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 #include <linux/kcov.h>
  79
  80 #include <asm/pgtable.h>
  81 #include <asm/pgalloc.h>
  82 #include <asm/uaccess.h>
  83 #include <asm/mmu_context.h>
  84 #include <asm/cacheflush.h>
  85 #include <asm/tlbflush.h>
  86
  87 #include <trace/events/sched.h>
  88
  89 #define CREATE_TRACE_POINTS
  90 #include <trace/events/task.h>
  91
  92 /*
  93  * Minimum number of threads to boot the kernel
  94  */
  95 #define MIN_THREADS 20
  96
  97 /*
  98  * Maximum number of threads
  99  */
 100 #define MAX_THREADS FUTEX_TID_MASK
 101
 102 /*
 103  * Protected counters by write_lock_irq(&tasklist_lock)
 104  */
 105 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 106 int nr_threads;                 /* The idle threads do not count.. */
 107
 108 int max_threads;                /* tunable limit on nr_threads */
 109
 110 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 111
 112 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 113
 114 #ifdef CONFIG_PROVE_RCU
 115 int lockdep_tasklist_lock_is_held(void)
 116 {
 117         return lockdep_is_held(&tasklist_lock);
 118 }
 119 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 120 #endif /* #ifdef CONFIG_PROVE_RCU */
 121
 122 int nr_processes(void)
 123 {
 124         int cpu;
 125         int total = 0;
 126
 127         for_each_possible_cpu(cpu)
 128                 total += per_cpu(process_counts, cpu);
 129
 130         return total;
 131 }
 132
 133 void __weak arch_release_task_struct(struct task_struct *tsk)
 134 {
 135 }
 136
 137 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 138 static struct kmem_cache *task_struct_cachep;
 139
 140 static inline struct task_struct *alloc_task_struct_node(int node)
 141 {
 142         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 143 }
 144
 145 static inline void free_task_struct(struct task_struct *tsk)
 146 {
 147         kmem_cache_free(task_struct_cachep, tsk);
 148 }
 149 #endif
 150
 151 void __weak arch_release_thread_stack(unsigned long *stack)
 152 {
 153 }
 154
 155 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 156
 157 /*
 158  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 159  * kmemcache based allocator.
 160  */
 161 # if THREAD_SIZE >= PAGE_SIZE
 162 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 163                                                   int node)
 164 {
 165         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 166                                              THREAD_SIZE_ORDER);
 167
 168         return page ? page_address(page) : NULL;
 169 }
 170
 171 static inline void free_thread_stack(unsigned long *stack)
 172 {
 173         __free_pages(virt_to_page(stack), THREAD_SIZE_ORDER);
 174 }
 175 # else
 176 static struct kmem_cache *thread_stack_cache;
 177
 178 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 179                                                   int node)
 180 {
 181         return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 182 }
 183
 184 static void free_thread_stack(unsigned long *stack)
 185 {
 186         kmem_cache_free(thread_stack_cache, stack);
 187 }
 188
 189 void thread_stack_cache_init(void)
 190 {
 191         thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
 192                                               THREAD_SIZE, 0, NULL);
 193         BUG_ON(thread_stack_cache == NULL);
 194 }
 195 # endif
 196 #endif
 197
 198 /* SLAB cache for signal_struct structures (tsk->signal) */
 199 static struct kmem_cache *signal_cachep;
 200
 201 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 202 struct kmem_cache *sighand_cachep;
 203
 204 /* SLAB cache for files_struct structures (tsk->files) */
 205 struct kmem_cache *files_cachep;
 206
 207 /* SLAB cache for fs_struct structures (tsk->fs) */
 208 struct kmem_cache *fs_cachep;
 209
 210 /* SLAB cache for vm_area_struct structures */
 211 struct kmem_cache *vm_area_cachep;
 212
 213 /* SLAB cache for mm_struct structures (tsk->mm) */
 214 static struct kmem_cache *mm_cachep;
 215
 216 static void account_kernel_stack(unsigned long *stack, int account)
 217 {
 218         /* All stack pages are in the same zone and belong to the same memcg. */
 219         struct page *first_page = virt_to_page(stack);
 220
 221         mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 222                             THREAD_SIZE / 1024 * account);
 223
 224         memcg_kmem_update_page_stat(
 225                 first_page, MEMCG_KERNEL_STACK_KB,
 226                 account * (THREAD_SIZE / 1024));
 227 }
 228
 229 void free_task(struct task_struct *tsk)
 230 {
 231         account_kernel_stack(tsk->stack, -1);
 232         arch_release_thread_stack(tsk->stack);
 233         free_thread_stack(tsk->stack);
 234         rt_mutex_debug_task_free(tsk);
 235         ftrace_graph_exit_task(tsk);
 236         put_seccomp_filter(tsk);
 237         arch_release_task_struct(tsk);
 238         free_task_struct(tsk);
 239 }
 240 EXPORT_SYMBOL(free_task);
 241
 242 static inline void free_signal_struct(struct signal_struct *sig)
 243 {
 244         taskstats_tgid_free(sig);
 245         sched_autogroup_exit(sig);
 246         kmem_cache_free(signal_cachep, sig);
 247 }
 248
 249 static inline void put_signal_struct(struct signal_struct *sig)
 250 {
 251         if (atomic_dec_and_test(&sig->sigcnt))
 252                 free_signal_struct(sig);
 253 }
 254
 255 void __put_task_struct(struct task_struct *tsk)
 256 {
 257         WARN_ON(!tsk->exit_state);
 258         WARN_ON(atomic_read(&tsk->usage));
 259         WARN_ON(tsk == current);
 260
 261         cgroup_free(tsk);
 262         task_numa_free(tsk);
 263         security_task_free(tsk);
 264         exit_creds(tsk);
 265         delayacct_tsk_free(tsk);
 266         put_signal_struct(tsk->signal);
 267
 268         if (!profile_handoff_task(tsk))
 269                 free_task(tsk);
 270 }
 271 EXPORT_SYMBOL_GPL(__put_task_struct);
 272
 273 void __init __weak arch_task_cache_init(void) { }
 274
 275 /*
 276  * set_max_threads
 277  */
 278 static void set_max_threads(unsigned int max_threads_suggested)
 279 {
 280         u64 threads;
 281
 282         /*
 283          * The number of threads shall be limited such that the thread
 284          * structures may only consume a small part of the available memory.
 285          */
 286         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 287                 threads = MAX_THREADS;
 288         else
 289                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 290                                     (u64) THREAD_SIZE * 8UL);
 291
 292         if (threads > max_threads_suggested)
 293                 threads = max_threads_suggested;
 294
 295         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 296 }
 297
 298 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 299 /* Initialized by the architecture: */
 300 int arch_task_struct_size __read_mostly;
 301 #endif
 302
 303 void __init fork_init(void)
 304 {
 305 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 306 #ifndef ARCH_MIN_TASKALIGN
 307 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 308 #endif
 309         /* create a slab on which task_structs can be allocated */
 310         task_struct_cachep = kmem_cache_create("task_struct",
 311                         arch_task_struct_size, ARCH_MIN_TASKALIGN,
 312                         SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
 313 #endif
 314
 315         /* do the arch specific task caches init */
 316         arch_task_cache_init();
 317
 318         set_max_threads(MAX_THREADS);
 319
 320         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 321         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 322         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 323                 init_task.signal->rlim[RLIMIT_NPROC];
 324 }
 325
 326 int __weak arch_dup_task_struct(struct task_struct *dst,
 327                                                struct task_struct *src)
 328 {
 329         *dst = *src;
 330         return 0;
 331 }
 332
 333 void set_task_stack_end_magic(struct task_struct *tsk)
 334 {
 335         unsigned long *stackend;
 336
 337         stackend = end_of_stack(tsk);
 338         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 339 }
 340
 341 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 342 {
 343         struct task_struct *tsk;
 344         unsigned long *stack;
 345         int err;
 346
 347         if (node == NUMA_NO_NODE)
 348                 node = tsk_fork_get_node(orig);
 349         tsk = alloc_task_struct_node(node);
 350         if (!tsk)
 351                 return NULL;
 352
 353         stack = alloc_thread_stack_node(tsk, node);
 354         if (!stack)
 355                 goto free_tsk;
 356
 357         err = arch_dup_task_struct(tsk, orig);
 358         if (err)
 359                 goto free_stack;
 360
 361         tsk->stack = stack;
 362 #ifdef CONFIG_SECCOMP
 363         /*
 364          * We must handle setting up seccomp filters once we're under
 365          * the sighand lock in case orig has changed between now and
 366          * then. Until then, filter must be NULL to avoid messing up
 367          * the usage counts on the error path calling free_task.
 368          */
 369         tsk->seccomp.filter = NULL;
 370 #endif
 371
 372         setup_thread_stack(tsk, orig);
 373         clear_user_return_notifier(tsk);
 374         clear_tsk_need_resched(tsk);
 375         set_task_stack_end_magic(tsk);
 376
 377 #ifdef CONFIG_CC_STACKPROTECTOR
 378         tsk->stack_canary = get_random_int();
 379 #endif
 380
 381         /*
 382          * One for us, one for whoever does the "release_task()" (usually
 383          * parent)
 384          */
 385         atomic_set(&tsk->usage, 2);
 386 #ifdef CONFIG_BLK_DEV_IO_TRACE
 387         tsk->btrace_seq = 0;
 388 #endif
 389         tsk->splice_pipe = NULL;
 390         tsk->task_frag.page = NULL;
 391         tsk->wake_q.next = NULL;
 392
 393         account_kernel_stack(stack, 1);
 394
 395         kcov_task_init(tsk);
 396
 397         return tsk;
 398
 399 free_stack:
 400         free_thread_stack(stack);
 401 free_tsk:
 402         free_task_struct(tsk);
 403         return NULL;
 404 }
 405
 406 #ifdef CONFIG_MMU
 407 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 408 {
 409         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 410         struct rb_node **rb_link, *rb_parent;
 411         int retval;
 412         unsigned long charge;
 413
 414         uprobe_start_dup_mmap();
 415         if (down_write_killable(&oldmm->mmap_sem)) {
 416                 retval = -EINTR;
 417                 goto fail_uprobe_end;
 418         }
 419         flush_cache_dup_mm(oldmm);
 420         uprobe_dup_mmap(oldmm, mm);
 421         /*
 422          * Not linked in yet - no deadlock potential:
 423          */
 424         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 425
 426         /* No ordering required: file already has been exposed. */
 427         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 428
 429         mm->total_vm = oldmm->total_vm;
 430         mm->data_vm = oldmm->data_vm;
 431         mm->exec_vm = oldmm->exec_vm;
 432         mm->stack_vm = oldmm->stack_vm;
 433
 434         rb_link = &mm->mm_rb.rb_node;
 435         rb_parent = NULL;
 436         pprev = &mm->mmap;
 437         retval = ksm_fork(mm, oldmm);
 438         if (retval)
 439                 goto out;
 440         retval = khugepaged_fork(mm, oldmm);
 441         if (retval)
 442                 goto out;
 443
 444         prev = NULL;
 445         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 446                 struct file *file;
 447
 448                 if (mpnt->vm_flags & VM_DONTCOPY) {
 449                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 450                         continue;
 451                 }
 452                 charge = 0;
 453                 if (mpnt->vm_flags & VM_ACCOUNT) {
 454                         unsigned long len = vma_pages(mpnt);
 455
 456                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 457                                 goto fail_nomem;
 458                         charge = len;
 459                 }
 460                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 461                 if (!tmp)
 462                         goto fail_nomem;
 463                 *tmp = *mpnt;
 464                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 465                 retval = vma_dup_policy(mpnt, tmp);
 466                 if (retval)
 467                         goto fail_nomem_policy;
 468                 tmp->vm_mm = mm;
 469                 if (anon_vma_fork(tmp, mpnt))
 470                         goto fail_nomem_anon_vma_fork;
 471                 tmp->vm_flags &=
 472                         ~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
 473                 tmp->vm_next = tmp->vm_prev = NULL;
 474                 tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
 475                 file = tmp->vm_file;
 476                 if (file) {
 477                         struct inode *inode = file_inode(file);
 478                         struct address_space *mapping = file->f_mapping;
 479
 480                         get_file(file);
 481                         if (tmp->vm_flags & VM_DENYWRITE)
 482                                 atomic_dec(&inode->i_writecount);
 483                         i_mmap_lock_write(mapping);
 484                         if (tmp->vm_flags & VM_SHARED)
 485                                 atomic_inc(&mapping->i_mmap_writable);
 486                         flush_dcache_mmap_lock(mapping);
 487                         /* insert tmp into the share list, just after mpnt */
 488                         vma_interval_tree_insert_after(tmp, mpnt,
 489                                         &mapping->i_mmap);
 490                         flush_dcache_mmap_unlock(mapping);
 491                         i_mmap_unlock_write(mapping);
 492                 }
 493
 494                 /*
 495                  * Clear hugetlb-related page reserves for children. This only
 496                  * affects MAP_PRIVATE mappings. Faults generated by the child
 497                  * are not guaranteed to succeed, even if read-only
 498                  */
 499                 if (is_vm_hugetlb_page(tmp))
 500                         reset_vma_resv_huge_pages(tmp);
 501
 502                 /*
 503                  * Link in the new vma and copy the page table entries.
 504                  */
 505                 *pprev = tmp;
 506                 pprev = &tmp->vm_next;
 507                 tmp->vm_prev = prev;
 508                 prev = tmp;
 509
 510                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 511                 rb_link = &tmp->vm_rb.rb_right;
 512                 rb_parent = &tmp->vm_rb;
 513
 514                 mm->map_count++;
 515                 retval = copy_page_range(mm, oldmm, mpnt);
 516
 517                 if (tmp->vm_ops && tmp->vm_ops->open)
 518                         tmp->vm_ops->open(tmp);
 519
 520                 if (retval)
 521                         goto out;
 522         }
 523         /* a new mm has just been created */
 524         arch_dup_mmap(oldmm, mm);
 525         retval = 0;
 526 out:
 527         up_write(&mm->mmap_sem);
 528         flush_tlb_mm(oldmm);
 529         up_write(&oldmm->mmap_sem);
 530 fail_uprobe_end:
 531         uprobe_end_dup_mmap();
 532         return retval;
 533 fail_nomem_anon_vma_fork:
 534         mpol_put(vma_policy(tmp));
 535 fail_nomem_policy:
 536         kmem_cache_free(vm_area_cachep, tmp);
 537 fail_nomem:
 538         retval = -ENOMEM;
 539         vm_unacct_memory(charge);
 540         goto out;
 541 }
 542
 543 static inline int mm_alloc_pgd(struct mm_struct *mm)
 544 {
 545         mm->pgd = pgd_alloc(mm);
 546         if (unlikely(!mm->pgd))
 547                 return -ENOMEM;
 548         return 0;
 549 }
 550
 551 static inline void mm_free_pgd(struct mm_struct *mm)
 552 {
 553         pgd_free(mm, mm->pgd);
 554 }
 555 #else
 556 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 557 {
 558         down_write(&oldmm->mmap_sem);
 559         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 560         up_write(&oldmm->mmap_sem);
 561         return 0;
 562 }
 563 #define mm_alloc_pgd(mm)        (0)
 564 #define mm_free_pgd(mm)
 565 #endif /* CONFIG_MMU */
 566
 567 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 568
 569 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 570 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 571
 572 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 573
 574 static int __init coredump_filter_setup(char *s)
 575 {
 576         default_dump_filter =
 577                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 578                 MMF_DUMP_FILTER_MASK;
 579         return 1;
 580 }
 581
 582 __setup("coredump_filter=", coredump_filter_setup);
 583
 584 #include <linux/init_task.h>
 585
 586 static void mm_init_aio(struct mm_struct *mm)
 587 {
 588 #ifdef CONFIG_AIO
 589         spin_lock_init(&mm->ioctx_lock);
 590         mm->ioctx_table = NULL;
 591 #endif
 592 }
 593
 594 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 595 {
 596 #ifdef CONFIG_MEMCG
 597         mm->owner = p;
 598 #endif
 599 }
 600
 601 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 602 {
 603         mm->mmap = NULL;
 604         mm->mm_rb = RB_ROOT;
 605         mm->vmacache_seqnum = 0;
 606         atomic_set(&mm->mm_users, 1);
 607         atomic_set(&mm->mm_count, 1);
 608         init_rwsem(&mm->mmap_sem);
 609         INIT_LIST_HEAD(&mm->mmlist);
 610         mm->core_state = NULL;
 611         atomic_long_set(&mm->nr_ptes, 0);
 612         mm_nr_pmds_init(mm);
 613         mm->map_count = 0;
 614         mm->locked_vm = 0;
 615         mm->pinned_vm = 0;
 616         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 617         spin_lock_init(&mm->page_table_lock);
 618         mm_init_cpumask(mm);
 619         mm_init_aio(mm);
 620         mm_init_owner(mm, p);
 621         mmu_notifier_mm_init(mm);
 622         clear_tlb_flush_pending(mm);
 623 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 624         mm->pmd_huge_pte = NULL;
 625 #endif
 626
 627         if (current->mm) {
 628                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 629                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 630         } else {
 631                 mm->flags = default_dump_filter;
 632                 mm->def_flags = 0;
 633         }
 634
 635         if (mm_alloc_pgd(mm))
 636                 goto fail_nopgd;
 637
 638         if (init_new_context(p, mm))
 639                 goto fail_nocontext;
 640
 641         return mm;
 642
 643 fail_nocontext:
 644         mm_free_pgd(mm);
 645 fail_nopgd:
 646         free_mm(mm);
 647         return NULL;
 648 }
 649
 650 static void check_mm(struct mm_struct *mm)
 651 {
 652         int i;
 653
 654         for (i = 0; i < NR_MM_COUNTERS; i++) {
 655                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 656
 657                 if (unlikely(x))
 658                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 659                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 660         }
 661
 662         if (atomic_long_read(&mm->nr_ptes))
 663                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 664                                 atomic_long_read(&mm->nr_ptes));
 665         if (mm_nr_pmds(mm))
 666                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 667                                 mm_nr_pmds(mm));
 668
 669 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 670         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 671 #endif
 672 }
 673
 674 /*
 675  * Allocate and initialize an mm_struct.
 676  */
 677 struct mm_struct *mm_alloc(void)
 678 {
 679         struct mm_struct *mm;
 680
 681         mm = allocate_mm();
 682         if (!mm)
 683                 return NULL;
 684
 685         memset(mm, 0, sizeof(*mm));
 686         return mm_init(mm, current);
 687 }
 688
 689 /*
 690  * Called when the last reference to the mm
 691  * is dropped: either by a lazy thread or by
 692  * mmput. Free the page directory and the mm.
 693  */
 694 void __mmdrop(struct mm_struct *mm)
 695 {
 696         BUG_ON(mm == &init_mm);
 697         mm_free_pgd(mm);
 698         destroy_context(mm);
 699         mmu_notifier_mm_destroy(mm);
 700         check_mm(mm);
 701         free_mm(mm);
 702 }
 703 EXPORT_SYMBOL_GPL(__mmdrop);
 704
 705 static inline void __mmput(struct mm_struct *mm)
 706 {
 707         VM_BUG_ON(atomic_read(&mm->mm_users));
 708
 709         uprobe_clear_state(mm);
 710         exit_aio(mm);
 711         ksm_exit(mm);
 712         khugepaged_exit(mm); /* must run before exit_mmap */
 713         exit_mmap(mm);
 714         set_mm_exe_file(mm, NULL);
 715         if (!list_empty(&mm->mmlist)) {
 716                 spin_lock(&mmlist_lock);
 717                 list_del(&mm->mmlist);
 718                 spin_unlock(&mmlist_lock);
 719         }
 720         if (mm->binfmt)
 721                 module_put(mm->binfmt->module);
 722         mmdrop(mm);
 723 }
 724
 725 /*
 726  * Decrement the use count and release all resources for an mm.
 727  */
 728 void mmput(struct mm_struct *mm)
 729 {
 730         might_sleep();
 731
 732         if (atomic_dec_and_test(&mm->mm_users))
 733                 __mmput(mm);
 734 }
 735 EXPORT_SYMBOL_GPL(mmput);
 736
 737 #ifdef CONFIG_MMU
 738 static void mmput_async_fn(struct work_struct *work)
 739 {
 740         struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
 741         __mmput(mm);
 742 }
 743
 744 void mmput_async(struct mm_struct *mm)
 745 {
 746         if (atomic_dec_and_test(&mm->mm_users)) {
 747                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
 748                 schedule_work(&mm->async_put_work);
 749         }
 750 }
 751 #endif
 752
 753 /**
 754  * set_mm_exe_file - change a reference to the mm's executable file
 755  *
 756  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 757  *
 758  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 759  * invocations: in mmput() nobody alive left, in execve task is single
 760  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 761  * mm->exe_file, but does so without using set_mm_exe_file() in order
 762  * to do avoid the need for any locks.
 763  */
 764 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 765 {
 766         struct file *old_exe_file;
 767
 768         /*
 769          * It is safe to dereference the exe_file without RCU as
 770          * this function is only called if nobody else can access
 771          * this mm -- see comment above for justification.
 772          */
 773         old_exe_file = rcu_dereference_raw(mm->exe_file);
 774
 775         if (new_exe_file)
 776                 get_file(new_exe_file);
 777         rcu_assign_pointer(mm->exe_file, new_exe_file);
 778         if (old_exe_file)
 779                 fput(old_exe_file);
 780 }
 781
 782 /**
 783  * get_mm_exe_file - acquire a reference to the mm's executable file
 784  *
 785  * Returns %NULL if mm has no associated executable file.
 786  * User must release file via fput().
 787  */
 788 struct file *get_mm_exe_file(struct mm_struct *mm)
 789 {
 790         struct file *exe_file;
 791
 792         rcu_read_lock();
 793         exe_file = rcu_dereference(mm->exe_file);
 794         if (exe_file && !get_file_rcu(exe_file))
 795                 exe_file = NULL;
 796         rcu_read_unlock();
 797         return exe_file;
 798 }
 799 EXPORT_SYMBOL(get_mm_exe_file);
 800
 801 /**
 802  * get_task_mm - acquire a reference to the task's mm
 803  *
 804  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 805  * this kernel workthread has transiently adopted a user mm with use_mm,
 806  * to do its AIO) is not set and if so returns a reference to it, after
 807  * bumping up the use count.  User must release the mm via mmput()
 808  * after use.  Typically used by /proc and ptrace.
 809  */
 810 struct mm_struct *get_task_mm(struct task_struct *task)
 811 {
 812         struct mm_struct *mm;
 813
 814         task_lock(task);
 815         mm = task->mm;
 816         if (mm) {
 817                 if (task->flags & PF_KTHREAD)
 818                         mm = NULL;
 819                 else
 820                         atomic_inc(&mm->mm_users);
 821         }
 822         task_unlock(task);
 823         return mm;
 824 }
 825 EXPORT_SYMBOL_GPL(get_task_mm);
 826
 827 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 828 {
 829         struct mm_struct *mm;
 830         int err;
 831
 832         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 833         if (err)
 834                 return ERR_PTR(err);
 835
 836         mm = get_task_mm(task);
 837         if (mm && mm != current->mm &&
 838                         !ptrace_may_access(task, mode)) {
 839                 mmput(mm);
 840                 mm = ERR_PTR(-EACCES);
 841         }
 842         mutex_unlock(&task->signal->cred_guard_mutex);
 843
 844         return mm;
 845 }
 846
 847 static void complete_vfork_done(struct task_struct *tsk)
 848 {
 849         struct completion *vfork;
 850
 851         task_lock(tsk);
 852         vfork = tsk->vfork_done;
 853         if (likely(vfork)) {
 854                 tsk->vfork_done = NULL;
 855                 complete(vfork);
 856         }
 857         task_unlock(tsk);
 858 }
 859
 860 static int wait_for_vfork_done(struct task_struct *child,
 861                                 struct completion *vfork)
 862 {
 863         int killed;
 864
 865         freezer_do_not_count();
 866         killed = wait_for_completion_killable(vfork);
 867         freezer_count();
 868
 869         if (killed) {
 870                 task_lock(child);
 871                 child->vfork_done = NULL;
 872                 task_unlock(child);
 873         }
 874
 875         put_task_struct(child);
 876         return killed;
 877 }
 878
 879 /* Please note the differences between mmput and mm_release.
 880  * mmput is called whenever we stop holding onto a mm_struct,
 881  * error success whatever.
 882  *
 883  * mm_release is called after a mm_struct has been removed
 884  * from the current process.
 885  *
 886  * This difference is important for error handling, when we
 887  * only half set up a mm_struct for a new process and need to restore
 888  * the old one.  Because we mmput the new mm_struct before
 889  * restoring the old one. . .
 890  * Eric Biederman 10 January 1998
 891  */
 892 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 893 {
 894         /* Get rid of any futexes when releasing the mm */
 895 #ifdef CONFIG_FUTEX
 896         if (unlikely(tsk->robust_list)) {
 897                 exit_robust_list(tsk);
 898                 tsk->robust_list = NULL;
 899         }
 900 #ifdef CONFIG_COMPAT
 901         if (unlikely(tsk->compat_robust_list)) {
 902                 compat_exit_robust_list(tsk);
 903                 tsk->compat_robust_list = NULL;
 904         }
 905 #endif
 906         if (unlikely(!list_empty(&tsk->pi_state_list)))
 907                 exit_pi_state_list(tsk);
 908 #endif
 909
 910         uprobe_free_utask(tsk);
 911
 912         /* Get rid of any cached register state */
 913         deactivate_mm(tsk, mm);
 914
 915         /*
 916          * If we're exiting normally, clear a user-space tid field if
 917          * requested.  We leave this alone when dying by signal, to leave
 918          * the value intact in a core dump, and to save the unnecessary
 919          * trouble, say, a killed vfork parent shouldn't touch this mm.
 920          * Userland only wants this done for a sys_exit.
 921          */
 922         if (tsk->clear_child_tid) {
 923                 if (!(tsk->flags & PF_SIGNALED) &&
 924                     atomic_read(&mm->mm_users) > 1) {
 925                         /*
 926                          * We don't check the error code - if userspace has
 927                          * not set up a proper pointer then tough luck.
 928                          */
 929                         put_user(0, tsk->clear_child_tid);
 930                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 931                                         1, NULL, NULL, 0);
 932                 }
 933                 tsk->clear_child_tid = NULL;
 934         }
 935
 936         /*
 937          * All done, finally we can wake up parent and return this mm to him.
 938          * Also kthread_stop() uses this completion for synchronization.
 939          */
 940         if (tsk->vfork_done)
 941                 complete_vfork_done(tsk);
 942 }
 943
 944 /*
 945  * Allocate a new mm structure and copy contents from the
 946  * mm structure of the passed in task structure.
 947  */
 948 static struct mm_struct *dup_mm(struct task_struct *tsk)
 949 {
 950         struct mm_struct *mm, *oldmm = current->mm;
 951         int err;
 952
 953         mm = allocate_mm();
 954         if (!mm)
 955                 goto fail_nomem;
 956
 957         memcpy(mm, oldmm, sizeof(*mm));
 958
 959         if (!mm_init(mm, tsk))
 960                 goto fail_nomem;
 961
 962         err = dup_mmap(mm, oldmm);
 963         if (err)
 964                 goto free_pt;
 965
 966         mm->hiwater_rss = get_mm_rss(mm);
 967         mm->hiwater_vm = mm->total_vm;
 968
 969         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 970                 goto free_pt;
 971
 972         return mm;
 973
 974 free_pt:
 975         /* don't put binfmt in mmput, we haven't got module yet */
 976         mm->binfmt = NULL;
 977         mmput(mm);
 978
 979 fail_nomem:
 980         return NULL;
 981 }
 982
 983 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 984 {
 985         struct mm_struct *mm, *oldmm;
 986         int retval;
 987
 988         tsk->min_flt = tsk->maj_flt = 0;
 989         tsk->nvcsw = tsk->nivcsw = 0;
 990 #ifdef CONFIG_DETECT_HUNG_TASK
 991         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 992 #endif
 993
 994         tsk->mm = NULL;
 995         tsk->active_mm = NULL;
 996
 997         /*
 998          * Are we cloning a kernel thread?
 999          *
1000          * We need to steal a active VM for that..
1001          */
1002         oldmm = current->mm;
1003         if (!oldmm)
1004                 return 0;
1005
1006         /* initialize the new vmacache entries */
1007         vmacache_flush(tsk);
1008
1009         if (clone_flags & CLONE_VM) {
1010                 atomic_inc(&oldmm->mm_users);
1011                 mm = oldmm;
1012                 goto good_mm;
1013         }
1014
1015         retval = -ENOMEM;
1016         mm = dup_mm(tsk);
1017         if (!mm)
1018                 goto fail_nomem;
1019
1020 good_mm:
1021         tsk->mm = mm;
1022         tsk->active_mm = mm;
1023         return 0;
1024
1025 fail_nomem:
1026         return retval;
1027 }
1028
1029 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1030 {
1031         struct fs_struct *fs = current->fs;
1032         if (clone_flags & CLONE_FS) {
1033                 /* tsk->fs is already what we want */
1034                 spin_lock(&fs->lock);
1035                 if (fs->in_exec) {
1036                         spin_unlock(&fs->lock);
1037                         return -EAGAIN;
1038                 }
1039                 fs->users++;
1040                 spin_unlock(&fs->lock);
1041                 return 0;
1042         }
1043         tsk->fs = copy_fs_struct(fs);
1044         if (!tsk->fs)
1045                 return -ENOMEM;
1046         return 0;
1047 }
1048
1049 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1050 {
1051         struct files_struct *oldf, *newf;
1052         int error = 0;
1053
1054         /*
1055          * A background process may not have any files ...
1056          */
1057         oldf = current->files;
1058         if (!oldf)
1059                 goto out;
1060
1061         if (clone_flags & CLONE_FILES) {
1062                 atomic_inc(&oldf->count);
1063                 goto out;
1064         }
1065
1066         newf = dup_fd(oldf, &error);
1067         if (!newf)
1068                 goto out;
1069
1070         tsk->files = newf;
1071         error = 0;
1072 out:
1073         return error;
1074 }
1075
1076 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1077 {
1078 #ifdef CONFIG_BLOCK
1079         struct io_context *ioc = current->io_context;
1080         struct io_context *new_ioc;
1081
1082         if (!ioc)
1083                 return 0;
1084         /*
1085          * Share io context with parent, if CLONE_IO is set
1086          */
1087         if (clone_flags & CLONE_IO) {
1088                 ioc_task_link(ioc);
1089                 tsk->io_context = ioc;
1090         } else if (ioprio_valid(ioc->ioprio)) {
1091                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1092                 if (unlikely(!new_ioc))
1093                         return -ENOMEM;
1094
1095                 new_ioc->ioprio = ioc->ioprio;
1096                 put_io_context(new_ioc);
1097         }
1098 #endif
1099         return 0;
1100 }
1101
1102 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1103 {
1104         struct sighand_struct *sig;
1105
1106         if (clone_flags & CLONE_SIGHAND) {
1107                 atomic_inc(&current->sighand->count);
1108                 return 0;
1109         }
1110         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1111         rcu_assign_pointer(tsk->sighand, sig);
1112         if (!sig)
1113                 return -ENOMEM;
1114
1115         atomic_set(&sig->count, 1);
1116         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1117         return 0;
1118 }
1119
1120 void __cleanup_sighand(struct sighand_struct *sighand)
1121 {
1122         if (atomic_dec_and_test(&sighand->count)) {
1123                 signalfd_cleanup(sighand);
1124                 /*
1125                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1126                  * without an RCU grace period, see __lock_task_sighand().
1127                  */
1128                 kmem_cache_free(sighand_cachep, sighand);
1129         }
1130 }
1131
1132 /*
1133  * Initialize POSIX timer handling for a thread group.
1134  */
1135 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1136 {
1137         unsigned long cpu_limit;
1138
1139         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1140         if (cpu_limit != RLIM_INFINITY) {
1141                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1142                 sig->cputimer.running = true;
1143         }
1144
1145         /* The timer lists. */
1146         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1147         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1148         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1149 }
1150
1151 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1152 {
1153         struct signal_struct *sig;
1154
1155         if (clone_flags & CLONE_THREAD)
1156                 return 0;
1157
1158         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1159         tsk->signal = sig;
1160         if (!sig)
1161                 return -ENOMEM;
1162
1163         sig->nr_threads = 1;
1164         atomic_set(&sig->live, 1);
1165         atomic_set(&sig->sigcnt, 1);
1166
1167         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1168         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1169         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1170
1171         init_waitqueue_head(&sig->wait_chldexit);
1172         sig->curr_target = tsk;
1173         init_sigpending(&sig->shared_pending);
1174         INIT_LIST_HEAD(&sig->posix_timers);
1175         seqlock_init(&sig->stats_lock);
1176         prev_cputime_init(&sig->prev_cputime);
1177
1178         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1179         sig->real_timer.function = it_real_fn;
1180
1181         task_lock(current->group_leader);
1182         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1183         task_unlock(current->group_leader);
1184
1185         posix_cpu_timers_init_group(sig);
1186
1187         tty_audit_fork(sig);
1188         sched_autogroup_fork(sig);
1189
1190         sig->oom_score_adj = current->signal->oom_score_adj;
1191         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1192
1193         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1194                                    current->signal->is_child_subreaper;
1195
1196         mutex_init(&sig->cred_guard_mutex);
1197
1198         return 0;
1199 }
1200
1201 static void copy_seccomp(struct task_struct *p)
1202 {
1203 #ifdef CONFIG_SECCOMP
1204         /*
1205          * Must be called with sighand->lock held, which is common to
1206          * all threads in the group. Holding cred_guard_mutex is not
1207          * needed because this new task is not yet running and cannot
1208          * be racing exec.
1209          */
1210         assert_spin_locked(&current->sighand->siglock);
1211
1212         /* Ref-count the new filter user, and assign it. */
1213         get_seccomp_filter(current);
1214         p->seccomp = current->seccomp;
1215
1216         /*
1217          * Explicitly enable no_new_privs here in case it got set
1218          * between the task_struct being duplicated and holding the
1219          * sighand lock. The seccomp state and nnp must be in sync.
1220          */
1221         if (task_no_new_privs(current))
1222                 task_set_no_new_privs(p);
1223
1224         /*
1225          * If the parent gained a seccomp mode after copying thread
1226          * flags and between before we held the sighand lock, we have
1227          * to manually enable the seccomp thread flag here.
1228          */
1229         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1230                 set_tsk_thread_flag(p, TIF_SECCOMP);
1231 #endif
1232 }
1233
1234 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1235 {
1236         current->clear_child_tid = tidptr;
1237
1238         return task_pid_vnr(current);
1239 }
1240
1241 static void rt_mutex_init_task(struct task_struct *p)
1242 {
1243         raw_spin_lock_init(&p->pi_lock);
1244 #ifdef CONFIG_RT_MUTEXES
1245         p->pi_waiters = RB_ROOT;
1246         p->pi_waiters_leftmost = NULL;
1247         p->pi_blocked_on = NULL;
1248 #endif
1249 }
1250
1251 /*
1252  * Initialize POSIX timer handling for a single task.
1253  */
1254 static void posix_cpu_timers_init(struct task_struct *tsk)
1255 {
1256         tsk->cputime_expires.prof_exp = 0;
1257         tsk->cputime_expires.virt_exp = 0;
1258         tsk->cputime_expires.sched_exp = 0;
1259         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1260         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1261         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1262 }
1263
1264 static inline void
1265 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1266 {
1267          task->pids[type].pid = pid;
1268 }
1269
1270 /*
1271  * This creates a new process as a copy of the old one,
1272  * but does not actually start it yet.
1273  *
1274  * It copies the registers, and all the appropriate
1275  * parts of the process environment (as per the clone
1276  * flags). The actual kick-off is left to the caller.
1277  */
1278 static struct task_struct *copy_process(unsigned long clone_flags,
1279                                         unsigned long stack_start,
1280                                         unsigned long stack_size,
1281                                         int __user *child_tidptr,
1282                                         struct pid *pid,
1283                                         int trace,
1284                                         unsigned long tls,
1285                                         int node)
1286 {
1287         int retval;
1288         struct task_struct *p;
1289
1290         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1291                 return ERR_PTR(-EINVAL);
1292
1293         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1294                 return ERR_PTR(-EINVAL);
1295
1296         /*
1297          * Thread groups must share signals as well, and detached threads
1298          * can only be started up within the thread group.
1299          */
1300         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1301                 return ERR_PTR(-EINVAL);
1302
1303         /*
1304          * Shared signal handlers imply shared VM. By way of the above,
1305          * thread groups also imply shared VM. Blocking this case allows
1306          * for various simplifications in other code.
1307          */
1308         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1309                 return ERR_PTR(-EINVAL);
1310
1311         /*
1312          * Siblings of global init remain as zombies on exit since they are
1313          * not reaped by their parent (swapper). To solve this and to avoid
1314          * multi-rooted process trees, prevent global and container-inits
1315          * from creating siblings.
1316          */
1317         if ((clone_flags & CLONE_PARENT) &&
1318                                 current->signal->flags & SIGNAL_UNKILLABLE)
1319                 return ERR_PTR(-EINVAL);
1320
1321         /*
1322          * If the new process will be in a different pid or user namespace
1323          * do not allow it to share a thread group with the forking task.
1324          */
1325         if (clone_flags & CLONE_THREAD) {
1326                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1327                     (task_active_pid_ns(current) !=
1328                                 current->nsproxy->pid_ns_for_children))
1329                         return ERR_PTR(-EINVAL);
1330         }
1331
1332         retval = security_task_create(clone_flags);
1333         if (retval)
1334                 goto fork_out;
1335
1336         retval = -ENOMEM;
1337         p = dup_task_struct(current, node);
1338         if (!p)
1339                 goto fork_out;
1340
1341         ftrace_graph_init_task(p);
1342
1343         rt_mutex_init_task(p);
1344
1345 #ifdef CONFIG_PROVE_LOCKING
1346         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1347         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1348 #endif
1349         retval = -EAGAIN;
1350         if (atomic_read(&p->real_cred->user->processes) >=
1351                         task_rlimit(p, RLIMIT_NPROC)) {
1352                 if (p->real_cred->user != INIT_USER &&
1353                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1354                         goto bad_fork_free;
1355         }
1356         current->flags &= ~PF_NPROC_EXCEEDED;
1357
1358         retval = copy_creds(p, clone_flags);
1359         if (retval < 0)
1360                 goto bad_fork_free;
1361
1362         /*
1363          * If multiple threads are within copy_process(), then this check
1364          * triggers too late. This doesn't hurt, the check is only there
1365          * to stop root fork bombs.
1366          */
1367         retval = -EAGAIN;
1368         if (nr_threads >= max_threads)
1369                 goto bad_fork_cleanup_count;
1370
1371         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1372         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1373         p->flags |= PF_FORKNOEXEC;
1374         INIT_LIST_HEAD(&p->children);
1375         INIT_LIST_HEAD(&p->sibling);
1376         rcu_copy_process(p);
1377         p->vfork_done = NULL;
1378         spin_lock_init(&p->alloc_lock);
1379
1380         init_sigpending(&p->pending);
1381
1382         p->utime = p->stime = p->gtime = 0;
1383         p->utimescaled = p->stimescaled = 0;
1384         prev_cputime_init(&p->prev_cputime);
1385
1386 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1387         seqcount_init(&p->vtime_seqcount);
1388         p->vtime_snap = 0;
1389         p->vtime_snap_whence = VTIME_INACTIVE;
1390 #endif
1391
1392 #if defined(SPLIT_RSS_COUNTING)
1393         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1394 #endif
1395
1396         p->default_timer_slack_ns = current->timer_slack_ns;
1397
1398         task_io_accounting_init(&p->ioac);
1399         acct_clear_integrals(p);
1400
1401         posix_cpu_timers_init(p);
1402
1403         p->start_time = ktime_get_ns();
1404         p->real_start_time = ktime_get_boot_ns();
1405         p->io_context = NULL;
1406         p->audit_context = NULL;
1407         threadgroup_change_begin(current);
1408         cgroup_fork(p);
1409 #ifdef CONFIG_NUMA
1410         p->mempolicy = mpol_dup(p->mempolicy);
1411         if (IS_ERR(p->mempolicy)) {
1412                 retval = PTR_ERR(p->mempolicy);
1413                 p->mempolicy = NULL;
1414                 goto bad_fork_cleanup_threadgroup_lock;
1415         }
1416 #endif
1417 #ifdef CONFIG_CPUSETS
1418         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1419         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1420         seqcount_init(&p->mems_allowed_seq);
1421 #endif
1422 #ifdef CONFIG_TRACE_IRQFLAGS
1423         p->irq_events = 0;
1424         p->hardirqs_enabled = 0;
1425         p->hardirq_enable_ip = 0;
1426         p->hardirq_enable_event = 0;
1427         p->hardirq_disable_ip = _THIS_IP_;
1428         p->hardirq_disable_event = 0;
1429         p->softirqs_enabled = 1;
1430         p->softirq_enable_ip = _THIS_IP_;
1431         p->softirq_enable_event = 0;
1432         p->softirq_disable_ip = 0;
1433         p->softirq_disable_event = 0;
1434         p->hardirq_context = 0;
1435         p->softirq_context = 0;
1436 #endif
1437
1438         p->pagefault_disabled = 0;
1439
1440 #ifdef CONFIG_LOCKDEP
1441         p->lockdep_depth = 0; /* no locks held yet */
1442         p->curr_chain_key = 0;
1443         p->lockdep_recursion = 0;
1444 #endif
1445
1446 #ifdef CONFIG_DEBUG_MUTEXES
1447         p->blocked_on = NULL; /* not blocked yet */
1448 #endif
1449 #ifdef CONFIG_BCACHE
1450         p->sequential_io        = 0;
1451         p->sequential_io_avg    = 0;
1452 #endif
1453
1454         /* Perform scheduler related setup. Assign this task to a CPU. */
1455         retval = sched_fork(clone_flags, p);
1456         if (retval)
1457                 goto bad_fork_cleanup_policy;
1458
1459         retval = perf_event_init_task(p);
1460         if (retval)
1461                 goto bad_fork_cleanup_policy;
1462         retval = audit_alloc(p);
1463         if (retval)
1464                 goto bad_fork_cleanup_perf;
1465         /* copy all the process information */
1466         shm_init_task(p);
1467         retval = copy_semundo(clone_flags, p);
1468         if (retval)
1469                 goto bad_fork_cleanup_audit;
1470         retval = copy_files(clone_flags, p);
1471         if (retval)
1472                 goto bad_fork_cleanup_semundo;
1473         retval = copy_fs(clone_flags, p);
1474         if (retval)
1475                 goto bad_fork_cleanup_files;
1476         retval = copy_sighand(clone_flags, p);
1477         if (retval)
1478                 goto bad_fork_cleanup_fs;
1479         retval = copy_signal(clone_flags, p);
1480         if (retval)
1481                 goto bad_fork_cleanup_sighand;
1482         retval = copy_mm(clone_flags, p);
1483         if (retval)
1484                 goto bad_fork_cleanup_signal;
1485         retval = copy_namespaces(clone_flags, p);
1486         if (retval)
1487                 goto bad_fork_cleanup_mm;
1488         retval = copy_io(clone_flags, p);
1489         if (retval)
1490                 goto bad_fork_cleanup_namespaces;
1491         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1492         if (retval)
1493                 goto bad_fork_cleanup_io;
1494
1495         if (pid != &init_struct_pid) {
1496                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1497                 if (IS_ERR(pid)) {
1498                         retval = PTR_ERR(pid);
1499                         goto bad_fork_cleanup_thread;
1500                 }
1501         }
1502
1503         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1504         /*
1505          * Clear TID on mm_release()?
1506          */
1507         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1508 #ifdef CONFIG_BLOCK
1509         p->plug = NULL;
1510 #endif
1511 #ifdef CONFIG_FUTEX
1512         p->robust_list = NULL;
1513 #ifdef CONFIG_COMPAT
1514         p->compat_robust_list = NULL;
1515 #endif
1516         INIT_LIST_HEAD(&p->pi_state_list);
1517         p->pi_state_cache = NULL;
1518 #endif
1519         /*
1520          * sigaltstack should be cleared when sharing the same VM
1521          */
1522         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1523                 sas_ss_reset(p);
1524
1525         /*
1526          * Syscall tracing and stepping should be turned off in the
1527          * child regardless of CLONE_PTRACE.
1528          */
1529         user_disable_single_step(p);
1530         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1531 #ifdef TIF_SYSCALL_EMU
1532         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1533 #endif
1534         clear_all_latency_tracing(p);
1535
1536         /* ok, now we should be set up.. */
1537         p->pid = pid_nr(pid);
1538         if (clone_flags & CLONE_THREAD) {
1539                 p->exit_signal = -1;
1540                 p->group_leader = current->group_leader;
1541                 p->tgid = current->tgid;
1542         } else {
1543                 if (clone_flags & CLONE_PARENT)
1544                         p->exit_signal = current->group_leader->exit_signal;
1545                 else
1546                         p->exit_signal = (clone_flags & CSIGNAL);
1547                 p->group_leader = p;
1548                 p->tgid = p->pid;
1549         }
1550
1551         p->nr_dirtied = 0;
1552         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1553         p->dirty_paused_when = 0;
1554
1555         p->pdeath_signal = 0;
1556         INIT_LIST_HEAD(&p->thread_group);
1557         p->task_works = NULL;
1558
1559         /*
1560          * Ensure that the cgroup subsystem policies allow the new process to be
1561          * forked. It should be noted the the new process's css_set can be changed
1562          * between here and cgroup_post_fork() if an organisation operation is in
1563          * progress.
1564          */
1565         retval = cgroup_can_fork(p);
1566         if (retval)
1567                 goto bad_fork_free_pid;
1568
1569         /*
1570          * Make it visible to the rest of the system, but dont wake it up yet.
1571          * Need tasklist lock for parent etc handling!
1572          */
1573         write_lock_irq(&tasklist_lock);
1574
1575         /* CLONE_PARENT re-uses the old parent */
1576         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1577                 p->real_parent = current->real_parent;
1578                 p->parent_exec_id = current->parent_exec_id;
1579         } else {
1580                 p->real_parent = current;
1581                 p->parent_exec_id = current->self_exec_id;
1582         }
1583
1584         spin_lock(&current->sighand->siglock);
1585
1586         /*
1587          * Copy seccomp details explicitly here, in case they were changed
1588          * before holding sighand lock.
1589          */
1590         copy_seccomp(p);
1591
1592         /*
1593          * Process group and session signals need to be delivered to just the
1594          * parent before the fork or both the parent and the child after the
1595          * fork. Restart if a signal comes in before we add the new process to
1596          * it's process group.
1597          * A fatal signal pending means that current will exit, so the new
1598          * thread can't slip out of an OOM kill (or normal SIGKILL).
1599         */
1600         recalc_sigpending();
1601         if (signal_pending(current)) {
1602                 spin_unlock(&current->sighand->siglock);
1603                 write_unlock_irq(&tasklist_lock);
1604                 retval = -ERESTARTNOINTR;
1605                 goto bad_fork_cancel_cgroup;
1606         }
1607
1608         if (likely(p->pid)) {
1609                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1610
1611                 init_task_pid(p, PIDTYPE_PID, pid);
1612                 if (thread_group_leader(p)) {
1613                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1614                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1615
1616                         if (is_child_reaper(pid)) {
1617                                 ns_of_pid(pid)->child_reaper = p;
1618                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1619                         }
1620
1621                         p->signal->leader_pid = pid;
1622                         p->signal->tty = tty_kref_get(current->signal->tty);
1623                         list_add_tail(&p->sibling, &p->real_parent->children);
1624                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1625                         attach_pid(p, PIDTYPE_PGID);
1626                         attach_pid(p, PIDTYPE_SID);
1627                         __this_cpu_inc(process_counts);
1628                 } else {
1629                         current->signal->nr_threads++;
1630                         atomic_inc(&current->signal->live);
1631                         atomic_inc(&current->signal->sigcnt);
1632                         list_add_tail_rcu(&p->thread_group,
1633                                           &p->group_leader->thread_group);
1634                         list_add_tail_rcu(&p->thread_node,
1635                                           &p->signal->thread_head);
1636                 }
1637                 attach_pid(p, PIDTYPE_PID);
1638                 nr_threads++;
1639         }
1640
1641         total_forks++;
1642         spin_unlock(&current->sighand->siglock);
1643         syscall_tracepoint_update(p);
1644         write_unlock_irq(&tasklist_lock);
1645
1646         proc_fork_connector(p);
1647         cgroup_post_fork(p);
1648         threadgroup_change_end(current);
1649         perf_event_fork(p);
1650
1651         trace_task_newtask(p, clone_flags);
1652         uprobe_copy_process(p, clone_flags);
1653
1654         return p;
1655
1656 bad_fork_cancel_cgroup:
1657         cgroup_cancel_fork(p);
1658 bad_fork_free_pid:
1659         if (pid != &init_struct_pid)
1660                 free_pid(pid);
1661 bad_fork_cleanup_thread:
1662         exit_thread(p);
1663 bad_fork_cleanup_io:
1664         if (p->io_context)
1665                 exit_io_context(p);
1666 bad_fork_cleanup_namespaces:
1667         exit_task_namespaces(p);
1668 bad_fork_cleanup_mm:
1669         if (p->mm)
1670                 mmput(p->mm);
1671 bad_fork_cleanup_signal:
1672         if (!(clone_flags & CLONE_THREAD))
1673                 free_signal_struct(p->signal);
1674 bad_fork_cleanup_sighand:
1675         __cleanup_sighand(p->sighand);
1676 bad_fork_cleanup_fs:
1677         exit_fs(p); /* blocking */
1678 bad_fork_cleanup_files:
1679         exit_files(p); /* blocking */
1680 bad_fork_cleanup_semundo:
1681         exit_sem(p);
1682 bad_fork_cleanup_audit:
1683         audit_free(p);
1684 bad_fork_cleanup_perf:
1685         perf_event_free_task(p);
1686 bad_fork_cleanup_policy:
1687 #ifdef CONFIG_NUMA
1688         mpol_put(p->mempolicy);
1689 bad_fork_cleanup_threadgroup_lock:
1690 #endif
1691         threadgroup_change_end(current);
1692         delayacct_tsk_free(p);
1693 bad_fork_cleanup_count:
1694         atomic_dec(&p->cred->user->processes);
1695         exit_creds(p);
1696 bad_fork_free:
1697         free_task(p);
1698 fork_out:
1699         return ERR_PTR(retval);
1700 }
1701
1702 static inline void init_idle_pids(struct pid_link *links)
1703 {
1704         enum pid_type type;
1705
1706         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1707                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1708                 links[type].pid = &init_struct_pid;
1709         }
1710 }
1711
1712 struct task_struct *fork_idle(int cpu)
1713 {
1714         struct task_struct *task;
1715         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1716                             cpu_to_node(cpu));
1717         if (!IS_ERR(task)) {
1718                 init_idle_pids(task->pids);
1719                 init_idle(task, cpu);
1720         }
1721
1722         return task;
1723 }
1724
1725 /*
1726  *  Ok, this is the main fork-routine.
1727  *
1728  * It copies the process, and if successful kick-starts
1729  * it and waits for it to finish using the VM if required.
1730  */
1731 long _do_fork(unsigned long clone_flags,
1732               unsigned long stack_start,
1733               unsigned long stack_size,
1734               int __user *parent_tidptr,
1735               int __user *child_tidptr,
1736               unsigned long tls)
1737 {
1738         struct task_struct *p;
1739         int trace = 0;
1740         long nr;
1741
1742         /*
1743          * Determine whether and which event to report to ptracer.  When
1744          * called from kernel_thread or CLONE_UNTRACED is explicitly
1745          * requested, no event is reported; otherwise, report if the event
1746          * for the type of forking is enabled.
1747          */
1748         if (!(clone_flags & CLONE_UNTRACED)) {
1749                 if (clone_flags & CLONE_VFORK)
1750                         trace = PTRACE_EVENT_VFORK;
1751                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1752                         trace = PTRACE_EVENT_CLONE;
1753                 else
1754                         trace = PTRACE_EVENT_FORK;
1755
1756                 if (likely(!ptrace_event_enabled(current, trace)))
1757                         trace = 0;
1758         }
1759
1760         p = copy_process(clone_flags, stack_start, stack_size,
1761                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
1762         /*
1763          * Do this prior waking up the new thread - the thread pointer
1764          * might get invalid after that point, if the thread exits quickly.
1765          */
1766         if (!IS_ERR(p)) {
1767                 struct completion vfork;
1768                 struct pid *pid;
1769
1770                 trace_sched_process_fork(current, p);
1771
1772                 pid = get_task_pid(p, PIDTYPE_PID);
1773                 nr = pid_vnr(pid);
1774
1775                 if (clone_flags & CLONE_PARENT_SETTID)
1776                         put_user(nr, parent_tidptr);
1777
1778                 if (clone_flags & CLONE_VFORK) {
1779                         p->vfork_done = &vfork;
1780                         init_completion(&vfork);
1781                         get_task_struct(p);
1782                 }
1783
1784                 wake_up_new_task(p);
1785
1786                 /* forking complete and child started to run, tell ptracer */
1787                 if (unlikely(trace))
1788                         ptrace_event_pid(trace, pid);
1789
1790                 if (clone_flags & CLONE_VFORK) {
1791                         if (!wait_for_vfork_done(p, &vfork))
1792                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1793                 }
1794
1795                 put_pid(pid);
1796         } else {
1797                 nr = PTR_ERR(p);
1798         }
1799         return nr;
1800 }
1801
1802 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1803 /* For compatibility with architectures that call do_fork directly rather than
1804  * using the syscall entry points below. */
1805 long do_fork(unsigned long clone_flags,
1806               unsigned long stack_start,
1807               unsigned long stack_size,
1808               int __user *parent_tidptr,
1809               int __user *child_tidptr)
1810 {
1811         return _do_fork(clone_flags, stack_start, stack_size,
1812                         parent_tidptr, child_tidptr, 0);
1813 }
1814 #endif
1815
1816 /*
1817  * Create a kernel thread.
1818  */
1819 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1820 {
1821         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1822                 (unsigned long)arg, NULL, NULL, 0);
1823 }
1824
1825 #ifdef __ARCH_WANT_SYS_FORK
1826 SYSCALL_DEFINE0(fork)
1827 {
1828 #ifdef CONFIG_MMU
1829         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1830 #else
1831         /* can not support in nommu mode */
1832         return -EINVAL;
1833 #endif
1834 }
1835 #endif
1836
1837 #ifdef __ARCH_WANT_SYS_VFORK
1838 SYSCALL_DEFINE0(vfork)
1839 {
1840         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1841                         0, NULL, NULL, 0);
1842 }
1843 #endif
1844
1845 #ifdef __ARCH_WANT_SYS_CLONE
1846 #ifdef CONFIG_CLONE_BACKWARDS
1847 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1848                  int __user *, parent_tidptr,
1849                  unsigned long, tls,
1850                  int __user *, child_tidptr)
1851 #elif defined(CONFIG_CLONE_BACKWARDS2)
1852 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1853                  int __user *, parent_tidptr,
1854                  int __user *, child_tidptr,
1855                  unsigned long, tls)
1856 #elif defined(CONFIG_CLONE_BACKWARDS3)
1857 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1858                 int, stack_size,
1859                 int __user *, parent_tidptr,
1860                 int __user *, child_tidptr,
1861                 unsigned long, tls)
1862 #else
1863 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1864                  int __user *, parent_tidptr,
1865                  int __user *, child_tidptr,
1866                  unsigned long, tls)
1867 #endif
1868 {
1869         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
1870 }
1871 #endif
1872
1873 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1874 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1875 #endif
1876
1877 static void sighand_ctor(void *data)
1878 {
1879         struct sighand_struct *sighand = data;
1880
1881         spin_lock_init(&sighand->siglock);
1882         init_waitqueue_head(&sighand->signalfd_wqh);
1883 }
1884
1885 void __init proc_caches_init(void)
1886 {
1887         sighand_cachep = kmem_cache_create("sighand_cache",
1888                         sizeof(struct sighand_struct), 0,
1889                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1890                         SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
1891         signal_cachep = kmem_cache_create("signal_cache",
1892                         sizeof(struct signal_struct), 0,
1893                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1894                         NULL);
1895         files_cachep = kmem_cache_create("files_cache",
1896                         sizeof(struct files_struct), 0,
1897                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1898                         NULL);
1899         fs_cachep = kmem_cache_create("fs_cache",
1900                         sizeof(struct fs_struct), 0,
1901                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1902                         NULL);
1903         /*
1904          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1905          * whole struct cpumask for the OFFSTACK case. We could change
1906          * this to *only* allocate as much of it as required by the
1907          * maximum number of CPU's we can ever have.  The cpumask_allocation
1908          * is at the end of the structure, exactly for that reason.
1909          */
1910         mm_cachep = kmem_cache_create("mm_struct",
1911                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1912                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1913                         NULL);
1914         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
1915         mmap_init();
1916         nsproxy_cache_init();
1917 }
1918
1919 /*
1920  * Check constraints on flags passed to the unshare system call.
1921  */
1922 static int check_unshare_flags(unsigned long unshare_flags)
1923 {
1924         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1925                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1926                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1927                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
1928                 return -EINVAL;
1929         /*
1930          * Not implemented, but pretend it works if there is nothing
1931          * to unshare.  Note that unsharing the address space or the
1932          * signal handlers also need to unshare the signal queues (aka
1933          * CLONE_THREAD).
1934          */
1935         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1936                 if (!thread_group_empty(current))
1937                         return -EINVAL;
1938         }
1939         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
1940                 if (atomic_read(&current->sighand->count) > 1)
1941                         return -EINVAL;
1942         }
1943         if (unshare_flags & CLONE_VM) {
1944                 if (!current_is_single_threaded())
1945                         return -EINVAL;
1946         }
1947
1948         return 0;
1949 }
1950
1951 /*
1952  * Unshare the filesystem structure if it is being shared
1953  */
1954 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1955 {
1956         struct fs_struct *fs = current->fs;
1957
1958         if (!(unshare_flags & CLONE_FS) || !fs)
1959                 return 0;
1960
1961         /* don't need lock here; in the worst case we'll do useless copy */
1962         if (fs->users == 1)
1963                 return 0;
1964
1965         *new_fsp = copy_fs_struct(fs);
1966         if (!*new_fsp)
1967                 return -ENOMEM;
1968
1969         return 0;
1970 }
1971
1972 /*
1973  * Unshare file descriptor table if it is being shared
1974  */
1975 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1976 {
1977         struct files_struct *fd = current->files;
1978         int error = 0;
1979
1980         if ((unshare_flags & CLONE_FILES) &&
1981             (fd && atomic_read(&fd->count) > 1)) {
1982                 *new_fdp = dup_fd(fd, &error);
1983                 if (!*new_fdp)
1984                         return error;
1985         }
1986
1987         return 0;
1988 }
1989
1990 /*
1991  * unshare allows a process to 'unshare' part of the process
1992  * context which was originally shared using clone.  copy_*
1993  * functions used by do_fork() cannot be used here directly
1994  * because they modify an inactive task_struct that is being
1995  * constructed. Here we are modifying the current, active,
1996  * task_struct.
1997  */
1998 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1999 {
2000         struct fs_struct *fs, *new_fs = NULL;
2001         struct files_struct *fd, *new_fd = NULL;
2002         struct cred *new_cred = NULL;
2003         struct nsproxy *new_nsproxy = NULL;
2004         int do_sysvsem = 0;
2005         int err;
2006
2007         /*
2008          * If unsharing a user namespace must also unshare the thread group
2009          * and unshare the filesystem root and working directories.
2010          */
2011         if (unshare_flags & CLONE_NEWUSER)
2012                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2013         /*
2014          * If unsharing vm, must also unshare signal handlers.
2015          */
2016         if (unshare_flags & CLONE_VM)
2017                 unshare_flags |= CLONE_SIGHAND;
2018         /*
2019          * If unsharing a signal handlers, must also unshare the signal queues.
2020          */
2021         if (unshare_flags & CLONE_SIGHAND)
2022                 unshare_flags |= CLONE_THREAD;
2023         /*
2024          * If unsharing namespace, must also unshare filesystem information.
2025          */
2026         if (unshare_flags & CLONE_NEWNS)
2027                 unshare_flags |= CLONE_FS;
2028
2029         err = check_unshare_flags(unshare_flags);
2030         if (err)
2031                 goto bad_unshare_out;
2032         /*
2033          * CLONE_NEWIPC must also detach from the undolist: after switching
2034          * to a new ipc namespace, the semaphore arrays from the old
2035          * namespace are unreachable.
2036          */
2037         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2038                 do_sysvsem = 1;
2039         err = unshare_fs(unshare_flags, &new_fs);
2040         if (err)
2041                 goto bad_unshare_out;
2042         err = unshare_fd(unshare_flags, &new_fd);
2043         if (err)
2044                 goto bad_unshare_cleanup_fs;
2045         err = unshare_userns(unshare_flags, &new_cred);
2046         if (err)
2047                 goto bad_unshare_cleanup_fd;
2048         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2049                                          new_cred, new_fs);
2050         if (err)
2051                 goto bad_unshare_cleanup_cred;
2052
2053         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2054                 if (do_sysvsem) {
2055                         /*
2056                          * CLONE_SYSVSEM is equivalent to sys_exit().
2057                          */
2058                         exit_sem(current);
2059                 }
2060                 if (unshare_flags & CLONE_NEWIPC) {
2061                         /* Orphan segments in old ns (see sem above). */
2062                         exit_shm(current);
2063                         shm_init_task(current);
2064                 }
2065
2066                 if (new_nsproxy)
2067                         switch_task_namespaces(current, new_nsproxy);
2068
2069                 task_lock(current);
2070
2071                 if (new_fs) {
2072                         fs = current->fs;
2073                         spin_lock(&fs->lock);
2074                         current->fs = new_fs;
2075                         if (--fs->users)
2076                                 new_fs = NULL;
2077                         else
2078                                 new_fs = fs;
2079                         spin_unlock(&fs->lock);
2080                 }
2081
2082                 if (new_fd) {
2083                         fd = current->files;
2084                         current->files = new_fd;
2085                         new_fd = fd;
2086                 }
2087
2088                 task_unlock(current);
2089
2090                 if (new_cred) {
2091                         /* Install the new user namespace */
2092                         commit_creds(new_cred);
2093                         new_cred = NULL;
2094                 }
2095         }
2096
2097 bad_unshare_cleanup_cred:
2098         if (new_cred)
2099                 put_cred(new_cred);
2100 bad_unshare_cleanup_fd:
2101         if (new_fd)
2102                 put_files_struct(new_fd);
2103
2104 bad_unshare_cleanup_fs:
2105         if (new_fs)
2106                 free_fs_struct(new_fs);
2107
2108 bad_unshare_out:
2109         return err;
2110 }
2111
2112 /*
2113  *      Helper to unshare the files of the current task.
2114  *      We don't want to expose copy_files internals to
2115  *      the exec layer of the kernel.
2116  */
2117
2118 int unshare_files(struct files_struct **displaced)
2119 {
2120         struct task_struct *task = current;
2121         struct files_struct *copy = NULL;
2122         int error;
2123
2124         error = unshare_fd(CLONE_FILES, &copy);
2125         if (error || !copy) {
2126                 *displaced = NULL;
2127                 return error;
2128         }
2129         *displaced = task->files;
2130         task_lock(task);
2131         task->files = copy;
2132         task_unlock(task);
2133         return 0;
2134 }
2135
2136 int sysctl_max_threads(struct ctl_table *table, int write,
2137                        void __user *buffer, size_t *lenp, loff_t *ppos)
2138 {
2139         struct ctl_table t;
2140         int ret;
2141         int threads = max_threads;
2142         int min = MIN_THREADS;
2143         int max = MAX_THREADS;
2144
2145         t = *table;
2146         t.data = &threads;
2147         t.extra1 = &min;
2148         t.extra2 = &max;
2149
2150         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2151         if (ret || !write)
2152                 return ret;
2153
2154         set_max_threads(threads);
2155
2156         return 0;
2157 }