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 EXPORT_SYMBOL(get_mm_exe_file);
 756
 757 /**
 758  * get_task_mm - acquire a reference to the task's mm
 759  *
 760  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 761  * this kernel workthread has transiently adopted a user mm with use_mm,
 762  * to do its AIO) is not set and if so returns a reference to it, after
 763  * bumping up the use count.  User must release the mm via mmput()
 764  * after use.  Typically used by /proc and ptrace.
 765  */
 766 struct mm_struct *get_task_mm(struct task_struct *task)
 767 {
 768         struct mm_struct *mm;
 769
 770         task_lock(task);
 771         mm = task->mm;
 772         if (mm) {
 773                 if (task->flags & PF_KTHREAD)
 774                         mm = NULL;
 775                 else
 776                         atomic_inc(&mm->mm_users);
 777         }
 778         task_unlock(task);
 779         return mm;
 780 }
 781 EXPORT_SYMBOL_GPL(get_task_mm);
 782
 783 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 784 {
 785         struct mm_struct *mm;
 786         int err;
 787
 788         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 789         if (err)
 790                 return ERR_PTR(err);
 791
 792         mm = get_task_mm(task);
 793         if (mm && mm != current->mm &&
 794                         !ptrace_may_access(task, mode)) {
 795                 mmput(mm);
 796                 mm = ERR_PTR(-EACCES);
 797         }
 798         mutex_unlock(&task->signal->cred_guard_mutex);
 799
 800         return mm;
 801 }
 802
 803 static void complete_vfork_done(struct task_struct *tsk)
 804 {
 805         struct completion *vfork;
 806
 807         task_lock(tsk);
 808         vfork = tsk->vfork_done;
 809         if (likely(vfork)) {
 810                 tsk->vfork_done = NULL;
 811                 complete(vfork);
 812         }
 813         task_unlock(tsk);
 814 }
 815
 816 static int wait_for_vfork_done(struct task_struct *child,
 817                                 struct completion *vfork)
 818 {
 819         int killed;
 820
 821         freezer_do_not_count();
 822         killed = wait_for_completion_killable(vfork);
 823         freezer_count();
 824
 825         if (killed) {
 826                 task_lock(child);
 827                 child->vfork_done = NULL;
 828                 task_unlock(child);
 829         }
 830
 831         put_task_struct(child);
 832         return killed;
 833 }
 834
 835 /* Please note the differences between mmput and mm_release.
 836  * mmput is called whenever we stop holding onto a mm_struct,
 837  * error success whatever.
 838  *
 839  * mm_release is called after a mm_struct has been removed
 840  * from the current process.
 841  *
 842  * This difference is important for error handling, when we
 843  * only half set up a mm_struct for a new process and need to restore
 844  * the old one.  Because we mmput the new mm_struct before
 845  * restoring the old one. . .
 846  * Eric Biederman 10 January 1998
 847  */
 848 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 849 {
 850         /* Get rid of any futexes when releasing the mm */
 851 #ifdef CONFIG_FUTEX
 852         if (unlikely(tsk->robust_list)) {
 853                 exit_robust_list(tsk);
 854                 tsk->robust_list = NULL;
 855         }
 856 #ifdef CONFIG_COMPAT
 857         if (unlikely(tsk->compat_robust_list)) {
 858                 compat_exit_robust_list(tsk);
 859                 tsk->compat_robust_list = NULL;
 860         }
 861 #endif
 862         if (unlikely(!list_empty(&tsk->pi_state_list)))
 863                 exit_pi_state_list(tsk);
 864 #endif
 865
 866         uprobe_free_utask(tsk);
 867
 868         /* Get rid of any cached register state */
 869         deactivate_mm(tsk, mm);
 870
 871         /*
 872          * If we're exiting normally, clear a user-space tid field if
 873          * requested.  We leave this alone when dying by signal, to leave
 874          * the value intact in a core dump, and to save the unnecessary
 875          * trouble, say, a killed vfork parent shouldn't touch this mm.
 876          * Userland only wants this done for a sys_exit.
 877          */
 878         if (tsk->clear_child_tid) {
 879                 if (!(tsk->flags & PF_SIGNALED) &&
 880                     atomic_read(&mm->mm_users) > 1) {
 881                         /*
 882                          * We don't check the error code - if userspace has
 883                          * not set up a proper pointer then tough luck.
 884                          */
 885                         put_user(0, tsk->clear_child_tid);
 886                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 887                                         1, NULL, NULL, 0);
 888                 }
 889                 tsk->clear_child_tid = NULL;
 890         }
 891
 892         /*
 893          * All done, finally we can wake up parent and return this mm to him.
 894          * Also kthread_stop() uses this completion for synchronization.
 895          */
 896         if (tsk->vfork_done)
 897                 complete_vfork_done(tsk);
 898 }
 899
 900 /*
 901  * Allocate a new mm structure and copy contents from the
 902  * mm structure of the passed in task structure.
 903  */
 904 static struct mm_struct *dup_mm(struct task_struct *tsk)
 905 {
 906         struct mm_struct *mm, *oldmm = current->mm;
 907         int err;
 908
 909         mm = allocate_mm();
 910         if (!mm)
 911                 goto fail_nomem;
 912
 913         memcpy(mm, oldmm, sizeof(*mm));
 914
 915         if (!mm_init(mm, tsk))
 916                 goto fail_nomem;
 917
 918         err = dup_mmap(mm, oldmm);
 919         if (err)
 920                 goto free_pt;
 921
 922         mm->hiwater_rss = get_mm_rss(mm);
 923         mm->hiwater_vm = mm->total_vm;
 924
 925         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 926                 goto free_pt;
 927
 928         return mm;
 929
 930 free_pt:
 931         /* don't put binfmt in mmput, we haven't got module yet */
 932         mm->binfmt = NULL;
 933         mmput(mm);
 934
 935 fail_nomem:
 936         return NULL;
 937 }
 938
 939 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 940 {
 941         struct mm_struct *mm, *oldmm;
 942         int retval;
 943
 944         tsk->min_flt = tsk->maj_flt = 0;
 945         tsk->nvcsw = tsk->nivcsw = 0;
 946 #ifdef CONFIG_DETECT_HUNG_TASK
 947         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 948 #endif
 949
 950         tsk->mm = NULL;
 951         tsk->active_mm = NULL;
 952
 953         /*
 954          * Are we cloning a kernel thread?
 955          *
 956          * We need to steal a active VM for that..
 957          */
 958         oldmm = current->mm;
 959         if (!oldmm)
 960                 return 0;
 961
 962         /* initialize the new vmacache entries */
 963         vmacache_flush(tsk);
 964
 965         if (clone_flags & CLONE_VM) {
 966                 atomic_inc(&oldmm->mm_users);
 967                 mm = oldmm;
 968                 goto good_mm;
 969         }
 970
 971         retval = -ENOMEM;
 972         mm = dup_mm(tsk);
 973         if (!mm)
 974                 goto fail_nomem;
 975
 976 good_mm:
 977         tsk->mm = mm;
 978         tsk->active_mm = mm;
 979         return 0;
 980
 981 fail_nomem:
 982         return retval;
 983 }
 984
 985 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 986 {
 987         struct fs_struct *fs = current->fs;
 988         if (clone_flags & CLONE_FS) {
 989                 /* tsk->fs is already what we want */
 990                 spin_lock(&fs->lock);
 991                 if (fs->in_exec) {
 992                         spin_unlock(&fs->lock);
 993                         return -EAGAIN;
 994                 }
 995                 fs->users++;
 996                 spin_unlock(&fs->lock);
 997                 return 0;
 998         }
 999         tsk->fs = copy_fs_struct(fs);
1000         if (!tsk->fs)
1001                 return -ENOMEM;
1002         return 0;
1003 }
1004
1005 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1006 {
1007         struct files_struct *oldf, *newf;
1008         int error = 0;
1009
1010         /*
1011          * A background process may not have any files ...
1012          */
1013         oldf = current->files;
1014         if (!oldf)
1015                 goto out;
1016
1017         if (clone_flags & CLONE_FILES) {
1018                 atomic_inc(&oldf->count);
1019                 goto out;
1020         }
1021
1022         newf = dup_fd(oldf, &error);
1023         if (!newf)
1024                 goto out;
1025
1026         tsk->files = newf;
1027         error = 0;
1028 out:
1029         return error;
1030 }
1031
1032 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1033 {
1034 #ifdef CONFIG_BLOCK
1035         struct io_context *ioc = current->io_context;
1036         struct io_context *new_ioc;
1037
1038         if (!ioc)
1039                 return 0;
1040         /*
1041          * Share io context with parent, if CLONE_IO is set
1042          */
1043         if (clone_flags & CLONE_IO) {
1044                 ioc_task_link(ioc);
1045                 tsk->io_context = ioc;
1046         } else if (ioprio_valid(ioc->ioprio)) {
1047                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1048                 if (unlikely(!new_ioc))
1049                         return -ENOMEM;
1050
1051                 new_ioc->ioprio = ioc->ioprio;
1052                 put_io_context(new_ioc);
1053         }
1054 #endif
1055         return 0;
1056 }
1057
1058 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1059 {
1060         struct sighand_struct *sig;
1061
1062         if (clone_flags & CLONE_SIGHAND) {
1063                 atomic_inc(&current->sighand->count);
1064                 return 0;
1065         }
1066         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1067         rcu_assign_pointer(tsk->sighand, sig);
1068         if (!sig)
1069                 return -ENOMEM;
1070         atomic_set(&sig->count, 1);
1071         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1072         return 0;
1073 }
1074
1075 void __cleanup_sighand(struct sighand_struct *sighand)
1076 {
1077         if (atomic_dec_and_test(&sighand->count)) {
1078                 signalfd_cleanup(sighand);
1079                 /*
1080                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1081                  * without an RCU grace period, see __lock_task_sighand().
1082                  */
1083                 kmem_cache_free(sighand_cachep, sighand);
1084         }
1085 }
1086
1087 /*
1088  * Initialize POSIX timer handling for a thread group.
1089  */
1090 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1091 {
1092         unsigned long cpu_limit;
1093
1094         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1095         if (cpu_limit != RLIM_INFINITY) {
1096                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1097                 sig->cputimer.running = 1;
1098         }
1099
1100         /* The timer lists. */
1101         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1102         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1103         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1104 }
1105
1106 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1107 {
1108         struct signal_struct *sig;
1109
1110         if (clone_flags & CLONE_THREAD)
1111                 return 0;
1112
1113         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1114         tsk->signal = sig;
1115         if (!sig)
1116                 return -ENOMEM;
1117
1118         sig->nr_threads = 1;
1119         atomic_set(&sig->live, 1);
1120         atomic_set(&sig->sigcnt, 1);
1121
1122         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1123         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1124         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1125
1126         init_waitqueue_head(&sig->wait_chldexit);
1127         sig->curr_target = tsk;
1128         init_sigpending(&sig->shared_pending);
1129         INIT_LIST_HEAD(&sig->posix_timers);
1130         seqlock_init(&sig->stats_lock);
1131
1132         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1133         sig->real_timer.function = it_real_fn;
1134
1135         task_lock(current->group_leader);
1136         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1137         task_unlock(current->group_leader);
1138
1139         posix_cpu_timers_init_group(sig);
1140
1141         tty_audit_fork(sig);
1142         sched_autogroup_fork(sig);
1143
1144         sig->oom_score_adj = current->signal->oom_score_adj;
1145         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1146
1147         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1148                                    current->signal->is_child_subreaper;
1149
1150         mutex_init(&sig->cred_guard_mutex);
1151
1152         return 0;
1153 }
1154
1155 static void copy_seccomp(struct task_struct *p)
1156 {
1157 #ifdef CONFIG_SECCOMP
1158         /*
1159          * Must be called with sighand->lock held, which is common to
1160          * all threads in the group. Holding cred_guard_mutex is not
1161          * needed because this new task is not yet running and cannot
1162          * be racing exec.
1163          */
1164         assert_spin_locked(&current->sighand->siglock);
1165
1166         /* Ref-count the new filter user, and assign it. */
1167         get_seccomp_filter(current);
1168         p->seccomp = current->seccomp;
1169
1170         /*
1171          * Explicitly enable no_new_privs here in case it got set
1172          * between the task_struct being duplicated and holding the
1173          * sighand lock. The seccomp state and nnp must be in sync.
1174          */
1175         if (task_no_new_privs(current))
1176                 task_set_no_new_privs(p);
1177
1178         /*
1179          * If the parent gained a seccomp mode after copying thread
1180          * flags and between before we held the sighand lock, we have
1181          * to manually enable the seccomp thread flag here.
1182          */
1183         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1184                 set_tsk_thread_flag(p, TIF_SECCOMP);
1185 #endif
1186 }
1187
1188 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1189 {
1190         current->clear_child_tid = tidptr;
1191
1192         return task_pid_vnr(current);
1193 }
1194
1195 static void rt_mutex_init_task(struct task_struct *p)
1196 {
1197         raw_spin_lock_init(&p->pi_lock);
1198 #ifdef CONFIG_RT_MUTEXES
1199         p->pi_waiters = RB_ROOT;
1200         p->pi_waiters_leftmost = NULL;
1201         p->pi_blocked_on = NULL;
1202 #endif
1203 }
1204
1205 /*
1206  * Initialize POSIX timer handling for a single task.
1207  */
1208 static void posix_cpu_timers_init(struct task_struct *tsk)
1209 {
1210         tsk->cputime_expires.prof_exp = 0;
1211         tsk->cputime_expires.virt_exp = 0;
1212         tsk->cputime_expires.sched_exp = 0;
1213         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1214         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1215         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1216 }
1217
1218 static inline void
1219 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1220 {
1221          task->pids[type].pid = pid;
1222 }
1223
1224 /*
1225  * This creates a new process as a copy of the old one,
1226  * but does not actually start it yet.
1227  *
1228  * It copies the registers, and all the appropriate
1229  * parts of the process environment (as per the clone
1230  * flags). The actual kick-off is left to the caller.
1231  */
1232 static struct task_struct *copy_process(unsigned long clone_flags,
1233                                         unsigned long stack_start,
1234                                         unsigned long stack_size,
1235                                         int __user *child_tidptr,
1236                                         struct pid *pid,
1237                                         int trace,
1238                                         unsigned long tls)
1239 {
1240         int retval;
1241         struct task_struct *p;
1242
1243         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1244                 return ERR_PTR(-EINVAL);
1245
1246         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1247                 return ERR_PTR(-EINVAL);
1248
1249         /*
1250          * Thread groups must share signals as well, and detached threads
1251          * can only be started up within the thread group.
1252          */
1253         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1254                 return ERR_PTR(-EINVAL);
1255
1256         /*
1257          * Shared signal handlers imply shared VM. By way of the above,
1258          * thread groups also imply shared VM. Blocking this case allows
1259          * for various simplifications in other code.
1260          */
1261         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1262                 return ERR_PTR(-EINVAL);
1263
1264         /*
1265          * Siblings of global init remain as zombies on exit since they are
1266          * not reaped by their parent (swapper). To solve this and to avoid
1267          * multi-rooted process trees, prevent global and container-inits
1268          * from creating siblings.
1269          */
1270         if ((clone_flags & CLONE_PARENT) &&
1271                                 current->signal->flags & SIGNAL_UNKILLABLE)
1272                 return ERR_PTR(-EINVAL);
1273
1274         /*
1275          * If the new process will be in a different pid or user namespace
1276          * do not allow it to share a thread group or signal handlers or
1277          * parent with the forking task.
1278          */
1279         if (clone_flags & CLONE_SIGHAND) {
1280                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1281                     (task_active_pid_ns(current) !=
1282                                 current->nsproxy->pid_ns_for_children))
1283                         return ERR_PTR(-EINVAL);
1284         }
1285
1286         retval = security_task_create(clone_flags);
1287         if (retval)
1288                 goto fork_out;
1289
1290         retval = -ENOMEM;
1291         p = dup_task_struct(current);
1292         if (!p)
1293                 goto fork_out;
1294
1295         ftrace_graph_init_task(p);
1296
1297         rt_mutex_init_task(p);
1298
1299 #ifdef CONFIG_PROVE_LOCKING
1300         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1301         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1302 #endif
1303         retval = -EAGAIN;
1304         if (atomic_read(&p->real_cred->user->processes) >=
1305                         task_rlimit(p, RLIMIT_NPROC)) {
1306                 if (p->real_cred->user != INIT_USER &&
1307                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1308                         goto bad_fork_free;
1309         }
1310         current->flags &= ~PF_NPROC_EXCEEDED;
1311
1312         retval = copy_creds(p, clone_flags);
1313         if (retval < 0)
1314                 goto bad_fork_free;
1315
1316         /*
1317          * If multiple threads are within copy_process(), then this check
1318          * triggers too late. This doesn't hurt, the check is only there
1319          * to stop root fork bombs.
1320          */
1321         retval = -EAGAIN;
1322         if (nr_threads >= max_threads)
1323                 goto bad_fork_cleanup_count;
1324
1325         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1326         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1327         p->flags |= PF_FORKNOEXEC;
1328         INIT_LIST_HEAD(&p->children);
1329         INIT_LIST_HEAD(&p->sibling);
1330         rcu_copy_process(p);
1331         p->vfork_done = NULL;
1332         spin_lock_init(&p->alloc_lock);
1333
1334         init_sigpending(&p->pending);
1335
1336         p->utime = p->stime = p->gtime = 0;
1337         p->utimescaled = p->stimescaled = 0;
1338 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1339         p->prev_cputime.utime = p->prev_cputime.stime = 0;
1340 #endif
1341 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1342         seqlock_init(&p->vtime_seqlock);
1343         p->vtime_snap = 0;
1344         p->vtime_snap_whence = VTIME_SLEEPING;
1345 #endif
1346
1347 #if defined(SPLIT_RSS_COUNTING)
1348         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1349 #endif
1350
1351         p->default_timer_slack_ns = current->timer_slack_ns;
1352
1353         task_io_accounting_init(&p->ioac);
1354         acct_clear_integrals(p);
1355
1356         posix_cpu_timers_init(p);
1357
1358         p->start_time = ktime_get_ns();
1359         p->real_start_time = ktime_get_boot_ns();
1360         p->io_context = NULL;
1361         p->audit_context = NULL;
1362         if (clone_flags & CLONE_THREAD)
1363                 threadgroup_change_begin(current);
1364         cgroup_fork(p);
1365 #ifdef CONFIG_NUMA
1366         p->mempolicy = mpol_dup(p->mempolicy);
1367         if (IS_ERR(p->mempolicy)) {
1368                 retval = PTR_ERR(p->mempolicy);
1369                 p->mempolicy = NULL;
1370                 goto bad_fork_cleanup_threadgroup_lock;
1371         }
1372 #endif
1373 #ifdef CONFIG_CPUSETS
1374         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1375         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1376         seqcount_init(&p->mems_allowed_seq);
1377 #endif
1378 #ifdef CONFIG_TRACE_IRQFLAGS
1379         p->irq_events = 0;
1380         p->hardirqs_enabled = 0;
1381         p->hardirq_enable_ip = 0;
1382         p->hardirq_enable_event = 0;
1383         p->hardirq_disable_ip = _THIS_IP_;
1384         p->hardirq_disable_event = 0;
1385         p->softirqs_enabled = 1;
1386         p->softirq_enable_ip = _THIS_IP_;
1387         p->softirq_enable_event = 0;
1388         p->softirq_disable_ip = 0;
1389         p->softirq_disable_event = 0;
1390         p->hardirq_context = 0;
1391         p->softirq_context = 0;
1392 #endif
1393
1394         p->pagefault_disabled = 0;
1395
1396 #ifdef CONFIG_LOCKDEP
1397         p->lockdep_depth = 0; /* no locks held yet */
1398         p->curr_chain_key = 0;
1399         p->lockdep_recursion = 0;
1400 #endif
1401
1402 #ifdef CONFIG_DEBUG_MUTEXES
1403         p->blocked_on = NULL; /* not blocked yet */
1404 #endif
1405 #ifdef CONFIG_BCACHE
1406         p->sequential_io        = 0;
1407         p->sequential_io_avg    = 0;
1408 #endif
1409
1410         /* Perform scheduler related setup. Assign this task to a CPU. */
1411         retval = sched_fork(clone_flags, p);
1412         if (retval)
1413                 goto bad_fork_cleanup_policy;
1414
1415         retval = perf_event_init_task(p);
1416         if (retval)
1417                 goto bad_fork_cleanup_policy;
1418         retval = audit_alloc(p);
1419         if (retval)
1420                 goto bad_fork_cleanup_perf;
1421         /* copy all the process information */
1422         shm_init_task(p);
1423         retval = copy_semundo(clone_flags, p);
1424         if (retval)
1425                 goto bad_fork_cleanup_audit;
1426         retval = copy_files(clone_flags, p);
1427         if (retval)
1428                 goto bad_fork_cleanup_semundo;
1429         retval = copy_fs(clone_flags, p);
1430         if (retval)
1431                 goto bad_fork_cleanup_files;
1432         retval = copy_sighand(clone_flags, p);
1433         if (retval)
1434                 goto bad_fork_cleanup_fs;
1435         retval = copy_signal(clone_flags, p);
1436         if (retval)
1437                 goto bad_fork_cleanup_sighand;
1438         retval = copy_mm(clone_flags, p);
1439         if (retval)
1440                 goto bad_fork_cleanup_signal;
1441         retval = copy_namespaces(clone_flags, p);
1442         if (retval)
1443                 goto bad_fork_cleanup_mm;
1444         retval = copy_io(clone_flags, p);
1445         if (retval)
1446                 goto bad_fork_cleanup_namespaces;
1447         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1448         if (retval)
1449                 goto bad_fork_cleanup_io;
1450
1451         if (pid != &init_struct_pid) {
1452                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1453                 if (IS_ERR(pid)) {
1454                         retval = PTR_ERR(pid);
1455                         goto bad_fork_cleanup_io;
1456                 }
1457         }
1458
1459         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1460         /*
1461          * Clear TID on mm_release()?
1462          */
1463         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1464 #ifdef CONFIG_BLOCK
1465         p->plug = NULL;
1466 #endif
1467 #ifdef CONFIG_FUTEX
1468         p->robust_list = NULL;
1469 #ifdef CONFIG_COMPAT
1470         p->compat_robust_list = NULL;
1471 #endif
1472         INIT_LIST_HEAD(&p->pi_state_list);
1473         p->pi_state_cache = NULL;
1474 #endif
1475         /*
1476          * sigaltstack should be cleared when sharing the same VM
1477          */
1478         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1479                 p->sas_ss_sp = p->sas_ss_size = 0;
1480
1481         /*
1482          * Syscall tracing and stepping should be turned off in the
1483          * child regardless of CLONE_PTRACE.
1484          */
1485         user_disable_single_step(p);
1486         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1487 #ifdef TIF_SYSCALL_EMU
1488         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1489 #endif
1490         clear_all_latency_tracing(p);
1491
1492         /* ok, now we should be set up.. */
1493         p->pid = pid_nr(pid);
1494         if (clone_flags & CLONE_THREAD) {
1495                 p->exit_signal = -1;
1496                 p->group_leader = current->group_leader;
1497                 p->tgid = current->tgid;
1498         } else {
1499                 if (clone_flags & CLONE_PARENT)
1500                         p->exit_signal = current->group_leader->exit_signal;
1501                 else
1502                         p->exit_signal = (clone_flags & CSIGNAL);
1503                 p->group_leader = p;
1504                 p->tgid = p->pid;
1505         }
1506
1507         p->nr_dirtied = 0;
1508         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1509         p->dirty_paused_when = 0;
1510
1511         p->pdeath_signal = 0;
1512         INIT_LIST_HEAD(&p->thread_group);
1513         p->task_works = NULL;
1514
1515         /*
1516          * Make it visible to the rest of the system, but dont wake it up yet.
1517          * Need tasklist lock for parent etc handling!
1518          */
1519         write_lock_irq(&tasklist_lock);
1520
1521         /* CLONE_PARENT re-uses the old parent */
1522         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1523                 p->real_parent = current->real_parent;
1524                 p->parent_exec_id = current->parent_exec_id;
1525         } else {
1526                 p->real_parent = current;
1527                 p->parent_exec_id = current->self_exec_id;
1528         }
1529
1530         spin_lock(&current->sighand->siglock);
1531
1532         /*
1533          * Copy seccomp details explicitly here, in case they were changed
1534          * before holding sighand lock.
1535          */
1536         copy_seccomp(p);
1537
1538         /*
1539          * Process group and session signals need to be delivered to just the
1540          * parent before the fork or both the parent and the child after the
1541          * fork. Restart if a signal comes in before we add the new process to
1542          * it's process group.
1543          * A fatal signal pending means that current will exit, so the new
1544          * thread can't slip out of an OOM kill (or normal SIGKILL).
1545         */
1546         recalc_sigpending();
1547         if (signal_pending(current)) {
1548                 spin_unlock(&current->sighand->siglock);
1549                 write_unlock_irq(&tasklist_lock);
1550                 retval = -ERESTARTNOINTR;
1551                 goto bad_fork_free_pid;
1552         }
1553
1554         if (likely(p->pid)) {
1555                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1556
1557                 init_task_pid(p, PIDTYPE_PID, pid);
1558                 if (thread_group_leader(p)) {
1559                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1560                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1561
1562                         if (is_child_reaper(pid)) {
1563                                 ns_of_pid(pid)->child_reaper = p;
1564                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1565                         }
1566
1567                         p->signal->leader_pid = pid;
1568                         p->signal->tty = tty_kref_get(current->signal->tty);
1569                         list_add_tail(&p->sibling, &p->real_parent->children);
1570                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1571                         attach_pid(p, PIDTYPE_PGID);
1572                         attach_pid(p, PIDTYPE_SID);
1573                         __this_cpu_inc(process_counts);
1574                 } else {
1575                         current->signal->nr_threads++;
1576                         atomic_inc(&current->signal->live);
1577                         atomic_inc(&current->signal->sigcnt);
1578                         list_add_tail_rcu(&p->thread_group,
1579                                           &p->group_leader->thread_group);
1580                         list_add_tail_rcu(&p->thread_node,
1581                                           &p->signal->thread_head);
1582                 }
1583                 attach_pid(p, PIDTYPE_PID);
1584                 nr_threads++;
1585         }
1586
1587         total_forks++;
1588         spin_unlock(&current->sighand->siglock);
1589         syscall_tracepoint_update(p);
1590         write_unlock_irq(&tasklist_lock);
1591
1592         proc_fork_connector(p);
1593         cgroup_post_fork(p);
1594         if (clone_flags & CLONE_THREAD)
1595                 threadgroup_change_end(current);
1596         perf_event_fork(p);
1597
1598         trace_task_newtask(p, clone_flags);
1599         uprobe_copy_process(p, clone_flags);
1600
1601         return p;
1602
1603 bad_fork_free_pid:
1604         if (pid != &init_struct_pid)
1605                 free_pid(pid);
1606 bad_fork_cleanup_io:
1607         if (p->io_context)
1608                 exit_io_context(p);
1609 bad_fork_cleanup_namespaces:
1610         exit_task_namespaces(p);
1611 bad_fork_cleanup_mm:
1612         if (p->mm)
1613                 mmput(p->mm);
1614 bad_fork_cleanup_signal:
1615         if (!(clone_flags & CLONE_THREAD))
1616                 free_signal_struct(p->signal);
1617 bad_fork_cleanup_sighand:
1618         __cleanup_sighand(p->sighand);
1619 bad_fork_cleanup_fs:
1620         exit_fs(p); /* blocking */
1621 bad_fork_cleanup_files:
1622         exit_files(p); /* blocking */
1623 bad_fork_cleanup_semundo:
1624         exit_sem(p);
1625 bad_fork_cleanup_audit:
1626         audit_free(p);
1627 bad_fork_cleanup_perf:
1628         perf_event_free_task(p);
1629 bad_fork_cleanup_policy:
1630 #ifdef CONFIG_NUMA
1631         mpol_put(p->mempolicy);
1632 bad_fork_cleanup_threadgroup_lock:
1633 #endif
1634         if (clone_flags & CLONE_THREAD)
1635                 threadgroup_change_end(current);
1636         delayacct_tsk_free(p);
1637 bad_fork_cleanup_count:
1638         atomic_dec(&p->cred->user->processes);
1639         exit_creds(p);
1640 bad_fork_free:
1641         free_task(p);
1642 fork_out:
1643         return ERR_PTR(retval);
1644 }
1645
1646 static inline void init_idle_pids(struct pid_link *links)
1647 {
1648         enum pid_type type;
1649
1650         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1651                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1652                 links[type].pid = &init_struct_pid;
1653         }
1654 }
1655
1656 struct task_struct *fork_idle(int cpu)
1657 {
1658         struct task_struct *task;
1659         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0);
1660         if (!IS_ERR(task)) {
1661                 init_idle_pids(task->pids);
1662                 init_idle(task, cpu);
1663         }
1664
1665         return task;
1666 }
1667
1668 /*
1669  *  Ok, this is the main fork-routine.
1670  *
1671  * It copies the process, and if successful kick-starts
1672  * it and waits for it to finish using the VM if required.
1673  */
1674 long _do_fork(unsigned long clone_flags,
1675               unsigned long stack_start,
1676               unsigned long stack_size,
1677               int __user *parent_tidptr,
1678               int __user *child_tidptr,
1679               unsigned long tls)
1680 {
1681         struct task_struct *p;
1682         int trace = 0;
1683         long nr;
1684
1685         /*
1686          * Determine whether and which event to report to ptracer.  When
1687          * called from kernel_thread or CLONE_UNTRACED is explicitly
1688          * requested, no event is reported; otherwise, report if the event
1689          * for the type of forking is enabled.
1690          */
1691         if (!(clone_flags & CLONE_UNTRACED)) {
1692                 if (clone_flags & CLONE_VFORK)
1693                         trace = PTRACE_EVENT_VFORK;
1694                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1695                         trace = PTRACE_EVENT_CLONE;
1696                 else
1697                         trace = PTRACE_EVENT_FORK;
1698
1699                 if (likely(!ptrace_event_enabled(current, trace)))
1700                         trace = 0;
1701         }
1702
1703         p = copy_process(clone_flags, stack_start, stack_size,
1704                          child_tidptr, NULL, trace, tls);
1705         /*
1706          * Do this prior waking up the new thread - the thread pointer
1707          * might get invalid after that point, if the thread exits quickly.
1708          */
1709         if (!IS_ERR(p)) {
1710                 struct completion vfork;
1711                 struct pid *pid;
1712
1713                 trace_sched_process_fork(current, p);
1714
1715                 pid = get_task_pid(p, PIDTYPE_PID);
1716                 nr = pid_vnr(pid);
1717
1718                 if (clone_flags & CLONE_PARENT_SETTID)
1719                         put_user(nr, parent_tidptr);
1720
1721                 if (clone_flags & CLONE_VFORK) {
1722                         p->vfork_done = &vfork;
1723                         init_completion(&vfork);
1724                         get_task_struct(p);
1725                 }
1726
1727                 wake_up_new_task(p);
1728
1729                 /* forking complete and child started to run, tell ptracer */
1730                 if (unlikely(trace))
1731                         ptrace_event_pid(trace, pid);
1732
1733                 if (clone_flags & CLONE_VFORK) {
1734                         if (!wait_for_vfork_done(p, &vfork))
1735                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1736                 }
1737
1738                 put_pid(pid);
1739         } else {
1740                 nr = PTR_ERR(p);
1741         }
1742         return nr;
1743 }
1744
1745 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1746 /* For compatibility with architectures that call do_fork directly rather than
1747  * using the syscall entry points below. */
1748 long do_fork(unsigned long clone_flags,
1749               unsigned long stack_start,
1750               unsigned long stack_size,
1751               int __user *parent_tidptr,
1752               int __user *child_tidptr)
1753 {
1754         return _do_fork(clone_flags, stack_start, stack_size,
1755                         parent_tidptr, child_tidptr, 0);
1756 }
1757 #endif
1758
1759 /*
1760  * Create a kernel thread.
1761  */
1762 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1763 {
1764         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1765                 (unsigned long)arg, NULL, NULL, 0);
1766 }
1767
1768 #ifdef __ARCH_WANT_SYS_FORK
1769 SYSCALL_DEFINE0(fork)
1770 {
1771 #ifdef CONFIG_MMU
1772         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1773 #else
1774         /* can not support in nommu mode */
1775         return -EINVAL;
1776 #endif
1777 }
1778 #endif
1779
1780 #ifdef __ARCH_WANT_SYS_VFORK
1781 SYSCALL_DEFINE0(vfork)
1782 {
1783         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1784                         0, NULL, NULL, 0);
1785 }
1786 #endif
1787
1788 #ifdef __ARCH_WANT_SYS_CLONE
1789 #ifdef CONFIG_CLONE_BACKWARDS
1790 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1791                  int __user *, parent_tidptr,
1792                  unsigned long, tls,
1793                  int __user *, child_tidptr)
1794 #elif defined(CONFIG_CLONE_BACKWARDS2)
1795 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1796                  int __user *, parent_tidptr,
1797                  int __user *, child_tidptr,
1798                  unsigned long, tls)
1799 #elif defined(CONFIG_CLONE_BACKWARDS3)
1800 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1801                 int, stack_size,
1802                 int __user *, parent_tidptr,
1803                 int __user *, child_tidptr,
1804                 unsigned long, tls)
1805 #else
1806 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1807                  int __user *, parent_tidptr,
1808                  int __user *, child_tidptr,
1809                  unsigned long, tls)
1810 #endif
1811 {
1812         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
1813 }
1814 #endif
1815
1816 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1817 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1818 #endif
1819
1820 static void sighand_ctor(void *data)
1821 {
1822         struct sighand_struct *sighand = data;
1823
1824         spin_lock_init(&sighand->siglock);
1825         init_waitqueue_head(&sighand->signalfd_wqh);
1826 }
1827
1828 void __init proc_caches_init(void)
1829 {
1830         sighand_cachep = kmem_cache_create("sighand_cache",
1831                         sizeof(struct sighand_struct), 0,
1832                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1833                         SLAB_NOTRACK, sighand_ctor);
1834         signal_cachep = kmem_cache_create("signal_cache",
1835                         sizeof(struct signal_struct), 0,
1836                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1837         files_cachep = kmem_cache_create("files_cache",
1838                         sizeof(struct files_struct), 0,
1839                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1840         fs_cachep = kmem_cache_create("fs_cache",
1841                         sizeof(struct fs_struct), 0,
1842                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1843         /*
1844          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1845          * whole struct cpumask for the OFFSTACK case. We could change
1846          * this to *only* allocate as much of it as required by the
1847          * maximum number of CPU's we can ever have.  The cpumask_allocation
1848          * is at the end of the structure, exactly for that reason.
1849          */
1850         mm_cachep = kmem_cache_create("mm_struct",
1851                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1852                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1853         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1854         mmap_init();
1855         nsproxy_cache_init();
1856 }
1857
1858 /*
1859  * Check constraints on flags passed to the unshare system call.
1860  */
1861 static int check_unshare_flags(unsigned long unshare_flags)
1862 {
1863         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1864                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1865                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1866                                 CLONE_NEWUSER|CLONE_NEWPID))
1867                 return -EINVAL;
1868         /*
1869          * Not implemented, but pretend it works if there is nothing to
1870          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1871          * needs to unshare vm.
1872          */
1873         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1874                 /* FIXME: get_task_mm() increments ->mm_users */
1875                 if (atomic_read(&current->mm->mm_users) > 1)
1876                         return -EINVAL;
1877         }
1878
1879         return 0;
1880 }
1881
1882 /*
1883  * Unshare the filesystem structure if it is being shared
1884  */
1885 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1886 {
1887         struct fs_struct *fs = current->fs;
1888
1889         if (!(unshare_flags & CLONE_FS) || !fs)
1890                 return 0;
1891
1892         /* don't need lock here; in the worst case we'll do useless copy */
1893         if (fs->users == 1)
1894                 return 0;
1895
1896         *new_fsp = copy_fs_struct(fs);
1897         if (!*new_fsp)
1898                 return -ENOMEM;
1899
1900         return 0;
1901 }
1902
1903 /*
1904  * Unshare file descriptor table if it is being shared
1905  */
1906 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1907 {
1908         struct files_struct *fd = current->files;
1909         int error = 0;
1910
1911         if ((unshare_flags & CLONE_FILES) &&
1912             (fd && atomic_read(&fd->count) > 1)) {
1913                 *new_fdp = dup_fd(fd, &error);
1914                 if (!*new_fdp)
1915                         return error;
1916         }
1917
1918         return 0;
1919 }
1920
1921 /*
1922  * unshare allows a process to 'unshare' part of the process
1923  * context which was originally shared using clone.  copy_*
1924  * functions used by do_fork() cannot be used here directly
1925  * because they modify an inactive task_struct that is being
1926  * constructed. Here we are modifying the current, active,
1927  * task_struct.
1928  */
1929 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1930 {
1931         struct fs_struct *fs, *new_fs = NULL;
1932         struct files_struct *fd, *new_fd = NULL;
1933         struct cred *new_cred = NULL;
1934         struct nsproxy *new_nsproxy = NULL;
1935         int do_sysvsem = 0;
1936         int err;
1937
1938         /*
1939          * If unsharing a user namespace must also unshare the thread.
1940          */
1941         if (unshare_flags & CLONE_NEWUSER)
1942                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1943         /*
1944          * If unsharing a thread from a thread group, must also unshare vm.
1945          */
1946         if (unshare_flags & CLONE_THREAD)
1947                 unshare_flags |= CLONE_VM;
1948         /*
1949          * If unsharing vm, must also unshare signal handlers.
1950          */
1951         if (unshare_flags & CLONE_VM)
1952                 unshare_flags |= CLONE_SIGHAND;
1953         /*
1954          * If unsharing namespace, must also unshare filesystem information.
1955          */
1956         if (unshare_flags & CLONE_NEWNS)
1957                 unshare_flags |= CLONE_FS;
1958
1959         err = check_unshare_flags(unshare_flags);
1960         if (err)
1961                 goto bad_unshare_out;
1962         /*
1963          * CLONE_NEWIPC must also detach from the undolist: after switching
1964          * to a new ipc namespace, the semaphore arrays from the old
1965          * namespace are unreachable.
1966          */
1967         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1968                 do_sysvsem = 1;
1969         err = unshare_fs(unshare_flags, &new_fs);
1970         if (err)
1971                 goto bad_unshare_out;
1972         err = unshare_fd(unshare_flags, &new_fd);
1973         if (err)
1974                 goto bad_unshare_cleanup_fs;
1975         err = unshare_userns(unshare_flags, &new_cred);
1976         if (err)
1977                 goto bad_unshare_cleanup_fd;
1978         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1979                                          new_cred, new_fs);
1980         if (err)
1981                 goto bad_unshare_cleanup_cred;
1982
1983         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1984                 if (do_sysvsem) {
1985                         /*
1986                          * CLONE_SYSVSEM is equivalent to sys_exit().
1987                          */
1988                         exit_sem(current);
1989                 }
1990                 if (unshare_flags & CLONE_NEWIPC) {
1991                         /* Orphan segments in old ns (see sem above). */
1992                         exit_shm(current);
1993                         shm_init_task(current);
1994                 }
1995
1996                 if (new_nsproxy)
1997                         switch_task_namespaces(current, new_nsproxy);
1998
1999                 task_lock(current);
2000
2001                 if (new_fs) {
2002                         fs = current->fs;
2003                         spin_lock(&fs->lock);
2004                         current->fs = new_fs;
2005                         if (--fs->users)
2006                                 new_fs = NULL;
2007                         else
2008                                 new_fs = fs;
2009                         spin_unlock(&fs->lock);
2010                 }
2011
2012                 if (new_fd) {
2013                         fd = current->files;
2014                         current->files = new_fd;
2015                         new_fd = fd;
2016                 }
2017
2018                 task_unlock(current);
2019
2020                 if (new_cred) {
2021                         /* Install the new user namespace */
2022                         commit_creds(new_cred);
2023                         new_cred = NULL;
2024                 }
2025         }
2026
2027 bad_unshare_cleanup_cred:
2028         if (new_cred)
2029                 put_cred(new_cred);
2030 bad_unshare_cleanup_fd:
2031         if (new_fd)
2032                 put_files_struct(new_fd);
2033
2034 bad_unshare_cleanup_fs:
2035         if (new_fs)
2036                 free_fs_struct(new_fs);
2037
2038 bad_unshare_out:
2039         return err;
2040 }
2041
2042 /*
2043  *      Helper to unshare the files of the current task.
2044  *      We don't want to expose copy_files internals to
2045  *      the exec layer of the kernel.
2046  */
2047
2048 int unshare_files(struct files_struct **displaced)
2049 {
2050         struct task_struct *task = current;
2051         struct files_struct *copy = NULL;
2052         int error;
2053
2054         error = unshare_fd(CLONE_FILES, &copy);
2055         if (error || !copy) {
2056                 *displaced = NULL;
2057                 return error;
2058         }
2059         *displaced = task->files;
2060         task_lock(task);
2061         task->files = copy;
2062         task_unlock(task);
2063         return 0;
2064 }
2065
2066 int sysctl_max_threads(struct ctl_table *table, int write,
2067                        void __user *buffer, size_t *lenp, loff_t *ppos)
2068 {
2069         struct ctl_table t;
2070         int ret;
2071         int threads = max_threads;
2072         int min = MIN_THREADS;
2073         int max = MAX_THREADS;
2074
2075         t = *table;
2076         t.data = &threads;
2077         t.extra1 = &min;
2078         t.extra2 = &max;
2079
2080         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2081         if (ret || !write)
2082                 return ret;
2083
2084         set_max_threads(threads);
2085
2086         return 0;
2087 }