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/swap.h>
  38 #include <linux/syscalls.h>
  39 #include <linux/jiffies.h>
  40 #include <linux/futex.h>
  41 #include <linux/compat.h>
  42 #include <linux/kthread.h>
  43 #include <linux/task_io_accounting_ops.h>
  44 #include <linux/rcupdate.h>
  45 #include <linux/ptrace.h>
  46 #include <linux/mount.h>
  47 #include <linux/audit.h>
  48 #include <linux/memcontrol.h>
  49 #include <linux/ftrace.h>
  50 #include <linux/proc_fs.h>
  51 #include <linux/profile.h>
  52 #include <linux/rmap.h>
  53 #include <linux/ksm.h>
  54 #include <linux/acct.h>
  55 #include <linux/tsacct_kern.h>
  56 #include <linux/cn_proc.h>
  57 #include <linux/freezer.h>
  58 #include <linux/delayacct.h>
  59 #include <linux/taskstats_kern.h>
  60 #include <linux/random.h>
  61 #include <linux/tty.h>
  62 #include <linux/blkdev.h>
  63 #include <linux/fs_struct.h>
  64 #include <linux/magic.h>
  65 #include <linux/perf_event.h>
  66 #include <linux/posix-timers.h>
  67 #include <linux/user-return-notifier.h>
  68 #include <linux/oom.h>
  69 #include <linux/khugepaged.h>
  70 #include <linux/signalfd.h>
  71
  72 #include <asm/pgtable.h>
  73 #include <asm/pgalloc.h>
  74 #include <asm/uaccess.h>
  75 #include <asm/mmu_context.h>
  76 #include <asm/cacheflush.h>
  77 #include <asm/tlbflush.h>
  78
  79 #include <trace/events/sched.h>
  80
  81 #define CREATE_TRACE_POINTS
  82 #include <trace/events/task.h>
  83
  84 /*
  85  * Protected counters by write_lock_irq(&tasklist_lock)
  86  */
  87 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  88 int nr_threads;                 /* The idle threads do not count.. */
  89
  90 int max_threads;                /* tunable limit on nr_threads */
  91
  92 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  93
  94 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  95
  96 #ifdef CONFIG_PROVE_RCU
  97 int lockdep_tasklist_lock_is_held(void)
  98 {
  99         return lockdep_is_held(&tasklist_lock);
 100 }
 101 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 102 #endif /* #ifdef CONFIG_PROVE_RCU */
 103
 104 int nr_processes(void)
 105 {
 106         int cpu;
 107         int total = 0;
 108
 109         for_each_possible_cpu(cpu)
 110                 total += per_cpu(process_counts, cpu);
 111
 112         return total;
 113 }
 114
 115 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 116 # define alloc_task_struct_node(node)           \
 117                 kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node)
 118 # define free_task_struct(tsk)                  \
 119                 kmem_cache_free(task_struct_cachep, (tsk))
 120 static struct kmem_cache *task_struct_cachep;
 121 #endif
 122
 123 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 124 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 125                                                   int node)
 126 {
 127 #ifdef CONFIG_DEBUG_STACK_USAGE
 128         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 129 #else
 130         gfp_t mask = GFP_KERNEL;
 131 #endif
 132         struct page *page = alloc_pages_node(node, mask, THREAD_SIZE_ORDER);
 133
 134         return page ? page_address(page) : NULL;
 135 }
 136
 137 static inline void free_thread_info(struct thread_info *ti)
 138 {
 139         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 140 }
 141 #endif
 142
 143 /* SLAB cache for signal_struct structures (tsk->signal) */
 144 static struct kmem_cache *signal_cachep;
 145
 146 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 147 struct kmem_cache *sighand_cachep;
 148
 149 /* SLAB cache for files_struct structures (tsk->files) */
 150 struct kmem_cache *files_cachep;
 151
 152 /* SLAB cache for fs_struct structures (tsk->fs) */
 153 struct kmem_cache *fs_cachep;
 154
 155 /* SLAB cache for vm_area_struct structures */
 156 struct kmem_cache *vm_area_cachep;
 157
 158 /* SLAB cache for mm_struct structures (tsk->mm) */
 159 static struct kmem_cache *mm_cachep;
 160
 161 static void account_kernel_stack(struct thread_info *ti, int account)
 162 {
 163         struct zone *zone = page_zone(virt_to_page(ti));
 164
 165         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 166 }
 167
 168 void free_task(struct task_struct *tsk)
 169 {
 170         account_kernel_stack(tsk->stack, -1);
 171         free_thread_info(tsk->stack);
 172         rt_mutex_debug_task_free(tsk);
 173         ftrace_graph_exit_task(tsk);
 174         free_task_struct(tsk);
 175 }
 176 EXPORT_SYMBOL(free_task);
 177
 178 static inline void free_signal_struct(struct signal_struct *sig)
 179 {
 180         taskstats_tgid_free(sig);
 181         sched_autogroup_exit(sig);
 182         kmem_cache_free(signal_cachep, sig);
 183 }
 184
 185 static inline void put_signal_struct(struct signal_struct *sig)
 186 {
 187         if (atomic_dec_and_test(&sig->sigcnt))
 188                 free_signal_struct(sig);
 189 }
 190
 191 void __put_task_struct(struct task_struct *tsk)
 192 {
 193         WARN_ON(!tsk->exit_state);
 194         WARN_ON(atomic_read(&tsk->usage));
 195         WARN_ON(tsk == current);
 196
 197         exit_creds(tsk);
 198         delayacct_tsk_free(tsk);
 199         put_signal_struct(tsk->signal);
 200
 201         if (!profile_handoff_task(tsk))
 202                 free_task(tsk);
 203 }
 204 EXPORT_SYMBOL_GPL(__put_task_struct);
 205
 206 /*
 207  * macro override instead of weak attribute alias, to workaround
 208  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 209  */
 210 #ifndef arch_task_cache_init
 211 #define arch_task_cache_init()
 212 #endif
 213
 214 void __init fork_init(unsigned long mempages)
 215 {
 216 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 217 #ifndef ARCH_MIN_TASKALIGN
 218 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 219 #endif
 220         /* create a slab on which task_structs can be allocated */
 221         task_struct_cachep =
 222                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 223                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 224 #endif
 225
 226         /* do the arch specific task caches init */
 227         arch_task_cache_init();
 228
 229         /*
 230          * The default maximum number of threads is set to a safe
 231          * value: the thread structures can take up at most half
 232          * of memory.
 233          */
 234         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 235
 236         /*
 237          * we need to allow at least 20 threads to boot a system
 238          */
 239         if (max_threads < 20)
 240                 max_threads = 20;
 241
 242         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 243         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 244         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 245                 init_task.signal->rlim[RLIMIT_NPROC];
 246 }
 247
 248 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 249                                                struct task_struct *src)
 250 {
 251         *dst = *src;
 252         return 0;
 253 }
 254
 255 static struct task_struct *dup_task_struct(struct task_struct *orig)
 256 {
 257         struct task_struct *tsk;
 258         struct thread_info *ti;
 259         unsigned long *stackend;
 260         int node = tsk_fork_get_node(orig);
 261         int err;
 262
 263         prepare_to_copy(orig);
 264
 265         tsk = alloc_task_struct_node(node);
 266         if (!tsk)
 267                 return NULL;
 268
 269         ti = alloc_thread_info_node(tsk, node);
 270         if (!ti) {
 271                 free_task_struct(tsk);
 272                 return NULL;
 273         }
 274
 275         err = arch_dup_task_struct(tsk, orig);
 276         if (err)
 277                 goto out;
 278
 279         tsk->stack = ti;
 280
 281         setup_thread_stack(tsk, orig);
 282         clear_user_return_notifier(tsk);
 283         clear_tsk_need_resched(tsk);
 284         stackend = end_of_stack(tsk);
 285         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 286
 287 #ifdef CONFIG_CC_STACKPROTECTOR
 288         tsk->stack_canary = get_random_int();
 289 #endif
 290
 291         /*
 292          * One for us, one for whoever does the "release_task()" (usually
 293          * parent)
 294          */
 295         atomic_set(&tsk->usage, 2);
 296 #ifdef CONFIG_BLK_DEV_IO_TRACE
 297         tsk->btrace_seq = 0;
 298 #endif
 299         tsk->splice_pipe = NULL;
 300
 301         account_kernel_stack(ti, 1);
 302
 303         return tsk;
 304
 305 out:
 306         free_thread_info(ti);
 307         free_task_struct(tsk);
 308         return NULL;
 309 }
 310
 311 #ifdef CONFIG_MMU
 312 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 313 {
 314         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 315         struct rb_node **rb_link, *rb_parent;
 316         int retval;
 317         unsigned long charge;
 318         struct mempolicy *pol;
 319
 320         down_write(&oldmm->mmap_sem);
 321         flush_cache_dup_mm(oldmm);
 322         /*
 323          * Not linked in yet - no deadlock potential:
 324          */
 325         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 326
 327         mm->locked_vm = 0;
 328         mm->mmap = NULL;
 329         mm->mmap_cache = NULL;
 330         mm->free_area_cache = oldmm->mmap_base;
 331         mm->cached_hole_size = ~0UL;
 332         mm->map_count = 0;
 333         cpumask_clear(mm_cpumask(mm));
 334         mm->mm_rb = RB_ROOT;
 335         rb_link = &mm->mm_rb.rb_node;
 336         rb_parent = NULL;
 337         pprev = &mm->mmap;
 338         retval = ksm_fork(mm, oldmm);
 339         if (retval)
 340                 goto out;
 341         retval = khugepaged_fork(mm, oldmm);
 342         if (retval)
 343                 goto out;
 344
 345         prev = NULL;
 346         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 347                 struct file *file;
 348
 349                 if (mpnt->vm_flags & VM_DONTCOPY) {
 350                         long pages = vma_pages(mpnt);
 351                         mm->total_vm -= pages;
 352                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 353                                                                 -pages);
 354                         continue;
 355                 }
 356                 charge = 0;
 357                 if (mpnt->vm_flags & VM_ACCOUNT) {
 358                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 359                         if (security_vm_enough_memory(len))
 360                                 goto fail_nomem;
 361                         charge = len;
 362                 }
 363                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 364                 if (!tmp)
 365                         goto fail_nomem;
 366                 *tmp = *mpnt;
 367                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 368                 pol = mpol_dup(vma_policy(mpnt));
 369                 retval = PTR_ERR(pol);
 370                 if (IS_ERR(pol))
 371                         goto fail_nomem_policy;
 372                 vma_set_policy(tmp, pol);
 373                 tmp->vm_mm = mm;
 374                 if (anon_vma_fork(tmp, mpnt))
 375                         goto fail_nomem_anon_vma_fork;
 376                 tmp->vm_flags &= ~VM_LOCKED;
 377                 tmp->vm_next = tmp->vm_prev = NULL;
 378                 file = tmp->vm_file;
 379                 if (file) {
 380                         struct inode *inode = file->f_path.dentry->d_inode;
 381                         struct address_space *mapping = file->f_mapping;
 382
 383                         get_file(file);
 384                         if (tmp->vm_prfile)
 385                                 get_file(tmp->vm_prfile);
 386                         if (tmp->vm_flags & VM_DENYWRITE)
 387                                 atomic_dec(&inode->i_writecount);
 388                         mutex_lock(&mapping->i_mmap_mutex);
 389                         if (tmp->vm_flags & VM_SHARED)
 390                                 mapping->i_mmap_writable++;
 391                         flush_dcache_mmap_lock(mapping);
 392                         /* insert tmp into the share list, just after mpnt */
 393                         vma_prio_tree_add(tmp, mpnt);
 394                         flush_dcache_mmap_unlock(mapping);
 395                         mutex_unlock(&mapping->i_mmap_mutex);
 396                 }
 397
 398                 /*
 399                  * Clear hugetlb-related page reserves for children. This only
 400                  * affects MAP_PRIVATE mappings. Faults generated by the child
 401                  * are not guaranteed to succeed, even if read-only
 402                  */
 403                 if (is_vm_hugetlb_page(tmp))
 404                         reset_vma_resv_huge_pages(tmp);
 405
 406                 /*
 407                  * Link in the new vma and copy the page table entries.
 408                  */
 409                 *pprev = tmp;
 410                 pprev = &tmp->vm_next;
 411                 tmp->vm_prev = prev;
 412                 prev = tmp;
 413
 414                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 415                 rb_link = &tmp->vm_rb.rb_right;
 416                 rb_parent = &tmp->vm_rb;
 417
 418                 mm->map_count++;
 419                 retval = copy_page_range(mm, oldmm, mpnt);
 420
 421                 if (tmp->vm_ops && tmp->vm_ops->open)
 422                         tmp->vm_ops->open(tmp);
 423
 424                 if (retval)
 425                         goto out;
 426         }
 427         /* a new mm has just been created */
 428         arch_dup_mmap(oldmm, mm);
 429         retval = 0;
 430 out:
 431         up_write(&mm->mmap_sem);
 432         flush_tlb_mm(oldmm);
 433         up_write(&oldmm->mmap_sem);
 434         return retval;
 435 fail_nomem_anon_vma_fork:
 436         mpol_put(pol);
 437 fail_nomem_policy:
 438         kmem_cache_free(vm_area_cachep, tmp);
 439 fail_nomem:
 440         retval = -ENOMEM;
 441         vm_unacct_memory(charge);
 442         goto out;
 443 }
 444
 445 static inline int mm_alloc_pgd(struct mm_struct *mm)
 446 {
 447         mm->pgd = pgd_alloc(mm);
 448         if (unlikely(!mm->pgd))
 449                 return -ENOMEM;
 450         return 0;
 451 }
 452
 453 static inline void mm_free_pgd(struct mm_struct *mm)
 454 {
 455         pgd_free(mm, mm->pgd);
 456 }
 457 #else
 458 #define dup_mmap(mm, oldmm)     (0)
 459 #define mm_alloc_pgd(mm)        (0)
 460 #define mm_free_pgd(mm)
 461 #endif /* CONFIG_MMU */
 462
 463 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 464
 465 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 466 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 467
 468 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 469
 470 static int __init coredump_filter_setup(char *s)
 471 {
 472         default_dump_filter =
 473                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 474                 MMF_DUMP_FILTER_MASK;
 475         return 1;
 476 }
 477
 478 __setup("coredump_filter=", coredump_filter_setup);
 479
 480 #include <linux/init_task.h>
 481
 482 static void mm_init_aio(struct mm_struct *mm)
 483 {
 484 #ifdef CONFIG_AIO
 485         spin_lock_init(&mm->ioctx_lock);
 486         INIT_HLIST_HEAD(&mm->ioctx_list);
 487 #endif
 488 }
 489
 490 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 491 {
 492         atomic_set(&mm->mm_users, 1);
 493         atomic_set(&mm->mm_count, 1);
 494         init_rwsem(&mm->mmap_sem);
 495         INIT_LIST_HEAD(&mm->mmlist);
 496         mm->flags = (current->mm) ?
 497                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 498         mm->core_state = NULL;
 499         mm->nr_ptes = 0;
 500         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 501         spin_lock_init(&mm->page_table_lock);
 502         mm->free_area_cache = TASK_UNMAPPED_BASE;
 503         mm->cached_hole_size = ~0UL;
 504         mm_init_aio(mm);
 505         mm_init_owner(mm, p);
 506
 507         if (likely(!mm_alloc_pgd(mm))) {
 508                 mm->def_flags = 0;
 509                 mmu_notifier_mm_init(mm);
 510                 return mm;
 511         }
 512
 513         free_mm(mm);
 514         return NULL;
 515 }
 516
 517 /*
 518  * Allocate and initialize an mm_struct.
 519  */
 520 struct mm_struct *mm_alloc(void)
 521 {
 522         struct mm_struct *mm;
 523
 524         mm = allocate_mm();
 525         if (!mm)
 526                 return NULL;
 527
 528         memset(mm, 0, sizeof(*mm));
 529         mm_init_cpumask(mm);
 530         return mm_init(mm, current);
 531 }
 532
 533 /*
 534  * Called when the last reference to the mm
 535  * is dropped: either by a lazy thread or by
 536  * mmput. Free the page directory and the mm.
 537  */
 538 void __mmdrop(struct mm_struct *mm)
 539 {
 540         BUG_ON(mm == &init_mm);
 541         mm_free_pgd(mm);
 542         destroy_context(mm);
 543         mmu_notifier_mm_destroy(mm);
 544 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 545         VM_BUG_ON(mm->pmd_huge_pte);
 546 #endif
 547         free_mm(mm);
 548 }
 549 EXPORT_SYMBOL_GPL(__mmdrop);
 550
 551 /*
 552  * Decrement the use count and release all resources for an mm.
 553  */
 554 void mmput(struct mm_struct *mm)
 555 {
 556         might_sleep();
 557
 558         if (atomic_dec_and_test(&mm->mm_users)) {
 559                 exit_aio(mm);
 560                 ksm_exit(mm);
 561                 khugepaged_exit(mm); /* must run before exit_mmap */
 562                 exit_mmap(mm);
 563                 set_mm_exe_file(mm, NULL);
 564                 if (!list_empty(&mm->mmlist)) {
 565                         spin_lock(&mmlist_lock);
 566                         list_del(&mm->mmlist);
 567                         spin_unlock(&mmlist_lock);
 568                 }
 569                 put_swap_token(mm);
 570                 if (mm->binfmt)
 571                         module_put(mm->binfmt->module);
 572                 mmdrop(mm);
 573         }
 574 }
 575 EXPORT_SYMBOL_GPL(mmput);
 576
 577 /*
 578  * We added or removed a vma mapping the executable. The vmas are only mapped
 579  * during exec and are not mapped with the mmap system call.
 580  * Callers must hold down_write() on the mm's mmap_sem for these
 581  */
 582 void added_exe_file_vma(struct mm_struct *mm)
 583 {
 584         mm->num_exe_file_vmas++;
 585 }
 586
 587 void removed_exe_file_vma(struct mm_struct *mm)
 588 {
 589         mm->num_exe_file_vmas--;
 590         if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
 591                 fput(mm->exe_file);
 592                 mm->exe_file = NULL;
 593         }
 594
 595 }
 596
 597 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 598 {
 599         if (new_exe_file)
 600                 get_file(new_exe_file);
 601         if (mm->exe_file)
 602                 fput(mm->exe_file);
 603         mm->exe_file = new_exe_file;
 604         mm->num_exe_file_vmas = 0;
 605 }
 606
 607 struct file *get_mm_exe_file(struct mm_struct *mm)
 608 {
 609         struct file *exe_file;
 610
 611         /* We need mmap_sem to protect against races with removal of
 612          * VM_EXECUTABLE vmas */
 613         down_read(&mm->mmap_sem);
 614         exe_file = mm->exe_file;
 615         if (exe_file)
 616                 get_file(exe_file);
 617         up_read(&mm->mmap_sem);
 618         return exe_file;
 619 }
 620
 621 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 622 {
 623         /* It's safe to write the exe_file pointer without exe_file_lock because
 624          * this is called during fork when the task is not yet in /proc */
 625         newmm->exe_file = get_mm_exe_file(oldmm);
 626 }
 627
 628 /**
 629  * get_task_mm - acquire a reference to the task's mm
 630  *
 631  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 632  * this kernel workthread has transiently adopted a user mm with use_mm,
 633  * to do its AIO) is not set and if so returns a reference to it, after
 634  * bumping up the use count.  User must release the mm via mmput()
 635  * after use.  Typically used by /proc and ptrace.
 636  */
 637 struct mm_struct *get_task_mm(struct task_struct *task)
 638 {
 639         struct mm_struct *mm;
 640
 641         task_lock(task);
 642         mm = task->mm;
 643         if (mm) {
 644                 if (task->flags & PF_KTHREAD)
 645                         mm = NULL;
 646                 else
 647                         atomic_inc(&mm->mm_users);
 648         }
 649         task_unlock(task);
 650         return mm;
 651 }
 652 EXPORT_SYMBOL_GPL(get_task_mm);
 653
 654 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 655 {
 656         struct mm_struct *mm;
 657         int err;
 658
 659         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 660         if (err)
 661                 return ERR_PTR(err);
 662
 663         mm = get_task_mm(task);
 664         if (mm && mm != current->mm &&
 665                         !ptrace_may_access(task, mode)) {
 666                 mmput(mm);
 667                 mm = ERR_PTR(-EACCES);
 668         }
 669         mutex_unlock(&task->signal->cred_guard_mutex);
 670
 671         return mm;
 672 }
 673
 674 static void complete_vfork_done(struct task_struct *tsk)
 675 {
 676         struct completion *vfork;
 677
 678         task_lock(tsk);
 679         vfork = tsk->vfork_done;
 680         if (likely(vfork)) {
 681                 tsk->vfork_done = NULL;
 682                 complete(vfork);
 683         }
 684         task_unlock(tsk);
 685 }
 686
 687 static int wait_for_vfork_done(struct task_struct *child,
 688                                 struct completion *vfork)
 689 {
 690         int killed;
 691
 692         freezer_do_not_count();
 693         killed = wait_for_completion_killable(vfork);
 694         freezer_count();
 695
 696         if (killed) {
 697                 task_lock(child);
 698                 child->vfork_done = NULL;
 699                 task_unlock(child);
 700         }
 701
 702         put_task_struct(child);
 703         return killed;
 704 }
 705
 706 /* Please note the differences between mmput and mm_release.
 707  * mmput is called whenever we stop holding onto a mm_struct,
 708  * error success whatever.
 709  *
 710  * mm_release is called after a mm_struct has been removed
 711  * from the current process.
 712  *
 713  * This difference is important for error handling, when we
 714  * only half set up a mm_struct for a new process and need to restore
 715  * the old one.  Because we mmput the new mm_struct before
 716  * restoring the old one. . .
 717  * Eric Biederman 10 January 1998
 718  */
 719 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 720 {
 721         /* Get rid of any futexes when releasing the mm */
 722 #ifdef CONFIG_FUTEX
 723         if (unlikely(tsk->robust_list)) {
 724                 exit_robust_list(tsk);
 725                 tsk->robust_list = NULL;
 726         }
 727 #ifdef CONFIG_COMPAT
 728         if (unlikely(tsk->compat_robust_list)) {
 729                 compat_exit_robust_list(tsk);
 730                 tsk->compat_robust_list = NULL;
 731         }
 732 #endif
 733         if (unlikely(!list_empty(&tsk->pi_state_list)))
 734                 exit_pi_state_list(tsk);
 735 #endif
 736
 737         /* Get rid of any cached register state */
 738         deactivate_mm(tsk, mm);
 739
 740         if (tsk->vfork_done)
 741                 complete_vfork_done(tsk);
 742
 743         /*
 744          * If we're exiting normally, clear a user-space tid field if
 745          * requested.  We leave this alone when dying by signal, to leave
 746          * the value intact in a core dump, and to save the unnecessary
 747          * trouble, say, a killed vfork parent shouldn't touch this mm.
 748          * Userland only wants this done for a sys_exit.
 749          */
 750         if (tsk->clear_child_tid) {
 751                 if (!(tsk->flags & PF_SIGNALED) &&
 752                     atomic_read(&mm->mm_users) > 1) {
 753                         /*
 754                          * We don't check the error code - if userspace has
 755                          * not set up a proper pointer then tough luck.
 756                          */
 757                         put_user(0, tsk->clear_child_tid);
 758                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 759                                         1, NULL, NULL, 0);
 760                 }
 761                 tsk->clear_child_tid = NULL;
 762         }
 763 }
 764
 765 /*
 766  * Allocate a new mm structure and copy contents from the
 767  * mm structure of the passed in task structure.
 768  */
 769 struct mm_struct *dup_mm(struct task_struct *tsk)
 770 {
 771         struct mm_struct *mm, *oldmm = current->mm;
 772         int err;
 773
 774         if (!oldmm)
 775                 return NULL;
 776
 777         mm = allocate_mm();
 778         if (!mm)
 779                 goto fail_nomem;
 780
 781         memcpy(mm, oldmm, sizeof(*mm));
 782         mm_init_cpumask(mm);
 783
 784         /* Initializing for Swap token stuff */
 785         mm->token_priority = 0;
 786         mm->last_interval = 0;
 787
 788 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 789         mm->pmd_huge_pte = NULL;
 790 #endif
 791
 792         if (!mm_init(mm, tsk))
 793                 goto fail_nomem;
 794
 795         if (init_new_context(tsk, mm))
 796                 goto fail_nocontext;
 797
 798         dup_mm_exe_file(oldmm, mm);
 799
 800         err = dup_mmap(mm, oldmm);
 801         if (err)
 802                 goto free_pt;
 803
 804         mm->hiwater_rss = get_mm_rss(mm);
 805         mm->hiwater_vm = mm->total_vm;
 806
 807         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 808                 goto free_pt;
 809
 810         return mm;
 811
 812 free_pt:
 813         /* don't put binfmt in mmput, we haven't got module yet */
 814         mm->binfmt = NULL;
 815         mmput(mm);
 816
 817 fail_nomem:
 818         return NULL;
 819
 820 fail_nocontext:
 821         /*
 822          * If init_new_context() failed, we cannot use mmput() to free the mm
 823          * because it calls destroy_context()
 824          */
 825         mm_free_pgd(mm);
 826         free_mm(mm);
 827         return NULL;
 828 }
 829
 830 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 831 {
 832         struct mm_struct *mm, *oldmm;
 833         int retval;
 834
 835         tsk->min_flt = tsk->maj_flt = 0;
 836         tsk->nvcsw = tsk->nivcsw = 0;
 837 #ifdef CONFIG_DETECT_HUNG_TASK
 838         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 839 #endif
 840
 841         tsk->mm = NULL;
 842         tsk->active_mm = NULL;
 843
 844         /*
 845          * Are we cloning a kernel thread?
 846          *
 847          * We need to steal a active VM for that..
 848          */
 849         oldmm = current->mm;
 850         if (!oldmm)
 851                 return 0;
 852
 853         if (clone_flags & CLONE_VM) {
 854                 atomic_inc(&oldmm->mm_users);
 855                 mm = oldmm;
 856                 goto good_mm;
 857         }
 858
 859         retval = -ENOMEM;
 860         mm = dup_mm(tsk);
 861         if (!mm)
 862                 goto fail_nomem;
 863
 864 good_mm:
 865         /* Initializing for Swap token stuff */
 866         mm->token_priority = 0;
 867         mm->last_interval = 0;
 868
 869         tsk->mm = mm;
 870         tsk->active_mm = mm;
 871         return 0;
 872
 873 fail_nomem:
 874         return retval;
 875 }
 876
 877 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 878 {
 879         struct fs_struct *fs = current->fs;
 880         if (clone_flags & CLONE_FS) {
 881                 /* tsk->fs is already what we want */
 882                 spin_lock(&fs->lock);
 883                 if (fs->in_exec) {
 884                         spin_unlock(&fs->lock);
 885                         return -EAGAIN;
 886                 }
 887                 fs->users++;
 888                 spin_unlock(&fs->lock);
 889                 return 0;
 890         }
 891         tsk->fs = copy_fs_struct(fs);
 892         if (!tsk->fs)
 893                 return -ENOMEM;
 894         return 0;
 895 }
 896
 897 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 898 {
 899         struct files_struct *oldf, *newf;
 900         int error = 0;
 901
 902         /*
 903          * A background process may not have any files ...
 904          */
 905         oldf = current->files;
 906         if (!oldf)
 907                 goto out;
 908
 909         if (clone_flags & CLONE_FILES) {
 910                 atomic_inc(&oldf->count);
 911                 goto out;
 912         }
 913
 914         newf = dup_fd(oldf, &error);
 915         if (!newf)
 916                 goto out;
 917
 918         tsk->files = newf;
 919         error = 0;
 920 out:
 921         return error;
 922 }
 923
 924 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 925 {
 926 #ifdef CONFIG_BLOCK
 927         struct io_context *ioc = current->io_context;
 928         struct io_context *new_ioc;
 929
 930         if (!ioc)
 931                 return 0;
 932         /*
 933          * Share io context with parent, if CLONE_IO is set
 934          */
 935         if (clone_flags & CLONE_IO) {
 936                 tsk->io_context = ioc_task_link(ioc);
 937                 if (unlikely(!tsk->io_context))
 938                         return -ENOMEM;
 939         } else if (ioprio_valid(ioc->ioprio)) {
 940                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
 941                 if (unlikely(!new_ioc))
 942                         return -ENOMEM;
 943
 944                 new_ioc->ioprio = ioc->ioprio;
 945                 put_io_context(new_ioc);
 946         }
 947 #endif
 948         return 0;
 949 }
 950
 951 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 952 {
 953         struct sighand_struct *sig;
 954
 955         if (clone_flags & CLONE_SIGHAND) {
 956                 atomic_inc(&current->sighand->count);
 957                 return 0;
 958         }
 959         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 960         rcu_assign_pointer(tsk->sighand, sig);
 961         if (!sig)
 962                 return -ENOMEM;
 963         atomic_set(&sig->count, 1);
 964         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 965         return 0;
 966 }
 967
 968 void __cleanup_sighand(struct sighand_struct *sighand)
 969 {
 970         if (atomic_dec_and_test(&sighand->count)) {
 971                 signalfd_cleanup(sighand);
 972                 kmem_cache_free(sighand_cachep, sighand);
 973         }
 974 }
 975
 976
 977 /*
 978  * Initialize POSIX timer handling for a thread group.
 979  */
 980 static void posix_cpu_timers_init_group(struct signal_struct *sig)
 981 {
 982         unsigned long cpu_limit;
 983
 984         /* Thread group counters. */
 985         thread_group_cputime_init(sig);
 986
 987         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
 988         if (cpu_limit != RLIM_INFINITY) {
 989                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
 990                 sig->cputimer.running = 1;
 991         }
 992
 993         /* The timer lists. */
 994         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 995         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 996         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 997 }
 998
 999 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1000 {
1001         struct signal_struct *sig;
1002
1003         if (clone_flags & CLONE_THREAD)
1004                 return 0;
1005
1006         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1007         tsk->signal = sig;
1008         if (!sig)
1009                 return -ENOMEM;
1010
1011         sig->nr_threads = 1;
1012         atomic_set(&sig->live, 1);
1013         atomic_set(&sig->sigcnt, 1);
1014         init_waitqueue_head(&sig->wait_chldexit);
1015         if (clone_flags & CLONE_NEWPID)
1016                 sig->flags |= SIGNAL_UNKILLABLE;
1017         sig->curr_target = tsk;
1018         init_sigpending(&sig->shared_pending);
1019         INIT_LIST_HEAD(&sig->posix_timers);
1020
1021         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1022         sig->real_timer.function = it_real_fn;
1023
1024         task_lock(current->group_leader);
1025         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1026         task_unlock(current->group_leader);
1027
1028         posix_cpu_timers_init_group(sig);
1029
1030         tty_audit_fork(sig);
1031         sched_autogroup_fork(sig);
1032
1033 #ifdef CONFIG_CGROUPS
1034         init_rwsem(&sig->group_rwsem);
1035 #endif
1036
1037         sig->oom_adj = current->signal->oom_adj;
1038         sig->oom_score_adj = current->signal->oom_score_adj;
1039         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1040
1041         mutex_init(&sig->cred_guard_mutex);
1042
1043         return 0;
1044 }
1045
1046 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1047 {
1048         unsigned long new_flags = p->flags;
1049
1050         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1051         new_flags |= PF_FORKNOEXEC;
1052         p->flags = new_flags;
1053 }
1054
1055 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1056 {
1057         current->clear_child_tid = tidptr;
1058
1059         return task_pid_vnr(current);
1060 }
1061
1062 static void rt_mutex_init_task(struct task_struct *p)
1063 {
1064         raw_spin_lock_init(&p->pi_lock);
1065 #ifdef CONFIG_RT_MUTEXES
1066         plist_head_init(&p->pi_waiters);
1067         p->pi_blocked_on = NULL;
1068 #endif
1069 }
1070
1071 #ifdef CONFIG_MM_OWNER
1072 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1073 {
1074         mm->owner = p;
1075 }
1076 #endif /* CONFIG_MM_OWNER */
1077
1078 /*
1079  * Initialize POSIX timer handling for a single task.
1080  */
1081 static void posix_cpu_timers_init(struct task_struct *tsk)
1082 {
1083         tsk->cputime_expires.prof_exp = 0;
1084         tsk->cputime_expires.virt_exp = 0;
1085         tsk->cputime_expires.sched_exp = 0;
1086         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1087         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1088         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1089 }
1090
1091 /*
1092  * This creates a new process as a copy of the old one,
1093  * but does not actually start it yet.
1094  *
1095  * It copies the registers, and all the appropriate
1096  * parts of the process environment (as per the clone
1097  * flags). The actual kick-off is left to the caller.
1098  */
1099 static struct task_struct *copy_process(unsigned long clone_flags,
1100                                         unsigned long stack_start,
1101                                         struct pt_regs *regs,
1102                                         unsigned long stack_size,
1103                                         int __user *child_tidptr,
1104                                         struct pid *pid,
1105                                         int trace)
1106 {
1107         int retval;
1108         struct task_struct *p;
1109         int cgroup_callbacks_done = 0;
1110
1111         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1112                 return ERR_PTR(-EINVAL);
1113
1114         /*
1115          * Thread groups must share signals as well, and detached threads
1116          * can only be started up within the thread group.
1117          */
1118         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1119                 return ERR_PTR(-EINVAL);
1120
1121         /*
1122          * Shared signal handlers imply shared VM. By way of the above,
1123          * thread groups also imply shared VM. Blocking this case allows
1124          * for various simplifications in other code.
1125          */
1126         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1127                 return ERR_PTR(-EINVAL);
1128
1129         /*
1130          * Siblings of global init remain as zombies on exit since they are
1131          * not reaped by their parent (swapper). To solve this and to avoid
1132          * multi-rooted process trees, prevent global and container-inits
1133          * from creating siblings.
1134          */
1135         if ((clone_flags & CLONE_PARENT) &&
1136                                 current->signal->flags & SIGNAL_UNKILLABLE)
1137                 return ERR_PTR(-EINVAL);
1138
1139         retval = security_task_create(clone_flags);
1140         if (retval)
1141                 goto fork_out;
1142
1143         retval = -ENOMEM;
1144         p = dup_task_struct(current);
1145         if (!p)
1146                 goto fork_out;
1147
1148         ftrace_graph_init_task(p);
1149
1150         rt_mutex_init_task(p);
1151
1152 #ifdef CONFIG_PROVE_LOCKING
1153         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1154         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1155 #endif
1156         retval = -EAGAIN;
1157         if (atomic_read(&p->real_cred->user->processes) >=
1158                         task_rlimit(p, RLIMIT_NPROC)) {
1159                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1160                     p->real_cred->user != INIT_USER)
1161                         goto bad_fork_free;
1162         }
1163         current->flags &= ~PF_NPROC_EXCEEDED;
1164
1165         retval = copy_creds(p, clone_flags);
1166         if (retval < 0)
1167                 goto bad_fork_free;
1168
1169         /*
1170          * If multiple threads are within copy_process(), then this check
1171          * triggers too late. This doesn't hurt, the check is only there
1172          * to stop root fork bombs.
1173          */
1174         retval = -EAGAIN;
1175         if (nr_threads >= max_threads)
1176                 goto bad_fork_cleanup_count;
1177
1178         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1179                 goto bad_fork_cleanup_count;
1180
1181         p->did_exec = 0;
1182         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1183         copy_flags(clone_flags, p);
1184         INIT_LIST_HEAD(&p->children);
1185         INIT_LIST_HEAD(&p->sibling);
1186         rcu_copy_process(p);
1187         p->vfork_done = NULL;
1188         spin_lock_init(&p->alloc_lock);
1189
1190         init_sigpending(&p->pending);
1191
1192         p->utime = p->stime = p->gtime = 0;
1193         p->utimescaled = p->stimescaled = 0;
1194 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1195         p->prev_utime = p->prev_stime = 0;
1196 #endif
1197 #if defined(SPLIT_RSS_COUNTING)
1198         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1199 #endif
1200
1201         p->default_timer_slack_ns = current->timer_slack_ns;
1202
1203         task_io_accounting_init(&p->ioac);
1204         acct_clear_integrals(p);
1205
1206         posix_cpu_timers_init(p);
1207
1208         do_posix_clock_monotonic_gettime(&p->start_time);
1209         p->real_start_time = p->start_time;
1210         monotonic_to_bootbased(&p->real_start_time);
1211         p->io_context = NULL;
1212         p->audit_context = NULL;
1213         if (clone_flags & CLONE_THREAD)
1214                 threadgroup_change_begin(current);
1215         cgroup_fork(p);
1216 #ifdef CONFIG_NUMA
1217         p->mempolicy = mpol_dup(p->mempolicy);
1218         if (IS_ERR(p->mempolicy)) {
1219                 retval = PTR_ERR(p->mempolicy);
1220                 p->mempolicy = NULL;
1221                 goto bad_fork_cleanup_cgroup;
1222         }
1223         mpol_fix_fork_child_flag(p);
1224 #endif
1225 #ifdef CONFIG_CPUSETS
1226         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1227         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1228 #endif
1229 #ifdef CONFIG_TRACE_IRQFLAGS
1230         p->irq_events = 0;
1231 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1232         p->hardirqs_enabled = 1;
1233 #else
1234         p->hardirqs_enabled = 0;
1235 #endif
1236         p->hardirq_enable_ip = 0;
1237         p->hardirq_enable_event = 0;
1238         p->hardirq_disable_ip = _THIS_IP_;
1239         p->hardirq_disable_event = 0;
1240         p->softirqs_enabled = 1;
1241         p->softirq_enable_ip = _THIS_IP_;
1242         p->softirq_enable_event = 0;
1243         p->softirq_disable_ip = 0;
1244         p->softirq_disable_event = 0;
1245         p->hardirq_context = 0;
1246         p->softirq_context = 0;
1247 #endif
1248 #ifdef CONFIG_LOCKDEP
1249         p->lockdep_depth = 0; /* no locks held yet */
1250         p->curr_chain_key = 0;
1251         p->lockdep_recursion = 0;
1252 #endif
1253
1254 #ifdef CONFIG_DEBUG_MUTEXES
1255         p->blocked_on = NULL; /* not blocked yet */
1256 #endif
1257 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1258         p->memcg_batch.do_batch = 0;
1259         p->memcg_batch.memcg = NULL;
1260 #endif
1261
1262         /* Perform scheduler related setup. Assign this task to a CPU. */
1263         sched_fork(p);
1264
1265         retval = perf_event_init_task(p);
1266         if (retval)
1267                 goto bad_fork_cleanup_policy;
1268         retval = audit_alloc(p);
1269         if (retval)
1270                 goto bad_fork_cleanup_policy;
1271         /* copy all the process information */
1272         retval = copy_semundo(clone_flags, p);
1273         if (retval)
1274                 goto bad_fork_cleanup_audit;
1275         retval = copy_files(clone_flags, p);
1276         if (retval)
1277                 goto bad_fork_cleanup_semundo;
1278         retval = copy_fs(clone_flags, p);
1279         if (retval)
1280                 goto bad_fork_cleanup_files;
1281         retval = copy_sighand(clone_flags, p);
1282         if (retval)
1283                 goto bad_fork_cleanup_fs;
1284         retval = copy_signal(clone_flags, p);
1285         if (retval)
1286                 goto bad_fork_cleanup_sighand;
1287         retval = copy_mm(clone_flags, p);
1288         if (retval)
1289                 goto bad_fork_cleanup_signal;
1290         retval = copy_namespaces(clone_flags, p);
1291         if (retval)
1292                 goto bad_fork_cleanup_mm;
1293         retval = copy_io(clone_flags, p);
1294         if (retval)
1295                 goto bad_fork_cleanup_namespaces;
1296         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1297         if (retval)
1298                 goto bad_fork_cleanup_io;
1299
1300         if (pid != &init_struct_pid) {
1301                 retval = -ENOMEM;
1302                 pid = alloc_pid(p->nsproxy->pid_ns);
1303                 if (!pid)
1304                         goto bad_fork_cleanup_io;
1305         }
1306
1307         p->pid = pid_nr(pid);
1308         p->tgid = p->pid;
1309         if (clone_flags & CLONE_THREAD)
1310                 p->tgid = current->tgid;
1311
1312         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1313         /*
1314          * Clear TID on mm_release()?
1315          */
1316         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1317 #ifdef CONFIG_BLOCK
1318         p->plug = NULL;
1319 #endif
1320 #ifdef CONFIG_FUTEX
1321         p->robust_list = NULL;
1322 #ifdef CONFIG_COMPAT
1323         p->compat_robust_list = NULL;
1324 #endif
1325         INIT_LIST_HEAD(&p->pi_state_list);
1326         p->pi_state_cache = NULL;
1327 #endif
1328         /*
1329          * sigaltstack should be cleared when sharing the same VM
1330          */
1331         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1332                 p->sas_ss_sp = p->sas_ss_size = 0;
1333
1334         /*
1335          * Syscall tracing and stepping should be turned off in the
1336          * child regardless of CLONE_PTRACE.
1337          */
1338         user_disable_single_step(p);
1339         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1340 #ifdef TIF_SYSCALL_EMU
1341         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1342 #endif
1343         clear_all_latency_tracing(p);
1344
1345         /* ok, now we should be set up.. */
1346         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1347         p->pdeath_signal = 0;
1348         p->exit_state = 0;
1349
1350         p->nr_dirtied = 0;
1351         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1352         p->dirty_paused_when = 0;
1353
1354         /*
1355          * Ok, make it visible to the rest of the system.
1356          * We dont wake it up yet.
1357          */
1358         p->group_leader = p;
1359         INIT_LIST_HEAD(&p->thread_group);
1360
1361         /* Now that the task is set up, run cgroup callbacks if
1362          * necessary. We need to run them before the task is visible
1363          * on the tasklist. */
1364         cgroup_fork_callbacks(p);
1365         cgroup_callbacks_done = 1;
1366
1367         /* Need tasklist lock for parent etc handling! */
1368         write_lock_irq(&tasklist_lock);
1369
1370         /* CLONE_PARENT re-uses the old parent */
1371         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1372                 p->real_parent = current->real_parent;
1373                 p->parent_exec_id = current->parent_exec_id;
1374         } else {
1375                 p->real_parent = current;
1376                 p->parent_exec_id = current->self_exec_id;
1377         }
1378
1379         spin_lock(&current->sighand->siglock);
1380
1381         /*
1382          * Process group and session signals need to be delivered to just the
1383          * parent before the fork or both the parent and the child after the
1384          * fork. Restart if a signal comes in before we add the new process to
1385          * it's process group.
1386          * A fatal signal pending means that current will exit, so the new
1387          * thread can't slip out of an OOM kill (or normal SIGKILL).
1388         */
1389         recalc_sigpending();
1390         if (signal_pending(current)) {
1391                 spin_unlock(&current->sighand->siglock);
1392                 write_unlock_irq(&tasklist_lock);
1393                 retval = -ERESTARTNOINTR;
1394                 goto bad_fork_free_pid;
1395         }
1396
1397         if (clone_flags & CLONE_THREAD) {
1398                 current->signal->nr_threads++;
1399                 atomic_inc(&current->signal->live);
1400                 atomic_inc(&current->signal->sigcnt);
1401                 p->group_leader = current->group_leader;
1402                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1403         }
1404
1405         if (likely(p->pid)) {
1406                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1407
1408                 if (thread_group_leader(p)) {
1409                         if (is_child_reaper(pid))
1410                                 p->nsproxy->pid_ns->child_reaper = p;
1411
1412                         p->signal->leader_pid = pid;
1413                         p->signal->tty = tty_kref_get(current->signal->tty);
1414                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1415                         attach_pid(p, PIDTYPE_SID, task_session(current));
1416                         list_add_tail(&p->sibling, &p->real_parent->children);
1417                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1418                         __this_cpu_inc(process_counts);
1419                 }
1420                 attach_pid(p, PIDTYPE_PID, pid);
1421                 nr_threads++;
1422         }
1423
1424         total_forks++;
1425         spin_unlock(&current->sighand->siglock);
1426         write_unlock_irq(&tasklist_lock);
1427         proc_fork_connector(p);
1428         cgroup_post_fork(p);
1429         if (clone_flags & CLONE_THREAD)
1430                 threadgroup_change_end(current);
1431         perf_event_fork(p);
1432
1433         trace_task_newtask(p, clone_flags);
1434
1435         return p;
1436
1437 bad_fork_free_pid:
1438         if (pid != &init_struct_pid)
1439                 free_pid(pid);
1440 bad_fork_cleanup_io:
1441         if (p->io_context)
1442                 exit_io_context(p);
1443 bad_fork_cleanup_namespaces:
1444         if (unlikely(clone_flags & CLONE_NEWPID))
1445                 pid_ns_release_proc(p->nsproxy->pid_ns);
1446         exit_task_namespaces(p);
1447 bad_fork_cleanup_mm:
1448         if (p->mm)
1449                 mmput(p->mm);
1450 bad_fork_cleanup_signal:
1451         if (!(clone_flags & CLONE_THREAD))
1452                 free_signal_struct(p->signal);
1453 bad_fork_cleanup_sighand:
1454         __cleanup_sighand(p->sighand);
1455 bad_fork_cleanup_fs:
1456         exit_fs(p); /* blocking */
1457 bad_fork_cleanup_files:
1458         exit_files(p); /* blocking */
1459 bad_fork_cleanup_semundo:
1460         exit_sem(p);
1461 bad_fork_cleanup_audit:
1462         audit_free(p);
1463 bad_fork_cleanup_policy:
1464         perf_event_free_task(p);
1465 #ifdef CONFIG_NUMA
1466         mpol_put(p->mempolicy);
1467 bad_fork_cleanup_cgroup:
1468 #endif
1469         if (clone_flags & CLONE_THREAD)
1470                 threadgroup_change_end(current);
1471         cgroup_exit(p, cgroup_callbacks_done);
1472         delayacct_tsk_free(p);
1473         module_put(task_thread_info(p)->exec_domain->module);
1474 bad_fork_cleanup_count:
1475         atomic_dec(&p->cred->user->processes);
1476         exit_creds(p);
1477 bad_fork_free:
1478         free_task(p);
1479 fork_out:
1480         return ERR_PTR(retval);
1481 }
1482
1483 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1484 {
1485         memset(regs, 0, sizeof(struct pt_regs));
1486         return regs;
1487 }
1488
1489 static inline void init_idle_pids(struct pid_link *links)
1490 {
1491         enum pid_type type;
1492
1493         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1494                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1495                 links[type].pid = &init_struct_pid;
1496         }
1497 }
1498
1499 struct task_struct * __cpuinit fork_idle(int cpu)
1500 {
1501         struct task_struct *task;
1502         struct pt_regs regs;
1503
1504         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1505                             &init_struct_pid, 0);
1506         if (!IS_ERR(task)) {
1507                 init_idle_pids(task->pids);
1508                 init_idle(task, cpu);
1509         }
1510
1511         return task;
1512 }
1513
1514 /*
1515  *  Ok, this is the main fork-routine.
1516  *
1517  * It copies the process, and if successful kick-starts
1518  * it and waits for it to finish using the VM if required.
1519  */
1520 long do_fork(unsigned long clone_flags,
1521               unsigned long stack_start,
1522               struct pt_regs *regs,
1523               unsigned long stack_size,
1524               int __user *parent_tidptr,
1525               int __user *child_tidptr)
1526 {
1527         struct task_struct *p;
1528         int trace = 0;
1529         long nr;
1530
1531         /*
1532          * Do some preliminary argument and permissions checking before we
1533          * actually start allocating stuff
1534          */
1535         if (clone_flags & CLONE_NEWUSER) {
1536                 if (clone_flags & CLONE_THREAD)
1537                         return -EINVAL;
1538                 /* hopefully this check will go away when userns support is
1539                  * complete
1540                  */
1541                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1542                                 !capable(CAP_SETGID))
1543                         return -EPERM;
1544         }
1545
1546         /*
1547          * Determine whether and which event to report to ptracer.  When
1548          * called from kernel_thread or CLONE_UNTRACED is explicitly
1549          * requested, no event is reported; otherwise, report if the event
1550          * for the type of forking is enabled.
1551          */
1552         if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1553                 if (clone_flags & CLONE_VFORK)
1554                         trace = PTRACE_EVENT_VFORK;
1555                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1556                         trace = PTRACE_EVENT_CLONE;
1557                 else
1558                         trace = PTRACE_EVENT_FORK;
1559
1560                 if (likely(!ptrace_event_enabled(current, trace)))
1561                         trace = 0;
1562         }
1563
1564         p = copy_process(clone_flags, stack_start, regs, stack_size,
1565                          child_tidptr, NULL, trace);
1566         /*
1567          * Do this prior waking up the new thread - the thread pointer
1568          * might get invalid after that point, if the thread exits quickly.
1569          */
1570         if (!IS_ERR(p)) {
1571                 struct completion vfork;
1572
1573                 trace_sched_process_fork(current, p);
1574
1575                 nr = task_pid_vnr(p);
1576
1577                 if (clone_flags & CLONE_PARENT_SETTID)
1578                         put_user(nr, parent_tidptr);
1579
1580                 if (clone_flags & CLONE_VFORK) {
1581                         p->vfork_done = &vfork;
1582                         init_completion(&vfork);
1583                         get_task_struct(p);
1584                 }
1585
1586                 wake_up_new_task(p);
1587
1588                 /* forking complete and child started to run, tell ptracer */
1589                 if (unlikely(trace))
1590                         ptrace_event(trace, nr);
1591
1592                 if (clone_flags & CLONE_VFORK) {
1593                         if (!wait_for_vfork_done(p, &vfork))
1594                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1595                 }
1596         } else {
1597                 nr = PTR_ERR(p);
1598         }
1599         return nr;
1600 }
1601
1602 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1603 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1604 #endif
1605
1606 static void sighand_ctor(void *data)
1607 {
1608         struct sighand_struct *sighand = data;
1609
1610         spin_lock_init(&sighand->siglock);
1611         init_waitqueue_head(&sighand->signalfd_wqh);
1612 }
1613
1614 void __init proc_caches_init(void)
1615 {
1616         sighand_cachep = kmem_cache_create("sighand_cache",
1617                         sizeof(struct sighand_struct), 0,
1618                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1619                         SLAB_NOTRACK, sighand_ctor);
1620         signal_cachep = kmem_cache_create("signal_cache",
1621                         sizeof(struct signal_struct), 0,
1622                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1623         files_cachep = kmem_cache_create("files_cache",
1624                         sizeof(struct files_struct), 0,
1625                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1626         fs_cachep = kmem_cache_create("fs_cache",
1627                         sizeof(struct fs_struct), 0,
1628                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1629         /*
1630          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1631          * whole struct cpumask for the OFFSTACK case. We could change
1632          * this to *only* allocate as much of it as required by the
1633          * maximum number of CPU's we can ever have.  The cpumask_allocation
1634          * is at the end of the structure, exactly for that reason.
1635          */
1636         mm_cachep = kmem_cache_create("mm_struct",
1637                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1638                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1639         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1640         mmap_init();
1641         nsproxy_cache_init();
1642 }
1643
1644 /*
1645  * Check constraints on flags passed to the unshare system call.
1646  */
1647 static int check_unshare_flags(unsigned long unshare_flags)
1648 {
1649         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1650                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1651                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1652                 return -EINVAL;
1653         /*
1654          * Not implemented, but pretend it works if there is nothing to
1655          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1656          * needs to unshare vm.
1657          */
1658         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1659                 /* FIXME: get_task_mm() increments ->mm_users */
1660                 if (atomic_read(&current->mm->mm_users) > 1)
1661                         return -EINVAL;
1662         }
1663
1664         return 0;
1665 }
1666
1667 /*
1668  * Unshare the filesystem structure if it is being shared
1669  */
1670 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1671 {
1672         struct fs_struct *fs = current->fs;
1673
1674         if (!(unshare_flags & CLONE_FS) || !fs)
1675                 return 0;
1676
1677         /* don't need lock here; in the worst case we'll do useless copy */
1678         if (fs->users == 1)
1679                 return 0;
1680
1681         *new_fsp = copy_fs_struct(fs);
1682         if (!*new_fsp)
1683                 return -ENOMEM;
1684
1685         return 0;
1686 }
1687
1688 /*
1689  * Unshare file descriptor table if it is being shared
1690  */
1691 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1692 {
1693         struct files_struct *fd = current->files;
1694         int error = 0;
1695
1696         if ((unshare_flags & CLONE_FILES) &&
1697             (fd && atomic_read(&fd->count) > 1)) {
1698                 *new_fdp = dup_fd(fd, &error);
1699                 if (!*new_fdp)
1700                         return error;
1701         }
1702
1703         return 0;
1704 }
1705
1706 /*
1707  * unshare allows a process to 'unshare' part of the process
1708  * context which was originally shared using clone.  copy_*
1709  * functions used by do_fork() cannot be used here directly
1710  * because they modify an inactive task_struct that is being
1711  * constructed. Here we are modifying the current, active,
1712  * task_struct.
1713  */
1714 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1715 {
1716         struct fs_struct *fs, *new_fs = NULL;
1717         struct files_struct *fd, *new_fd = NULL;
1718         struct nsproxy *new_nsproxy = NULL;
1719         int do_sysvsem = 0;
1720         int err;
1721
1722         err = check_unshare_flags(unshare_flags);
1723         if (err)
1724                 goto bad_unshare_out;
1725
1726         /*
1727          * If unsharing namespace, must also unshare filesystem information.
1728          */
1729         if (unshare_flags & CLONE_NEWNS)
1730                 unshare_flags |= CLONE_FS;
1731         /*
1732          * CLONE_NEWIPC must also detach from the undolist: after switching
1733          * to a new ipc namespace, the semaphore arrays from the old
1734          * namespace are unreachable.
1735          */
1736         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1737                 do_sysvsem = 1;
1738         err = unshare_fs(unshare_flags, &new_fs);
1739         if (err)
1740                 goto bad_unshare_out;
1741         err = unshare_fd(unshare_flags, &new_fd);
1742         if (err)
1743                 goto bad_unshare_cleanup_fs;
1744         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1745         if (err)
1746                 goto bad_unshare_cleanup_fd;
1747
1748         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1749                 if (do_sysvsem) {
1750                         /*
1751                          * CLONE_SYSVSEM is equivalent to sys_exit().
1752                          */
1753                         exit_sem(current);
1754                 }
1755
1756                 if (new_nsproxy) {
1757                         switch_task_namespaces(current, new_nsproxy);
1758                         new_nsproxy = NULL;
1759                 }
1760
1761                 task_lock(current);
1762
1763                 if (new_fs) {
1764                         fs = current->fs;
1765                         spin_lock(&fs->lock);
1766                         current->fs = new_fs;
1767                         if (--fs->users)
1768                                 new_fs = NULL;
1769                         else
1770                                 new_fs = fs;
1771                         spin_unlock(&fs->lock);
1772                 }
1773
1774                 if (new_fd) {
1775                         fd = current->files;
1776                         current->files = new_fd;
1777                         new_fd = fd;
1778                 }
1779
1780                 task_unlock(current);
1781         }
1782
1783         if (new_nsproxy)
1784                 put_nsproxy(new_nsproxy);
1785
1786 bad_unshare_cleanup_fd:
1787         if (new_fd)
1788                 put_files_struct(new_fd);
1789
1790 bad_unshare_cleanup_fs:
1791         if (new_fs)
1792                 free_fs_struct(new_fs);
1793
1794 bad_unshare_out:
1795         return err;
1796 }
1797
1798 /*
1799  *      Helper to unshare the files of the current task.
1800  *      We don't want to expose copy_files internals to
1801  *      the exec layer of the kernel.
1802  */
1803
1804 int unshare_files(struct files_struct **displaced)
1805 {
1806         struct task_struct *task = current;
1807         struct files_struct *copy = NULL;
1808         int error;
1809
1810         error = unshare_fd(CLONE_FILES, &copy);
1811         if (error || !copy) {
1812                 *displaced = NULL;
1813                 return error;
1814         }
1815         *displaced = task->files;
1816         task_lock(task);
1817         task->files = copy;
1818         task_unlock(task);
1819         return 0;
1820 }