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