p);
1525 #ifdef CONFIG_NUMA
1526         mpol_put(p->mempolicy);
1527 bad_fork_cleanup_cgroup:
1528 #endif
1529         if (clone_flags & CLONE_THREAD)
1530                 threadgroup_change_end(current);
1531         cgroup_exit(p, cgroup_callbacks_done);
1532         delayacct_tsk_free(p);
1533         module_put(task_thread_info(p)->exec_domain->module);
1534 bad_fork_cleanup_count:
1535         atomic_dec(&p->cred->user->processes);
1536         exit_creds(p);
1537 bad_fork_free:
1538         free_task(p);
1539 fork_out:
1540         return ERR_PTR(retval);
1541 }
1542
1543 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1544 {
1545         memset(regs, 0, sizeof(struct pt_regs));
1546         return regs;
1547 }
1548
1549 static inline void init_idle_pids(struct pid_link *links)
1550 {
1551         enum pid_type type;
1552
1553         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1554                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1555                 links[type].pid = &init_struct_pid;
1556         }
1557 }
1558
1559 struct task_struct * __cpuinit fork_idle(int cpu)
1560 {
1561         struct task_struct *task;
1562         struct pt_regs regs;
1563
1564         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1565                             &init_struct_pid, 0);
1566         if (!IS_ERR(task)) {
1567                 init_idle_pids(task->pids);
1568                 init_idle(task, cpu);
1569         }
1570
1571         return task;
1572 }
1573
1574 /*
1575  *  Ok, this is the main fork-routine.
1576  *
1577  * It copies the process, and if successful kick-starts
1578  * it and waits for it to finish using the VM if required.
1579  */
1580 long do_fork(unsigned long clone_flags,
1581               unsigned long stack_start,
1582               struct pt_regs *regs,
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) {
1596                 if (clone_flags & CLONE_THREAD)
1597                         return -EINVAL;
1598                 /* hopefully this check will go away when userns support is
1599                  * complete
1600                  */
1601                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1602                                 !capable(CAP_SETGID))
1603                         return -EPERM;
1604         }
1605
1606         /*
1607          * Determine whether and which event to report to ptracer.  When
1608          * called from kernel_thread or CLONE_UNTRACED is explicitly
1609          * requested, no event is reported; otherwise, report if the event
1610          * for the type of forking is enabled.
1611          */
1612         if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1613                 if (clone_flags & CLONE_VFORK)
1614                         trace = PTRACE_EVENT_VFORK;
1615                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1616                         trace = PTRACE_EVENT_CLONE;
1617                 else
1618                         trace = PTRACE_EVENT_FORK;
1619
1620                 if (likely(!ptrace_event_enabled(current, trace)))
1621                         trace = 0;
1622         }
1623
1624         p = copy_process(clone_flags, stack_start, regs, stack_size,
1625                          child_tidptr, NULL, trace);
1626         /*
1627          * Do this prior waking up the new thread - the thread pointer
1628          * might get invalid after that point, if the thread exits quickly.
1629          */
1630         if (!IS_ERR(p)) {
1631                 struct completion vfork;
1632
1633                 trace_sched_process_fork(current, p);
1634
1635                 nr = task_pid_vnr(p);
1636
1637                 if (clone_flags & CLONE_PARENT_SETTID)
1638                         put_user(nr, parent_tidptr);
1639
1640                 if (clone_flags & CLONE_VFORK) {
1641                         p->vfork_done = &vfork;
1642                         init_completion(&vfork);
1643                         get_task_struct(p);
1644                 }
1645
1646                 wake_up_new_task(p);
1647
1648                 /* forking complete and child started to run, tell ptracer */
1649                 if (unlikely(trace))
1650                         ptrace_event(trace, nr);
1651
1652                 if (clone_flags & CLONE_VFORK) {
1653                         if (!wait_for_vfork_done(p, &vfork))
1654                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1655                 }
1656         } else {
1657                 nr = PTR_ERR(p);
1658         }
1659         return nr;
1660 }
1661
1662 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1663 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1664 #endif
1665
1666 static void sighand_ctor(void *data)
1667 {
1668         struct sighand_struct *sighand = data;
1669
1670         spin_lock_init(&sighand->siglock);
1671         init_waitqueue_head(&sighand->signalfd_wqh);
1672 }
1673
1674 void __init proc_caches_init(void)
1675 {
1676         sighand_cachep = kmem_cache_create("sighand_cache",
1677                         sizeof(struct sighand_struct), 0,
1678                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1679                         SLAB_NOTRACK, sighand_ctor);
1680         signal_cachep = kmem_cache_create("signal_cache",
1681                         sizeof(struct signal_struct), 0,
1682                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1683         files_cachep = kmem_cache_create("files_cache",
1684                         sizeof(struct files_struct), 0,
1685                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1686         fs_cachep = kmem_cache_create("fs_cache",
1687                         sizeof(struct fs_struct), 0,
1688                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1689         /*
1690          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1691          * whole struct cpumask for the OFFSTACK case. We could change
1692          * this to *only* allocate as much of it as required by the
1693          * maximum number of CPU's we can ever have.  The cpumask_allocation
1694          * is at the end of the structure, exactly for that reason.
1695          */
1696         mm_cachep = kmem_cache_create("mm_struct",
1697                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1698                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1699         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1700         mmap_init();
1701         nsproxy_cache_init();
1702 }
1703
1704 /*
1705  * Check constraints on flags passed to the unshare system call.
1706  */
1707 static int check_unshare_flags(unsigned long unshare_flags)
1708 {
1709         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1710                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1711                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1712                 return -EINVAL;
1713         /*
1714          * Not implemented, but pretend it works if there is nothing to
1715          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1716          * needs to unshare vm.
1717          */
1718         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1719                 /* FIXME: get_task_mm() increments ->mm_users */
1720                 if (atomic_read(&current->mm->mm_users) > 1)
1721                         return -EINVAL;
1722         }
1723
1724         return 0;
1725 }
1726
1727 /*
1728  * Unshare the filesystem structure if it is being shared
1729  */
1730 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1731 {
1732         struct fs_struct *fs = current->fs;
1733
1734         if (!(unshare_flags & CLONE_FS) || !fs)
1735                 return 0;
1736
1737         /* don't need lock here; in the worst case we'll do useless copy */
1738         if (fs->users == 1)
1739                 return 0;
1740
1741         *new_fsp = copy_fs_struct(fs);
1742         if (!*new_fsp)
1743                 return -ENOMEM;
1744
1745         return 0;
1746 }
1747
1748 /*
1749  * Unshare file descriptor table if it is being shared
1750  */
1751 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1752 {
1753         struct files_struct *fd = current->files;
1754         int error = 0;
1755
1756         if ((unshare_flags & CLONE_FILES) &&
1757             (fd && atomic_read(&fd->count) > 1)) {
1758                 *new_fdp = dup_fd(fd, &error);
1759                 if (!*new_fdp)
1760                         return error;
1761         }
1762
1763         return 0;
1764 }
1765
1766 /*
1767  * unshare allows a process to 'unshare' part of the process
1768  * context which was originally shared using clone.  copy_*
1769  * functions used by do_fork() cannot be used here directly
1770  * because they modify an inactive task_struct that is being
1771  * constructed. Here we are modifying the current, active,
1772  * task_struct.
1773  */
1774 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1775 {
1776         struct fs_struct *fs, *new_fs = NULL;
1777         struct files_struct *fd, *new_fd = NULL;
1778         struct nsproxy *new_nsproxy = NULL;
1779         int do_sysvsem = 0;
1780         int err;
1781
1782         err = check_unshare_flags(unshare_flags);
1783         if (err)
1784                 goto bad_unshare_out;
1785
1786         /*
1787          * If unsharing namespace, must also unshare filesystem information.
1788          */
1789         if (unshare_flags & CLONE_NEWNS)
1790                 unshare_flags |= CLONE_FS;
1791         /*
1792          * CLONE_NEWIPC must also detach from the undolist: after switching
1793          * to a new ipc namespace, the semaphore arrays from the old
1794          * namespace are unreachable.
1795          */
1796         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1797                 do_sysvsem = 1;
1798         err = unshare_fs(unshare_flags, &new_fs);
1799         if (err)
1800                 goto bad_unshare_out;
1801         err = unshare_fd(unshare_flags, &new_fd);
1802         if (err)
1803                 goto bad_unshare_cleanup_fs;
1804         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1805         if (err)
1806                 goto bad_unshare_cleanup_fd;
1807
1808         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1809                 if (do_sysvsem) {
1810                         /*
1811                          * CLONE_SYSVSEM is equivalent to sys_exit().
1812                          */
1813                         exit_sem(current);
1814                 }
1815
1816                 if (new_nsproxy) {
1817                         switch_task_namespaces(current, new_nsproxy);
1818                         new_nsproxy = NULL;
1819                 }
1820
1821                 task_lock(current);
1822
1823                 if (new_fs) {
1824                         fs = current->fs;
1825                         spin_lock(&fs->lock);
1826                         current->fs = new_fs;
1827                         if (--fs->users)
1828                                 new_fs = NULL;
1829                         else
1830                                 new_fs = fs;
1831                         spin_unlock(&fs->lock);
1832                 }
1833
1834                 if (new_fd) {
1835                         fd = current->files;
1836                         current->files = new_fd;
1837                         new_fd = fd;
1838                 }
1839
1840                 task_unlock(current);
1841         }
1842
1843         if (new_nsproxy)
1844                 put_nsproxy(new_nsproxy);
1845
1846 bad_unshare_cleanup_fd:
1847         if (new_fd)
1848                 put_files_struct(new_fd);
1849
1850 bad_unshare_cleanup_fs:
1851         if (new_fs)
1852                 free_fs_struct(new_fs);
1853
1854 bad_unshare_out:
1855         return err;
1856 }
1857
1858 /*
1859  *      Helper to unshare the files of the current task.
1860  *      We don't want to expose copy_files internals to
1861  *      the exec layer of the kernel.
1862  */
1863
1864 int unshare_files(struct files_struct **displaced)
1865 {
1866         struct task_struct *task = current;
1867         struct files_struct *copy = NULL;
1868         int error;
1869
1870         error = unshare_fd(CLONE_FILES, &copy);
1871         if (error || !copy) {
1872                 *displaced = NULL;
1873                 return error;
1874         }
1875         *displaced = task->files;
1876         task_lock(task);
1877         task->files = copy;
1878         task_unlock(task);
1879         return 0;
1880 }