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/mm.h>
  32 #include <linux/vmacache.h>
  33 #include <linux/nsproxy.h>
  34 #include <linux/capability.h>
  35 #include <linux/cpu.h>
  36 #include <linux/cgroup.h>
  37 #include <linux/security.h>
  38 #include <linux/hugetlb.h>
  39 #include <linux/seccomp.h>
  40 #include <linux/swap.h>
  41 #include <linux/syscalls.h>
  42 #include <linux/jiffies.h>
  43 #include <linux/futex.h>
  44 #include <linux/compat.h>
  45 #include <linux/kthread.h>
  46 #include <linux/task_io_accounting_ops.h>
  47 #include <linux/rcupdate.h>
  48 #include <linux/ptrace.h>
  49 #include <linux/mount.h>
  50 #include <linux/audit.h>
  51 #include <linux/memcontrol.h>
  52 #include <linux/ftrace.h>
  53 #include <linux/proc_fs.h>
  54 #include <linux/profile.h>
  55 #include <linux/rmap.h>
  56 #include <linux/ksm.h>
  57 #include <linux/acct.h>
  58 #include <linux/tsacct_kern.h>
  59 #include <linux/cn_proc.h>
  60 #include <linux/freezer.h>
  61 #include <linux/delayacct.h>
  62 #include <linux/taskstats_kern.h>
  63 #include <linux/random.h>
  64 #include <linux/tty.h>
  65 #include <linux/blkdev.h>
  66 #include <linux/fs_struct.h>
  67 #include <linux/magic.h>
  68 #include <linux/perf_event.h>
  69 #include <linux/posix-timers.h>
  70 #include <linux/user-return-notifier.h>
  71 #include <linux/oom.h>
  72 #include <linux/khugepaged.h>
  73 #include <linux/signalfd.h>
  74 #include <linux/uprobes.h>
  75 #include <linux/aio.h>
  76 #include <linux/compiler.h>
  77 #include <linux/sysctl.h>
  78
  79 #include <asm/pgtable.h>
  80 #include <asm/pgalloc.h>
  81 #include <asm/uaccess.h>
  82 #include <asm/mmu_context.h>
  83 #include <asm/cacheflush.h>
  84 #include <asm/tlbflush.h>
  85
  86 #include <trace/events/sched.h>
  87
  88 #define CREATE_TRACE_POINTS
  89 #include <trace/events/task.h>
  90
  91 /*
  92  * Minimum number of threads to boot the kernel
  93  */
  94 #define MIN_THREADS 20
  95
  96 /*
  97  * Maximum number of threads
  98  */
  99 #define MAX_THREADS FUTEX_TID_MASK
 100
 101 /*
 102  * Protected counters by write_lock_irq(&tasklist_lock)
 103  */
 104 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 105 int nr_threads;                 /* The idle threads do not count.. */
 106
 107 int max_threads;                /* tunable limit on nr_threads */
 108
 109 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 110
 111 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 112
 113 #ifdef CONFIG_PROVE_RCU
 114 int lockdep_tasklist_lock_is_held(void)
 115 {
 116         return lockdep_is_held(&tasklist_lock);
 117 }
 118 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 119 #endif /* #ifdef CONFIG_PROVE_RCU */
 120
 121 int nr_processes(void)
 122 {
 123         int cpu;
 124         int total = 0;
 125
 126         for_each_possible_cpu(cpu)
 127                 total += per_cpu(process_counts, cpu);
 128
 129         return total;
 130 }
 131
 132 void __weak arch_release_task_struct(struct task_struct *tsk)
 133 {
 134 }
 135
 136 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 137 static struct kmem_cache *task_struct_cachep;
 138
 139 static inline struct task_struct *alloc_task_struct_node(int node)
 140 {
 141         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 142 }
 143
 144 static inline void free_task_struct(struct task_struct *tsk)
 145 {
 146         kmem_cache_free(task_struct_cachep, tsk);
 147 }
 148 #endif
 149
 150 void __weak arch_release_thread_info(struct thread_info *ti)
 151 {
 152 }
 153
 154 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
 155
 156 /*
 157  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 158  * kmemcache based allocator.
 159  */
 160 # if THREAD_SIZE >= PAGE_SIZE
 161 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 162                                                   int node)
 163 {
 164         struct page *page = alloc_kmem_pages_node(node, THREADINFO_GFP,
 165                                                   THREAD_SIZE_ORDER);
 166
 167         return page ? page_address(page) : NULL;
 168 }
 169
 170 static inline void free_thread_info(struct thread_info *ti)
 171 {
 172         free_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 173 }
 174 # else
 175 static struct kmem_cache *thread_info_cache;
 176
 177 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 178                                                   int node)
 179 {
 180         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
 181 }
 182
 183 static void free_thread_info(struct thread_info *ti)
 184 {
 185         kmem_cache_free(thread_info_cache, ti);
 186 }
 187
 188 void thread_info_cache_init(void)
 189 {
 190         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
 191                                               THREAD_SIZE, 0, NULL);
 192         BUG_ON(thread_info_cache == NULL);
 193 }
 194 # endif
 195 #endif
 196
 197 /* SLAB cache for signal_struct structures (tsk->signal) */
 198 static struct kmem_cache *signal_cachep;
 199
 200 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 201 struct kmem_cache *sighand_cachep;
 202
 203 /* SLAB cache for files_struct structures (tsk->files) */
 204 struct kmem_cache *files_cachep;
 205
 206 /* SLAB cache for fs_struct structures (tsk->fs) */
 207 struct kmem_cache *fs_cachep;
 208
 209 /* SLAB cache for vm_area_struct structures */
 210 struct kmem_cache *vm_area_cachep;
 211
 212 /* SLAB cache for mm_struct structures (tsk->mm) */
 213 static struct kmem_cache *mm_cachep;
 214
 215 static void account_kernel_stack(struct thread_info *ti, int account)
 216 {
 217         struct zone *zone = page_zone(virt_to_page(ti));
 218
 219         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 220 }
 221
 222 void free_task(struct task_struct *tsk)
 223 {
 224         account_kernel_stack(tsk->stack, -1);
 225         arch_release_thread_info(tsk->stack);
 226         free_thread_info(tsk->stack);
 227         rt_mutex_debug_task_free(tsk);
 228         ftrace_graph_exit_task(tsk);
 229         put_seccomp_filter(tsk);
 230         arch_release_task_struct(tsk);
 231         free_task_struct(tsk);
 232 }
 233 EXPORT_SYMBOL(free_task);
 234
 235 static inline void free_signal_struct(struct signal_struct *sig)
 236 {
 237         taskstats_tgid_free(sig);
 238         sched_autogroup_exit(sig);
 239         kmem_cache_free(signal_cachep, sig);
 240 }
 241
 242 static inline void put_signal_struct(struct signal_struct *sig)
 243 {
 244         if (atomic_dec_and_test(&sig->sigcnt))
 245                 free_signal_struct(sig);
 246 }
 247
 248 void __put_task_struct(struct task_struct *tsk)
 249 {
 250         WARN_ON(!tsk->exit_state);
 251         WARN_ON(atomic_read(&tsk->usage));
 252         WARN_ON(tsk == current);
 253
 254         cgroup_free(tsk);
 255         task_numa_free(tsk);
 256         security_task_free(tsk);
 257         exit_creds(tsk);
 258         delayacct_tsk_free(tsk);
 259         put_signal_struct(tsk->signal);
 260
 261         if (!profile_handoff_task(tsk))
 262                 free_task(tsk);
 263 }
 264 EXPORT_SYMBOL_GPL(__put_task_struct);
 265
 266 void __init __weak arch_task_cache_init(void) { }
 267
 268 /*
 269  * set_max_threads
 270  */
 271 static void set_max_threads(unsigned int max_threads_suggested)
 272 {
 273         u64 threads;
 274
 275         /*
 276          * The number of threads shall be limited such that the thread
 277          * structures may only consume a small part of the available memory.
 278          */
 279         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 280                 threads = MAX_THREADS;
 281         else
 282                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 283                                     (u64) THREAD_SIZE * 8UL);
 284
 285         if (threads > max_threads_suggested)
 286                 threads = max_threads_suggested;
 287
 288         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 289 }
 290
 291 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 292 /* Initialized by the architecture: */
 293 int arch_task_struct_size __read_mostly;
 294 #endif
 295
 296 void __init fork_init(void)
 297 {
 298 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 299 #ifndef ARCH_MIN_TASKALIGN
 300 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 301 #endif
 302         /* create a slab on which task_structs can be allocated */
 303         task_struct_cachep =
 304                 kmem_cache_create("task_struct", arch_task_struct_size,
 305                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 306 #endif
 307
 308         /* do the arch specific task caches init */
 309         arch_task_cache_init();
 310
 311         set_max_threads(MAX_THREADS);
 312
 313         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 314         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 315         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 316                 init_task.signal->rlim[RLIMIT_NPROC];
 317 }
 318
 319 int __weak arch_dup_task_struct(struct task_struct *dst,
 320                                                struct task_struct *src)
 321 {
 322         *dst = *src;
 323         return 0;
 324 }
 325
 326 void set_task_stack_end_magic(struct task_struct *tsk)
 327 {
 328         unsigned long *stackend;
 329
 330         stackend = end_of_stack(tsk);
 331         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 332 }
 333
 334 static struct task_struct *dup_task_struct(struct task_struct *orig)
 335 {
 336         struct task_struct *tsk;
 337         struct thread_info *ti;
 338         int node = tsk_fork_get_node(orig);
 339         int err;
 340
 341         tsk = alloc_task_struct_node(node);
 342         if (!tsk)
 343                 return NULL;
 344
 345         ti = alloc_thread_info_node(tsk, node);
 346         if (!ti)
 347                 goto free_tsk;
 348
 349         err = arch_dup_task_struct(tsk, orig);
 350         if (err)
 351                 goto free_ti;
 352
 353         tsk->stack = ti;
 354 #ifdef CONFIG_SECCOMP
 355         /*
 356          * We must handle setting up seccomp filters once we're under
 357          * the sighand lock in case orig has changed between now and
 358          * then. Until then, filter must be NULL to avoid messing up
 359          * the usage counts on the error path calling free_task.
 360          */
 361         tsk->seccomp.filter = NULL;
 362 #endif
 363
 364         setup_thread_stack(tsk, orig);
 365         clear_user_return_notifier(tsk);
 366         clear_tsk_need_resched(tsk);
 367         set_task_stack_end_magic(tsk);
 368
 369 #ifdef CONFIG_CC_STACKPROTECTOR
 370         tsk->stack_canary = get_random_int();
 371 #endif
 372
 373         /*
 374          * One for us, one for whoever does the "release_task()" (usually
 375          * parent)
 376          */
 377         atomic_set(&tsk->usage, 2);
 378 #ifdef CONFIG_BLK_DEV_IO_TRACE
 379         tsk->btrace_seq = 0;
 380 #endif
 381         tsk->splice_pipe = NULL;
 382         tsk->task_frag.page = NULL;
 383
 384         account_kernel_stack(ti, 1);
 385
 386         return tsk;
 387
 388 free_ti:
 389         free_thread_info(ti);
 390 free_tsk:
 391         free_task_struct(tsk);
 392         return NULL;
 393 }
 394
 395 #ifdef CONFIG_MMU
 396 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 397 {
 398         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 399         struct rb_node **rb_link, *rb_parent;
 400         int retval;
 401         unsigned long charge;
 402
 403         uprobe_start_dup_mmap();
 404         down_write(&oldmm->mmap_sem);
 405         flush_cache_dup_mm(oldmm);
 406         uprobe_dup_mmap(oldmm, mm);
 407         /*
 408          * Not linked in yet - no deadlock potential:
 409          */
 410         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 411
 412         /* No ordering required: file already has been exposed. */
 413         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 414
 415         mm->total_vm = oldmm->total_vm;
 416         mm->shared_vm = oldmm->shared_vm;
 417         mm->exec_vm = oldmm->exec_vm;
 418         mm->stack_vm = oldmm->stack_vm;
 419
 420         rb_link = &mm->mm_rb.rb_node;
 421         rb_parent = NULL;
 422         pprev = &mm->mmap;
 423         retval = ksm_fork(mm, oldmm);
 424         if (retval)
 425                 goto out;
 426         retval = khugepaged_fork(mm, oldmm);
 427         if (retval)
 428                 goto out;
 429
 430         prev = NULL;
 431         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 432                 struct file *file;
 433
 434                 if (mpnt->vm_flags & VM_DONTCOPY) {
 435                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 436                                                         -vma_pages(mpnt));
 437                         continue;
 438                 }
 439                 charge = 0;
 440                 if (mpnt->vm_flags & VM_ACCOUNT) {
 441                         unsigned long len = vma_pages(mpnt);
 442
 443                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 444                                 goto fail_nomem;
 445                         charge = len;
 446                 }
 447                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 448                 if (!tmp)
 449                         goto fail_nomem;
 450                 *tmp = *mpnt;
 451                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 452                 retval = vma_dup_policy(mpnt, tmp);
 453                 if (retval)
 454                         goto fail_nomem_policy;
 455                 tmp->vm_mm = mm;
 456                 if (anon_vma_fork(tmp, mpnt))
 457                         goto fail_nomem_anon_vma_fork;
 458                 tmp->vm_flags &= ~(VM_LOCKED|VM_UFFD_MISSING|VM_UFFD_WP);
 459                 tmp->vm_next = tmp->vm_prev = NULL;
 460                 tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
 461                 file = tmp->vm_file;
 462                 if (file) {
 463                         struct inode *inode = file_inode(file);
 464                         struct address_space *mapping = file->f_mapping;
 465
 466                         get_file(file);
 467                         if (tmp->vm_flags & VM_DENYWRITE)
 468                                 atomic_dec(&inode->i_writecount);
 469                         i_mmap_lock_write(mapping);
 470                         if (tmp->vm_flags & VM_SHARED)
 471                                 atomic_inc(&mapping->i_mmap_writable);
 472                         flush_dcache_mmap_lock(mapping);
 473                         /* insert tmp into the share list, just after mpnt */
 474                         vma_interval_tree_insert_after(tmp, mpnt,
 475                                         &mapping->i_mmap);
 476                         flush_dcache_mmap_unlock(mapping);
 477                         i_mmap_unlock_write(mapping);
 478                 }
 479
 480                 /*
 481                  * Clear hugetlb-related page reserves for children. This only
 482                  * affects MAP_PRIVATE mappings. Faults generated by the child
 483                  * are not guaranteed to succeed, even if read-only
 484                  */
 485                 if (is_vm_hugetlb_page(tmp))
 486                         reset_vma_resv_huge_pages(tmp);
 487
 488                 /*
 489                  * Link in the new vma and copy the page table entries.
 490                  */
 491                 *pprev = tmp;
 492                 pprev = &tmp->vm_next;
 493                 tmp->vm_prev = prev;
 494                 prev = tmp;
 495
 496                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 497                 rb_link = &tmp->vm_rb.rb_right;
 498                 rb_parent = &tmp->vm_rb;
 499
 500                 mm->map_count++;
 501                 retval = copy_page_range(mm, oldmm, mpnt);
 502
 503                 if (tmp->vm_ops && tmp->vm_ops->open)
 504                         tmp->vm_ops->open(tmp);
 505
 506                 if (retval)
 507                         goto out;
 508         }
 509         /* a new mm has just been created */
 510         arch_dup_mmap(oldmm, mm);
 511         retval = 0;
 512 out:
 513         up_write(&mm->mmap_sem);
 514         flush_tlb_mm(oldmm);
 515         up_write(&oldmm->mmap_sem);
 516         uprobe_end_dup_mmap();
 517         return retval;
 518 fail_nomem_anon_vma_fork:
 519         mpol_put(vma_policy(tmp));
 520 fail_nomem_policy:
 521         kmem_cache_free(vm_area_cachep, tmp);
 522 fail_nomem:
 523         retval = -ENOMEM;
 524         vm_unacct_memory(charge);
 525         goto out;
 526 }
 527
 528 static inline int mm_alloc_pgd(struct mm_struct *mm)
 529 {
 530         mm->pgd = pgd_alloc(mm);
 531         if (unlikely(!mm->pgd))
 532                 return -ENOMEM;
 533         return 0;
 534 }
 535
 536 static inline void mm_free_pgd(struct mm_struct *mm)
 537 {
 538         pgd_free(mm, mm->pgd);
 539 }
 540 #else
 541 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 542 {
 543         down_write(&oldmm->mmap_sem);
 544         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 545         up_write(&oldmm->mmap_sem);
 546         return 0;
 547 }
 548 #define mm_alloc_pgd(mm)        (0)
 549 #define mm_free_pgd(mm)
 550 #endif /* CONFIG_MMU */
 551
 552 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 553
 554 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 555 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 556
 557 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 558
 559 static int __init coredump_filter_setup(char *s)
 560 {
 561         default_dump_filter =
 562                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 563                 MMF_DUMP_FILTER_MASK;
 564         return 1;
 565 }
 566
 567 __setup("coredump_filter=", coredump_filter_setup);
 568
 569 #include <linux/init_task.h>
 570
 571 static void mm_init_aio(struct mm_struct *mm)
 572 {
 573 #ifdef CONFIG_AIO
 574         spin_lock_init(&mm->ioctx_lock);
 575         mm->ioctx_table = NULL;
 576 #endif
 577 }
 578
 579 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 580 {
 581 #ifdef CONFIG_MEMCG
 582         mm->owner = p;
 583 #endif
 584 }
 585
 586 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 587 {
 588         mm->mmap = NULL;
 589         mm->mm_rb = RB_ROOT;
 590         mm->vmacache_seqnum = 0;
 591         atomic_set(&mm->mm_users, 1);
 592         atomic_set(&mm->mm_count, 1);
 593         init_rwsem(&mm->mmap_sem);
 594         INIT_LIST_HEAD(&mm->mmlist);
 595         mm->core_state = NULL;
 596         atomic_long_set(&mm->nr_ptes, 0);
 597         mm_nr_pmds_init(mm);
 598         mm->map_count = 0;
 599         mm->locked_vm = 0;
 600         mm->pinned_vm = 0;
 601         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 602         spin_lock_init(&mm->page_table_lock);
 603         mm_init_cpumask(mm);
 604         mm_init_aio(mm);
 605         mm_init_owner(mm, p);
 606         mmu_notifier_mm_init(mm);
 607         clear_tlb_flush_pending(mm);
 608 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 609         mm->pmd_huge_pte = NULL;
 610 #endif
 611
 612         if (current->mm) {
 613                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 614                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 615         } else {
 616                 mm->flags = default_dump_filter;
 617                 mm->def_flags = 0;
 618         }
 619
 620         if (mm_alloc_pgd(mm))
 621                 goto fail_nopgd;
 622
 623         if (init_new_context(p, mm))
 624                 goto fail_nocontext;
 625
 626         return mm;
 627
 628 fail_nocontext:
 629         mm_free_pgd(mm);
 630 fail_nopgd:
 631         free_mm(mm);
 632         return NULL;
 633 }
 634
 635 static void check_mm(struct mm_struct *mm)
 636 {
 637         int i;
 638
 639         for (i = 0; i < NR_MM_COUNTERS; i++) {
 640                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 641
 642                 if (unlikely(x))
 643                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 644                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 645         }
 646
 647         if (atomic_long_read(&mm->nr_ptes))
 648                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 649                                 atomic_long_read(&mm->nr_ptes));
 650         if (mm_nr_pmds(mm))
 651                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 652                                 mm_nr_pmds(mm));
 653
 654 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 655         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 656 #endif
 657 }
 658
 659 /*
 660  * Allocate and initialize an mm_struct.
 661  */
 662 struct mm_struct *mm_alloc(void)
 663 {
 664         struct mm_struct *mm;
 665
 666         mm = allocate_mm();
 667         if (!mm)
 668                 return NULL;
 669
 670         memset(mm, 0, sizeof(*mm));
 671         return mm_init(mm, current);
 672 }
 673
 674 /*
 675  * Called when the last reference to the mm
 676  * is dropped: either by a lazy thread or by
 677  * mmput. Free the page directory and the mm.
 678  */
 679 void __mmdrop(struct mm_struct *mm)
 680 {
 681         BUG_ON(mm == &init_mm);
 682         mm_free_pgd(mm);
 683         destroy_context(mm);
 684         mmu_notifier_mm_destroy(mm);
 685         check_mm(mm);
 686         free_mm(mm);
 687 }
 688 EXPORT_SYMBOL_GPL(__mmdrop);
 689
 690 /*
 691  * Decrement the use count and release all resources for an mm.
 692  */
 693 void mmput(struct mm_struct *mm)
 694 {
 695         might_sleep();
 696
 697         if (atomic_dec_and_test(&mm->mm_users)) {
 698                 uprobe_clear_state(mm);
 699                 exit_aio(mm);
 700                 ksm_exit(mm);
 701                 khugepaged_exit(mm); /* must run before exit_mmap */
 702                 exit_mmap(mm);
 703                 set_mm_exe_file(mm, NULL);
 704                 if (!list_empty(&mm->mmlist)) {
 705                         spin_lock(&mmlist_lock);
 706                         list_del(&mm->mmlist);
 707                         spin_unlock(&mmlist_lock);
 708                 }
 709                 if (mm->binfmt)
 710                         module_put(mm->binfmt->module);
 711                 mmdrop(mm);
 712         }
 713 }
 714 EXPORT_SYMBOL_GPL(mmput);
 715
 716 /**
 717  * set_mm_exe_file - change a reference to the mm's executable file
 718  *
 719  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 720  *
 721  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 722  * invocations: in mmput() nobody alive left, in execve task is single
 723  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 724  * mm->exe_file, but does so without using set_mm_exe_file() in order
 725  * to do avoid the need for any locks.
 726  */
 727 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 728 {
 729         struct file *old_exe_file;
 730
 731         /*
 732          * It is safe to dereference the exe_file without RCU as
 733          * this function is only called if nobody else can access
 734          * this mm -- see comment above for justification.
 735          */
 736         old_exe_file = rcu_dereference_raw(mm->exe_file);
 737
 738         if (new_exe_file)
 739                 get_file(new_exe_file);
 740         rcu_assign_pointer(mm->exe_file, new_exe_file);
 741         if (old_exe_file)
 742                 fput(old_exe_file);
 743 }
 744
 745 /**
 746  * get_mm_exe_file - acquire a reference to the mm's executable file
 747  *
 748  * Returns %NULL if mm has no associated executable file.
 749  * User must release file via fput().
 750  */
 751 struct file *get_mm_exe_file(struct mm_struct *mm)
 752 {
 753         struct file *exe_file;
 754
 755         rcu_read_lock();
 756         exe_file = rcu_dereference(mm->exe_file);
 757         if (exe_file && !get_file_rcu(exe_file))
 758                 exe_file = NULL;
 759         rcu_read_unlock();
 760         return exe_file;
 761 }
 762 EXPORT_SYMBOL(get_mm_exe_file);
 763
 764 /**
 765  * get_task_mm - acquire a reference to the task's mm
 766  *
 767  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 768  * this kernel workthread has transiently adopted a user mm with use_mm,
 769  * to do its AIO) is not set and if so returns a reference to it, after
 770  * bumping up the use count.  User must release the mm via mmput()
 771  * after use.  Typically used by /proc and ptrace.
 772  */
 773 struct mm_struct *get_task_mm(struct task_struct *task)
 774 {
 775         struct mm_struct *mm;
 776
 777         task_lock(task);
 778         mm = task->mm;
 779         if (mm) {
 780                 if (task->flags & PF_KTHREAD)
 781                         mm = NULL;
 782                 else
 783                         atomic_inc(&mm->mm_users);
 784         }
 785         task_unlock(task);
 786         return mm;
 787 }
 788 EXPORT_SYMBOL_GPL(get_task_mm);
 789
 790 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 791 {
 792         struct mm_struct *mm;
 793         int err;
 794
 795         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 796         if (err)
 797                 return ERR_PTR(err);
 798
 799         mm = get_task_mm(task);
 800         if (mm && mm != current->mm &&
 801                         !ptrace_may_access(task, mode)) {
 802                 mmput(mm);
 803                 mm = ERR_PTR(-EACCES);
 804         }
 805         mutex_unlock(&task->signal->cred_guard_mutex);
 806
 807         return mm;
 808 }
 809
 810 static void complete_vfork_done(struct task_struct *tsk)
 811 {
 812         struct completion *vfork;
 813
 814         task_lock(tsk);
 815         vfork = tsk->vfork_done;
 816         if (likely(vfork)) {
 817                 tsk->vfork_done = NULL;
 818                 complete(vfork);
 819         }
 820         task_unlock(tsk);
 821 }
 822
 823 static int wait_for_vfork_done(struct task_struct *child,
 824                                 struct completion *vfork)
 825 {
 826         int killed;
 827
 828         freezer_do_not_count();
 829         killed = wait_for_completion_killable(vfork);
 830         freezer_count();
 831
 832         if (killed) {
 833                 task_lock(child);
 834                 child->vfork_done = NULL;
 835                 task_unlock(child);
 836         }
 837
 838         put_task_struct(child);
 839         return killed;
 840 }
 841
 842 /* Please note the differences between mmput and mm_release.
 843  * mmput is called whenever we stop holding onto a mm_struct,
 844  * error success whatever.
 845  *
 846  * mm_release is called after a mm_struct has been removed
 847  * from the current process.
 848  *
 849  * This difference is important for error handling, when we
 850  * only half set up a mm_struct for a new process and need to restore
 851  * the old one.  Because we mmput the new mm_struct before
 852  * restoring the old one. . .
 853  * Eric Biederman 10 January 1998
 854  */
 855 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 856 {
 857         /* Get rid of any futexes when releasing the mm */
 858 #ifdef CONFIG_FUTEX
 859         if (unlikely(tsk->robust_list)) {
 860                 exit_robust_list(tsk);
 861                 tsk->robust_list = NULL;
 862         }
 863 #ifdef CONFIG_COMPAT
 864         if (unlikely(tsk->compat_robust_list)) {
 865                 compat_exit_robust_list(tsk);
 866                 tsk->compat_robust_list = NULL;
 867         }
 868 #endif
 869         if (unlikely(!list_empty(&tsk->pi_state_list)))
 870                 exit_pi_state_list(tsk);
 871 #endif
 872
 873         uprobe_free_utask(tsk);
 874
 875         /* Get rid of any cached register state */
 876         deactivate_mm(tsk, mm);
 877
 878         /*
 879          * If we're exiting normally, clear a user-space tid field if
 880          * requested.  We leave this alone when dying by signal, to leave
 881          * the value intact in a core dump, and to save the unnecessary
 882          * trouble, say, a killed vfork parent shouldn't touch this mm.
 883          * Userland only wants this done for a sys_exit.
 884          */
 885         if (tsk->clear_child_tid) {
 886                 if (!(tsk->flags & PF_SIGNALED) &&
 887                     atomic_read(&mm->mm_users) > 1) {
 888                         /*
 889                          * We don't check the error code - if userspace has
 890                          * not set up a proper pointer then tough luck.
 891                          */
 892                         put_user(0, tsk->clear_child_tid);
 893                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 894                                         1, NULL, NULL, 0);
 895                 }
 896                 tsk->clear_child_tid = NULL;
 897         }
 898
 899         /*
 900          * All done, finally we can wake up parent and return this mm to him.
 901          * Also kthread_stop() uses this completion for synchronization.
 902          */
 903         if (tsk->vfork_done)
 904                 complete_vfork_done(tsk);
 905 }
 906
 907 /*
 908  * Allocate a new mm structure and copy contents from the
 909  * mm structure of the passed in task structure.
 910  */
 911 static struct mm_struct *dup_mm(struct task_struct *tsk)
 912 {
 913         struct mm_struct *mm, *oldmm = current->mm;
 914         int err;
 915
 916         mm = allocate_mm();
 917         if (!mm)
 918                 goto fail_nomem;
 919
 920         memcpy(mm, oldmm, sizeof(*mm));
 921
 922         if (!mm_init(mm, tsk))
 923                 goto fail_nomem;
 924
 925         err = dup_mmap(mm, oldmm);
 926         if (err)
 927                 goto free_pt;
 928
 929         mm->hiwater_rss = get_mm_rss(mm);
 930         mm->hiwater_vm = mm->total_vm;
 931
 932         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 933                 goto free_pt;
 934
 935         return mm;
 936
 937 free_pt:
 938         /* don't put binfmt in mmput, we haven't got module yet */
 939         mm->binfmt = NULL;
 940         mmput(mm);
 941
 942 fail_nomem:
 943         return NULL;
 944 }
 945
 946 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 947 {
 948         struct mm_struct *mm, *oldmm;
 949         int retval;
 950
 951         tsk->min_flt = tsk->maj_flt = 0;
 952         tsk->nvcsw = tsk->nivcsw = 0;
 953 #ifdef CONFIG_DETECT_HUNG_TASK
 954         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 955 #endif
 956
 957         tsk->mm = NULL;
 958         tsk->active_mm = NULL;
 959
 960         /*
 961          * Are we cloning a kernel thread?
 962          *
 963          * We need to steal a active VM for that..
 964          */
 965         oldmm = current->mm;
 966         if (!oldmm)
 967                 return 0;
 968
 969         /* initialize the new vmacache entries */
 970         vmacache_flush(tsk);
 971
 972         if (clone_flags & CLONE_VM) {
 973                 atomic_inc(&oldmm->mm_users);
 974                 mm = oldmm;
 975                 goto good_mm;
 976         }
 977
 978         retval = -ENOMEM;
 979         mm = dup_mm(tsk);
 980         if (!mm)
 981                 goto fail_nomem;
 982
 983 good_mm:
 984         tsk->mm = mm;
 985         tsk->active_mm = mm;
 986         return 0;
 987
 988 fail_nomem:
 989         return retval;
 990 }
 991
 992 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 993 {
 994         struct fs_struct *fs = current->fs;
 995         if (clone_flags & CLONE_FS) {
 996                 /* tsk->fs is already what we want */
 997                 spin_lock(&fs->lock);
 998                 if (fs->in_exec) {
 999                         spin_unlock(&fs->lock);
1000                         return -EAGAIN;
1001                 }
1002                 fs->users++;
1003                 spin_unlock(&fs->lock);
1004                 return 0;
1005         }
1006         tsk->fs = copy_fs_struct(fs);
1007         if (!tsk->fs)
1008                 return -ENOMEM;
1009         return 0;
1010 }
1011
1012 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1013 {
1014         struct files_struct *oldf, *newf;
1015         int error = 0;
1016
1017         /*
1018          * A background process may not have any files ...
1019          */
1020         oldf = current->files;
1021         if (!oldf)
1022                 goto out;
1023
1024         if (clone_flags & CLONE_FILES) {
1025                 atomic_inc(&oldf->count);
1026                 goto out;
1027         }
1028
1029         newf = dup_fd(oldf, &error);
1030         if (!newf)
1031                 goto out;
1032
1033         tsk->files = newf;
1034         error = 0;
1035 out:
1036         return error;
1037 }
1038
1039 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1040 {
1041 #ifdef CONFIG_BLOCK
1042         struct io_context *ioc = current->io_context;
1043         struct io_context *new_ioc;
1044
1045         if (!ioc)
1046                 return 0;
1047         /*
1048          * Share io context with parent, if CLONE_IO is set
1049          */
1050         if (clone_flags & CLONE_IO) {
1051                 ioc_task_link(ioc);
1052                 tsk->io_context = ioc;
1053         } else if (ioprio_valid(ioc->ioprio)) {
1054                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1055                 if (unlikely(!new_ioc))
1056                         return -ENOMEM;
1057
1058                 new_ioc->ioprio = ioc->ioprio;
1059                 put_io_context(new_ioc);
1060         }
1061 #endif
1062         return 0;
1063 }
1064
1065 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1066 {
1067         struct sighand_struct *sig;
1068
1069         if (clone_flags & CLONE_SIGHAND) {
1070                 atomic_inc(&current->sighand->count);
1071                 return 0;
1072         }
1073         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1074         rcu_assign_pointer(tsk->sighand, sig);
1075         if (!sig)
1076                 return -ENOMEM;
1077
1078         atomic_set(&sig->count, 1);
1079         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1080         return 0;
1081 }
1082
1083 void __cleanup_sighand(struct sighand_struct *sighand)
1084 {
1085         if (atomic_dec_and_test(&sighand->count)) {
1086                 signalfd_cleanup(sighand);
1087                 /*
1088                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1089                  * without an RCU grace period, see __lock_task_sighand().
1090                  */
1091                 kmem_cache_free(sighand_cachep, sighand);
1092         }
1093 }
1094
1095 /*
1096  * Initialize POSIX timer handling for a thread group.
1097  */
1098 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1099 {
1100         unsigned long cpu_limit;
1101
1102         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1103         if (cpu_limit != RLIM_INFINITY) {
1104                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1105                 sig->cputimer.running = true;
1106         }
1107
1108         /* The timer lists. */
1109         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1110         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1111         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1112 }
1113
1114 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1115 {
1116         struct signal_struct *sig;
1117
1118         if (clone_flags & CLONE_THREAD)
1119                 return 0;
1120
1121         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1122         tsk->signal = sig;
1123         if (!sig)
1124                 return -ENOMEM;
1125
1126         sig->nr_threads = 1;
1127         atomic_set(&sig->live, 1);
1128         atomic_set(&sig->sigcnt, 1);
1129
1130         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1131         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1132         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1133
1134         init_waitqueue_head(&sig->wait_chldexit);
1135         sig->curr_target = tsk;
1136         init_sigpending(&sig->shared_pending);
1137         INIT_LIST_HEAD(&sig->posix_timers);
1138         seqlock_init(&sig->stats_lock);
1139         prev_cputime_init(&sig->prev_cputime);
1140
1141         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1142         sig->real_timer.function = it_real_fn;
1143
1144         task_lock(current->group_leader);
1145         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1146         task_unlock(current->group_leader);
1147
1148         posix_cpu_timers_init_group(sig);
1149
1150         tty_audit_fork(sig);
1151         sched_autogroup_fork(sig);
1152
1153         sig->oom_score_adj = current->signal->oom_score_adj;
1154         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1155
1156         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1157                                    current->signal->is_child_subreaper;
1158
1159         mutex_init(&sig->cred_guard_mutex);
1160
1161         return 0;
1162 }
1163
1164 static void copy_seccomp(struct task_struct *p)
1165 {
1166 #ifdef CONFIG_SECCOMP
1167         /*
1168          * Must be called with sighand->lock held, which is common to
1169          * all threads in the group. Holding cred_guard_mutex is not
1170          * needed because this new task is not yet running and cannot
1171          * be racing exec.
1172          */
1173         assert_spin_locked(&current->sighand->siglock);
1174
1175         /* Ref-count the new filter user, and assign it. */
1176         get_seccomp_filter(current);
1177         p->seccomp = current->seccomp;
1178
1179         /*
1180          * Explicitly enable no_new_privs here in case it got set
1181          * between the task_struct being duplicated and holding the
1182          * sighand lock. The seccomp state and nnp must be in sync.
1183          */
1184         if (task_no_new_privs(current))
1185                 task_set_no_new_privs(p);
1186
1187         /*
1188          * If the parent gained a seccomp mode after copying thread
1189          * flags and between before we held the sighand lock, we have
1190          * to manually enable the seccomp thread flag here.
1191          */
1192         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1193                 set_tsk_thread_flag(p, TIF_SECCOMP);
1194 #endif
1195 }
1196
1197 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1198 {
1199         current->clear_child_tid = tidptr;
1200
1201         return task_pid_vnr(current);
1202 }
1203
1204 static void rt_mutex_init_task(struct task_struct *p)
1205 {
1206         raw_spin_lock_init(&p->pi_lock);
1207 #ifdef CONFIG_RT_MUTEXES
1208         p->pi_waiters = RB_ROOT;
1209         p->pi_waiters_leftmost = NULL;
1210         p->pi_blocked_on = NULL;
1211 #endif
1212 }
1213
1214 /*
1215  * Initialize POSIX timer handling for a single task.
1216  */
1217 static void posix_cpu_timers_init(struct task_struct *tsk)
1218 {
1219         tsk->cputime_expires.prof_exp = 0;
1220         tsk->cputime_expires.virt_exp = 0;
1221         tsk->cputime_expires.sched_exp = 0;
1222         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1223         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1224         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1225 }
1226
1227 static inline void
1228 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1229 {
1230          task->pids[type].pid = pid;
1231 }
1232
1233 /*
1234  * This creates a new process as a copy of the old one,
1235  * but does not actually start it yet.
1236  *
1237  * It copies the registers, and all the appropriate
1238  * parts of the process environment (as per the clone
1239  * flags). The actual kick-off is left to the caller.
1240  */
1241 static struct task_struct *copy_process(unsigned long clone_flags,
1242                                         unsigned long stack_start,
1243                                         unsigned long stack_size,
1244                                         int __user *child_tidptr,
1245                                         struct pid *pid,
1246                                         int trace,
1247                                         unsigned long tls)
1248 {
1249         int retval;
1250         struct task_struct *p;
1251         void *cgrp_ss_priv[CGROUP_CANFORK_COUNT] = {};
1252
1253         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1254                 return ERR_PTR(-EINVAL);
1255
1256         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1257                 return ERR_PTR(-EINVAL);
1258
1259         /*
1260          * Thread groups must share signals as well, and detached threads
1261          * can only be started up within the thread group.
1262          */
1263         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1264                 return ERR_PTR(-EINVAL);
1265
1266         /*
1267          * Shared signal handlers imply shared VM. By way of the above,
1268          * thread groups also imply shared VM. Blocking this case allows
1269          * for various simplifications in other code.
1270          */
1271         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1272                 return ERR_PTR(-EINVAL);
1273
1274         /*
1275          * Siblings of global init remain as zombies on exit since they are
1276          * not reaped by their parent (swapper). To solve this and to avoid
1277          * multi-rooted process trees, prevent global and container-inits
1278          * from creating siblings.
1279          */
1280         if ((clone_flags & CLONE_PARENT) &&
1281                                 current->signal->flags & SIGNAL_UNKILLABLE)
1282                 return ERR_PTR(-EINVAL);
1283
1284         /*
1285          * If the new process will be in a different pid or user namespace
1286          * do not allow it to share a thread group with the forking task.
1287          */
1288         if (clone_flags & CLONE_THREAD) {
1289                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1290                     (task_active_pid_ns(current) !=
1291                                 current->nsproxy->pid_ns_for_children))
1292                         return ERR_PTR(-EINVAL);
1293         }
1294
1295         retval = security_task_create(clone_flags);
1296         if (retval)
1297                 goto fork_out;
1298
1299         retval = -ENOMEM;
1300         p = dup_task_struct(current);
1301         if (!p)
1302                 goto fork_out;
1303
1304         ftrace_graph_init_task(p);
1305
1306         rt_mutex_init_task(p);
1307
1308 #ifdef CONFIG_PROVE_LOCKING
1309         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1310         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1311 #endif
1312         retval = -EAGAIN;
1313         if (atomic_read(&p->real_cred->user->processes) >=
1314                         task_rlimit(p, RLIMIT_NPROC)) {
1315                 if (p->real_cred->user != INIT_USER &&
1316                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1317                         goto bad_fork_free;
1318         }
1319         current->flags &= ~PF_NPROC_EXCEEDED;
1320
1321         retval = copy_creds(p, clone_flags);
1322         if (retval < 0)
1323                 goto bad_fork_free;
1324
1325         /*
1326          * If multiple threads are within copy_process(), then this check
1327          * triggers too late. This doesn't hurt, the check is only there
1328          * to stop root fork bombs.
1329          */
1330         retval = -EAGAIN;
1331         if (nr_threads >= max_threads)
1332                 goto bad_fork_cleanup_count;
1333
1334         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1335         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1336         p->flags |= PF_FORKNOEXEC;
1337         INIT_LIST_HEAD(&p->children);
1338         INIT_LIST_HEAD(&p->sibling);
1339         rcu_copy_process(p);
1340         p->vfork_done = NULL;
1341         spin_lock_init(&p->alloc_lock);
1342
1343         init_sigpending(&p->pending);
1344
1345         p->utime = p->stime = p->gtime = 0;
1346         p->utimescaled = p->stimescaled = 0;
1347         prev_cputime_init(&p->prev_cputime);
1348
1349 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1350         seqlock_init(&p->vtime_seqlock);
1351         p->vtime_snap = 0;
1352         p->vtime_snap_whence = VTIME_SLEEPING;
1353 #endif
1354
1355 #if defined(SPLIT_RSS_COUNTING)
1356         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1357 #endif
1358
1359         p->default_timer_slack_ns = current->timer_slack_ns;
1360
1361         task_io_accounting_init(&p->ioac);
1362         acct_clear_integrals(p);
1363
1364         posix_cpu_timers_init(p);
1365
1366         p->start_time = ktime_get_ns();
1367         p->real_start_time = ktime_get_boot_ns();
1368         p->io_context = NULL;
1369         p->audit_context = NULL;
1370         if (clone_flags & CLONE_THREAD)
1371                 threadgroup_change_begin(current);
1372         cgroup_fork(p);
1373 #ifdef CONFIG_NUMA
1374         p->mempolicy = mpol_dup(p->mempolicy);
1375         if (IS_ERR(p->mempolicy)) {
1376                 retval = PTR_ERR(p->mempolicy);
1377                 p->mempolicy = NULL;
1378                 goto bad_fork_cleanup_threadgroup_lock;
1379         }
1380 #endif
1381 #ifdef CONFIG_CPUSETS
1382         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1383         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1384         seqcount_init(&p->mems_allowed_seq);
1385 #endif
1386 #ifdef CONFIG_TRACE_IRQFLAGS
1387         p->irq_events = 0;
1388         p->hardirqs_enabled = 0;
1389         p->hardirq_enable_ip = 0;
1390         p->hardirq_enable_event = 0;
1391         p->hardirq_disable_ip = _THIS_IP_;
1392         p->hardirq_disable_event = 0;
1393         p->softirqs_enabled = 1;
1394         p->softirq_enable_ip = _THIS_IP_;
1395         p->softirq_enable_event = 0;
1396         p->softirq_disable_ip = 0;
1397         p->softirq_disable_event = 0;
1398         p->hardirq_context = 0;
1399         p->softirq_context = 0;
1400 #endif
1401
1402         p->pagefault_disabled = 0;
1403
1404 #ifdef CONFIG_LOCKDEP
1405         p->lockdep_depth = 0; /* no locks held yet */
1406         p->curr_chain_key = 0;
1407         p->lockdep_recursion = 0;
1408 #endif
1409
1410 #ifdef CONFIG_DEBUG_MUTEXES
1411         p->blocked_on = NULL; /* not blocked yet */
1412 #endif
1413 #ifdef CONFIG_BCACHE
1414         p->sequential_io        = 0;
1415         p->sequential_io_avg    = 0;
1416 #endif
1417
1418         /* Perform scheduler related setup. Assign this task to a CPU. */
1419         retval = sched_fork(clone_flags, p);
1420         if (retval)
1421                 goto bad_fork_cleanup_policy;
1422
1423         retval = perf_event_init_task(p);
1424         if (retval)
1425                 goto bad_fork_cleanup_policy;
1426         retval = audit_alloc(p);
1427         if (retval)
1428                 goto bad_fork_cleanup_perf;
1429         /* copy all the process information */
1430         shm_init_task(p);
1431         retval = copy_semundo(clone_flags, p);
1432         if (retval)
1433                 goto bad_fork_cleanup_audit;
1434         retval = copy_files(clone_flags, p);
1435         if (retval)
1436                 goto bad_fork_cleanup_semundo;
1437         retval = copy_fs(clone_flags, p);
1438         if (retval)
1439                 goto bad_fork_cleanup_files;
1440         retval = copy_sighand(clone_flags, p);
1441         if (retval)
1442                 goto bad_fork_cleanup_fs;
1443         retval = copy_signal(clone_flags, p);
1444         if (retval)
1445                 goto bad_fork_cleanup_sighand;
1446         retval = copy_mm(clone_flags, p);
1447         if (retval)
1448                 goto bad_fork_cleanup_signal;
1449         retval = copy_namespaces(clone_flags, p);
1450         if (retval)
1451                 goto bad_fork_cleanup_mm;
1452         retval = copy_io(clone_flags, p);
1453         if (retval)
1454                 goto bad_fork_cleanup_namespaces;
1455         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1456         if (retval)
1457                 goto bad_fork_cleanup_io;
1458
1459         if (pid != &init_struct_pid) {
1460                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1461                 if (IS_ERR(pid)) {
1462                         retval = PTR_ERR(pid);
1463                         goto bad_fork_cleanup_io;
1464                 }
1465         }
1466
1467         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1468         /*
1469          * Clear TID on mm_release()?
1470          */
1471         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1472 #ifdef CONFIG_BLOCK
1473         p->plug = NULL;
1474 #endif
1475 #ifdef CONFIG_FUTEX
1476         p->robust_list = NULL;
1477 #ifdef CONFIG_COMPAT
1478         p->compat_robust_list = NULL;
1479 #endif
1480         INIT_LIST_HEAD(&p->pi_state_list);
1481         p->pi_state_cache = NULL;
1482 #endif
1483         /*
1484          * sigaltstack should be cleared when sharing the same VM
1485          */
1486         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1487                 p->sas_ss_sp = p->sas_ss_size = 0;
1488
1489         /*
1490          * Syscall tracing and stepping should be turned off in the
1491          * child regardless of CLONE_PTRACE.
1492          */
1493         user_disable_single_step(p);
1494         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1495 #ifdef TIF_SYSCALL_EMU
1496         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1497 #endif
1498         clear_all_latency_tracing(p);
1499
1500         /* ok, now we should be set up.. */
1501         p->pid = pid_nr(pid);
1502         if (clone_flags & CLONE_THREAD) {
1503                 p->exit_signal = -1;
1504                 p->group_leader = current->group_leader;
1505                 p->tgid = current->tgid;
1506         } else {
1507                 if (clone_flags & CLONE_PARENT)
1508                         p->exit_signal = current->group_leader->exit_signal;
1509                 else
1510                         p->exit_signal = (clone_flags & CSIGNAL);
1511                 p->group_leader = p;
1512                 p->tgid = p->pid;
1513         }
1514
1515         p->nr_dirtied = 0;
1516         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1517         p->dirty_paused_when = 0;
1518
1519         p->pdeath_signal = 0;
1520         INIT_LIST_HEAD(&p->thread_group);
1521         p->task_works = NULL;
1522
1523         /*
1524          * Ensure that the cgroup subsystem policies allow the new process to be
1525          * forked. It should be noted the the new process's css_set can be changed
1526          * between here and cgroup_post_fork() if an organisation operation is in
1527          * progress.
1528          */
1529         retval = cgroup_can_fork(p, cgrp_ss_priv);
1530         if (retval)
1531                 goto bad_fork_free_pid;
1532
1533         /*
1534          * Make it visible to the rest of the system, but dont wake it up yet.
1535          * Need tasklist lock for parent etc handling!
1536          */
1537         write_lock_irq(&tasklist_lock);
1538
1539         /* CLONE_PARENT re-uses the old parent */
1540         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1541                 p->real_parent = current->real_parent;
1542                 p->parent_exec_id = current->parent_exec_id;
1543         } else {
1544                 p->real_parent = current;
1545                 p->parent_exec_id = current->self_exec_id;
1546         }
1547
1548         spin_lock(&current->sighand->siglock);
1549
1550         /*
1551          * Copy seccomp details explicitly here, in case they were changed
1552          * before holding sighand lock.
1553          */
1554         copy_seccomp(p);
1555
1556         /*
1557          * Process group and session signals need to be delivered to just the
1558          * parent before the fork or both the parent and the child after the
1559          * fork. Restart if a signal comes in before we add the new process to
1560          * it's process group.
1561          * A fatal signal pending means that current will exit, so the new
1562          * thread can't slip out of an OOM kill (or normal SIGKILL).
1563         */
1564         recalc_sigpending();
1565         if (signal_pending(current)) {
1566                 spin_unlock(&current->sighand->siglock);
1567                 write_unlock_irq(&tasklist_lock);
1568                 retval = -ERESTARTNOINTR;
1569                 goto bad_fork_cancel_cgroup;
1570         }
1571
1572         if (likely(p->pid)) {
1573                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1574
1575                 init_task_pid(p, PIDTYPE_PID, pid);
1576                 if (thread_group_leader(p)) {
1577                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1578                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1579
1580                         if (is_child_reaper(pid)) {
1581                                 ns_of_pid(pid)->child_reaper = p;
1582                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1583                         }
1584
1585                         p->signal->leader_pid = pid;
1586                         p->signal->tty = tty_kref_get(current->signal->tty);
1587                         list_add_tail(&p->sibling, &p->real_parent->children);
1588                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1589                         attach_pid(p, PIDTYPE_PGID);
1590                         attach_pid(p, PIDTYPE_SID);
1591                         __this_cpu_inc(process_counts);
1592                 } else {
1593                         current->signal->nr_threads++;
1594                         atomic_inc(&current->signal->live);
1595                         atomic_inc(&current->signal->sigcnt);
1596                         list_add_tail_rcu(&p->thread_group,
1597                                           &p->group_leader->thread_group);
1598                         list_add_tail_rcu(&p->thread_node,
1599                                           &p->signal->thread_head);
1600                 }
1601                 attach_pid(p, PIDTYPE_PID);
1602                 nr_threads++;
1603         }
1604
1605         total_forks++;
1606         spin_unlock(&current->sighand->siglock);
1607         syscall_tracepoint_update(p);
1608         write_unlock_irq(&tasklist_lock);
1609
1610         proc_fork_connector(p);
1611         cgroup_post_fork(p, cgrp_ss_priv);
1612         if (clone_flags & CLONE_THREAD)
1613                 threadgroup_change_end(current);
1614         perf_event_fork(p);
1615
1616         trace_task_newtask(p, clone_flags);
1617         uprobe_copy_process(p, clone_flags);
1618
1619         return p;
1620
1621 bad_fork_cancel_cgroup:
1622         cgroup_cancel_fork(p, cgrp_ss_priv);
1623 bad_fork_free_pid:
1624         if (pid != &init_struct_pid)
1625                 free_pid(pid);
1626 bad_fork_cleanup_io:
1627         if (p->io_context)
1628                 exit_io_context(p);
1629 bad_fork_cleanup_namespaces:
1630         exit_task_namespaces(p);
1631 bad_fork_cleanup_mm:
1632         if (p->mm)
1633                 mmput(p->mm);
1634 bad_fork_cleanup_signal:
1635         if (!(clone_flags & CLONE_THREAD))
1636                 free_signal_struct(p->signal);
1637 bad_fork_cleanup_sighand:
1638         __cleanup_sighand(p->sighand);
1639 bad_fork_cleanup_fs:
1640         exit_fs(p); /* blocking */
1641 bad_fork_cleanup_files:
1642         exit_files(p); /* blocking */
1643 bad_fork_cleanup_semundo:
1644         exit_sem(p);
1645 bad_fork_cleanup_audit:
1646         audit_free(p);
1647 bad_fork_cleanup_perf:
1648         perf_event_free_task(p);
1649 bad_fork_cleanup_policy:
1650 #ifdef CONFIG_NUMA
1651         mpol_put(p->mempolicy);
1652 bad_fork_cleanup_threadgroup_lock:
1653 #endif
1654         if (clone_flags & CLONE_THREAD)
1655                 threadgroup_change_end(current);
1656         delayacct_tsk_free(p);
1657 bad_fork_cleanup_count:
1658         atomic_dec(&p->cred->user->processes);
1659         exit_creds(p);
1660 bad_fork_free:
1661         free_task(p);
1662 fork_out:
1663         return ERR_PTR(retval);
1664 }
1665
1666 static inline void init_idle_pids(struct pid_link *links)
1667 {
1668         enum pid_type type;
1669
1670         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1671                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1672                 links[type].pid = &init_struct_pid;
1673         }
1674 }
1675
1676 struct task_struct *fork_idle(int cpu)
1677 {
1678         struct task_struct *task;
1679         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0);
1680         if (!IS_ERR(task)) {
1681                 init_idle_pids(task->pids);
1682                 init_idle(task, cpu);
1683         }
1684
1685         return task;
1686 }
1687
1688 /*
1689  *  Ok, this is the main fork-routine.
1690  *
1691  * It copies the process, and if successful kick-starts
1692  * it and waits for it to finish using the VM if required.
1693  */
1694 long _do_fork(unsigned long clone_flags,
1695               unsigned long stack_start,
1696               unsigned long stack_size,
1697               int __user *parent_tidptr,
1698               int __user *child_tidptr,
1699               unsigned long tls)
1700 {
1701         struct task_struct *p;
1702         int trace = 0;
1703         long nr;
1704
1705         /*
1706          * Determine whether and which event to report to ptracer.  When
1707          * called from kernel_thread or CLONE_UNTRACED is explicitly
1708          * requested, no event is reported; otherwise, report if the event
1709          * for the type of forking is enabled.
1710          */
1711         if (!(clone_flags & CLONE_UNTRACED)) {
1712                 if (clone_flags & CLONE_VFORK)
1713                         trace = PTRACE_EVENT_VFORK;
1714                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1715                         trace = PTRACE_EVENT_CLONE;
1716                 else
1717                         trace = PTRACE_EVENT_FORK;
1718
1719                 if (likely(!ptrace_event_enabled(current, trace)))
1720                         trace = 0;
1721         }
1722
1723         p = copy_process(clone_flags, stack_start, stack_size,
1724                          child_tidptr, NULL, trace, tls);
1725         /*
1726          * Do this prior waking up the new thread - the thread pointer
1727          * might get invalid after that point, if the thread exits quickly.
1728          */
1729         if (!IS_ERR(p)) {
1730                 struct completion vfork;
1731                 struct pid *pid;
1732
1733                 trace_sched_process_fork(current, p);
1734
1735                 pid = get_task_pid(p, PIDTYPE_PID);
1736                 nr = pid_vnr(pid);
1737
1738                 if (clone_flags & CLONE_PARENT_SETTID)
1739                         put_user(nr, parent_tidptr);
1740
1741                 if (clone_flags & CLONE_VFORK) {
1742                         p->vfork_done = &vfork;
1743                         init_completion(&vfork);
1744                         get_task_struct(p);
1745                 }
1746
1747                 wake_up_new_task(p);
1748
1749                 /* forking complete and child started to run, tell ptracer */
1750                 if (unlikely(trace))
1751                         ptrace_event_pid(trace, pid);
1752
1753                 if (clone_flags & CLONE_VFORK) {
1754                         if (!wait_for_vfork_done(p, &vfork))
1755                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1756                 }
1757
1758                 put_pid(pid);
1759         } else {
1760                 nr = PTR_ERR(p);
1761         }
1762         return nr;
1763 }
1764
1765 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1766 /* For compatibility with architectures that call do_fork directly rather than
1767  * using the syscall entry points below. */
1768 long do_fork(unsigned long clone_flags,
1769               unsigned long stack_start,
1770               unsigned long stack_size,
1771               int __user *parent_tidptr,
1772               int __user *child_tidptr)
1773 {
1774         return _do_fork(clone_flags, stack_start, stack_size,
1775                         parent_tidptr, child_tidptr, 0);
1776 }
1777 #endif
1778
1779 /*
1780  * Create a kernel thread.
1781  */
1782 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1783 {
1784         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1785                 (unsigned long)arg, NULL, NULL, 0);
1786 }
1787
1788 #ifdef __ARCH_WANT_SYS_FORK
1789 SYSCALL_DEFINE0(fork)
1790 {
1791 #ifdef CONFIG_MMU
1792         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1793 #else
1794         /* can not support in nommu mode */
1795         return -EINVAL;
1796 #endif
1797 }
1798 #endif
1799
1800 #ifdef __ARCH_WANT_SYS_VFORK
1801 SYSCALL_DEFINE0(vfork)
1802 {
1803         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1804                         0, NULL, NULL, 0);
1805 }
1806 #endif
1807
1808 #ifdef __ARCH_WANT_SYS_CLONE
1809 #ifdef CONFIG_CLONE_BACKWARDS
1810 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1811                  int __user *, parent_tidptr,
1812                  unsigned long, tls,
1813                  int __user *, child_tidptr)
1814 #elif defined(CONFIG_CLONE_BACKWARDS2)
1815 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1816                  int __user *, parent_tidptr,
1817                  int __user *, child_tidptr,
1818                  unsigned long, tls)
1819 #elif defined(CONFIG_CLONE_BACKWARDS3)
1820 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1821                 int, stack_size,
1822                 int __user *, parent_tidptr,
1823                 int __user *, child_tidptr,
1824                 unsigned long, tls)
1825 #else
1826 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1827                  int __user *, parent_tidptr,
1828                  int __user *, child_tidptr,
1829                  unsigned long, tls)
1830 #endif
1831 {
1832         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
1833 }
1834 #endif
1835
1836 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1837 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1838 #endif
1839
1840 static void sighand_ctor(void *data)
1841 {
1842         struct sighand_struct *sighand = data;
1843
1844         spin_lock_init(&sighand->siglock);
1845         init_waitqueue_head(&sighand->signalfd_wqh);
1846 }
1847
1848 void __init proc_caches_init(void)
1849 {
1850         sighand_cachep = kmem_cache_create("sighand_cache",
1851                         sizeof(struct sighand_struct), 0,
1852                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1853                         SLAB_NOTRACK, sighand_ctor);
1854         signal_cachep = kmem_cache_create("signal_cache",
1855                         sizeof(struct signal_struct), 0,
1856                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1857         files_cachep = kmem_cache_create("files_cache",
1858                         sizeof(struct files_struct), 0,
1859                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1860         fs_cachep = kmem_cache_create("fs_cache",
1861                         sizeof(struct fs_struct), 0,
1862                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1863         /*
1864          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1865          * whole struct cpumask for the OFFSTACK case. We could change
1866          * this to *only* allocate as much of it as required by the
1867          * maximum number of CPU's we can ever have.  The cpumask_allocation
1868          * is at the end of the structure, exactly for that reason.
1869          */
1870         mm_cachep = kmem_cache_create("mm_struct",
1871                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1872                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1873         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1874         mmap_init();
1875         nsproxy_cache_init();
1876 }
1877
1878 /*
1879  * Check constraints on flags passed to the unshare system call.
1880  */
1881 static int check_unshare_flags(unsigned long unshare_flags)
1882 {
1883         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1884                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1885                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1886                                 CLONE_NEWUSER|CLONE_NEWPID))
1887                 return -EINVAL;
1888         /*
1889          * Not implemented, but pretend it works if there is nothing
1890          * to unshare.  Note that unsharing the address space or the
1891          * signal handlers also need to unshare the signal queues (aka
1892          * CLONE_THREAD).
1893          */
1894         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1895                 if (!thread_group_empty(current))
1896                         return -EINVAL;
1897         }
1898         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
1899                 if (atomic_read(&current->sighand->count) > 1)
1900                         return -EINVAL;
1901         }
1902         if (unshare_flags & CLONE_VM) {
1903                 if (!current_is_single_threaded())
1904                         return -EINVAL;
1905         }
1906
1907         return 0;
1908 }
1909
1910 /*
1911  * Unshare the filesystem structure if it is being shared
1912  */
1913 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1914 {
1915         struct fs_struct *fs = current->fs;
1916
1917         if (!(unshare_flags & CLONE_FS) || !fs)
1918                 return 0;
1919
1920         /* don't need lock here; in the worst case we'll do useless copy */
1921         if (fs->users == 1)
1922                 return 0;
1923
1924         *new_fsp = copy_fs_struct(fs);
1925         if (!*new_fsp)
1926                 return -ENOMEM;
1927
1928         return 0;
1929 }
1930
1931 /*
1932  * Unshare file descriptor table if it is being shared
1933  */
1934 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1935 {
1936         struct files_struct *fd = current->files;
1937         int error = 0;
1938
1939         if ((unshare_flags & CLONE_FILES) &&
1940             (fd && atomic_read(&fd->count) > 1)) {
1941                 *new_fdp = dup_fd(fd, &error);
1942                 if (!*new_fdp)
1943                         return error;
1944         }
1945
1946         return 0;
1947 }
1948
1949 /*
1950  * unshare allows a process to 'unshare' part of the process
1951  * context which was originally shared using clone.  copy_*
1952  * functions used by do_fork() cannot be used here directly
1953  * because they modify an inactive task_struct that is being
1954  * constructed. Here we are modifying the current, active,
1955  * task_struct.
1956  */
1957 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1958 {
1959         struct fs_struct *fs, *new_fs = NULL;
1960         struct files_struct *fd, *new_fd = NULL;
1961         struct cred *new_cred = NULL;
1962         struct nsproxy *new_nsproxy = NULL;
1963         int do_sysvsem = 0;
1964         int err;
1965
1966         /*
1967          * If unsharing a user namespace must also unshare the thread group
1968          * and unshare the filesystem root and working directories.
1969          */
1970         if (unshare_flags & CLONE_NEWUSER)
1971                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1972         /*
1973          * If unsharing vm, must also unshare signal handlers.
1974          */
1975         if (unshare_flags & CLONE_VM)
1976                 unshare_flags |= CLONE_SIGHAND;
1977         /*
1978          * If unsharing a signal handlers, must also unshare the signal queues.
1979          */
1980         if (unshare_flags & CLONE_SIGHAND)
1981                 unshare_flags |= CLONE_THREAD;
1982         /*
1983          * If unsharing namespace, must also unshare filesystem information.
1984          */
1985         if (unshare_flags & CLONE_NEWNS)
1986                 unshare_flags |= CLONE_FS;
1987
1988         err = check_unshare_flags(unshare_flags);
1989         if (err)
1990                 goto bad_unshare_out;
1991         /*
1992          * CLONE_NEWIPC must also detach from the undolist: after switching
1993          * to a new ipc namespace, the semaphore arrays from the old
1994          * namespace are unreachable.
1995          */
1996         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1997                 do_sysvsem = 1;
1998         err = unshare_fs(unshare_flags, &new_fs);
1999         if (err)
2000                 goto bad_unshare_out;
2001         err = unshare_fd(unshare_flags, &new_fd);
2002         if (err)
2003                 goto bad_unshare_cleanup_fs;
2004         err = unshare_userns(unshare_flags, &new_cred);
2005         if (err)
2006                 goto bad_unshare_cleanup_fd;
2007         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2008                                          new_cred, new_fs);
2009         if (err)
2010                 goto bad_unshare_cleanup_cred;
2011
2012         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2013                 if (do_sysvsem) {
2014                         /*
2015                          * CLONE_SYSVSEM is equivalent to sys_exit().
2016                          */
2017                         exit_sem(current);
2018                 }
2019                 if (unshare_flags & CLONE_NEWIPC) {
2020                         /* Orphan segments in old ns (see sem above). */
2021                         exit_shm(current);
2022                         shm_init_task(current);
2023                 }
2024
2025                 if (new_nsproxy)
2026                         switch_task_namespaces(current, new_nsproxy);
2027
2028                 task_lock(current);
2029
2030                 if (new_fs) {
2031                         fs = current->fs;
2032                         spin_lock(&fs->lock);
2033                         current->fs = new_fs;
2034                         if (--fs->users)
2035                                 new_fs = NULL;
2036                         else
2037                                 new_fs = fs;
2038                         spin_unlock(&fs->lock);
2039                 }
2040
2041                 if (new_fd) {
2042                         fd = current->files;
2043                         current->files = new_fd;
2044                         new_fd = fd;
2045                 }
2046
2047                 task_unlock(current);
2048
2049                 if (new_cred) {
2050                         /* Install the new user namespace */
2051                         commit_creds(new_cred);
2052                         new_cred = NULL;
2053                 }
2054         }
2055
2056 bad_unshare_cleanup_cred:
2057         if (new_cred)
2058                 put_cred(new_cred);
2059 bad_unshare_cleanup_fd:
2060         if (new_fd)
2061                 put_files_struct(new_fd);
2062
2063 bad_unshare_cleanup_fs:
2064         if (new_fs)
2065                 free_fs_struct(new_fs);
2066
2067 bad_unshare_out:
2068         return err;
2069 }
2070
2071 /*
2072  *      Helper to unshare the files of the current task.
2073  *      We don't want to expose copy_files internals to
2074  *      the exec layer of the kernel.
2075  */
2076
2077 int unshare_files(struct files_struct **displaced)
2078 {
2079         struct task_struct *task = current;
2080         struct files_struct *copy = NULL;
2081         int error;
2082
2083         error = unshare_fd(CLONE_FILES, &copy);
2084         if (error || !copy) {
2085                 *displaced = NULL;
2086                 return error;
2087         }
2088         *displaced = task->files;
2089         task_lock(task);
2090         task->files = copy;
2091         task_unlock(task);
2092         return 0;
2093 }
2094
2095 int sysctl_max_threads(struct ctl_table *table, int write,
2096                        void __user *buffer, size_t *lenp, loff_t *ppos)
2097 {
2098         struct ctl_table t;
2099         int ret;
2100         int threads = max_threads;
2101         int min = MIN_THREADS;
2102         int max = MAX_THREADS;
2103
2104         t = *table;
2105         t.data = &threads;
2106         t.extra1 = &min;
2107         t.extra2 = &max;
2108
2109         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2110         if (ret || !write)
2111                 return ret;
2112
2113         set_max_threads(threads);
2114
2115         return 0;
2116 }