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         atomic_set(&mm->active_swap_token, 0);
 808
 809 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 810         mm->pmd_huge_pte = NULL;
 811 #endif
 812         uprobe_reset_state(mm);
 813
 814         if (!mm_init(mm, tsk))
 815                 goto fail_nomem;
 816
 817         if (init_new_context(tsk, mm))
 818                 goto fail_nocontext;
 819
 820         dup_mm_exe_file(oldmm, mm);
 821
 822         err = dup_mmap(mm, oldmm);
 823         if (err)
 824                 goto free_pt;
 825
 826         mm->hiwater_rss = get_mm_rss(mm);
 827         mm->hiwater_vm = mm->total_vm;
 828
 829         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 830                 goto free_pt;
 831
 832         return mm;
 833
 834 free_pt:
 835         /* don't put binfmt in mmput, we haven't got module yet */
 836         mm->binfmt = NULL;
 837         mmput(mm);
 838
 839 fail_nomem:
 840         return NULL;
 841
 842 fail_nocontext:
 843         /*
 844          * If init_new_context() failed, we cannot use mmput() to free the mm
 845          * because it calls destroy_context()
 846          */
 847         mm_free_pgd(mm);
 848         free_mm(mm);
 849         return NULL;
 850 }
 851
 852 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 853 {
 854         struct mm_struct *mm, *oldmm;
 855         int retval;
 856
 857         tsk->min_flt = tsk->maj_flt = 0;
 858         tsk->nvcsw = tsk->nivcsw = 0;
 859 #ifdef CONFIG_DETECT_HUNG_TASK
 860         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 861 #endif
 862
 863         tsk->mm = NULL;
 864         tsk->active_mm = NULL;
 865
 866         /*
 867          * Are we cloning a kernel thread?
 868          *
 869          * We need to steal a active VM for that..
 870          */
 871         oldmm = current->mm;
 872         if (!oldmm)
 873                 return 0;
 874
 875         if (clone_flags & CLONE_VM) {
 876                 atomic_inc(&oldmm->mm_users);
 877                 mm = oldmm;
 878                 goto good_mm;
 879         }
 880
 881         retval = -ENOMEM;
 882         mm = dup_mm(tsk);
 883         if (!mm)
 884                 goto fail_nomem;
 885
 886 good_mm:
 887         /* Initializing for Swap token stuff */
 888         mm->token_priority = 0;
 889         mm->last_interval = 0;
 890
 891         tsk->mm = mm;
 892         tsk->active_mm = mm;
 893         return 0;
 894
 895 fail_nomem:
 896         return retval;
 897 }
 898
 899 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 900 {
 901         struct fs_struct *fs = current->fs;
 902         if (clone_flags & CLONE_FS) {
 903                 /* tsk->fs is already what we want */
 904                 spin_lock(&fs->lock);
 905                 if (fs->in_exec) {
 906                         spin_unlock(&fs->lock);
 907                         return -EAGAIN;
 908                 }
 909                 fs->users++;
 910                 spin_unlock(&fs->lock);
 911                 return 0;
 912         }
 913         tsk->fs = copy_fs_struct(fs);
 914         if (!tsk->fs)
 915                 return -ENOMEM;
 916         return 0;
 917 }
 918
 919 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
 920 {
 921         struct files_struct *oldf, *newf;
 922         int error = 0;
 923
 924         /*
 925          * A background process may not have any files ...
 926          */
 927         oldf = current->files;
 928         if (!oldf)
 929                 goto out;
 930
 931         if (clone_flags & CLONE_FILES) {
 932                 atomic_inc(&oldf->count);
 933                 goto out;
 934         }
 935
 936         newf = dup_fd(oldf, &error);
 937         if (!newf)
 938                 goto out;
 939
 940         tsk->files = newf;
 941         error = 0;
 942 out:
 943         return error;
 944 }
 945
 946 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
 947 {
 948 #ifdef CONFIG_BLOCK
 949         struct io_context *ioc = current->io_context;
 950         struct io_context *new_ioc;
 951
 952         if (!ioc)
 953                 return 0;
 954         /*
 955          * Share io context with parent, if CLONE_IO is set
 956          */
 957         if (clone_flags & CLONE_IO) {
 958                 tsk->io_context = ioc_task_link(ioc);
 959                 if (unlikely(!tsk->io_context))
 960                         return -ENOMEM;
 961         } else if (ioprio_valid(ioc->ioprio)) {
 962                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
 963                 if (unlikely(!new_ioc))
 964                         return -ENOMEM;
 965
 966                 new_ioc->ioprio = ioc->ioprio;
 967                 put_io_context(new_ioc);
 968         }
 969 #endif
 970         return 0;
 971 }
 972
 973 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
 974 {
 975         struct sighand_struct *sig;
 976
 977         if (clone_flags & CLONE_SIGHAND) {
 978                 atomic_inc(&current->sighand->count);
 979                 return 0;
 980         }
 981         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 982         rcu_assign_pointer(tsk->sighand, sig);
 983         if (!sig)
 984                 return -ENOMEM;
 985         atomic_set(&sig->count, 1);
 986         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
 987         return 0;
 988 }
 989
 990 void __cleanup_sighand(struct sighand_struct *sighand)
 991 {
 992         if (atomic_dec_and_test(&sighand->count)) {
 993                 signalfd_cleanup(sighand);
 994                 kmem_cache_free(sighand_cachep, sighand);
 995         }
 996 }
 997
 998
 999 /*
1000  * Initialize POSIX timer handling for a thread group.
1001  */
1002 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1003 {
1004         unsigned long cpu_limit;
1005
1006         /* Thread group counters. */
1007         thread_group_cputime_init(sig);
1008
1009         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1010         if (cpu_limit != RLIM_INFINITY) {
1011                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1012                 sig->cputimer.running = 1;
1013         }
1014
1015         /* The timer lists. */
1016         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1017         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1018         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1019 }
1020
1021 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1022 {
1023         struct signal_struct *sig;
1024
1025         if (clone_flags & CLONE_THREAD)
1026                 return 0;
1027
1028         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1029         tsk->signal = sig;
1030         if (!sig)
1031                 return -ENOMEM;
1032
1033         sig->nr_threads = 1;
1034         atomic_set(&sig->live, 1);
1035         atomic_set(&sig->sigcnt, 1);
1036         init_waitqueue_head(&sig->wait_chldexit);
1037         if (clone_flags & CLONE_NEWPID)
1038                 sig->flags |= SIGNAL_UNKILLABLE;
1039         sig->curr_target = tsk;
1040         init_sigpending(&sig->shared_pending);
1041         INIT_LIST_HEAD(&sig->posix_timers);
1042
1043         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1044         sig->real_timer.function = it_real_fn;
1045
1046         task_lock(current->group_leader);
1047         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1048         task_unlock(current->group_leader);
1049
1050         posix_cpu_timers_init_group(sig);
1051
1052         tty_audit_fork(sig);
1053         sched_autogroup_fork(sig);
1054
1055 #ifdef CONFIG_CGROUPS
1056         init_rwsem(&sig->group_rwsem);
1057 #endif
1058
1059         sig->oom_adj = current->signal->oom_adj;
1060         sig->oom_score_adj = current->signal->oom_score_adj;
1061         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1062
1063         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1064                                    current->signal->is_child_subreaper;
1065
1066         mutex_init(&sig->cred_guard_mutex);
1067
1068         return 0;
1069 }
1070
1071 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1072 {
1073         unsigned long new_flags = p->flags;
1074
1075         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1076         new_flags |= PF_FORKNOEXEC;
1077         p->flags = new_flags;
1078 }
1079
1080 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1081 {
1082         current->clear_child_tid = tidptr;
1083
1084         return task_pid_vnr(current);
1085 }
1086
1087 static void rt_mutex_init_task(struct task_struct *p)
1088 {
1089         raw_spin_lock_init(&p->pi_lock);
1090 #ifdef CONFIG_RT_MUTEXES
1091         plist_head_init(&p->pi_waiters);
1092         p->pi_blocked_on = NULL;
1093 #endif
1094 }
1095
1096 #ifdef CONFIG_MM_OWNER
1097 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1098 {
1099         mm->owner = p;
1100 }
1101 #endif /* CONFIG_MM_OWNER */
1102
1103 /*
1104  * Initialize POSIX timer handling for a single task.
1105  */
1106 static void posix_cpu_timers_init(struct task_struct *tsk)
1107 {
1108         tsk->cputime_expires.prof_exp = 0;
1109         tsk->cputime_expires.virt_exp = 0;
1110         tsk->cputime_expires.sched_exp = 0;
1111         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1112         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1113         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1114 }
1115
1116 /*
1117  * This creates a new process as a copy of the old one,
1118  * but does not actually start it yet.
1119  *
1120  * It copies the registers, and all the appropriate
1121  * parts of the process environment (as per the clone
1122  * flags). The actual kick-off is left to the caller.
1123  */
1124 static struct task_struct *copy_process(unsigned long clone_flags,
1125                                         unsigned long stack_start,
1126                                         struct pt_regs *regs,
1127                                         unsigned long stack_size,
1128                                         int __user *child_tidptr,
1129                                         struct pid *pid,
1130                                         int trace)
1131 {
1132         int retval;
1133         struct task_struct *p;
1134         int cgroup_callbacks_done = 0;
1135
1136         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1137                 return ERR_PTR(-EINVAL);
1138
1139         /*
1140          * Thread groups must share signals as well, and detached threads
1141          * can only be started up within the thread group.
1142          */
1143         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1144                 return ERR_PTR(-EINVAL);
1145
1146         /*
1147          * Shared signal handlers imply shared VM. By way of the above,
1148          * thread groups also imply shared VM. Blocking this case allows
1149          * for various simplifications in other code.
1150          */
1151         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1152                 return ERR_PTR(-EINVAL);
1153
1154         /*
1155          * Siblings of global init remain as zombies on exit since they are
1156          * not reaped by their parent (swapper). To solve this and to avoid
1157          * multi-rooted process trees, prevent global and container-inits
1158          * from creating siblings.
1159          */
1160         if ((clone_flags & CLONE_PARENT) &&
1161                                 current->signal->flags & SIGNAL_UNKILLABLE)
1162                 return ERR_PTR(-EINVAL);
1163
1164         retval = security_task_create(clone_flags);
1165         if (retval)
1166                 goto fork_out;
1167
1168         retval = -ENOMEM;
1169         p = dup_task_struct(current);
1170         if (!p)
1171                 goto fork_out;
1172
1173         ftrace_graph_init_task(p);
1174
1175         rt_mutex_init_task(p);
1176
1177 #ifdef CONFIG_PROVE_LOCKING
1178         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1179         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1180 #endif
1181         retval = -EAGAIN;
1182         if (atomic_read(&p->real_cred->user->processes) >=
1183                         task_rlimit(p, RLIMIT_NPROC)) {
1184                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1185                     p->real_cred->user != INIT_USER)
1186                         goto bad_fork_free;
1187         }
1188         current->flags &= ~PF_NPROC_EXCEEDED;
1189
1190         retval = copy_creds(p, clone_flags);
1191         if (retval < 0)
1192                 goto bad_fork_free;
1193
1194         /*
1195          * If multiple threads are within copy_process(), then this check
1196          * triggers too late. This doesn't hurt, the check is only there
1197          * to stop root fork bombs.
1198          */
1199         retval = -EAGAIN;
1200         if (nr_threads >= max_threads)
1201                 goto bad_fork_cleanup_count;
1202
1203         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1204                 goto bad_fork_cleanup_count;
1205
1206         p->did_exec = 0;
1207         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1208         copy_flags(clone_flags, p);
1209         INIT_LIST_HEAD(&p->children);
1210         INIT_LIST_HEAD(&p->sibling);
1211         rcu_copy_process(p);
1212         p->vfork_done = NULL;
1213         spin_lock_init(&p->alloc_lock);
1214
1215         init_sigpending(&p->pending);
1216
1217         p->utime = p->stime = p->gtime = 0;
1218         p->utimescaled = p->stimescaled = 0;
1219 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1220         p->prev_utime = p->prev_stime = 0;
1221 #endif
1222 #if defined(SPLIT_RSS_COUNTING)
1223         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1224 #endif
1225
1226         p->default_timer_slack_ns = current->timer_slack_ns;
1227
1228         task_io_accounting_init(&p->ioac);
1229         acct_clear_integrals(p);
1230
1231         posix_cpu_timers_init(p);
1232
1233         do_posix_clock_monotonic_gettime(&p->start_time);
1234         p->real_start_time = p->start_time;
1235         monotonic_to_bootbased(&p->real_start_time);
1236         p->io_context = NULL;
1237         p->audit_context = NULL;
1238         if (clone_flags & CLONE_THREAD)
1239                 threadgroup_change_begin(current);
1240         cgroup_fork(p);
1241 #ifdef CONFIG_NUMA
1242         p->mempolicy = mpol_dup(p->mempolicy);
1243         if (IS_ERR(p->mempolicy)) {
1244                 retval = PTR_ERR(p->mempolicy);
1245                 p->mempolicy = NULL;
1246                 goto bad_fork_cleanup_cgroup;
1247         }
1248         mpol_fix_fork_child_flag(p);
1249 #endif
1250 #ifdef CONFIG_CPUSETS
1251         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1252         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1253         seqcount_init(&p->mems_allowed_seq);
1254 #endif
1255 #ifdef CONFIG_TRACE_IRQFLAGS
1256         p->irq_events = 0;
1257 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1258         p->hardirqs_enabled = 1;
1259 #else
1260         p->hardirqs_enabled = 0;
1261 #endif
1262         p->hardirq_enable_ip = 0;
1263         p->hardirq_enable_event = 0;
1264         p->hardirq_disable_ip = _THIS_IP_;
1265         p->hardirq_disable_event = 0;
1266         p->softirqs_enabled = 1;
1267         p->softirq_enable_ip = _THIS_IP_;
1268         p->softirq_enable_event = 0;
1269         p->softirq_disable_ip = 0;
1270         p->softirq_disable_event = 0;
1271         p->hardirq_context = 0;
1272         p->softirq_context = 0;
1273 #endif
1274 #ifdef CONFIG_LOCKDEP
1275         p->lockdep_depth = 0; /* no locks held yet */
1276         p->curr_chain_key = 0;
1277         p->lockdep_recursion = 0;
1278 #endif
1279
1280 #ifdef CONFIG_DEBUG_MUTEXES
1281         p->blocked_on = NULL; /* not blocked yet */
1282 #endif
1283 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1284         p->memcg_batch.do_batch = 0;
1285         p->memcg_batch.memcg = NULL;
1286 #endif
1287
1288         /* Perform scheduler related setup. Assign this task to a CPU. */
1289         sched_fork(p);
1290
1291         retval = perf_event_init_task(p);
1292         if (retval)
1293                 goto bad_fork_cleanup_policy;
1294         retval = audit_alloc(p);
1295         if (retval)
1296                 goto bad_fork_cleanup_policy;
1297         /* copy all the process information */
1298         retval = copy_semundo(clone_flags, p);
1299         if (retval)
1300                 goto bad_fork_cleanup_audit;
1301         retval = copy_files(clone_flags, p);
1302         if (retval)
1303                 goto bad_fork_cleanup_semundo;
1304         retval = copy_fs(clone_flags, p);
1305         if (retval)
1306                 goto bad_fork_cleanup_files;
1307         retval = copy_sighand(clone_flags, p);
1308         if (retval)
1309                 goto bad_fork_cleanup_fs;
1310         retval = copy_signal(clone_flags, p);
1311         if (retval)
1312                 goto bad_fork_cleanup_sighand;
1313         retval = copy_mm(clone_flags, p);
1314         if (retval)
1315                 goto bad_fork_cleanup_signal;
1316         retval = copy_namespaces(clone_flags, p);
1317         if (retval)
1318                 goto bad_fork_cleanup_mm;
1319         retval = copy_io(clone_flags, p);
1320         if (retval)
1321                 goto bad_fork_cleanup_namespaces;
1322         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1323         if (retval)
1324                 goto bad_fork_cleanup_io;
1325
1326         if (pid != &init_struct_pid) {
1327                 retval = -ENOMEM;
1328                 pid = alloc_pid(p->nsproxy->pid_ns);
1329                 if (!pid)
1330                         goto bad_fork_cleanup_io;
1331         }
1332
1333         p->pid = pid_nr(pid);
1334         p->tgid = p->pid;
1335         if (clone_flags & CLONE_THREAD)
1336                 p->tgid = current->tgid;
1337
1338         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1339         /*
1340          * Clear TID on mm_release()?
1341          */
1342         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1343 #ifdef CONFIG_BLOCK
1344         p->plug = NULL;
1345 #endif
1346 #ifdef CONFIG_FUTEX
1347         p->robust_list = NULL;
1348 #ifdef CONFIG_COMPAT
1349         p->compat_robust_list = NULL;
1350 #endif
1351         INIT_LIST_HEAD(&p->pi_state_list);
1352         p->pi_state_cache = NULL;
1353 #endif
1354         uprobe_copy_process(p);
1355         /*
1356          * sigaltstack should be cleared when sharing the same VM
1357          */
1358         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1359                 p->sas_ss_sp = p->sas_ss_size = 0;
1360
1361         /*
1362          * Syscall tracing and stepping should be turned off in the
1363          * child regardless of CLONE_PTRACE.
1364          */
1365         user_disable_single_step(p);
1366         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1367 #ifdef TIF_SYSCALL_EMU
1368         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1369 #endif
1370         clear_all_latency_tracing(p);
1371
1372         /* ok, now we should be set up.. */
1373         if (clone_flags & CLONE_THREAD)
1374                 p->exit_signal = -1;
1375         else if (clone_flags & CLONE_PARENT)
1376                 p->exit_signal = current->group_leader->exit_signal;
1377         else
1378                 p->exit_signal = (clone_flags & CSIGNAL);
1379
1380         p->pdeath_signal = 0;
1381         p->exit_state = 0;
1382
1383         p->nr_dirtied = 0;
1384         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1385         p->dirty_paused_when = 0;
1386
1387         /*
1388          * Ok, make it visible to the rest of the system.
1389          * We dont wake it up yet.
1390          */
1391         p->group_leader = p;
1392         INIT_LIST_HEAD(&p->thread_group);
1393
1394         /* Now that the task is set up, run cgroup callbacks if
1395          * necessary. We need to run them before the task is visible
1396          * on the tasklist. */
1397         cgroup_fork_callbacks(p);
1398         cgroup_callbacks_done = 1;
1399
1400         /* Need tasklist lock for parent etc handling! */
1401         write_lock_irq(&tasklist_lock);
1402
1403         /* CLONE_PARENT re-uses the old parent */
1404         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1405                 p->real_parent = current->real_parent;
1406                 p->parent_exec_id = current->parent_exec_id;
1407         } else {
1408                 p->real_parent = current;
1409                 p->parent_exec_id = current->self_exec_id;
1410         }
1411
1412         spin_lock(&current->sighand->siglock);
1413
1414         /*
1415          * Process group and session signals need to be delivered to just the
1416          * parent before the fork or both the parent and the child after the
1417          * fork. Restart if a signal comes in before we add the new process to
1418          * it's process group.
1419          * A fatal signal pending means that current will exit, so the new
1420          * thread can't slip out of an OOM kill (or normal SIGKILL).
1421         */
1422         recalc_sigpending();
1423         if (signal_pending(current)) {
1424                 spin_unlock(&current->sighand->siglock);
1425                 write_unlock_irq(&tasklist_lock);
1426                 retval = -ERESTARTNOINTR;
1427                 goto bad_fork_free_pid;
1428         }
1429
1430         if (clone_flags & CLONE_THREAD) {
1431                 current->signal->nr_threads++;
1432                 atomic_inc(&current->signal->live);
1433                 atomic_inc(&current->signal->sigcnt);
1434                 p->group_leader = current->group_leader;
1435                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1436         }
1437
1438         if (likely(p->pid)) {
1439                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1440
1441                 if (thread_group_leader(p)) {
1442                         if (is_child_reaper(pid))
1443                                 p->nsproxy->pid_ns->child_reaper = p;
1444
1445                         p->signal->leader_pid = pid;
1446                         p->signal->tty = tty_kref_get(current->signal->tty);
1447                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1448                         attach_pid(p, PIDTYPE_SID, task_session(current));
1449                         list_add_tail(&p->sibling, &p->real_parent->children);
1450                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1451                         __this_cpu_inc(process_counts);
1452                 }
1453                 attach_pid(p, PIDTYPE_PID, pid);
1454                 nr_threads++;
1455         }
1456
1457         total_forks++;
1458         spin_unlock(&current->sighand->siglock);
1459         write_unlock_irq(&tasklist_lock);
1460         proc_fork_connector(p);
1461         cgroup_post_fork(p);
1462         if (clone_flags & CLONE_THREAD)
1463                 threadgroup_change_end(current);
1464         perf_event_fork(p);
1465
1466         trace_task_newtask(p, clone_flags);
1467
1468         return p;
1469
1470 bad_fork_free_pid:
1471         if (pid != &init_struct_pid)
1472                 free_pid(pid);
1473 bad_fork_cleanup_io:
1474         if (p->io_context)
1475                 exit_io_context(p);
1476 bad_fork_cleanup_namespaces:
1477         exit_task_namespaces(p);
1478 bad_fork_cleanup_mm:
1479         if (p->mm)
1480                 mmput(p->mm);
1481 bad_fork_cleanup_signal:
1482         if (!(clone_flags & CLONE_THREAD))
1483                 free_signal_struct(p->signal);
1484 bad_fork_cleanup_sighand:
1485         __cleanup_sighand(p->sighand);
1486 bad_fork_cleanup_fs:
1487         exit_fs(p); /* blocking */
1488 bad_fork_cleanup_files:
1489         exit_files(p); /* blocking */
1490 bad_fork_cleanup_semundo:
1491         exit_sem(p);
1492 bad_fork_cleanup_audit:
1493         audit_free(p);
1494 bad_fork_cleanup_policy:
1495         perf_event_free_task(p);
1496 #ifdef CONFIG_NUMA
1497         mpol_put(p->mempolicy);
1498 bad_fork_cleanup_cgroup:
1499 #endif
1500         if (clone_flags & CLONE_THREAD)
1501                 threadgroup_change_end(current);
1502         cgroup_exit(p, cgroup_callbacks_done);
1503         delayacct_tsk_free(p);
1504         module_put(task_thread_info(p)->exec_domain->module);
1505 bad_fork_cleanup_count:
1506         atomic_dec(&p->cred->user->processes);
1507         exit_creds(p);
1508 bad_fork_free:
1509         free_task(p);
1510 fork_out:
1511         return ERR_PTR(retval);
1512 }
1513
1514 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1515 {
1516         memset(regs, 0, sizeof(struct pt_regs));
1517         return regs;
1518 }
1519
1520 static inline void init_idle_pids(struct pid_link *links)
1521 {
1522         enum pid_type type;
1523
1524         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1525                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1526                 links[type].pid = &init_struct_pid;
1527         }
1528 }
1529
1530 struct task_struct * __cpuinit fork_idle(int cpu)
1531 {
1532         struct task_struct *task;
1533         struct pt_regs regs;
1534
1535         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1536                             &init_struct_pid, 0);
1537         if (!IS_ERR(task)) {
1538                 init_idle_pids(task->pids);
1539                 init_idle(task, cpu);
1540         }
1541
1542         return task;
1543 }
1544
1545 /*
1546  *  Ok, this is the main fork-routine.
1547  *
1548  * It copies the process, and if successful kick-starts
1549  * it and waits for it to finish using the VM if required.
1550  */
1551 long do_fork(unsigned long clone_flags,
1552               unsigned long stack_start,
1553               struct pt_regs *regs,
1554               unsigned long stack_size,
1555               int __user *parent_tidptr,
1556               int __user *child_tidptr)
1557 {
1558         struct task_struct *p;
1559         int trace = 0;
1560         long nr;
1561
1562         /*
1563          * Do some preliminary argument and permissions checking before we
1564          * actually start allocating stuff
1565          */
1566         if (clone_flags & CLONE_NEWUSER) {
1567                 if (clone_flags & CLONE_THREAD)
1568                         return -EINVAL;
1569                 /* hopefully this check will go away when userns support is
1570                  * complete
1571                  */
1572                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1573                                 !capable(CAP_SETGID))
1574                         return -EPERM;
1575         }
1576
1577         /*
1578          * Determine whether and which event to report to ptracer.  When
1579          * called from kernel_thread or CLONE_UNTRACED is explicitly
1580          * requested, no event is reported; otherwise, report if the event
1581          * for the type of forking is enabled.
1582          */
1583         if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1584                 if (clone_flags & CLONE_VFORK)
1585                         trace = PTRACE_EVENT_VFORK;
1586                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1587                         trace = PTRACE_EVENT_CLONE;
1588                 else
1589                         trace = PTRACE_EVENT_FORK;
1590
1591                 if (likely(!ptrace_event_enabled(current, trace)))
1592                         trace = 0;
1593         }
1594
1595         p = copy_process(clone_flags, stack_start, regs, stack_size,
1596                          child_tidptr, NULL, trace);
1597         /*
1598          * Do this prior waking up the new thread - the thread pointer
1599          * might get invalid after that point, if the thread exits quickly.
1600          */
1601         if (!IS_ERR(p)) {
1602                 struct completion vfork;
1603
1604                 trace_sched_process_fork(current, p);
1605
1606                 nr = task_pid_vnr(p);
1607
1608                 if (clone_flags & CLONE_PARENT_SETTID)
1609                         put_user(nr, parent_tidptr);
1610
1611                 if (clone_flags & CLONE_VFORK) {
1612                         p->vfork_done = &vfork;
1613                         init_completion(&vfork);
1614                         get_task_struct(p);
1615                 }
1616
1617                 wake_up_new_task(p);
1618
1619                 /* forking complete and child started to run, tell ptracer */
1620                 if (unlikely(trace))
1621                         ptrace_event(trace, nr);
1622
1623                 if (clone_flags & CLONE_VFORK) {
1624                         if (!wait_for_vfork_done(p, &vfork))
1625                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1626                 }
1627         } else {
1628                 nr = PTR_ERR(p);
1629         }
1630         return nr;
1631 }
1632
1633 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1634 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1635 #endif
1636
1637 static void sighand_ctor(void *data)
1638 {
1639         struct sighand_struct *sighand = data;
1640
1641         spin_lock_init(&sighand->siglock);
1642         init_waitqueue_head(&sighand->signalfd_wqh);
1643 }
1644
1645 void __init proc_caches_init(void)
1646 {
1647         sighand_cachep = kmem_cache_create("sighand_cache",
1648                         sizeof(struct sighand_struct), 0,
1649                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1650                         SLAB_NOTRACK, sighand_ctor);
1651         signal_cachep = kmem_cache_create("signal_cache",
1652                         sizeof(struct signal_struct), 0,
1653                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1654         files_cachep = kmem_cache_create("files_cache",
1655                         sizeof(struct files_struct), 0,
1656                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1657         fs_cachep = kmem_cache_create("fs_cache",
1658                         sizeof(struct fs_struct), 0,
1659                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1660         /*
1661          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1662          * whole struct cpumask for the OFFSTACK case. We could change
1663          * this to *only* allocate as much of it as required by the
1664          * maximum number of CPU's we can ever have.  The cpumask_allocation
1665          * is at the end of the structure, exactly for that reason.
1666          */
1667         mm_cachep = kmem_cache_create("mm_struct",
1668                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1669                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1670         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1671         mmap_init();
1672         nsproxy_cache_init();
1673 }
1674
1675 /*
1676  * Check constraints on flags passed to the unshare system call.
1677  */
1678 static int check_unshare_flags(unsigned long unshare_flags)
1679 {
1680         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1681                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1682                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1683                 return -EINVAL;
1684         /*
1685          * Not implemented, but pretend it works if there is nothing to
1686          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1687          * needs to unshare vm.
1688          */
1689         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1690                 /* FIXME: get_task_mm() increments ->mm_users */
1691                 if (atomic_read(&current->mm->mm_users) > 1)
1692                         return -EINVAL;
1693         }
1694
1695         return 0;
1696 }
1697
1698 /*
1699  * Unshare the filesystem structure if it is being shared
1700  */
1701 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1702 {
1703         struct fs_struct *fs = current->fs;
1704
1705         if (!(unshare_flags & CLONE_FS) || !fs)
1706                 return 0;
1707
1708         /* don't need lock here; in the worst case we'll do useless copy */
1709         if (fs->users == 1)
1710                 return 0;
1711
1712         *new_fsp = copy_fs_struct(fs);
1713         if (!*new_fsp)
1714                 return -ENOMEM;
1715
1716         return 0;
1717 }
1718
1719 /*
1720  * Unshare file descriptor table if it is being shared
1721  */
1722 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1723 {
1724         struct files_struct *fd = current->files;
1725         int error = 0;
1726
1727         if ((unshare_flags & CLONE_FILES) &&
1728             (fd && atomic_read(&fd->count) > 1)) {
1729                 *new_fdp = dup_fd(fd, &error);
1730                 if (!*new_fdp)
1731                         return error;
1732         }
1733
1734         return 0;
1735 }
1736
1737 /*
1738  * unshare allows a process to 'unshare' part of the process
1739  * context which was originally shared using clone.  copy_*
1740  * functions used by do_fork() cannot be used here directly
1741  * because they modify an inactive task_struct that is being
1742  * constructed. Here we are modifying the current, active,
1743  * task_struct.
1744  */
1745 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1746 {
1747         struct fs_struct *fs, *new_fs = NULL;
1748         struct files_struct *fd, *new_fd = NULL;
1749         struct nsproxy *new_nsproxy = NULL;
1750         int do_sysvsem = 0;
1751         int err;
1752
1753         err = check_unshare_flags(unshare_flags);
1754         if (err)
1755                 goto bad_unshare_out;
1756
1757         /*
1758          * If unsharing namespace, must also unshare filesystem information.
1759          */
1760         if (unshare_flags & CLONE_NEWNS)
1761                 unshare_flags |= CLONE_FS;
1762         /*
1763          * CLONE_NEWIPC must also detach from the undolist: after switching
1764          * to a new ipc namespace, the semaphore arrays from the old
1765          * namespace are unreachable.
1766          */
1767         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1768                 do_sysvsem = 1;
1769         err = unshare_fs(unshare_flags, &new_fs);
1770         if (err)
1771                 goto bad_unshare_out;
1772         err = unshare_fd(unshare_flags, &new_fd);
1773         if (err)
1774                 goto bad_unshare_cleanup_fs;
1775         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1776         if (err)
1777                 goto bad_unshare_cleanup_fd;
1778
1779         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1780                 if (do_sysvsem) {
1781                         /*
1782                          * CLONE_SYSVSEM is equivalent to sys_exit().
1783                          */
1784                         exit_sem(current);
1785                 }
1786
1787                 if (new_nsproxy) {
1788                         switch_task_namespaces(current, new_nsproxy);
1789                         new_nsproxy = NULL;
1790                 }
1791
1792                 task_lock(current);
1793
1794                 if (new_fs) {
1795                         fs = current->fs;
1796                         spin_lock(&fs->lock);
1797                         current->fs = new_fs;
1798                         if (--fs->users)
1799                                 new_fs = NULL;
1800                         else
1801                                 new_fs = fs;
1802                         spin_unlock(&fs->lock);
1803                 }
1804
1805                 if (new_fd) {
1806                         fd = current->files;
1807                         current->files = new_fd;
1808                         new_fd = fd;
1809                 }
1810
1811                 task_unlock(current);
1812         }
1813
1814         if (new_nsproxy)
1815                 put_nsproxy(new_nsproxy);
1816
1817 bad_unshare_cleanup_fd:
1818         if (new_fd)
1819                 put_files_struct(new_fd);
1820
1821 bad_unshare_cleanup_fs:
1822         if (new_fs)
1823                 free_fs_struct(new_fs);
1824
1825 bad_unshare_out:
1826         return err;
1827 }
1828
1829 /*
1830  *      Helper to unshare the files of the current task.
1831  *      We don't want to expose copy_files internals to
1832  *      the exec layer of the kernel.
1833  */
1834
1835 int unshare_files(struct files_struct **displaced)
1836 {
1837         struct task_struct *task = current;
1838         struct files_struct *copy = NULL;
1839         int error;
1840
1841         error = unshare_fd(CLONE_FILES, &copy);
1842         if (error || !copy) {
1843                 *displaced = NULL;
1844                 return error;
1845         }
1846         *displaced = task->files;
1847         task_lock(task);
1848         task->files = copy;
1849         task_unlock(task);
1850         return 0;
1851 }