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