kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/mm.h>
  32 #include <linux/vmacache.h>
  33 #include <linux/nsproxy.h>
  34 #include <linux/capability.h>
  35 #include <linux/cpu.h>
  36 #include <linux/cgroup.h>
  37 #include <linux/security.h>
  38 #include <linux/hugetlb.h>
  39 #include <linux/seccomp.h>
  40 #include <linux/swap.h>
  41 #include <linux/syscalls.h>
  42 #include <linux/jiffies.h>
  43 #include <linux/futex.h>
  44 #include <linux/compat.h>
  45 #include <linux/kthread.h>
  46 #include <linux/task_io_accounting_ops.h>
  47 #include <linux/rcupdate.h>
  48 #include <linux/ptrace.h>
  49 #include <linux/mount.h>
  50 #include <linux/audit.h>
  51 #include <linux/memcontrol.h>
  52 #include <linux/ftrace.h>
  53 #include <linux/proc_fs.h>
  54 #include <linux/profile.h>
  55 #include <linux/rmap.h>
  56 #include <linux/ksm.h>
  57 #include <linux/acct.h>
  58 #include <linux/tsacct_kern.h>
  59 #include <linux/cn_proc.h>
  60 #include <linux/freezer.h>
  61 #include <linux/delayacct.h>
  62 #include <linux/taskstats_kern.h>
  63 #include <linux/random.h>
  64 #include <linux/tty.h>
  65 #include <linux/blkdev.h>
  66 #include <linux/fs_struct.h>
  67 #include <linux/magic.h>
  68 #include <linux/perf_event.h>
  69 #include <linux/posix-timers.h>
  70 #include <linux/user-return-notifier.h>
  71 #include <linux/oom.h>
  72 #include <linux/khugepaged.h>
  73 #include <linux/signalfd.h>
  74 #include <linux/uprobes.h>
  75 #include <linux/aio.h>
  76 #include <linux/compiler.h>
  77 #include <linux/sysctl.h>
  78
  79 #include <asm/pgtable.h>
  80 #include <asm/pgalloc.h>
  81 #include <asm/uaccess.h>
  82 #include <asm/mmu_context.h>
  83 #include <asm/cacheflush.h>
  84 #include <asm/tlbflush.h>
  85
  86 #include <trace/events/sched.h>
  87
  88 #define CREATE_TRACE_POINTS
  89 #include <trace/events/task.h>
  90
  91 /*
  92  * Minimum number of threads to boot the kernel
  93  */
  94 #define MIN_THREADS 20
  95
  96 /*
  97  * Maximum number of threads
  98  */
  99 #define MAX_THREADS FUTEX_TID_MASK
 100
 101 /*
 102  * Protected counters by write_lock_irq(&tasklist_lock)
 103  */
 104 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 105 int nr_threads;                 /* The idle threads do not count.. */
 106
 107 int max_threads;                /* tunable limit on nr_threads */
 108
 109 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 110
 111 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 112
 113 #ifdef CONFIG_PROVE_RCU
 114 int lockdep_tasklist_lock_is_held(void)
 115 {
 116         return lockdep_is_held(&tasklist_lock);
 117 }
 118 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 119 #endif /* #ifdef CONFIG_PROVE_RCU */
 120
 121 int nr_processes(void)
 122 {
 123         int cpu;
 124         int total = 0;
 125
 126         for_each_possible_cpu(cpu)
 127                 total += per_cpu(process_counts, cpu);
 128
 129         return total;
 130 }
 131
 132 void __weak arch_release_task_struct(struct task_struct *tsk)
 133 {
 134 }
 135
 136 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 137 static struct kmem_cache *task_struct_cachep;
 138
 139 static inline struct task_struct *alloc_task_struct_node(int node)
 140 {
 141         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 142 }
 143
 144 static inline void free_task_struct(struct task_struct *tsk)
 145 {
 146         kmem_cache_free(task_struct_cachep, tsk);
 147 }
 148 #endif
 149
 150 void __weak arch_release_thread_info(struct thread_info *ti)
 151 {
 152 }
 153
 154 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
 155
 156 /*
 157  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 158  * kmemcache based allocator.
 159  */
 160 # if THREAD_SIZE >= PAGE_SIZE
 161 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 162                                                   int node)
 163 {
 164         struct page *page = alloc_kmem_pages_node(node, THREADINFO_GFP,
 165                                                   THREAD_SIZE_ORDER);
 166
 167         return page ? page_address(page) : NULL;
 168 }
 169
 170 static inline void free_thread_info(struct thread_info *ti)
 171 {
 172         free_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 173 }
 174 # else
 175 static struct kmem_cache *thread_info_cache;
 176
 177 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 178                                                   int node)
 179 {
 180         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
 181 }
 182
 183 static void free_thread_info(struct thread_info *ti)
 184 {
 185         kmem_cache_free(thread_info_cache, ti);
 186 }
 187
 188 void thread_info_cache_init(void)
 189 {
 190         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
 191                                               THREAD_SIZE, 0, NULL);
 192         BUG_ON(thread_info_cache == NULL);
 193 }
 194 # endif
 195 #endif
 196
 197 /* SLAB cache for signal_struct structures (tsk->signal) */
 198 static struct kmem_cache *signal_cachep;
 199
 200 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 201 struct kmem_cache *sighand_cachep;
 202
 203 /* SLAB cache for files_struct structures (tsk->files) */
 204 struct kmem_cache *files_cachep;
 205
 206 /* SLAB cache for fs_struct structures (tsk->fs) */
 207 struct kmem_cache *fs_cachep;
 208
 209 /* SLAB cache for vm_area_struct structures */
 210 struct kmem_cache *vm_area_cachep;
 211
 212 /* SLAB cache for mm_struct structures (tsk->mm) */
 213 static struct kmem_cache *mm_cachep;
 214
 215 static void account_kernel_stack(struct thread_info *ti, int account)
 216 {
 217         struct zone *zone = page_zone(virt_to_page(ti));
 218
 219         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 220 }
 221
 222 void free_task(struct task_struct *tsk)
 223 {
 224         account_kernel_stack(tsk->stack, -1);
 225         arch_release_thread_info(tsk->stack);
 226         free_thread_info(tsk->stack);
 227         rt_mutex_debug_task_free(tsk);
 228         ftrace_graph_exit_task(tsk);
 229         put_seccomp_filter(tsk);
 230         arch_release_task_struct(tsk);
 231         free_task_struct(tsk);
 232 }
 233 EXPORT_SYMBOL(free_task);
 234
 235 static inline void free_signal_struct(struct signal_struct *sig)
 236 {
 237         taskstats_tgid_free(sig);
 238         sched_autogroup_exit(sig);
 239         kmem_cache_free(signal_cachep, sig);
 240 }
 241
 242 static inline void put_signal_struct(struct signal_struct *sig)
 243 {
 244         if (atomic_dec_and_test(&sig->sigcnt))
 245                 free_signal_struct(sig);
 246 }
 247
 248 void __put_task_struct(struct task_struct *tsk)
 249 {
 250         WARN_ON(!tsk->exit_state);
 251         WARN_ON(atomic_read(&tsk->usage));
 252         WARN_ON(tsk == current);
 253
 254         cgroup_free(tsk);
 255         task_numa_free(tsk);
 256         security_task_free(tsk);
 257         exit_creds(tsk);
 258         delayacct_tsk_free(tsk);
 259         put_signal_struct(tsk->signal);
 260
 261         if (!profile_handoff_task(tsk))
 262                 free_task(tsk);
 263 }
 264 EXPORT_SYMBOL_GPL(__put_task_struct);
 265
 266 void __init __weak arch_task_cache_init(void) { }
 267
 268 /*
 269  * set_max_threads
 270  */
 271 static void set_max_threads(unsigned int max_threads_suggested)
 272 {
 273         u64 threads;
 274
 275         /*
 276          * The number of threads shall be limited such that the thread
 277          * structures may only consume a small part of the available memory.
 278          */
 279         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 280                 threads = MAX_THREADS;
 281         else
 282                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 283                                     (u64) THREAD_SIZE * 8UL);
 284
 285         if (threads > max_threads_suggested)
 286                 threads = max_threads_suggested;
 287
 288         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 289 }
 290
 291 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 292 /* Initialized by the architecture: */
 293 int arch_task_struct_size __read_mostly;
 294 #endif
 295
 296 void __init fork_init(void)
 297 {
 298 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 299 #ifndef ARCH_MIN_TASKALIGN
 300 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 301 #endif
 302         /* create a slab on which task_structs can be allocated */
 303         task_struct_cachep =
 304                 kmem_cache_create("task_struct", arch_task_struct_size,
 305                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 306 #endif
 307
 308         /* do the arch specific task caches init */
 309         arch_task_cache_init();
 310
 311         set_max_threads(MAX_THREADS);
 312
 313         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 314         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 315         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 316                 init_task.signal->rlim[RLIMIT_NPROC];
 317 }
 318
 319 int __weak arch_dup_task_struct(struct task_struct *dst,
 320                                                struct task_struct *src)
 321 {
 322         *dst = *src;
 323         return 0;
 324 }
 325
 326 void set_task_stack_end_magic(struct task_struct *tsk)
 327 {
 328         unsigned long *stackend;
 329
 330         stackend = end_of_stack(tsk);
 331         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 332 }
 333
 334 static struct task_struct *dup_task_struct(struct task_struct *orig)
 335 {
 336         struct task_struct *tsk;
 337         struct thread_info *ti;
 338         int node = tsk_fork_get_node(orig);
 339         int err;
 340
 341         tsk = alloc_task_struct_node(node);
 342         if (!tsk)
 343                 return NULL;
 344
 345         ti = alloc_thread_info_node(tsk, node);
 346         if (!ti)
 347                 goto free_tsk;
 348
 349         err = arch_dup_task_struct(tsk, orig);
 350         if (err)
 351                 goto free_ti;
 352
 353         tsk->stack = ti;
 354 #ifdef CONFIG_SECCOMP
 355         /*
 356          * We must handle setting up seccomp filters once we're under
 357          * the sighand lock in case orig has changed between now and
 358          * then. Until then, filter must be NULL to avoid messing up
 359          * the usage counts on the error path calling free_task.
 360          */
 361         tsk->seccomp.filter = NULL;
 362 #endif
 363
 364         setup_thread_stack(tsk, orig);
 365         clear_user_return_notifier(tsk);
 366         clear_tsk_need_resched(tsk);
 367         set_task_stack_end_magic(tsk);
 368
 369 #ifdef CONFIG_CC_STACKPROTECTOR
 370         tsk->stack_canary = get_random_int();
 371 #endif
 372
 373         /*
 374          * One for us, one for whoever does the "release_task()" (usually
 375          * parent)
 376          */
 377         atomic_set(&tsk->usage, 2);
 378 #ifdef CONFIG_BLK_DEV_IO_TRACE
 379         tsk->btrace_seq = 0;
 380 #endif
 381         tsk->splice_pipe = NULL;
 382         tsk->task_frag.page = NULL;
 383         tsk->wake_q.next = NULL;
 384
 385         account_kernel_stack(ti, 1);
 386
 387         return tsk;
 388
 389 free_ti:
 390         free_thread_info(ti);
 391 free_tsk:
 392         free_task_struct(tsk);
 393         return NULL;
 394 }
 395
 396 #ifdef CONFIG_MMU
 397 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 398 {
 399         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 400         struct rb_node **rb_link, *rb_parent;
 401         int retval;
 402         unsigned long charge;
 403
 404         uprobe_start_dup_mmap();
 405         down_write(&oldmm->mmap_sem);
 406         flush_cache_dup_mm(oldmm);
 407         uprobe_dup_mmap(oldmm, mm);
 408         /*
 409          * Not linked in yet - no deadlock potential:
 410          */
 411         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 412
 413         /* No ordering required: file already has been exposed. */
 414         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 415
 416         mm->total_vm = oldmm->total_vm;
 417         mm->shared_vm = oldmm->shared_vm;
 418         mm->exec_vm = oldmm->exec_vm;
 419         mm->stack_vm = oldmm->stack_vm;
 420
 421         rb_link = &mm->mm_rb.rb_node;
 422         rb_parent = NULL;
 423         pprev = &mm->mmap;
 424         retval = ksm_fork(mm, oldmm);
 425         if (retval)
 426                 goto out;
 427         retval = khugepaged_fork(mm, oldmm);
 428         if (retval)
 429                 goto out;
 430
 431         prev = NULL;
 432         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 433                 struct file *file;
 434
 435                 if (mpnt->vm_flags & VM_DONTCOPY) {
 436                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 437                                                         -vma_pages(mpnt));
 438                         continue;
 439                 }
 440                 charge = 0;
 441                 if (mpnt->vm_flags & VM_ACCOUNT) {
 442                         unsigned long len = vma_pages(mpnt);
 443
 444                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 445                                 goto fail_nomem;
 446                         charge = len;
 447                 }
 448                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 449                 if (!tmp)
 450                         goto fail_nomem;
 451                 *tmp = *mpnt;
 452                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 453                 retval = vma_dup_policy(mpnt, tmp);
 454                 if (retval)
 455                         goto fail_nomem_policy;
 456                 tmp->vm_mm = mm;
 457                 if (anon_vma_fork(tmp, mpnt))
 458                         goto fail_nomem_anon_vma_fork;
 459                 tmp->vm_flags &=
 460                         ~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
 461                 tmp->vm_next = tmp->vm_prev = NULL;
 462                 tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
 463                 file = tmp->vm_file;
 464                 if (file) {
 465                         struct inode *inode = file_inode(file);
 466                         struct address_space *mapping = file->f_mapping;
 467
 468                         get_file(file);
 469                         if (tmp->vm_flags & VM_DENYWRITE)
 470                                 atomic_dec(&inode->i_writecount);
 471                         i_mmap_lock_write(mapping);
 472                         if (tmp->vm_flags & VM_SHARED)
 473                                 atomic_inc(&mapping->i_mmap_writable);
 474                         flush_dcache_mmap_lock(mapping);
 475                         /* insert tmp into the share list, just after mpnt */
 476                         vma_interval_tree_insert_after(tmp, mpnt,
 477                                         &mapping->i_mmap);
 478                         flush_dcache_mmap_unlock(mapping);
 479                         i_mmap_unlock_write(mapping);
 480                 }
 481
 482                 /*
 483                  * Clear hugetlb-related page reserves for children. This only
 484                  * affects MAP_PRIVATE mappings. Faults generated by the child
 485                  * are not guaranteed to succeed, even if read-only
 486                  */
 487                 if (is_vm_hugetlb_page(tmp))
 488                         reset_vma_resv_huge_pages(tmp);
 489
 490                 /*
 491                  * Link in the new vma and copy the page table entries.
 492                  */
 493                 *pprev = tmp;
 494                 pprev = &tmp->vm_next;
 495                 tmp->vm_prev = prev;
 496                 prev = tmp;
 497
 498                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 499                 rb_link = &tmp->vm_rb.rb_right;
 500                 rb_parent = &tmp->vm_rb;
 501
 502                 mm->map_count++;
 503                 retval = copy_page_range(mm, oldmm, mpnt);
 504
 505                 if (tmp->vm_ops && tmp->vm_ops->open)
 506                         tmp->vm_ops->open(tmp);
 507
 508                 if (retval)
 509                         goto out;
 510         }
 511         /* a new mm has just been created */
 512         arch_dup_mmap(oldmm, mm);
 513         retval = 0;
 514 out:
 515         up_write(&mm->mmap_sem);
 516         flush_tlb_mm(oldmm);
 517         up_write(&oldmm->mmap_sem);
 518         uprobe_end_dup_mmap();
 519         return retval;
 520 fail_nomem_anon_vma_fork:
 521         mpol_put(vma_policy(tmp));
 522 fail_nomem_policy:
 523         kmem_cache_free(vm_area_cachep, tmp);
 524 fail_nomem:
 525         retval = -ENOMEM;
 526         vm_unacct_memory(charge);
 527         goto out;
 528 }
 529
 530 static inline int mm_alloc_pgd(struct mm_struct *mm)
 531 {
 532         mm->pgd = pgd_alloc(mm);
 533         if (unlikely(!mm->pgd))
 534                 return -ENOMEM;
 535         return 0;
 536 }
 537
 538 static inline void mm_free_pgd(struct mm_struct *mm)
 539 {
 540         pgd_free(mm, mm->pgd);
 541 }
 542 #else
 543 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 544 {
 545         down_write(&oldmm->mmap_sem);
 546         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 547         up_write(&oldmm->mmap_sem);
 548         return 0;
 549 }
 550 #define mm_alloc_pgd(mm)        (0)
 551 #define mm_free_pgd(mm)
 552 #endif /* CONFIG_MMU */
 553
 554 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 555
 556 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 557 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 558
 559 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 560
 561 static int __init coredump_filter_setup(char *s)
 562 {
 563         default_dump_filter =
 564                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 565                 MMF_DUMP_FILTER_MASK;
 566         return 1;
 567 }
 568
 569 __setup("coredump_filter=", coredump_filter_setup);
 570
 571 #include <linux/init_task.h>
 572
 573 static void mm_init_aio(struct mm_struct *mm)
 574 {
 575 #ifdef CONFIG_AIO
 576         spin_lock_init(&mm->ioctx_lock);
 577         mm->ioctx_table = NULL;
 578 #endif
 579 }
 580
 581 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 582 {
 583 #ifdef CONFIG_MEMCG
 584         mm->owner = p;
 585 #endif
 586 }
 587
 588 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 589 {
 590         mm->mmap = NULL;
 591         mm->mm_rb = RB_ROOT;
 592         mm->vmacache_seqnum = 0;
 593         atomic_set(&mm->mm_users, 1);
 594         atomic_set(&mm->mm_count, 1);
 595         init_rwsem(&mm->mmap_sem);
 596         INIT_LIST_HEAD(&mm->mmlist);
 597         mm->core_state = NULL;
 598         atomic_long_set(&mm->nr_ptes, 0);
 599         mm_nr_pmds_init(mm);
 600         mm->map_count = 0;
 601         mm->locked_vm = 0;
 602         mm->pinned_vm = 0;
 603         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 604         spin_lock_init(&mm->page_table_lock);
 605         mm_init_cpumask(mm);
 606         mm_init_aio(mm);
 607         mm_init_owner(mm, p);
 608         mmu_notifier_mm_init(mm);
 609         clear_tlb_flush_pending(mm);
 610 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 611         mm->pmd_huge_pte = NULL;
 612 #endif
 613
 614         if (current->mm) {
 615                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 616                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 617         } else {
 618                 mm->flags = default_dump_filter;
 619                 mm->def_flags = 0;
 620         }
 621
 622         if (mm_alloc_pgd(mm))
 623                 goto fail_nopgd;
 624
 625         if (init_new_context(p, mm))
 626                 goto fail_nocontext;
 627
 628         return mm;
 629
 630 fail_nocontext:
 631         mm_free_pgd(mm);
 632 fail_nopgd:
 633         free_mm(mm);
 634         return NULL;
 635 }
 636
 637 static void check_mm(struct mm_struct *mm)
 638 {
 639         int i;
 640
 641         for (i = 0; i < NR_MM_COUNTERS; i++) {
 642                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 643
 644                 if (unlikely(x))
 645                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 646                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 647         }
 648
 649         if (atomic_long_read(&mm->nr_ptes))
 650                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 651                                 atomic_long_read(&mm->nr_ptes));
 652         if (mm_nr_pmds(mm))
 653                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 654                                 mm_nr_pmds(mm));
 655
 656 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 657         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 658 #endif
 659 }
 660
 661 /*
 662  * Allocate and initialize an mm_struct.
 663  */
 664 struct mm_struct *mm_alloc(void)
 665 {
 666         struct mm_struct *mm;
 667
 668         mm = allocate_mm();
 669         if (!mm)
 670                 return NULL;
 671
 672         memset(mm, 0, sizeof(*mm));
 673         return mm_init(mm, current);
 674 }
 675
 676 /*
 677  * Called when the last reference to the mm
 678  * is dropped: either by a lazy thread or by
 679  * mmput. Free the page directory and the mm.
 680  */
 681 void __mmdrop(struct mm_struct *mm)
 682 {
 683         BUG_ON(mm == &init_mm);
 684         mm_free_pgd(mm);
 685         destroy_context(mm);
 686         mmu_notifier_mm_destroy(mm);
 687         check_mm(mm);
 688         free_mm(mm);
 689 }
 690 EXPORT_SYMBOL_GPL(__mmdrop);
 691
 692 /*
 693  * Decrement the use count and release all resources for an mm.
 694  */
 695 void mmput(struct mm_struct *mm)
 696 {
 697         might_sleep();
 698
 699         if (atomic_dec_and_test(&mm->mm_users)) {
 700                 uprobe_clear_state(mm);
 701                 exit_aio(mm);
 702                 ksm_exit(mm);
 703                 khugepaged_exit(mm); /* must run before exit_mmap */
 704                 exit_mmap(mm);
 705                 set_mm_exe_file(mm, NULL);
 706                 if (!list_empty(&mm->mmlist)) {
 707                         spin_lock(&mmlist_lock);
 708                         list_del(&mm->mmlist);
 709                         spin_unlock(&mmlist_lock);
 710                 }
 711                 if (mm->binfmt)
 712                         module_put(mm->binfmt->module);
 713                 mmdrop(mm);
 714         }
 715 }
 716 EXPORT_SYMBOL_GPL(mmput);
 717
 718 /**
 719  * set_mm_exe_file - change a reference to the mm's executable file
 720  *
 721  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 722  *
 723  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 724  * invocations: in mmput() nobody alive left, in execve task is single
 725  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 726  * mm->exe_file, but does so without using set_mm_exe_file() in order
 727  * to do avoid the need for any locks.
 728  */
 729 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 730 {
 731         struct file *old_exe_file;
 732
 733         /*
 734          * It is safe to dereference the exe_file without RCU as
 735          * this function is only called if nobody else can access
 736          * this mm -- see comment above for justification.
 737          */
 738         old_exe_file = rcu_dereference_raw(mm->exe_file);
 739
 740         if (new_exe_file)
 741                 get_file(new_exe_file);
 742         rcu_assign_pointer(mm->exe_file, new_exe_file);
 743         if (old_exe_file)
 744                 fput(old_exe_file);
 745 }
 746
 747 /**
 748  * get_mm_exe_file - acquire a reference to the mm's executable file
 749  *
 750  * Returns %NULL if mm has no associated executable file.
 751  * User must release file via fput().
 752  */
 753 struct file *get_mm_exe_file(struct mm_struct *mm)
 754 {
 755         struct file *exe_file;
 756
 757         rcu_read_lock();
 758         exe_file = rcu_dereference(mm->exe_file);
 759         if (exe_file && !get_file_rcu(exe_file))
 760                 exe_file = NULL;
 761         rcu_read_unlock();
 762         return exe_file;
 763 }
 764 EXPORT_SYMBOL(get_mm_exe_file);
 765
 766 /**
 767  * get_task_exe_file - acquire a reference to the task's executable file
 768  *
 769  * Returns %NULL if task's mm (if any) has no associated executable file or
 770  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
 771  * User must release file via fput().
 772  */
 773 struct file *get_task_exe_file(struct task_struct *task)
 774 {
 775         struct file *exe_file = NULL;
 776         struct mm_struct *mm;
 777
 778         task_lock(task);
 779         mm = task->mm;
 780         if (mm) {
 781                 if (!(task->flags & PF_KTHREAD))
 782                         exe_file = get_mm_exe_file(mm);
 783         }
 784         task_unlock(task);
 785         return exe_file;
 786 }
 787 EXPORT_SYMBOL(get_task_exe_file);
 788
 789 /**
 790  * get_task_mm - acquire a reference to the task's mm
 791  *
 792  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 793  * this kernel workthread has transiently adopted a user mm with use_mm,
 794  * to do its AIO) is not set and if so returns a reference to it, after
 795  * bumping up the use count.  User must release the mm via mmput()
 796  * after use.  Typically used by /proc and ptrace.
 797  */
 798 struct mm_struct *get_task_mm(struct task_struct *task)
 799 {
 800         struct mm_struct *mm;
 801
 802         task_lock(task);
 803         mm = task->mm;
 804         if (mm) {
 805                 if (task->flags & PF_KTHREAD)
 806                         mm = NULL;
 807                 else
 808                         atomic_inc(&mm->mm_users);
 809         }
 810         task_unlock(task);
 811         return mm;
 812 }
 813 EXPORT_SYMBOL_GPL(get_task_mm);
 814
 815 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 816 {
 817         struct mm_struct *mm;
 818         int err;
 819
 820         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 821         if (err)
 822                 return ERR_PTR(err);
 823
 824         mm = get_task_mm(task);
 825         if (mm && mm != current->mm &&
 826                         !ptrace_may_access(task, mode)) {
 827                 mmput(mm);
 828                 mm = ERR_PTR(-EACCES);
 829         }
 830         mutex_unlock(&task->signal->cred_guard_mutex);
 831
 832         return mm;
 833 }
 834
 835 static void complete_vfork_done(struct task_struct *tsk)
 836 {
 837         struct completion *vfork;
 838
 839         task_lock(tsk);
 840         vfork = tsk->vfork_done;
 841         if (likely(vfork)) {
 842                 tsk->vfork_done = NULL;
 843                 complete(vfork);
 844         }
 845         task_unlock(tsk);
 846 }
 847
 848 static int wait_for_vfork_done(struct task_struct *child,
 849                                 struct completion *vfork)
 850 {
 851         int killed;
 852
 853         freezer_do_not_count();
 854         killed = wait_for_completion_killable(vfork);
 855         freezer_count();
 856
 857         if (killed) {
 858                 task_lock(child);
 859                 child->vfork_done = NULL;
 860                 task_unlock(child);
 861         }
 862
 863         put_task_struct(child);
 864         return killed;
 865 }
 866
 867 /* Please note the differences between mmput and mm_release.
 868  * mmput is called whenever we stop holding onto a mm_struct,
 869  * error success whatever.
 870  *
 871  * mm_release is called after a mm_struct has been removed
 872  * from the current process.
 873  *
 874  * This difference is important for error handling, when we
 875  * only half set up a mm_struct for a new process and need to restore
 876  * the old one.  Because we mmput the new mm_struct before
 877  * restoring the old one. . .
 878  * Eric Biederman 10 January 1998
 879  */
 880 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 881 {
 882         /* Get rid of any futexes when releasing the mm */
 883 #ifdef CONFIG_FUTEX
 884         if (unlikely(tsk->robust_list)) {
 885                 exit_robust_list(tsk);
 886                 tsk->robust_list = NULL;
 887         }
 888 #ifdef CONFIG_COMPAT
 889         if (unlikely(tsk->compat_robust_list)) {
 890                 compat_exit_robust_list(tsk);
 891                 tsk->compat_robust_list = NULL;
 892         }
 893 #endif
 894         if (unlikely(!list_empty(&tsk->pi_state_list)))
 895                 exit_pi_state_list(tsk);
 896 #endif
 897
 898         uprobe_free_utask(tsk);
 899
 900         /* Get rid of any cached register state */
 901         deactivate_mm(tsk, mm);
 902
 903         /*
 904          * Signal userspace if we're not exiting with a core dump
 905          * because we want to leave the value intact for debugging
 906          * purposes.
 907          */
 908         if (tsk->clear_child_tid) {
 909                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
 910                     atomic_read(&mm->mm_users) > 1) {
 911                         /*
 912                          * We don't check the error code - if userspace has
 913                          * not set up a proper pointer then tough luck.
 914                          */
 915                         put_user(0, tsk->clear_child_tid);
 916                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 917                                         1, NULL, NULL, 0);
 918                 }
 919                 tsk->clear_child_tid = NULL;
 920         }
 921
 922         /*
 923          * All done, finally we can wake up parent and return this mm to him.
 924          * Also kthread_stop() uses this completion for synchronization.
 925          */
 926         if (tsk->vfork_done)
 927                 complete_vfork_done(tsk);
 928 }
 929
 930 /*
 931  * Allocate a new mm structure and copy contents from the
 932  * mm structure of the passed in task structure.
 933  */
 934 static struct mm_struct *dup_mm(struct task_struct *tsk)
 935 {
 936         struct mm_struct *mm, *oldmm = current->mm;
 937         int err;
 938
 939         mm = allocate_mm();
 940         if (!mm)
 941                 goto fail_nomem;
 942
 943         memcpy(mm, oldmm, sizeof(*mm));
 944
 945         if (!mm_init(mm, tsk))
 946                 goto fail_nomem;
 947
 948         err = dup_mmap(mm, oldmm);
 949         if (err)
 950                 goto free_pt;
 951
 952         mm->hiwater_rss = get_mm_rss(mm);
 953         mm->hiwater_vm = mm->total_vm;
 954
 955         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 956                 goto free_pt;
 957
 958         return mm;
 959
 960 free_pt:
 961         /* don't put binfmt in mmput, we haven't got module yet */
 962         mm->binfmt = NULL;
 963         mmput(mm);
 964
 965 fail_nomem:
 966         return NULL;
 967 }
 968
 969 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 970 {
 971         struct mm_struct *mm, *oldmm;
 972         int retval;
 973
 974         tsk->min_flt = tsk->maj_flt = 0;
 975         tsk->nvcsw = tsk->nivcsw = 0;
 976 #ifdef CONFIG_DETECT_HUNG_TASK
 977         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 978 #endif
 979
 980         tsk->mm = NULL;
 981         tsk->active_mm = NULL;
 982
 983         /*
 984          * Are we cloning a kernel thread?
 985          *
 986          * We need to steal a active VM for that..
 987          */
 988         oldmm = current->mm;
 989         if (!oldmm)
 990                 return 0;
 991
 992         /* initialize the new vmacache entries */
 993         vmacache_flush(tsk);
 994
 995         if (clone_flags & CLONE_VM) {
 996                 atomic_inc(&oldmm->mm_users);
 997                 mm = oldmm;
 998                 goto good_mm;
 999         }
1000
1001         retval = -ENOMEM;
1002         mm = dup_mm(tsk);
1003         if (!mm)
1004                 goto fail_nomem;
1005
1006 good_mm:
1007         tsk->mm = mm;
1008         tsk->active_mm = mm;
1009         return 0;
1010
1011 fail_nomem:
1012         return retval;
1013 }
1014
1015 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1016 {
1017         struct fs_struct *fs = current->fs;
1018         if (clone_flags & CLONE_FS) {
1019                 /* tsk->fs is already what we want */
1020                 spin_lock(&fs->lock);
1021                 if (fs->in_exec) {
1022                         spin_unlock(&fs->lock);
1023                         return -EAGAIN;
1024                 }
1025                 fs->users++;
1026                 spin_unlock(&fs->lock);
1027                 return 0;
1028         }
1029         tsk->fs = copy_fs_struct(fs);
1030         if (!tsk->fs)
1031                 return -ENOMEM;
1032         return 0;
1033 }
1034
1035 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1036 {
1037         struct files_struct *oldf, *newf;
1038         int error = 0;
1039
1040         /*
1041          * A background process may not have any files ...
1042          */
1043         oldf = current->files;
1044         if (!oldf)
1045                 goto out;
1046
1047         if (clone_flags & CLONE_FILES) {
1048                 atomic_inc(&oldf->count);
1049                 goto out;
1050         }
1051
1052         newf = dup_fd(oldf, &error);
1053         if (!newf)
1054                 goto out;
1055
1056         tsk->files = newf;
1057         error = 0;
1058 out:
1059         return error;
1060 }
1061
1062 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1063 {
1064 #ifdef CONFIG_BLOCK
1065         struct io_context *ioc = current->io_context;
1066         struct io_context *new_ioc;
1067
1068         if (!ioc)
1069                 return 0;
1070         /*
1071          * Share io context with parent, if CLONE_IO is set
1072          */
1073         if (clone_flags & CLONE_IO) {
1074                 ioc_task_link(ioc);
1075                 tsk->io_context = ioc;
1076         } else if (ioprio_valid(ioc->ioprio)) {
1077                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1078                 if (unlikely(!new_ioc))
1079                         return -ENOMEM;
1080
1081                 new_ioc->ioprio = ioc->ioprio;
1082                 put_io_context(new_ioc);
1083         }
1084 #endif
1085         return 0;
1086 }
1087
1088 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1089 {
1090         struct sighand_struct *sig;
1091
1092         if (clone_flags & CLONE_SIGHAND) {
1093                 atomic_inc(&current->sighand->count);
1094                 return 0;
1095         }
1096         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1097         rcu_assign_pointer(tsk->sighand, sig);
1098         if (!sig)
1099                 return -ENOMEM;
1100
1101         atomic_set(&sig->count, 1);
1102         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1103         return 0;
1104 }
1105
1106 void __cleanup_sighand(struct sighand_struct *sighand)
1107 {
1108         if (atomic_dec_and_test(&sighand->count)) {
1109                 signalfd_cleanup(sighand);
1110                 /*
1111                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1112                  * without an RCU grace period, see __lock_task_sighand().
1113                  */
1114                 kmem_cache_free(sighand_cachep, sighand);
1115         }
1116 }
1117
1118 /*
1119  * Initialize POSIX timer handling for a thread group.
1120  */
1121 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1122 {
1123         unsigned long cpu_limit;
1124
1125         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1126         if (cpu_limit != RLIM_INFINITY) {
1127                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1128                 sig->cputimer.running = true;
1129         }
1130
1131         /* The timer lists. */
1132         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1133         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1134         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1135 }
1136
1137 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1138 {
1139         struct signal_struct *sig;
1140
1141         if (clone_flags & CLONE_THREAD)
1142                 return 0;
1143
1144         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1145         tsk->signal = sig;
1146         if (!sig)
1147                 return -ENOMEM;
1148
1149         sig->nr_threads = 1;
1150         atomic_set(&sig->live, 1);
1151         atomic_set(&sig->sigcnt, 1);
1152
1153         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1154         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1155         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1156
1157         init_waitqueue_head(&sig->wait_chldexit);
1158         sig->curr_target = tsk;
1159         init_sigpending(&sig->shared_pending);
1160         INIT_LIST_HEAD(&sig->posix_timers);
1161         seqlock_init(&sig->stats_lock);
1162         prev_cputime_init(&sig->prev_cputime);
1163
1164         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1165         sig->real_timer.function = it_real_fn;
1166
1167         task_lock(current->group_leader);
1168         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1169         task_unlock(current->group_leader);
1170
1171         posix_cpu_timers_init_group(sig);
1172
1173         tty_audit_fork(sig);
1174         sched_autogroup_fork(sig);
1175
1176         sig->oom_score_adj = current->signal->oom_score_adj;
1177         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1178
1179         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1180                                    current->signal->is_child_subreaper;
1181
1182         mutex_init(&sig->cred_guard_mutex);
1183
1184         return 0;
1185 }
1186
1187 static void copy_seccomp(struct task_struct *p)
1188 {
1189 #ifdef CONFIG_SECCOMP
1190         /*
1191          * Must be called with sighand->lock held, which is common to
1192          * all threads in the group. Holding cred_guard_mutex is not
1193          * needed because this new task is not yet running and cannot
1194          * be racing exec.
1195          */
1196         assert_spin_locked(&current->sighand->siglock);
1197
1198         /* Ref-count the new filter user, and assign it. */
1199         get_seccomp_filter(current);
1200         p->seccomp = current->seccomp;
1201
1202         /*
1203          * Explicitly enable no_new_privs here in case it got set
1204          * between the task_struct being duplicated and holding the
1205          * sighand lock. The seccomp state and nnp must be in sync.
1206          */
1207         if (task_no_new_privs(current))
1208                 task_set_no_new_privs(p);
1209
1210         /*
1211          * If the parent gained a seccomp mode after copying thread
1212          * flags and between before we held the sighand lock, we have
1213          * to manually enable the seccomp thread flag here.
1214          */
1215         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1216                 set_tsk_thread_flag(p, TIF_SECCOMP);
1217 #endif
1218 }
1219
1220 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1221 {
1222         current->clear_child_tid = tidptr;
1223
1224         return task_pid_vnr(current);
1225 }
1226
1227 static void rt_mutex_init_task(struct task_struct *p)
1228 {
1229         raw_spin_lock_init(&p->pi_lock);
1230 #ifdef CONFIG_RT_MUTEXES
1231         p->pi_waiters = RB_ROOT;
1232         p->pi_waiters_leftmost = NULL;
1233         p->pi_blocked_on = NULL;
1234 #endif
1235 }
1236
1237 /*
1238  * Initialize POSIX timer handling for a single task.
1239  */
1240 static void posix_cpu_timers_init(struct task_struct *tsk)
1241 {
1242         tsk->cputime_expires.prof_exp = 0;
1243         tsk->cputime_expires.virt_exp = 0;
1244         tsk->cputime_expires.sched_exp = 0;
1245         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1246         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1247         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1248 }
1249
1250 static inline void
1251 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1252 {
1253          task->pids[type].pid = pid;
1254 }
1255
1256 /*
1257  * This creates a new process as a copy of the old one,
1258  * but does not actually start it yet.
1259  *
1260  * It copies the registers, and all the appropriate
1261  * parts of the process environment (as per the clone
1262  * flags). The actual kick-off is left to the caller.
1263  */
1264 static struct task_struct *copy_process(unsigned long clone_flags,
1265                                         unsigned long stack_start,
1266                                         unsigned long stack_size,
1267                                         int __user *child_tidptr,
1268                                         struct pid *pid,
1269                                         int trace,
1270                                         unsigned long tls)
1271 {
1272         int retval;
1273         struct task_struct *p;
1274         void *cgrp_ss_priv[CGROUP_CANFORK_COUNT] = {};
1275
1276         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1277                 return ERR_PTR(-EINVAL);
1278
1279         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1280                 return ERR_PTR(-EINVAL);
1281
1282         /*
1283          * Thread groups must share signals as well, and detached threads
1284          * can only be started up within the thread group.
1285          */
1286         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1287                 return ERR_PTR(-EINVAL);
1288
1289         /*
1290          * Shared signal handlers imply shared VM. By way of the above,
1291          * thread groups also imply shared VM. Blocking this case allows
1292          * for various simplifications in other code.
1293          */
1294         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1295                 return ERR_PTR(-EINVAL);
1296
1297         /*
1298          * Siblings of global init remain as zombies on exit since they are
1299          * not reaped by their parent (swapper). To solve this and to avoid
1300          * multi-rooted process trees, prevent global and container-inits
1301          * from creating siblings.
1302          */
1303         if ((clone_flags & CLONE_PARENT) &&
1304                                 current->signal->flags & SIGNAL_UNKILLABLE)
1305                 return ERR_PTR(-EINVAL);
1306
1307         /*
1308          * If the new process will be in a different pid or user namespace
1309          * do not allow it to share a thread group with the forking task.
1310          */
1311         if (clone_flags & CLONE_THREAD) {
1312                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1313                     (task_active_pid_ns(current) !=
1314                                 current->nsproxy->pid_ns_for_children))
1315                         return ERR_PTR(-EINVAL);
1316         }
1317
1318         retval = security_task_create(clone_flags);
1319         if (retval)
1320                 goto fork_out;
1321
1322         retval = -ENOMEM;
1323         p = dup_task_struct(current);
1324         if (!p)
1325                 goto fork_out;
1326
1327         ftrace_graph_init_task(p);
1328
1329         rt_mutex_init_task(p);
1330
1331 #ifdef CONFIG_PROVE_LOCKING
1332         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1333         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1334 #endif
1335         retval = -EAGAIN;
1336         if (atomic_read(&p->real_cred->user->processes) >=
1337                         task_rlimit(p, RLIMIT_NPROC)) {
1338                 if (p->real_cred->user != INIT_USER &&
1339                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1340                         goto bad_fork_free;
1341         }
1342         current->flags &= ~PF_NPROC_EXCEEDED;
1343
1344         retval = copy_creds(p, clone_flags);
1345         if (retval < 0)
1346                 goto bad_fork_free;
1347
1348         /*
1349          * If multiple threads are within copy_process(), then this check
1350          * triggers too late. This doesn't hurt, the check is only there
1351          * to stop root fork bombs.
1352          */
1353         retval = -EAGAIN;
1354         if (nr_threads >= max_threads)
1355                 goto bad_fork_cleanup_count;
1356
1357         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1358         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1359         p->flags |= PF_FORKNOEXEC;
1360         INIT_LIST_HEAD(&p->children);
1361         INIT_LIST_HEAD(&p->sibling);
1362         rcu_copy_process(p);
1363         p->vfork_done = NULL;
1364         spin_lock_init(&p->alloc_lock);
1365
1366         init_sigpending(&p->pending);
1367
1368         p->utime = p->stime = p->gtime = 0;
1369         p->utimescaled = p->stimescaled = 0;
1370         prev_cputime_init(&p->prev_cputime);
1371
1372 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1373         seqlock_init(&p->vtime_seqlock);
1374         p->vtime_snap = 0;
1375         p->vtime_snap_whence = VTIME_SLEEPING;
1376 #endif
1377
1378 #if defined(SPLIT_RSS_COUNTING)
1379         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1380 #endif
1381
1382         p->default_timer_slack_ns = current->timer_slack_ns;
1383
1384         task_io_accounting_init(&p->ioac);
1385         acct_clear_integrals(p);
1386
1387         posix_cpu_timers_init(p);
1388
1389         p->start_time = ktime_get_ns();
1390         p->real_start_time = ktime_get_boot_ns();
1391         p->io_context = NULL;
1392         p->audit_context = NULL;
1393         cgroup_fork(p);
1394 #ifdef CONFIG_NUMA
1395         p->mempolicy = mpol_dup(p->mempolicy);
1396         if (IS_ERR(p->mempolicy)) {
1397                 retval = PTR_ERR(p->mempolicy);
1398                 p->mempolicy = NULL;
1399                 goto bad_fork_cleanup_threadgroup_lock;
1400         }
1401 #endif
1402 #ifdef CONFIG_CPUSETS
1403         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1404         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1405         seqcount_init(&p->mems_allowed_seq);
1406 #endif
1407 #ifdef CONFIG_TRACE_IRQFLAGS
1408         p->irq_events = 0;
1409         p->hardirqs_enabled = 0;
1410         p->hardirq_enable_ip = 0;
1411         p->hardirq_enable_event = 0;
1412         p->hardirq_disable_ip = _THIS_IP_;
1413         p->hardirq_disable_event = 0;
1414         p->softirqs_enabled = 1;
1415         p->softirq_enable_ip = _THIS_IP_;
1416         p->softirq_enable_event = 0;
1417         p->softirq_disable_ip = 0;
1418         p->softirq_disable_event = 0;
1419         p->hardirq_context = 0;
1420         p->softirq_context = 0;
1421 #endif
1422
1423         p->pagefault_disabled = 0;
1424
1425 #ifdef CONFIG_LOCKDEP
1426         p->lockdep_depth = 0; /* no locks held yet */
1427         p->curr_chain_key = 0;
1428         p->lockdep_recursion = 0;
1429 #endif
1430
1431 #ifdef CONFIG_DEBUG_MUTEXES
1432         p->blocked_on = NULL; /* not blocked yet */
1433 #endif
1434 #ifdef CONFIG_BCACHE
1435         p->sequential_io        = 0;
1436         p->sequential_io_avg    = 0;
1437 #endif
1438
1439         /* Perform scheduler related setup. Assign this task to a CPU. */
1440         retval = sched_fork(clone_flags, p);
1441         if (retval)
1442                 goto bad_fork_cleanup_policy;
1443
1444         retval = perf_event_init_task(p);
1445         if (retval)
1446                 goto bad_fork_cleanup_policy;
1447         retval = audit_alloc(p);
1448         if (retval)
1449                 goto bad_fork_cleanup_perf;
1450         /* copy all the process information */
1451         shm_init_task(p);
1452         retval = copy_semundo(clone_flags, p);
1453         if (retval)
1454                 goto bad_fork_cleanup_audit;
1455         retval = copy_files(clone_flags, p);
1456         if (retval)
1457                 goto bad_fork_cleanup_semundo;
1458         retval = copy_fs(clone_flags, p);
1459         if (retval)
1460                 goto bad_fork_cleanup_files;
1461         retval = copy_sighand(clone_flags, p);
1462         if (retval)
1463                 goto bad_fork_cleanup_fs;
1464         retval = copy_signal(clone_flags, p);
1465         if (retval)
1466                 goto bad_fork_cleanup_sighand;
1467         retval = copy_mm(clone_flags, p);
1468         if (retval)
1469                 goto bad_fork_cleanup_signal;
1470         retval = copy_namespaces(clone_flags, p);
1471         if (retval)
1472                 goto bad_fork_cleanup_mm;
1473         retval = copy_io(clone_flags, p);
1474         if (retval)
1475                 goto bad_fork_cleanup_namespaces;
1476         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1477         if (retval)
1478                 goto bad_fork_cleanup_io;
1479
1480         if (pid != &init_struct_pid) {
1481                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1482                 if (IS_ERR(pid)) {
1483                         retval = PTR_ERR(pid);
1484                         goto bad_fork_cleanup_io;
1485                 }
1486         }
1487
1488         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1489         /*
1490          * Clear TID on mm_release()?
1491          */
1492         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1493 #ifdef CONFIG_BLOCK
1494         p->plug = NULL;
1495 #endif
1496 #ifdef CONFIG_FUTEX
1497         p->robust_list = NULL;
1498 #ifdef CONFIG_COMPAT
1499         p->compat_robust_list = NULL;
1500 #endif
1501         INIT_LIST_HEAD(&p->pi_state_list);
1502         p->pi_state_cache = NULL;
1503 #endif
1504         /*
1505          * sigaltstack should be cleared when sharing the same VM
1506          */
1507         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1508                 p->sas_ss_sp = p->sas_ss_size = 0;
1509
1510         /*
1511          * Syscall tracing and stepping should be turned off in the
1512          * child regardless of CLONE_PTRACE.
1513          */
1514         user_disable_single_step(p);
1515         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1516 #ifdef TIF_SYSCALL_EMU
1517         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1518 #endif
1519         clear_all_latency_tracing(p);
1520
1521         /* ok, now we should be set up.. */
1522         p->pid = pid_nr(pid);
1523         if (clone_flags & CLONE_THREAD) {
1524                 p->exit_signal = -1;
1525                 p->group_leader = current->group_leader;
1526                 p->tgid = current->tgid;
1527         } else {
1528                 if (clone_flags & CLONE_PARENT)
1529                         p->exit_signal = current->group_leader->exit_signal;
1530                 else
1531                         p->exit_signal = (clone_flags & CSIGNAL);
1532                 p->group_leader = p;
1533                 p->tgid = p->pid;
1534         }
1535
1536         p->nr_dirtied = 0;
1537         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1538         p->dirty_paused_when = 0;
1539
1540         p->pdeath_signal = 0;
1541         INIT_LIST_HEAD(&p->thread_group);
1542         p->task_works = NULL;
1543
1544         threadgroup_change_begin(current);
1545         /*
1546          * Ensure that the cgroup subsystem policies allow the new process to be
1547          * forked. It should be noted the the new process's css_set can be changed
1548          * between here and cgroup_post_fork() if an organisation operation is in
1549          * progress.
1550          */
1551         retval = cgroup_can_fork(p, cgrp_ss_priv);
1552         if (retval)
1553                 goto bad_fork_free_pid;
1554
1555         /*
1556          * Make it visible to the rest of the system, but dont wake it up yet.
1557          * Need tasklist lock for parent etc handling!
1558          */
1559         write_lock_irq(&tasklist_lock);
1560
1561         /* CLONE_PARENT re-uses the old parent */
1562         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1563                 p->real_parent = current->real_parent;
1564                 p->parent_exec_id = current->parent_exec_id;
1565         } else {
1566                 p->real_parent = current;
1567                 p->parent_exec_id = current->self_exec_id;
1568         }
1569
1570         spin_lock(&current->sighand->siglock);
1571
1572         /*
1573          * Copy seccomp details explicitly here, in case they were changed
1574          * before holding sighand lock.
1575          */
1576         copy_seccomp(p);
1577
1578         /*
1579          * Process group and session signals need to be delivered to just the
1580          * parent before the fork or both the parent and the child after the
1581          * fork. Restart if a signal comes in before we add the new process to
1582          * it's process group.
1583          * A fatal signal pending means that current will exit, so the new
1584          * thread can't slip out of an OOM kill (or normal SIGKILL).
1585         */
1586         recalc_sigpending();
1587         if (signal_pending(current)) {
1588                 spin_unlock(&current->sighand->siglock);
1589                 write_unlock_irq(&tasklist_lock);
1590                 retval = -ERESTARTNOINTR;
1591                 goto bad_fork_cancel_cgroup;
1592         }
1593
1594         if (likely(p->pid)) {
1595                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1596
1597                 init_task_pid(p, PIDTYPE_PID, pid);
1598                 if (thread_group_leader(p)) {
1599                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1600                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1601
1602                         if (is_child_reaper(pid)) {
1603                                 ns_of_pid(pid)->child_reaper = p;
1604                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1605                         }
1606
1607                         p->signal->leader_pid = pid;
1608                         p->signal->tty = tty_kref_get(current->signal->tty);
1609                         list_add_tail(&p->sibling, &p->real_parent->children);
1610                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1611                         attach_pid(p, PIDTYPE_PGID);
1612                         attach_pid(p, PIDTYPE_SID);
1613                         __this_cpu_inc(process_counts);
1614                 } else {
1615                         current->signal->nr_threads++;
1616                         atomic_inc(&current->signal->live);
1617                         atomic_inc(&current->signal->sigcnt);
1618                         list_add_tail_rcu(&p->thread_group,
1619                                           &p->group_leader->thread_group);
1620                         list_add_tail_rcu(&p->thread_node,
1621                                           &p->signal->thread_head);
1622                 }
1623                 attach_pid(p, PIDTYPE_PID);
1624                 nr_threads++;
1625         }
1626
1627         total_forks++;
1628         spin_unlock(&current->sighand->siglock);
1629         syscall_tracepoint_update(p);
1630         write_unlock_irq(&tasklist_lock);
1631
1632         proc_fork_connector(p);
1633         cgroup_post_fork(p, cgrp_ss_priv);
1634         threadgroup_change_end(current);
1635         perf_event_fork(p);
1636
1637         trace_task_newtask(p, clone_flags);
1638         uprobe_copy_process(p, clone_flags);
1639
1640         return p;
1641
1642 bad_fork_cancel_cgroup:
1643         cgroup_cancel_fork(p, cgrp_ss_priv);
1644 bad_fork_free_pid:
1645         threadgroup_change_end(current);
1646         if (pid != &init_struct_pid)
1647                 free_pid(pid);
1648 bad_fork_cleanup_io:
1649         if (p->io_context)
1650                 exit_io_context(p);
1651 bad_fork_cleanup_namespaces:
1652         exit_task_namespaces(p);
1653 bad_fork_cleanup_mm:
1654         if (p->mm)
1655                 mmput(p->mm);
1656 bad_fork_cleanup_signal:
1657         if (!(clone_flags & CLONE_THREAD))
1658                 free_signal_struct(p->signal);
1659 bad_fork_cleanup_sighand:
1660         __cleanup_sighand(p->sighand);
1661 bad_fork_cleanup_fs:
1662         exit_fs(p); /* blocking */
1663 bad_fork_cleanup_files:
1664         exit_files(p); /* blocking */
1665 bad_fork_cleanup_semundo:
1666         exit_sem(p);
1667 bad_fork_cleanup_audit:
1668         audit_free(p);
1669 bad_fork_cleanup_perf:
1670         perf_event_free_task(p);
1671 bad_fork_cleanup_policy:
1672 #ifdef CONFIG_NUMA
1673         mpol_put(p->mempolicy);
1674 bad_fork_cleanup_threadgroup_lock:
1675 #endif
1676         delayacct_tsk_free(p);
1677 bad_fork_cleanup_count:
1678         atomic_dec(&p->cred->user->processes);
1679         exit_creds(p);
1680 bad_fork_free:
1681         free_task(p);
1682 fork_out:
1683         return ERR_PTR(retval);
1684 }
1685
1686 static inline void init_idle_pids(struct pid_link *links)
1687 {
1688         enum pid_type type;
1689
1690         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1691                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1692                 links[type].pid = &init_struct_pid;
1693         }
1694 }
1695
1696 struct task_struct *fork_idle(int cpu)
1697 {
1698         struct task_struct *task;
1699         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0);
1700         if (!IS_ERR(task)) {
1701                 init_idle_pids(task->pids);
1702                 init_idle(task, cpu);
1703         }
1704
1705         return task;
1706 }
1707
1708 /*
1709  *  Ok, this is the main fork-routine.
1710  *
1711  * It copies the process, and if successful kick-starts
1712  * it and waits for it to finish using the VM if required.
1713  */
1714 long _do_fork(unsigned long clone_flags,
1715               unsigned long stack_start,
1716               unsigned long stack_size,
1717               int __user *parent_tidptr,
1718               int __user *child_tidptr,
1719               unsigned long tls)
1720 {
1721         struct task_struct *p;
1722         int trace = 0;
1723         long nr;
1724
1725         /*
1726          * Determine whether and which event to report to ptracer.  When
1727          * called from kernel_thread or CLONE_UNTRACED is explicitly
1728          * requested, no event is reported; otherwise, report if the event
1729          * for the type of forking is enabled.
1730          */
1731         if (!(clone_flags & CLONE_UNTRACED)) {
1732                 if (clone_flags & CLONE_VFORK)
1733                         trace = PTRACE_EVENT_VFORK;
1734                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1735                         trace = PTRACE_EVENT_CLONE;
1736                 else
1737                         trace = PTRACE_EVENT_FORK;
1738
1739                 if (likely(!ptrace_event_enabled(current, trace)))
1740                         trace = 0;
1741         }
1742
1743         p = copy_process(clone_flags, stack_start, stack_size,
1744                          child_tidptr, NULL, trace, tls);
1745         /*
1746          * Do this prior waking up the new thread - the thread pointer
1747          * might get invalid after that point, if the thread exits quickly.
1748          */
1749         if (!IS_ERR(p)) {
1750                 struct completion vfork;
1751                 struct pid *pid;
1752
1753                 trace_sched_process_fork(current, p);
1754
1755                 pid = get_task_pid(p, PIDTYPE_PID);
1756                 nr = pid_vnr(pid);
1757
1758                 if (clone_flags & CLONE_PARENT_SETTID)
1759                         put_user(nr, parent_tidptr);
1760
1761                 if (clone_flags & CLONE_VFORK) {
1762                         p->vfork_done = &vfork;
1763                         init_completion(&vfork);
1764                         get_task_struct(p);
1765                 }
1766
1767                 wake_up_new_task(p);
1768
1769                 /* forking complete and child started to run, tell ptracer */
1770                 if (unlikely(trace))
1771                         ptrace_event_pid(trace, pid);
1772
1773                 if (clone_flags & CLONE_VFORK) {
1774                         if (!wait_for_vfork_done(p, &vfork))
1775                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1776                 }
1777
1778                 put_pid(pid);
1779         } else {
1780                 nr = PTR_ERR(p);
1781         }
1782         return nr;
1783 }
1784
1785 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1786 /* For compatibility with architectures that call do_fork directly rather than
1787  * using the syscall entry points below. */
1788 long do_fork(unsigned long clone_flags,
1789               unsigned long stack_start,
1790               unsigned long stack_size,
1791               int __user *parent_tidptr,
1792               int __user *child_tidptr)
1793 {
1794         return _do_fork(clone_flags, stack_start, stack_size,
1795                         parent_tidptr, child_tidptr, 0);
1796 }
1797 #endif
1798
1799 /*
1800  * Create a kernel thread.
1801  */
1802 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1803 {
1804         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1805                 (unsigned long)arg, NULL, NULL, 0);
1806 }
1807
1808 #ifdef __ARCH_WANT_SYS_FORK
1809 SYSCALL_DEFINE0(fork)
1810 {
1811 #ifdef CONFIG_MMU
1812         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1813 #else
1814         /* can not support in nommu mode */
1815         return -EINVAL;
1816 #endif
1817 }
1818 #endif
1819
1820 #ifdef __ARCH_WANT_SYS_VFORK
1821 SYSCALL_DEFINE0(vfork)
1822 {
1823         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1824                         0, NULL, NULL, 0);
1825 }
1826 #endif
1827
1828 #ifdef __ARCH_WANT_SYS_CLONE
1829 #ifdef CONFIG_CLONE_BACKWARDS
1830 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1831                  int __user *, parent_tidptr,
1832                  unsigned long, tls,
1833                  int __user *, child_tidptr)
1834 #elif defined(CONFIG_CLONE_BACKWARDS2)
1835 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1836                  int __user *, parent_tidptr,
1837                  int __user *, child_tidptr,
1838                  unsigned long, tls)
1839 #elif defined(CONFIG_CLONE_BACKWARDS3)
1840 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1841                 int, stack_size,
1842                 int __user *, parent_tidptr,
1843                 int __user *, child_tidptr,
1844                 unsigned long, tls)
1845 #else
1846 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1847                  int __user *, parent_tidptr,
1848                  int __user *, child_tidptr,
1849                  unsigned long, tls)
1850 #endif
1851 {
1852         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
1853 }
1854 #endif
1855
1856 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1857 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1858 #endif
1859
1860 static void sighand_ctor(void *data)
1861 {
1862         struct sighand_struct *sighand = data;
1863
1864         spin_lock_init(&sighand->siglock);
1865         init_waitqueue_head(&sighand->signalfd_wqh);
1866 }
1867
1868 void __init proc_caches_init(void)
1869 {
1870         sighand_cachep = kmem_cache_create("sighand_cache",
1871                         sizeof(struct sighand_struct), 0,
1872                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1873                         SLAB_NOTRACK, sighand_ctor);
1874         signal_cachep = kmem_cache_create("signal_cache",
1875                         sizeof(struct signal_struct), 0,
1876                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1877         files_cachep = kmem_cache_create("files_cache",
1878                         sizeof(struct files_struct), 0,
1879                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1880         fs_cachep = kmem_cache_create("fs_cache",
1881                         sizeof(struct fs_struct), 0,
1882                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1883         /*
1884          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1885          * whole struct cpumask for the OFFSTACK case. We could change
1886          * this to *only* allocate as much of it as required by the
1887          * maximum number of CPU's we can ever have.  The cpumask_allocation
1888          * is at the end of the structure, exactly for that reason.
1889          */
1890         mm_cachep = kmem_cache_create("mm_struct",
1891                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1892                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1893         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1894         mmap_init();
1895         nsproxy_cache_init();
1896 }
1897
1898 /*
1899  * Check constraints on flags passed to the unshare system call.
1900  */
1901 static int check_unshare_flags(unsigned long unshare_flags)
1902 {
1903         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1904                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1905                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1906                                 CLONE_NEWUSER|CLONE_NEWPID))
1907                 return -EINVAL;
1908         /*
1909          * Not implemented, but pretend it works if there is nothing
1910          * to unshare.  Note that unsharing the address space or the
1911          * signal handlers also need to unshare the signal queues (aka
1912          * CLONE_THREAD).
1913          */
1914         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1915                 if (!thread_group_empty(current))
1916                         return -EINVAL;
1917         }
1918         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
1919                 if (atomic_read(&current->sighand->count) > 1)
1920                         return -EINVAL;
1921         }
1922         if (unshare_flags & CLONE_VM) {
1923                 if (!current_is_single_threaded())
1924                         return -EINVAL;
1925         }
1926
1927         return 0;
1928 }
1929
1930 /*
1931  * Unshare the filesystem structure if it is being shared
1932  */
1933 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1934 {
1935         struct fs_struct *fs = current->fs;
1936
1937         if (!(unshare_flags & CLONE_FS) || !fs)
1938                 return 0;
1939
1940         /* don't need lock here; in the worst case we'll do useless copy */
1941         if (fs->users == 1)
1942                 return 0;
1943
1944         *new_fsp = copy_fs_struct(fs);
1945         if (!*new_fsp)
1946                 return -ENOMEM;
1947
1948         return 0;
1949 }
1950
1951 /*
1952  * Unshare file descriptor table if it is being shared
1953  */
1954 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1955 {
1956         struct files_struct *fd = current->files;
1957         int error = 0;
1958
1959         if ((unshare_flags & CLONE_FILES) &&
1960             (fd && atomic_read(&fd->count) > 1)) {
1961                 *new_fdp = dup_fd(fd, &error);
1962                 if (!*new_fdp)
1963                         return error;
1964         }
1965
1966         return 0;
1967 }
1968
1969 /*
1970  * unshare allows a process to 'unshare' part of the process
1971  * context which was originally shared using clone.  copy_*
1972  * functions used by do_fork() cannot be used here directly
1973  * because they modify an inactive task_struct that is being
1974  * constructed. Here we are modifying the current, active,
1975  * task_struct.
1976  */
1977 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1978 {
1979         struct fs_struct *fs, *new_fs = NULL;
1980         struct files_struct *fd, *new_fd = NULL;
1981         struct cred *new_cred = NULL;
1982         struct nsproxy *new_nsproxy = NULL;
1983         int do_sysvsem = 0;
1984         int err;
1985
1986         /*
1987          * If unsharing a user namespace must also unshare the thread group
1988          * and unshare the filesystem root and working directories.
1989          */
1990         if (unshare_flags & CLONE_NEWUSER)
1991                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1992         /*
1993          * If unsharing vm, must also unshare signal handlers.
1994          */
1995         if (unshare_flags & CLONE_VM)
1996                 unshare_flags |= CLONE_SIGHAND;
1997         /*
1998          * If unsharing a signal handlers, must also unshare the signal queues.
1999          */
2000         if (unshare_flags & CLONE_SIGHAND)
2001                 unshare_flags |= CLONE_THREAD;
2002         /*
2003          * If unsharing namespace, must also unshare filesystem information.
2004          */
2005         if (unshare_flags & CLONE_NEWNS)
2006                 unshare_flags |= CLONE_FS;
2007
2008         err = check_unshare_flags(unshare_flags);
2009         if (err)
2010                 goto bad_unshare_out;
2011         /*
2012          * CLONE_NEWIPC must also detach from the undolist: after switching
2013          * to a new ipc namespace, the semaphore arrays from the old
2014          * namespace are unreachable.
2015          */
2016         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2017                 do_sysvsem = 1;
2018         err = unshare_fs(unshare_flags, &new_fs);
2019         if (err)
2020                 goto bad_unshare_out;
2021         err = unshare_fd(unshare_flags, &new_fd);
2022         if (err)
2023                 goto bad_unshare_cleanup_fs;
2024         err = unshare_userns(unshare_flags, &new_cred);
2025         if (err)
2026                 goto bad_unshare_cleanup_fd;
2027         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2028                                          new_cred, new_fs);
2029         if (err)
2030                 goto bad_unshare_cleanup_cred;
2031
2032         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2033                 if (do_sysvsem) {
2034                         /*
2035                          * CLONE_SYSVSEM is equivalent to sys_exit().
2036                          */
2037                         exit_sem(current);
2038                 }
2039                 if (unshare_flags & CLONE_NEWIPC) {
2040                         /* Orphan segments in old ns (see sem above). */
2041                         exit_shm(current);
2042                         shm_init_task(current);
2043                 }
2044
2045                 if (new_nsproxy)
2046                         switch_task_namespaces(current, new_nsproxy);
2047
2048                 task_lock(current);
2049
2050                 if (new_fs) {
2051                         fs = current->fs;
2052                         spin_lock(&fs->lock);
2053                         current->fs = new_fs;
2054                         if (--fs->users)
2055                                 new_fs = NULL;
2056                         else
2057                                 new_fs = fs;
2058                         spin_unlock(&fs->lock);
2059                 }
2060
2061                 if (new_fd) {
2062                         fd = current->files;
2063                         current->files = new_fd;
2064                         new_fd = fd;
2065                 }
2066
2067                 task_unlock(current);
2068
2069                 if (new_cred) {
2070                         /* Install the new user namespace */
2071                         commit_creds(new_cred);
2072                         new_cred = NULL;
2073                 }
2074         }
2075
2076 bad_unshare_cleanup_cred:
2077         if (new_cred)
2078                 put_cred(new_cred);
2079 bad_unshare_cleanup_fd:
2080         if (new_fd)
2081                 put_files_struct(new_fd);
2082
2083 bad_unshare_cleanup_fs:
2084         if (new_fs)
2085                 free_fs_struct(new_fs);
2086
2087 bad_unshare_out:
2088         return err;
2089 }
2090
2091 /*
2092  *      Helper to unshare the files of the current task.
2093  *      We don't want to expose copy_files internals to
2094  *      the exec layer of the kernel.
2095  */
2096
2097 int unshare_files(struct files_struct **displaced)
2098 {
2099         struct task_struct *task = current;
2100         struct files_struct *copy = NULL;
2101         int error;
2102
2103         error = unshare_fd(CLONE_FILES, &copy);
2104         if (error || !copy) {
2105                 *displaced = NULL;
2106                 return error;
2107         }
2108         *displaced = task->files;
2109         task_lock(task);
2110         task->files = copy;
2111         task_unlock(task);
2112         return 0;
2113 }
2114
2115 int sysctl_max_threads(struct ctl_table *table, int write,
2116                        void __user *buffer, size_t *lenp, loff_t *ppos)
2117 {
2118         struct ctl_table t;
2119         int ret;
2120         int threads = max_threads;
2121         int min = MIN_THREADS;
2122         int max = MAX_THREADS;
2123
2124         t = *table;
2125         t.data = &threads;
2126         t.extra1 = &min;
2127         t.extra2 = &max;
2128
2129         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2130         if (ret || !write)
2131                 return ret;
2132
2133         set_max_threads(threads);
2134
2135         return 0;
2136 }