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, int node)
 335 {
 336         struct task_struct *tsk;
 337         struct thread_info *ti;
 338         int err;
 339
 340         if (node == NUMA_NO_NODE)
 341                 node = tsk_fork_get_node(orig);
 342         tsk = alloc_task_struct_node(node);
 343         if (!tsk)
 344                 return NULL;
 345
 346         ti = alloc_thread_info_node(tsk, node);
 347         if (!ti)
 348                 goto free_tsk;
 349
 350         err = arch_dup_task_struct(tsk, orig);
 351         if (err)
 352                 goto free_ti;
 353
 354         tsk->stack = ti;
 355 #ifdef CONFIG_SECCOMP
 356         /*
 357          * We must handle setting up seccomp filters once we're under
 358          * the sighand lock in case orig has changed between now and
 359          * then. Until then, filter must be NULL to avoid messing up
 360          * the usage counts on the error path calling free_task.
 361          */
 362         tsk->seccomp.filter = NULL;
 363 #endif
 364
 365         setup_thread_stack(tsk, orig);
 366         clear_user_return_notifier(tsk);
 367         clear_tsk_need_resched(tsk);
 368         set_task_stack_end_magic(tsk);
 369
 370 #ifdef CONFIG_CC_STACKPROTECTOR
 371         tsk->stack_canary = get_random_int();
 372 #endif
 373
 374         /*
 375          * One for us, one for whoever does the "release_task()" (usually
 376          * parent)
 377          */
 378         atomic_set(&tsk->usage, 2);
 379 #ifdef CONFIG_BLK_DEV_IO_TRACE
 380         tsk->btrace_seq = 0;
 381 #endif
 382         tsk->splice_pipe = NULL;
 383         tsk->task_frag.page = NULL;
 384         tsk->wake_q.next = NULL;
 385
 386         account_kernel_stack(ti, 1);
 387
 388         return tsk;
 389
 390 free_ti:
 391         free_thread_info(ti);
 392 free_tsk:
 393         free_task_struct(tsk);
 394         return NULL;
 395 }
 396
 397 #ifdef CONFIG_MMU
 398 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 399 {
 400         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 401         struct rb_node **rb_link, *rb_parent;
 402         int retval;
 403         unsigned long charge;
 404
 405         uprobe_start_dup_mmap();
 406         down_write(&oldmm->mmap_sem);
 407         flush_cache_dup_mm(oldmm);
 408         uprobe_dup_mmap(oldmm, mm);
 409         /*
 410          * Not linked in yet - no deadlock potential:
 411          */
 412         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 413
 414         /* No ordering required: file already has been exposed. */
 415         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 416
 417         mm->total_vm = oldmm->total_vm;
 418         mm->shared_vm = oldmm->shared_vm;
 419         mm->exec_vm = oldmm->exec_vm;
 420         mm->stack_vm = oldmm->stack_vm;
 421
 422         rb_link = &mm->mm_rb.rb_node;
 423         rb_parent = NULL;
 424         pprev = &mm->mmap;
 425         retval = ksm_fork(mm, oldmm);
 426         if (retval)
 427                 goto out;
 428         retval = khugepaged_fork(mm, oldmm);
 429         if (retval)
 430                 goto out;
 431
 432         prev = NULL;
 433         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 434                 struct file *file;
 435
 436                 if (mpnt->vm_flags & VM_DONTCOPY) {
 437                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 438                                                         -vma_pages(mpnt));
 439                         continue;
 440                 }
 441                 charge = 0;
 442                 if (mpnt->vm_flags & VM_ACCOUNT) {
 443                         unsigned long len = vma_pages(mpnt);
 444
 445                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 446                                 goto fail_nomem;
 447                         charge = len;
 448                 }
 449                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 450                 if (!tmp)
 451                         goto fail_nomem;
 452                 *tmp = *mpnt;
 453                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 454                 retval = vma_dup_policy(mpnt, tmp);
 455                 if (retval)
 456                         goto fail_nomem_policy;
 457                 tmp->vm_mm = mm;
 458                 if (anon_vma_fork(tmp, mpnt))
 459                         goto fail_nomem_anon_vma_fork;
 460                 tmp->vm_flags &=
 461                         ~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
 462                 tmp->vm_next = tmp->vm_prev = NULL;
 463                 tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
 464                 file = tmp->vm_file;
 465                 if (file) {
 466                         struct inode *inode = file_inode(file);
 467                         struct address_space *mapping = file->f_mapping;
 468
 469                         get_file(file);
 470                         if (tmp->vm_flags & VM_DENYWRITE)
 471                                 atomic_dec(&inode->i_writecount);
 472                         i_mmap_lock_write(mapping);
 473                         if (tmp->vm_flags & VM_SHARED)
 474                                 atomic_inc(&mapping->i_mmap_writable);
 475                         flush_dcache_mmap_lock(mapping);
 476                         /* insert tmp into the share list, just after mpnt */
 477                         vma_interval_tree_insert_after(tmp, mpnt,
 478                                         &mapping->i_mmap);
 479                         flush_dcache_mmap_unlock(mapping);
 480                         i_mmap_unlock_write(mapping);
 481                 }
 482
 483                 /*
 484                  * Clear hugetlb-related page reserves for children. This only
 485                  * affects MAP_PRIVATE mappings. Faults generated by the child
 486                  * are not guaranteed to succeed, even if read-only
 487                  */
 488                 if (is_vm_hugetlb_page(tmp))
 489                         reset_vma_resv_huge_pages(tmp);
 490
 491                 /*
 492                  * Link in the new vma and copy the page table entries.
 493                  */
 494                 *pprev = tmp;
 495                 pprev = &tmp->vm_next;
 496                 tmp->vm_prev = prev;
 497                 prev = tmp;
 498
 499                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 500                 rb_link = &tmp->vm_rb.rb_right;
 501                 rb_parent = &tmp->vm_rb;
 502
 503                 mm->map_count++;
 504                 retval = copy_page_range(mm, oldmm, mpnt);
 505
 506                 if (tmp->vm_ops && tmp->vm_ops->open)
 507                         tmp->vm_ops->open(tmp);
 508
 509                 if (retval)
 510                         goto out;
 511         }
 512         /* a new mm has just been created */
 513         arch_dup_mmap(oldmm, mm);
 514         retval = 0;
 515 out:
 516         up_write(&mm->mmap_sem);
 517         flush_tlb_mm(oldmm);
 518         up_write(&oldmm->mmap_sem);
 519         uprobe_end_dup_mmap();
 520         return retval;
 521 fail_nomem_anon_vma_fork:
 522         mpol_put(vma_policy(tmp));
 523 fail_nomem_policy:
 524         kmem_cache_free(vm_area_cachep, tmp);
 525 fail_nomem:
 526         retval = -ENOMEM;
 527         vm_unacct_memory(charge);
 528         goto out;
 529 }
 530
 531 static inline int mm_alloc_pgd(struct mm_struct *mm)
 532 {
 533         mm->pgd = pgd_alloc(mm);
 534         if (unlikely(!mm->pgd))
 535                 return -ENOMEM;
 536         return 0;
 537 }
 538
 539 static inline void mm_free_pgd(struct mm_struct *mm)
 540 {
 541         pgd_free(mm, mm->pgd);
 542 }
 543 #else
 544 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 545 {
 546         down_write(&oldmm->mmap_sem);
 547         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 548         up_write(&oldmm->mmap_sem);
 549         return 0;
 550 }
 551 #define mm_alloc_pgd(mm)        (0)
 552 #define mm_free_pgd(mm)
 553 #endif /* CONFIG_MMU */
 554
 555 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 556
 557 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 558 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 559
 560 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 561
 562 static int __init coredump_filter_setup(char *s)
 563 {
 564         default_dump_filter =
 565                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 566                 MMF_DUMP_FILTER_MASK;
 567         return 1;
 568 }
 569
 570 __setup("coredump_filter=", coredump_filter_setup);
 571
 572 #include <linux/init_task.h>
 573
 574 static void mm_init_aio(struct mm_struct *mm)
 575 {
 576 #ifdef CONFIG_AIO
 577         spin_lock_init(&mm->ioctx_lock);
 578         mm->ioctx_table = NULL;
 579 #endif
 580 }
 581
 582 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 583 {
 584 #ifdef CONFIG_MEMCG
 585         mm->owner = p;
 586 #endif
 587 }
 588
 589 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 590         struct user_namespace *user_ns)
 591 {
 592         mm->mmap = NULL;
 593         mm->mm_rb = RB_ROOT;
 594         mm->vmacache_seqnum = 0;
 595         atomic_set(&mm->mm_users, 1);
 596         atomic_set(&mm->mm_count, 1);
 597         init_rwsem(&mm->mmap_sem);
 598         INIT_LIST_HEAD(&mm->mmlist);
 599         mm->core_state = NULL;
 600         atomic_long_set(&mm->nr_ptes, 0);
 601         mm_nr_pmds_init(mm);
 602         mm->map_count = 0;
 603         mm->locked_vm = 0;
 604         mm->pinned_vm = 0;
 605         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 606         spin_lock_init(&mm->page_table_lock);
 607         mm_init_cpumask(mm);
 608         mm_init_aio(mm);
 609         mm_init_owner(mm, p);
 610         mmu_notifier_mm_init(mm);
 611         clear_tlb_flush_pending(mm);
 612 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 613         mm->pmd_huge_pte = NULL;
 614 #endif
 615
 616         if (current->mm) {
 617                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 618                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 619         } else {
 620                 mm->flags = default_dump_filter;
 621                 mm->def_flags = 0;
 622         }
 623
 624         if (mm_alloc_pgd(mm))
 625                 goto fail_nopgd;
 626
 627         if (init_new_context(p, mm))
 628                 goto fail_nocontext;
 629
 630         mm->user_ns = get_user_ns(user_ns);
 631         return mm;
 632
 633 fail_nocontext:
 634         mm_free_pgd(mm);
 635 fail_nopgd:
 636         free_mm(mm);
 637         return NULL;
 638 }
 639
 640 static void check_mm(struct mm_struct *mm)
 641 {
 642         int i;
 643
 644         for (i = 0; i < NR_MM_COUNTERS; i++) {
 645                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 646
 647                 if (unlikely(x))
 648                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 649                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 650         }
 651
 652         if (atomic_long_read(&mm->nr_ptes))
 653                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 654                                 atomic_long_read(&mm->nr_ptes));
 655         if (mm_nr_pmds(mm))
 656                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 657                                 mm_nr_pmds(mm));
 658
 659 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 660         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 661 #endif
 662 }
 663
 664 /*
 665  * Allocate and initialize an mm_struct.
 666  */
 667 struct mm_struct *mm_alloc(void)
 668 {
 669         struct mm_struct *mm;
 670
 671         mm = allocate_mm();
 672         if (!mm)
 673                 return NULL;
 674
 675         memset(mm, 0, sizeof(*mm));
 676         return mm_init(mm, current, current_user_ns());
 677 }
 678
 679 /*
 680  * Called when the last reference to the mm
 681  * is dropped: either by a lazy thread or by
 682  * mmput. Free the page directory and the mm.
 683  */
 684 void __mmdrop(struct mm_struct *mm)
 685 {
 686         BUG_ON(mm == &init_mm);
 687         mm_free_pgd(mm);
 688         destroy_context(mm);
 689         mmu_notifier_mm_destroy(mm);
 690         check_mm(mm);
 691         put_user_ns(mm->user_ns);
 692         free_mm(mm);
 693 }
 694 EXPORT_SYMBOL_GPL(__mmdrop);
 695
 696 /*
 697  * Decrement the use count and release all resources for an mm.
 698  */
 699 void mmput(struct mm_struct *mm)
 700 {
 701         might_sleep();
 702
 703         if (atomic_dec_and_test(&mm->mm_users)) {
 704                 uprobe_clear_state(mm);
 705                 exit_aio(mm);
 706                 ksm_exit(mm);
 707                 khugepaged_exit(mm); /* must run before exit_mmap */
 708                 exit_mmap(mm);
 709                 set_mm_exe_file(mm, NULL);
 710                 if (!list_empty(&mm->mmlist)) {
 711                         spin_lock(&mmlist_lock);
 712                         list_del(&mm->mmlist);
 713                         spin_unlock(&mmlist_lock);
 714                 }
 715                 if (mm->binfmt)
 716                         module_put(mm->binfmt->module);
 717                 mmdrop(mm);
 718         }
 719 }
 720 EXPORT_SYMBOL_GPL(mmput);
 721
 722 /**
 723  * set_mm_exe_file - change a reference to the mm's executable file
 724  *
 725  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 726  *
 727  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 728  * invocations: in mmput() nobody alive left, in execve task is single
 729  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 730  * mm->exe_file, but does so without using set_mm_exe_file() in order
 731  * to do avoid the need for any locks.
 732  */
 733 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 734 {
 735         struct file *old_exe_file;
 736
 737         /*
 738          * It is safe to dereference the exe_file without RCU as
 739          * this function is only called if nobody else can access
 740          * this mm -- see comment above for justification.
 741          */
 742         old_exe_file = rcu_dereference_raw(mm->exe_file);
 743
 744         if (new_exe_file)
 745                 get_file(new_exe_file);
 746         rcu_assign_pointer(mm->exe_file, new_exe_file);
 747         if (old_exe_file)
 748                 fput(old_exe_file);
 749 }
 750
 751 /**
 752  * get_mm_exe_file - acquire a reference to the mm's executable file
 753  *
 754  * Returns %NULL if mm has no associated executable file.
 755  * User must release file via fput().
 756  */
 757 struct file *get_mm_exe_file(struct mm_struct *mm)
 758 {
 759         struct file *exe_file;
 760
 761         rcu_read_lock();
 762         exe_file = rcu_dereference(mm->exe_file);
 763         if (exe_file && !get_file_rcu(exe_file))
 764                 exe_file = NULL;
 765         rcu_read_unlock();
 766         return exe_file;
 767 }
 768 EXPORT_SYMBOL(get_mm_exe_file);
 769
 770 /**
 771  * get_task_exe_file - acquire a reference to the task's executable file
 772  *
 773  * Returns %NULL if task's mm (if any) has no associated executable file or
 774  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
 775  * User must release file via fput().
 776  */
 777 struct file *get_task_exe_file(struct task_struct *task)
 778 {
 779         struct file *exe_file = NULL;
 780         struct mm_struct *mm;
 781
 782         task_lock(task);
 783         mm = task->mm;
 784         if (mm) {
 785                 if (!(task->flags & PF_KTHREAD))
 786                         exe_file = get_mm_exe_file(mm);
 787         }
 788         task_unlock(task);
 789         return exe_file;
 790 }
 791 EXPORT_SYMBOL(get_task_exe_file);
 792
 793 /**
 794  * get_task_mm - acquire a reference to the task's mm
 795  *
 796  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 797  * this kernel workthread has transiently adopted a user mm with use_mm,
 798  * to do its AIO) is not set and if so returns a reference to it, after
 799  * bumping up the use count.  User must release the mm via mmput()
 800  * after use.  Typically used by /proc and ptrace.
 801  */
 802 struct mm_struct *get_task_mm(struct task_struct *task)
 803 {
 804         struct mm_struct *mm;
 805
 806         task_lock(task);
 807         mm = task->mm;
 808         if (mm) {
 809                 if (task->flags & PF_KTHREAD)
 810                         mm = NULL;
 811                 else
 812                         atomic_inc(&mm->mm_users);
 813         }
 814         task_unlock(task);
 815         return mm;
 816 }
 817 EXPORT_SYMBOL_GPL(get_task_mm);
 818
 819 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 820 {
 821         struct mm_struct *mm;
 822         int err;
 823
 824         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 825         if (err)
 826                 return ERR_PTR(err);
 827
 828         mm = get_task_mm(task);
 829         if (mm && mm != current->mm &&
 830                         !ptrace_may_access(task, mode)) {
 831                 mmput(mm);
 832                 mm = ERR_PTR(-EACCES);
 833         }
 834         mutex_unlock(&task->signal->cred_guard_mutex);
 835
 836         return mm;
 837 }
 838
 839 static void complete_vfork_done(struct task_struct *tsk)
 840 {
 841         struct completion *vfork;
 842
 843         task_lock(tsk);
 844         vfork = tsk->vfork_done;
 845         if (likely(vfork)) {
 846                 tsk->vfork_done = NULL;
 847                 complete(vfork);
 848         }
 849         task_unlock(tsk);
 850 }
 851
 852 static int wait_for_vfork_done(struct task_struct *child,
 853                                 struct completion *vfork)
 854 {
 855         int killed;
 856
 857         freezer_do_not_count();
 858         killed = wait_for_completion_killable(vfork);
 859         freezer_count();
 860
 861         if (killed) {
 862                 task_lock(child);
 863                 child->vfork_done = NULL;
 864                 task_unlock(child);
 865         }
 866
 867         put_task_struct(child);
 868         return killed;
 869 }
 870
 871 /* Please note the differences between mmput and mm_release.
 872  * mmput is called whenever we stop holding onto a mm_struct,
 873  * error success whatever.
 874  *
 875  * mm_release is called after a mm_struct has been removed
 876  * from the current process.
 877  *
 878  * This difference is important for error handling, when we
 879  * only half set up a mm_struct for a new process and need to restore
 880  * the old one.  Because we mmput the new mm_struct before
 881  * restoring the old one. . .
 882  * Eric Biederman 10 January 1998
 883  */
 884 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 885 {
 886         /* Get rid of any futexes when releasing the mm */
 887 #ifdef CONFIG_FUTEX
 888         if (unlikely(tsk->robust_list)) {
 889                 exit_robust_list(tsk);
 890                 tsk->robust_list = NULL;
 891         }
 892 #ifdef CONFIG_COMPAT
 893         if (unlikely(tsk->compat_robust_list)) {
 894                 compat_exit_robust_list(tsk);
 895                 tsk->compat_robust_list = NULL;
 896         }
 897 #endif
 898         if (unlikely(!list_empty(&tsk->pi_state_list)))
 899                 exit_pi_state_list(tsk);
 900 #endif
 901
 902         uprobe_free_utask(tsk);
 903
 904         /* Get rid of any cached register state */
 905         deactivate_mm(tsk, mm);
 906
 907         /*
 908          * Signal userspace if we're not exiting with a core dump
 909          * because we want to leave the value intact for debugging
 910          * purposes.
 911          */
 912         if (tsk->clear_child_tid) {
 913                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
 914                     atomic_read(&mm->mm_users) > 1) {
 915                         /*
 916                          * We don't check the error code - if userspace has
 917                          * not set up a proper pointer then tough luck.
 918                          */
 919                         put_user(0, tsk->clear_child_tid);
 920                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 921                                         1, NULL, NULL, 0);
 922                 }
 923                 tsk->clear_child_tid = NULL;
 924         }
 925
 926         /*
 927          * All done, finally we can wake up parent and return this mm to him.
 928          * Also kthread_stop() uses this completion for synchronization.
 929          */
 930         if (tsk->vfork_done)
 931                 complete_vfork_done(tsk);
 932 }
 933
 934 /*
 935  * Allocate a new mm structure and copy contents from the
 936  * mm structure of the passed in task structure.
 937  */
 938 static struct mm_struct *dup_mm(struct task_struct *tsk)
 939 {
 940         struct mm_struct *mm, *oldmm = current->mm;
 941         int err;
 942
 943         mm = allocate_mm();
 944         if (!mm)
 945                 goto fail_nomem;
 946
 947         memcpy(mm, oldmm, sizeof(*mm));
 948
 949         if (!mm_init(mm, tsk, mm->user_ns))
 950                 goto fail_nomem;
 951
 952         err = dup_mmap(mm, oldmm);
 953         if (err)
 954                 goto free_pt;
 955
 956         mm->hiwater_rss = get_mm_rss(mm);
 957         mm->hiwater_vm = mm->total_vm;
 958
 959         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 960                 goto free_pt;
 961
 962         return mm;
 963
 964 free_pt:
 965         /* don't put binfmt in mmput, we haven't got module yet */
 966         mm->binfmt = NULL;
 967         mmput(mm);
 968
 969 fail_nomem:
 970         return NULL;
 971 }
 972
 973 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 974 {
 975         struct mm_struct *mm, *oldmm;
 976         int retval;
 977
 978         tsk->min_flt = tsk->maj_flt = 0;
 979         tsk->nvcsw = tsk->nivcsw = 0;
 980 #ifdef CONFIG_DETECT_HUNG_TASK
 981         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 982 #endif
 983
 984         tsk->mm = NULL;
 985         tsk->active_mm = NULL;
 986
 987         /*
 988          * Are we cloning a kernel thread?
 989          *
 990          * We need to steal a active VM for that..
 991          */
 992         oldmm = current->mm;
 993         if (!oldmm)
 994                 return 0;
 995
 996         /* initialize the new vmacache entries */
 997         vmacache_flush(tsk);
 998
 999         if (clone_flags & CLONE_VM) {
1000                 atomic_inc(&oldmm->mm_users);
1001                 mm = oldmm;
1002                 goto good_mm;
1003         }
1004
1005         retval = -ENOMEM;
1006         mm = dup_mm(tsk);
1007         if (!mm)
1008                 goto fail_nomem;
1009
1010 good_mm:
1011         tsk->mm = mm;
1012         tsk->active_mm = mm;
1013         return 0;
1014
1015 fail_nomem:
1016         return retval;
1017 }
1018
1019 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1020 {
1021         struct fs_struct *fs = current->fs;
1022         if (clone_flags & CLONE_FS) {
1023                 /* tsk->fs is already what we want */
1024                 spin_lock(&fs->lock);
1025                 if (fs->in_exec) {
1026                         spin_unlock(&fs->lock);
1027                         return -EAGAIN;
1028                 }
1029                 fs->users++;
1030                 spin_unlock(&fs->lock);
1031                 return 0;
1032         }
1033         tsk->fs = copy_fs_struct(fs);
1034         if (!tsk->fs)
1035                 return -ENOMEM;
1036         return 0;
1037 }
1038
1039 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1040 {
1041         struct files_struct *oldf, *newf;
1042         int error = 0;
1043
1044         /*
1045          * A background process may not have any files ...
1046          */
1047         oldf = current->files;
1048         if (!oldf)
1049                 goto out;
1050
1051         if (clone_flags & CLONE_FILES) {
1052                 atomic_inc(&oldf->count);
1053                 goto out;
1054         }
1055
1056         newf = dup_fd(oldf, &error);
1057         if (!newf)
1058                 goto out;
1059
1060         tsk->files = newf;
1061         error = 0;
1062 out:
1063         return error;
1064 }
1065
1066 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1067 {
1068 #ifdef CONFIG_BLOCK
1069         struct io_context *ioc = current->io_context;
1070         struct io_context *new_ioc;
1071
1072         if (!ioc)
1073                 return 0;
1074         /*
1075          * Share io context with parent, if CLONE_IO is set
1076          */
1077         if (clone_flags & CLONE_IO) {
1078                 ioc_task_link(ioc);
1079                 tsk->io_context = ioc;
1080         } else if (ioprio_valid(ioc->ioprio)) {
1081                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1082                 if (unlikely(!new_ioc))
1083                         return -ENOMEM;
1084
1085                 new_ioc->ioprio = ioc->ioprio;
1086                 put_io_context(new_ioc);
1087         }
1088 #endif
1089         return 0;
1090 }
1091
1092 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1093 {
1094         struct sighand_struct *sig;
1095
1096         if (clone_flags & CLONE_SIGHAND) {
1097                 atomic_inc(&current->sighand->count);
1098                 return 0;
1099         }
1100         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1101         rcu_assign_pointer(tsk->sighand, sig);
1102         if (!sig)
1103                 return -ENOMEM;
1104
1105         atomic_set(&sig->count, 1);
1106         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1107         return 0;
1108 }
1109
1110 void __cleanup_sighand(struct sighand_struct *sighand)
1111 {
1112         if (atomic_dec_and_test(&sighand->count)) {
1113                 signalfd_cleanup(sighand);
1114                 /*
1115                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1116                  * without an RCU grace period, see __lock_task_sighand().
1117                  */
1118                 kmem_cache_free(sighand_cachep, sighand);
1119         }
1120 }
1121
1122 /*
1123  * Initialize POSIX timer handling for a thread group.
1124  */
1125 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1126 {
1127         unsigned long cpu_limit;
1128
1129         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1130         if (cpu_limit != RLIM_INFINITY) {
1131                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1132                 sig->cputimer.running = true;
1133         }
1134
1135         /* The timer lists. */
1136         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1137         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1138         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1139 }
1140
1141 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1142 {
1143         struct signal_struct *sig;
1144
1145         if (clone_flags & CLONE_THREAD)
1146                 return 0;
1147
1148         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1149         tsk->signal = sig;
1150         if (!sig)
1151                 return -ENOMEM;
1152
1153         sig->nr_threads = 1;
1154         atomic_set(&sig->live, 1);
1155         atomic_set(&sig->sigcnt, 1);
1156
1157         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1158         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1159         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1160
1161         init_waitqueue_head(&sig->wait_chldexit);
1162         sig->curr_target = tsk;
1163         init_sigpending(&sig->shared_pending);
1164         INIT_LIST_HEAD(&sig->posix_timers);
1165         seqlock_init(&sig->stats_lock);
1166         prev_cputime_init(&sig->prev_cputime);
1167
1168         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1169         sig->real_timer.function = it_real_fn;
1170
1171         task_lock(current->group_leader);
1172         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1173         task_unlock(current->group_leader);
1174
1175         posix_cpu_timers_init_group(sig);
1176
1177         tty_audit_fork(sig);
1178         sched_autogroup_fork(sig);
1179
1180         sig->oom_score_adj = current->signal->oom_score_adj;
1181         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1182
1183         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1184                                    current->signal->is_child_subreaper;
1185
1186         mutex_init(&sig->cred_guard_mutex);
1187
1188         return 0;
1189 }
1190
1191 static void copy_seccomp(struct task_struct *p)
1192 {
1193 #ifdef CONFIG_SECCOMP
1194         /*
1195          * Must be called with sighand->lock held, which is common to
1196          * all threads in the group. Holding cred_guard_mutex is not
1197          * needed because this new task is not yet running and cannot
1198          * be racing exec.
1199          */
1200         assert_spin_locked(&current->sighand->siglock);
1201
1202         /* Ref-count the new filter user, and assign it. */
1203         get_seccomp_filter(current);
1204         p->seccomp = current->seccomp;
1205
1206         /*
1207          * Explicitly enable no_new_privs here in case it got set
1208          * between the task_struct being duplicated and holding the
1209          * sighand lock. The seccomp state and nnp must be in sync.
1210          */
1211         if (task_no_new_privs(current))
1212                 task_set_no_new_privs(p);
1213
1214         /*
1215          * If the parent gained a seccomp mode after copying thread
1216          * flags and between before we held the sighand lock, we have
1217          * to manually enable the seccomp thread flag here.
1218          */
1219         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1220                 set_tsk_thread_flag(p, TIF_SECCOMP);
1221 #endif
1222 }
1223
1224 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1225 {
1226         current->clear_child_tid = tidptr;
1227
1228         return task_pid_vnr(current);
1229 }
1230
1231 static void rt_mutex_init_task(struct task_struct *p)
1232 {
1233         raw_spin_lock_init(&p->pi_lock);
1234 #ifdef CONFIG_RT_MUTEXES
1235         p->pi_waiters = RB_ROOT;
1236         p->pi_waiters_leftmost = NULL;
1237         p->pi_blocked_on = NULL;
1238 #endif
1239 }
1240
1241 /*
1242  * Initialize POSIX timer handling for a single task.
1243  */
1244 static void posix_cpu_timers_init(struct task_struct *tsk)
1245 {
1246         tsk->cputime_expires.prof_exp = 0;
1247         tsk->cputime_expires.virt_exp = 0;
1248         tsk->cputime_expires.sched_exp = 0;
1249         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1250         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1251         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1252 }
1253
1254 static inline void
1255 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1256 {
1257          task->pids[type].pid = pid;
1258 }
1259
1260 /*
1261  * This creates a new process as a copy of the old one,
1262  * but does not actually start it yet.
1263  *
1264  * It copies the registers, and all the appropriate
1265  * parts of the process environment (as per the clone
1266  * flags). The actual kick-off is left to the caller.
1267  */
1268 static struct task_struct *copy_process(unsigned long clone_flags,
1269                                         unsigned long stack_start,
1270                                         unsigned long stack_size,
1271                                         int __user *child_tidptr,
1272                                         struct pid *pid,
1273                                         int trace,
1274                                         unsigned long tls,
1275                                         int node)
1276 {
1277         int retval;
1278         struct task_struct *p;
1279         void *cgrp_ss_priv[CGROUP_CANFORK_COUNT] = {};
1280
1281         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1282                 return ERR_PTR(-EINVAL);
1283
1284         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1285                 return ERR_PTR(-EINVAL);
1286
1287         /*
1288          * Thread groups must share signals as well, and detached threads
1289          * can only be started up within the thread group.
1290          */
1291         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1292                 return ERR_PTR(-EINVAL);
1293
1294         /*
1295          * Shared signal handlers imply shared VM. By way of the above,
1296          * thread groups also imply shared VM. Blocking this case allows
1297          * for various simplifications in other code.
1298          */
1299         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1300                 return ERR_PTR(-EINVAL);
1301
1302         /*
1303          * Siblings of global init remain as zombies on exit since they are
1304          * not reaped by their parent (swapper). To solve this and to avoid
1305          * multi-rooted process trees, prevent global and container-inits
1306          * from creating siblings.
1307          */
1308         if ((clone_flags & CLONE_PARENT) &&
1309                                 current->signal->flags & SIGNAL_UNKILLABLE)
1310                 return ERR_PTR(-EINVAL);
1311
1312         /*
1313          * If the new process will be in a different pid or user namespace
1314          * do not allow it to share a thread group with the forking task.
1315          */
1316         if (clone_flags & CLONE_THREAD) {
1317                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1318                     (task_active_pid_ns(current) !=
1319                                 current->nsproxy->pid_ns_for_children))
1320                         return ERR_PTR(-EINVAL);
1321         }
1322
1323         retval = security_task_create(clone_flags);
1324         if (retval)
1325                 goto fork_out;
1326
1327         retval = -ENOMEM;
1328         p = dup_task_struct(current, node);
1329         if (!p)
1330                 goto fork_out;
1331
1332         ftrace_graph_init_task(p);
1333
1334         rt_mutex_init_task(p);
1335
1336 #ifdef CONFIG_PROVE_LOCKING
1337         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1338         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1339 #endif
1340         retval = -EAGAIN;
1341         if (atomic_read(&p->real_cred->user->processes) >=
1342                         task_rlimit(p, RLIMIT_NPROC)) {
1343                 if (p->real_cred->user != INIT_USER &&
1344                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1345                         goto bad_fork_free;
1346         }
1347         current->flags &= ~PF_NPROC_EXCEEDED;
1348
1349         retval = copy_creds(p, clone_flags);
1350         if (retval < 0)
1351                 goto bad_fork_free;
1352
1353         /*
1354          * If multiple threads are within copy_process(), then this check
1355          * triggers too late. This doesn't hurt, the check is only there
1356          * to stop root fork bombs.
1357          */
1358         retval = -EAGAIN;
1359         if (nr_threads >= max_threads)
1360                 goto bad_fork_cleanup_count;
1361
1362         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1363         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1364         p->flags |= PF_FORKNOEXEC;
1365         INIT_LIST_HEAD(&p->children);
1366         INIT_LIST_HEAD(&p->sibling);
1367         rcu_copy_process(p);
1368         p->vfork_done = NULL;
1369         spin_lock_init(&p->alloc_lock);
1370
1371         init_sigpending(&p->pending);
1372
1373         p->utime = p->stime = p->gtime = 0;
1374         p->utimescaled = p->stimescaled = 0;
1375         prev_cputime_init(&p->prev_cputime);
1376
1377 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1378         seqlock_init(&p->vtime_seqlock);
1379         p->vtime_snap = 0;
1380         p->vtime_snap_whence = VTIME_SLEEPING;
1381 #endif
1382
1383 #if defined(SPLIT_RSS_COUNTING)
1384         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1385 #endif
1386
1387         p->default_timer_slack_ns = current->timer_slack_ns;
1388
1389         task_io_accounting_init(&p->ioac);
1390         acct_clear_integrals(p);
1391
1392         posix_cpu_timers_init(p);
1393
1394         p->start_time = ktime_get_ns();
1395         p->real_start_time = ktime_get_boot_ns();
1396         p->io_context = NULL;
1397         p->audit_context = NULL;
1398         cgroup_fork(p);
1399 #ifdef CONFIG_NUMA
1400         p->mempolicy = mpol_dup(p->mempolicy);
1401         if (IS_ERR(p->mempolicy)) {
1402                 retval = PTR_ERR(p->mempolicy);
1403                 p->mempolicy = NULL;
1404                 goto bad_fork_cleanup_threadgroup_lock;
1405         }
1406 #endif
1407 #ifdef CONFIG_CPUSETS
1408         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1409         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1410         seqcount_init(&p->mems_allowed_seq);
1411 #endif
1412 #ifdef CONFIG_TRACE_IRQFLAGS
1413         p->irq_events = 0;
1414         p->hardirqs_enabled = 0;
1415         p->hardirq_enable_ip = 0;
1416         p->hardirq_enable_event = 0;
1417         p->hardirq_disable_ip = _THIS_IP_;
1418         p->hardirq_disable_event = 0;
1419         p->softirqs_enabled = 1;
1420         p->softirq_enable_ip = _THIS_IP_;
1421         p->softirq_enable_event = 0;
1422         p->softirq_disable_ip = 0;
1423         p->softirq_disable_event = 0;
1424         p->hardirq_context = 0;
1425         p->softirq_context = 0;
1426 #endif
1427
1428         p->pagefault_disabled = 0;
1429
1430 #ifdef CONFIG_LOCKDEP
1431         p->lockdep_depth = 0; /* no locks held yet */
1432         p->curr_chain_key = 0;
1433         p->lockdep_recursion = 0;
1434 #endif
1435
1436 #ifdef CONFIG_DEBUG_MUTEXES
1437         p->blocked_on = NULL; /* not blocked yet */
1438 #endif
1439 #ifdef CONFIG_BCACHE
1440         p->sequential_io        = 0;
1441         p->sequential_io_avg    = 0;
1442 #endif
1443
1444         /* Perform scheduler related setup. Assign this task to a CPU. */
1445         retval = sched_fork(clone_flags, p);
1446         if (retval)
1447                 goto bad_fork_cleanup_policy;
1448
1449         retval = perf_event_init_task(p);
1450         if (retval)
1451                 goto bad_fork_cleanup_policy;
1452         retval = audit_alloc(p);
1453         if (retval)
1454                 goto bad_fork_cleanup_perf;
1455         /* copy all the process information */
1456         shm_init_task(p);
1457         retval = copy_semundo(clone_flags, p);
1458         if (retval)
1459                 goto bad_fork_cleanup_audit;
1460         retval = copy_files(clone_flags, p);
1461         if (retval)
1462                 goto bad_fork_cleanup_semundo;
1463         retval = copy_fs(clone_flags, p);
1464         if (retval)
1465                 goto bad_fork_cleanup_files;
1466         retval = copy_sighand(clone_flags, p);
1467         if (retval)
1468                 goto bad_fork_cleanup_fs;
1469         retval = copy_signal(clone_flags, p);
1470         if (retval)
1471                 goto bad_fork_cleanup_sighand;
1472         retval = copy_mm(clone_flags, p);
1473         if (retval)
1474                 goto bad_fork_cleanup_signal;
1475         retval = copy_namespaces(clone_flags, p);
1476         if (retval)
1477                 goto bad_fork_cleanup_mm;
1478         retval = copy_io(clone_flags, p);
1479         if (retval)
1480                 goto bad_fork_cleanup_namespaces;
1481         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1482         if (retval)
1483                 goto bad_fork_cleanup_io;
1484
1485         if (pid != &init_struct_pid) {
1486                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1487                 if (IS_ERR(pid)) {
1488                         retval = PTR_ERR(pid);
1489                         goto bad_fork_cleanup_io;
1490                 }
1491         }
1492
1493         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1494         /*
1495          * Clear TID on mm_release()?
1496          */
1497         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1498 #ifdef CONFIG_BLOCK
1499         p->plug = NULL;
1500 #endif
1501 #ifdef CONFIG_FUTEX
1502         p->robust_list = NULL;
1503 #ifdef CONFIG_COMPAT
1504         p->compat_robust_list = NULL;
1505 #endif
1506         INIT_LIST_HEAD(&p->pi_state_list);
1507         p->pi_state_cache = NULL;
1508 #endif
1509         /*
1510          * sigaltstack should be cleared when sharing the same VM
1511          */
1512         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1513                 p->sas_ss_sp = p->sas_ss_size = 0;
1514
1515         /*
1516          * Syscall tracing and stepping should be turned off in the
1517          * child regardless of CLONE_PTRACE.
1518          */
1519         user_disable_single_step(p);
1520         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1521 #ifdef TIF_SYSCALL_EMU
1522         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1523 #endif
1524         clear_all_latency_tracing(p);
1525
1526         /* ok, now we should be set up.. */
1527         p->pid = pid_nr(pid);
1528         if (clone_flags & CLONE_THREAD) {
1529                 p->exit_signal = -1;
1530                 p->group_leader = current->group_leader;
1531                 p->tgid = current->tgid;
1532         } else {
1533                 if (clone_flags & CLONE_PARENT)
1534                         p->exit_signal = current->group_leader->exit_signal;
1535                 else
1536                         p->exit_signal = (clone_flags & CSIGNAL);
1537                 p->group_leader = p;
1538                 p->tgid = p->pid;
1539         }
1540
1541         p->nr_dirtied = 0;
1542         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1543         p->dirty_paused_when = 0;
1544
1545         p->pdeath_signal = 0;
1546         INIT_LIST_HEAD(&p->thread_group);
1547         p->task_works = NULL;
1548
1549         threadgroup_change_begin(current);
1550         /*
1551          * Ensure that the cgroup subsystem policies allow the new process to be
1552          * forked. It should be noted the the new process's css_set can be changed
1553          * between here and cgroup_post_fork() if an organisation operation is in
1554          * progress.
1555          */
1556         retval = cgroup_can_fork(p, cgrp_ss_priv);
1557         if (retval)
1558                 goto bad_fork_free_pid;
1559
1560         /*
1561          * Make it visible to the rest of the system, but dont wake it up yet.
1562          * Need tasklist lock for parent etc handling!
1563          */
1564         write_lock_irq(&tasklist_lock);
1565
1566         /* CLONE_PARENT re-uses the old parent */
1567         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1568                 p->real_parent = current->real_parent;
1569                 p->parent_exec_id = current->parent_exec_id;
1570         } else {
1571                 p->real_parent = current;
1572                 p->parent_exec_id = current->self_exec_id;
1573         }
1574
1575         spin_lock(&current->sighand->siglock);
1576
1577         /*
1578          * Copy seccomp details explicitly here, in case they were changed
1579          * before holding sighand lock.
1580          */
1581         copy_seccomp(p);
1582
1583         /*
1584          * Process group and session signals need to be delivered to just the
1585          * parent before the fork or both the parent and the child after the
1586          * fork. Restart if a signal comes in before we add the new process to
1587          * it's process group.
1588          * A fatal signal pending means that current will exit, so the new
1589          * thread can't slip out of an OOM kill (or normal SIGKILL).
1590         */
1591         recalc_sigpending();
1592         if (signal_pending(current)) {
1593                 spin_unlock(&current->sighand->siglock);
1594                 write_unlock_irq(&tasklist_lock);
1595                 retval = -ERESTARTNOINTR;
1596                 goto bad_fork_cancel_cgroup;
1597         }
1598
1599         if (likely(p->pid)) {
1600                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1601
1602                 init_task_pid(p, PIDTYPE_PID, pid);
1603                 if (thread_group_leader(p)) {
1604                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1605                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1606
1607                         if (is_child_reaper(pid)) {
1608                                 ns_of_pid(pid)->child_reaper = p;
1609                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1610                         }
1611
1612                         p->signal->leader_pid = pid;
1613                         p->signal->tty = tty_kref_get(current->signal->tty);
1614                         list_add_tail(&p->sibling, &p->real_parent->children);
1615                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1616                         attach_pid(p, PIDTYPE_PGID);
1617                         attach_pid(p, PIDTYPE_SID);
1618                         __this_cpu_inc(process_counts);
1619                 } else {
1620                         current->signal->nr_threads++;
1621                         atomic_inc(&current->signal->live);
1622                         atomic_inc(&current->signal->sigcnt);
1623                         list_add_tail_rcu(&p->thread_group,
1624                                           &p->group_leader->thread_group);
1625                         list_add_tail_rcu(&p->thread_node,
1626                                           &p->signal->thread_head);
1627                 }
1628                 attach_pid(p, PIDTYPE_PID);
1629                 nr_threads++;
1630         }
1631
1632         total_forks++;
1633         spin_unlock(&current->sighand->siglock);
1634         syscall_tracepoint_update(p);
1635         write_unlock_irq(&tasklist_lock);
1636
1637         proc_fork_connector(p);
1638         cgroup_post_fork(p, cgrp_ss_priv);
1639         threadgroup_change_end(current);
1640         perf_event_fork(p);
1641
1642         trace_task_newtask(p, clone_flags);
1643         uprobe_copy_process(p, clone_flags);
1644
1645         return p;
1646
1647 bad_fork_cancel_cgroup:
1648         cgroup_cancel_fork(p, cgrp_ss_priv);
1649 bad_fork_free_pid:
1650         threadgroup_change_end(current);
1651         if (pid != &init_struct_pid)
1652                 free_pid(pid);
1653 bad_fork_cleanup_io:
1654         if (p->io_context)
1655                 exit_io_context(p);
1656 bad_fork_cleanup_namespaces:
1657         exit_task_namespaces(p);
1658 bad_fork_cleanup_mm:
1659         if (p->mm)
1660                 mmput(p->mm);
1661 bad_fork_cleanup_signal:
1662         if (!(clone_flags & CLONE_THREAD))
1663                 free_signal_struct(p->signal);
1664 bad_fork_cleanup_sighand:
1665         __cleanup_sighand(p->sighand);
1666 bad_fork_cleanup_fs:
1667         exit_fs(p); /* blocking */
1668 bad_fork_cleanup_files:
1669         exit_files(p); /* blocking */
1670 bad_fork_cleanup_semundo:
1671         exit_sem(p);
1672 bad_fork_cleanup_audit:
1673         audit_free(p);
1674 bad_fork_cleanup_perf:
1675         perf_event_free_task(p);
1676 bad_fork_cleanup_policy:
1677 #ifdef CONFIG_NUMA
1678         mpol_put(p->mempolicy);
1679 bad_fork_cleanup_threadgroup_lock:
1680 #endif
1681         delayacct_tsk_free(p);
1682 bad_fork_cleanup_count:
1683         atomic_dec(&p->cred->user->processes);
1684         exit_creds(p);
1685 bad_fork_free:
1686         free_task(p);
1687 fork_out:
1688         return ERR_PTR(retval);
1689 }
1690
1691 static inline void init_idle_pids(struct pid_link *links)
1692 {
1693         enum pid_type type;
1694
1695         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1696                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1697                 links[type].pid = &init_struct_pid;
1698         }
1699 }
1700
1701 struct task_struct *fork_idle(int cpu)
1702 {
1703         struct task_struct *task;
1704         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1705                             cpu_to_node(cpu));
1706         if (!IS_ERR(task)) {
1707                 init_idle_pids(task->pids);
1708                 init_idle(task, cpu);
1709         }
1710
1711         return task;
1712 }
1713
1714 /*
1715  *  Ok, this is the main fork-routine.
1716  *
1717  * It copies the process, and if successful kick-starts
1718  * it and waits for it to finish using the VM if required.
1719  */
1720 long _do_fork(unsigned long clone_flags,
1721               unsigned long stack_start,
1722               unsigned long stack_size,
1723               int __user *parent_tidptr,
1724               int __user *child_tidptr,
1725               unsigned long tls)
1726 {
1727         struct task_struct *p;
1728         int trace = 0;
1729         long nr;
1730
1731         /*
1732          * Determine whether and which event to report to ptracer.  When
1733          * called from kernel_thread or CLONE_UNTRACED is explicitly
1734          * requested, no event is reported; otherwise, report if the event
1735          * for the type of forking is enabled.
1736          */
1737         if (!(clone_flags & CLONE_UNTRACED)) {
1738                 if (clone_flags & CLONE_VFORK)
1739                         trace = PTRACE_EVENT_VFORK;
1740                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1741                         trace = PTRACE_EVENT_CLONE;
1742                 else
1743                         trace = PTRACE_EVENT_FORK;
1744
1745                 if (likely(!ptrace_event_enabled(current, trace)))
1746                         trace = 0;
1747         }
1748
1749         p = copy_process(clone_flags, stack_start, stack_size,
1750                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
1751         /*
1752          * Do this prior waking up the new thread - the thread pointer
1753          * might get invalid after that point, if the thread exits quickly.
1754          */
1755         if (!IS_ERR(p)) {
1756                 struct completion vfork;
1757                 struct pid *pid;
1758
1759                 trace_sched_process_fork(current, p);
1760
1761                 pid = get_task_pid(p, PIDTYPE_PID);
1762                 nr = pid_vnr(pid);
1763
1764                 if (clone_flags & CLONE_PARENT_SETTID)
1765                         put_user(nr, parent_tidptr);
1766
1767                 if (clone_flags & CLONE_VFORK) {
1768                         p->vfork_done = &vfork;
1769                         init_completion(&vfork);
1770                         get_task_struct(p);
1771                 }
1772
1773                 wake_up_new_task(p);
1774
1775                 /* forking complete and child started to run, tell ptracer */
1776                 if (unlikely(trace))
1777                         ptrace_event_pid(trace, pid);
1778
1779                 if (clone_flags & CLONE_VFORK) {
1780                         if (!wait_for_vfork_done(p, &vfork))
1781                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1782                 }
1783
1784                 put_pid(pid);
1785         } else {
1786                 nr = PTR_ERR(p);
1787         }
1788         return nr;
1789 }
1790
1791 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1792 /* For compatibility with architectures that call do_fork directly rather than
1793  * using the syscall entry points below. */
1794 long do_fork(unsigned long clone_flags,
1795               unsigned long stack_start,
1796               unsigned long stack_size,
1797               int __user *parent_tidptr,
1798               int __user *child_tidptr)
1799 {
1800         return _do_fork(clone_flags, stack_start, stack_size,
1801                         parent_tidptr, child_tidptr, 0);
1802 }
1803 #endif
1804
1805 /*
1806  * Create a kernel thread.
1807  */
1808 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1809 {
1810         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1811                 (unsigned long)arg, NULL, NULL, 0);
1812 }
1813
1814 #ifdef __ARCH_WANT_SYS_FORK
1815 SYSCALL_DEFINE0(fork)
1816 {
1817 #ifdef CONFIG_MMU
1818         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1819 #else
1820         /* can not support in nommu mode */
1821         return -EINVAL;
1822 #endif
1823 }
1824 #endif
1825
1826 #ifdef __ARCH_WANT_SYS_VFORK
1827 SYSCALL_DEFINE0(vfork)
1828 {
1829         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1830                         0, NULL, NULL, 0);
1831 }
1832 #endif
1833
1834 #ifdef __ARCH_WANT_SYS_CLONE
1835 #ifdef CONFIG_CLONE_BACKWARDS
1836 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1837                  int __user *, parent_tidptr,
1838                  unsigned long, tls,
1839                  int __user *, child_tidptr)
1840 #elif defined(CONFIG_CLONE_BACKWARDS2)
1841 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1842                  int __user *, parent_tidptr,
1843                  int __user *, child_tidptr,
1844                  unsigned long, tls)
1845 #elif defined(CONFIG_CLONE_BACKWARDS3)
1846 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1847                 int, stack_size,
1848                 int __user *, parent_tidptr,
1849                 int __user *, child_tidptr,
1850                 unsigned long, tls)
1851 #else
1852 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1853                  int __user *, parent_tidptr,
1854                  int __user *, child_tidptr,
1855                  unsigned long, tls)
1856 #endif
1857 {
1858         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
1859 }
1860 #endif
1861
1862 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1863 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1864 #endif
1865
1866 static void sighand_ctor(void *data)
1867 {
1868         struct sighand_struct *sighand = data;
1869
1870         spin_lock_init(&sighand->siglock);
1871         init_waitqueue_head(&sighand->signalfd_wqh);
1872 }
1873
1874 void __init proc_caches_init(void)
1875 {
1876         sighand_cachep = kmem_cache_create("sighand_cache",
1877                         sizeof(struct sighand_struct), 0,
1878                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1879                         SLAB_NOTRACK, sighand_ctor);
1880         signal_cachep = kmem_cache_create("signal_cache",
1881                         sizeof(struct signal_struct), 0,
1882                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1883         files_cachep = kmem_cache_create("files_cache",
1884                         sizeof(struct files_struct), 0,
1885                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1886         fs_cachep = kmem_cache_create("fs_cache",
1887                         sizeof(struct fs_struct), 0,
1888                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1889         /*
1890          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1891          * whole struct cpumask for the OFFSTACK case. We could change
1892          * this to *only* allocate as much of it as required by the
1893          * maximum number of CPU's we can ever have.  The cpumask_allocation
1894          * is at the end of the structure, exactly for that reason.
1895          */
1896         mm_cachep = kmem_cache_create("mm_struct",
1897                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1898                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1899         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1900         mmap_init();
1901         nsproxy_cache_init();
1902 }
1903
1904 /*
1905  * Check constraints on flags passed to the unshare system call.
1906  */
1907 static int check_unshare_flags(unsigned long unshare_flags)
1908 {
1909         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1910                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1911                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1912                                 CLONE_NEWUSER|CLONE_NEWPID))
1913                 return -EINVAL;
1914         /*
1915          * Not implemented, but pretend it works if there is nothing
1916          * to unshare.  Note that unsharing the address space or the
1917          * signal handlers also need to unshare the signal queues (aka
1918          * CLONE_THREAD).
1919          */
1920         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1921                 if (!thread_group_empty(current))
1922                         return -EINVAL;
1923         }
1924         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
1925                 if (atomic_read(&current->sighand->count) > 1)
1926                         return -EINVAL;
1927         }
1928         if (unshare_flags & CLONE_VM) {
1929                 if (!current_is_single_threaded())
1930                         return -EINVAL;
1931         }
1932
1933         return 0;
1934 }
1935
1936 /*
1937  * Unshare the filesystem structure if it is being shared
1938  */
1939 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1940 {
1941         struct fs_struct *fs = current->fs;
1942
1943         if (!(unshare_flags & CLONE_FS) || !fs)
1944                 return 0;
1945
1946         /* don't need lock here; in the worst case we'll do useless copy */
1947         if (fs->users == 1)
1948                 return 0;
1949
1950         *new_fsp = copy_fs_struct(fs);
1951         if (!*new_fsp)
1952                 return -ENOMEM;
1953
1954         return 0;
1955 }
1956
1957 /*
1958  * Unshare file descriptor table if it is being shared
1959  */
1960 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1961 {
1962         struct files_struct *fd = current->files;
1963         int error = 0;
1964
1965         if ((unshare_flags & CLONE_FILES) &&
1966             (fd && atomic_read(&fd->count) > 1)) {
1967                 *new_fdp = dup_fd(fd, &error);
1968                 if (!*new_fdp)
1969                         return error;
1970         }
1971
1972         return 0;
1973 }
1974
1975 /*
1976  * unshare allows a process to 'unshare' part of the process
1977  * context which was originally shared using clone.  copy_*
1978  * functions used by do_fork() cannot be used here directly
1979  * because they modify an inactive task_struct that is being
1980  * constructed. Here we are modifying the current, active,
1981  * task_struct.
1982  */
1983 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1984 {
1985         struct fs_struct *fs, *new_fs = NULL;
1986         struct files_struct *fd, *new_fd = NULL;
1987         struct cred *new_cred = NULL;
1988         struct nsproxy *new_nsproxy = NULL;
1989         int do_sysvsem = 0;
1990         int err;
1991
1992         /*
1993          * If unsharing a user namespace must also unshare the thread group
1994          * and unshare the filesystem root and working directories.
1995          */
1996         if (unshare_flags & CLONE_NEWUSER)
1997                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1998         /*
1999          * If unsharing vm, must also unshare signal handlers.
2000          */
2001         if (unshare_flags & CLONE_VM)
2002                 unshare_flags |= CLONE_SIGHAND;
2003         /*
2004          * If unsharing a signal handlers, must also unshare the signal queues.
2005          */
2006         if (unshare_flags & CLONE_SIGHAND)
2007                 unshare_flags |= CLONE_THREAD;
2008         /*
2009          * If unsharing namespace, must also unshare filesystem information.
2010          */
2011         if (unshare_flags & CLONE_NEWNS)
2012                 unshare_flags |= CLONE_FS;
2013
2014         err = check_unshare_flags(unshare_flags);
2015         if (err)
2016                 goto bad_unshare_out;
2017         /*
2018          * CLONE_NEWIPC must also detach from the undolist: after switching
2019          * to a new ipc namespace, the semaphore arrays from the old
2020          * namespace are unreachable.
2021          */
2022         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2023                 do_sysvsem = 1;
2024         err = unshare_fs(unshare_flags, &new_fs);
2025         if (err)
2026                 goto bad_unshare_out;
2027         err = unshare_fd(unshare_flags, &new_fd);
2028         if (err)
2029                 goto bad_unshare_cleanup_fs;
2030         err = unshare_userns(unshare_flags, &new_cred);
2031         if (err)
2032                 goto bad_unshare_cleanup_fd;
2033         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2034                                          new_cred, new_fs);
2035         if (err)
2036                 goto bad_unshare_cleanup_cred;
2037
2038         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2039                 if (do_sysvsem) {
2040                         /*
2041                          * CLONE_SYSVSEM is equivalent to sys_exit().
2042                          */
2043                         exit_sem(current);
2044                 }
2045                 if (unshare_flags & CLONE_NEWIPC) {
2046                         /* Orphan segments in old ns (see sem above). */
2047                         exit_shm(current);
2048                         shm_init_task(current);
2049                 }
2050
2051                 if (new_nsproxy)
2052                         switch_task_namespaces(current, new_nsproxy);
2053
2054                 task_lock(current);
2055
2056                 if (new_fs) {
2057                         fs = current->fs;
2058                         spin_lock(&fs->lock);
2059                         current->fs = new_fs;
2060                         if (--fs->users)
2061                                 new_fs = NULL;
2062                         else
2063                                 new_fs = fs;
2064                         spin_unlock(&fs->lock);
2065                 }
2066
2067                 if (new_fd) {
2068                         fd = current->files;
2069                         current->files = new_fd;
2070                         new_fd = fd;
2071                 }
2072
2073                 task_unlock(current);
2074
2075                 if (new_cred) {
2076                         /* Install the new user namespace */
2077                         commit_creds(new_cred);
2078                         new_cred = NULL;
2079                 }
2080         }
2081
2082 bad_unshare_cleanup_cred:
2083         if (new_cred)
2084                 put_cred(new_cred);
2085 bad_unshare_cleanup_fd:
2086         if (new_fd)
2087                 put_files_struct(new_fd);
2088
2089 bad_unshare_cleanup_fs:
2090         if (new_fs)
2091                 free_fs_struct(new_fs);
2092
2093 bad_unshare_out:
2094         return err;
2095 }
2096
2097 /*
2098  *      Helper to unshare the files of the current task.
2099  *      We don't want to expose copy_files internals to
2100  *      the exec layer of the kernel.
2101  */
2102
2103 int unshare_files(struct files_struct **displaced)
2104 {
2105         struct task_struct *task = current;
2106         struct files_struct *copy = NULL;
2107         int error;
2108
2109         error = unshare_fd(CLONE_FILES, &copy);
2110         if (error || !copy) {
2111                 *displaced = NULL;
2112                 return error;
2113         }
2114         *displaced = task->files;
2115         task_lock(task);
2116         task->files = copy;
2117         task_unlock(task);
2118         return 0;
2119 }
2120
2121 int sysctl_max_threads(struct ctl_table *table, int write,
2122                        void __user *buffer, size_t *lenp, loff_t *ppos)
2123 {
2124         struct ctl_table t;
2125         int ret;
2126         int threads = max_threads;
2127         int min = MIN_THREADS;
2128         int max = MAX_THREADS;
2129
2130         t = *table;
2131         t.data = &threads;
2132         t.extra1 = &min;
2133         t.extra2 = &max;
2134
2135         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2136         if (ret || !write)
2137                 return ret;
2138
2139         set_max_threads(threads);
2140
2141         return 0;
2142 }