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() and sys_execve(). Callers prevent concurrent
 715  * invocations: in mmput() nobody alive left, in execve task is single
 716  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 717  * mm->exe_file, but does so without using set_mm_exe_file() in order
 718  * to do avoid the need for any locks.
 719  */
 720 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 721 {
 722         struct file *old_exe_file;
 723
 724         /*
 725          * It is safe to dereference the exe_file without RCU as
 726          * this function is only called if nobody else can access
 727          * this mm -- see comment above for justification.
 728          */
 729         old_exe_file = rcu_dereference_raw(mm->exe_file);
 730
 731         if (new_exe_file)
 732                 get_file(new_exe_file);
 733         rcu_assign_pointer(mm->exe_file, new_exe_file);
 734         if (old_exe_file)
 735                 fput(old_exe_file);
 736 }
 737
 738 /**
 739  * get_mm_exe_file - acquire a reference to the mm's executable file
 740  *
 741  * Returns %NULL if mm has no associated executable file.
 742  * User must release file via fput().
 743  */
 744 struct file *get_mm_exe_file(struct mm_struct *mm)
 745 {
 746         struct file *exe_file;
 747
 748         rcu_read_lock();
 749         exe_file = rcu_dereference(mm->exe_file);
 750         if (exe_file && !get_file_rcu(exe_file))
 751                 exe_file = NULL;
 752         rcu_read_unlock();
 753         return exe_file;
 754 }
 755
 756 /**
 757  * get_task_mm - acquire a reference to the task's mm
 758  *
 759  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 760  * this kernel workthread has transiently adopted a user mm with use_mm,
 761  * to do its AIO) is not set and if so returns a reference to it, after
 762  * bumping up the use count.  User must release the mm via mmput()
 763  * after use.  Typically used by /proc and ptrace.
 764  */
 765 struct mm_struct *get_task_mm(struct task_struct *task)
 766 {
 767         struct mm_struct *mm;
 768
 769         task_lock(task);
 770         mm = task->mm;
 771         if (mm) {
 772                 if (task->flags & PF_KTHREAD)
 773                         mm = NULL;
 774                 else
 775                         atomic_inc(&mm->mm_users);
 776         }
 777         task_unlock(task);
 778         return mm;
 779 }
 780 EXPORT_SYMBOL_GPL(get_task_mm);
 781
 782 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 783 {
 784         struct mm_struct *mm;
 785         int err;
 786
 787         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 788         if (err)
 789                 return ERR_PTR(err);
 790
 791         mm = get_task_mm(task);
 792         if (mm && mm != current->mm &&
 793                         !ptrace_may_access(task, mode)) {
 794                 mmput(mm);
 795                 mm = ERR_PTR(-EACCES);
 796         }
 797         mutex_unlock(&task->signal->cred_guard_mutex);
 798
 799         return mm;
 800 }
 801
 802 static void complete_vfork_done(struct task_struct *tsk)
 803 {
 804         struct completion *vfork;
 805
 806         task_lock(tsk);
 807         vfork = tsk->vfork_done;
 808         if (likely(vfork)) {
 809                 tsk->vfork_done = NULL;
 810                 complete(vfork);
 811         }
 812         task_unlock(tsk);
 813 }
 814
 815 static int wait_for_vfork_done(struct task_struct *child,
 816                                 struct completion *vfork)
 817 {
 818         int killed;
 819
 820         freezer_do_not_count();
 821         killed = wait_for_completion_killable(vfork);
 822         freezer_count();
 823
 824         if (killed) {
 825                 task_lock(child);
 826                 child->vfork_done = NULL;
 827                 task_unlock(child);
 828         }
 829
 830         put_task_struct(child);
 831         return killed;
 832 }
 833
 834 /* Please note the differences between mmput and mm_release.
 835  * mmput is called whenever we stop holding onto a mm_struct,
 836  * error success whatever.
 837  *
 838  * mm_release is called after a mm_struct has been removed
 839  * from the current process.
 840  *
 841  * This difference is important for error handling, when we
 842  * only half set up a mm_struct for a new process and need to restore
 843  * the old one.  Because we mmput the new mm_struct before
 844  * restoring the old one. . .
 845  * Eric Biederman 10 January 1998
 846  */
 847 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 848 {
 849         /* Get rid of any futexes when releasing the mm */
 850 #ifdef CONFIG_FUTEX
 851         if (unlikely(tsk->robust_list)) {
 852                 exit_robust_list(tsk);
 853                 tsk->robust_list = NULL;
 854         }
 855 #ifdef CONFIG_COMPAT
 856         if (unlikely(tsk->compat_robust_list)) {
 857                 compat_exit_robust_list(tsk);
 858                 tsk->compat_robust_list = NULL;
 859         }
 860 #endif
 861         if (unlikely(!list_empty(&tsk->pi_state_list)))
 862                 exit_pi_state_list(tsk);
 863 #endif
 864
 865         uprobe_free_utask(tsk);
 866
 867         /* Get rid of any cached register state */
 868         deactivate_mm(tsk, mm);
 869
 870         /*
 871          * If we're exiting normally, clear a user-space tid field if
 872          * requested.  We leave this alone when dying by signal, to leave
 873          * the value intact in a core dump, and to save the unnecessary
 874          * trouble, say, a killed vfork parent shouldn't touch this mm.
 875          * Userland only wants this done for a sys_exit.
 876          */
 877         if (tsk->clear_child_tid) {
 878                 if (!(tsk->flags & PF_SIGNALED) &&
 879                     atomic_read(&mm->mm_users) > 1) {
 880                         /*
 881                          * We don't check the error code - if userspace has
 882                          * not set up a proper pointer then tough luck.
 883                          */
 884                         put_user(0, tsk->clear_child_tid);
 885                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 886                                         1, NULL, NULL, 0);
 887                 }
 888                 tsk->clear_child_tid = NULL;
 889         }
 890
 891         /*
 892          * All done, finally we can wake up parent and return this mm to him.
 893          * Also kthread_stop() uses this completion for synchronization.
 894          */
 895         if (tsk->vfork_done)
 896                 complete_vfork_done(tsk);
 897 }
 898
 899 /*
 900  * Allocate a new mm structure and copy contents from the
 901  * mm structure of the passed in task structure.
 902  */
 903 static struct mm_struct *dup_mm(struct task_struct *tsk)
 904 {
 905         struct mm_struct *mm, *oldmm = current->mm;
 906         int err;
 907
 908         mm = allocate_mm();
 909         if (!mm)
 910                 goto fail_nomem;
 911
 912         memcpy(mm, oldmm, sizeof(*mm));
 913
 914         if (!mm_init(mm, tsk))
 915                 goto fail_nomem;
 916
 917         err = dup_mmap(mm, oldmm);
 918         if (err)
 919                 goto free_pt;
 920
 921         mm->hiwater_rss = get_mm_rss(mm);
 922         mm->hiwater_vm = mm->total_vm;
 923
 924         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 925                 goto free_pt;
 926
 927         return mm;
 928
 929 free_pt:
 930         /* don't put binfmt in mmput, we haven't got module yet */
 931         mm->binfmt = NULL;
 932         mmput(mm);
 933
 934 fail_nomem:
 935         return NULL;
 936 }
 937
 938 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 939 {
 940         struct mm_struct *mm, *oldmm;
 941         int retval;
 942
 943         tsk->min_flt = tsk->maj_flt = 0;
 944         tsk->nvcsw = tsk->nivcsw = 0;
 945 #ifdef CONFIG_DETECT_HUNG_TASK
 946         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 947 #endif
 948
 949         tsk->mm = NULL;
 950         tsk->active_mm = NULL;
 951
 952         /*
 953          * Are we cloning a kernel thread?
 954          *
 955          * We need to steal a active VM for that..
 956          */
 957         oldmm = current->mm;
 958         if (!oldmm)
 959                 return 0;
 960
 961         /* initialize the new vmacache entries */
 962         vmacache_flush(tsk);
 963
 964         if (clone_flags & CLONE_VM) {
 965                 atomic_inc(&oldmm->mm_users);
 966                 mm = oldmm;
 967                 goto good_mm;
 968         }
 969
 970         retval = -ENOMEM;
 971         mm = dup_mm(tsk);
 972         if (!mm)
 973                 goto fail_nomem;
 974
 975 good_mm:
 976         tsk->mm = mm;
 977         tsk->active_mm = mm;
 978         return 0;
 979
 980 fail_nomem:
 981         return retval;
 982 }
 983
 984 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 985 {
 986         struct fs_struct *fs = current->fs;
 987         if (clone_flags & CLONE_FS) {
 988                 /* tsk->fs is already what we want */
 989                 spin_lock(&fs->lock);
 990                 if (fs->in_exec) {
 991                         spin_unlock(&fs->lock);
 992                         return -EAGAIN;
 993                 }
 994                 fs->users++;
 995                 spin_unlock(&fs->lock);
 996                 return 0;
 997         }
 998         tsk->fs = copy_fs_struct(fs);
 999         if (!tsk->fs)
1000                 return -ENOMEM;
1001         return 0;
1002 }
1003
1004 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1005 {
1006         struct files_struct *oldf, *newf;
1007         int error = 0;
1008
1009         /*
1010          * A background process may not have any files ...
1011          */
1012         oldf = current->files;
1013         if (!oldf)
1014                 goto out;
1015
1016         if (clone_flags & CLONE_FILES) {
1017                 atomic_inc(&oldf->count);
1018                 goto out;
1019         }
1020
1021         newf = dup_fd(oldf, &error);
1022         if (!newf)
1023                 goto out;
1024
1025         tsk->files = newf;
1026         error = 0;
1027 out:
1028         return error;
1029 }
1030
1031 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1032 {
1033 #ifdef CONFIG_BLOCK
1034         struct io_context *ioc = current->io_context;
1035         struct io_context *new_ioc;
1036
1037         if (!ioc)
1038                 return 0;
1039         /*
1040          * Share io context with parent, if CLONE_IO is set
1041          */
1042         if (clone_flags & CLONE_IO) {
1043                 ioc_task_link(ioc);
1044                 tsk->io_context = ioc;
1045         } else if (ioprio_valid(ioc->ioprio)) {
1046                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1047                 if (unlikely(!new_ioc))
1048                         return -ENOMEM;
1049
1050                 new_ioc->ioprio = ioc->ioprio;
1051                 put_io_context(new_ioc);
1052         }
1053 #endif
1054         return 0;
1055 }
1056
1057 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1058 {
1059         struct sighand_struct *sig;
1060
1061         if (clone_flags & CLONE_SIGHAND) {
1062                 atomic_inc(&current->sighand->count);
1063                 return 0;
1064         }
1065         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1066         rcu_assign_pointer(tsk->sighand, sig);
1067         if (!sig)
1068                 return -ENOMEM;
1069         atomic_set(&sig->count, 1);
1070         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1071         return 0;
1072 }
1073
1074 void __cleanup_sighand(struct sighand_struct *sighand)
1075 {
1076         if (atomic_dec_and_test(&sighand->count)) {
1077                 signalfd_cleanup(sighand);
1078                 /*
1079                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1080                  * without an RCU grace period, see __lock_task_sighand().
1081                  */
1082                 kmem_cache_free(sighand_cachep, sighand);
1083         }
1084 }
1085
1086 /*
1087  * Initialize POSIX timer handling for a thread group.
1088  */
1089 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1090 {
1091         unsigned long cpu_limit;
1092
1093         /* Thread group counters. */
1094         thread_group_cputime_init(sig);
1095
1096         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1097         if (cpu_limit != RLIM_INFINITY) {
1098                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1099                 sig->cputimer.running = 1;
1100         }
1101
1102         /* The timer lists. */
1103         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1104         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1105         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1106 }
1107
1108 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1109 {
1110         struct signal_struct *sig;
1111
1112         if (clone_flags & CLONE_THREAD)
1113                 return 0;
1114
1115         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1116         tsk->signal = sig;
1117         if (!sig)
1118                 return -ENOMEM;
1119
1120         sig->nr_threads = 1;
1121         atomic_set(&sig->live, 1);
1122         atomic_set(&sig->sigcnt, 1);
1123
1124         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1125         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1126         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1127
1128         init_waitqueue_head(&sig->wait_chldexit);
1129         sig->curr_target = tsk;
1130         init_sigpending(&sig->shared_pending);
1131         INIT_LIST_HEAD(&sig->posix_timers);
1132         seqlock_init(&sig->stats_lock);
1133
1134         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1135         sig->real_timer.function = it_real_fn;
1136
1137         task_lock(current->group_leader);
1138         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1139         task_unlock(current->group_leader);
1140
1141         posix_cpu_timers_init_group(sig);
1142
1143         tty_audit_fork(sig);
1144         sched_autogroup_fork(sig);
1145
1146 #ifdef CONFIG_CGROUPS
1147         init_rwsem(&sig->group_rwsem);
1148 #endif
1149
1150         sig->oom_score_adj = current->signal->oom_score_adj;
1151         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1152
1153         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1154                                    current->signal->is_child_subreaper;
1155
1156         mutex_init(&sig->cred_guard_mutex);
1157
1158         return 0;
1159 }
1160
1161 static void copy_seccomp(struct task_struct *p)
1162 {
1163 #ifdef CONFIG_SECCOMP
1164         /*
1165          * Must be called with sighand->lock held, which is common to
1166          * all threads in the group. Holding cred_guard_mutex is not
1167          * needed because this new task is not yet running and cannot
1168          * be racing exec.
1169          */
1170         assert_spin_locked(&current->sighand->siglock);
1171
1172         /* Ref-count the new filter user, and assign it. */
1173         get_seccomp_filter(current);
1174         p->seccomp = current->seccomp;
1175
1176         /*
1177          * Explicitly enable no_new_privs here in case it got set
1178          * between the task_struct being duplicated and holding the
1179          * sighand lock. The seccomp state and nnp must be in sync.
1180          */
1181         if (task_no_new_privs(current))
1182                 task_set_no_new_privs(p);
1183
1184         /*
1185          * If the parent gained a seccomp mode after copying thread
1186          * flags and between before we held the sighand lock, we have
1187          * to manually enable the seccomp thread flag here.
1188          */
1189         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1190                 set_tsk_thread_flag(p, TIF_SECCOMP);
1191 #endif
1192 }
1193
1194 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1195 {
1196         current->clear_child_tid = tidptr;
1197
1198         return task_pid_vnr(current);
1199 }
1200
1201 static void rt_mutex_init_task(struct task_struct *p)
1202 {
1203         raw_spin_lock_init(&p->pi_lock);
1204 #ifdef CONFIG_RT_MUTEXES
1205         p->pi_waiters = RB_ROOT;
1206         p->pi_waiters_leftmost = NULL;
1207         p->pi_blocked_on = NULL;
1208 #endif
1209 }
1210
1211 /*
1212  * Initialize POSIX timer handling for a single task.
1213  */
1214 static void posix_cpu_timers_init(struct task_struct *tsk)
1215 {
1216         tsk->cputime_expires.prof_exp = 0;
1217         tsk->cputime_expires.virt_exp = 0;
1218         tsk->cputime_expires.sched_exp = 0;
1219         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1220         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1221         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1222 }
1223
1224 static inline void
1225 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1226 {
1227          task->pids[type].pid = pid;
1228 }
1229
1230 /*
1231  * This creates a new process as a copy of the old one,
1232  * but does not actually start it yet.
1233  *
1234  * It copies the registers, and all the appropriate
1235  * parts of the process environment (as per the clone
1236  * flags). The actual kick-off is left to the caller.
1237  */
1238 static struct task_struct *copy_process(unsigned long clone_flags,
1239                                         unsigned long stack_start,
1240                                         unsigned long stack_size,
1241                                         int __user *child_tidptr,
1242                                         struct pid *pid,
1243                                         int trace)
1244 {
1245         int retval;
1246         struct task_struct *p;
1247
1248         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1249                 return ERR_PTR(-EINVAL);
1250
1251         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1252                 return ERR_PTR(-EINVAL);
1253
1254         /*
1255          * Thread groups must share signals as well, and detached threads
1256          * can only be started up within the thread group.
1257          */
1258         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1259                 return ERR_PTR(-EINVAL);
1260
1261         /*
1262          * Shared signal handlers imply shared VM. By way of the above,
1263          * thread groups also imply shared VM. Blocking this case allows
1264          * for various simplifications in other code.
1265          */
1266         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1267                 return ERR_PTR(-EINVAL);
1268
1269         /*
1270          * Siblings of global init remain as zombies on exit since they are
1271          * not reaped by their parent (swapper). To solve this and to avoid
1272          * multi-rooted process trees, prevent global and container-inits
1273          * from creating siblings.
1274          */
1275         if ((clone_flags & CLONE_PARENT) &&
1276                                 current->signal->flags & SIGNAL_UNKILLABLE)
1277                 return ERR_PTR(-EINVAL);
1278
1279         /*
1280          * If the new process will be in a different pid or user namespace
1281          * do not allow it to share a thread group or signal handlers or
1282          * parent with the forking task.
1283          */
1284         if (clone_flags & CLONE_SIGHAND) {
1285                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1286                     (task_active_pid_ns(current) !=
1287                                 current->nsproxy->pid_ns_for_children))
1288                         return ERR_PTR(-EINVAL);
1289         }
1290
1291         retval = security_task_create(clone_flags);
1292         if (retval)
1293                 goto fork_out;
1294
1295         retval = -ENOMEM;
1296         p = dup_task_struct(current);
1297         if (!p)
1298                 goto fork_out;
1299
1300         ftrace_graph_init_task(p);
1301
1302         rt_mutex_init_task(p);
1303
1304 #ifdef CONFIG_PROVE_LOCKING
1305         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1306         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1307 #endif
1308         retval = -EAGAIN;
1309         if (atomic_read(&p->real_cred->user->processes) >=
1310                         task_rlimit(p, RLIMIT_NPROC)) {
1311                 if (p->real_cred->user != INIT_USER &&
1312                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1313                         goto bad_fork_free;
1314         }
1315         current->flags &= ~PF_NPROC_EXCEEDED;
1316
1317         retval = copy_creds(p, clone_flags);
1318         if (retval < 0)
1319                 goto bad_fork_free;
1320
1321         /*
1322          * If multiple threads are within copy_process(), then this check
1323          * triggers too late. This doesn't hurt, the check is only there
1324          * to stop root fork bombs.
1325          */
1326         retval = -EAGAIN;
1327         if (nr_threads >= max_threads)
1328                 goto bad_fork_cleanup_count;
1329
1330         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1331         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1332         p->flags |= PF_FORKNOEXEC;
1333         INIT_LIST_HEAD(&p->children);
1334         INIT_LIST_HEAD(&p->sibling);
1335         rcu_copy_process(p);
1336         p->vfork_done = NULL;
1337         spin_lock_init(&p->alloc_lock);
1338
1339         init_sigpending(&p->pending);
1340
1341         p->utime = p->stime = p->gtime = 0;
1342         p->utimescaled = p->stimescaled = 0;
1343 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1344         p->prev_cputime.utime = p->prev_cputime.stime = 0;
1345 #endif
1346 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1347         seqlock_init(&p->vtime_seqlock);
1348         p->vtime_snap = 0;
1349         p->vtime_snap_whence = VTIME_SLEEPING;
1350 #endif
1351
1352 #if defined(SPLIT_RSS_COUNTING)
1353         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1354 #endif
1355
1356         p->default_timer_slack_ns = current->timer_slack_ns;
1357
1358         task_io_accounting_init(&p->ioac);
1359         acct_clear_integrals(p);
1360
1361         posix_cpu_timers_init(p);
1362
1363         p->start_time = ktime_get_ns();
1364         p->real_start_time = ktime_get_boot_ns();
1365         p->io_context = NULL;
1366         p->audit_context = NULL;
1367         if (clone_flags & CLONE_THREAD)
1368                 threadgroup_change_begin(current);
1369         cgroup_fork(p);
1370 #ifdef CONFIG_NUMA
1371         p->mempolicy = mpol_dup(p->mempolicy);
1372         if (IS_ERR(p->mempolicy)) {
1373                 retval = PTR_ERR(p->mempolicy);
1374                 p->mempolicy = NULL;
1375                 goto bad_fork_cleanup_threadgroup_lock;
1376         }
1377 #endif
1378 #ifdef CONFIG_CPUSETS
1379         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1380         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1381         seqcount_init(&p->mems_allowed_seq);
1382 #endif
1383 #ifdef CONFIG_TRACE_IRQFLAGS
1384         p->irq_events = 0;
1385         p->hardirqs_enabled = 0;
1386         p->hardirq_enable_ip = 0;
1387         p->hardirq_enable_event = 0;
1388         p->hardirq_disable_ip = _THIS_IP_;
1389         p->hardirq_disable_event = 0;
1390         p->softirqs_enabled = 1;
1391         p->softirq_enable_ip = _THIS_IP_;
1392         p->softirq_enable_event = 0;
1393         p->softirq_disable_ip = 0;
1394         p->softirq_disable_event = 0;
1395         p->hardirq_context = 0;
1396         p->softirq_context = 0;
1397 #endif
1398 #ifdef CONFIG_LOCKDEP
1399         p->lockdep_depth = 0; /* no locks held yet */
1400         p->curr_chain_key = 0;
1401         p->lockdep_recursion = 0;
1402 #endif
1403
1404 #ifdef CONFIG_DEBUG_MUTEXES
1405         p->blocked_on = NULL; /* not blocked yet */
1406 #endif
1407 #ifdef CONFIG_BCACHE
1408         p->sequential_io        = 0;
1409         p->sequential_io_avg    = 0;
1410 #endif
1411
1412         /* Perform scheduler related setup. Assign this task to a CPU. */
1413         retval = sched_fork(clone_flags, p);
1414         if (retval)
1415                 goto bad_fork_cleanup_policy;
1416
1417         retval = perf_event_init_task(p);
1418         if (retval)
1419                 goto bad_fork_cleanup_policy;
1420         retval = audit_alloc(p);
1421         if (retval)
1422                 goto bad_fork_cleanup_perf;
1423         /* copy all the process information */
1424         shm_init_task(p);
1425         retval = copy_semundo(clone_flags, p);
1426         if (retval)
1427                 goto bad_fork_cleanup_audit;
1428         retval = copy_files(clone_flags, p);
1429         if (retval)
1430                 goto bad_fork_cleanup_semundo;
1431         retval = copy_fs(clone_flags, p);
1432         if (retval)
1433                 goto bad_fork_cleanup_files;
1434         retval = copy_sighand(clone_flags, p);
1435         if (retval)
1436                 goto bad_fork_cleanup_fs;
1437         retval = copy_signal(clone_flags, p);
1438         if (retval)
1439                 goto bad_fork_cleanup_sighand;
1440         retval = copy_mm(clone_flags, p);
1441         if (retval)
1442                 goto bad_fork_cleanup_signal;
1443         retval = copy_namespaces(clone_flags, p);
1444         if (retval)
1445                 goto bad_fork_cleanup_mm;
1446         retval = copy_io(clone_flags, p);
1447         if (retval)
1448                 goto bad_fork_cleanup_namespaces;
1449         retval = copy_thread(clone_flags, stack_start, stack_size, p);
1450         if (retval)
1451                 goto bad_fork_cleanup_io;
1452
1453         if (pid != &init_struct_pid) {
1454                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1455                 if (IS_ERR(pid)) {
1456                         retval = PTR_ERR(pid);
1457                         goto bad_fork_cleanup_io;
1458                 }
1459         }
1460
1461         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1462         /*
1463          * Clear TID on mm_release()?
1464          */
1465         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1466 #ifdef CONFIG_BLOCK
1467         p->plug = NULL;
1468 #endif
1469 #ifdef CONFIG_FUTEX
1470         p->robust_list = NULL;
1471 #ifdef CONFIG_COMPAT
1472         p->compat_robust_list = NULL;
1473 #endif
1474         INIT_LIST_HEAD(&p->pi_state_list);
1475         p->pi_state_cache = NULL;
1476 #endif
1477         /*
1478          * sigaltstack should be cleared when sharing the same VM
1479          */
1480         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1481                 p->sas_ss_sp = p->sas_ss_size = 0;
1482
1483         /*
1484          * Syscall tracing and stepping should be turned off in the
1485          * child regardless of CLONE_PTRACE.
1486          */
1487         user_disable_single_step(p);
1488         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1489 #ifdef TIF_SYSCALL_EMU
1490         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1491 #endif
1492         clear_all_latency_tracing(p);
1493
1494         /* ok, now we should be set up.. */
1495         p->pid = pid_nr(pid);
1496         if (clone_flags & CLONE_THREAD) {
1497                 p->exit_signal = -1;
1498                 p->group_leader = current->group_leader;
1499                 p->tgid = current->tgid;
1500         } else {
1501                 if (clone_flags & CLONE_PARENT)
1502                         p->exit_signal = current->group_leader->exit_signal;
1503                 else
1504                         p->exit_signal = (clone_flags & CSIGNAL);
1505                 p->group_leader = p;
1506                 p->tgid = p->pid;
1507         }
1508
1509         p->nr_dirtied = 0;
1510         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1511         p->dirty_paused_when = 0;
1512
1513         p->pdeath_signal = 0;
1514         INIT_LIST_HEAD(&p->thread_group);
1515         p->task_works = NULL;
1516
1517         /*
1518          * Make it visible to the rest of the system, but dont wake it up yet.
1519          * Need tasklist lock for parent etc handling!
1520          */
1521         write_lock_irq(&tasklist_lock);
1522
1523         /* CLONE_PARENT re-uses the old parent */
1524         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1525                 p->real_parent = current->real_parent;
1526                 p->parent_exec_id = current->parent_exec_id;
1527         } else {
1528                 p->real_parent = current;
1529                 p->parent_exec_id = current->self_exec_id;
1530         }
1531
1532         spin_lock(&current->sighand->siglock);
1533
1534         /*
1535          * Copy seccomp details explicitly here, in case they were changed
1536          * before holding sighand lock.
1537          */
1538         copy_seccomp(p);
1539
1540         /*
1541          * Process group and session signals need to be delivered to just the
1542          * parent before the fork or both the parent and the child after the
1543          * fork. Restart if a signal comes in before we add the new process to
1544          * it's process group.
1545          * A fatal signal pending means that current will exit, so the new
1546          * thread can't slip out of an OOM kill (or normal SIGKILL).
1547         */
1548         recalc_sigpending();
1549         if (signal_pending(current)) {
1550                 spin_unlock(&current->sighand->siglock);
1551                 write_unlock_irq(&tasklist_lock);
1552                 retval = -ERESTARTNOINTR;
1553                 goto bad_fork_free_pid;
1554         }
1555
1556         if (likely(p->pid)) {
1557                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1558
1559                 init_task_pid(p, PIDTYPE_PID, pid);
1560                 if (thread_group_leader(p)) {
1561                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1562                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1563
1564                         if (is_child_reaper(pid)) {
1565                                 ns_of_pid(pid)->child_reaper = p;
1566                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1567                         }
1568
1569                         p->signal->leader_pid = pid;
1570                         p->signal->tty = tty_kref_get(current->signal->tty);
1571                         list_add_tail(&p->sibling, &p->real_parent->children);
1572                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1573                         attach_pid(p, PIDTYPE_PGID);
1574                         attach_pid(p, PIDTYPE_SID);
1575                         __this_cpu_inc(process_counts);
1576                 } else {
1577                         current->signal->nr_threads++;
1578                         atomic_inc(&current->signal->live);
1579                         atomic_inc(&current->signal->sigcnt);
1580                         list_add_tail_rcu(&p->thread_group,
1581                                           &p->group_leader->thread_group);
1582                         list_add_tail_rcu(&p->thread_node,
1583                                           &p->signal->thread_head);
1584                 }
1585                 attach_pid(p, PIDTYPE_PID);
1586                 nr_threads++;
1587         }
1588
1589         total_forks++;
1590         spin_unlock(&current->sighand->siglock);
1591         syscall_tracepoint_update(p);
1592         write_unlock_irq(&tasklist_lock);
1593
1594         proc_fork_connector(p);
1595         cgroup_post_fork(p);
1596         if (clone_flags & CLONE_THREAD)
1597                 threadgroup_change_end(current);
1598         perf_event_fork(p);
1599
1600         trace_task_newtask(p, clone_flags);
1601         uprobe_copy_process(p, clone_flags);
1602
1603         return p;
1604
1605 bad_fork_free_pid:
1606         if (pid != &init_struct_pid)
1607                 free_pid(pid);
1608 bad_fork_cleanup_io:
1609         if (p->io_context)
1610                 exit_io_context(p);
1611 bad_fork_cleanup_namespaces:
1612         exit_task_namespaces(p);
1613 bad_fork_cleanup_mm:
1614         if (p->mm)
1615                 mmput(p->mm);
1616 bad_fork_cleanup_signal:
1617         if (!(clone_flags & CLONE_THREAD))
1618                 free_signal_struct(p->signal);
1619 bad_fork_cleanup_sighand:
1620         __cleanup_sighand(p->sighand);
1621 bad_fork_cleanup_fs:
1622         exit_fs(p); /* blocking */
1623 bad_fork_cleanup_files:
1624         exit_files(p); /* blocking */
1625 bad_fork_cleanup_semundo:
1626         exit_sem(p);
1627 bad_fork_cleanup_audit:
1628         audit_free(p);
1629 bad_fork_cleanup_perf:
1630         perf_event_free_task(p);
1631 bad_fork_cleanup_policy:
1632 #ifdef CONFIG_NUMA
1633         mpol_put(p->mempolicy);
1634 bad_fork_cleanup_threadgroup_lock:
1635 #endif
1636         if (clone_flags & CLONE_THREAD)
1637                 threadgroup_change_end(current);
1638         delayacct_tsk_free(p);
1639 bad_fork_cleanup_count:
1640         atomic_dec(&p->cred->user->processes);
1641         exit_creds(p);
1642 bad_fork_free:
1643         free_task(p);
1644 fork_out:
1645         return ERR_PTR(retval);
1646 }
1647
1648 static inline void init_idle_pids(struct pid_link *links)
1649 {
1650         enum pid_type type;
1651
1652         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1653                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1654                 links[type].pid = &init_struct_pid;
1655         }
1656 }
1657
1658 struct task_struct *fork_idle(int cpu)
1659 {
1660         struct task_struct *task;
1661         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1662         if (!IS_ERR(task)) {
1663                 init_idle_pids(task->pids);
1664                 init_idle(task, cpu);
1665         }
1666
1667         return task;
1668 }
1669
1670 /*
1671  *  Ok, this is the main fork-routine.
1672  *
1673  * It copies the process, and if successful kick-starts
1674  * it and waits for it to finish using the VM if required.
1675  */
1676 long do_fork(unsigned long clone_flags,
1677               unsigned long stack_start,
1678               unsigned long stack_size,
1679               int __user *parent_tidptr,
1680               int __user *child_tidptr)
1681 {
1682         struct task_struct *p;
1683         int trace = 0;
1684         long nr;
1685
1686         /*
1687          * Determine whether and which event to report to ptracer.  When
1688          * called from kernel_thread or CLONE_UNTRACED is explicitly
1689          * requested, no event is reported; otherwise, report if the event
1690          * for the type of forking is enabled.
1691          */
1692         if (!(clone_flags & CLONE_UNTRACED)) {
1693                 if (clone_flags & CLONE_VFORK)
1694                         trace = PTRACE_EVENT_VFORK;
1695                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1696                         trace = PTRACE_EVENT_CLONE;
1697                 else
1698                         trace = PTRACE_EVENT_FORK;
1699
1700                 if (likely(!ptrace_event_enabled(current, trace)))
1701                         trace = 0;
1702         }
1703
1704         p = copy_process(clone_flags, stack_start, stack_size,
1705                          child_tidptr, NULL, trace);
1706         /*
1707          * Do this prior waking up the new thread - the thread pointer
1708          * might get invalid after that point, if the thread exits quickly.
1709          */
1710         if (!IS_ERR(p)) {
1711                 struct completion vfork;
1712                 struct pid *pid;
1713
1714                 trace_sched_process_fork(current, p);
1715
1716                 pid = get_task_pid(p, PIDTYPE_PID);
1717                 nr = pid_vnr(pid);
1718
1719                 if (clone_flags & CLONE_PARENT_SETTID)
1720                         put_user(nr, parent_tidptr);
1721
1722                 if (clone_flags & CLONE_VFORK) {
1723                         p->vfork_done = &vfork;
1724                         init_completion(&vfork);
1725                         get_task_struct(p);
1726                 }
1727
1728                 wake_up_new_task(p);
1729
1730                 /* forking complete and child started to run, tell ptracer */
1731                 if (unlikely(trace))
1732                         ptrace_event_pid(trace, pid);
1733
1734                 if (clone_flags & CLONE_VFORK) {
1735                         if (!wait_for_vfork_done(p, &vfork))
1736                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1737                 }
1738
1739                 put_pid(pid);
1740         } else {
1741                 nr = PTR_ERR(p);
1742         }
1743         return nr;
1744 }
1745
1746 /*
1747  * Create a kernel thread.
1748  */
1749 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1750 {
1751         return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1752                 (unsigned long)arg, NULL, NULL);
1753 }
1754
1755 #ifdef __ARCH_WANT_SYS_FORK
1756 SYSCALL_DEFINE0(fork)
1757 {
1758 #ifdef CONFIG_MMU
1759         return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1760 #else
1761         /* can not support in nommu mode */
1762         return -EINVAL;
1763 #endif
1764 }
1765 #endif
1766
1767 #ifdef __ARCH_WANT_SYS_VFORK
1768 SYSCALL_DEFINE0(vfork)
1769 {
1770         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1771                         0, NULL, NULL);
1772 }
1773 #endif
1774
1775 #ifdef __ARCH_WANT_SYS_CLONE
1776 #ifdef CONFIG_CLONE_BACKWARDS
1777 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1778                  int __user *, parent_tidptr,
1779                  int, tls_val,
1780                  int __user *, child_tidptr)
1781 #elif defined(CONFIG_CLONE_BACKWARDS2)
1782 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1783                  int __user *, parent_tidptr,
1784                  int __user *, child_tidptr,
1785                  int, tls_val)
1786 #elif defined(CONFIG_CLONE_BACKWARDS3)
1787 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1788                 int, stack_size,
1789                 int __user *, parent_tidptr,
1790                 int __user *, child_tidptr,
1791                 int, tls_val)
1792 #else
1793 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1794                  int __user *, parent_tidptr,
1795                  int __user *, child_tidptr,
1796                  int, tls_val)
1797 #endif
1798 {
1799         return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1800 }
1801 #endif
1802
1803 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1804 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1805 #endif
1806
1807 static void sighand_ctor(void *data)
1808 {
1809         struct sighand_struct *sighand = data;
1810
1811         spin_lock_init(&sighand->siglock);
1812         init_waitqueue_head(&sighand->signalfd_wqh);
1813 }
1814
1815 void __init proc_caches_init(void)
1816 {
1817         sighand_cachep = kmem_cache_create("sighand_cache",
1818                         sizeof(struct sighand_struct), 0,
1819                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1820                         SLAB_NOTRACK, sighand_ctor);
1821         signal_cachep = kmem_cache_create("signal_cache",
1822                         sizeof(struct signal_struct), 0,
1823                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1824         files_cachep = kmem_cache_create("files_cache",
1825                         sizeof(struct files_struct), 0,
1826                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1827         fs_cachep = kmem_cache_create("fs_cache",
1828                         sizeof(struct fs_struct), 0,
1829                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1830         /*
1831          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1832          * whole struct cpumask for the OFFSTACK case. We could change
1833          * this to *only* allocate as much of it as required by the
1834          * maximum number of CPU's we can ever have.  The cpumask_allocation
1835          * is at the end of the structure, exactly for that reason.
1836          */
1837         mm_cachep = kmem_cache_create("mm_struct",
1838                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1839                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1840         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1841         mmap_init();
1842         nsproxy_cache_init();
1843 }
1844
1845 /*
1846  * Check constraints on flags passed to the unshare system call.
1847  */
1848 static int check_unshare_flags(unsigned long unshare_flags)
1849 {
1850         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1851                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1852                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1853                                 CLONE_NEWUSER|CLONE_NEWPID))
1854                 return -EINVAL;
1855         /*
1856          * Not implemented, but pretend it works if there is nothing to
1857          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1858          * needs to unshare vm.
1859          */
1860         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1861                 /* FIXME: get_task_mm() increments ->mm_users */
1862                 if (atomic_read(&current->mm->mm_users) > 1)
1863                         return -EINVAL;
1864         }
1865
1866         return 0;
1867 }
1868
1869 /*
1870  * Unshare the filesystem structure if it is being shared
1871  */
1872 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1873 {
1874         struct fs_struct *fs = current->fs;
1875
1876         if (!(unshare_flags & CLONE_FS) || !fs)
1877                 return 0;
1878
1879         /* don't need lock here; in the worst case we'll do useless copy */
1880         if (fs->users == 1)
1881                 return 0;
1882
1883         *new_fsp = copy_fs_struct(fs);
1884         if (!*new_fsp)
1885                 return -ENOMEM;
1886
1887         return 0;
1888 }
1889
1890 /*
1891  * Unshare file descriptor table if it is being shared
1892  */
1893 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1894 {
1895         struct files_struct *fd = current->files;
1896         int error = 0;
1897
1898         if ((unshare_flags & CLONE_FILES) &&
1899             (fd && atomic_read(&fd->count) > 1)) {
1900                 *new_fdp = dup_fd(fd, &error);
1901                 if (!*new_fdp)
1902                         return error;
1903         }
1904
1905         return 0;
1906 }
1907
1908 /*
1909  * unshare allows a process to 'unshare' part of the process
1910  * context which was originally shared using clone.  copy_*
1911  * functions used by do_fork() cannot be used here directly
1912  * because they modify an inactive task_struct that is being
1913  * constructed. Here we are modifying the current, active,
1914  * task_struct.
1915  */
1916 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1917 {
1918         struct fs_struct *fs, *new_fs = NULL;
1919         struct files_struct *fd, *new_fd = NULL;
1920         struct cred *new_cred = NULL;
1921         struct nsproxy *new_nsproxy = NULL;
1922         int do_sysvsem = 0;
1923         int err;
1924
1925         /*
1926          * If unsharing a user namespace must also unshare the thread.
1927          */
1928         if (unshare_flags & CLONE_NEWUSER)
1929                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1930         /*
1931          * If unsharing a thread from a thread group, must also unshare vm.
1932          */
1933         if (unshare_flags & CLONE_THREAD)
1934                 unshare_flags |= CLONE_VM;
1935         /*
1936          * If unsharing vm, must also unshare signal handlers.
1937          */
1938         if (unshare_flags & CLONE_VM)
1939                 unshare_flags |= CLONE_SIGHAND;
1940         /*
1941          * If unsharing namespace, must also unshare filesystem information.
1942          */
1943         if (unshare_flags & CLONE_NEWNS)
1944                 unshare_flags |= CLONE_FS;
1945
1946         err = check_unshare_flags(unshare_flags);
1947         if (err)
1948                 goto bad_unshare_out;
1949         /*
1950          * CLONE_NEWIPC must also detach from the undolist: after switching
1951          * to a new ipc namespace, the semaphore arrays from the old
1952          * namespace are unreachable.
1953          */
1954         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1955                 do_sysvsem = 1;
1956         err = unshare_fs(unshare_flags, &new_fs);
1957         if (err)
1958                 goto bad_unshare_out;
1959         err = unshare_fd(unshare_flags, &new_fd);
1960         if (err)
1961                 goto bad_unshare_cleanup_fs;
1962         err = unshare_userns(unshare_flags, &new_cred);
1963         if (err)
1964                 goto bad_unshare_cleanup_fd;
1965         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1966                                          new_cred, new_fs);
1967         if (err)
1968                 goto bad_unshare_cleanup_cred;
1969
1970         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1971                 if (do_sysvsem) {
1972                         /*
1973                          * CLONE_SYSVSEM is equivalent to sys_exit().
1974                          */
1975                         exit_sem(current);
1976                 }
1977                 if (unshare_flags & CLONE_NEWIPC) {
1978                         /* Orphan segments in old ns (see sem above). */
1979                         exit_shm(current);
1980                         shm_init_task(current);
1981                 }
1982
1983                 if (new_nsproxy)
1984                         switch_task_namespaces(current, new_nsproxy);
1985
1986                 task_lock(current);
1987
1988                 if (new_fs) {
1989                         fs = current->fs;
1990                         spin_lock(&fs->lock);
1991                         current->fs = new_fs;
1992                         if (--fs->users)
1993                                 new_fs = NULL;
1994                         else
1995                                 new_fs = fs;
1996                         spin_unlock(&fs->lock);
1997                 }
1998
1999                 if (new_fd) {
2000                         fd = current->files;
2001                         current->files = new_fd;
2002                         new_fd = fd;
2003                 }
2004
2005                 task_unlock(current);
2006
2007                 if (new_cred) {
2008                         /* Install the new user namespace */
2009                         commit_creds(new_cred);
2010                         new_cred = NULL;
2011                 }
2012         }
2013
2014 bad_unshare_cleanup_cred:
2015         if (new_cred)
2016                 put_cred(new_cred);
2017 bad_unshare_cleanup_fd:
2018         if (new_fd)
2019                 put_files_struct(new_fd);
2020
2021 bad_unshare_cleanup_fs:
2022         if (new_fs)
2023                 free_fs_struct(new_fs);
2024
2025 bad_unshare_out:
2026         return err;
2027 }
2028
2029 /*
2030  *      Helper to unshare the files of the current task.
2031  *      We don't want to expose copy_files internals to
2032  *      the exec layer of the kernel.
2033  */
2034
2035 int unshare_files(struct files_struct **displaced)
2036 {
2037         struct task_struct *task = current;
2038         struct files_struct *copy = NULL;
2039         int error;
2040
2041         error = unshare_fd(CLONE_FILES, &copy);
2042         if (error || !copy) {
2043                 *displaced = NULL;
2044                 return error;
2045         }
2046         *displaced = task->files;
2047         task_lock(task);
2048         task->files = copy;
2049         task_unlock(task);
2050         return 0;
2051 }
2052
2053 int sysctl_max_threads(struct ctl_table *table, int write,
2054                        void __user *buffer, size_t *lenp, loff_t *ppos)
2055 {
2056         struct ctl_table t;
2057         int ret;
2058         int threads = max_threads;
2059         int min = MIN_THREADS;
2060         int max = MAX_THREADS;
2061
2062         t = *table;
2063         t.data = &threads;
2064         t.extra1 = &min;
2065         t.extra2 = &max;
2066
2067         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2068         if (ret || !write)
2069                 return ret;
2070
2071         set_max_threads(threads);
2072
2073         return 0;
2074 }