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/task_io_accounting_ops.h>
  44 #include <linux/rcupdate.h>
  45 #include <linux/ptrace.h>
  46 #include <linux/mount.h>
  47 #include <linux/audit.h>
  48 #include <linux/memcontrol.h>
  49 #include <linux/ftrace.h>
  50 #include <linux/profile.h>
  51 #include <linux/rmap.h>
  52 #include <linux/ksm.h>
  53 #include <linux/acct.h>
  54 #include <linux/tsacct_kern.h>
  55 #include <linux/cn_proc.h>
  56 #include <linux/freezer.h>
  57 #include <linux/delayacct.h>
  58 #include <linux/taskstats_kern.h>
  59 #include <linux/random.h>
  60 #include <linux/tty.h>
  61 #include <linux/proc_fs.h>
  62 #include <linux/blkdev.h>
  63 #include <linux/fs_struct.h>
  64 #include <linux/magic.h>
  65 #include <linux/perf_event.h>
  66 #include <linux/posix-timers.h>
  67 #include <linux/user-return-notifier.h>
  68
  69 #include <asm/pgtable.h>
  70 #include <asm/pgalloc.h>
  71 #include <asm/uaccess.h>
  72 #include <asm/mmu_context.h>
  73 #include <asm/cacheflush.h>
  74 #include <asm/tlbflush.h>
  75
  76 #include <trace/events/sched.h>
  77
  78 /*
  79  * Protected counters by write_lock_irq(&tasklist_lock)
  80  */
  81 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  82 int nr_threads;                 /* The idle threads do not count.. */
  83
  84 int max_threads;                /* tunable limit on nr_threads */
  85
  86 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  87
  88 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  89
  90 #ifdef CONFIG_PROVE_RCU
  91 int lockdep_tasklist_lock_is_held(void)
  92 {
  93         return lockdep_is_held(&tasklist_lock);
  94 }
  95 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
  96 #endif /* #ifdef CONFIG_PROVE_RCU */
  97
  98 int nr_processes(void)
  99 {
 100         int cpu;
 101         int total = 0;
 102
 103         for_each_possible_cpu(cpu)
 104                 total += per_cpu(process_counts, cpu);
 105
 106         return total;
 107 }
 108
 109 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 110 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
 111 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
 112 static struct kmem_cache *task_struct_cachep;
 113 #endif
 114
 115 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 116 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
 117 {
 118 #ifdef CONFIG_DEBUG_STACK_USAGE
 119         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 120 #else
 121         gfp_t mask = GFP_KERNEL;
 122 #endif
 123         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
 124 }
 125
 126 static inline void free_thread_info(struct thread_info *ti)
 127 {
 128         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 129 }
 130 #endif
 131
 132 /* SLAB cache for signal_struct structures (tsk->signal) */
 133 static struct kmem_cache *signal_cachep;
 134
 135 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 136 struct kmem_cache *sighand_cachep;
 137
 138 /* SLAB cache for files_struct structures (tsk->files) */
 139 struct kmem_cache *files_cachep;
 140
 141 /* SLAB cache for fs_struct structures (tsk->fs) */
 142 struct kmem_cache *fs_cachep;
 143
 144 /* SLAB cache for vm_area_struct structures */
 145 struct kmem_cache *vm_area_cachep;
 146
 147 /* SLAB cache for mm_struct structures (tsk->mm) */
 148 static struct kmem_cache *mm_cachep;
 149
 150 /* Notifier list called when a task struct is freed */
 151 static ATOMIC_NOTIFIER_HEAD(task_free_notifier);
 152
 153 static void account_kernel_stack(struct thread_info *ti, int account)
 154 {
 155         struct zone *zone = page_zone(virt_to_page(ti));
 156
 157         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 158 }
 159
 160 void free_task(struct task_struct *tsk)
 161 {
 162         prop_local_destroy_single(&tsk->dirties);
 163         account_kernel_stack(tsk->stack, -1);
 164         free_thread_info(tsk->stack);
 165         rt_mutex_debug_task_free(tsk);
 166         ftrace_graph_exit_task(tsk);
 167         free_task_struct(tsk);
 168 }
 169 EXPORT_SYMBOL(free_task);
 170
 171 static inline void free_signal_struct(struct signal_struct *sig)
 172 {
 173         taskstats_tgid_free(sig);
 174         kmem_cache_free(signal_cachep, sig);
 175 }
 176
 177 static inline void put_signal_struct(struct signal_struct *sig)
 178 {
 179         if (atomic_dec_and_test(&sig->sigcnt))
 180                 free_signal_struct(sig);
 181 }
 182
 183 int task_free_register(struct notifier_block *n)
 184 {
 185         return atomic_notifier_chain_register(&task_free_notifier, n);
 186 }
 187 EXPORT_SYMBOL(task_free_register);
 188
 189 int task_free_unregister(struct notifier_block *n)
 190 {
 191         return atomic_notifier_chain_unregister(&task_free_notifier, n);
 192 }
 193 EXPORT_SYMBOL(task_free_unregister);
 194
 195 void __put_task_struct(struct task_struct *tsk)
 196 {
 197         WARN_ON(!tsk->exit_state);
 198         WARN_ON(atomic_read(&tsk->usage));
 199         WARN_ON(tsk == current);
 200
 201         exit_creds(tsk);
 202         delayacct_tsk_free(tsk);
 203         put_signal_struct(tsk->signal);
 204
 205         atomic_notifier_call_chain(&task_free_notifier, 0, tsk);
 206         if (!profile_handoff_task(tsk))
 207                 free_task(tsk);
 208 }
 209
 210 /*
 211  * macro override instead of weak attribute alias, to workaround
 212  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 213  */
 214 #ifndef arch_task_cache_init
 215 #define arch_task_cache_init()
 216 #endif
 217
 218 void __init fork_init(unsigned long mempages)
 219 {
 220 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 221 #ifndef ARCH_MIN_TASKALIGN
 222 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 223 #endif
 224         /* create a slab on which task_structs can be allocated */
 225         task_struct_cachep =
 226                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 227                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 228 #endif
 229
 230         /* do the arch specific task caches init */
 231         arch_task_cache_init();
 232
 233         /*
 234          * The default maximum number of threads is set to a safe
 235          * value: the thread structures can take up at most half
 236          * of memory.
 237          */
 238         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 239
 240         /*
 241          * we need to allow at least 20 threads to boot a system
 242          */
 243         if(max_threads < 20)
 244                 max_threads = 20;
 245
 246         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 247         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 248         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 249                 init_task.signal->rlim[RLIMIT_NPROC];
 250 }
 251
 252 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 253                                                struct task_struct *src)
 254 {
 255         *dst = *src;
 256         return 0;
 257 }
 258
 259 static struct task_struct *dup_task_struct(struct task_struct *orig)
 260 {
 261         struct task_struct *tsk;
 262         struct thread_info *ti;
 263         unsigned long *stackend;
 264
 265         int err;
 266
 267         prepare_to_copy(orig);
 268
 269         tsk = alloc_task_struct();
 270         if (!tsk)
 271                 return NULL;
 272
 273         ti = alloc_thread_info(tsk);
 274         if (!ti) {
 275                 free_task_struct(tsk);
 276                 return NULL;
 277         }
 278
 279         err = arch_dup_task_struct(tsk, orig);
 280         if (err)
 281                 goto out;
 282
 283         tsk->stack = ti;
 284
 285         err = prop_local_init_single(&tsk->dirties);
 286         if (err)
 287                 goto out;
 288
 289         setup_thread_stack(tsk, orig);
 290         clear_user_return_notifier(tsk);
 291         clear_tsk_need_resched(tsk);
 292         stackend = end_of_stack(tsk);
 293         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 294
 295 #ifdef CONFIG_CC_STACKPROTECTOR
 296         tsk->stack_canary = get_random_int();
 297 #endif
 298
 299         /* One for us, one for whoever does the "release_task()" (usually parent) */
 300         atomic_set(&tsk->usage,2);
 301         atomic_set(&tsk->fs_excl, 0);
 302 #ifdef CONFIG_BLK_DEV_IO_TRACE
 303         tsk->btrace_seq = 0;
 304 #endif
 305         tsk->splice_pipe = NULL;
 306
 307         account_kernel_stack(ti, 1);
 308
 309         return tsk;
 310
 311 out:
 312         free_thread_info(ti);
 313         free_task_struct(tsk);
 314         return NULL;
 315 }
 316
 317 #ifdef CONFIG_MMU
 318 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 319 {
 320         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 321         struct rb_node **rb_link, *rb_parent;
 322         int retval;
 323         unsigned long charge;
 324         struct mempolicy *pol;
 325
 326         down_write(&oldmm->mmap_sem);
 327         flush_cache_dup_mm(oldmm);
 328         /*
 329          * Not linked in yet - no deadlock potential:
 330          */
 331         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 332
 333         mm->locked_vm = 0;
 334         mm->mmap = NULL;
 335         mm->mmap_cache = NULL;
 336         mm->free_area_cache = oldmm->mmap_base;
 337         mm->cached_hole_size = ~0UL;
 338         mm->map_count = 0;
 339         cpumask_clear(mm_cpumask(mm));
 340         mm->mm_rb = RB_ROOT;
 341         rb_link = &mm->mm_rb.rb_node;
 342         rb_parent = NULL;
 343         pprev = &mm->mmap;
 344         retval = ksm_fork(mm, oldmm);
 345         if (retval)
 346                 goto out;
 347
 348         prev = NULL;
 349         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 350                 struct file *file;
 351
 352                 if (mpnt->vm_flags & VM_DONTCOPY) {
 353                         long pages = vma_pages(mpnt);
 354                         mm->total_vm -= pages;
 355                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 356                                                                 -pages);
 357                         continue;
 358                 }
 359                 charge = 0;
 360                 if (mpnt->vm_flags & VM_ACCOUNT) {
 361                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 362                         if (security_vm_enough_memory(len))
 363                                 goto fail_nomem;
 364                         charge = len;
 365                 }
 366                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 367                 if (!tmp)
 368                         goto fail_nomem;
 369                 *tmp = *mpnt;
 370                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 371                 pol = mpol_dup(vma_policy(mpnt));
 372                 retval = PTR_ERR(pol);
 373                 if (IS_ERR(pol))
 374                         goto fail_nomem_policy;
 375                 vma_set_policy(tmp, pol);
 376                 tmp->vm_mm = mm;
 377                 if (anon_vma_fork(tmp, mpnt))
 378                         goto fail_nomem_anon_vma_fork;
 379                 tmp->vm_flags &= ~VM_LOCKED;
 380                 tmp->vm_next = tmp->vm_prev = NULL;
 381                 file = tmp->vm_file;
 382                 if (file) {
 383                         struct inode *inode = file->f_path.dentry->d_inode;
 384                         struct address_space *mapping = file->f_mapping;
 385
 386                         get_file(file);
 387                         if (tmp->vm_flags & VM_DENYWRITE)
 388                                 atomic_dec(&inode->i_writecount);
 389                         spin_lock(&mapping->i_mmap_lock);
 390                         if (tmp->vm_flags & VM_SHARED)
 391                                 mapping->i_mmap_writable++;
 392                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
 393                         flush_dcache_mmap_lock(mapping);
 394                         /* insert tmp into the share list, just after mpnt */
 395                         vma_prio_tree_add(tmp, mpnt);
 396                         flush_dcache_mmap_unlock(mapping);
 397                         spin_unlock(&mapping->i_mmap_lock);
 398                 }
 399
 400                 /*
 401                  * Clear hugetlb-related page reserves for children. This only
 402                  * affects MAP_PRIVATE mappings. Faults generated by the child
 403                  * are not guaranteed to succeed, even if read-only
 404                  */
 405                 if (is_vm_hugetlb_page(tmp))
 406                         reset_vma_resv_huge_pages(tmp);
 407
 408                 /*
 409                  * Link in the new vma and copy the page table entries.
 410                  */
 411                 *pprev = tmp;
 412                 pprev = &tmp->vm_next;
 413                 tmp->vm_prev = prev;
 414                 prev = tmp;
 415
 416                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 417                 rb_link = &tmp->vm_rb.rb_right;
 418                 rb_parent = &tmp->vm_rb;
 419
 420                 mm->map_count++;
 421                 retval = copy_page_range(mm, oldmm, mpnt);
 422
 423                 if (tmp->vm_ops && tmp->vm_ops->open)
 424                         tmp->vm_ops->open(tmp);
 425
 426                 if (retval)
 427                         goto out;
 428         }
 429         /* a new mm has just been created */
 430         arch_dup_mmap(oldmm, mm);
 431         retval = 0;
 432 out:
 433         up_write(&mm->mmap_sem);
 434         flush_tlb_mm(oldmm);
 435         up_write(&oldmm->mmap_sem);
 436         return retval;
 437 fail_nomem_anon_vma_fork:
 438         mpol_put(pol);
 439 fail_nomem_policy:
 440         kmem_cache_free(vm_area_cachep, tmp);
 441 fail_nomem:
 442         retval = -ENOMEM;
 443         vm_unacct_memory(charge);
 444         goto out;
 445 }
 446
 447 static inline int mm_alloc_pgd(struct mm_struct * mm)
 448 {
 449         mm->pgd = pgd_alloc(mm);
 450         if (unlikely(!mm->pgd))
 451                 return -ENOMEM;
 452         return 0;
 453 }
 454
 455 static inline void mm_free_pgd(struct mm_struct * mm)
 456 {
 457         pgd_free(mm, mm->pgd);
 458 }
 459 #else
 460 #define dup_mmap(mm, oldmm)     (0)
 461 #define mm_alloc_pgd(mm)        (0)
 462 #define mm_free_pgd(mm)
 463 #endif /* CONFIG_MMU */
 464
 465 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 466
 467 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 468 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 469
 470 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 471
 472 static int __init coredump_filter_setup(char *s)
 473 {
 474         default_dump_filter =
 475                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 476                 MMF_DUMP_FILTER_MASK;
 477         return 1;
 478 }
 479
 480 __setup("coredump_filter=", coredump_filter_setup);
 481
 482 #include <linux/init_task.h>
 483
 484 static void mm_init_aio(struct mm_struct *mm)
 485 {
 486 #ifdef CONFIG_AIO
 487         spin_lock_init(&mm->ioctx_lock);
 488         INIT_HLIST_HEAD(&mm->ioctx_list);
 489 #endif
 490 }
 491
 492 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
 493 {
 494         atomic_set(&mm->mm_users, 1);
 495         atomic_set(&mm->mm_count, 1);
 496         init_rwsem(&mm->mmap_sem);
 497         INIT_LIST_HEAD(&mm->mmlist);
 498         mm->flags = (current->mm) ?
 499                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 500         mm->core_state = NULL;
 501         mm->nr_ptes = 0;
 502         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 503         spin_lock_init(&mm->page_table_lock);
 504         mm->free_area_cache = TASK_UNMAPPED_BASE;
 505         mm->cached_hole_size = ~0UL;
 506         mm_init_aio(mm);
 507         mm_init_owner(mm, p);
 508
 509         if (likely(!mm_alloc_pgd(mm))) {
 510                 mm->def_flags = 0;
 511                 mmu_notifier_mm_init(mm);
 512                 return mm;
 513         }
 514
 515         free_mm(mm);
 516         return NULL;
 517 }
 518
 519 /*
 520  * Allocate and initialize an mm_struct.
 521  */
 522 struct mm_struct * mm_alloc(void)
 523 {
 524         struct mm_struct * mm;
 525
 526         mm = allocate_mm();
 527         if (mm) {
 528                 memset(mm, 0, sizeof(*mm));
 529                 mm = mm_init(mm, current);
 530         }
 531         return mm;
 532 }
 533
 534 /*
 535  * Called when the last reference to the mm
 536  * is dropped: either by a lazy thread or by
 537  * mmput. Free the page directory and the mm.
 538  */
 539 void __mmdrop(struct mm_struct *mm)
 540 {
 541         BUG_ON(mm == &init_mm);
 542         mm_free_pgd(mm);
 543         destroy_context(mm);
 544         mmu_notifier_mm_destroy(mm);
 545         free_mm(mm);
 546 }
 547 EXPORT_SYMBOL_GPL(__mmdrop);
 548
 549 /*
 550  * Decrement the use count and release all resources for an mm.
 551  */
 552 void mmput(struct mm_struct *mm)
 553 {
 554         might_sleep();
 555
 556         if (atomic_dec_and_test(&mm->mm_users)) {
 557                 exit_aio(mm);
 558                 ksm_exit(mm);
 559                 exit_mmap(mm);
 560                 set_mm_exe_file(mm, NULL);
 561                 if (!list_empty(&mm->mmlist)) {
 562                         spin_lock(&mmlist_lock);
 563                         list_del(&mm->mmlist);
 564                         spin_unlock(&mmlist_lock);
 565                 }
 566                 put_swap_token(mm);
 567                 if (mm->binfmt)
 568                         module_put(mm->binfmt->module);
 569                 mmdrop(mm);
 570         }
 571 }
 572 EXPORT_SYMBOL_GPL(mmput);
 573
 574 /**
 575  * get_task_mm - acquire a reference to the task's mm
 576  *
 577  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 578  * this kernel workthread has transiently adopted a user mm with use_mm,
 579  * to do its AIO) is not set and if so returns a reference to it, after
 580  * bumping up the use count.  User must release the mm via mmput()
 581  * after use.  Typically used by /proc and ptrace.
 582  */
 583 struct mm_struct *get_task_mm(struct task_struct *task)
 584 {
 585         struct mm_struct *mm;
 586
 587         task_lock(task);
 588         mm = task->mm;
 589         if (mm) {
 590                 if (task->flags & PF_KTHREAD)
 591                         mm = NULL;
 592                 else
 593                         atomic_inc(&mm->mm_users);
 594         }
 595         task_unlock(task);
 596         return mm;
 597 }
 598 EXPORT_SYMBOL_GPL(get_task_mm);
 599
 600 /* Please note the differences between mmput and mm_release.
 601  * mmput is called whenever we stop holding onto a mm_struct,
 602  * error success whatever.
 603  *
 604  * mm_release is called after a mm_struct has been removed
 605  * from the current process.
 606  *
 607  * This difference is important for error handling, when we
 608  * only half set up a mm_struct for a new process and need to restore
 609  * the old one.  Because we mmput the new mm_struct before
 610  * restoring the old one. . .
 611  * Eric Biederman 10 January 1998
 612  */
 613 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 614 {
 615         struct completion *vfork_done = tsk->vfork_done;
 616
 617         /* Get rid of any futexes when releasing the mm */
 618 #ifdef CONFIG_FUTEX
 619         if (unlikely(tsk->robust_list)) {
 620                 exit_robust_list(tsk);
 621                 tsk->robust_list = NULL;
 622         }
 623 #ifdef CONFIG_COMPAT
 624         if (unlikely(tsk->compat_robust_list)) {
 625                 compat_exit_robust_list(tsk);
 626                 tsk->compat_robust_list = NULL;
 627         }
 628 #endif
 629         if (unlikely(!list_empty(&tsk->pi_state_list)))
 630                 exit_pi_state_list(tsk);
 631 #endif
 632
 633         /* Get rid of any cached register state */
 634         deactivate_mm(tsk, mm);
 635
 636         /* notify parent sleeping on vfork() */
 637         if (vfork_done) {
 638                 tsk->vfork_done = NULL;
 639                 complete(vfork_done);
 640         }
 641
 642         /*
 643          * If we're exiting normally, clear a user-space tid field if
 644          * requested.  We leave this alone when dying by signal, to leave
 645          * the value intact in a core dump, and to save the unnecessary
 646          * trouble otherwise.  Userland only wants this done for a sys_exit.
 647          */
 648         if (tsk->clear_child_tid) {
 649                 if (!(tsk->flags & PF_SIGNALED) &&
 650                     atomic_read(&mm->mm_users) > 1) {
 651                         /*
 652                          * We don't check the error code - if userspace has
 653                          * not set up a proper pointer then tough luck.
 654                          */
 655                         put_user(0, tsk->clear_child_tid);
 656                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 657                                         1, NULL, NULL, 0);
 658                 }
 659                 tsk->clear_child_tid = NULL;
 660         }
 661 }
 662
 663 /*
 664  * Allocate a new mm structure and copy contents from the
 665  * mm structure of the passed in task structure.
 666  */
 667 struct mm_struct *dup_mm(struct task_struct *tsk)
 668 {
 669         struct mm_struct *mm, *oldmm = current->mm;
 670         int err;
 671
 672         if (!oldmm)
 673                 return NULL;
 674
 675         mm = allocate_mm();
 676         if (!mm)
 677                 goto fail_nomem;
 678
 679         memcpy(mm, oldmm, sizeof(*mm));
 680
 681         /* Initializing for Swap token stuff */
 682         mm->token_priority = 0;
 683         mm->last_interval = 0;
 684
 685         if (!mm_init(mm, tsk))
 686                 goto fail_nomem;
 687
 688         if (init_new_context(tsk, mm))
 689                 goto fail_nocontext;
 690
 691         dup_mm_exe_file(oldmm, mm);
 692
 693         err = dup_mmap(mm, oldmm);
 694         if (err)
 695                 goto free_pt;
 696
 697         mm->hiwater_rss = get_mm_rss(mm);
 698         mm->hiwater_vm = mm->total_vm;
 699
 700         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 701                 goto free_pt;
 702
 703         return mm;
 704
 705 free_pt:
 706         /* don't put binfmt in mmput, we haven't got module yet */
 707         mm->binfmt = NULL;
 708         mmput(mm);
 709
 710 fail_nomem:
 711         return NULL;
 712
 713 fail_nocontext:
 714         /*
 715          * If init_new_context() failed, we cannot use mmput() to free the mm
 716          * because it calls destroy_context()
 717          */
 718         mm_free_pgd(mm);
 719         free_mm(mm);
 720         return NULL;
 721 }
 722
 723 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 724 {
 725         struct mm_struct * mm, *oldmm;
 726         int retval;
 727
 728         tsk->min_flt = tsk->maj_flt = 0;
 729         tsk->nvcsw = tsk->nivcsw = 0;
 730 #ifdef CONFIG_DETECT_HUNG_TASK
 731         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 732 #endif
 733
 734         tsk->mm = NULL;
 735         tsk->active_mm = NULL;
 736
 737         /*
 738          * Are we cloning a kernel thread?
 739          *
 740          * We need to steal a active VM for that..
 741          */
 742         oldmm = current->mm;
 743         if (!oldmm)
 744                 return 0;
 745
 746         if (clone_flags & CLONE_VM) {
 747                 atomic_inc(&oldmm->mm_users);
 748                 mm = oldmm;
 749                 goto good_mm;
 750         }
 751
 752         retval = -ENOMEM;
 753         mm = dup_mm(tsk);
 754         if (!mm)
 755                 goto fail_nomem;
 756
 757 good_mm:
 758         /* Initializing for Swap token stuff */
 759         mm->token_priority = 0;
 760         mm->last_interval = 0;
 761
 762         tsk->mm = mm;
 763         tsk->active_mm = mm;
 764         return 0;
 765
 766 fail_nomem:
 767         return retval;
 768 }
 769
 770 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 771 {
 772         struct fs_struct *fs = current->fs;
 773         if (clone_flags & CLONE_FS) {
 774                 /* tsk->fs is already what we want */
 775                 spin_lock(&fs->lock);
 776                 if (fs->in_exec) {
 777                         spin_unlock(&fs->lock);
 778                         return -EAGAIN;
 779                 }
 780                 fs->users++;
 781                 spin_unlock(&fs->lock);
 782                 return 0;
 783         }
 784         tsk->fs = copy_fs_struct(fs);
 785         if (!tsk->fs)
 786                 return -ENOMEM;
 787         return 0;
 788 }
 789
 790 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 791 {
 792         struct files_struct *oldf, *newf;
 793         int error = 0;
 794
 795         /*
 796          * A background process may not have any files ...
 797          */
 798         oldf = current->files;
 799         if (!oldf)
 800                 goto out;
 801
 802         if (clone_flags & CLONE_FILES) {
 803                 atomic_inc(&oldf->count);
 804                 goto out;
 805         }
 806
 807         newf = dup_fd(oldf, &error);
 808         if (!newf)
 809                 goto out;
 810
 811         tsk->files = newf;
 812         error = 0;
 813 out:
 814         return error;
 815 }
 816
 817 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 818 {
 819 #ifdef CONFIG_BLOCK
 820         struct io_context *ioc = current->io_context;
 821
 822         if (!ioc)
 823                 return 0;
 824         /*
 825          * Share io context with parent, if CLONE_IO is set
 826          */
 827         if (clone_flags & CLONE_IO) {
 828                 tsk->io_context = ioc_task_link(ioc);
 829                 if (unlikely(!tsk->io_context))
 830                         return -ENOMEM;
 831         } else if (ioprio_valid(ioc->ioprio)) {
 832                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
 833                 if (unlikely(!tsk->io_context))
 834                         return -ENOMEM;
 835
 836                 tsk->io_context->ioprio = ioc->ioprio;
 837         }
 838 #endif
 839         return 0;
 840 }
 841
 842 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 843 {
 844         struct sighand_struct *sig;
 845
 846         if (clone_flags & CLONE_SIGHAND) {
 847                 atomic_inc(&current->sighand->count);
 848                 return 0;
 849         }
 850         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 851         rcu_assign_pointer(tsk->sighand, sig);
 852         if (!sig)
 853                 return -ENOMEM;
 854         atomic_set(&sig->count, 1);
 855         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 856         return 0;
 857 }
 858
 859 void __cleanup_sighand(struct sighand_struct *sighand)
 860 {
 861         if (atomic_dec_and_test(&sighand->count))
 862                 kmem_cache_free(sighand_cachep, sighand);
 863 }
 864
 865
 866 /*
 867  * Initialize POSIX timer handling for a thread group.
 868  */
 869 static void posix_cpu_timers_init_group(struct signal_struct *sig)
 870 {
 871         unsigned long cpu_limit;
 872
 873         /* Thread group counters. */
 874         thread_group_cputime_init(sig);
 875
 876         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
 877         if (cpu_limit != RLIM_INFINITY) {
 878                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
 879                 sig->cputimer.running = 1;
 880         }
 881
 882         /* The timer lists. */
 883         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 884         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 885         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 886 }
 887
 888 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 889 {
 890         struct signal_struct *sig;
 891
 892         if (clone_flags & CLONE_THREAD)
 893                 return 0;
 894
 895         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
 896         tsk->signal = sig;
 897         if (!sig)
 898                 return -ENOMEM;
 899
 900         sig->nr_threads = 1;
 901         atomic_set(&sig->live, 1);
 902         atomic_set(&sig->sigcnt, 1);
 903         init_waitqueue_head(&sig->wait_chldexit);
 904         if (clone_flags & CLONE_NEWPID)
 905                 sig->flags |= SIGNAL_UNKILLABLE;
 906         sig->curr_target = tsk;
 907         init_sigpending(&sig->shared_pending);
 908         INIT_LIST_HEAD(&sig->posix_timers);
 909
 910         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 911         sig->real_timer.function = it_real_fn;
 912
 913         task_lock(current->group_leader);
 914         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 915         task_unlock(current->group_leader);
 916
 917         posix_cpu_timers_init_group(sig);
 918
 919         tty_audit_fork(sig);
 920
 921         sig->oom_adj = current->signal->oom_adj;
 922         sig->oom_score_adj = current->signal->oom_score_adj;
 923
 924         return 0;
 925 }
 926
 927 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 928 {
 929         unsigned long new_flags = p->flags;
 930
 931         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
 932         new_flags |= PF_FORKNOEXEC;
 933         new_flags |= PF_STARTING;
 934         p->flags = new_flags;
 935         clear_freeze_flag(p);
 936 }
 937
 938 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
 939 {
 940         current->clear_child_tid = tidptr;
 941
 942         return task_pid_vnr(current);
 943 }
 944
 945 static void rt_mutex_init_task(struct task_struct *p)
 946 {
 947         raw_spin_lock_init(&p->pi_lock);
 948 #ifdef CONFIG_RT_MUTEXES
 949         plist_head_init_raw(&p->pi_waiters, &p->pi_lock);
 950         p->pi_blocked_on = NULL;
 951 #endif
 952 }
 953
 954 #ifdef CONFIG_MM_OWNER
 955 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 956 {
 957         mm->owner = p;
 958 }
 959 #endif /* CONFIG_MM_OWNER */
 960
 961 /*
 962  * Initialize POSIX timer handling for a single task.
 963  */
 964 static void posix_cpu_timers_init(struct task_struct *tsk)
 965 {
 966         tsk->cputime_expires.prof_exp = cputime_zero;
 967         tsk->cputime_expires.virt_exp = cputime_zero;
 968         tsk->cputime_expires.sched_exp = 0;
 969         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
 970         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
 971         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
 972 }
 973
 974 /*
 975  * This creates a new process as a copy of the old one,
 976  * but does not actually start it yet.
 977  *
 978  * It copies the registers, and all the appropriate
 979  * parts of the process environment (as per the clone
 980  * flags). The actual kick-off is left to the caller.
 981  */
 982 static struct task_struct *copy_process(unsigned long clone_flags,
 983                                         unsigned long stack_start,
 984                                         struct pt_regs *regs,
 985                                         unsigned long stack_size,
 986                                         int __user *child_tidptr,
 987                                         struct pid *pid,
 988                                         int trace)
 989 {
 990         int retval;
 991         struct task_struct *p;
 992         int cgroup_callbacks_done = 0;
 993
 994         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
 995                 return ERR_PTR(-EINVAL);
 996
 997         /*
 998          * Thread groups must share signals as well, and detached threads
 999          * can only be started up within the thread group.
1000          */
1001         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1002                 return ERR_PTR(-EINVAL);
1003
1004         /*
1005          * Shared signal handlers imply shared VM. By way of the above,
1006          * thread groups also imply shared VM. Blocking this case allows
1007          * for various simplifications in other code.
1008          */
1009         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1010                 return ERR_PTR(-EINVAL);
1011
1012         /*
1013          * Siblings of global init remain as zombies on exit since they are
1014          * not reaped by their parent (swapper). To solve this and to avoid
1015          * multi-rooted process trees, prevent global and container-inits
1016          * from creating siblings.
1017          */
1018         if ((clone_flags & CLONE_PARENT) &&
1019                                 current->signal->flags & SIGNAL_UNKILLABLE)
1020                 return ERR_PTR(-EINVAL);
1021
1022         retval = security_task_create(clone_flags);
1023         if (retval)
1024                 goto fork_out;
1025
1026         retval = -ENOMEM;
1027         p = dup_task_struct(current);
1028         if (!p)
1029                 goto fork_out;
1030
1031         ftrace_graph_init_task(p);
1032
1033         rt_mutex_init_task(p);
1034
1035 #ifdef CONFIG_PROVE_LOCKING
1036         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1037         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1038 #endif
1039         retval = -EAGAIN;
1040         if (atomic_read(&p->real_cred->user->processes) >=
1041                         task_rlimit(p, RLIMIT_NPROC)) {
1042                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1043                     p->real_cred->user != INIT_USER)
1044                         goto bad_fork_free;
1045         }
1046
1047         retval = copy_creds(p, clone_flags);
1048         if (retval < 0)
1049                 goto bad_fork_free;
1050
1051         /*
1052          * If multiple threads are within copy_process(), then this check
1053          * triggers too late. This doesn't hurt, the check is only there
1054          * to stop root fork bombs.
1055          */
1056         retval = -EAGAIN;
1057         if (nr_threads >= max_threads)
1058                 goto bad_fork_cleanup_count;
1059
1060         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1061                 goto bad_fork_cleanup_count;
1062
1063         p->did_exec = 0;
1064         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1065         copy_flags(clone_flags, p);
1066         INIT_LIST_HEAD(&p->children);
1067         INIT_LIST_HEAD(&p->sibling);
1068         rcu_copy_process(p);
1069         p->vfork_done = NULL;
1070         spin_lock_init(&p->alloc_lock);
1071
1072         init_sigpending(&p->pending);
1073
1074         p->utime = cputime_zero;
1075         p->stime = cputime_zero;
1076         p->gtime = cputime_zero;
1077         p->utimescaled = cputime_zero;
1078         p->stimescaled = cputime_zero;
1079 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1080         p->prev_utime = cputime_zero;
1081         p->prev_stime = cputime_zero;
1082 #endif
1083 #if defined(SPLIT_RSS_COUNTING)
1084         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1085 #endif
1086
1087         p->default_timer_slack_ns = current->timer_slack_ns;
1088
1089         task_io_accounting_init(&p->ioac);
1090         acct_clear_integrals(p);
1091
1092         posix_cpu_timers_init(p);
1093
1094         p->lock_depth = -1;             /* -1 = no lock */
1095         do_posix_clock_monotonic_gettime(&p->start_time);
1096         p->real_start_time = p->start_time;
1097         monotonic_to_bootbased(&p->real_start_time);
1098         p->io_context = NULL;
1099         p->audit_context = NULL;
1100         cgroup_fork(p);
1101 #ifdef CONFIG_NUMA
1102         p->mempolicy = mpol_dup(p->mempolicy);
1103         if (IS_ERR(p->mempolicy)) {
1104                 retval = PTR_ERR(p->mempolicy);
1105                 p->mempolicy = NULL;
1106                 goto bad_fork_cleanup_cgroup;
1107         }
1108         mpol_fix_fork_child_flag(p);
1109 #endif
1110 #ifdef CONFIG_TRACE_IRQFLAGS
1111         p->irq_events = 0;
1112 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1113         p->hardirqs_enabled = 1;
1114 #else
1115         p->hardirqs_enabled = 0;
1116 #endif
1117         p->hardirq_enable_ip = 0;
1118         p->hardirq_enable_event = 0;
1119         p->hardirq_disable_ip = _THIS_IP_;
1120         p->hardirq_disable_event = 0;
1121         p->softirqs_enabled = 1;
1122         p->softirq_enable_ip = _THIS_IP_;
1123         p->softirq_enable_event = 0;
1124         p->softirq_disable_ip = 0;
1125         p->softirq_disable_event = 0;
1126         p->hardirq_context = 0;
1127         p->softirq_context = 0;
1128 #endif
1129 #ifdef CONFIG_LOCKDEP
1130         p->lockdep_depth = 0; /* no locks held yet */
1131         p->curr_chain_key = 0;
1132         p->lockdep_recursion = 0;
1133 #endif
1134
1135 #ifdef CONFIG_DEBUG_MUTEXES
1136         p->blocked_on = NULL; /* not blocked yet */
1137 #endif
1138 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1139         p->memcg_batch.do_batch = 0;
1140         p->memcg_batch.memcg = NULL;
1141 #endif
1142
1143         /* Perform scheduler related setup. Assign this task to a CPU. */
1144         sched_fork(p, clone_flags);
1145
1146         retval = perf_event_init_task(p);
1147         if (retval)
1148                 goto bad_fork_cleanup_policy;
1149
1150         if ((retval = audit_alloc(p)))
1151                 goto bad_fork_cleanup_policy;
1152         /* copy all the process information */
1153         if ((retval = copy_semundo(clone_flags, p)))
1154                 goto bad_fork_cleanup_audit;
1155         if ((retval = copy_files(clone_flags, p)))
1156                 goto bad_fork_cleanup_semundo;
1157         if ((retval = copy_fs(clone_flags, p)))
1158                 goto bad_fork_cleanup_files;
1159         if ((retval = copy_sighand(clone_flags, p)))
1160                 goto bad_fork_cleanup_fs;
1161         if ((retval = copy_signal(clone_flags, p)))
1162                 goto bad_fork_cleanup_sighand;
1163         if ((retval = copy_mm(clone_flags, p)))
1164                 goto bad_fork_cleanup_signal;
1165         if ((retval = copy_namespaces(clone_flags, p)))
1166                 goto bad_fork_cleanup_mm;
1167         if ((retval = copy_io(clone_flags, p)))
1168                 goto bad_fork_cleanup_namespaces;
1169         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1170         if (retval)
1171                 goto bad_fork_cleanup_io;
1172
1173         if (pid != &init_struct_pid) {
1174                 retval = -ENOMEM;
1175                 pid = alloc_pid(p->nsproxy->pid_ns);
1176                 if (!pid)
1177                         goto bad_fork_cleanup_io;
1178
1179                 if (clone_flags & CLONE_NEWPID) {
1180                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1181                         if (retval < 0)
1182                                 goto bad_fork_free_pid;
1183                 }
1184         }
1185
1186         p->pid = pid_nr(pid);
1187         p->tgid = p->pid;
1188         if (clone_flags & CLONE_THREAD)
1189                 p->tgid = current->tgid;
1190
1191         if (current->nsproxy != p->nsproxy) {
1192                 retval = ns_cgroup_clone(p, pid);
1193                 if (retval)
1194                         goto bad_fork_free_pid;
1195         }
1196
1197         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1198         /*
1199          * Clear TID on mm_release()?
1200          */
1201         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1202 #ifdef CONFIG_FUTEX
1203         p->robust_list = NULL;
1204 #ifdef CONFIG_COMPAT
1205         p->compat_robust_list = NULL;
1206 #endif
1207         INIT_LIST_HEAD(&p->pi_state_list);
1208         p->pi_state_cache = NULL;
1209 #endif
1210         /*
1211          * sigaltstack should be cleared when sharing the same VM
1212          */
1213         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1214                 p->sas_ss_sp = p->sas_ss_size = 0;
1215
1216         /*
1217          * Syscall tracing and stepping should be turned off in the
1218          * child regardless of CLONE_PTRACE.
1219          */
1220         user_disable_single_step(p);
1221         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1222 #ifdef TIF_SYSCALL_EMU
1223         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1224 #endif
1225         clear_all_latency_tracing(p);
1226
1227         /* ok, now we should be set up.. */
1228         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1229         p->pdeath_signal = 0;
1230         p->exit_state = 0;
1231
1232         /*
1233          * Ok, make it visible to the rest of the system.
1234          * We dont wake it up yet.
1235          */
1236         p->group_leader = p;
1237         INIT_LIST_HEAD(&p->thread_group);
1238
1239         /* Now that the task is set up, run cgroup callbacks if
1240          * necessary. We need to run them before the task is visible
1241          * on the tasklist. */
1242         cgroup_fork_callbacks(p);
1243         cgroup_callbacks_done = 1;
1244
1245         /* Need tasklist lock for parent etc handling! */
1246         write_lock_irq(&tasklist_lock);
1247
1248         /* CLONE_PARENT re-uses the old parent */
1249         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1250                 p->real_parent = current->real_parent;
1251                 p->parent_exec_id = current->parent_exec_id;
1252         } else {
1253                 p->real_parent = current;
1254                 p->parent_exec_id = current->self_exec_id;
1255         }
1256
1257         spin_lock(&current->sighand->siglock);
1258
1259         /*
1260          * Process group and session signals need to be delivered to just the
1261          * parent before the fork or both the parent and the child after the
1262          * fork. Restart if a signal comes in before we add the new process to
1263          * it's process group.
1264          * A fatal signal pending means that current will exit, so the new
1265          * thread can't slip out of an OOM kill (or normal SIGKILL).
1266          */
1267         recalc_sigpending();
1268         if (signal_pending(current)) {
1269                 spin_unlock(&current->sighand->siglock);
1270                 write_unlock_irq(&tasklist_lock);
1271                 retval = -ERESTARTNOINTR;
1272                 goto bad_fork_free_pid;
1273         }
1274
1275         if (clone_flags & CLONE_THREAD) {
1276                 current->signal->nr_threads++;
1277                 atomic_inc(&current->signal->live);
1278                 atomic_inc(&current->signal->sigcnt);
1279                 p->group_leader = current->group_leader;
1280                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1281         }
1282
1283         if (likely(p->pid)) {
1284                 tracehook_finish_clone(p, clone_flags, trace);
1285
1286                 if (thread_group_leader(p)) {
1287                         if (clone_flags & CLONE_NEWPID)
1288                                 p->nsproxy->pid_ns->child_reaper = p;
1289
1290                         p->signal->leader_pid = pid;
1291                         p->signal->tty = tty_kref_get(current->signal->tty);
1292                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1293                         attach_pid(p, PIDTYPE_SID, task_session(current));
1294                         list_add_tail(&p->sibling, &p->real_parent->children);
1295                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1296                         __get_cpu_var(process_counts)++;
1297                 }
1298                 attach_pid(p, PIDTYPE_PID, pid);
1299                 nr_threads++;
1300         }
1301
1302         total_forks++;
1303         spin_unlock(&current->sighand->siglock);
1304         write_unlock_irq(&tasklist_lock);
1305         proc_fork_connector(p);
1306         cgroup_post_fork(p);
1307         perf_event_fork(p);
1308         return p;
1309
1310 bad_fork_free_pid:
1311         if (pid != &init_struct_pid)
1312                 free_pid(pid);
1313 bad_fork_cleanup_io:
1314         if (p->io_context)
1315                 exit_io_context(p);
1316 bad_fork_cleanup_namespaces:
1317         exit_task_namespaces(p);
1318 bad_fork_cleanup_mm:
1319         if (p->mm)
1320                 mmput(p->mm);
1321 bad_fork_cleanup_signal:
1322         if (!(clone_flags & CLONE_THREAD))
1323                 free_signal_struct(p->signal);
1324 bad_fork_cleanup_sighand:
1325         __cleanup_sighand(p->sighand);
1326 bad_fork_cleanup_fs:
1327         exit_fs(p); /* blocking */
1328 bad_fork_cleanup_files:
1329         exit_files(p); /* blocking */
1330 bad_fork_cleanup_semundo:
1331         exit_sem(p);
1332 bad_fork_cleanup_audit:
1333         audit_free(p);
1334 bad_fork_cleanup_policy:
1335         perf_event_free_task(p);
1336 #ifdef CONFIG_NUMA
1337         mpol_put(p->mempolicy);
1338 bad_fork_cleanup_cgroup:
1339 #endif
1340         cgroup_exit(p, cgroup_callbacks_done);
1341         delayacct_tsk_free(p);
1342         module_put(task_thread_info(p)->exec_domain->module);
1343 bad_fork_cleanup_count:
1344         atomic_dec(&p->cred->user->processes);
1345         exit_creds(p);
1346 bad_fork_free:
1347         free_task(p);
1348 fork_out:
1349         return ERR_PTR(retval);
1350 }
1351
1352 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1353 {
1354         memset(regs, 0, sizeof(struct pt_regs));
1355         return regs;
1356 }
1357
1358 static inline void init_idle_pids(struct pid_link *links)
1359 {
1360         enum pid_type type;
1361
1362         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1363                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1364                 links[type].pid = &init_struct_pid;
1365         }
1366 }
1367
1368 struct task_struct * __cpuinit fork_idle(int cpu)
1369 {
1370         struct task_struct *task;
1371         struct pt_regs regs;
1372
1373         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1374                             &init_struct_pid, 0);
1375         if (!IS_ERR(task)) {
1376                 init_idle_pids(task->pids);
1377                 init_idle(task, cpu);
1378         }
1379
1380         return task;
1381 }
1382
1383 /*
1384  *  Ok, this is the main fork-routine.
1385  *
1386  * It copies the process, and if successful kick-starts
1387  * it and waits for it to finish using the VM if required.
1388  */
1389 long do_fork(unsigned long clone_flags,
1390               unsigned long stack_start,
1391               struct pt_regs *regs,
1392               unsigned long stack_size,
1393               int __user *parent_tidptr,
1394               int __user *child_tidptr)
1395 {
1396         struct task_struct *p;
1397         int trace = 0;
1398         long nr;
1399
1400         /*
1401          * Do some preliminary argument and permissions checking before we
1402          * actually start allocating stuff
1403          */
1404         if (clone_flags & CLONE_NEWUSER) {
1405                 if (clone_flags & CLONE_THREAD)
1406                         return -EINVAL;
1407                 /* hopefully this check will go away when userns support is
1408                  * complete
1409                  */
1410                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1411                                 !capable(CAP_SETGID))
1412                         return -EPERM;
1413         }
1414
1415         /*
1416          * We hope to recycle these flags after 2.6.26
1417          */
1418         if (unlikely(clone_flags & CLONE_STOPPED)) {
1419                 static int __read_mostly count = 100;
1420
1421                 if (count > 0 && printk_ratelimit()) {
1422                         char comm[TASK_COMM_LEN];
1423
1424                         count--;
1425                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1426                                         "clone flags 0x%lx\n",
1427                                 get_task_comm(comm, current),
1428                                 clone_flags & CLONE_STOPPED);
1429                 }
1430         }
1431
1432         /*
1433          * When called from kernel_thread, don't do user tracing stuff.
1434          */
1435         if (likely(user_mode(regs)))
1436                 trace = tracehook_prepare_clone(clone_flags);
1437
1438         p = copy_process(clone_flags, stack_start, regs, stack_size,
1439                          child_tidptr, NULL, trace);
1440         /*
1441          * Do this prior waking up the new thread - the thread pointer
1442          * might get invalid after that point, if the thread exits quickly.
1443          */
1444         if (!IS_ERR(p)) {
1445                 struct completion vfork;
1446
1447                 trace_sched_process_fork(current, p);
1448
1449                 nr = task_pid_vnr(p);
1450
1451                 if (clone_flags & CLONE_PARENT_SETTID)
1452                         put_user(nr, parent_tidptr);
1453
1454                 if (clone_flags & CLONE_VFORK) {
1455                         p->vfork_done = &vfork;
1456                         init_completion(&vfork);
1457                 }
1458
1459                 audit_finish_fork(p);
1460                 tracehook_report_clone(regs, clone_flags, nr, p);
1461
1462                 /*
1463                  * We set PF_STARTING at creation in case tracing wants to
1464                  * use this to distinguish a fully live task from one that
1465                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1466                  * clear it and set the child going.
1467                  */
1468                 p->flags &= ~PF_STARTING;
1469
1470                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1471                         /*
1472                          * We'll start up with an immediate SIGSTOP.
1473                          */
1474                         sigaddset(&p->pending.signal, SIGSTOP);
1475                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1476                         __set_task_state(p, TASK_STOPPED);
1477                 } else {
1478                         wake_up_new_task(p, clone_flags);
1479                 }
1480
1481                 tracehook_report_clone_complete(trace, regs,
1482                                                 clone_flags, nr, p);
1483
1484                 if (clone_flags & CLONE_VFORK) {
1485                         freezer_do_not_count();
1486                         wait_for_completion(&vfork);
1487                         freezer_count();
1488                         tracehook_report_vfork_done(p, nr);
1489                 }
1490         } else {
1491                 nr = PTR_ERR(p);
1492         }
1493         return nr;
1494 }
1495
1496 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1497 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1498 #endif
1499
1500 static void sighand_ctor(void *data)
1501 {
1502         struct sighand_struct *sighand = data;
1503
1504         spin_lock_init(&sighand->siglock);
1505         init_waitqueue_head(&sighand->signalfd_wqh);
1506 }
1507
1508 void __init proc_caches_init(void)
1509 {
1510         sighand_cachep = kmem_cache_create("sighand_cache",
1511                         sizeof(struct sighand_struct), 0,
1512                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1513                         SLAB_NOTRACK, sighand_ctor);
1514         signal_cachep = kmem_cache_create("signal_cache",
1515                         sizeof(struct signal_struct), 0,
1516                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1517         files_cachep = kmem_cache_create("files_cache",
1518                         sizeof(struct files_struct), 0,
1519                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1520         fs_cachep = kmem_cache_create("fs_cache",
1521                         sizeof(struct fs_struct), 0,
1522                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1523         mm_cachep = kmem_cache_create("mm_struct",
1524                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1525                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1526         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1527         mmap_init();
1528 }
1529
1530 /*
1531  * Check constraints on flags passed to the unshare system call and
1532  * force unsharing of additional process context as appropriate.
1533  */
1534 static void check_unshare_flags(unsigned long *flags_ptr)
1535 {
1536         /*
1537          * If unsharing a thread from a thread group, must also
1538          * unshare vm.
1539          */
1540         if (*flags_ptr & CLONE_THREAD)
1541                 *flags_ptr |= CLONE_VM;
1542
1543         /*
1544          * If unsharing vm, must also unshare signal handlers.
1545          */
1546         if (*flags_ptr & CLONE_VM)
1547                 *flags_ptr |= CLONE_SIGHAND;
1548
1549         /*
1550          * If unsharing namespace, must also unshare filesystem information.
1551          */
1552         if (*flags_ptr & CLONE_NEWNS)
1553                 *flags_ptr |= CLONE_FS;
1554 }
1555
1556 /*
1557  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1558  */
1559 static int unshare_thread(unsigned long unshare_flags)
1560 {
1561         if (unshare_flags & CLONE_THREAD)
1562                 return -EINVAL;
1563
1564         return 0;
1565 }
1566
1567 /*
1568  * Unshare the filesystem structure if it is being shared
1569  */
1570 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1571 {
1572         struct fs_struct *fs = current->fs;
1573
1574         if (!(unshare_flags & CLONE_FS) || !fs)
1575                 return 0;
1576
1577         /* don't need lock here; in the worst case we'll do useless copy */
1578         if (fs->users == 1)
1579                 return 0;
1580
1581         *new_fsp = copy_fs_struct(fs);
1582         if (!*new_fsp)
1583                 return -ENOMEM;
1584
1585         return 0;
1586 }
1587
1588 /*
1589  * Unsharing of sighand is not supported yet
1590  */
1591 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1592 {
1593         struct sighand_struct *sigh = current->sighand;
1594
1595         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1596                 return -EINVAL;
1597         else
1598                 return 0;
1599 }
1600
1601 /*
1602  * Unshare vm if it is being shared
1603  */
1604 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1605 {
1606         struct mm_struct *mm = current->mm;
1607
1608         if ((unshare_flags & CLONE_VM) &&
1609             (mm && atomic_read(&mm->mm_users) > 1)) {
1610                 return -EINVAL;
1611         }
1612
1613         return 0;
1614 }
1615
1616 /*
1617  * Unshare file descriptor table if it is being shared
1618  */
1619 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1620 {
1621         struct files_struct *fd = current->files;
1622         int error = 0;
1623
1624         if ((unshare_flags & CLONE_FILES) &&
1625             (fd && atomic_read(&fd->count) > 1)) {
1626                 *new_fdp = dup_fd(fd, &error);
1627                 if (!*new_fdp)
1628                         return error;
1629         }
1630
1631         return 0;
1632 }
1633
1634 /*
1635  * unshare allows a process to 'unshare' part of the process
1636  * context which was originally shared using clone.  copy_*
1637  * functions used by do_fork() cannot be used here directly
1638  * because they modify an inactive task_struct that is being
1639  * constructed. Here we are modifying the current, active,
1640  * task_struct.
1641  */
1642 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1643 {
1644         int err = 0;
1645         struct fs_struct *fs, *new_fs = NULL;
1646         struct sighand_struct *new_sigh = NULL;
1647         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1648         struct files_struct *fd, *new_fd = NULL;
1649         struct nsproxy *new_nsproxy = NULL;
1650         int do_sysvsem = 0;
1651
1652         check_unshare_flags(&unshare_flags);
1653
1654         /* Return -EINVAL for all unsupported flags */
1655         err = -EINVAL;
1656         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1657                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1658                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1659                 goto bad_unshare_out;
1660
1661         /*
1662          * CLONE_NEWIPC must also detach from the undolist: after switching
1663          * to a new ipc namespace, the semaphore arrays from the old
1664          * namespace are unreachable.
1665          */
1666         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1667                 do_sysvsem = 1;
1668         if ((err = unshare_thread(unshare_flags)))
1669                 goto bad_unshare_out;
1670         if ((err = unshare_fs(unshare_flags, &new_fs)))
1671                 goto bad_unshare_cleanup_thread;
1672         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1673                 goto bad_unshare_cleanup_fs;
1674         if ((err = unshare_vm(unshare_flags, &new_mm)))
1675                 goto bad_unshare_cleanup_sigh;
1676         if ((err = unshare_fd(unshare_flags, &new_fd)))
1677                 goto bad_unshare_cleanup_vm;
1678         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1679                         new_fs)))
1680                 goto bad_unshare_cleanup_fd;
1681
1682         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1683                 if (do_sysvsem) {
1684                         /*
1685                          * CLONE_SYSVSEM is equivalent to sys_exit().
1686                          */
1687                         exit_sem(current);
1688                 }
1689
1690                 if (new_nsproxy) {
1691                         switch_task_namespaces(current, new_nsproxy);
1692                         new_nsproxy = NULL;
1693                 }
1694
1695                 task_lock(current);
1696
1697                 if (new_fs) {
1698                         fs = current->fs;
1699                         spin_lock(&fs->lock);
1700                         current->fs = new_fs;
1701                         if (--fs->users)
1702                                 new_fs = NULL;
1703                         else
1704                                 new_fs = fs;
1705                         spin_unlock(&fs->lock);
1706                 }
1707
1708                 if (new_mm) {
1709                         mm = current->mm;
1710                         active_mm = current->active_mm;
1711                         current->mm = new_mm;
1712                         current->active_mm = new_mm;
1713                         activate_mm(active_mm, new_mm);
1714                         new_mm = mm;
1715                 }
1716
1717                 if (new_fd) {
1718                         fd = current->files;
1719                         current->files = new_fd;
1720                         new_fd = fd;
1721                 }
1722
1723                 task_unlock(current);
1724         }
1725
1726         if (new_nsproxy)
1727                 put_nsproxy(new_nsproxy);
1728
1729 bad_unshare_cleanup_fd:
1730         if (new_fd)
1731                 put_files_struct(new_fd);
1732
1733 bad_unshare_cleanup_vm:
1734         if (new_mm)
1735                 mmput(new_mm);
1736
1737 bad_unshare_cleanup_sigh:
1738         if (new_sigh)
1739                 if (atomic_dec_and_test(&new_sigh->count))
1740                         kmem_cache_free(sighand_cachep, new_sigh);
1741
1742 bad_unshare_cleanup_fs:
1743         if (new_fs)
1744                 free_fs_struct(new_fs);
1745
1746 bad_unshare_cleanup_thread:
1747 bad_unshare_out:
1748         return err;
1749 }
1750
1751 /*
1752  *      Helper to unshare the files of the current task.
1753  *      We don't want to expose copy_files internals to
1754  *      the exec layer of the kernel.
1755  */
1756
1757 int unshare_files(struct files_struct **displaced)
1758 {
1759         struct task_struct *task = current;
1760         struct files_struct *copy = NULL;
1761         int error;
1762
1763         error = unshare_fd(CLONE_FILES, &copy);
1764         if (error || !copy) {
1765                 *displaced = NULL;
1766                 return error;
1767         }
1768         *displaced = task->files;
1769         task_lock(task);
1770         task->files = copy;
1771         task_unlock(task);
1772         return 0;
1773 }