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