kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/nsproxy.h>
  32 #include <linux/capability.h>
  33 #include <linux/cpu.h>
  34 #include <linux/cgroup.h>
  35 #include <linux/security.h>
  36 #include <linux/hugetlb.h>
  37 #include <linux/swap.h>
  38 #include <linux/syscalls.h>
  39 #include <linux/jiffies.h>
  40 #include <linux/futex.h>
  41 #include <linux/compat.h>
  42 #include <linux/kthread.h>
  43 #include <linux/task_io_accounting_ops.h>
  44 #include <linux/rcupdate.h>
  45 #include <linux/ptrace.h>
  46 #include <linux/mount.h>
  47 #include <linux/audit.h>
  48 #include <linux/memcontrol.h>
  49 #include <linux/ftrace.h>
  50 #include <linux/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/blkdev.h>
  62 #include <linux/fs_struct.h>
  63 #include <linux/magic.h>
  64 #include <linux/perf_event.h>
  65 #include <linux/posix-timers.h>
  66 #include <linux/user-return-notifier.h>
  67 #include <linux/oom.h>
  68 #include <linux/khugepaged.h>
  69 #include <linux/signalfd.h>
  70
  71 #include <asm/pgtable.h>
  72 #include <asm/pgalloc.h>
  73 #include <asm/uaccess.h>
  74 #include <asm/mmu_context.h>
  75 #include <asm/cacheflush.h>
  76 #include <asm/tlbflush.h>
  77
  78 #include <trace/events/sched.h>
  79
  80 #define CREATE_TRACE_POINTS
  81 #include <trace/events/task.h>
  82
  83 /*
  84  * Protected counters by write_lock_irq(&tasklist_lock)
  85  */
  86 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  87 int nr_threads;                 /* The idle threads do not count.. */
  88
  89 int max_threads;                /* tunable limit on nr_threads */
  90
  91 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  92
  93 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  94
  95 #ifdef CONFIG_PROVE_RCU
  96 int lockdep_tasklist_lock_is_held(void)
  97 {
  98         return lockdep_is_held(&tasklist_lock);
  99 }
 100 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 101 #endif /* #ifdef CONFIG_PROVE_RCU */
 102
 103 int nr_processes(void)
 104 {
 105         int cpu;
 106         int total = 0;
 107
 108         for_each_possible_cpu(cpu)
 109                 total += per_cpu(process_counts, cpu);
 110
 111         return total;
 112 }
 113
 114 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 115 # define alloc_task_struct_node(node)           \
 116                 kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node)
 117 # define free_task_struct(tsk)                  \
 118                 kmem_cache_free(task_struct_cachep, (tsk))
 119 static struct kmem_cache *task_struct_cachep;
 120 #endif
 121
 122 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 123 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 124                                                   int node)
 125 {
 126 #ifdef CONFIG_DEBUG_STACK_USAGE
 127         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 128 #else
 129         gfp_t mask = GFP_KERNEL;
 130 #endif
 131         struct page *page = alloc_pages_node(node, mask, THREAD_SIZE_ORDER);
 132
 133         return page ? page_address(page) : NULL;
 134 }
 135
 136 static inline void free_thread_info(struct thread_info *ti)
 137 {
 138         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 139 }
 140 #endif
 141
 142 /* SLAB cache for signal_struct structures (tsk->signal) */
 143 static struct kmem_cache *signal_cachep;
 144
 145 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 146 struct kmem_cache *sighand_cachep;
 147
 148 /* SLAB cache for files_struct structures (tsk->files) */
 149 struct kmem_cache *files_cachep;
 150
 151 /* SLAB cache for fs_struct structures (tsk->fs) */
 152 struct kmem_cache *fs_cachep;
 153
 154 /* SLAB cache for vm_area_struct structures */
 155 struct kmem_cache *vm_area_cachep;
 156
 157 /* SLAB cache for mm_struct structures (tsk->mm) */
 158 static struct kmem_cache *mm_cachep;
 159
 160 static void account_kernel_stack(struct thread_info *ti, int account)
 161 {
 162         struct zone *zone = page_zone(virt_to_page(ti));
 163
 164         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 165 }
 166
 167 void free_task(struct task_struct *tsk)
 168 {
 169         account_kernel_stack(tsk->stack, -1);
 170         free_thread_info(tsk->stack);
 171         rt_mutex_debug_task_free(tsk);
 172         ftrace_graph_exit_task(tsk);
 173         free_task_struct(tsk);
 174 }
 175 EXPORT_SYMBOL(free_task);
 176
 177 static inline void free_signal_struct(struct signal_struct *sig)
 178 {
 179         taskstats_tgid_free(sig);
 180         sched_autogroup_exit(sig);
 181         kmem_cache_free(signal_cachep, sig);
 182 }
 183
 184 static inline void put_signal_struct(struct signal_struct *sig)
 185 {
 186         if (atomic_dec_and_test(&sig->sigcnt))
 187                 free_signal_struct(sig);
 188 }
 189
 190 void __put_task_struct(struct task_struct *tsk)
 191 {
 192         WARN_ON(!tsk->exit_state);
 193         WARN_ON(atomic_read(&tsk->usage));
 194         WARN_ON(tsk == current);
 195
 196         exit_creds(tsk);
 197         delayacct_tsk_free(tsk);
 198         put_signal_struct(tsk->signal);
 199
 200         if (!profile_handoff_task(tsk))
 201                 free_task(tsk);
 202 }
 203 EXPORT_SYMBOL_GPL(__put_task_struct);
 204
 205 /*
 206  * macro override instead of weak attribute alias, to workaround
 207  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 208  */
 209 #ifndef arch_task_cache_init
 210 #define arch_task_cache_init()
 211 #endif
 212
 213 void __init fork_init(unsigned long mempages)
 214 {
 215 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 216 #ifndef ARCH_MIN_TASKALIGN
 217 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 218 #endif
 219         /* create a slab on which task_structs can be allocated */
 220         task_struct_cachep =
 221                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 222                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 223 #endif
 224
 225         /* do the arch specific task caches init */
 226         arch_task_cache_init();
 227
 228         /*
 229          * The default maximum number of threads is set to a safe
 230          * value: the thread structures can take up at most half
 231          * of memory.
 232          */
 233         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 234
 235         /*
 236          * we need to allow at least 20 threads to boot a system
 237          */
 238         if (max_threads < 20)
 239                 max_threads = 20;
 240
 241         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 242         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 243         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 244                 init_task.signal->rlim[RLIMIT_NPROC];
 245 }
 246
 247 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 248                                                struct task_struct *src)
 249 {
 250         *dst = *src;
 251         return 0;
 252 }
 253
 254 static struct task_struct *dup_task_struct(struct task_struct *orig)
 255 {
 256         struct task_struct *tsk;
 257         struct thread_info *ti;
 258         unsigned long *stackend;
 259         int node = tsk_fork_get_node(orig);
 260         int err;
 261
 262         prepare_to_copy(orig);
 263
 264         tsk = alloc_task_struct_node(node);
 265         if (!tsk)
 266                 return NULL;
 267
 268         ti = alloc_thread_info_node(tsk, node);
 269         if (!ti) {
 270                 free_task_struct(tsk);
 271                 return NULL;
 272         }
 273
 274         err = arch_dup_task_struct(tsk, orig);
 275         if (err)
 276                 goto out;
 277
 278         tsk->stack = ti;
 279
 280         setup_thread_stack(tsk, orig);
 281         clear_user_return_notifier(tsk);
 282         clear_tsk_need_resched(tsk);
 283         stackend = end_of_stack(tsk);
 284         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 285
 286 #ifdef CONFIG_CC_STACKPROTECTOR
 287         tsk->stack_canary = get_random_int();
 288 #endif
 289
 290         /*
 291          * One for us, one for whoever does the "release_task()" (usually
 292          * parent)
 293          */
 294         atomic_set(&tsk->usage, 2);
 295 #ifdef CONFIG_BLK_DEV_IO_TRACE
 296         tsk->btrace_seq = 0;
 297 #endif
 298         tsk->splice_pipe = NULL;
 299
 300         account_kernel_stack(ti, 1);
 301
 302         return tsk;
 303
 304 out:
 305         free_thread_info(ti);
 306         free_task_struct(tsk);
 307         return NULL;
 308 }
 309
 310 #ifdef CONFIG_MMU
 311 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 312 {
 313         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 314         struct rb_node **rb_link, *rb_parent;
 315         int retval;
 316         unsigned long charge;
 317         struct mempolicy *pol;
 318
 319         down_write(&oldmm->mmap_sem);
 320         flush_cache_dup_mm(oldmm);
 321         /*
 322          * Not linked in yet - no deadlock potential:
 323          */
 324         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 325
 326         mm->locked_vm = 0;
 327         mm->mmap = NULL;
 328         mm->mmap_cache = NULL;
 329         mm->free_area_cache = oldmm->mmap_base;
 330         mm->cached_hole_size = ~0UL;
 331         mm->map_count = 0;
 332         cpumask_clear(mm_cpumask(mm));
 333         mm->mm_rb = RB_ROOT;
 334         rb_link = &mm->mm_rb.rb_node;
 335         rb_parent = NULL;
 336         pprev = &mm->mmap;
 337         retval = ksm_fork(mm, oldmm);
 338         if (retval)
 339                 goto out;
 340         retval = khugepaged_fork(mm, oldmm);
 341         if (retval)
 342                 goto out;
 343
 344         prev = NULL;
 345         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 346                 struct file *file;
 347
 348                 if (mpnt->vm_flags & VM_DONTCOPY) {
 349                         long pages = vma_pages(mpnt);
 350                         mm->total_vm -= pages;
 351                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 352                                                                 -pages);
 353                         continue;
 354                 }
 355                 charge = 0;
 356                 if (mpnt->vm_flags & VM_ACCOUNT) {
 357                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 358                         if (security_vm_enough_memory(len))
 359                                 goto fail_nomem;
 360                         charge = len;
 361                 }
 362                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 363                 if (!tmp)
 364                         goto fail_nomem;
 365                 *tmp = *mpnt;
 366                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 367                 pol = mpol_dup(vma_policy(mpnt));
 368                 retval = PTR_ERR(pol);
 369                 if (IS_ERR(pol))
 370                         goto fail_nomem_policy;
 371                 vma_set_policy(tmp, pol);
 372                 tmp->vm_mm = mm;
 373                 if (anon_vma_fork(tmp, mpnt))
 374                         goto fail_nomem_anon_vma_fork;
 375                 tmp->vm_flags &= ~VM_LOCKED;
 376                 tmp->vm_next = tmp->vm_prev = NULL;
 377                 file = tmp->vm_file;
 378                 if (file) {
 379                         struct inode *inode = file->f_path.dentry->d_inode;
 380                         struct address_space *mapping = file->f_mapping;
 381
 382                         get_file(file);
 383                         if (tmp->vm_flags & VM_DENYWRITE)
 384                                 atomic_dec(&inode->i_writecount);
 385                         mutex_lock(&mapping->i_mmap_mutex);
 386                         if (tmp->vm_flags & VM_SHARED)
 387                                 mapping->i_mmap_writable++;
 388                         flush_dcache_mmap_lock(mapping);
 389                         /* insert tmp into the share list, just after mpnt */
 390                         vma_prio_tree_add(tmp, mpnt);
 391                         flush_dcache_mmap_unlock(mapping);
 392                         mutex_unlock(&mapping->i_mmap_mutex);
 393                 }
 394
 395                 /*
 396                  * Clear hugetlb-related page reserves for children. This only
 397                  * affects MAP_PRIVATE mappings. Faults generated by the child
 398                  * are not guaranteed to succeed, even if read-only
 399                  */
 400                 if (is_vm_hugetlb_page(tmp))
 401                         reset_vma_resv_huge_pages(tmp);
 402
 403                 /*
 404                  * Link in the new vma and copy the page table entries.
 405                  */
 406                 *pprev = tmp;
 407                 pprev = &tmp->vm_next;
 408                 tmp->vm_prev = prev;
 409                 prev = tmp;
 410
 411                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 412                 rb_link = &tmp->vm_rb.rb_right;
 413                 rb_parent = &tmp->vm_rb;
 414
 415                 mm->map_count++;
 416                 retval = copy_page_range(mm, oldmm, mpnt);
 417
 418                 if (tmp->vm_ops && tmp->vm_ops->open)
 419                         tmp->vm_ops->open(tmp);
 420
 421                 if (retval)
 422                         goto out;
 423         }
 424         /* a new mm has just been created */
 425         arch_dup_mmap(oldmm, mm);
 426         retval = 0;
 427 out:
 428         up_write(&mm->mmap_sem);
 429         flush_tlb_mm(oldmm);
 430         up_write(&oldmm->mmap_sem);
 431         return retval;
 432 fail_nomem_anon_vma_fork:
 433         mpol_put(pol);
 434 fail_nomem_policy:
 435         kmem_cache_free(vm_area_cachep, tmp);
 436 fail_nomem:
 437         retval = -ENOMEM;
 438         vm_unacct_memory(charge);
 439         goto out;
 440 }
 441
 442 static inline int mm_alloc_pgd(struct mm_struct *mm)
 443 {
 444         mm->pgd = pgd_alloc(mm);
 445         if (unlikely(!mm->pgd))
 446                 return -ENOMEM;
 447         return 0;
 448 }
 449
 450 static inline void mm_free_pgd(struct mm_struct *mm)
 451 {
 452         pgd_free(mm, mm->pgd);
 453 }
 454 #else
 455 #define dup_mmap(mm, oldmm)     (0)
 456 #define mm_alloc_pgd(mm)        (0)
 457 #define mm_free_pgd(mm)
 458 #endif /* CONFIG_MMU */
 459
 460 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 461
 462 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 463 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 464
 465 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 466
 467 static int __init coredump_filter_setup(char *s)
 468 {
 469         default_dump_filter =
 470                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 471                 MMF_DUMP_FILTER_MASK;
 472         return 1;
 473 }
 474
 475 __setup("coredump_filter=", coredump_filter_setup);
 476
 477 #include <linux/init_task.h>
 478
 479 static void mm_init_aio(struct mm_struct *mm)
 480 {
 481 #ifdef CONFIG_AIO
 482         spin_lock_init(&mm->ioctx_lock);
 483         INIT_HLIST_HEAD(&mm->ioctx_list);
 484 #endif
 485 }
 486
 487 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 488 {
 489         atomic_set(&mm->mm_users, 1);
 490         atomic_set(&mm->mm_count, 1);
 491         init_rwsem(&mm->mmap_sem);
 492         INIT_LIST_HEAD(&mm->mmlist);
 493         mm->flags = (current->mm) ?
 494                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
 495         mm->core_state = NULL;
 496         mm->nr_ptes = 0;
 497         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 498         spin_lock_init(&mm->page_table_lock);
 499         mm->free_area_cache = TASK_UNMAPPED_BASE;
 500         mm->cached_hole_size = ~0UL;
 501         mm_init_aio(mm);
 502         mm_init_owner(mm, p);
 503
 504         if (likely(!mm_alloc_pgd(mm))) {
 505                 mm->def_flags = 0;
 506                 mmu_notifier_mm_init(mm);
 507                 return mm;
 508         }
 509
 510         free_mm(mm);
 511         return NULL;
 512 }
 513
 514 /*
 515  * Allocate and initialize an mm_struct.
 516  */
 517 struct mm_struct *mm_alloc(void)
 518 {
 519         struct mm_struct *mm;
 520
 521         mm = allocate_mm();
 522         if (!mm)
 523                 return NULL;
 524
 525         memset(mm, 0, sizeof(*mm));
 526         mm_init_cpumask(mm);
 527         return mm_init(mm, current);
 528 }
 529
 530 /*
 531  * Called when the last reference to the mm
 532  * is dropped: either by a lazy thread or by
 533  * mmput. Free the page directory and the mm.
 534  */
 535 void __mmdrop(struct mm_struct *mm)
 536 {
 537         BUG_ON(mm == &init_mm);
 538         mm_free_pgd(mm);
 539         destroy_context(mm);
 540         mmu_notifier_mm_destroy(mm);
 541 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 542         VM_BUG_ON(mm->pmd_huge_pte);
 543 #endif
 544         free_mm(mm);
 545 }
 546 EXPORT_SYMBOL_GPL(__mmdrop);
 547
 548 /*
 549  * Decrement the use count and release all resources for an mm.
 550  */
 551 void mmput(struct mm_struct *mm)
 552 {
 553         might_sleep();
 554
 555         if (atomic_dec_and_test(&mm->mm_users)) {
 556                 exit_aio(mm);
 557                 ksm_exit(mm);
 558                 khugepaged_exit(mm); /* must run before exit_mmap */
 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  * We added or removed a vma mapping the executable. The vmas are only mapped
 576  * during exec and are not mapped with the mmap system call.
 577  * Callers must hold down_write() on the mm's mmap_sem for these
 578  */
 579 void added_exe_file_vma(struct mm_struct *mm)
 580 {
 581         mm->num_exe_file_vmas++;
 582 }
 583
 584 void removed_exe_file_vma(struct mm_struct *mm)
 585 {
 586         mm->num_exe_file_vmas--;
 587         if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
 588                 fput(mm->exe_file);
 589                 mm->exe_file = NULL;
 590         }
 591
 592 }
 593
 594 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 595 {
 596         if (new_exe_file)
 597                 get_file(new_exe_file);
 598         if (mm->exe_file)
 599                 fput(mm->exe_file);
 600         mm->exe_file = new_exe_file;
 601         mm->num_exe_file_vmas = 0;
 602 }
 603
 604 struct file *get_mm_exe_file(struct mm_struct *mm)
 605 {
 606         struct file *exe_file;
 607
 608         /* We need mmap_sem to protect against races with removal of
 609          * VM_EXECUTABLE vmas */
 610         down_read(&mm->mmap_sem);
 611         exe_file = mm->exe_file;
 612         if (exe_file)
 613                 get_file(exe_file);
 614         up_read(&mm->mmap_sem);
 615         return exe_file;
 616 }
 617
 618 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 619 {
 620         /* It's safe to write the exe_file pointer without exe_file_lock because
 621          * this is called during fork when the task is not yet in /proc */
 622         newmm->exe_file = get_mm_exe_file(oldmm);
 623 }
 624
 625 /**
 626  * get_task_mm - acquire a reference to the task's mm
 627  *
 628  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 629  * this kernel workthread has transiently adopted a user mm with use_mm,
 630  * to do its AIO) is not set and if so returns a reference to it, after
 631  * bumping up the use count.  User must release the mm via mmput()
 632  * after use.  Typically used by /proc and ptrace.
 633  */
 634 struct mm_struct *get_task_mm(struct task_struct *task)
 635 {
 636         struct mm_struct *mm;
 637
 638         task_lock(task);
 639         mm = task->mm;
 640         if (mm) {
 641                 if (task->flags & PF_KTHREAD)
 642                         mm = NULL;
 643                 else
 644                         atomic_inc(&mm->mm_users);
 645         }
 646         task_unlock(task);
 647         return mm;
 648 }
 649 EXPORT_SYMBOL_GPL(get_task_mm);
 650
 651 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 652 {
 653         struct mm_struct *mm;
 654         int err;
 655
 656         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 657         if (err)
 658                 return ERR_PTR(err);
 659
 660         mm = get_task_mm(task);
 661         if (mm && mm != current->mm &&
 662                         !ptrace_may_access(task, mode)) {
 663                 mmput(mm);
 664                 mm = ERR_PTR(-EACCES);
 665         }
 666         mutex_unlock(&task->signal->cred_guard_mutex);
 667
 668         return mm;
 669 }
 670
 671 void complete_vfork_done(struct task_struct *tsk)
 672 {
 673         struct completion *vfork_done = tsk->vfork_done;
 674
 675         tsk->vfork_done = NULL;
 676         complete(vfork_done);
 677 }
 678
 679 /* Please note the differences between mmput and mm_release.
 680  * mmput is called whenever we stop holding onto a mm_struct,
 681  * error success whatever.
 682  *
 683  * mm_release is called after a mm_struct has been removed
 684  * from the current process.
 685  *
 686  * This difference is important for error handling, when we
 687  * only half set up a mm_struct for a new process and need to restore
 688  * the old one.  Because we mmput the new mm_struct before
 689  * restoring the old one. . .
 690  * Eric Biederman 10 January 1998
 691  */
 692 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 693 {
 694         /* Get rid of any futexes when releasing the mm */
 695 #ifdef CONFIG_FUTEX
 696         if (unlikely(tsk->robust_list)) {
 697                 exit_robust_list(tsk);
 698                 tsk->robust_list = NULL;
 699         }
 700 #ifdef CONFIG_COMPAT
 701         if (unlikely(tsk->compat_robust_list)) {
 702                 compat_exit_robust_list(tsk);
 703                 tsk->compat_robust_list = NULL;
 704         }
 705 #endif
 706         if (unlikely(!list_empty(&tsk->pi_state_list)))
 707                 exit_pi_state_list(tsk);
 708 #endif
 709
 710         /* Get rid of any cached register state */
 711         deactivate_mm(tsk, mm);
 712
 713         if (tsk->vfork_done)
 714                 complete_vfork_done(tsk);
 715
 716         /*
 717          * If we're exiting normally, clear a user-space tid field if
 718          * requested.  We leave this alone when dying by signal, to leave
 719          * the value intact in a core dump, and to save the unnecessary
 720          * trouble otherwise.  Userland only wants this done for a sys_exit.
 721          */
 722         if (tsk->clear_child_tid) {
 723                 if (!(tsk->flags & PF_SIGNALED) &&
 724                     atomic_read(&mm->mm_users) > 1) {
 725                         /*
 726                          * We don't check the error code - if userspace has
 727                          * not set up a proper pointer then tough luck.
 728                          */
 729                         put_user(0, tsk->clear_child_tid);
 730                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 731                                         1, NULL, NULL, 0);
 732                 }
 733                 tsk->clear_child_tid = NULL;
 734         }
 735 }
 736
 737 /*
 738  * Allocate a new mm structure and copy contents from the
 739  * mm structure of the passed in task structure.
 740  */
 741 struct mm_struct *dup_mm(struct task_struct *tsk)
 742 {
 743         struct mm_struct *mm, *oldmm = current->mm;
 744         int err;
 745
 746         if (!oldmm)
 747                 return NULL;
 748
 749         mm = allocate_mm();
 750         if (!mm)
 751                 goto fail_nomem;
 752
 753         memcpy(mm, oldmm, sizeof(*mm));
 754         mm_init_cpumask(mm);
 755
 756         /* Initializing for Swap token stuff */
 757         mm->token_priority = 0;
 758         mm->last_interval = 0;
 759
 760 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 761         mm->pmd_huge_pte = NULL;
 762 #endif
 763
 764         if (!mm_init(mm, tsk))
 765                 goto fail_nomem;
 766
 767         if (init_new_context(tsk, mm))
 768                 goto fail_nocontext;
 769
 770         dup_mm_exe_file(oldmm, mm);
 771
 772         err = dup_mmap(mm, oldmm);
 773         if (err)
 774                 goto free_pt;
 775
 776         mm->hiwater_rss = get_mm_rss(mm);
 777         mm->hiwater_vm = mm->total_vm;
 778
 779         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 780                 goto free_pt;
 781
 782         return mm;
 783
 784 free_pt:
 785         /* don't put binfmt in mmput, we haven't got module yet */
 786         mm->binfmt = NULL;
 787         mmput(mm);
 788
 789 fail_nomem:
 790         return NULL;
 791
 792 fail_nocontext:
 793         /*
 794          * If init_new_context() failed, we cannot use mmput() to free the mm
 795          * because it calls destroy_context()
 796          */
 797         mm_free_pgd(mm);
 798         free_mm(mm);
 799         return NULL;
 800 }
 801
 802 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 803 {
 804         struct mm_struct *mm, *oldmm;
 805         int retval;
 806
 807         tsk->min_flt = tsk->maj_flt = 0;
 808         tsk->nvcsw = tsk->nivcsw = 0;
 809 #ifdef CONFIG_DETECT_HUNG_TASK
 810         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 811 #endif
 812
 813         tsk->mm = NULL;
 814         tsk->active_mm = NULL;
 815
 816         /*
 817          * Are we cloning a kernel thread?
 818          *
 819          * We need to steal a active VM for that..
 820          */
 821         oldmm = current->mm;
 822         if (!oldmm)
 823                 return 0;
 824
 825         if (clone_flags & CLONE_VM) {
 826                 atomic_inc(&oldmm->mm_users);
 827                 mm = oldmm;
 828                 goto good_mm;
 829         }
 830
 831         retval = -ENOMEM;
 832         mm = dup_mm(tsk);
 833         if (!mm)
 834                 goto fail_nomem;
 835
 836 good_mm:
 837         /* Initializing for Swap token stuff */
 838         mm->token_priority = 0;
 839         mm->last_interval = 0;
 840
 841         tsk->mm = mm;
 842         tsk->active_mm = mm;
 843         return 0;
 844
 845 fail_nomem:
 846         return retval;
 847 }
 848
 849 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 850 {
 851         struct fs_struct *fs = current->fs;
 852         if (clone_flags & CLONE_FS) {
 853                 /* tsk->fs is already what we want */
 854                 spin_lock(&fs->lock);
 855                 if (fs->in_exec) {
 856                         spin_unlock(&fs->lock);
 857                         return -EAGAIN;
 858                 }
 859                 fs->users++;
 860                 spin_unlock(&fs->lock);
 861                 return 0;
 862         }
 863         tsk->fs = copy_fs_struct(fs);
 864         if (!tsk->fs)
 865                 return -ENOMEM;
 866         return 0;
 867 }
 868
 869 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 870 {
 871         struct files_struct *oldf, *newf;
 872         int error = 0;
 873
 874         /*
 875          * A background process may not have any files ...
 876          */
 877         oldf = current->files;
 878         if (!oldf)
 879                 goto out;
 880
 881         if (clone_flags & CLONE_FILES) {
 882                 atomic_inc(&oldf->count);
 883                 goto out;
 884         }
 885
 886         newf = dup_fd(oldf, &error);
 887         if (!newf)
 888                 goto out;
 889
 890         tsk->files = newf;
 891         error = 0;
 892 out:
 893         return error;
 894 }
 895
 896 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 897 {
 898 #ifdef CONFIG_BLOCK
 899         struct io_context *ioc = current->io_context;
 900         struct io_context *new_ioc;
 901
 902         if (!ioc)
 903                 return 0;
 904         /*
 905          * Share io context with parent, if CLONE_IO is set
 906          */
 907         if (clone_flags & CLONE_IO) {
 908                 tsk->io_context = ioc_task_link(ioc);
 909                 if (unlikely(!tsk->io_context))
 910                         return -ENOMEM;
 911         } else if (ioprio_valid(ioc->ioprio)) {
 912                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
 913                 if (unlikely(!new_ioc))
 914                         return -ENOMEM;
 915
 916                 new_ioc->ioprio = ioc->ioprio;
 917                 put_io_context(new_ioc);
 918         }
 919 #endif
 920         return 0;
 921 }
 922
 923 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 924 {
 925         struct sighand_struct *sig;
 926
 927         if (clone_flags & CLONE_SIGHAND) {
 928                 atomic_inc(&current->sighand->count);
 929                 return 0;
 930         }
 931         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 932         rcu_assign_pointer(tsk->sighand, sig);
 933         if (!sig)
 934                 return -ENOMEM;
 935         atomic_set(&sig->count, 1);
 936         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 937         return 0;
 938 }
 939
 940 void __cleanup_sighand(struct sighand_struct *sighand)
 941 {
 942         if (atomic_dec_and_test(&sighand->count)) {
 943                 signalfd_cleanup(sighand);
 944                 kmem_cache_free(sighand_cachep, sighand);
 945         }
 946 }
 947
 948
 949 /*
 950  * Initialize POSIX timer handling for a thread group.
 951  */
 952 static void posix_cpu_timers_init_group(struct signal_struct *sig)
 953 {
 954         unsigned long cpu_limit;
 955
 956         /* Thread group counters. */
 957         thread_group_cputime_init(sig);
 958
 959         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
 960         if (cpu_limit != RLIM_INFINITY) {
 961                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
 962                 sig->cputimer.running = 1;
 963         }
 964
 965         /* The timer lists. */
 966         INIT_LIST_HEAD(&sig->cpu_timers[0]);
 967         INIT_LIST_HEAD(&sig->cpu_timers[1]);
 968         INIT_LIST_HEAD(&sig->cpu_timers[2]);
 969 }
 970
 971 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 972 {
 973         struct signal_struct *sig;
 974
 975         if (clone_flags & CLONE_THREAD)
 976                 return 0;
 977
 978         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
 979         tsk->signal = sig;
 980         if (!sig)
 981                 return -ENOMEM;
 982
 983         sig->nr_threads = 1;
 984         atomic_set(&sig->live, 1);
 985         atomic_set(&sig->sigcnt, 1);
 986         init_waitqueue_head(&sig->wait_chldexit);
 987         if (clone_flags & CLONE_NEWPID)
 988                 sig->flags |= SIGNAL_UNKILLABLE;
 989         sig->curr_target = tsk;
 990         init_sigpending(&sig->shared_pending);
 991         INIT_LIST_HEAD(&sig->posix_timers);
 992
 993         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 994         sig->real_timer.function = it_real_fn;
 995
 996         task_lock(current->group_leader);
 997         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
 998         task_unlock(current->group_leader);
 999
1000         posix_cpu_timers_init_group(sig);
1001
1002         tty_audit_fork(sig);
1003         sched_autogroup_fork(sig);
1004
1005 #ifdef CONFIG_CGROUPS
1006         init_rwsem(&sig->group_rwsem);
1007 #endif
1008
1009         sig->oom_adj = current->signal->oom_adj;
1010         sig->oom_score_adj = current->signal->oom_score_adj;
1011         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1012
1013         mutex_init(&sig->cred_guard_mutex);
1014
1015         return 0;
1016 }
1017
1018 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1019 {
1020         unsigned long new_flags = p->flags;
1021
1022         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1023         new_flags |= PF_FORKNOEXEC;
1024         new_flags |= PF_STARTING;
1025         p->flags = new_flags;
1026 }
1027
1028 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1029 {
1030         current->clear_child_tid = tidptr;
1031
1032         return task_pid_vnr(current);
1033 }
1034
1035 static void rt_mutex_init_task(struct task_struct *p)
1036 {
1037         raw_spin_lock_init(&p->pi_lock);
1038 #ifdef CONFIG_RT_MUTEXES
1039         plist_head_init(&p->pi_waiters);
1040         p->pi_blocked_on = NULL;
1041 #endif
1042 }
1043
1044 #ifdef CONFIG_MM_OWNER
1045 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1046 {
1047         mm->owner = p;
1048 }
1049 #endif /* CONFIG_MM_OWNER */
1050
1051 /*
1052  * Initialize POSIX timer handling for a single task.
1053  */
1054 static void posix_cpu_timers_init(struct task_struct *tsk)
1055 {
1056         tsk->cputime_expires.prof_exp = 0;
1057         tsk->cputime_expires.virt_exp = 0;
1058         tsk->cputime_expires.sched_exp = 0;
1059         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1060         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1061         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1062 }
1063
1064 /*
1065  * This creates a new process as a copy of the old one,
1066  * but does not actually start it yet.
1067  *
1068  * It copies the registers, and all the appropriate
1069  * parts of the process environment (as per the clone
1070  * flags). The actual kick-off is left to the caller.
1071  */
1072 static struct task_struct *copy_process(unsigned long clone_flags,
1073                                         unsigned long stack_start,
1074                                         struct pt_regs *regs,
1075                                         unsigned long stack_size,
1076                                         int __user *child_tidptr,
1077                                         struct pid *pid,
1078                                         int trace)
1079 {
1080         int retval;
1081         struct task_struct *p;
1082         int cgroup_callbacks_done = 0;
1083
1084         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1085                 return ERR_PTR(-EINVAL);
1086
1087         /*
1088          * Thread groups must share signals as well, and detached threads
1089          * can only be started up within the thread group.
1090          */
1091         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1092                 return ERR_PTR(-EINVAL);
1093
1094         /*
1095          * Shared signal handlers imply shared VM. By way of the above,
1096          * thread groups also imply shared VM. Blocking this case allows
1097          * for various simplifications in other code.
1098          */
1099         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1100                 return ERR_PTR(-EINVAL);
1101
1102         /*
1103          * Siblings of global init remain as zombies on exit since they are
1104          * not reaped by their parent (swapper). To solve this and to avoid
1105          * multi-rooted process trees, prevent global and container-inits
1106          * from creating siblings.
1107          */
1108         if ((clone_flags & CLONE_PARENT) &&
1109                                 current->signal->flags & SIGNAL_UNKILLABLE)
1110                 return ERR_PTR(-EINVAL);
1111
1112         retval = security_task_create(clone_flags);
1113         if (retval)
1114                 goto fork_out;
1115
1116         retval = -ENOMEM;
1117         p = dup_task_struct(current);
1118         if (!p)
1119                 goto fork_out;
1120
1121         ftrace_graph_init_task(p);
1122
1123         rt_mutex_init_task(p);
1124
1125 #ifdef CONFIG_PROVE_LOCKING
1126         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1127         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1128 #endif
1129         retval = -EAGAIN;
1130         if (atomic_read(&p->real_cred->user->processes) >=
1131                         task_rlimit(p, RLIMIT_NPROC)) {
1132                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1133                     p->real_cred->user != INIT_USER)
1134                         goto bad_fork_free;
1135         }
1136         current->flags &= ~PF_NPROC_EXCEEDED;
1137
1138         retval = copy_creds(p, clone_flags);
1139         if (retval < 0)
1140                 goto bad_fork_free;
1141
1142         /*
1143          * If multiple threads are within copy_process(), then this check
1144          * triggers too late. This doesn't hurt, the check is only there
1145          * to stop root fork bombs.
1146          */
1147         retval = -EAGAIN;
1148         if (nr_threads >= max_threads)
1149                 goto bad_fork_cleanup_count;
1150
1151         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1152                 goto bad_fork_cleanup_count;
1153
1154         p->did_exec = 0;
1155         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1156         copy_flags(clone_flags, p);
1157         INIT_LIST_HEAD(&p->children);
1158         INIT_LIST_HEAD(&p->sibling);
1159         rcu_copy_process(p);
1160         p->vfork_done = NULL;
1161         spin_lock_init(&p->alloc_lock);
1162
1163         init_sigpending(&p->pending);
1164
1165         p->utime = p->stime = p->gtime = 0;
1166         p->utimescaled = p->stimescaled = 0;
1167 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1168         p->prev_utime = p->prev_stime = 0;
1169 #endif
1170 #if defined(SPLIT_RSS_COUNTING)
1171         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1172 #endif
1173
1174         p->default_timer_slack_ns = current->timer_slack_ns;
1175
1176         task_io_accounting_init(&p->ioac);
1177         acct_clear_integrals(p);
1178
1179         posix_cpu_timers_init(p);
1180
1181         do_posix_clock_monotonic_gettime(&p->start_time);
1182         p->real_start_time = p->start_time;
1183         monotonic_to_bootbased(&p->real_start_time);
1184         p->io_context = NULL;
1185         p->audit_context = NULL;
1186         if (clone_flags & CLONE_THREAD)
1187                 threadgroup_change_begin(current);
1188         cgroup_fork(p);
1189 #ifdef CONFIG_NUMA
1190         p->mempolicy = mpol_dup(p->mempolicy);
1191         if (IS_ERR(p->mempolicy)) {
1192                 retval = PTR_ERR(p->mempolicy);
1193                 p->mempolicy = NULL;
1194                 goto bad_fork_cleanup_cgroup;
1195         }
1196         mpol_fix_fork_child_flag(p);
1197 #endif
1198 #ifdef CONFIG_CPUSETS
1199         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1200         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1201 #endif
1202 #ifdef CONFIG_TRACE_IRQFLAGS
1203         p->irq_events = 0;
1204 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1205         p->hardirqs_enabled = 1;
1206 #else
1207         p->hardirqs_enabled = 0;
1208 #endif
1209         p->hardirq_enable_ip = 0;
1210         p->hardirq_enable_event = 0;
1211         p->hardirq_disable_ip = _THIS_IP_;
1212         p->hardirq_disable_event = 0;
1213         p->softirqs_enabled = 1;
1214         p->softirq_enable_ip = _THIS_IP_;
1215         p->softirq_enable_event = 0;
1216         p->softirq_disable_ip = 0;
1217         p->softirq_disable_event = 0;
1218         p->hardirq_context = 0;
1219         p->softirq_context = 0;
1220 #endif
1221 #ifdef CONFIG_LOCKDEP
1222         p->lockdep_depth = 0; /* no locks held yet */
1223         p->curr_chain_key = 0;
1224         p->lockdep_recursion = 0;
1225 #endif
1226
1227 #ifdef CONFIG_DEBUG_MUTEXES
1228         p->blocked_on = NULL; /* not blocked yet */
1229 #endif
1230 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1231         p->memcg_batch.do_batch = 0;
1232         p->memcg_batch.memcg = NULL;
1233 #endif
1234
1235         /* Perform scheduler related setup. Assign this task to a CPU. */
1236         sched_fork(p);
1237
1238         retval = perf_event_init_task(p);
1239         if (retval)
1240                 goto bad_fork_cleanup_policy;
1241         retval = audit_alloc(p);
1242         if (retval)
1243                 goto bad_fork_cleanup_policy;
1244         /* copy all the process information */
1245         retval = copy_semundo(clone_flags, p);
1246         if (retval)
1247                 goto bad_fork_cleanup_audit;
1248         retval = copy_files(clone_flags, p);
1249         if (retval)
1250                 goto bad_fork_cleanup_semundo;
1251         retval = copy_fs(clone_flags, p);
1252         if (retval)
1253                 goto bad_fork_cleanup_files;
1254         retval = copy_sighand(clone_flags, p);
1255         if (retval)
1256                 goto bad_fork_cleanup_fs;
1257         retval = copy_signal(clone_flags, p);
1258         if (retval)
1259                 goto bad_fork_cleanup_sighand;
1260         retval = copy_mm(clone_flags, p);
1261         if (retval)
1262                 goto bad_fork_cleanup_signal;
1263         retval = copy_namespaces(clone_flags, p);
1264         if (retval)
1265                 goto bad_fork_cleanup_mm;
1266         retval = copy_io(clone_flags, p);
1267         if (retval)
1268                 goto bad_fork_cleanup_namespaces;
1269         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1270         if (retval)
1271                 goto bad_fork_cleanup_io;
1272
1273         if (pid != &init_struct_pid) {
1274                 retval = -ENOMEM;
1275                 pid = alloc_pid(p->nsproxy->pid_ns);
1276                 if (!pid)
1277                         goto bad_fork_cleanup_io;
1278         }
1279
1280         p->pid = pid_nr(pid);
1281         p->tgid = p->pid;
1282         if (clone_flags & CLONE_THREAD)
1283                 p->tgid = current->tgid;
1284
1285         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1286         /*
1287          * Clear TID on mm_release()?
1288          */
1289         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1290 #ifdef CONFIG_BLOCK
1291         p->plug = NULL;
1292 #endif
1293 #ifdef CONFIG_FUTEX
1294         p->robust_list = NULL;
1295 #ifdef CONFIG_COMPAT
1296         p->compat_robust_list = NULL;
1297 #endif
1298         INIT_LIST_HEAD(&p->pi_state_list);
1299         p->pi_state_cache = NULL;
1300 #endif
1301         /*
1302          * sigaltstack should be cleared when sharing the same VM
1303          */
1304         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1305                 p->sas_ss_sp = p->sas_ss_size = 0;
1306
1307         /*
1308          * Syscall tracing and stepping should be turned off in the
1309          * child regardless of CLONE_PTRACE.
1310          */
1311         user_disable_single_step(p);
1312         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1313 #ifdef TIF_SYSCALL_EMU
1314         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1315 #endif
1316         clear_all_latency_tracing(p);
1317
1318         /* ok, now we should be set up.. */
1319         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1320         p->pdeath_signal = 0;
1321         p->exit_state = 0;
1322
1323         p->nr_dirtied = 0;
1324         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1325         p->dirty_paused_when = 0;
1326
1327         /*
1328          * Ok, make it visible to the rest of the system.
1329          * We dont wake it up yet.
1330          */
1331         p->group_leader = p;
1332         INIT_LIST_HEAD(&p->thread_group);
1333
1334         /* Now that the task is set up, run cgroup callbacks if
1335          * necessary. We need to run them before the task is visible
1336          * on the tasklist. */
1337         cgroup_fork_callbacks(p);
1338         cgroup_callbacks_done = 1;
1339
1340         /* Need tasklist lock for parent etc handling! */
1341         write_lock_irq(&tasklist_lock);
1342
1343         /* CLONE_PARENT re-uses the old parent */
1344         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1345                 p->real_parent = current->real_parent;
1346                 p->parent_exec_id = current->parent_exec_id;
1347         } else {
1348                 p->real_parent = current;
1349                 p->parent_exec_id = current->self_exec_id;
1350         }
1351
1352         spin_lock(&current->sighand->siglock);
1353
1354         /*
1355          * Process group and session signals need to be delivered to just the
1356          * parent before the fork or both the parent and the child after the
1357          * fork. Restart if a signal comes in before we add the new process to
1358          * it's process group.
1359          * A fatal signal pending means that current will exit, so the new
1360          * thread can't slip out of an OOM kill (or normal SIGKILL).
1361         */
1362         recalc_sigpending();
1363         if (signal_pending(current)) {
1364                 spin_unlock(&current->sighand->siglock);
1365                 write_unlock_irq(&tasklist_lock);
1366                 retval = -ERESTARTNOINTR;
1367                 goto bad_fork_free_pid;
1368         }
1369
1370         if (clone_flags & CLONE_THREAD) {
1371                 current->signal->nr_threads++;
1372                 atomic_inc(&current->signal->live);
1373                 atomic_inc(&current->signal->sigcnt);
1374                 p->group_leader = current->group_leader;
1375                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1376         }
1377
1378         if (likely(p->pid)) {
1379                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1380
1381                 if (thread_group_leader(p)) {
1382                         if (is_child_reaper(pid))
1383                                 p->nsproxy->pid_ns->child_reaper = p;
1384
1385                         p->signal->leader_pid = pid;
1386                         p->signal->tty = tty_kref_get(current->signal->tty);
1387                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1388                         attach_pid(p, PIDTYPE_SID, task_session(current));
1389                         list_add_tail(&p->sibling, &p->real_parent->children);
1390                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1391                         __this_cpu_inc(process_counts);
1392                 }
1393                 attach_pid(p, PIDTYPE_PID, pid);
1394                 nr_threads++;
1395         }
1396
1397         total_forks++;
1398         spin_unlock(&current->sighand->siglock);
1399         write_unlock_irq(&tasklist_lock);
1400         proc_fork_connector(p);
1401         cgroup_post_fork(p);
1402         if (clone_flags & CLONE_THREAD)
1403                 threadgroup_change_end(current);
1404         perf_event_fork(p);
1405
1406         trace_task_newtask(p, clone_flags);
1407
1408         return p;
1409
1410 bad_fork_free_pid:
1411         if (pid != &init_struct_pid)
1412                 free_pid(pid);
1413 bad_fork_cleanup_io:
1414         if (p->io_context)
1415                 exit_io_context(p);
1416 bad_fork_cleanup_namespaces:
1417         exit_task_namespaces(p);
1418 bad_fork_cleanup_mm:
1419         if (p->mm)
1420                 mmput(p->mm);
1421 bad_fork_cleanup_signal:
1422         if (!(clone_flags & CLONE_THREAD))
1423                 free_signal_struct(p->signal);
1424 bad_fork_cleanup_sighand:
1425         __cleanup_sighand(p->sighand);
1426 bad_fork_cleanup_fs:
1427         exit_fs(p); /* blocking */
1428 bad_fork_cleanup_files:
1429         exit_files(p); /* blocking */
1430 bad_fork_cleanup_semundo:
1431         exit_sem(p);
1432 bad_fork_cleanup_audit:
1433         audit_free(p);
1434 bad_fork_cleanup_policy:
1435         perf_event_free_task(p);
1436 #ifdef CONFIG_NUMA
1437         mpol_put(p->mempolicy);
1438 bad_fork_cleanup_cgroup:
1439 #endif
1440         if (clone_flags & CLONE_THREAD)
1441                 threadgroup_change_end(current);
1442         cgroup_exit(p, cgroup_callbacks_done);
1443         delayacct_tsk_free(p);
1444         module_put(task_thread_info(p)->exec_domain->module);
1445 bad_fork_cleanup_count:
1446         atomic_dec(&p->cred->user->processes);
1447         exit_creds(p);
1448 bad_fork_free:
1449         free_task(p);
1450 fork_out:
1451         return ERR_PTR(retval);
1452 }
1453
1454 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1455 {
1456         memset(regs, 0, sizeof(struct pt_regs));
1457         return regs;
1458 }
1459
1460 static inline void init_idle_pids(struct pid_link *links)
1461 {
1462         enum pid_type type;
1463
1464         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1465                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1466                 links[type].pid = &init_struct_pid;
1467         }
1468 }
1469
1470 struct task_struct * __cpuinit fork_idle(int cpu)
1471 {
1472         struct task_struct *task;
1473         struct pt_regs regs;
1474
1475         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1476                             &init_struct_pid, 0);
1477         if (!IS_ERR(task)) {
1478                 init_idle_pids(task->pids);
1479                 init_idle(task, cpu);
1480         }
1481
1482         return task;
1483 }
1484
1485 /*
1486  *  Ok, this is the main fork-routine.
1487  *
1488  * It copies the process, and if successful kick-starts
1489  * it and waits for it to finish using the VM if required.
1490  */
1491 long do_fork(unsigned long clone_flags,
1492               unsigned long stack_start,
1493               struct pt_regs *regs,
1494               unsigned long stack_size,
1495               int __user *parent_tidptr,
1496               int __user *child_tidptr)
1497 {
1498         struct task_struct *p;
1499         int trace = 0;
1500         long nr;
1501
1502         /*
1503          * Do some preliminary argument and permissions checking before we
1504          * actually start allocating stuff
1505          */
1506         if (clone_flags & CLONE_NEWUSER) {
1507                 if (clone_flags & CLONE_THREAD)
1508                         return -EINVAL;
1509                 /* hopefully this check will go away when userns support is
1510                  * complete
1511                  */
1512                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1513                                 !capable(CAP_SETGID))
1514                         return -EPERM;
1515         }
1516
1517         /*
1518          * Determine whether and which event to report to ptracer.  When
1519          * called from kernel_thread or CLONE_UNTRACED is explicitly
1520          * requested, no event is reported; otherwise, report if the event
1521          * for the type of forking is enabled.
1522          */
1523         if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1524                 if (clone_flags & CLONE_VFORK)
1525                         trace = PTRACE_EVENT_VFORK;
1526                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1527                         trace = PTRACE_EVENT_CLONE;
1528                 else
1529                         trace = PTRACE_EVENT_FORK;
1530
1531                 if (likely(!ptrace_event_enabled(current, trace)))
1532                         trace = 0;
1533         }
1534
1535         p = copy_process(clone_flags, stack_start, regs, stack_size,
1536                          child_tidptr, NULL, trace);
1537         /*
1538          * Do this prior waking up the new thread - the thread pointer
1539          * might get invalid after that point, if the thread exits quickly.
1540          */
1541         if (!IS_ERR(p)) {
1542                 struct completion vfork;
1543
1544                 trace_sched_process_fork(current, p);
1545
1546                 nr = task_pid_vnr(p);
1547
1548                 if (clone_flags & CLONE_PARENT_SETTID)
1549                         put_user(nr, parent_tidptr);
1550
1551                 if (clone_flags & CLONE_VFORK) {
1552                         p->vfork_done = &vfork;
1553                         init_completion(&vfork);
1554                 }
1555
1556                 /*
1557                  * We set PF_STARTING at creation in case tracing wants to
1558                  * use this to distinguish a fully live task from one that
1559                  * hasn't finished SIGSTOP raising yet.  Now we clear it
1560                  * and set the child going.
1561                  */
1562                 p->flags &= ~PF_STARTING;
1563
1564                 wake_up_new_task(p);
1565
1566                 /* forking complete and child started to run, tell ptracer */
1567                 if (unlikely(trace))
1568                         ptrace_event(trace, nr);
1569
1570                 if (clone_flags & CLONE_VFORK) {
1571                         freezer_do_not_count();
1572                         wait_for_completion(&vfork);
1573                         freezer_count();
1574                         ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1575                 }
1576         } else {
1577                 nr = PTR_ERR(p);
1578         }
1579         return nr;
1580 }
1581
1582 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1583 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1584 #endif
1585
1586 static void sighand_ctor(void *data)
1587 {
1588         struct sighand_struct *sighand = data;
1589
1590         spin_lock_init(&sighand->siglock);
1591         init_waitqueue_head(&sighand->signalfd_wqh);
1592 }
1593
1594 void __init proc_caches_init(void)
1595 {
1596         sighand_cachep = kmem_cache_create("sighand_cache",
1597                         sizeof(struct sighand_struct), 0,
1598                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1599                         SLAB_NOTRACK, sighand_ctor);
1600         signal_cachep = kmem_cache_create("signal_cache",
1601                         sizeof(struct signal_struct), 0,
1602                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1603         files_cachep = kmem_cache_create("files_cache",
1604                         sizeof(struct files_struct), 0,
1605                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1606         fs_cachep = kmem_cache_create("fs_cache",
1607                         sizeof(struct fs_struct), 0,
1608                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1609         /*
1610          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1611          * whole struct cpumask for the OFFSTACK case. We could change
1612          * this to *only* allocate as much of it as required by the
1613          * maximum number of CPU's we can ever have.  The cpumask_allocation
1614          * is at the end of the structure, exactly for that reason.
1615          */
1616         mm_cachep = kmem_cache_create("mm_struct",
1617                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1618                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1619         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1620         mmap_init();
1621         nsproxy_cache_init();
1622 }
1623
1624 /*
1625  * Check constraints on flags passed to the unshare system call.
1626  */
1627 static int check_unshare_flags(unsigned long unshare_flags)
1628 {
1629         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1630                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1631                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1632                 return -EINVAL;
1633         /*
1634          * Not implemented, but pretend it works if there is nothing to
1635          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1636          * needs to unshare vm.
1637          */
1638         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1639                 /* FIXME: get_task_mm() increments ->mm_users */
1640                 if (atomic_read(&current->mm->mm_users) > 1)
1641                         return -EINVAL;
1642         }
1643
1644         return 0;
1645 }
1646
1647 /*
1648  * Unshare the filesystem structure if it is being shared
1649  */
1650 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1651 {
1652         struct fs_struct *fs = current->fs;
1653
1654         if (!(unshare_flags & CLONE_FS) || !fs)
1655                 return 0;
1656
1657         /* don't need lock here; in the worst case we'll do useless copy */
1658         if (fs->users == 1)
1659                 return 0;
1660
1661         *new_fsp = copy_fs_struct(fs);
1662         if (!*new_fsp)
1663                 return -ENOMEM;
1664
1665         return 0;
1666 }
1667
1668 /*
1669  * Unshare file descriptor table if it is being shared
1670  */
1671 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1672 {
1673         struct files_struct *fd = current->files;
1674         int error = 0;
1675
1676         if ((unshare_flags & CLONE_FILES) &&
1677             (fd && atomic_read(&fd->count) > 1)) {
1678                 *new_fdp = dup_fd(fd, &error);
1679                 if (!*new_fdp)
1680                         return error;
1681         }
1682
1683         return 0;
1684 }
1685
1686 /*
1687  * unshare allows a process to 'unshare' part of the process
1688  * context which was originally shared using clone.  copy_*
1689  * functions used by do_fork() cannot be used here directly
1690  * because they modify an inactive task_struct that is being
1691  * constructed. Here we are modifying the current, active,
1692  * task_struct.
1693  */
1694 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1695 {
1696         struct fs_struct *fs, *new_fs = NULL;
1697         struct files_struct *fd, *new_fd = NULL;
1698         struct nsproxy *new_nsproxy = NULL;
1699         int do_sysvsem = 0;
1700         int err;
1701
1702         err = check_unshare_flags(unshare_flags);
1703         if (err)
1704                 goto bad_unshare_out;
1705
1706         /*
1707          * If unsharing namespace, must also unshare filesystem information.
1708          */
1709         if (unshare_flags & CLONE_NEWNS)
1710                 unshare_flags |= CLONE_FS;
1711         /*
1712          * CLONE_NEWIPC must also detach from the undolist: after switching
1713          * to a new ipc namespace, the semaphore arrays from the old
1714          * namespace are unreachable.
1715          */
1716         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1717                 do_sysvsem = 1;
1718         err = unshare_fs(unshare_flags, &new_fs);
1719         if (err)
1720                 goto bad_unshare_out;
1721         err = unshare_fd(unshare_flags, &new_fd);
1722         if (err)
1723                 goto bad_unshare_cleanup_fs;
1724         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1725         if (err)
1726                 goto bad_unshare_cleanup_fd;
1727
1728         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1729                 if (do_sysvsem) {
1730                         /*
1731                          * CLONE_SYSVSEM is equivalent to sys_exit().
1732                          */
1733                         exit_sem(current);
1734                 }
1735
1736                 if (new_nsproxy) {
1737                         switch_task_namespaces(current, new_nsproxy);
1738                         new_nsproxy = NULL;
1739                 }
1740
1741                 task_lock(current);
1742
1743                 if (new_fs) {
1744                         fs = current->fs;
1745                         spin_lock(&fs->lock);
1746                         current->fs = new_fs;
1747                         if (--fs->users)
1748                                 new_fs = NULL;
1749                         else
1750                                 new_fs = fs;
1751                         spin_unlock(&fs->lock);
1752                 }
1753
1754                 if (new_fd) {
1755                         fd = current->files;
1756                         current->files = new_fd;
1757                         new_fd = fd;
1758                 }
1759
1760                 task_unlock(current);
1761         }
1762
1763         if (new_nsproxy)
1764                 put_nsproxy(new_nsproxy);
1765
1766 bad_unshare_cleanup_fd:
1767         if (new_fd)
1768                 put_files_struct(new_fd);
1769
1770 bad_unshare_cleanup_fs:
1771         if (new_fs)
1772                 free_fs_struct(new_fs);
1773
1774 bad_unshare_out:
1775         return err;
1776 }
1777
1778 /*
1779  *      Helper to unshare the files of the current task.
1780  *      We don't want to expose copy_files internals to
1781  *      the exec layer of the kernel.
1782  */
1783
1784 int unshare_files(struct files_struct **displaced)
1785 {
1786         struct task_struct *task = current;
1787         struct files_struct *copy = NULL;
1788         int error;
1789
1790         error = unshare_fd(CLONE_FILES, &copy);
1791         if (error || !copy) {
1792                 *displaced = NULL;
1793                 return error;
1794         }
1795         *displaced = task->files;
1796         task_lock(task);
1797         task->files = copy;
1798         task_unlock(task);
1799         return 0;
1800 }