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_mm - acquire a reference to the task's mm
 768  *
 769  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 770  * this kernel workthread has transiently adopted a user mm with use_mm,
 771  * to do its AIO) is not set and if so returns a reference to it, after
 772  * bumping up the use count.  User must release the mm via mmput()
 773  * after use.  Typically used by /proc and ptrace.
 774  */
 775 struct mm_struct *get_task_mm(struct task_struct *task)
 776 {
 777         struct mm_struct *mm;
 778
 779         task_lock(task);
 780         mm = task->mm;
 781         if (mm) {
 782                 if (task->flags & PF_KTHREAD)
 783                         mm = NULL;
 784                 else
 785                         atomic_inc(&mm->mm_users);
 786         }
 787         task_unlock(task);
 788         return mm;
 789 }
 790 EXPORT_SYMBOL_GPL(get_task_mm);
 791
 792 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 793 {
 794         struct mm_struct *mm;
 795         int err;
 796
 797         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 798         if (err)
 799                 return ERR_PTR(err);
 800
 801         mm = get_task_mm(task);
 802         if (mm && mm != current->mm &&
 803                         !ptrace_may_access(task, mode) &&
 804                         !capable(CAP_SYS_RESOURCE)) {
 805                 mmput(mm);
 806                 mm = ERR_PTR(-EACCES);
 807         }
 808         mutex_unlock(&task->signal->cred_guard_mutex);
 809
 810         return mm;
 811 }
 812
 813 static void complete_vfork_done(struct task_struct *tsk)
 814 {
 815         struct completion *vfork;
 816
 817         task_lock(tsk);
 818         vfork = tsk->vfork_done;
 819         if (likely(vfork)) {
 820                 tsk->vfork_done = NULL;
 821                 complete(vfork);
 822         }
 823         task_unlock(tsk);
 824 }
 825
 826 static int wait_for_vfork_done(struct task_struct *child,
 827                                 struct completion *vfork)
 828 {
 829         int killed;
 830
 831         freezer_do_not_count();
 832         killed = wait_for_completion_killable(vfork);
 833         freezer_count();
 834
 835         if (killed) {
 836                 task_lock(child);
 837                 child->vfork_done = NULL;
 838                 task_unlock(child);
 839         }
 840
 841         put_task_struct(child);
 842         return killed;
 843 }
 844
 845 /* Please note the differences between mmput and mm_release.
 846  * mmput is called whenever we stop holding onto a mm_struct,
 847  * error success whatever.
 848  *
 849  * mm_release is called after a mm_struct has been removed
 850  * from the current process.
 851  *
 852  * This difference is important for error handling, when we
 853  * only half set up a mm_struct for a new process and need to restore
 854  * the old one.  Because we mmput the new mm_struct before
 855  * restoring the old one. . .
 856  * Eric Biederman 10 January 1998
 857  */
 858 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 859 {
 860         /* Get rid of any futexes when releasing the mm */
 861 #ifdef CONFIG_FUTEX
 862         if (unlikely(tsk->robust_list)) {
 863                 exit_robust_list(tsk);
 864                 tsk->robust_list = NULL;
 865         }
 866 #ifdef CONFIG_COMPAT
 867         if (unlikely(tsk->compat_robust_list)) {
 868                 compat_exit_robust_list(tsk);
 869                 tsk->compat_robust_list = NULL;
 870         }
 871 #endif
 872         if (unlikely(!list_empty(&tsk->pi_state_list)))
 873                 exit_pi_state_list(tsk);
 874 #endif
 875
 876         uprobe_free_utask(tsk);
 877
 878         /* Get rid of any cached register state */
 879         deactivate_mm(tsk, mm);
 880
 881         /*
 882          * If we're exiting normally, clear a user-space tid field if
 883          * requested.  We leave this alone when dying by signal, to leave
 884          * the value intact in a core dump, and to save the unnecessary
 885          * trouble, say, a killed vfork parent shouldn't touch this mm.
 886          * Userland only wants this done for a sys_exit.
 887          */
 888         if (tsk->clear_child_tid) {
 889                 if (!(tsk->flags & PF_SIGNALED) &&
 890                     atomic_read(&mm->mm_users) > 1) {
 891                         /*
 892                          * We don't check the error code - if userspace has
 893                          * not set up a proper pointer then tough luck.
 894                          */
 895                         put_user(0, tsk->clear_child_tid);
 896                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 897                                         1, NULL, NULL, 0);
 898                 }
 899                 tsk->clear_child_tid = NULL;
 900         }
 901
 902         /*
 903          * All done, finally we can wake up parent and return this mm to him.
 904          * Also kthread_stop() uses this completion for synchronization.
 905          */
 906         if (tsk->vfork_done)
 907                 complete_vfork_done(tsk);
 908 }
 909
 910 /*
 911  * Allocate a new mm structure and copy contents from the
 912  * mm structure of the passed in task structure.
 913  */
 914 static struct mm_struct *dup_mm(struct task_struct *tsk)
 915 {
 916         struct mm_struct *mm, *oldmm = current->mm;
 917         int err;
 918
 919         mm = allocate_mm();
 920         if (!mm)
 921                 goto fail_nomem;
 922
 923         memcpy(mm, oldmm, sizeof(*mm));
 924
 925         if (!mm_init(mm, tsk))
 926                 goto fail_nomem;
 927
 928         err = dup_mmap(mm, oldmm);
 929         if (err)
 930                 goto free_pt;
 931
 932         mm->hiwater_rss = get_mm_rss(mm);
 933         mm->hiwater_vm = mm->total_vm;
 934
 935         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 936                 goto free_pt;
 937
 938         return mm;
 939
 940 free_pt:
 941         /* don't put binfmt in mmput, we haven't got module yet */
 942         mm->binfmt = NULL;
 943         mmput(mm);
 944
 945 fail_nomem:
 946         return NULL;
 947 }
 948
 949 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 950 {
 951         struct mm_struct *mm, *oldmm;
 952         int retval;
 953
 954         tsk->min_flt = tsk->maj_flt = 0;
 955         tsk->nvcsw = tsk->nivcsw = 0;
 956 #ifdef CONFIG_DETECT_HUNG_TASK
 957         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 958 #endif
 959
 960         tsk->mm = NULL;
 961         tsk->active_mm = NULL;
 962
 963         /*
 964          * Are we cloning a kernel thread?
 965          *
 966          * We need to steal a active VM for that..
 967          */
 968         oldmm = current->mm;
 969         if (!oldmm)
 970                 return 0;
 971
 972         /* initialize the new vmacache entries */
 973         vmacache_flush(tsk);
 974
 975         if (clone_flags & CLONE_VM) {
 976                 atomic_inc(&oldmm->mm_users);
 977                 mm = oldmm;
 978                 goto good_mm;
 979         }
 980
 981         retval = -ENOMEM;
 982         mm = dup_mm(tsk);
 983         if (!mm)
 984                 goto fail_nomem;
 985
 986 good_mm:
 987         tsk->mm = mm;
 988         tsk->active_mm = mm;
 989         return 0;
 990
 991 fail_nomem:
 992         return retval;
 993 }
 994
 995 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 996 {
 997         struct fs_struct *fs = current->fs;
 998         if (clone_flags & CLONE_FS) {
 999                 /* tsk->fs is already what we want */
1000                 spin_lock(&fs->lock);
1001                 if (fs->in_exec) {
1002                         spin_unlock(&fs->lock);
1003                         return -EAGAIN;
1004                 }
1005                 fs->users++;
1006                 spin_unlock(&fs->lock);
1007                 return 0;
1008         }
1009         tsk->fs = copy_fs_struct(fs);
1010         if (!tsk->fs)
1011                 return -ENOMEM;
1012         return 0;
1013 }
1014
1015 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1016 {
1017         struct files_struct *oldf, *newf;
1018         int error = 0;
1019
1020         /*
1021          * A background process may not have any files ...
1022          */
1023         oldf = current->files;
1024         if (!oldf)
1025                 goto out;
1026
1027         if (clone_flags & CLONE_FILES) {
1028                 atomic_inc(&oldf->count);
1029                 goto out;
1030         }
1031
1032         newf = dup_fd(oldf, &error);
1033         if (!newf)
1034                 goto out;
1035
1036         tsk->files = newf;
1037         error = 0;
1038 out:
1039         return error;
1040 }
1041
1042 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1043 {
1044 #ifdef CONFIG_BLOCK
1045         struct io_context *ioc = current->io_context;
1046         struct io_context *new_ioc;
1047
1048         if (!ioc)
1049                 return 0;
1050         /*
1051          * Share io context with parent, if CLONE_IO is set
1052          */
1053         if (clone_flags & CLONE_IO) {
1054                 ioc_task_link(ioc);
1055                 tsk->io_context = ioc;
1056         } else if (ioprio_valid(ioc->ioprio)) {
1057                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1058                 if (unlikely(!new_ioc))
1059                         return -ENOMEM;
1060
1061                 new_ioc->ioprio = ioc->ioprio;
1062                 put_io_context(new_ioc);
1063         }
1064 #endif
1065         return 0;
1066 }
1067
1068 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1069 {
1070         struct sighand_struct *sig;
1071
1072         if (clone_flags & CLONE_SIGHAND) {
1073                 atomic_inc(&current->sighand->count);
1074                 return 0;
1075         }
1076         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1077         rcu_assign_pointer(tsk->sighand, sig);
1078         if (!sig)
1079                 return -ENOMEM;
1080
1081         atomic_set(&sig->count, 1);
1082         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1083         return 0;
1084 }
1085
1086 void __cleanup_sighand(struct sighand_struct *sighand)
1087 {
1088         if (atomic_dec_and_test(&sighand->count)) {
1089                 signalfd_cleanup(sighand);
1090                 /*
1091                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1092                  * without an RCU grace period, see __lock_task_sighand().
1093                  */
1094                 kmem_cache_free(sighand_cachep, sighand);
1095         }
1096 }
1097
1098 /*
1099  * Initialize POSIX timer handling for a thread group.
1100  */
1101 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1102 {
1103         unsigned long cpu_limit;
1104
1105         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1106         if (cpu_limit != RLIM_INFINITY) {
1107                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1108                 sig->cputimer.running = true;
1109         }
1110
1111         /* The timer lists. */
1112         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1113         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1114         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1115 }
1116
1117 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1118 {
1119         struct signal_struct *sig;
1120
1121         if (clone_flags & CLONE_THREAD)
1122                 return 0;
1123
1124         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1125         tsk->signal = sig;
1126         if (!sig)
1127                 return -ENOMEM;
1128
1129         sig->nr_threads = 1;
1130         atomic_set(&sig->live, 1);
1131         atomic_set(&sig->sigcnt, 1);
1132
1133         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1134         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1135         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1136
1137         init_waitqueue_head(&sig->wait_chldexit);
1138         sig->curr_target = tsk;
1139         init_sigpending(&sig->shared_pending);
1140         INIT_LIST_HEAD(&sig->posix_timers);
1141         seqlock_init(&sig->stats_lock);
1142         prev_cputime_init(&sig->prev_cputime);
1143
1144         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1145         sig->real_timer.function = it_real_fn;
1146
1147         task_lock(current->group_leader);
1148         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1149         task_unlock(current->group_leader);
1150
1151         posix_cpu_timers_init_group(sig);
1152
1153         tty_audit_fork(sig);
1154         sched_autogroup_fork(sig);
1155
1156         sig->oom_score_adj = current->signal->oom_score_adj;
1157         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1158
1159         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1160                                    current->signal->is_child_subreaper;
1161
1162         mutex_init(&sig->cred_guard_mutex);
1163
1164         return 0;
1165 }
1166
1167 static void copy_seccomp(struct task_struct *p)
1168 {
1169 #ifdef CONFIG_SECCOMP
1170         /*
1171          * Must be called with sighand->lock held, which is common to
1172          * all threads in the group. Holding cred_guard_mutex is not
1173          * needed because this new task is not yet running and cannot
1174          * be racing exec.
1175          */
1176         assert_spin_locked(&current->sighand->siglock);
1177
1178         /* Ref-count the new filter user, and assign it. */
1179         get_seccomp_filter(current);
1180         p->seccomp = current->seccomp;
1181
1182         /*
1183          * Explicitly enable no_new_privs here in case it got set
1184          * between the task_struct being duplicated and holding the
1185          * sighand lock. The seccomp state and nnp must be in sync.
1186          */
1187         if (task_no_new_privs(current))
1188                 task_set_no_new_privs(p);
1189
1190         /*
1191          * If the parent gained a seccomp mode after copying thread
1192          * flags and between before we held the sighand lock, we have
1193          * to manually enable the seccomp thread flag here.
1194          */
1195         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1196                 set_tsk_thread_flag(p, TIF_SECCOMP);
1197 #endif
1198 }
1199
1200 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1201 {
1202         current->clear_child_tid = tidptr;
1203
1204         return task_pid_vnr(current);
1205 }
1206
1207 static void rt_mutex_init_task(struct task_struct *p)
1208 {
1209         raw_spin_lock_init(&p->pi_lock);
1210 #ifdef CONFIG_RT_MUTEXES
1211         p->pi_waiters = RB_ROOT;
1212         p->pi_waiters_leftmost = NULL;
1213         p->pi_blocked_on = NULL;
1214 #endif
1215 }
1216
1217 /*
1218  * Initialize POSIX timer handling for a single task.
1219  */
1220 static void posix_cpu_timers_init(struct task_struct *tsk)
1221 {
1222         tsk->cputime_expires.prof_exp = 0;
1223         tsk->cputime_expires.virt_exp = 0;
1224         tsk->cputime_expires.sched_exp = 0;
1225         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1226         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1227         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1228 }
1229
1230 static inline void
1231 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1232 {
1233          task->pids[type].pid = pid;
1234 }
1235
1236 /*
1237  * This creates a new process as a copy of the old one,
1238  * but does not actually start it yet.
1239  *
1240  * It copies the registers, and all the appropriate
1241  * parts of the process environment (as per the clone
1242  * flags). The actual kick-off is left to the caller.
1243  */
1244 static struct task_struct *copy_process(unsigned long clone_flags,
1245                                         unsigned long stack_start,
1246                                         unsigned long stack_size,
1247                                         int __user *child_tidptr,
1248                                         struct pid *pid,
1249                                         int trace,
1250                                         unsigned long tls)
1251 {
1252         int retval;
1253         struct task_struct *p;
1254         void *cgrp_ss_priv[CGROUP_CANFORK_COUNT] = {};
1255
1256         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1257                 return ERR_PTR(-EINVAL);
1258
1259         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1260                 return ERR_PTR(-EINVAL);
1261
1262         /*
1263          * Thread groups must share signals as well, and detached threads
1264          * can only be started up within the thread group.
1265          */
1266         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1267                 return ERR_PTR(-EINVAL);
1268
1269         /*
1270          * Shared signal handlers imply shared VM. By way of the above,
1271          * thread groups also imply shared VM. Blocking this case allows
1272          * for various simplifications in other code.
1273          */
1274         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1275                 return ERR_PTR(-EINVAL);
1276
1277         /*
1278          * Siblings of global init remain as zombies on exit since they are
1279          * not reaped by their parent (swapper). To solve this and to avoid
1280          * multi-rooted process trees, prevent global and container-inits
1281          * from creating siblings.
1282          */
1283         if ((clone_flags & CLONE_PARENT) &&
1284                                 current->signal->flags & SIGNAL_UNKILLABLE)
1285                 return ERR_PTR(-EINVAL);
1286
1287         /*
1288          * If the new process will be in a different pid or user namespace
1289          * do not allow it to share a thread group with the forking task.
1290          */
1291         if (clone_flags & CLONE_THREAD) {
1292                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1293                     (task_active_pid_ns(current) !=
1294                                 current->nsproxy->pid_ns_for_children))
1295                         return ERR_PTR(-EINVAL);
1296         }
1297
1298         retval = security_task_create(clone_flags);
1299         if (retval)
1300                 goto fork_out;
1301
1302         retval = -ENOMEM;
1303         p = dup_task_struct(current);
1304         if (!p)
1305                 goto fork_out;
1306
1307         ftrace_graph_init_task(p);
1308
1309         rt_mutex_init_task(p);
1310
1311 #ifdef CONFIG_PROVE_LOCKING
1312         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1313         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1314 #endif
1315         retval = -EAGAIN;
1316         if (atomic_read(&p->real_cred->user->processes) >=
1317                         task_rlimit(p, RLIMIT_NPROC)) {
1318                 if (p->real_cred->user != INIT_USER &&
1319                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1320                         goto bad_fork_free;
1321         }
1322         current->flags &= ~PF_NPROC_EXCEEDED;
1323
1324         retval = copy_creds(p, clone_flags);
1325         if (retval < 0)
1326                 goto bad_fork_free;
1327
1328         /*
1329          * If multiple threads are within copy_process(), then this check
1330          * triggers too late. This doesn't hurt, the check is only there
1331          * to stop root fork bombs.
1332          */
1333         retval = -EAGAIN;
1334         if (nr_threads >= max_threads)
1335                 goto bad_fork_cleanup_count;
1336
1337         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1338         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1339         p->flags |= PF_FORKNOEXEC;
1340         INIT_LIST_HEAD(&p->children);
1341         INIT_LIST_HEAD(&p->sibling);
1342         rcu_copy_process(p);
1343         p->vfork_done = NULL;
1344         spin_lock_init(&p->alloc_lock);
1345
1346         init_sigpending(&p->pending);
1347
1348         p->utime = p->stime = p->gtime = 0;
1349         p->utimescaled = p->stimescaled = 0;
1350         prev_cputime_init(&p->prev_cputime);
1351
1352 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1353         seqlock_init(&p->vtime_seqlock);
1354         p->vtime_snap = 0;
1355         p->vtime_snap_whence = VTIME_SLEEPING;
1356 #endif
1357
1358 #if defined(SPLIT_RSS_COUNTING)
1359         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1360 #endif
1361
1362         p->default_timer_slack_ns = current->timer_slack_ns;
1363
1364         task_io_accounting_init(&p->ioac);
1365         acct_clear_integrals(p);
1366
1367         posix_cpu_timers_init(p);
1368
1369         p->start_time = ktime_get_ns();
1370         p->real_start_time = ktime_get_boot_ns();
1371         p->io_context = NULL;
1372         p->audit_context = NULL;
1373         threadgroup_change_begin(current);
1374         cgroup_fork(p);
1375 #ifdef CONFIG_NUMA
1376         p->mempolicy = mpol_dup(p->mempolicy);
1377         if (IS_ERR(p->mempolicy)) {
1378                 retval = PTR_ERR(p->mempolicy);
1379                 p->mempolicy = NULL;
1380                 goto bad_fork_cleanup_threadgroup_lock;
1381         }
1382 #endif
1383 #ifdef CONFIG_CPUSETS
1384         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1385         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1386         seqcount_init(&p->mems_allowed_seq);
1387 #endif
1388 #ifdef CONFIG_TRACE_IRQFLAGS
1389         p->irq_events = 0;
1390         p->hardirqs_enabled = 0;
1391         p->hardirq_enable_ip = 0;
1392         p->hardirq_enable_event = 0;
1393         p->hardirq_disable_ip = _THIS_IP_;
1394         p->hardirq_disable_event = 0;
1395         p->softirqs_enabled = 1;
1396         p->softirq_enable_ip = _THIS_IP_;
1397         p->softirq_enable_event = 0;
1398         p->softirq_disable_ip = 0;
1399         p->softirq_disable_event = 0;
1400         p->hardirq_context = 0;
1401         p->softirq_context = 0;
1402 #endif
1403
1404         p->pagefault_disabled = 0;
1405
1406 #ifdef CONFIG_LOCKDEP
1407         p->lockdep_depth = 0; /* no locks held yet */
1408         p->curr_chain_key = 0;
1409         p->lockdep_recursion = 0;
1410 #endif
1411
1412 #ifdef CONFIG_DEBUG_MUTEXES
1413         p->blocked_on = NULL; /* not blocked yet */
1414 #endif
1415 #ifdef CONFIG_BCACHE
1416         p->sequential_io        = 0;
1417         p->sequential_io_avg    = 0;
1418 #endif
1419
1420         /* Perform scheduler related setup. Assign this task to a CPU. */
1421         retval = sched_fork(clone_flags, p);
1422         if (retval)
1423                 goto bad_fork_cleanup_policy;
1424
1425         retval = perf_event_init_task(p);
1426         if (retval)
1427                 goto bad_fork_cleanup_policy;
1428         retval = audit_alloc(p);
1429         if (retval)
1430                 goto bad_fork_cleanup_perf;
1431         /* copy all the process information */
1432         shm_init_task(p);
1433         retval = copy_semundo(clone_flags, p);
1434         if (retval)
1435                 goto bad_fork_cleanup_audit;
1436         retval = copy_files(clone_flags, p);
1437         if (retval)
1438                 goto bad_fork_cleanup_semundo;
1439         retval = copy_fs(clone_flags, p);
1440         if (retval)
1441                 goto bad_fork_cleanup_files;
1442         retval = copy_sighand(clone_flags, p);
1443         if (retval)
1444                 goto bad_fork_cleanup_fs;
1445         retval = copy_signal(clone_flags, p);
1446         if (retval)
1447                 goto bad_fork_cleanup_sighand;
1448         retval = copy_mm(clone_flags, p);
1449         if (retval)
1450                 goto bad_fork_cleanup_signal;
1451         retval = copy_namespaces(clone_flags, p);
1452         if (retval)
1453                 goto bad_fork_cleanup_mm;
1454         retval = copy_io(clone_flags, p);
1455         if (retval)
1456                 goto bad_fork_cleanup_namespaces;
1457         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1458         if (retval)
1459                 goto bad_fork_cleanup_io;
1460
1461         if (pid != &init_struct_pid) {
1462                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1463                 if (IS_ERR(pid)) {
1464                         retval = PTR_ERR(pid);
1465                         goto bad_fork_cleanup_io;
1466                 }
1467         }
1468
1469         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1470         /*
1471          * Clear TID on mm_release()?
1472          */
1473         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1474 #ifdef CONFIG_BLOCK
1475         p->plug = NULL;
1476 #endif
1477 #ifdef CONFIG_FUTEX
1478         p->robust_list = NULL;
1479 #ifdef CONFIG_COMPAT
1480         p->compat_robust_list = NULL;
1481 #endif
1482         INIT_LIST_HEAD(&p->pi_state_list);
1483         p->pi_state_cache = NULL;
1484 #endif
1485         /*
1486          * sigaltstack should be cleared when sharing the same VM
1487          */
1488         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1489                 p->sas_ss_sp = p->sas_ss_size = 0;
1490
1491         /*
1492          * Syscall tracing and stepping should be turned off in the
1493          * child regardless of CLONE_PTRACE.
1494          */
1495         user_disable_single_step(p);
1496         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1497 #ifdef TIF_SYSCALL_EMU
1498         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1499 #endif
1500         clear_all_latency_tracing(p);
1501
1502         /* ok, now we should be set up.. */
1503         p->pid = pid_nr(pid);
1504         if (clone_flags & CLONE_THREAD) {
1505                 p->exit_signal = -1;
1506                 p->group_leader = current->group_leader;
1507                 p->tgid = current->tgid;
1508         } else {
1509                 if (clone_flags & CLONE_PARENT)
1510                         p->exit_signal = current->group_leader->exit_signal;
1511                 else
1512                         p->exit_signal = (clone_flags & CSIGNAL);
1513                 p->group_leader = p;
1514                 p->tgid = p->pid;
1515         }
1516
1517         p->nr_dirtied = 0;
1518         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1519         p->dirty_paused_when = 0;
1520
1521         p->pdeath_signal = 0;
1522         INIT_LIST_HEAD(&p->thread_group);
1523         p->task_works = NULL;
1524
1525         /*
1526          * Ensure that the cgroup subsystem policies allow the new process to be
1527          * forked. It should be noted the the new process's css_set can be changed
1528          * between here and cgroup_post_fork() if an organisation operation is in
1529          * progress.
1530          */
1531         retval = cgroup_can_fork(p, cgrp_ss_priv);
1532         if (retval)
1533                 goto bad_fork_free_pid;
1534
1535         /*
1536          * Make it visible to the rest of the system, but dont wake it up yet.
1537          * Need tasklist lock for parent etc handling!
1538          */
1539         write_lock_irq(&tasklist_lock);
1540
1541         /* CLONE_PARENT re-uses the old parent */
1542         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1543                 p->real_parent = current->real_parent;
1544                 p->parent_exec_id = current->parent_exec_id;
1545         } else {
1546                 p->real_parent = current;
1547                 p->parent_exec_id = current->self_exec_id;
1548         }
1549
1550         spin_lock(&current->sighand->siglock);
1551
1552         /*
1553          * Copy seccomp details explicitly here, in case they were changed
1554          * before holding sighand lock.
1555          */
1556         copy_seccomp(p);
1557
1558         /*
1559          * Process group and session signals need to be delivered to just the
1560          * parent before the fork or both the parent and the child after the
1561          * fork. Restart if a signal comes in before we add the new process to
1562          * it's process group.
1563          * A fatal signal pending means that current will exit, so the new
1564          * thread can't slip out of an OOM kill (or normal SIGKILL).
1565         */
1566         recalc_sigpending();
1567         if (signal_pending(current)) {
1568                 spin_unlock(&current->sighand->siglock);
1569                 write_unlock_irq(&tasklist_lock);
1570                 retval = -ERESTARTNOINTR;
1571                 goto bad_fork_cancel_cgroup;
1572         }
1573
1574         if (likely(p->pid)) {
1575                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1576
1577                 init_task_pid(p, PIDTYPE_PID, pid);
1578                 if (thread_group_leader(p)) {
1579                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1580                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1581
1582                         if (is_child_reaper(pid)) {
1583                                 ns_of_pid(pid)->child_reaper = p;
1584                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1585                         }
1586
1587                         p->signal->leader_pid = pid;
1588                         p->signal->tty = tty_kref_get(current->signal->tty);
1589                         list_add_tail(&p->sibling, &p->real_parent->children);
1590                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1591                         attach_pid(p, PIDTYPE_PGID);
1592                         attach_pid(p, PIDTYPE_SID);
1593                         __this_cpu_inc(process_counts);
1594                 } else {
1595                         current->signal->nr_threads++;
1596                         atomic_inc(&current->signal->live);
1597                         atomic_inc(&current->signal->sigcnt);
1598                         list_add_tail_rcu(&p->thread_group,
1599                                           &p->group_leader->thread_group);
1600                         list_add_tail_rcu(&p->thread_node,
1601                                           &p->signal->thread_head);
1602                 }
1603                 attach_pid(p, PIDTYPE_PID);
1604                 nr_threads++;
1605         }
1606
1607         total_forks++;
1608         spin_unlock(&current->sighand->siglock);
1609         syscall_tracepoint_update(p);
1610         write_unlock_irq(&tasklist_lock);
1611
1612         proc_fork_connector(p);
1613         cgroup_post_fork(p, cgrp_ss_priv);
1614         threadgroup_change_end(current);
1615         perf_event_fork(p);
1616
1617         trace_task_newtask(p, clone_flags);
1618         uprobe_copy_process(p, clone_flags);
1619
1620         return p;
1621
1622 bad_fork_cancel_cgroup:
1623         cgroup_cancel_fork(p, cgrp_ss_priv);
1624 bad_fork_free_pid:
1625         if (pid != &init_struct_pid)
1626                 free_pid(pid);
1627 bad_fork_cleanup_io:
1628         if (p->io_context)
1629                 exit_io_context(p);
1630 bad_fork_cleanup_namespaces:
1631         exit_task_namespaces(p);
1632 bad_fork_cleanup_mm:
1633         if (p->mm)
1634                 mmput(p->mm);
1635 bad_fork_cleanup_signal:
1636         if (!(clone_flags & CLONE_THREAD))
1637                 free_signal_struct(p->signal);
1638 bad_fork_cleanup_sighand:
1639         __cleanup_sighand(p->sighand);
1640 bad_fork_cleanup_fs:
1641         exit_fs(p); /* blocking */
1642 bad_fork_cleanup_files:
1643         exit_files(p); /* blocking */
1644 bad_fork_cleanup_semundo:
1645         exit_sem(p);
1646 bad_fork_cleanup_audit:
1647         audit_free(p);
1648 bad_fork_cleanup_perf:
1649         perf_event_free_task(p);
1650 bad_fork_cleanup_policy:
1651 #ifdef CONFIG_NUMA
1652         mpol_put(p->mempolicy);
1653 bad_fork_cleanup_threadgroup_lock:
1654 #endif
1655         threadgroup_change_end(current);
1656         delayacct_tsk_free(p);
1657 bad_fork_cleanup_count:
1658         atomic_dec(&p->cred->user->processes);
1659         exit_creds(p);
1660 bad_fork_free:
1661         free_task(p);
1662 fork_out:
1663         return ERR_PTR(retval);
1664 }
1665
1666 static inline void init_idle_pids(struct pid_link *links)
1667 {
1668         enum pid_type type;
1669
1670         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1671                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1672                 links[type].pid = &init_struct_pid;
1673         }
1674 }
1675
1676 struct task_struct *fork_idle(int cpu)
1677 {
1678         struct task_struct *task;
1679         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0);
1680         if (!IS_ERR(task)) {
1681                 init_idle_pids(task->pids);
1682                 init_idle(task, cpu);
1683         }
1684
1685         return task;
1686 }
1687
1688 /*
1689  *  Ok, this is the main fork-routine.
1690  *
1691  * It copies the process, and if successful kick-starts
1692  * it and waits for it to finish using the VM if required.
1693  */
1694 long _do_fork(unsigned long clone_flags,
1695               unsigned long stack_start,
1696               unsigned long stack_size,
1697               int __user *parent_tidptr,
1698               int __user *child_tidptr,
1699               unsigned long tls)
1700 {
1701         struct task_struct *p;
1702         int trace = 0;
1703         long nr;
1704
1705         /*
1706          * Determine whether and which event to report to ptracer.  When
1707          * called from kernel_thread or CLONE_UNTRACED is explicitly
1708          * requested, no event is reported; otherwise, report if the event
1709          * for the type of forking is enabled.
1710          */
1711         if (!(clone_flags & CLONE_UNTRACED)) {
1712                 if (clone_flags & CLONE_VFORK)
1713                         trace = PTRACE_EVENT_VFORK;
1714                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1715                         trace = PTRACE_EVENT_CLONE;
1716                 else
1717                         trace = PTRACE_EVENT_FORK;
1718
1719                 if (likely(!ptrace_event_enabled(current, trace)))
1720                         trace = 0;
1721         }
1722
1723         p = copy_process(clone_flags, stack_start, stack_size,
1724                          child_tidptr, NULL, trace, tls);
1725         /*
1726          * Do this prior waking up the new thread - the thread pointer
1727          * might get invalid after that point, if the thread exits quickly.
1728          */
1729         if (!IS_ERR(p)) {
1730                 struct completion vfork;
1731                 struct pid *pid;
1732
1733                 trace_sched_process_fork(current, p);
1734
1735                 pid = get_task_pid(p, PIDTYPE_PID);
1736                 nr = pid_vnr(pid);
1737
1738                 if (clone_flags & CLONE_PARENT_SETTID)
1739                         put_user(nr, parent_tidptr);
1740
1741                 if (clone_flags & CLONE_VFORK) {
1742                         p->vfork_done = &vfork;
1743                         init_completion(&vfork);
1744                         get_task_struct(p);
1745                 }
1746
1747                 wake_up_new_task(p);
1748
1749                 /* forking complete and child started to run, tell ptracer */
1750                 if (unlikely(trace))
1751                         ptrace_event_pid(trace, pid);
1752
1753                 if (clone_flags & CLONE_VFORK) {
1754                         if (!wait_for_vfork_done(p, &vfork))
1755                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1756                 }
1757
1758                 put_pid(pid);
1759         } else {
1760                 nr = PTR_ERR(p);
1761         }
1762         return nr;
1763 }
1764
1765 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1766 /* For compatibility with architectures that call do_fork directly rather than
1767  * using the syscall entry points below. */
1768 long do_fork(unsigned long clone_flags,
1769               unsigned long stack_start,
1770               unsigned long stack_size,
1771               int __user *parent_tidptr,
1772               int __user *child_tidptr)
1773 {
1774         return _do_fork(clone_flags, stack_start, stack_size,
1775                         parent_tidptr, child_tidptr, 0);
1776 }
1777 #endif
1778
1779 /*
1780  * Create a kernel thread.
1781  */
1782 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1783 {
1784         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1785                 (unsigned long)arg, NULL, NULL, 0);
1786 }
1787
1788 #ifdef __ARCH_WANT_SYS_FORK
1789 SYSCALL_DEFINE0(fork)
1790 {
1791 #ifdef CONFIG_MMU
1792         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1793 #else
1794         /* can not support in nommu mode */
1795         return -EINVAL;
1796 #endif
1797 }
1798 #endif
1799
1800 #ifdef __ARCH_WANT_SYS_VFORK
1801 SYSCALL_DEFINE0(vfork)
1802 {
1803         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1804                         0, NULL, NULL, 0);
1805 }
1806 #endif
1807
1808 #ifdef __ARCH_WANT_SYS_CLONE
1809 #ifdef CONFIG_CLONE_BACKWARDS
1810 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1811                  int __user *, parent_tidptr,
1812                  unsigned long, tls,
1813                  int __user *, child_tidptr)
1814 #elif defined(CONFIG_CLONE_BACKWARDS2)
1815 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1816                  int __user *, parent_tidptr,
1817                  int __user *, child_tidptr,
1818                  unsigned long, tls)
1819 #elif defined(CONFIG_CLONE_BACKWARDS3)
1820 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1821                 int, stack_size,
1822                 int __user *, parent_tidptr,
1823                 int __user *, child_tidptr,
1824                 unsigned long, tls)
1825 #else
1826 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1827                  int __user *, parent_tidptr,
1828                  int __user *, child_tidptr,
1829                  unsigned long, tls)
1830 #endif
1831 {
1832         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
1833 }
1834 #endif
1835
1836 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1837 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1838 #endif
1839
1840 static void sighand_ctor(void *data)
1841 {
1842         struct sighand_struct *sighand = data;
1843
1844         spin_lock_init(&sighand->siglock);
1845         init_waitqueue_head(&sighand->signalfd_wqh);
1846 }
1847
1848 void __init proc_caches_init(void)
1849 {
1850         sighand_cachep = kmem_cache_create("sighand_cache",
1851                         sizeof(struct sighand_struct), 0,
1852                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1853                         SLAB_NOTRACK, sighand_ctor);
1854         signal_cachep = kmem_cache_create("signal_cache",
1855                         sizeof(struct signal_struct), 0,
1856                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1857         files_cachep = kmem_cache_create("files_cache",
1858                         sizeof(struct files_struct), 0,
1859                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1860         fs_cachep = kmem_cache_create("fs_cache",
1861                         sizeof(struct fs_struct), 0,
1862                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1863         /*
1864          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1865          * whole struct cpumask for the OFFSTACK case. We could change
1866          * this to *only* allocate as much of it as required by the
1867          * maximum number of CPU's we can ever have.  The cpumask_allocation
1868          * is at the end of the structure, exactly for that reason.
1869          */
1870         mm_cachep = kmem_cache_create("mm_struct",
1871                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1872                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1873         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1874         mmap_init();
1875         nsproxy_cache_init();
1876 }
1877
1878 /*
1879  * Check constraints on flags passed to the unshare system call.
1880  */
1881 static int check_unshare_flags(unsigned long unshare_flags)
1882 {
1883         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1884                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1885                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1886                                 CLONE_NEWUSER|CLONE_NEWPID))
1887                 return -EINVAL;
1888         /*
1889          * Not implemented, but pretend it works if there is nothing
1890          * to unshare.  Note that unsharing the address space or the
1891          * signal handlers also need to unshare the signal queues (aka
1892          * CLONE_THREAD).
1893          */
1894         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1895                 if (!thread_group_empty(current))
1896                         return -EINVAL;
1897         }
1898         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
1899                 if (atomic_read(&current->sighand->count) > 1)
1900                         return -EINVAL;
1901         }
1902         if (unshare_flags & CLONE_VM) {
1903                 if (!current_is_single_threaded())
1904                         return -EINVAL;
1905         }
1906
1907         return 0;
1908 }
1909
1910 /*
1911  * Unshare the filesystem structure if it is being shared
1912  */
1913 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1914 {
1915         struct fs_struct *fs = current->fs;
1916
1917         if (!(unshare_flags & CLONE_FS) || !fs)
1918                 return 0;
1919
1920         /* don't need lock here; in the worst case we'll do useless copy */
1921         if (fs->users == 1)
1922                 return 0;
1923
1924         *new_fsp = copy_fs_struct(fs);
1925         if (!*new_fsp)
1926                 return -ENOMEM;
1927
1928         return 0;
1929 }
1930
1931 /*
1932  * Unshare file descriptor table if it is being shared
1933  */
1934 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1935 {
1936         struct files_struct *fd = current->files;
1937         int error = 0;
1938
1939         if ((unshare_flags & CLONE_FILES) &&
1940             (fd && atomic_read(&fd->count) > 1)) {
1941                 *new_fdp = dup_fd(fd, &error);
1942                 if (!*new_fdp)
1943                         return error;
1944         }
1945
1946         return 0;
1947 }
1948
1949 /*
1950  * unshare allows a process to 'unshare' part of the process
1951  * context which was originally shared using clone.  copy_*
1952  * functions used by do_fork() cannot be used here directly
1953  * because they modify an inactive task_struct that is being
1954  * constructed. Here we are modifying the current, active,
1955  * task_struct.
1956  */
1957 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1958 {
1959         struct fs_struct *fs, *new_fs = NULL;
1960         struct files_struct *fd, *new_fd = NULL;
1961         struct cred *new_cred = NULL;
1962         struct nsproxy *new_nsproxy = NULL;
1963         int do_sysvsem = 0;
1964         int err;
1965
1966         /*
1967          * If unsharing a user namespace must also unshare the thread group
1968          * and unshare the filesystem root and working directories.
1969          */
1970         if (unshare_flags & CLONE_NEWUSER)
1971                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1972         /*
1973          * If unsharing vm, must also unshare signal handlers.
1974          */
1975         if (unshare_flags & CLONE_VM)
1976                 unshare_flags |= CLONE_SIGHAND;
1977         /*
1978          * If unsharing a signal handlers, must also unshare the signal queues.
1979          */
1980         if (unshare_flags & CLONE_SIGHAND)
1981                 unshare_flags |= CLONE_THREAD;
1982         /*
1983          * If unsharing namespace, must also unshare filesystem information.
1984          */
1985         if (unshare_flags & CLONE_NEWNS)
1986                 unshare_flags |= CLONE_FS;
1987
1988         err = check_unshare_flags(unshare_flags);
1989         if (err)
1990                 goto bad_unshare_out;
1991         /*
1992          * CLONE_NEWIPC must also detach from the undolist: after switching
1993          * to a new ipc namespace, the semaphore arrays from the old
1994          * namespace are unreachable.
1995          */
1996         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1997                 do_sysvsem = 1;
1998         err = unshare_fs(unshare_flags, &new_fs);
1999         if (err)
2000                 goto bad_unshare_out;
2001         err = unshare_fd(unshare_flags, &new_fd);
2002         if (err)
2003                 goto bad_unshare_cleanup_fs;
2004         err = unshare_userns(unshare_flags, &new_cred);
2005         if (err)
2006                 goto bad_unshare_cleanup_fd;
2007         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2008                                          new_cred, new_fs);
2009         if (err)
2010                 goto bad_unshare_cleanup_cred;
2011
2012         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2013                 if (do_sysvsem) {
2014                         /*
2015                          * CLONE_SYSVSEM is equivalent to sys_exit().
2016                          */
2017                         exit_sem(current);
2018                 }
2019                 if (unshare_flags & CLONE_NEWIPC) {
2020                         /* Orphan segments in old ns (see sem above). */
2021                         exit_shm(current);
2022                         shm_init_task(current);
2023                 }
2024
2025                 if (new_nsproxy)
2026                         switch_task_namespaces(current, new_nsproxy);
2027
2028                 task_lock(current);
2029
2030                 if (new_fs) {
2031                         fs = current->fs;
2032                         spin_lock(&fs->lock);
2033                         current->fs = new_fs;
2034                         if (--fs->users)
2035                                 new_fs = NULL;
2036                         else
2037                                 new_fs = fs;
2038                         spin_unlock(&fs->lock);
2039                 }
2040
2041                 if (new_fd) {
2042                         fd = current->files;
2043                         current->files = new_fd;
2044                         new_fd = fd;
2045                 }
2046
2047                 task_unlock(current);
2048
2049                 if (new_cred) {
2050                         /* Install the new user namespace */
2051                         commit_creds(new_cred);
2052                         new_cred = NULL;
2053                 }
2054         }
2055
2056 bad_unshare_cleanup_cred:
2057         if (new_cred)
2058                 put_cred(new_cred);
2059 bad_unshare_cleanup_fd:
2060         if (new_fd)
2061                 put_files_struct(new_fd);
2062
2063 bad_unshare_cleanup_fs:
2064         if (new_fs)
2065                 free_fs_struct(new_fs);
2066
2067 bad_unshare_out:
2068         return err;
2069 }
2070
2071 /*
2072  *      Helper to unshare the files of the current task.
2073  *      We don't want to expose copy_files internals to
2074  *      the exec layer of the kernel.
2075  */
2076
2077 int unshare_files(struct files_struct **displaced)
2078 {
2079         struct task_struct *task = current;
2080         struct files_struct *copy = NULL;
2081         int error;
2082
2083         error = unshare_fd(CLONE_FILES, &copy);
2084         if (error || !copy) {
2085                 *displaced = NULL;
2086                 return error;
2087         }
2088         *displaced = task->files;
2089         task_lock(task);
2090         task->files = copy;
2091         task_unlock(task);
2092         return 0;
2093 }
2094
2095 int sysctl_max_threads(struct ctl_table *table, int write,
2096                        void __user *buffer, size_t *lenp, loff_t *ppos)
2097 {
2098         struct ctl_table t;
2099         int ret;
2100         int threads = max_threads;
2101         int min = MIN_THREADS;
2102         int max = MAX_THREADS;
2103
2104         t = *table;
2105         t.data = &threads;
2106         t.extra1 = &min;
2107         t.extra2 = &max;
2108
2109         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2110         if (ret || !write)
2111                 return ret;
2112
2113         set_max_threads(threads);
2114
2115         return 0;
2116 }