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                         mutex_lock(&mapping->i_mmap_mutex);
 387                         if (tmp->vm_flags & VM_SHARED)
 388                                 mapping->i_mmap_writable++;
 389                         flush_dcache_mmap_lock(mapping);
 390                         /* insert tmp into the share list, just after mpnt */
 391                         vma_prio_tree_add(tmp, mpnt);
 392                         flush_dcache_mmap_unlock(mapping);
 393                         mutex_unlock(&mapping->i_mmap_mutex);
 394                 }
 395
 396                 /*
 397                  * Clear hugetlb-related page reserves for children. This only
 398                  * affects MAP_PRIVATE mappings. Faults generated by the child
 399                  * are not guaranteed to succeed, even if read-only
 400                  */
 401                 if (is_vm_hugetlb_page(tmp))
 402                         reset_vma_resv_huge_pages(tmp);
 403
 404                 /*
 405                  * Link in the new vma and copy the page table entries.
 406                  */
 407                 *pprev = tmp;
 408                 pprev = &tmp->vm_next;
 409                 tmp->vm_prev = prev;
 410                 prev = tmp;
 411
 412                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 413                 rb_link = &tmp->vm_rb.rb_right;
 414                 rb_parent = &tmp->vm_rb;
 415
 416                 mm->map_count++;
 417                 retval = copy_page_range(mm, oldmm, mpnt);
 418
 419                 if (tmp->vm_ops && tmp->vm_ops->open)
 420                         tmp->vm_ops->open(tmp);
 421
 422                 if (retval)
 423                         goto out;
 424         }
 425         /* a new mm has just been created */
 426         arch_dup_mmap(oldmm, mm);
 427         retval = 0;
 428 out:
 429         up_write(&mm->mmap_sem);
 430         flush_tlb_mm(oldmm);
 431         up_write(&oldmm->mmap_sem);
 432         return retval;
 433 fail_nomem_anon_vma_fork:
 434         mpol_put(pol);
 435 fail_nomem_policy:
 436         kmem_cache_free(vm_area_cachep, tmp);
 437 fail_nomem:
 438         retval = -ENOMEM;
 439         vm_unacct_memory(charge);
 440         goto out;
 441 }
 442
 443 static inline int mm_alloc_pgd(struct mm_struct * mm)
 444 {
 445         mm->pgd = pgd_alloc(mm);
 446         if (unlikely(!mm->pgd))
 447                 return -ENOMEM;
 448         return 0;
 449 }
 450
 451 static inline void mm_free_pgd(struct mm_struct * mm)
 452 {
 453         pgd_free(mm, mm->pgd);
 454 }
 455 #else
 456 #define dup_mmap(mm, oldmm)     (0)
 457 #define mm_alloc_pgd(mm)        (0)
 458 #define mm_free_pgd(mm)
 459 #endif /* CONFIG_MMU */
 460
 461 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 462
 463 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 464 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 465
 466 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 467
 468 static int __init coredump_filter_setup(char *s)
 469 {
 470         default_dump_filter =
 471                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 472                 MMF_DUMP_FILTER_MASK;
 473         return 1;
 474 }
 475
 476 __setup("coredump_filter=", coredump_filter_setup);
 477
 478 #include <linux/init_task.h>
 479
 480 static void mm_init_aio(struct mm_struct *mm)
 481 {
 482 #ifdef CONFIG_AIO
 483         spin_lock_init(&mm->ioctx_lock);
 484         INIT_HLIST_HEAD(&mm->ioctx_list);
 485 #endif
 486 }
 487
 488 int mm_init_cpumask(struct mm_struct *mm, struct mm_struct *oldmm)
 489 {
 490 #ifdef CONFIG_CPUMASK_OFFSTACK
 491         if (!alloc_cpumask_var(&mm->cpu_vm_mask_var, GFP_KERNEL))
 492                 return -ENOMEM;
 493
 494         if (oldmm)
 495                 cpumask_copy(mm_cpumask(mm), mm_cpumask(oldmm));
 496         else
 497                 memset(mm_cpumask(mm), 0, cpumask_size());
 498 #endif
 499         return 0;
 500 }
 501
 502 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
 503 {
 504         atomic_set(&mm->mm_users, 1);
 505         atomic_set(&mm->mm_count, 1);
 506         init_rwsem(&mm->mmap_sem);
 507         INIT_LIST_HEAD(&mm->mmlist);
 508         mm->flags = (current->mm) ?
 509                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 510         mm->core_state = NULL;
 511         mm->nr_ptes = 0;
 512         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 513         spin_lock_init(&mm->page_table_lock);
 514         mm->free_area_cache = TASK_UNMAPPED_BASE;
 515         mm->cached_hole_size = ~0UL;
 516         mm_init_aio(mm);
 517         mm_init_owner(mm, p);
 518         atomic_set(&mm->oom_disable_count, 0);
 519
 520         if (likely(!mm_alloc_pgd(mm))) {
 521                 mm->def_flags = 0;
 522                 mmu_notifier_mm_init(mm);
 523                 return mm;
 524         }
 525
 526         free_mm(mm);
 527         return NULL;
 528 }
 529
 530 /*
 531  * Allocate and initialize an mm_struct.
 532  */
 533 struct mm_struct * mm_alloc(void)
 534 {
 535         struct mm_struct * mm;
 536
 537         mm = allocate_mm();
 538         if (!mm)
 539                 return NULL;
 540
 541         memset(mm, 0, sizeof(*mm));
 542         mm = mm_init(mm, current);
 543         if (!mm)
 544                 return NULL;
 545
 546         if (mm_init_cpumask(mm, NULL)) {
 547                 mm_free_pgd(mm);
 548                 free_mm(mm);
 549                 return NULL;
 550         }
 551
 552         return mm;
 553 }
 554
 555 /*
 556  * Called when the last reference to the mm
 557  * is dropped: either by a lazy thread or by
 558  * mmput. Free the page directory and the mm.
 559  */
 560 void __mmdrop(struct mm_struct *mm)
 561 {
 562         BUG_ON(mm == &init_mm);
 563         free_cpumask_var(mm->cpu_vm_mask_var);
 564         mm_free_pgd(mm);
 565         destroy_context(mm);
 566         mmu_notifier_mm_destroy(mm);
 567 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 568         VM_BUG_ON(mm->pmd_huge_pte);
 569 #endif
 570         free_mm(mm);
 571 }
 572 EXPORT_SYMBOL_GPL(__mmdrop);
 573
 574 /*
 575  * Decrement the use count and release all resources for an mm.
 576  */
 577 void mmput(struct mm_struct *mm)
 578 {
 579         might_sleep();
 580
 581         if (atomic_dec_and_test(&mm->mm_users)) {
 582                 exit_aio(mm);
 583                 ksm_exit(mm);
 584                 khugepaged_exit(mm); /* must run before exit_mmap */
 585                 exit_mmap(mm);
 586                 set_mm_exe_file(mm, NULL);
 587                 if (!list_empty(&mm->mmlist)) {
 588                         spin_lock(&mmlist_lock);
 589                         list_del(&mm->mmlist);
 590                         spin_unlock(&mmlist_lock);
 591                 }
 592                 put_swap_token(mm);
 593                 if (mm->binfmt)
 594                         module_put(mm->binfmt->module);
 595                 mmdrop(mm);
 596         }
 597 }
 598 EXPORT_SYMBOL_GPL(mmput);
 599
 600 /**
 601  * get_task_mm - acquire a reference to the task's mm
 602  *
 603  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 604  * this kernel workthread has transiently adopted a user mm with use_mm,
 605  * to do its AIO) is not set and if so returns a reference to it, after
 606  * bumping up the use count.  User must release the mm via mmput()
 607  * after use.  Typically used by /proc and ptrace.
 608  */
 609 struct mm_struct *get_task_mm(struct task_struct *task)
 610 {
 611         struct mm_struct *mm;
 612
 613         task_lock(task);
 614         mm = task->mm;
 615         if (mm) {
 616                 if (task->flags & PF_KTHREAD)
 617                         mm = NULL;
 618                 else
 619                         atomic_inc(&mm->mm_users);
 620         }
 621         task_unlock(task);
 622         return mm;
 623 }
 624 EXPORT_SYMBOL_GPL(get_task_mm);
 625
 626 /* Please note the differences between mmput and mm_release.
 627  * mmput is called whenever we stop holding onto a mm_struct,
 628  * error success whatever.
 629  *
 630  * mm_release is called after a mm_struct has been removed
 631  * from the current process.
 632  *
 633  * This difference is important for error handling, when we
 634  * only half set up a mm_struct for a new process and need to restore
 635  * the old one.  Because we mmput the new mm_struct before
 636  * restoring the old one. . .
 637  * Eric Biederman 10 January 1998
 638  */
 639 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 640 {
 641         struct completion *vfork_done = tsk->vfork_done;
 642
 643         /* Get rid of any futexes when releasing the mm */
 644 #ifdef CONFIG_FUTEX
 645         if (unlikely(tsk->robust_list)) {
 646                 exit_robust_list(tsk);
 647                 tsk->robust_list = NULL;
 648         }
 649 #ifdef CONFIG_COMPAT
 650         if (unlikely(tsk->compat_robust_list)) {
 651                 compat_exit_robust_list(tsk);
 652                 tsk->compat_robust_list = NULL;
 653         }
 654 #endif
 655         if (unlikely(!list_empty(&tsk->pi_state_list)))
 656                 exit_pi_state_list(tsk);
 657 #endif
 658
 659         /* Get rid of any cached register state */
 660         deactivate_mm(tsk, mm);
 661
 662         /* notify parent sleeping on vfork() */
 663         if (vfork_done) {
 664                 tsk->vfork_done = NULL;
 665                 complete(vfork_done);
 666         }
 667
 668         /*
 669          * If we're exiting normally, clear a user-space tid field if
 670          * requested.  We leave this alone when dying by signal, to leave
 671          * the value intact in a core dump, and to save the unnecessary
 672          * trouble otherwise.  Userland only wants this done for a sys_exit.
 673          */
 674         if (tsk->clear_child_tid) {
 675                 if (!(tsk->flags & PF_SIGNALED) &&
 676                     atomic_read(&mm->mm_users) > 1) {
 677                         /*
 678                          * We don't check the error code - if userspace has
 679                          * not set up a proper pointer then tough luck.
 680                          */
 681                         put_user(0, tsk->clear_child_tid);
 682                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 683                                         1, NULL, NULL, 0);
 684                 }
 685                 tsk->clear_child_tid = NULL;
 686         }
 687 }
 688
 689 /*
 690  * Allocate a new mm structure and copy contents from the
 691  * mm structure of the passed in task structure.
 692  */
 693 struct mm_struct *dup_mm(struct task_struct *tsk)
 694 {
 695         struct mm_struct *mm, *oldmm = current->mm;
 696         int err;
 697
 698         if (!oldmm)
 699                 return NULL;
 700
 701         mm = allocate_mm();
 702         if (!mm)
 703                 goto fail_nomem;
 704
 705         memcpy(mm, oldmm, sizeof(*mm));
 706
 707         /* Initializing for Swap token stuff */
 708         mm->token_priority = 0;
 709         mm->last_interval = 0;
 710
 711 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 712         mm->pmd_huge_pte = NULL;
 713 #endif
 714
 715         if (!mm_init(mm, tsk))
 716                 goto fail_nomem;
 717
 718         if (mm_init_cpumask(mm, oldmm))
 719                 goto fail_nocpumask;
 720
 721         if (init_new_context(tsk, mm))
 722                 goto fail_nocontext;
 723
 724         dup_mm_exe_file(oldmm, mm);
 725
 726         err = dup_mmap(mm, oldmm);
 727         if (err)
 728                 goto free_pt;
 729
 730         mm->hiwater_rss = get_mm_rss(mm);
 731         mm->hiwater_vm = mm->total_vm;
 732
 733         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 734                 goto free_pt;
 735
 736         return mm;
 737
 738 free_pt:
 739         /* don't put binfmt in mmput, we haven't got module yet */
 740         mm->binfmt = NULL;
 741         mmput(mm);
 742
 743 fail_nomem:
 744         return NULL;
 745
 746 fail_nocontext:
 747         free_cpumask_var(mm->cpu_vm_mask_var);
 748
 749 fail_nocpumask:
 750         /*
 751          * If init_new_context() failed, we cannot use mmput() to free the mm
 752          * because it calls destroy_context()
 753          */
 754         mm_free_pgd(mm);
 755         free_mm(mm);
 756         return NULL;
 757 }
 758
 759 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 760 {
 761         struct mm_struct * mm, *oldmm;
 762         int retval;
 763
 764         tsk->min_flt = tsk->maj_flt = 0;
 765         tsk->nvcsw = tsk->nivcsw = 0;
 766 #ifdef CONFIG_DETECT_HUNG_TASK
 767         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 768 #endif
 769
 770         tsk->mm = NULL;
 771         tsk->active_mm = NULL;
 772
 773         /*
 774          * Are we cloning a kernel thread?
 775          *
 776          * We need to steal a active VM for that..
 777          */
 778         oldmm = current->mm;
 779         if (!oldmm)
 780                 return 0;
 781
 782         if (clone_flags & CLONE_VM) {
 783                 atomic_inc(&oldmm->mm_users);
 784                 mm = oldmm;
 785                 goto good_mm;
 786         }
 787
 788         retval = -ENOMEM;
 789         mm = dup_mm(tsk);
 790         if (!mm)
 791                 goto fail_nomem;
 792
 793 good_mm:
 794         /* Initializing for Swap token stuff */
 795         mm->token_priority = 0;
 796         mm->last_interval = 0;
 797         if (tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
 798                 atomic_inc(&mm->oom_disable_count);
 799
 800         tsk->mm = mm;
 801         tsk->active_mm = mm;
 802         return 0;
 803
 804 fail_nomem:
 805         return retval;
 806 }
 807
 808 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 809 {
 810         struct fs_struct *fs = current->fs;
 811         if (clone_flags & CLONE_FS) {
 812                 /* tsk->fs is already what we want */
 813                 spin_lock(&fs->lock);
 814                 if (fs->in_exec) {
 815                         spin_unlock(&fs->lock);
 816                         return -EAGAIN;
 817                 }
 818                 fs->users++;
 819                 spin_unlock(&fs->lock);
 820                 return 0;
 821         }
 822         tsk->fs = copy_fs_struct(fs);
 823         if (!tsk->fs)
 824                 return -ENOMEM;
 825         return 0;
 826 }
 827
 828 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 829 {
 830         struct files_struct *oldf, *newf;
 831         int error = 0;
 832
 833         /*
 834          * A background process may not have any files ...
 835          */
 836         oldf = current->files;
 837         if (!oldf)
 838                 goto out;
 839
 840         if (clone_flags & CLONE_FILES) {
 841                 atomic_inc(&oldf->count);
 842                 goto out;
 843         }
 844
 845         newf = dup_fd(oldf, &error);
 846         if (!newf)
 847                 goto out;
 848
 849         tsk->files = newf;
 850         error = 0;
 851 out:
 852         return error;
 853 }
 854
 855 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 856 {
 857 #ifdef CONFIG_BLOCK
 858         struct io_context *ioc = current->io_context;
 859
 860         if (!ioc)
 861                 return 0;
 862         /*
 863          * Share io context with parent, if CLONE_IO is set
 864          */
 865         if (clone_flags & CLONE_IO) {
 866                 tsk->io_context = ioc_task_link(ioc);
 867                 if (unlikely(!tsk->io_context))
 868                         return -ENOMEM;
 869         } else if (ioprio_valid(ioc->ioprio)) {
 870                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 871                 if (unlikely(!tsk->io_context))
 872                         return -ENOMEM;
 873
 874                 tsk->io_context->ioprio = ioc->ioprio;
 875         }
 876 #endif
 877         return 0;
 878 }
 879
 880 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 881 {
 882         struct sighand_struct *sig;
 883
 884         if (clone_flags & CLONE_SIGHAND) {
 885                 atomic_inc(&current->sighand->count);
 886                 return 0;
 887         }
 888         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 889         rcu_assign_pointer(tsk->sighand, sig);
 890         if (!sig)
 891                 return -ENOMEM;
 892         atomic_set(&sig->count, 1);
 893         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 894         return 0;
 895 }
 896
 897 void __cleanup_sighand(struct sighand_struct *sighand)
 898 {
 899         if (atomic_dec_and_test(&sighand->count))
 900                 kmem_cache_free(sighand_cachep, sighand);
 901 }
 902
 903
 904 /*
 905  * Initialize POSIX timer handling for a thread group.
 906  */
 907 static void posix_cpu_timers_init_group(struct signal_struct *sig)
 908 {
 909         unsigned long cpu_limit;
 910
 911         /* Thread group counters. */
 912         thread_group_cputime_init(sig);
 913
 914         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
 915         if (cpu_limit != RLIM_INFINITY) {
 916                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
 917                 sig->cputimer.running = 1;
 918         }
 919
 920         /* The timer lists. */
 921         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 922         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 923         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 924 }
 925
 926 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 927 {
 928         struct signal_struct *sig;
 929
 930         if (clone_flags & CLONE_THREAD)
 931                 return 0;
 932
 933         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
 934         tsk->signal = sig;
 935         if (!sig)
 936                 return -ENOMEM;
 937
 938         sig->nr_threads = 1;
 939         atomic_set(&sig->live, 1);
 940         atomic_set(&sig->sigcnt, 1);
 941         init_waitqueue_head(&sig->wait_chldexit);
 942         if (clone_flags & CLONE_NEWPID)
 943                 sig->flags |= SIGNAL_UNKILLABLE;
 944         sig->curr_target = tsk;
 945         init_sigpending(&sig->shared_pending);
 946         INIT_LIST_HEAD(&sig->posix_timers);
 947
 948         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 949         sig->real_timer.function = it_real_fn;
 950
 951         task_lock(current->group_leader);
 952         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 953         task_unlock(current->group_leader);
 954
 955         posix_cpu_timers_init_group(sig);
 956
 957         tty_audit_fork(sig);
 958         sched_autogroup_fork(sig);
 959
 960 #ifdef CONFIG_CGROUPS
 961         init_rwsem(&sig->threadgroup_fork_lock);
 962 #endif
 963
 964         sig->oom_adj = current->signal->oom_adj;
 965         sig->oom_score_adj = current->signal->oom_score_adj;
 966         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
 967
 968         mutex_init(&sig->cred_guard_mutex);
 969
 970         return 0;
 971 }
 972
 973 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 974 {
 975         unsigned long new_flags = p->flags;
 976
 977         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
 978         new_flags |= PF_FORKNOEXEC;
 979         new_flags |= PF_STARTING;
 980         p->flags = new_flags;
 981         clear_freeze_flag(p);
 982 }
 983
 984 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
 985 {
 986         current->clear_child_tid = tidptr;
 987
 988         return task_pid_vnr(current);
 989 }
 990
 991 static void rt_mutex_init_task(struct task_struct *p)
 992 {
 993         raw_spin_lock_init(&p->pi_lock);
 994 #ifdef CONFIG_RT_MUTEXES
 995         plist_head_init_raw(&p->pi_waiters, &p->pi_lock);
 996         p->pi_blocked_on = NULL;
 997 #endif
 998 }
 999
1000 #ifdef CONFIG_MM_OWNER
1001 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1002 {
1003         mm->owner = p;
1004 }
1005 #endif /* CONFIG_MM_OWNER */
1006
1007 /*
1008  * Initialize POSIX timer handling for a single task.
1009  */
1010 static void posix_cpu_timers_init(struct task_struct *tsk)
1011 {
1012         tsk->cputime_expires.prof_exp = cputime_zero;
1013         tsk->cputime_expires.virt_exp = cputime_zero;
1014         tsk->cputime_expires.sched_exp = 0;
1015         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1016         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1017         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1018 }
1019
1020 /*
1021  * This creates a new process as a copy of the old one,
1022  * but does not actually start it yet.
1023  *
1024  * It copies the registers, and all the appropriate
1025  * parts of the process environment (as per the clone
1026  * flags). The actual kick-off is left to the caller.
1027  */
1028 static struct task_struct *copy_process(unsigned long clone_flags,
1029                                         unsigned long stack_start,
1030                                         struct pt_regs *regs,
1031                                         unsigned long stack_size,
1032                                         int __user *child_tidptr,
1033                                         struct pid *pid,
1034                                         int trace)
1035 {
1036         int retval;
1037         struct task_struct *p;
1038         int cgroup_callbacks_done = 0;
1039
1040         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1041                 return ERR_PTR(-EINVAL);
1042
1043         /*
1044          * Thread groups must share signals as well, and detached threads
1045          * can only be started up within the thread group.
1046          */
1047         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1048                 return ERR_PTR(-EINVAL);
1049
1050         /*
1051          * Shared signal handlers imply shared VM. By way of the above,
1052          * thread groups also imply shared VM. Blocking this case allows
1053          * for various simplifications in other code.
1054          */
1055         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1056                 return ERR_PTR(-EINVAL);
1057
1058         /*
1059          * Siblings of global init remain as zombies on exit since they are
1060          * not reaped by their parent (swapper). To solve this and to avoid
1061          * multi-rooted process trees, prevent global and container-inits
1062          * from creating siblings.
1063          */
1064         if ((clone_flags & CLONE_PARENT) &&
1065                                 current->signal->flags & SIGNAL_UNKILLABLE)
1066                 return ERR_PTR(-EINVAL);
1067
1068         retval = security_task_create(clone_flags);
1069         if (retval)
1070                 goto fork_out;
1071
1072         retval = -ENOMEM;
1073         p = dup_task_struct(current);
1074         if (!p)
1075                 goto fork_out;
1076
1077         ftrace_graph_init_task(p);
1078
1079         rt_mutex_init_task(p);
1080
1081 #ifdef CONFIG_PROVE_LOCKING
1082         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1083         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1084 #endif
1085         retval = -EAGAIN;
1086         if (atomic_read(&p->real_cred->user->processes) >=
1087                         task_rlimit(p, RLIMIT_NPROC)) {
1088                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1089                     p->real_cred->user != INIT_USER)
1090                         goto bad_fork_free;
1091         }
1092
1093         retval = copy_creds(p, clone_flags);
1094         if (retval < 0)
1095                 goto bad_fork_free;
1096
1097         /*
1098          * If multiple threads are within copy_process(), then this check
1099          * triggers too late. This doesn't hurt, the check is only there
1100          * to stop root fork bombs.
1101          */
1102         retval = -EAGAIN;
1103         if (nr_threads >= max_threads)
1104                 goto bad_fork_cleanup_count;
1105
1106         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1107                 goto bad_fork_cleanup_count;
1108
1109         p->did_exec = 0;
1110         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1111         copy_flags(clone_flags, p);
1112         INIT_LIST_HEAD(&p->children);
1113         INIT_LIST_HEAD(&p->sibling);
1114         rcu_copy_process(p);
1115         p->vfork_done = NULL;
1116         spin_lock_init(&p->alloc_lock);
1117
1118         init_sigpending(&p->pending);
1119
1120         p->utime = cputime_zero;
1121         p->stime = cputime_zero;
1122         p->gtime = cputime_zero;
1123         p->utimescaled = cputime_zero;
1124         p->stimescaled = cputime_zero;
1125 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1126         p->prev_utime = cputime_zero;
1127         p->prev_stime = cputime_zero;
1128 #endif
1129 #if defined(SPLIT_RSS_COUNTING)
1130         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1131 #endif
1132
1133         p->default_timer_slack_ns = current->timer_slack_ns;
1134
1135         task_io_accounting_init(&p->ioac);
1136         acct_clear_integrals(p);
1137
1138         posix_cpu_timers_init(p);
1139
1140         do_posix_clock_monotonic_gettime(&p->start_time);
1141         p->real_start_time = p->start_time;
1142         monotonic_to_bootbased(&p->real_start_time);
1143         p->io_context = NULL;
1144         p->audit_context = NULL;
1145         if (clone_flags & CLONE_THREAD)
1146                 threadgroup_fork_read_lock(current);
1147         cgroup_fork(p);
1148 #ifdef CONFIG_NUMA
1149         p->mempolicy = mpol_dup(p->mempolicy);
1150         if (IS_ERR(p->mempolicy)) {
1151                 retval = PTR_ERR(p->mempolicy);
1152                 p->mempolicy = NULL;
1153                 goto bad_fork_cleanup_cgroup;
1154         }
1155         mpol_fix_fork_child_flag(p);
1156 #endif
1157 #ifdef CONFIG_TRACE_IRQFLAGS
1158         p->irq_events = 0;
1159 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1160         p->hardirqs_enabled = 1;
1161 #else
1162         p->hardirqs_enabled = 0;
1163 #endif
1164         p->hardirq_enable_ip = 0;
1165         p->hardirq_enable_event = 0;
1166         p->hardirq_disable_ip = _THIS_IP_;
1167         p->hardirq_disable_event = 0;
1168         p->softirqs_enabled = 1;
1169         p->softirq_enable_ip = _THIS_IP_;
1170         p->softirq_enable_event = 0;
1171         p->softirq_disable_ip = 0;
1172         p->softirq_disable_event = 0;
1173         p->hardirq_context = 0;
1174         p->softirq_context = 0;
1175 #endif
1176 #ifdef CONFIG_LOCKDEP
1177         p->lockdep_depth = 0; /* no locks held yet */
1178         p->curr_chain_key = 0;
1179         p->lockdep_recursion = 0;
1180 #endif
1181
1182 #ifdef CONFIG_DEBUG_MUTEXES
1183         p->blocked_on = NULL; /* not blocked yet */
1184 #endif
1185 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1186         p->memcg_batch.do_batch = 0;
1187         p->memcg_batch.memcg = NULL;
1188 #endif
1189
1190         /* Perform scheduler related setup. Assign this task to a CPU. */
1191         sched_fork(p);
1192
1193         retval = perf_event_init_task(p);
1194         if (retval)
1195                 goto bad_fork_cleanup_policy;
1196
1197         if ((retval = audit_alloc(p)))
1198                 goto bad_fork_cleanup_policy;
1199         /* copy all the process information */
1200         if ((retval = copy_semundo(clone_flags, p)))
1201                 goto bad_fork_cleanup_audit;
1202         if ((retval = copy_files(clone_flags, p)))
1203                 goto bad_fork_cleanup_semundo;
1204         if ((retval = copy_fs(clone_flags, p)))
1205                 goto bad_fork_cleanup_files;
1206         if ((retval = copy_sighand(clone_flags, p)))
1207                 goto bad_fork_cleanup_fs;
1208         if ((retval = copy_signal(clone_flags, p)))
1209                 goto bad_fork_cleanup_sighand;
1210         if ((retval = copy_mm(clone_flags, p)))
1211                 goto bad_fork_cleanup_signal;
1212         if ((retval = copy_namespaces(clone_flags, p)))
1213                 goto bad_fork_cleanup_mm;
1214         if ((retval = copy_io(clone_flags, p)))
1215                 goto bad_fork_cleanup_namespaces;
1216         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1217         if (retval)
1218                 goto bad_fork_cleanup_io;
1219
1220         if (pid != &init_struct_pid) {
1221                 retval = -ENOMEM;
1222                 pid = alloc_pid(p->nsproxy->pid_ns);
1223                 if (!pid)
1224                         goto bad_fork_cleanup_io;
1225         }
1226
1227         p->pid = pid_nr(pid);
1228         p->tgid = p->pid;
1229         if (clone_flags & CLONE_THREAD)
1230                 p->tgid = current->tgid;
1231
1232         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1233         /*
1234          * Clear TID on mm_release()?
1235          */
1236         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1237 #ifdef CONFIG_BLOCK
1238         p->plug = NULL;
1239 #endif
1240 #ifdef CONFIG_FUTEX
1241         p->robust_list = NULL;
1242 #ifdef CONFIG_COMPAT
1243         p->compat_robust_list = NULL;
1244 #endif
1245         INIT_LIST_HEAD(&p->pi_state_list);
1246         p->pi_state_cache = NULL;
1247 #endif
1248         /*
1249          * sigaltstack should be cleared when sharing the same VM
1250          */
1251         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1252                 p->sas_ss_sp = p->sas_ss_size = 0;
1253
1254         /*
1255          * Syscall tracing and stepping should be turned off in the
1256          * child regardless of CLONE_PTRACE.
1257          */
1258         user_disable_single_step(p);
1259         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1260 #ifdef TIF_SYSCALL_EMU
1261         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1262 #endif
1263         clear_all_latency_tracing(p);
1264
1265         /* ok, now we should be set up.. */
1266         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1267         p->pdeath_signal = 0;
1268         p->exit_state = 0;
1269
1270         /*
1271          * Ok, make it visible to the rest of the system.
1272          * We dont wake it up yet.
1273          */
1274         p->group_leader = p;
1275         INIT_LIST_HEAD(&p->thread_group);
1276
1277         /* Now that the task is set up, run cgroup callbacks if
1278          * necessary. We need to run them before the task is visible
1279          * on the tasklist. */
1280         cgroup_fork_callbacks(p);
1281         cgroup_callbacks_done = 1;
1282
1283         /* Need tasklist lock for parent etc handling! */
1284         write_lock_irq(&tasklist_lock);
1285
1286         /* CLONE_PARENT re-uses the old parent */
1287         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1288                 p->real_parent = current->real_parent;
1289                 p->parent_exec_id = current->parent_exec_id;
1290         } else {
1291                 p->real_parent = current;
1292                 p->parent_exec_id = current->self_exec_id;
1293         }
1294
1295         spin_lock(&current->sighand->siglock);
1296
1297         /*
1298          * Process group and session signals need to be delivered to just the
1299          * parent before the fork or both the parent and the child after the
1300          * fork. Restart if a signal comes in before we add the new process to
1301          * it's process group.
1302          * A fatal signal pending means that current will exit, so the new
1303          * thread can't slip out of an OOM kill (or normal SIGKILL).
1304          */
1305         recalc_sigpending();
1306         if (signal_pending(current)) {
1307                 spin_unlock(&current->sighand->siglock);
1308                 write_unlock_irq(&tasklist_lock);
1309                 retval = -ERESTARTNOINTR;
1310                 goto bad_fork_free_pid;
1311         }
1312
1313         if (clone_flags & CLONE_THREAD) {
1314                 current->signal->nr_threads++;
1315                 atomic_inc(&current->signal->live);
1316                 atomic_inc(&current->signal->sigcnt);
1317                 p->group_leader = current->group_leader;
1318                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1319         }
1320
1321         if (likely(p->pid)) {
1322                 tracehook_finish_clone(p, clone_flags, trace);
1323
1324                 if (thread_group_leader(p)) {
1325                         if (is_child_reaper(pid))
1326                                 p->nsproxy->pid_ns->child_reaper = p;
1327
1328                         p->signal->leader_pid = pid;
1329                         p->signal->tty = tty_kref_get(current->signal->tty);
1330                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1331                         attach_pid(p, PIDTYPE_SID, task_session(current));
1332                         list_add_tail(&p->sibling, &p->real_parent->children);
1333                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1334                         __this_cpu_inc(process_counts);
1335                 }
1336                 attach_pid(p, PIDTYPE_PID, pid);
1337                 nr_threads++;
1338         }
1339
1340         total_forks++;
1341         spin_unlock(&current->sighand->siglock);
1342         write_unlock_irq(&tasklist_lock);
1343         proc_fork_connector(p);
1344         cgroup_post_fork(p);
1345         if (clone_flags & CLONE_THREAD)
1346                 threadgroup_fork_read_unlock(current);
1347         perf_event_fork(p);
1348         return p;
1349
1350 bad_fork_free_pid:
1351         if (pid != &init_struct_pid)
1352                 free_pid(pid);
1353 bad_fork_cleanup_io:
1354         if (p->io_context)
1355                 exit_io_context(p);
1356 bad_fork_cleanup_namespaces:
1357         exit_task_namespaces(p);
1358 bad_fork_cleanup_mm:
1359         if (p->mm) {
1360                 task_lock(p);
1361                 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1362                         atomic_dec(&p->mm->oom_disable_count);
1363                 task_unlock(p);
1364                 mmput(p->mm);
1365         }
1366 bad_fork_cleanup_signal:
1367         if (!(clone_flags & CLONE_THREAD))
1368                 free_signal_struct(p->signal);
1369 bad_fork_cleanup_sighand:
1370         __cleanup_sighand(p->sighand);
1371 bad_fork_cleanup_fs:
1372         exit_fs(p); /* blocking */
1373 bad_fork_cleanup_files:
1374         exit_files(p); /* blocking */
1375 bad_fork_cleanup_semundo:
1376         exit_sem(p);
1377 bad_fork_cleanup_audit:
1378         audit_free(p);
1379 bad_fork_cleanup_policy:
1380         perf_event_free_task(p);
1381 #ifdef CONFIG_NUMA
1382         mpol_put(p->mempolicy);
1383 bad_fork_cleanup_cgroup:
1384 #endif
1385         if (clone_flags & CLONE_THREAD)
1386                 threadgroup_fork_read_unlock(current);
1387         cgroup_exit(p, cgroup_callbacks_done);
1388         delayacct_tsk_free(p);
1389         module_put(task_thread_info(p)->exec_domain->module);
1390 bad_fork_cleanup_count:
1391         atomic_dec(&p->cred->user->processes);
1392         exit_creds(p);
1393 bad_fork_free:
1394         free_task(p);
1395 fork_out:
1396         return ERR_PTR(retval);
1397 }
1398
1399 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1400 {
1401         memset(regs, 0, sizeof(struct pt_regs));
1402         return regs;
1403 }
1404
1405 static inline void init_idle_pids(struct pid_link *links)
1406 {
1407         enum pid_type type;
1408
1409         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1410                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1411                 links[type].pid = &init_struct_pid;
1412         }
1413 }
1414
1415 struct task_struct * __cpuinit fork_idle(int cpu)
1416 {
1417         struct task_struct *task;
1418         struct pt_regs regs;
1419
1420         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1421                             &init_struct_pid, 0);
1422         if (!IS_ERR(task)) {
1423                 init_idle_pids(task->pids);
1424                 init_idle(task, cpu);
1425         }
1426
1427         return task;
1428 }
1429
1430 /*
1431  *  Ok, this is the main fork-routine.
1432  *
1433  * It copies the process, and if successful kick-starts
1434  * it and waits for it to finish using the VM if required.
1435  */
1436 long do_fork(unsigned long clone_flags,
1437               unsigned long stack_start,
1438               struct pt_regs *regs,
1439               unsigned long stack_size,
1440               int __user *parent_tidptr,
1441               int __user *child_tidptr)
1442 {
1443         struct task_struct *p;
1444         int trace = 0;
1445         long nr;
1446
1447         /*
1448          * Do some preliminary argument and permissions checking before we
1449          * actually start allocating stuff
1450          */
1451         if (clone_flags & CLONE_NEWUSER) {
1452                 if (clone_flags & CLONE_THREAD)
1453                         return -EINVAL;
1454                 /* hopefully this check will go away when userns support is
1455                  * complete
1456                  */
1457                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1458                                 !capable(CAP_SETGID))
1459                         return -EPERM;
1460         }
1461
1462         /*
1463          * When called from kernel_thread, don't do user tracing stuff.
1464          */
1465         if (likely(user_mode(regs)))
1466                 trace = tracehook_prepare_clone(clone_flags);
1467
1468         p = copy_process(clone_flags, stack_start, regs, stack_size,
1469                          child_tidptr, NULL, trace);
1470         /*
1471          * Do this prior waking up the new thread - the thread pointer
1472          * might get invalid after that point, if the thread exits quickly.
1473          */
1474         if (!IS_ERR(p)) {
1475                 struct completion vfork;
1476
1477                 trace_sched_process_fork(current, p);
1478
1479                 nr = task_pid_vnr(p);
1480
1481                 if (clone_flags & CLONE_PARENT_SETTID)
1482                         put_user(nr, parent_tidptr);
1483
1484                 if (clone_flags & CLONE_VFORK) {
1485                         p->vfork_done = &vfork;
1486                         init_completion(&vfork);
1487                 }
1488
1489                 audit_finish_fork(p);
1490                 tracehook_report_clone(regs, clone_flags, nr, p);
1491
1492                 /*
1493                  * We set PF_STARTING at creation in case tracing wants to
1494                  * use this to distinguish a fully live task from one that
1495                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1496                  * clear it and set the child going.
1497                  */
1498                 p->flags &= ~PF_STARTING;
1499
1500                 wake_up_new_task(p);
1501
1502                 tracehook_report_clone_complete(trace, regs,
1503                                                 clone_flags, nr, p);
1504
1505                 if (clone_flags & CLONE_VFORK) {
1506                         freezer_do_not_count();
1507                         wait_for_completion(&vfork);
1508                         freezer_count();
1509                         tracehook_report_vfork_done(p, nr);
1510                 }
1511         } else {
1512                 nr = PTR_ERR(p);
1513         }
1514         return nr;
1515 }
1516
1517 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1518 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1519 #endif
1520
1521 static void sighand_ctor(void *data)
1522 {
1523         struct sighand_struct *sighand = data;
1524
1525         spin_lock_init(&sighand->siglock);
1526         init_waitqueue_head(&sighand->signalfd_wqh);
1527 }
1528
1529 void __init proc_caches_init(void)
1530 {
1531         sighand_cachep = kmem_cache_create("sighand_cache",
1532                         sizeof(struct sighand_struct), 0,
1533                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1534                         SLAB_NOTRACK, sighand_ctor);
1535         signal_cachep = kmem_cache_create("signal_cache",
1536                         sizeof(struct signal_struct), 0,
1537                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1538         files_cachep = kmem_cache_create("files_cache",
1539                         sizeof(struct files_struct), 0,
1540                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1541         fs_cachep = kmem_cache_create("fs_cache",
1542                         sizeof(struct fs_struct), 0,
1543                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1544         mm_cachep = kmem_cache_create("mm_struct",
1545                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1546                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1547         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1548         mmap_init();
1549 }
1550
1551 /*
1552  * Check constraints on flags passed to the unshare system call.
1553  */
1554 static int check_unshare_flags(unsigned long unshare_flags)
1555 {
1556         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1557                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1558                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1559                 return -EINVAL;
1560         /*
1561          * Not implemented, but pretend it works if there is nothing to
1562          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1563          * needs to unshare vm.
1564          */
1565         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1566                 /* FIXME: get_task_mm() increments ->mm_users */
1567                 if (atomic_read(&current->mm->mm_users) > 1)
1568                         return -EINVAL;
1569         }
1570
1571         return 0;
1572 }
1573
1574 /*
1575  * Unshare the filesystem structure if it is being shared
1576  */
1577 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1578 {
1579         struct fs_struct *fs = current->fs;
1580
1581         if (!(unshare_flags & CLONE_FS) || !fs)
1582                 return 0;
1583
1584         /* don't need lock here; in the worst case we'll do useless copy */
1585         if (fs->users == 1)
1586                 return 0;
1587
1588         *new_fsp = copy_fs_struct(fs);
1589         if (!*new_fsp)
1590                 return -ENOMEM;
1591
1592         return 0;
1593 }
1594
1595 /*
1596  * Unshare file descriptor table if it is being shared
1597  */
1598 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1599 {
1600         struct files_struct *fd = current->files;
1601         int error = 0;
1602
1603         if ((unshare_flags & CLONE_FILES) &&
1604             (fd && atomic_read(&fd->count) > 1)) {
1605                 *new_fdp = dup_fd(fd, &error);
1606                 if (!*new_fdp)
1607                         return error;
1608         }
1609
1610         return 0;
1611 }
1612
1613 /*
1614  * unshare allows a process to 'unshare' part of the process
1615  * context which was originally shared using clone.  copy_*
1616  * functions used by do_fork() cannot be used here directly
1617  * because they modify an inactive task_struct that is being
1618  * constructed. Here we are modifying the current, active,
1619  * task_struct.
1620  */
1621 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1622 {
1623         struct fs_struct *fs, *new_fs = NULL;
1624         struct files_struct *fd, *new_fd = NULL;
1625         struct nsproxy *new_nsproxy = NULL;
1626         int do_sysvsem = 0;
1627         int err;
1628
1629         err = check_unshare_flags(unshare_flags);
1630         if (err)
1631                 goto bad_unshare_out;
1632
1633         /*
1634          * If unsharing namespace, must also unshare filesystem information.
1635          */
1636         if (unshare_flags & CLONE_NEWNS)
1637                 unshare_flags |= CLONE_FS;
1638         /*
1639          * CLONE_NEWIPC must also detach from the undolist: after switching
1640          * to a new ipc namespace, the semaphore arrays from the old
1641          * namespace are unreachable.
1642          */
1643         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1644                 do_sysvsem = 1;
1645         if ((err = unshare_fs(unshare_flags, &new_fs)))
1646                 goto bad_unshare_out;
1647         if ((err = unshare_fd(unshare_flags, &new_fd)))
1648                 goto bad_unshare_cleanup_fs;
1649         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1650                         new_fs)))
1651                 goto bad_unshare_cleanup_fd;
1652
1653         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1654                 if (do_sysvsem) {
1655                         /*
1656                          * CLONE_SYSVSEM is equivalent to sys_exit().
1657                          */
1658                         exit_sem(current);
1659                 }
1660
1661                 if (new_nsproxy) {
1662                         switch_task_namespaces(current, new_nsproxy);
1663                         new_nsproxy = NULL;
1664                 }
1665
1666                 task_lock(current);
1667
1668                 if (new_fs) {
1669                         fs = current->fs;
1670                         spin_lock(&fs->lock);
1671                         current->fs = new_fs;
1672                         if (--fs->users)
1673                                 new_fs = NULL;
1674                         else
1675                                 new_fs = fs;
1676                         spin_unlock(&fs->lock);
1677                 }
1678
1679                 if (new_fd) {
1680                         fd = current->files;
1681                         current->files = new_fd;
1682                         new_fd = fd;
1683                 }
1684
1685                 task_unlock(current);
1686         }
1687
1688         if (new_nsproxy)
1689                 put_nsproxy(new_nsproxy);
1690
1691 bad_unshare_cleanup_fd:
1692         if (new_fd)
1693                 put_files_struct(new_fd);
1694
1695 bad_unshare_cleanup_fs:
1696         if (new_fs)
1697                 free_fs_struct(new_fs);
1698
1699 bad_unshare_out:
1700         return err;
1701 }
1702
1703 /*
1704  *      Helper to unshare the files of the current task.
1705  *      We don't want to expose copy_files internals to
1706  *      the exec layer of the kernel.
1707  */
1708
1709 int unshare_files(struct files_struct **displaced)
1710 {
1711         struct task_struct *task = current;
1712         struct files_struct *copy = NULL;
1713         int error;
1714
1715         error = unshare_fd(CLONE_FILES, &copy);
1716         if (error || !copy) {
1717                 *displaced = NULL;
1718                 return error;
1719         }
1720         *displaced = task->files;
1721         task_lock(task);
1722         task->files = copy;
1723         task_unlock(task);
1724         return 0;
1725 }