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 #include <linux/uprobes.h>
  71
  72 #include <asm/pgtable.h>
  73 #include <asm/pgalloc.h>
  74 #include <asm/uaccess.h>
  75 #include <asm/mmu_context.h>
  76 #include <asm/cacheflush.h>
  77 #include <asm/tlbflush.h>
  78
  79 #include <trace/events/sched.h>
  80
  81 #define CREATE_TRACE_POINTS
  82 #include <trace/events/task.h>
  83
  84 /*
  85  * Protected counters by write_lock_irq(&tasklist_lock)
  86  */
  87 unsigned long total_forks;      /* Handle normal Linux uptimes. */
  88 int nr_threads;                 /* The idle threads do not count.. */
  89
  90 int max_threads;                /* tunable limit on nr_threads */
  91
  92 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
  93
  94 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
  95
  96 #ifdef CONFIG_PROVE_RCU
  97 int lockdep_tasklist_lock_is_held(void)
  98 {
  99         return lockdep_is_held(&tasklist_lock);
 100 }
 101 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 102 #endif /* #ifdef CONFIG_PROVE_RCU */
 103
 104 int nr_processes(void)
 105 {
 106         int cpu;
 107         int total = 0;
 108
 109         for_each_possible_cpu(cpu)
 110                 total += per_cpu(process_counts, cpu);
 111
 112         return total;
 113 }
 114
 115 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 116 # define alloc_task_struct_node(node)           \
 117                 kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node)
 118 # define free_task_struct(tsk)                  \
 119                 kmem_cache_free(task_struct_cachep, (tsk))
 120 static struct kmem_cache *task_struct_cachep;
 121 #endif
 122
 123 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
 124 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 125                                                   int node)
 126 {
 127 #ifdef CONFIG_DEBUG_STACK_USAGE
 128         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
 129 #else
 130         gfp_t mask = GFP_KERNEL;
 131 #endif
 132         struct page *page = alloc_pages_node(node, mask, THREAD_SIZE_ORDER);
 133
 134         return page ? page_address(page) : NULL;
 135 }
 136
 137 static inline void free_thread_info(struct thread_info *ti)
 138 {
 139         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 140 }
 141 #endif
 142
 143 /* SLAB cache for signal_struct structures (tsk->signal) */
 144 static struct kmem_cache *signal_cachep;
 145
 146 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 147 struct kmem_cache *sighand_cachep;
 148
 149 /* SLAB cache for files_struct structures (tsk->files) */
 150 struct kmem_cache *files_cachep;
 151
 152 /* SLAB cache for fs_struct structures (tsk->fs) */
 153 struct kmem_cache *fs_cachep;
 154
 155 /* SLAB cache for vm_area_struct structures */
 156 struct kmem_cache *vm_area_cachep;
 157
 158 /* SLAB cache for mm_struct structures (tsk->mm) */
 159 static struct kmem_cache *mm_cachep;
 160
 161 static void account_kernel_stack(struct thread_info *ti, int account)
 162 {
 163         struct zone *zone = page_zone(virt_to_page(ti));
 164
 165         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 166 }
 167
 168 void free_task(struct task_struct *tsk)
 169 {
 170         account_kernel_stack(tsk->stack, -1);
 171         free_thread_info(tsk->stack);
 172         rt_mutex_debug_task_free(tsk);
 173         ftrace_graph_exit_task(tsk);
 174         free_task_struct(tsk);
 175 }
 176 EXPORT_SYMBOL(free_task);
 177
 178 static inline void free_signal_struct(struct signal_struct *sig)
 179 {
 180         taskstats_tgid_free(sig);
 181         sched_autogroup_exit(sig);
 182         kmem_cache_free(signal_cachep, sig);
 183 }
 184
 185 static inline void put_signal_struct(struct signal_struct *sig)
 186 {
 187         if (atomic_dec_and_test(&sig->sigcnt))
 188                 free_signal_struct(sig);
 189 }
 190
 191 void __put_task_struct(struct task_struct *tsk)
 192 {
 193         WARN_ON(!tsk->exit_state);
 194         WARN_ON(atomic_read(&tsk->usage));
 195         WARN_ON(tsk == current);
 196
 197         security_task_free(tsk);
 198         exit_creds(tsk);
 199         delayacct_tsk_free(tsk);
 200         put_signal_struct(tsk->signal);
 201
 202         if (!profile_handoff_task(tsk))
 203                 free_task(tsk);
 204 }
 205 EXPORT_SYMBOL_GPL(__put_task_struct);
 206
 207 /*
 208  * macro override instead of weak attribute alias, to workaround
 209  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
 210  */
 211 #ifndef arch_task_cache_init
 212 #define arch_task_cache_init()
 213 #endif
 214
 215 void __init fork_init(unsigned long mempages)
 216 {
 217 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
 218 #ifndef ARCH_MIN_TASKALIGN
 219 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 220 #endif
 221         /* create a slab on which task_structs can be allocated */
 222         task_struct_cachep =
 223                 kmem_cache_create("task_struct", sizeof(struct task_struct),
 224                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 225 #endif
 226
 227         /* do the arch specific task caches init */
 228         arch_task_cache_init();
 229
 230         /*
 231          * The default maximum number of threads is set to a safe
 232          * value: the thread structures can take up at most half
 233          * of memory.
 234          */
 235         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
 236
 237         /*
 238          * we need to allow at least 20 threads to boot a system
 239          */
 240         if (max_threads < 20)
 241                 max_threads = 20;
 242
 243         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 244         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 245         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 246                 init_task.signal->rlim[RLIMIT_NPROC];
 247 }
 248
 249 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
 250                                                struct task_struct *src)
 251 {
 252         *dst = *src;
 253         return 0;
 254 }
 255
 256 static struct task_struct *dup_task_struct(struct task_struct *orig)
 257 {
 258         struct task_struct *tsk;
 259         struct thread_info *ti;
 260         unsigned long *stackend;
 261         int node = tsk_fork_get_node(orig);
 262         int err;
 263
 264         prepare_to_copy(orig);
 265
 266         tsk = alloc_task_struct_node(node);
 267         if (!tsk)
 268                 return NULL;
 269
 270         ti = alloc_thread_info_node(tsk, node);
 271         if (!ti) {
 272                 free_task_struct(tsk);
 273                 return NULL;
 274         }
 275
 276         err = arch_dup_task_struct(tsk, orig);
 277         if (err)
 278                 goto out;
 279
 280         tsk->stack = ti;
 281
 282         setup_thread_stack(tsk, orig);
 283         clear_user_return_notifier(tsk);
 284         clear_tsk_need_resched(tsk);
 285         stackend = end_of_stack(tsk);
 286         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 287
 288 #ifdef CONFIG_CC_STACKPROTECTOR
 289         tsk->stack_canary = get_random_int();
 290 #endif
 291
 292         /*
 293          * One for us, one for whoever does the "release_task()" (usually
 294          * parent)
 295          */
 296         atomic_set(&tsk->usage, 2);
 297 #ifdef CONFIG_BLK_DEV_IO_TRACE
 298         tsk->btrace_seq = 0;
 299 #endif
 300         tsk->splice_pipe = NULL;
 301
 302         account_kernel_stack(ti, 1);
 303
 304         return tsk;
 305
 306 out:
 307         free_thread_info(ti);
 308         free_task_struct(tsk);
 309         return NULL;
 310 }
 311
 312 #ifdef CONFIG_MMU
 313 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 314 {
 315         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 316         struct rb_node **rb_link, *rb_parent;
 317         int retval;
 318         unsigned long charge;
 319         struct mempolicy *pol;
 320
 321         down_write(&oldmm->mmap_sem);
 322         flush_cache_dup_mm(oldmm);
 323         /*
 324          * Not linked in yet - no deadlock potential:
 325          */
 326         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 327
 328         mm->locked_vm = 0;
 329         mm->mmap = NULL;
 330         mm->mmap_cache = NULL;
 331         mm->free_area_cache = oldmm->mmap_base;
 332         mm->cached_hole_size = ~0UL;
 333         mm->map_count = 0;
 334         cpumask_clear(mm_cpumask(mm));
 335         mm->mm_rb = RB_ROOT;
 336         rb_link = &mm->mm_rb.rb_node;
 337         rb_parent = NULL;
 338         pprev = &mm->mmap;
 339         retval = ksm_fork(mm, oldmm);
 340         if (retval)
 341                 goto out;
 342         retval = khugepaged_fork(mm, oldmm);
 343         if (retval)
 344                 goto out;
 345
 346         prev = NULL;
 347         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 348                 struct file *file;
 349
 350                 if (mpnt->vm_flags & VM_DONTCOPY) {
 351                         long pages = vma_pages(mpnt);
 352                         mm->total_vm -= pages;
 353                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 354                                                                 -pages);
 355                         continue;
 356                 }
 357                 charge = 0;
 358                 if (mpnt->vm_flags & VM_ACCOUNT) {
 359                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 360                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 361                                 goto fail_nomem;
 362                         charge = len;
 363                 }
 364                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 365                 if (!tmp)
 366                         goto fail_nomem;
 367                 *tmp = *mpnt;
 368                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 369                 pol = mpol_dup(vma_policy(mpnt));
 370                 retval = PTR_ERR(pol);
 371                 if (IS_ERR(pol))
 372                         goto fail_nomem_policy;
 373                 vma_set_policy(tmp, pol);
 374                 tmp->vm_mm = mm;
 375                 if (anon_vma_fork(tmp, mpnt))
 376                         goto fail_nomem_anon_vma_fork;
 377                 tmp->vm_flags &= ~VM_LOCKED;
 378                 tmp->vm_next = tmp->vm_prev = NULL;
 379                 file = tmp->vm_file;
 380                 if (file) {
 381                         struct inode *inode = file->f_path.dentry->d_inode;
 382                         struct address_space *mapping = file->f_mapping;
 383
 384                         get_file(file);
 385                         if (tmp->vm_flags & VM_DENYWRITE)
 386                                 atomic_dec(&inode->i_writecount);
 387                         mutex_lock(&mapping->i_mmap_mutex);
 388                         if (tmp->vm_flags & VM_SHARED)
 389                                 mapping->i_mmap_writable++;
 390                         flush_dcache_mmap_lock(mapping);
 391                         /* insert tmp into the share list, just after mpnt */
 392                         vma_prio_tree_add(tmp, mpnt);
 393                         flush_dcache_mmap_unlock(mapping);
 394                         mutex_unlock(&mapping->i_mmap_mutex);
 395                 }
 396
 397                 /*
 398                  * Clear hugetlb-related page reserves for children. This only
 399                  * affects MAP_PRIVATE mappings. Faults generated by the child
 400                  * are not guaranteed to succeed, even if read-only
 401                  */
 402                 if (is_vm_hugetlb_page(tmp))
 403                         reset_vma_resv_huge_pages(tmp);
 404
 405                 /*
 406                  * Link in the new vma and copy the page table entries.
 407                  */
 408                 *pprev = tmp;
 409                 pprev = &tmp->vm_next;
 410                 tmp->vm_prev = prev;
 411                 prev = tmp;
 412
 413                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 414                 rb_link = &tmp->vm_rb.rb_right;
 415                 rb_parent = &tmp->vm_rb;
 416
 417                 mm->map_count++;
 418                 retval = copy_page_range(mm, oldmm, mpnt);
 419
 420                 if (tmp->vm_ops && tmp->vm_ops->open)
 421                         tmp->vm_ops->open(tmp);
 422
 423                 if (retval)
 424                         goto out;
 425
 426                 if (file && uprobe_mmap(tmp))
 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 static void check_mm(struct mm_struct *mm)
 520 {
 521         int i;
 522
 523         for (i = 0; i < NR_MM_COUNTERS; i++) {
 524                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 525
 526                 if (unlikely(x))
 527                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 528                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 529         }
 530
 531 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 532         VM_BUG_ON(mm->pmd_huge_pte);
 533 #endif
 534 }
 535
 536 /*
 537  * Allocate and initialize an mm_struct.
 538  */
 539 struct mm_struct *mm_alloc(void)
 540 {
 541         struct mm_struct *mm;
 542
 543         mm = allocate_mm();
 544         if (!mm)
 545                 return NULL;
 546
 547         memset(mm, 0, sizeof(*mm));
 548         mm_init_cpumask(mm);
 549         return mm_init(mm, current);
 550 }
 551
 552 /*
 553  * Called when the last reference to the mm
 554  * is dropped: either by a lazy thread or by
 555  * mmput. Free the page directory and the mm.
 556  */
 557 void __mmdrop(struct mm_struct *mm)
 558 {
 559         BUG_ON(mm == &init_mm);
 560         mm_free_pgd(mm);
 561         destroy_context(mm);
 562         mmu_notifier_mm_destroy(mm);
 563         check_mm(mm);
 564         free_mm(mm);
 565 }
 566 EXPORT_SYMBOL_GPL(__mmdrop);
 567
 568 /*
 569  * Decrement the use count and release all resources for an mm.
 570  */
 571 void mmput(struct mm_struct *mm)
 572 {
 573         might_sleep();
 574
 575         if (atomic_dec_and_test(&mm->mm_users)) {
 576                 uprobe_clear_state(mm);
 577                 exit_aio(mm);
 578                 ksm_exit(mm);
 579                 khugepaged_exit(mm); /* must run before exit_mmap */
 580                 exit_mmap(mm);
 581                 set_mm_exe_file(mm, NULL);
 582                 if (!list_empty(&mm->mmlist)) {
 583                         spin_lock(&mmlist_lock);
 584                         list_del(&mm->mmlist);
 585                         spin_unlock(&mmlist_lock);
 586                 }
 587                 put_swap_token(mm);
 588                 if (mm->binfmt)
 589                         module_put(mm->binfmt->module);
 590                 mmdrop(mm);
 591         }
 592 }
 593 EXPORT_SYMBOL_GPL(mmput);
 594
 595 /*
 596  * We added or removed a vma mapping the executable. The vmas are only mapped
 597  * during exec and are not mapped with the mmap system call.
 598  * Callers must hold down_write() on the mm's mmap_sem for these
 599  */
 600 void added_exe_file_vma(struct mm_struct *mm)
 601 {
 602         mm->num_exe_file_vmas++;
 603 }
 604
 605 void removed_exe_file_vma(struct mm_struct *mm)
 606 {
 607         mm->num_exe_file_vmas--;
 608         if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
 609                 fput(mm->exe_file);
 610                 mm->exe_file = NULL;
 611         }
 612
 613 }
 614
 615 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 616 {
 617         if (new_exe_file)
 618                 get_file(new_exe_file);
 619         if (mm->exe_file)
 620                 fput(mm->exe_file);
 621         mm->exe_file = new_exe_file;
 622         mm->num_exe_file_vmas = 0;
 623 }
 624
 625 struct file *get_mm_exe_file(struct mm_struct *mm)
 626 {
 627         struct file *exe_file;
 628
 629         /* We need mmap_sem to protect against races with removal of
 630          * VM_EXECUTABLE vmas */
 631         down_read(&mm->mmap_sem);
 632         exe_file = mm->exe_file;
 633         if (exe_file)
 634                 get_file(exe_file);
 635         up_read(&mm->mmap_sem);
 636         return exe_file;
 637 }
 638
 639 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
 640 {
 641         /* It's safe to write the exe_file pointer without exe_file_lock because
 642          * this is called during fork when the task is not yet in /proc */
 643         newmm->exe_file = get_mm_exe_file(oldmm);
 644 }
 645
 646 /**
 647  * get_task_mm - acquire a reference to the task's mm
 648  *
 649  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 650  * this kernel workthread has transiently adopted a user mm with use_mm,
 651  * to do its AIO) is not set and if so returns a reference to it, after
 652  * bumping up the use count.  User must release the mm via mmput()
 653  * after use.  Typically used by /proc and ptrace.
 654  */
 655 struct mm_struct *get_task_mm(struct task_struct *task)
 656 {
 657         struct mm_struct *mm;
 658
 659         task_lock(task);
 660         mm = task->mm;
 661         if (mm) {
 662                 if (task->flags & PF_KTHREAD)
 663                         mm = NULL;
 664                 else
 665                         atomic_inc(&mm->mm_users);
 666         }
 667         task_unlock(task);
 668         return mm;
 669 }
 670 EXPORT_SYMBOL_GPL(get_task_mm);
 671
 672 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 673 {
 674         struct mm_struct *mm;
 675         int err;
 676
 677         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 678         if (err)
 679                 return ERR_PTR(err);
 680
 681         mm = get_task_mm(task);
 682         if (mm && mm != current->mm &&
 683                         !ptrace_may_access(task, mode)) {
 684                 mmput(mm);
 685                 mm = ERR_PTR(-EACCES);
 686         }
 687         mutex_unlock(&task->signal->cred_guard_mutex);
 688
 689         return mm;
 690 }
 691
 692 static void complete_vfork_done(struct task_struct *tsk)
 693 {
 694         struct completion *vfork;
 695
 696         task_lock(tsk);
 697         vfork = tsk->vfork_done;
 698         if (likely(vfork)) {
 699                 tsk->vfork_done = NULL;
 700                 complete(vfork);
 701         }
 702         task_unlock(tsk);
 703 }
 704
 705 static int wait_for_vfork_done(struct task_struct *child,
 706                                 struct completion *vfork)
 707 {
 708         int killed;
 709
 710         freezer_do_not_count();
 711         killed = wait_for_completion_killable(vfork);
 712         freezer_count();
 713
 714         if (killed) {
 715                 task_lock(child);
 716                 child->vfork_done = NULL;
 717                 task_unlock(child);
 718         }
 719
 720         put_task_struct(child);
 721         return killed;
 722 }
 723
 724 /* Please note the differences between mmput and mm_release.
 725  * mmput is called whenever we stop holding onto a mm_struct,
 726  * error success whatever.
 727  *
 728  * mm_release is called after a mm_struct has been removed
 729  * from the current process.
 730  *
 731  * This difference is important for error handling, when we
 732  * only half set up a mm_struct for a new process and need to restore
 733  * the old one.  Because we mmput the new mm_struct before
 734  * restoring the old one. . .
 735  * Eric Biederman 10 January 1998
 736  */
 737 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 738 {
 739         /* Get rid of any futexes when releasing the mm */
 740 #ifdef CONFIG_FUTEX
 741         if (unlikely(tsk->robust_list)) {
 742                 exit_robust_list(tsk);
 743                 tsk->robust_list = NULL;
 744         }
 745 #ifdef CONFIG_COMPAT
 746         if (unlikely(tsk->compat_robust_list)) {
 747                 compat_exit_robust_list(tsk);
 748                 tsk->compat_robust_list = NULL;
 749         }
 750 #endif
 751         if (unlikely(!list_empty(&tsk->pi_state_list)))
 752                 exit_pi_state_list(tsk);
 753 #endif
 754
 755         uprobe_free_utask(tsk);
 756
 757         /* Get rid of any cached register state */
 758         deactivate_mm(tsk, mm);
 759
 760         if (tsk->vfork_done)
 761                 complete_vfork_done(tsk);
 762
 763         /*
 764          * If we're exiting normally, clear a user-space tid field if
 765          * requested.  We leave this alone when dying by signal, to leave
 766          * the value intact in a core dump, and to save the unnecessary
 767          * trouble, say, a killed vfork parent shouldn't touch this mm.
 768          * Userland only wants this done for a sys_exit.
 769          */
 770         if (tsk->clear_child_tid) {
 771                 if (!(tsk->flags & PF_SIGNALED) &&
 772                     atomic_read(&mm->mm_users) > 1) {
 773                         /*
 774                          * We don't check the error code - if userspace has
 775                          * not set up a proper pointer then tough luck.
 776                          */
 777                         put_user(0, tsk->clear_child_tid);
 778                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 779                                         1, NULL, NULL, 0);
 780                 }
 781                 tsk->clear_child_tid = NULL;
 782         }
 783 }
 784
 785 /*
 786  * Allocate a new mm structure and copy contents from the
 787  * mm structure of the passed in task structure.
 788  */
 789 struct mm_struct *dup_mm(struct task_struct *tsk)
 790 {
 791         struct mm_struct *mm, *oldmm = current->mm;
 792         int err;
 793
 794         if (!oldmm)
 795                 return NULL;
 796
 797         mm = allocate_mm();
 798         if (!mm)
 799                 goto fail_nomem;
 800
 801         memcpy(mm, oldmm, sizeof(*mm));
 802         mm_init_cpumask(mm);
 803
 804         /* Initializing for Swap token stuff */
 805         mm->token_priority = 0;
 806         mm->last_interval = 0;
 807
 808 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 809         mm->pmd_huge_pte = NULL;
 810 #endif
 811         uprobe_reset_state(mm);
 812
 813         if (!mm_init(mm, tsk))
 814                 goto fail_nomem;
 815
 816         if (init_new_context(tsk, mm))
 817                 goto fail_nocontext;
 818
 819         dup_mm_exe_file(oldmm, mm);
 820
 821         err = dup_mmap(mm, oldmm);
 822         if (err)
 823                 goto free_pt;
 824
 825         mm->hiwater_rss = get_mm_rss(mm);
 826         mm->hiwater_vm = mm->total_vm;
 827
 828         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 829                 goto free_pt;
 830
 831         return mm;
 832
 833 free_pt:
 834         /* don't put binfmt in mmput, we haven't got module yet */
 835         mm->binfmt = NULL;
 836         mmput(mm);
 837
 838 fail_nomem:
 839         return NULL;
 840
 841 fail_nocontext:
 842         /*
 843          * If init_new_context() failed, we cannot use mmput() to free the mm
 844          * because it calls destroy_context()
 845          */
 846         mm_free_pgd(mm);
 847         free_mm(mm);
 848         return NULL;
 849 }
 850
 851 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 852 {
 853         struct mm_struct *mm, *oldmm;
 854         int retval;
 855
 856         tsk->min_flt = tsk->maj_flt = 0;
 857         tsk->nvcsw = tsk->nivcsw = 0;
 858 #ifdef CONFIG_DETECT_HUNG_TASK
 859         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 860 #endif
 861
 862         tsk->mm = NULL;
 863         tsk->active_mm = NULL;
 864
 865         /*
 866          * Are we cloning a kernel thread?
 867          *
 868          * We need to steal a active VM for that..
 869          */
 870         oldmm = current->mm;
 871         if (!oldmm)
 872                 return 0;
 873
 874         if (clone_flags & CLONE_VM) {
 875                 atomic_inc(&oldmm->mm_users);
 876                 mm = oldmm;
 877                 goto good_mm;
 878         }
 879
 880         retval = -ENOMEM;
 881         mm = dup_mm(tsk);
 882         if (!mm)
 883                 goto fail_nomem;
 884
 885 good_mm:
 886         /* Initializing for Swap token stuff */
 887         mm->token_priority = 0;
 888         mm->last_interval = 0;
 889
 890         tsk->mm = mm;
 891         tsk->active_mm = mm;
 892         return 0;
 893
 894 fail_nomem:
 895         return retval;
 896 }
 897
 898 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 899 {
 900         struct fs_struct *fs = current->fs;
 901         if (clone_flags & CLONE_FS) {
 902                 /* tsk->fs is already what we want */
 903                 spin_lock(&fs->lock);
 904                 if (fs->in_exec) {
 905                         spin_unlock(&fs->lock);
 906                         return -EAGAIN;
 907                 }
 908                 fs->users++;
 909                 spin_unlock(&fs->lock);
 910                 return 0;
 911         }
 912         tsk->fs = copy_fs_struct(fs);
 913         if (!tsk->fs)
 914                 return -ENOMEM;
 915         return 0;
 916 }
 917
 918 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 919 {
 920         struct files_struct *oldf, *newf;
 921         int error = 0;
 922
 923         /*
 924          * A background process may not have any files ...
 925          */
 926         oldf = current->files;
 927         if (!oldf)
 928                 goto out;
 929
 930         if (clone_flags & CLONE_FILES) {
 931                 atomic_inc(&oldf->count);
 932                 goto out;
 933         }
 934
 935         newf = dup_fd(oldf, &error);
 936         if (!newf)
 937                 goto out;
 938
 939         tsk->files = newf;
 940         error = 0;
 941 out:
 942         return error;
 943 }
 944
 945 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 946 {
 947 #ifdef CONFIG_BLOCK
 948         struct io_context *ioc = current->io_context;
 949         struct io_context *new_ioc;
 950
 951         if (!ioc)
 952                 return 0;
 953         /*
 954          * Share io context with parent, if CLONE_IO is set
 955          */
 956         if (clone_flags & CLONE_IO) {
 957                 tsk->io_context = ioc_task_link(ioc);
 958                 if (unlikely(!tsk->io_context))
 959                         return -ENOMEM;
 960         } else if (ioprio_valid(ioc->ioprio)) {
 961                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
 962                 if (unlikely(!new_ioc))
 963                         return -ENOMEM;
 964
 965                 new_ioc->ioprio = ioc->ioprio;
 966                 put_io_context(new_ioc);
 967         }
 968 #endif
 969         return 0;
 970 }
 971
 972 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 973 {
 974         struct sighand_struct *sig;
 975
 976         if (clone_flags & CLONE_SIGHAND) {
 977                 atomic_inc(&current->sighand->count);
 978                 return 0;
 979         }
 980         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 981         rcu_assign_pointer(tsk->sighand, sig);
 982         if (!sig)
 983                 return -ENOMEM;
 984         atomic_set(&sig->count, 1);
 985         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 986         return 0;
 987 }
 988
 989 void __cleanup_sighand(struct sighand_struct *sighand)
 990 {
 991         if (atomic_dec_and_test(&sighand->count)) {
 992                 signalfd_cleanup(sighand);
 993                 kmem_cache_free(sighand_cachep, sighand);
 994         }
 995 }
 996
 997
 998 /*
 999  * Initialize POSIX timer handling for a thread group.
1000  */
1001 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1002 {
1003         unsigned long cpu_limit;
1004
1005         /* Thread group counters. */
1006         thread_group_cputime_init(sig);
1007
1008         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1009         if (cpu_limit != RLIM_INFINITY) {
1010                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1011                 sig->cputimer.running = 1;
1012         }
1013
1014         /* The timer lists. */
1015         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1016         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1017         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1018 }
1019
1020 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1021 {
1022         struct signal_struct *sig;
1023
1024         if (clone_flags & CLONE_THREAD)
1025                 return 0;
1026
1027         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1028         tsk->signal = sig;
1029         if (!sig)
1030                 return -ENOMEM;
1031
1032         sig->nr_threads = 1;
1033         atomic_set(&sig->live, 1);
1034         atomic_set(&sig->sigcnt, 1);
1035         init_waitqueue_head(&sig->wait_chldexit);
1036         if (clone_flags & CLONE_NEWPID)
1037                 sig->flags |= SIGNAL_UNKILLABLE;
1038         sig->curr_target = tsk;
1039         init_sigpending(&sig->shared_pending);
1040         INIT_LIST_HEAD(&sig->posix_timers);
1041
1042         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1043         sig->real_timer.function = it_real_fn;
1044
1045         task_lock(current->group_leader);
1046         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1047         task_unlock(current->group_leader);
1048
1049         posix_cpu_timers_init_group(sig);
1050
1051         tty_audit_fork(sig);
1052         sched_autogroup_fork(sig);
1053
1054 #ifdef CONFIG_CGROUPS
1055         init_rwsem(&sig->group_rwsem);
1056 #endif
1057
1058         sig->oom_adj = current->signal->oom_adj;
1059         sig->oom_score_adj = current->signal->oom_score_adj;
1060         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1061
1062         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1063                                    current->signal->is_child_subreaper;
1064
1065         mutex_init(&sig->cred_guard_mutex);
1066
1067         return 0;
1068 }
1069
1070 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1071 {
1072         unsigned long new_flags = p->flags;
1073
1074         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1075         new_flags |= PF_FORKNOEXEC;
1076         p->flags = new_flags;
1077 }
1078
1079 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1080 {
1081         current->clear_child_tid = tidptr;
1082
1083         return task_pid_vnr(current);
1084 }
1085
1086 static void rt_mutex_init_task(struct task_struct *p)
1087 {
1088         raw_spin_lock_init(&p->pi_lock);
1089 #ifdef CONFIG_RT_MUTEXES
1090         plist_head_init(&p->pi_waiters);
1091         p->pi_blocked_on = NULL;
1092 #endif
1093 }
1094
1095 #ifdef CONFIG_MM_OWNER
1096 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1097 {
1098         mm->owner = p;
1099 }
1100 #endif /* CONFIG_MM_OWNER */
1101
1102 /*
1103  * Initialize POSIX timer handling for a single task.
1104  */
1105 static void posix_cpu_timers_init(struct task_struct *tsk)
1106 {
1107         tsk->cputime_expires.prof_exp = 0;
1108         tsk->cputime_expires.virt_exp = 0;
1109         tsk->cputime_expires.sched_exp = 0;
1110         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1111         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1112         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1113 }
1114
1115 /*
1116  * This creates a new process as a copy of the old one,
1117  * but does not actually start it yet.
1118  *
1119  * It copies the registers, and all the appropriate
1120  * parts of the process environment (as per the clone
1121  * flags). The actual kick-off is left to the caller.
1122  */
1123 static struct task_struct *copy_process(unsigned long clone_flags,
1124                                         unsigned long stack_start,
1125                                         struct pt_regs *regs,
1126                                         unsigned long stack_size,
1127                                         int __user *child_tidptr,
1128                                         struct pid *pid,
1129                                         int trace)
1130 {
1131         int retval;
1132         struct task_struct *p;
1133         int cgroup_callbacks_done = 0;
1134
1135         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1136                 return ERR_PTR(-EINVAL);
1137
1138         /*
1139          * Thread groups must share signals as well, and detached threads
1140          * can only be started up within the thread group.
1141          */
1142         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1143                 return ERR_PTR(-EINVAL);
1144
1145         /*
1146          * Shared signal handlers imply shared VM. By way of the above,
1147          * thread groups also imply shared VM. Blocking this case allows
1148          * for various simplifications in other code.
1149          */
1150         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1151                 return ERR_PTR(-EINVAL);
1152
1153         /*
1154          * Siblings of global init remain as zombies on exit since they are
1155          * not reaped by their parent (swapper). To solve this and to avoid
1156          * multi-rooted process trees, prevent global and container-inits
1157          * from creating siblings.
1158          */
1159         if ((clone_flags & CLONE_PARENT) &&
1160                                 current->signal->flags & SIGNAL_UNKILLABLE)
1161                 return ERR_PTR(-EINVAL);
1162
1163         retval = security_task_create(clone_flags);
1164         if (retval)
1165                 goto fork_out;
1166
1167         retval = -ENOMEM;
1168         p = dup_task_struct(current);
1169         if (!p)
1170                 goto fork_out;
1171
1172         ftrace_graph_init_task(p);
1173
1174         rt_mutex_init_task(p);
1175
1176 #ifdef CONFIG_PROVE_LOCKING
1177         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1178         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1179 #endif
1180         retval = -EAGAIN;
1181         if (atomic_read(&p->real_cred->user->processes) >=
1182                         task_rlimit(p, RLIMIT_NPROC)) {
1183                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1184                     p->real_cred->user != INIT_USER)
1185                         goto bad_fork_free;
1186         }
1187         current->flags &= ~PF_NPROC_EXCEEDED;
1188
1189         retval = copy_creds(p, clone_flags);
1190         if (retval < 0)
1191                 goto bad_fork_free;
1192
1193         /*
1194          * If multiple threads are within copy_process(), then this check
1195          * triggers too late. This doesn't hurt, the check is only there
1196          * to stop root fork bombs.
1197          */
1198         retval = -EAGAIN;
1199         if (nr_threads >= max_threads)
1200                 goto bad_fork_cleanup_count;
1201
1202         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1203                 goto bad_fork_cleanup_count;
1204
1205         p->did_exec = 0;
1206         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1207         copy_flags(clone_flags, p);
1208         INIT_LIST_HEAD(&p->children);
1209         INIT_LIST_HEAD(&p->sibling);
1210         rcu_copy_process(p);
1211         p->vfork_done = NULL;
1212         spin_lock_init(&p->alloc_lock);
1213
1214         init_sigpending(&p->pending);
1215
1216         p->utime = p->stime = p->gtime = 0;
1217         p->utimescaled = p->stimescaled = 0;
1218 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1219         p->prev_utime = p->prev_stime = 0;
1220 #endif
1221 #if defined(SPLIT_RSS_COUNTING)
1222         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1223 #endif
1224
1225         p->default_timer_slack_ns = current->timer_slack_ns;
1226
1227         task_io_accounting_init(&p->ioac);
1228         acct_clear_integrals(p);
1229
1230         posix_cpu_timers_init(p);
1231
1232         do_posix_clock_monotonic_gettime(&p->start_time);
1233         p->real_start_time = p->start_time;
1234         monotonic_to_bootbased(&p->real_start_time);
1235         p->io_context = NULL;
1236         p->audit_context = NULL;
1237         if (clone_flags & CLONE_THREAD)
1238                 threadgroup_change_begin(current);
1239         cgroup_fork(p);
1240 #ifdef CONFIG_NUMA
1241         p->mempolicy = mpol_dup(p->mempolicy);
1242         if (IS_ERR(p->mempolicy)) {
1243                 retval = PTR_ERR(p->mempolicy);
1244                 p->mempolicy = NULL;
1245                 goto bad_fork_cleanup_cgroup;
1246         }
1247         mpol_fix_fork_child_flag(p);
1248 #endif
1249 #ifdef CONFIG_CPUSETS
1250         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1251         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1252         seqcount_init(&p->mems_allowed_seq);
1253 #endif
1254 #ifdef CONFIG_TRACE_IRQFLAGS
1255         p->irq_events = 0;
1256 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1257         p->hardirqs_enabled = 1;
1258 #else
1259         p->hardirqs_enabled = 0;
1260 #endif
1261         p->hardirq_enable_ip = 0;
1262         p->hardirq_enable_event = 0;
1263         p->hardirq_disable_ip = _THIS_IP_;
1264         p->hardirq_disable_event = 0;
1265         p->softirqs_enabled = 1;
1266         p->softirq_enable_ip = _THIS_IP_;
1267         p->softirq_enable_event = 0;
1268         p->softirq_disable_ip = 0;
1269         p->softirq_disable_event = 0;
1270         p->hardirq_context = 0;
1271         p->softirq_context = 0;
1272 #endif
1273 #ifdef CONFIG_LOCKDEP
1274         p->lockdep_depth = 0; /* no locks held yet */
1275         p->curr_chain_key = 0;
1276         p->lockdep_recursion = 0;
1277 #endif
1278
1279 #ifdef CONFIG_DEBUG_MUTEXES
1280         p->blocked_on = NULL; /* not blocked yet */
1281 #endif
1282 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1283         p->memcg_batch.do_batch = 0;
1284         p->memcg_batch.memcg = NULL;
1285 #endif
1286
1287         /* Perform scheduler related setup. Assign this task to a CPU. */
1288         sched_fork(p);
1289
1290         retval = perf_event_init_task(p);
1291         if (retval)
1292                 goto bad_fork_cleanup_policy;
1293         retval = audit_alloc(p);
1294         if (retval)
1295                 goto bad_fork_cleanup_policy;
1296         /* copy all the process information */
1297         retval = copy_semundo(clone_flags, p);
1298         if (retval)
1299                 goto bad_fork_cleanup_audit;
1300         retval = copy_files(clone_flags, p);
1301         if (retval)
1302                 goto bad_fork_cleanup_semundo;
1303         retval = copy_fs(clone_flags, p);
1304         if (retval)
1305                 goto bad_fork_cleanup_files;
1306         retval = copy_sighand(clone_flags, p);
1307         if (retval)
1308                 goto bad_fork_cleanup_fs;
1309         retval = copy_signal(clone_flags, p);
1310         if (retval)
1311                 goto bad_fork_cleanup_sighand;
1312         retval = copy_mm(clone_flags, p);
1313         if (retval)
1314                 goto bad_fork_cleanup_signal;
1315         retval = copy_namespaces(clone_flags, p);
1316         if (retval)
1317                 goto bad_fork_cleanup_mm;
1318         retval = copy_io(clone_flags, p);
1319         if (retval)
1320                 goto bad_fork_cleanup_namespaces;
1321         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1322         if (retval)
1323                 goto bad_fork_cleanup_io;
1324
1325         if (pid != &init_struct_pid) {
1326                 retval = -ENOMEM;
1327                 pid = alloc_pid(p->nsproxy->pid_ns);
1328                 if (!pid)
1329                         goto bad_fork_cleanup_io;
1330         }
1331
1332         p->pid = pid_nr(pid);
1333         p->tgid = p->pid;
1334         if (clone_flags & CLONE_THREAD)
1335                 p->tgid = current->tgid;
1336
1337         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1338         /*
1339          * Clear TID on mm_release()?
1340          */
1341         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1342 #ifdef CONFIG_BLOCK
1343         p->plug = NULL;
1344 #endif
1345 #ifdef CONFIG_FUTEX
1346         p->robust_list = NULL;
1347 #ifdef CONFIG_COMPAT
1348         p->compat_robust_list = NULL;
1349 #endif
1350         INIT_LIST_HEAD(&p->pi_state_list);
1351         p->pi_state_cache = NULL;
1352 #endif
1353         uprobe_copy_process(p);
1354         /*
1355          * sigaltstack should be cleared when sharing the same VM
1356          */
1357         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1358                 p->sas_ss_sp = p->sas_ss_size = 0;
1359
1360         /*
1361          * Syscall tracing and stepping should be turned off in the
1362          * child regardless of CLONE_PTRACE.
1363          */
1364         user_disable_single_step(p);
1365         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1366 #ifdef TIF_SYSCALL_EMU
1367         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1368 #endif
1369         clear_all_latency_tracing(p);
1370
1371         /* ok, now we should be set up.. */
1372         if (clone_flags & CLONE_THREAD)
1373                 p->exit_signal = -1;
1374         else if (clone_flags & CLONE_PARENT)
1375                 p->exit_signal = current->group_leader->exit_signal;
1376         else
1377                 p->exit_signal = (clone_flags & CSIGNAL);
1378
1379         p->pdeath_signal = 0;
1380         p->exit_state = 0;
1381
1382         p->nr_dirtied = 0;
1383         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1384         p->dirty_paused_when = 0;
1385
1386         /*
1387          * Ok, make it visible to the rest of the system.
1388          * We dont wake it up yet.
1389          */
1390         p->group_leader = p;
1391         INIT_LIST_HEAD(&p->thread_group);
1392
1393         /* Now that the task is set up, run cgroup callbacks if
1394          * necessary. We need to run them before the task is visible
1395          * on the tasklist. */
1396         cgroup_fork_callbacks(p);
1397         cgroup_callbacks_done = 1;
1398
1399         /* Need tasklist lock for parent etc handling! */
1400         write_lock_irq(&tasklist_lock);
1401
1402         /* CLONE_PARENT re-uses the old parent */
1403         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1404                 p->real_parent = current->real_parent;
1405                 p->parent_exec_id = current->parent_exec_id;
1406         } else {
1407                 p->real_parent = current;
1408                 p->parent_exec_id = current->self_exec_id;
1409         }
1410
1411         spin_lock(&current->sighand->siglock);
1412
1413         /*
1414          * Process group and session signals need to be delivered to just the
1415          * parent before the fork or both the parent and the child after the
1416          * fork. Restart if a signal comes in before we add the new process to
1417          * it's process group.
1418          * A fatal signal pending means that current will exit, so the new
1419          * thread can't slip out of an OOM kill (or normal SIGKILL).
1420         */
1421         recalc_sigpending();
1422         if (signal_pending(current)) {
1423                 spin_unlock(&current->sighand->siglock);
1424                 write_unlock_irq(&tasklist_lock);
1425                 retval = -ERESTARTNOINTR;
1426                 goto bad_fork_free_pid;
1427         }
1428
1429         if (clone_flags & CLONE_THREAD) {
1430                 current->signal->nr_threads++;
1431                 atomic_inc(&current->signal->live);
1432                 atomic_inc(&current->signal->sigcnt);
1433                 p->group_leader = current->group_leader;
1434                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1435         }
1436
1437         if (likely(p->pid)) {
1438                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1439
1440                 if (thread_group_leader(p)) {
1441                         if (is_child_reaper(pid))
1442                                 p->nsproxy->pid_ns->child_reaper = p;
1443
1444                         p->signal->leader_pid = pid;
1445                         p->signal->tty = tty_kref_get(current->signal->tty);
1446                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1447                         attach_pid(p, PIDTYPE_SID, task_session(current));
1448                         list_add_tail(&p->sibling, &p->real_parent->children);
1449                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1450                         __this_cpu_inc(process_counts);
1451                 }
1452                 attach_pid(p, PIDTYPE_PID, pid);
1453                 nr_threads++;
1454         }
1455
1456         total_forks++;
1457         spin_unlock(&current->sighand->siglock);
1458         write_unlock_irq(&tasklist_lock);
1459         proc_fork_connector(p);
1460         cgroup_post_fork(p);
1461         if (clone_flags & CLONE_THREAD)
1462                 threadgroup_change_end(current);
1463         perf_event_fork(p);
1464
1465         trace_task_newtask(p, clone_flags);
1466
1467         return p;
1468
1469 bad_fork_free_pid:
1470         if (pid != &init_struct_pid)
1471                 free_pid(pid);
1472 bad_fork_cleanup_io:
1473         if (p->io_context)
1474                 exit_io_context(p);
1475 bad_fork_cleanup_namespaces:
1476         exit_task_namespaces(p);
1477 bad_fork_cleanup_mm:
1478         if (p->mm)
1479                 mmput(p->mm);
1480 bad_fork_cleanup_signal:
1481         if (!(clone_flags & CLONE_THREAD))
1482                 free_signal_struct(p->signal);
1483 bad_fork_cleanup_sighand:
1484         __cleanup_sighand(p->sighand);
1485 bad_fork_cleanup_fs:
1486         exit_fs(p); /* blocking */
1487 bad_fork_cleanup_files:
1488         exit_files(p); /* blocking */
1489 bad_fork_cleanup_semundo:
1490         exit_sem(p);
1491 bad_fork_cleanup_audit:
1492         audit_free(p);
1493 bad_fork_cleanup_policy:
1494         perf_event_free_task(p);
1495 #ifdef CONFIG_NUMA
1496         mpol_put(p->mempolicy);
1497 bad_fork_cleanup_cgroup:
1498 #endif
1499         if (clone_flags & CLONE_THREAD)
1500                 threadgroup_change_end(current);
1501         cgroup_exit(p, cgroup_callbacks_done);
1502         delayacct_tsk_free(p);
1503         module_put(task_thread_info(p)->exec_domain->module);
1504 bad_fork_cleanup_count:
1505         atomic_dec(&p->cred->user->processes);
1506         exit_creds(p);
1507 bad_fork_free:
1508         free_task(p);
1509 fork_out:
1510         return ERR_PTR(retval);
1511 }
1512
1513 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1514 {
1515         memset(regs, 0, sizeof(struct pt_regs));
1516         return regs;
1517 }
1518
1519 static inline void init_idle_pids(struct pid_link *links)
1520 {
1521         enum pid_type type;
1522
1523         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1524                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1525                 links[type].pid = &init_struct_pid;
1526         }
1527 }
1528
1529 struct task_struct * __cpuinit fork_idle(int cpu)
1530 {
1531         struct task_struct *task;
1532         struct pt_regs regs;
1533
1534         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1535                             &init_struct_pid, 0);
1536         if (!IS_ERR(task)) {
1537                 init_idle_pids(task->pids);
1538                 init_idle(task, cpu);
1539         }
1540
1541         return task;
1542 }
1543
1544 /*
1545  *  Ok, this is the main fork-routine.
1546  *
1547  * It copies the process, and if successful kick-starts
1548  * it and waits for it to finish using the VM if required.
1549  */
1550 long do_fork(unsigned long clone_flags,
1551               unsigned long stack_start,
1552               struct pt_regs *regs,
1553               unsigned long stack_size,
1554               int __user *parent_tidptr,
1555               int __user *child_tidptr)
1556 {
1557         struct task_struct *p;
1558         int trace = 0;
1559         long nr;
1560
1561         /*
1562          * Do some preliminary argument and permissions checking before we
1563          * actually start allocating stuff
1564          */
1565         if (clone_flags & CLONE_NEWUSER) {
1566                 if (clone_flags & CLONE_THREAD)
1567                         return -EINVAL;
1568                 /* hopefully this check will go away when userns support is
1569                  * complete
1570                  */
1571                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1572                                 !capable(CAP_SETGID))
1573                         return -EPERM;
1574         }
1575
1576         /*
1577          * Determine whether and which event to report to ptracer.  When
1578          * called from kernel_thread or CLONE_UNTRACED is explicitly
1579          * requested, no event is reported; otherwise, report if the event
1580          * for the type of forking is enabled.
1581          */
1582         if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1583                 if (clone_flags & CLONE_VFORK)
1584                         trace = PTRACE_EVENT_VFORK;
1585                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1586                         trace = PTRACE_EVENT_CLONE;
1587                 else
1588                         trace = PTRACE_EVENT_FORK;
1589
1590                 if (likely(!ptrace_event_enabled(current, trace)))
1591                         trace = 0;
1592         }
1593
1594         p = copy_process(clone_flags, stack_start, regs, stack_size,
1595                          child_tidptr, NULL, trace);
1596         /*
1597          * Do this prior waking up the new thread - the thread pointer
1598          * might get invalid after that point, if the thread exits quickly.
1599          */
1600         if (!IS_ERR(p)) {
1601                 struct completion vfork;
1602
1603                 trace_sched_process_fork(current, p);
1604
1605                 nr = task_pid_vnr(p);
1606
1607                 if (clone_flags & CLONE_PARENT_SETTID)
1608                         put_user(nr, parent_tidptr);
1609
1610                 if (clone_flags & CLONE_VFORK) {
1611                         p->vfork_done = &vfork;
1612                         init_completion(&vfork);
1613                         get_task_struct(p);
1614                 }
1615
1616                 wake_up_new_task(p);
1617
1618                 /* forking complete and child started to run, tell ptracer */
1619                 if (unlikely(trace))
1620                         ptrace_event(trace, nr);
1621
1622                 if (clone_flags & CLONE_VFORK) {
1623                         if (!wait_for_vfork_done(p, &vfork))
1624                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1625                 }
1626         } else {
1627                 nr = PTR_ERR(p);
1628         }
1629         return nr;
1630 }
1631
1632 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1633 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1634 #endif
1635
1636 static void sighand_ctor(void *data)
1637 {
1638         struct sighand_struct *sighand = data;
1639
1640         spin_lock_init(&sighand->siglock);
1641         init_waitqueue_head(&sighand->signalfd_wqh);
1642 }
1643
1644 void __init proc_caches_init(void)
1645 {
1646         sighand_cachep = kmem_cache_create("sighand_cache",
1647                         sizeof(struct sighand_struct), 0,
1648                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1649                         SLAB_NOTRACK, sighand_ctor);
1650         signal_cachep = kmem_cache_create("signal_cache",
1651                         sizeof(struct signal_struct), 0,
1652                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1653         files_cachep = kmem_cache_create("files_cache",
1654                         sizeof(struct files_struct), 0,
1655                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1656         fs_cachep = kmem_cache_create("fs_cache",
1657                         sizeof(struct fs_struct), 0,
1658                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1659         /*
1660          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1661          * whole struct cpumask for the OFFSTACK case. We could change
1662          * this to *only* allocate as much of it as required by the
1663          * maximum number of CPU's we can ever have.  The cpumask_allocation
1664          * is at the end of the structure, exactly for that reason.
1665          */
1666         mm_cachep = kmem_cache_create("mm_struct",
1667                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1668                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1669         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1670         mmap_init();
1671         nsproxy_cache_init();
1672 }
1673
1674 /*
1675  * Check constraints on flags passed to the unshare system call.
1676  */
1677 static int check_unshare_flags(unsigned long unshare_flags)
1678 {
1679         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1680                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1681                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1682                 return -EINVAL;
1683         /*
1684          * Not implemented, but pretend it works if there is nothing to
1685          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1686          * needs to unshare vm.
1687          */
1688         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1689                 /* FIXME: get_task_mm() increments ->mm_users */
1690                 if (atomic_read(&current->mm->mm_users) > 1)
1691                         return -EINVAL;
1692         }
1693
1694         return 0;
1695 }
1696
1697 /*
1698  * Unshare the filesystem structure if it is being shared
1699  */
1700 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1701 {
1702         struct fs_struct *fs = current->fs;
1703
1704         if (!(unshare_flags & CLONE_FS) || !fs)
1705                 return 0;
1706
1707         /* don't need lock here; in the worst case we'll do useless copy */
1708         if (fs->users == 1)
1709                 return 0;
1710
1711         *new_fsp = copy_fs_struct(fs);
1712         if (!*new_fsp)
1713                 return -ENOMEM;
1714
1715         return 0;
1716 }
1717
1718 /*
1719  * Unshare file descriptor table if it is being shared
1720  */
1721 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1722 {
1723         struct files_struct *fd = current->files;
1724         int error = 0;
1725
1726         if ((unshare_flags & CLONE_FILES) &&
1727             (fd && atomic_read(&fd->count) > 1)) {
1728                 *new_fdp = dup_fd(fd, &error);
1729                 if (!*new_fdp)
1730                         return error;
1731         }
1732
1733         return 0;
1734 }
1735
1736 /*
1737  * unshare allows a process to 'unshare' part of the process
1738  * context which was originally shared using clone.  copy_*
1739  * functions used by do_fork() cannot be used here directly
1740  * because they modify an inactive task_struct that is being
1741  * constructed. Here we are modifying the current, active,
1742  * task_struct.
1743  */
1744 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1745 {
1746         struct fs_struct *fs, *new_fs = NULL;
1747         struct files_struct *fd, *new_fd = NULL;
1748         struct nsproxy *new_nsproxy = NULL;
1749         int do_sysvsem = 0;
1750         int err;
1751
1752         err = check_unshare_flags(unshare_flags);
1753         if (err)
1754                 goto bad_unshare_out;
1755
1756         /*
1757          * If unsharing namespace, must also unshare filesystem information.
1758          */
1759         if (unshare_flags & CLONE_NEWNS)
1760                 unshare_flags |= CLONE_FS;
1761         /*
1762          * CLONE_NEWIPC must also detach from the undolist: after switching
1763          * to a new ipc namespace, the semaphore arrays from the old
1764          * namespace are unreachable.
1765          */
1766         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1767                 do_sysvsem = 1;
1768         err = unshare_fs(unshare_flags, &new_fs);
1769         if (err)
1770                 goto bad_unshare_out;
1771         err = unshare_fd(unshare_flags, &new_fd);
1772         if (err)
1773                 goto bad_unshare_cleanup_fs;
1774         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1775         if (err)
1776                 goto bad_unshare_cleanup_fd;
1777
1778         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1779                 if (do_sysvsem) {
1780                         /*
1781                          * CLONE_SYSVSEM is equivalent to sys_exit().
1782                          */
1783                         exit_sem(current);
1784                 }
1785
1786                 if (new_nsproxy) {
1787                         switch_task_namespaces(current, new_nsproxy);
1788                         new_nsproxy = NULL;
1789                 }
1790
1791                 task_lock(current);
1792
1793                 if (new_fs) {
1794                         fs = current->fs;
1795                         spin_lock(&fs->lock);
1796                         current->fs = new_fs;
1797                         if (--fs->users)
1798                                 new_fs = NULL;
1799                         else
1800                                 new_fs = fs;
1801                         spin_unlock(&fs->lock);
1802                 }
1803
1804                 if (new_fd) {
1805                         fd = current->files;
1806                         current->files = new_fd;
1807                         new_fd = fd;
1808                 }
1809
1810                 task_unlock(current);
1811         }
1812
1813         if (new_nsproxy)
1814                 put_nsproxy(new_nsproxy);
1815
1816 bad_unshare_cleanup_fd:
1817         if (new_fd)
1818                 put_files_struct(new_fd);
1819
1820 bad_unshare_cleanup_fs:
1821         if (new_fs)
1822                 free_fs_struct(new_fs);
1823
1824 bad_unshare_out:
1825         return err;
1826 }
1827
1828 /*
1829  *      Helper to unshare the files of the current task.
1830  *      We don't want to expose copy_files internals to
1831  *      the exec layer of the kernel.
1832  */
1833
1834 int unshare_files(struct files_struct **displaced)
1835 {
1836         struct task_struct *task = current;
1837         struct files_struct *copy = NULL;
1838         int error;
1839
1840         error = unshare_fd(CLONE_FILES, &copy);
1841         if (error || !copy) {
1842                 *displaced = NULL;
1843                 return error;
1844         }
1845         *displaced = task->files;
1846         task_lock(task);
1847         task->files = copy;
1848         task_unlock(task);
1849         return 0;
1850 }