kernel/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 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 #ifdef CONFIG_PREEMPT_RT_BASE
 248 static
 249 #endif
 250 void __put_task_struct(struct task_struct *tsk)
 251 {
 252         WARN_ON(!tsk->exit_state);
 253         WARN_ON(atomic_read(&tsk->usage));
 254         WARN_ON(tsk == current);
 255
 256         cgroup_free(tsk);
 257         task_numa_free(tsk);
 258         security_task_free(tsk);
 259         exit_creds(tsk);
 260         delayacct_tsk_free(tsk);
 261         put_signal_struct(tsk->signal);
 262
 263         if (!profile_handoff_task(tsk))
 264                 free_task(tsk);
 265 }
 266 #ifndef CONFIG_PREEMPT_RT_BASE
 267 EXPORT_SYMBOL_GPL(__put_task_struct);
 268 #else
 269 void __put_task_struct_cb(struct rcu_head *rhp)
 270 {
 271         struct task_struct *tsk = container_of(rhp, struct task_struct, put_rcu);
 272
 273         __put_task_struct(tsk);
 274
 275 }
 276 EXPORT_SYMBOL_GPL(__put_task_struct_cb);
 277 #endif
 278
 279 void __init __weak arch_task_cache_init(void) { }
 280
 281 /*
 282  * set_max_threads
 283  */
 284 static void set_max_threads(unsigned int max_threads_suggested)
 285 {
 286         u64 threads;
 287
 288         /*
 289          * The number of threads shall be limited such that the thread
 290          * structures may only consume a small part of the available memory.
 291          */
 292         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 293                 threads = MAX_THREADS;
 294         else
 295                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 296                                     (u64) THREAD_SIZE * 8UL);
 297
 298         if (threads > max_threads_suggested)
 299                 threads = max_threads_suggested;
 300
 301         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 302 }
 303
 304 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 305 /* Initialized by the architecture: */
 306 int arch_task_struct_size __read_mostly;
 307 #endif
 308
 309 void __init fork_init(void)
 310 {
 311 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 312 #ifndef ARCH_MIN_TASKALIGN
 313 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 314 #endif
 315         /* create a slab on which task_structs can be allocated */
 316         task_struct_cachep =
 317                 kmem_cache_create("task_struct", arch_task_struct_size,
 318                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 319 #endif
 320
 321         /* do the arch specific task caches init */
 322         arch_task_cache_init();
 323
 324         set_max_threads(MAX_THREADS);
 325
 326         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 327         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 328         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 329                 init_task.signal->rlim[RLIMIT_NPROC];
 330 }
 331
 332 int __weak arch_dup_task_struct(struct task_struct *dst,
 333                                                struct task_struct *src)
 334 {
 335         *dst = *src;
 336         return 0;
 337 }
 338
 339 void set_task_stack_end_magic(struct task_struct *tsk)
 340 {
 341         unsigned long *stackend;
 342
 343         stackend = end_of_stack(tsk);
 344         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 345 }
 346
 347 static struct task_struct *dup_task_struct(struct task_struct *orig)
 348 {
 349         struct task_struct *tsk;
 350         struct thread_info *ti;
 351         int node = tsk_fork_get_node(orig);
 352         int err;
 353
 354         tsk = alloc_task_struct_node(node);
 355         if (!tsk)
 356                 return NULL;
 357
 358         ti = alloc_thread_info_node(tsk, node);
 359         if (!ti)
 360                 goto free_tsk;
 361
 362         err = arch_dup_task_struct(tsk, orig);
 363         if (err)
 364                 goto free_ti;
 365
 366         tsk->stack = ti;
 367 #ifdef CONFIG_SECCOMP
 368         /*
 369          * We must handle setting up seccomp filters once we're under
 370          * the sighand lock in case orig has changed between now and
 371          * then. Until then, filter must be NULL to avoid messing up
 372          * the usage counts on the error path calling free_task.
 373          */
 374         tsk->seccomp.filter = NULL;
 375 #endif
 376
 377         setup_thread_stack(tsk, orig);
 378         clear_user_return_notifier(tsk);
 379         clear_tsk_need_resched(tsk);
 380         set_task_stack_end_magic(tsk);
 381
 382 #ifdef CONFIG_CC_STACKPROTECTOR
 383         tsk->stack_canary = get_random_int();
 384 #endif
 385
 386         /*
 387          * One for us, one for whoever does the "release_task()" (usually
 388          * parent)
 389          */
 390         atomic_set(&tsk->usage, 2);
 391 #ifdef CONFIG_BLK_DEV_IO_TRACE
 392         tsk->btrace_seq = 0;
 393 #endif
 394         tsk->splice_pipe = NULL;
 395         tsk->task_frag.page = NULL;
 396         tsk->wake_q.next = NULL;
 397
 398         account_kernel_stack(ti, 1);
 399
 400         return tsk;
 401
 402 free_ti:
 403         free_thread_info(ti);
 404 free_tsk:
 405         free_task_struct(tsk);
 406         return NULL;
 407 }
 408
 409 #ifdef CONFIG_MMU
 410 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 411 {
 412         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 413         struct rb_node **rb_link, *rb_parent;
 414         int retval;
 415         unsigned long charge;
 416
 417         uprobe_start_dup_mmap();
 418         down_write(&oldmm->mmap_sem);
 419         flush_cache_dup_mm(oldmm);
 420         uprobe_dup_mmap(oldmm, mm);
 421         /*
 422          * Not linked in yet - no deadlock potential:
 423          */
 424         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 425
 426         /* No ordering required: file already has been exposed. */
 427         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 428
 429         mm->total_vm = oldmm->total_vm;
 430         mm->shared_vm = oldmm->shared_vm;
 431         mm->exec_vm = oldmm->exec_vm;
 432         mm->stack_vm = oldmm->stack_vm;
 433
 434         rb_link = &mm->mm_rb.rb_node;
 435         rb_parent = NULL;
 436         pprev = &mm->mmap;
 437         retval = ksm_fork(mm, oldmm);
 438         if (retval)
 439                 goto out;
 440         retval = khugepaged_fork(mm, oldmm);
 441         if (retval)
 442                 goto out;
 443
 444         prev = NULL;
 445         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 446                 struct file *file;
 447
 448                 if (mpnt->vm_flags & VM_DONTCOPY) {
 449                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 450                                                         -vma_pages(mpnt));
 451                         continue;
 452                 }
 453                 charge = 0;
 454                 if (mpnt->vm_flags & VM_ACCOUNT) {
 455                         unsigned long len = vma_pages(mpnt);
 456
 457                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 458                                 goto fail_nomem;
 459                         charge = len;
 460                 }
 461                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 462                 if (!tmp)
 463                         goto fail_nomem;
 464                 *tmp = *mpnt;
 465                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 466                 retval = vma_dup_policy(mpnt, tmp);
 467                 if (retval)
 468                         goto fail_nomem_policy;
 469                 tmp->vm_mm = mm;
 470                 if (anon_vma_fork(tmp, mpnt))
 471                         goto fail_nomem_anon_vma_fork;
 472                 tmp->vm_flags &=
 473                         ~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
 474                 tmp->vm_next = tmp->vm_prev = NULL;
 475                 tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
 476                 file = tmp->vm_file;
 477                 if (file) {
 478                         struct inode *inode = file_inode(file);
 479                         struct address_space *mapping = file->f_mapping;
 480
 481                         get_file(file);
 482                         if (tmp->vm_flags & VM_DENYWRITE)
 483                                 atomic_dec(&inode->i_writecount);
 484                         i_mmap_lock_write(mapping);
 485                         if (tmp->vm_flags & VM_SHARED)
 486                                 atomic_inc(&mapping->i_mmap_writable);
 487                         flush_dcache_mmap_lock(mapping);
 488                         /* insert tmp into the share list, just after mpnt */
 489                         vma_interval_tree_insert_after(tmp, mpnt,
 490                                         &mapping->i_mmap);
 491                         flush_dcache_mmap_unlock(mapping);
 492                         i_mmap_unlock_write(mapping);
 493                 }
 494
 495                 /*
 496                  * Clear hugetlb-related page reserves for children. This only
 497                  * affects MAP_PRIVATE mappings. Faults generated by the child
 498                  * are not guaranteed to succeed, even if read-only
 499                  */
 500                 if (is_vm_hugetlb_page(tmp))
 501                         reset_vma_resv_huge_pages(tmp);
 502
 503                 /*
 504                  * Link in the new vma and copy the page table entries.
 505                  */
 506                 *pprev = tmp;
 507                 pprev = &tmp->vm_next;
 508                 tmp->vm_prev = prev;
 509                 prev = tmp;
 510
 511                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 512                 rb_link = &tmp->vm_rb.rb_right;
 513                 rb_parent = &tmp->vm_rb;
 514
 515                 mm->map_count++;
 516                 retval = copy_page_range(mm, oldmm, mpnt);
 517
 518                 if (tmp->vm_ops && tmp->vm_ops->open)
 519                         tmp->vm_ops->open(tmp);
 520
 521                 if (retval)
 522                         goto out;
 523         }
 524         /* a new mm has just been created */
 525         arch_dup_mmap(oldmm, mm);
 526         retval = 0;
 527 out:
 528         up_write(&mm->mmap_sem);
 529         flush_tlb_mm(oldmm);
 530         up_write(&oldmm->mmap_sem);
 531         uprobe_end_dup_mmap();
 532         return retval;
 533 fail_nomem_anon_vma_fork:
 534         mpol_put(vma_policy(tmp));
 535 fail_nomem_policy:
 536         kmem_cache_free(vm_area_cachep, tmp);
 537 fail_nomem:
 538         retval = -ENOMEM;
 539         vm_unacct_memory(charge);
 540         goto out;
 541 }
 542
 543 static inline int mm_alloc_pgd(struct mm_struct *mm)
 544 {
 545         mm->pgd = pgd_alloc(mm);
 546         if (unlikely(!mm->pgd))
 547                 return -ENOMEM;
 548         return 0;
 549 }
 550
 551 static inline void mm_free_pgd(struct mm_struct *mm)
 552 {
 553         pgd_free(mm, mm->pgd);
 554 }
 555 #else
 556 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 557 {
 558         down_write(&oldmm->mmap_sem);
 559         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 560         up_write(&oldmm->mmap_sem);
 561         return 0;
 562 }
 563 #define mm_alloc_pgd(mm)        (0)
 564 #define mm_free_pgd(mm)
 565 #endif /* CONFIG_MMU */
 566
 567 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 568
 569 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 570 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 571
 572 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 573
 574 static int __init coredump_filter_setup(char *s)
 575 {
 576         default_dump_filter =
 577                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 578                 MMF_DUMP_FILTER_MASK;
 579         return 1;
 580 }
 581
 582 __setup("coredump_filter=", coredump_filter_setup);
 583
 584 #include <linux/init_task.h>
 585
 586 static void mm_init_aio(struct mm_struct *mm)
 587 {
 588 #ifdef CONFIG_AIO
 589         spin_lock_init(&mm->ioctx_lock);
 590         mm->ioctx_table = NULL;
 591 #endif
 592 }
 593
 594 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 595 {
 596 #ifdef CONFIG_MEMCG
 597         mm->owner = p;
 598 #endif
 599 }
 600
 601 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 602         struct user_namespace *user_ns)
 603 {
 604         mm->mmap = NULL;
 605         mm->mm_rb = RB_ROOT;
 606         mm->vmacache_seqnum = 0;
 607         atomic_set(&mm->mm_users, 1);
 608         atomic_set(&mm->mm_count, 1);
 609         init_rwsem(&mm->mmap_sem);
 610         INIT_LIST_HEAD(&mm->mmlist);
 611         mm->core_state = NULL;
 612         atomic_long_set(&mm->nr_ptes, 0);
 613         mm_nr_pmds_init(mm);
 614         mm->map_count = 0;
 615         mm->locked_vm = 0;
 616         mm->pinned_vm = 0;
 617         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 618         spin_lock_init(&mm->page_table_lock);
 619         mm_init_cpumask(mm);
 620         mm_init_aio(mm);
 621         mm_init_owner(mm, p);
 622         mmu_notifier_mm_init(mm);
 623         clear_tlb_flush_pending(mm);
 624 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 625         mm->pmd_huge_pte = NULL;
 626 #endif
 627
 628         if (current->mm) {
 629                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 630                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 631         } else {
 632                 mm->flags = default_dump_filter;
 633                 mm->def_flags = 0;
 634         }
 635
 636         if (mm_alloc_pgd(mm))
 637                 goto fail_nopgd;
 638
 639         if (init_new_context(p, mm))
 640                 goto fail_nocontext;
 641
 642         mm->user_ns = get_user_ns(user_ns);
 643         return mm;
 644
 645 fail_nocontext:
 646         mm_free_pgd(mm);
 647 fail_nopgd:
 648         free_mm(mm);
 649         return NULL;
 650 }
 651
 652 static void check_mm(struct mm_struct *mm)
 653 {
 654         int i;
 655
 656         for (i = 0; i < NR_MM_COUNTERS; i++) {
 657                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 658
 659                 if (unlikely(x))
 660                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 661                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 662         }
 663
 664         if (atomic_long_read(&mm->nr_ptes))
 665                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 666                                 atomic_long_read(&mm->nr_ptes));
 667         if (mm_nr_pmds(mm))
 668                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 669                                 mm_nr_pmds(mm));
 670
 671 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 672         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 673 #endif
 674 }
 675
 676 /*
 677  * Allocate and initialize an mm_struct.
 678  */
 679 struct mm_struct *mm_alloc(void)
 680 {
 681         struct mm_struct *mm;
 682
 683         mm = allocate_mm();
 684         if (!mm)
 685                 return NULL;
 686
 687         memset(mm, 0, sizeof(*mm));
 688         return mm_init(mm, current, current_user_ns());
 689 }
 690
 691 /*
 692  * Called when the last reference to the mm
 693  * is dropped: either by a lazy thread or by
 694  * mmput. Free the page directory and the mm.
 695  */
 696 void __mmdrop(struct mm_struct *mm)
 697 {
 698         BUG_ON(mm == &init_mm);
 699         mm_free_pgd(mm);
 700         destroy_context(mm);
 701         mmu_notifier_mm_destroy(mm);
 702         check_mm(mm);
 703         put_user_ns(mm->user_ns);
 704         free_mm(mm);
 705 }
 706 EXPORT_SYMBOL_GPL(__mmdrop);
 707
 708 #ifdef CONFIG_PREEMPT_RT_BASE
 709 /*
 710  * RCU callback for delayed mm drop. Not strictly rcu, but we don't
 711  * want another facility to make this work.
 712  */
 713 void __mmdrop_delayed(struct rcu_head *rhp)
 714 {
 715         struct mm_struct *mm = container_of(rhp, struct mm_struct, delayed_drop);
 716
 717         __mmdrop(mm);
 718 }
 719 #endif
 720
 721 /*
 722  * Decrement the use count and release all resources for an mm.
 723  */
 724 void mmput(struct mm_struct *mm)
 725 {
 726         might_sleep();
 727
 728         if (atomic_dec_and_test(&mm->mm_users)) {
 729                 uprobe_clear_state(mm);
 730                 exit_aio(mm);
 731                 ksm_exit(mm);
 732                 khugepaged_exit(mm); /* must run before exit_mmap */
 733                 exit_mmap(mm);
 734                 set_mm_exe_file(mm, NULL);
 735                 if (!list_empty(&mm->mmlist)) {
 736                         spin_lock(&mmlist_lock);
 737                         list_del(&mm->mmlist);
 738                         spin_unlock(&mmlist_lock);
 739                 }
 740                 if (mm->binfmt)
 741                         module_put(mm->binfmt->module);
 742                 mmdrop(mm);
 743         }
 744 }
 745 EXPORT_SYMBOL_GPL(mmput);
 746
 747 /**
 748  * set_mm_exe_file - change a reference to the mm's executable file
 749  *
 750  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 751  *
 752  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 753  * invocations: in mmput() nobody alive left, in execve task is single
 754  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 755  * mm->exe_file, but does so without using set_mm_exe_file() in order
 756  * to do avoid the need for any locks.
 757  */
 758 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 759 {
 760         struct file *old_exe_file;
 761
 762         /*
 763          * It is safe to dereference the exe_file without RCU as
 764          * this function is only called if nobody else can access
 765          * this mm -- see comment above for justification.
 766          */
 767         old_exe_file = rcu_dereference_raw(mm->exe_file);
 768
 769         if (new_exe_file)
 770                 get_file(new_exe_file);
 771         rcu_assign_pointer(mm->exe_file, new_exe_file);
 772         if (old_exe_file)
 773                 fput(old_exe_file);
 774 }
 775
 776 /**
 777  * get_mm_exe_file - acquire a reference to the mm's executable file
 778  *
 779  * Returns %NULL if mm has no associated executable file.
 780  * User must release file via fput().
 781  */
 782 struct file *get_mm_exe_file(struct mm_struct *mm)
 783 {
 784         struct file *exe_file;
 785
 786         rcu_read_lock();
 787         exe_file = rcu_dereference(mm->exe_file);
 788         if (exe_file && !get_file_rcu(exe_file))
 789                 exe_file = NULL;
 790         rcu_read_unlock();
 791         return exe_file;
 792 }
 793 EXPORT_SYMBOL(get_mm_exe_file);
 794
 795 /**
 796  * get_task_exe_file - acquire a reference to the task's executable file
 797  *
 798  * Returns %NULL if task's mm (if any) has no associated executable file or
 799  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
 800  * User must release file via fput().
 801  */
 802 struct file *get_task_exe_file(struct task_struct *task)
 803 {
 804         struct file *exe_file = NULL;
 805         struct mm_struct *mm;
 806
 807         task_lock(task);
 808         mm = task->mm;
 809         if (mm) {
 810                 if (!(task->flags & PF_KTHREAD))
 811                         exe_file = get_mm_exe_file(mm);
 812         }
 813         task_unlock(task);
 814         return exe_file;
 815 }
 816 EXPORT_SYMBOL(get_task_exe_file);
 817
 818 /**
 819  * get_task_mm - acquire a reference to the task's mm
 820  *
 821  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 822  * this kernel workthread has transiently adopted a user mm with use_mm,
 823  * to do its AIO) is not set and if so returns a reference to it, after
 824  * bumping up the use count.  User must release the mm via mmput()
 825  * after use.  Typically used by /proc and ptrace.
 826  */
 827 struct mm_struct *get_task_mm(struct task_struct *task)
 828 {
 829         struct mm_struct *mm;
 830
 831         task_lock(task);
 832         mm = task->mm;
 833         if (mm) {
 834                 if (task->flags & PF_KTHREAD)
 835                         mm = NULL;
 836                 else
 837                         atomic_inc(&mm->mm_users);
 838         }
 839         task_unlock(task);
 840         return mm;
 841 }
 842 EXPORT_SYMBOL_GPL(get_task_mm);
 843
 844 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 845 {
 846         struct mm_struct *mm;
 847         int err;
 848
 849         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 850         if (err)
 851                 return ERR_PTR(err);
 852
 853         mm = get_task_mm(task);
 854         if (mm && mm != current->mm &&
 855                         !ptrace_may_access(task, mode)) {
 856                 mmput(mm);
 857                 mm = ERR_PTR(-EACCES);
 858         }
 859         mutex_unlock(&task->signal->cred_guard_mutex);
 860
 861         return mm;
 862 }
 863
 864 static void complete_vfork_done(struct task_struct *tsk)
 865 {
 866         struct completion *vfork;
 867
 868         task_lock(tsk);
 869         vfork = tsk->vfork_done;
 870         if (likely(vfork)) {
 871                 tsk->vfork_done = NULL;
 872                 complete(vfork);
 873         }
 874         task_unlock(tsk);
 875 }
 876
 877 static int wait_for_vfork_done(struct task_struct *child,
 878                                 struct completion *vfork)
 879 {
 880         int killed;
 881
 882         freezer_do_not_count();
 883         killed = wait_for_completion_killable(vfork);
 884         freezer_count();
 885
 886         if (killed) {
 887                 task_lock(child);
 888                 child->vfork_done = NULL;
 889                 task_unlock(child);
 890         }
 891
 892         put_task_struct(child);
 893         return killed;
 894 }
 895
 896 /* Please note the differences between mmput and mm_release.
 897  * mmput is called whenever we stop holding onto a mm_struct,
 898  * error success whatever.
 899  *
 900  * mm_release is called after a mm_struct has been removed
 901  * from the current process.
 902  *
 903  * This difference is important for error handling, when we
 904  * only half set up a mm_struct for a new process and need to restore
 905  * the old one.  Because we mmput the new mm_struct before
 906  * restoring the old one. . .
 907  * Eric Biederman 10 January 1998
 908  */
 909 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 910 {
 911         /* Get rid of any futexes when releasing the mm */
 912 #ifdef CONFIG_FUTEX
 913         if (unlikely(tsk->robust_list)) {
 914                 exit_robust_list(tsk);
 915                 tsk->robust_list = NULL;
 916         }
 917 #ifdef CONFIG_COMPAT
 918         if (unlikely(tsk->compat_robust_list)) {
 919                 compat_exit_robust_list(tsk);
 920                 tsk->compat_robust_list = NULL;
 921         }
 922 #endif
 923         if (unlikely(!list_empty(&tsk->pi_state_list)))
 924                 exit_pi_state_list(tsk);
 925 #endif
 926
 927         uprobe_free_utask(tsk);
 928
 929         /* Get rid of any cached register state */
 930         deactivate_mm(tsk, mm);
 931
 932         /*
 933          * Signal userspace if we're not exiting with a core dump
 934          * because we want to leave the value intact for debugging
 935          * purposes.
 936          */
 937         if (tsk->clear_child_tid) {
 938                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
 939                     atomic_read(&mm->mm_users) > 1) {
 940                         /*
 941                          * We don't check the error code - if userspace has
 942                          * not set up a proper pointer then tough luck.
 943                          */
 944                         put_user(0, tsk->clear_child_tid);
 945                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 946                                         1, NULL, NULL, 0);
 947                 }
 948                 tsk->clear_child_tid = NULL;
 949         }
 950
 951         /*
 952          * All done, finally we can wake up parent and return this mm to him.
 953          * Also kthread_stop() uses this completion for synchronization.
 954          */
 955         if (tsk->vfork_done)
 956                 complete_vfork_done(tsk);
 957 }
 958
 959 /*
 960  * Allocate a new mm structure and copy contents from the
 961  * mm structure of the passed in task structure.
 962  */
 963 static struct mm_struct *dup_mm(struct task_struct *tsk)
 964 {
 965         struct mm_struct *mm, *oldmm = current->mm;
 966         int err;
 967
 968         mm = allocate_mm();
 969         if (!mm)
 970                 goto fail_nomem;
 971
 972         memcpy(mm, oldmm, sizeof(*mm));
 973
 974         if (!mm_init(mm, tsk, mm->user_ns))
 975                 goto fail_nomem;
 976
 977         err = dup_mmap(mm, oldmm);
 978         if (err)
 979                 goto free_pt;
 980
 981         mm->hiwater_rss = get_mm_rss(mm);
 982         mm->hiwater_vm = mm->total_vm;
 983
 984         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 985                 goto free_pt;
 986
 987         return mm;
 988
 989 free_pt:
 990         /* don't put binfmt in mmput, we haven't got module yet */
 991         mm->binfmt = NULL;
 992         mmput(mm);
 993
 994 fail_nomem:
 995         return NULL;
 996 }
 997
 998 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 999 {
1000         struct mm_struct *mm, *oldmm;
1001         int retval;
1002
1003         tsk->min_flt = tsk->maj_flt = 0;
1004         tsk->nvcsw = tsk->nivcsw = 0;
1005 #ifdef CONFIG_DETECT_HUNG_TASK
1006         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1007 #endif
1008
1009         tsk->mm = NULL;
1010         tsk->active_mm = NULL;
1011
1012         /*
1013          * Are we cloning a kernel thread?
1014          *
1015          * We need to steal a active VM for that..
1016          */
1017         oldmm = current->mm;
1018         if (!oldmm)
1019                 return 0;
1020
1021         /* initialize the new vmacache entries */
1022         vmacache_flush(tsk);
1023
1024         if (clone_flags & CLONE_VM) {
1025                 atomic_inc(&oldmm->mm_users);
1026                 mm = oldmm;
1027                 goto good_mm;
1028         }
1029
1030         retval = -ENOMEM;
1031         mm = dup_mm(tsk);
1032         if (!mm)
1033                 goto fail_nomem;
1034
1035 good_mm:
1036         tsk->mm = mm;
1037         tsk->active_mm = mm;
1038         return 0;
1039
1040 fail_nomem:
1041         return retval;
1042 }
1043
1044 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1045 {
1046         struct fs_struct *fs = current->fs;
1047         if (clone_flags & CLONE_FS) {
1048                 /* tsk->fs is already what we want */
1049                 spin_lock(&fs->lock);
1050                 if (fs->in_exec) {
1051                         spin_unlock(&fs->lock);
1052                         return -EAGAIN;
1053                 }
1054                 fs->users++;
1055                 spin_unlock(&fs->lock);
1056                 return 0;
1057         }
1058         tsk->fs = copy_fs_struct(fs);
1059         if (!tsk->fs)
1060                 return -ENOMEM;
1061         return 0;
1062 }
1063
1064 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1065 {
1066         struct files_struct *oldf, *newf;
1067         int error = 0;
1068
1069         /*
1070          * A background process may not have any files ...
1071          */
1072         oldf = current->files;
1073         if (!oldf)
1074                 goto out;
1075
1076         if (clone_flags & CLONE_FILES) {
1077                 atomic_inc(&oldf->count);
1078                 goto out;
1079         }
1080
1081         newf = dup_fd(oldf, &error);
1082         if (!newf)
1083                 goto out;
1084
1085         tsk->files = newf;
1086         error = 0;
1087 out:
1088         return error;
1089 }
1090
1091 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1092 {
1093 #ifdef CONFIG_BLOCK
1094         struct io_context *ioc = current->io_context;
1095         struct io_context *new_ioc;
1096
1097         if (!ioc)
1098                 return 0;
1099         /*
1100          * Share io context with parent, if CLONE_IO is set
1101          */
1102         if (clone_flags & CLONE_IO) {
1103                 ioc_task_link(ioc);
1104                 tsk->io_context = ioc;
1105         } else if (ioprio_valid(ioc->ioprio)) {
1106                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1107                 if (unlikely(!new_ioc))
1108                         return -ENOMEM;
1109
1110                 new_ioc->ioprio = ioc->ioprio;
1111                 put_io_context(new_ioc);
1112         }
1113 #endif
1114         return 0;
1115 }
1116
1117 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1118 {
1119         struct sighand_struct *sig;
1120
1121         if (clone_flags & CLONE_SIGHAND) {
1122                 atomic_inc(&current->sighand->count);
1123                 return 0;
1124         }
1125         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1126         rcu_assign_pointer(tsk->sighand, sig);
1127         if (!sig)
1128                 return -ENOMEM;
1129
1130         atomic_set(&sig->count, 1);
1131         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1132         return 0;
1133 }
1134
1135 void __cleanup_sighand(struct sighand_struct *sighand)
1136 {
1137         if (atomic_dec_and_test(&sighand->count)) {
1138                 signalfd_cleanup(sighand);
1139                 /*
1140                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1141                  * without an RCU grace period, see __lock_task_sighand().
1142                  */
1143                 kmem_cache_free(sighand_cachep, sighand);
1144         }
1145 }
1146
1147 /*
1148  * Initialize POSIX timer handling for a thread group.
1149  */
1150 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1151 {
1152         unsigned long cpu_limit;
1153
1154         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1155         if (cpu_limit != RLIM_INFINITY) {
1156                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1157                 sig->cputimer.running = true;
1158         }
1159
1160         /* The timer lists. */
1161         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1162         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1163         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1164 }
1165
1166 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1167 {
1168         struct signal_struct *sig;
1169
1170         if (clone_flags & CLONE_THREAD)
1171                 return 0;
1172
1173         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1174         tsk->signal = sig;
1175         if (!sig)
1176                 return -ENOMEM;
1177
1178         sig->nr_threads = 1;
1179         atomic_set(&sig->live, 1);
1180         atomic_set(&sig->sigcnt, 1);
1181
1182         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1183         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1184         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1185
1186         init_waitqueue_head(&sig->wait_chldexit);
1187         sig->curr_target = tsk;
1188         init_sigpending(&sig->shared_pending);
1189         INIT_LIST_HEAD(&sig->posix_timers);
1190         seqlock_init(&sig->stats_lock);
1191         prev_cputime_init(&sig->prev_cputime);
1192
1193         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1194         sig->real_timer.function = it_real_fn;
1195
1196         task_lock(current->group_leader);
1197         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1198         task_unlock(current->group_leader);
1199
1200         posix_cpu_timers_init_group(sig);
1201
1202         tty_audit_fork(sig);
1203         sched_autogroup_fork(sig);
1204
1205         sig->oom_score_adj = current->signal->oom_score_adj;
1206         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1207
1208         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1209                                    current->signal->is_child_subreaper;
1210
1211         mutex_init(&sig->cred_guard_mutex);
1212
1213         return 0;
1214 }
1215
1216 static void copy_seccomp(struct task_struct *p)
1217 {
1218 #ifdef CONFIG_SECCOMP
1219         /*
1220          * Must be called with sighand->lock held, which is common to
1221          * all threads in the group. Holding cred_guard_mutex is not
1222          * needed because this new task is not yet running and cannot
1223          * be racing exec.
1224          */
1225         assert_spin_locked(&current->sighand->siglock);
1226
1227         /* Ref-count the new filter user, and assign it. */
1228         get_seccomp_filter(current);
1229         p->seccomp = current->seccomp;
1230
1231         /*
1232          * Explicitly enable no_new_privs here in case it got set
1233          * between the task_struct being duplicated and holding the
1234          * sighand lock. The seccomp state and nnp must be in sync.
1235          */
1236         if (task_no_new_privs(current))
1237                 task_set_no_new_privs(p);
1238
1239         /*
1240          * If the parent gained a seccomp mode after copying thread
1241          * flags and between before we held the sighand lock, we have
1242          * to manually enable the seccomp thread flag here.
1243          */
1244         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1245                 set_tsk_thread_flag(p, TIF_SECCOMP);
1246 #endif
1247 }
1248
1249 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1250 {
1251         current->clear_child_tid = tidptr;
1252
1253         return task_pid_vnr(current);
1254 }
1255
1256 static void rt_mutex_init_task(struct task_struct *p)
1257 {
1258         raw_spin_lock_init(&p->pi_lock);
1259 #ifdef CONFIG_RT_MUTEXES
1260         p->pi_waiters = RB_ROOT;
1261         p->pi_waiters_leftmost = NULL;
1262         p->pi_blocked_on = NULL;
1263 #endif
1264 }
1265
1266 /*
1267  * Initialize POSIX timer handling for a single task.
1268  */
1269 static void posix_cpu_timers_init(struct task_struct *tsk)
1270 {
1271 #ifdef CONFIG_PREEMPT_RT_BASE
1272         tsk->posix_timer_list = NULL;
1273 #endif
1274         tsk->cputime_expires.prof_exp = 0;
1275         tsk->cputime_expires.virt_exp = 0;
1276         tsk->cputime_expires.sched_exp = 0;
1277         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1278         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1279         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1280 }
1281
1282 static inline void
1283 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1284 {
1285          task->pids[type].pid = pid;
1286 }
1287
1288 /*
1289  * This creates a new process as a copy of the old one,
1290  * but does not actually start it yet.
1291  *
1292  * It copies the registers, and all the appropriate
1293  * parts of the process environment (as per the clone
1294  * flags). The actual kick-off is left to the caller.
1295  */
1296 static struct task_struct *copy_process(unsigned long clone_flags,
1297                                         unsigned long stack_start,
1298                                         unsigned long stack_size,
1299                                         int __user *child_tidptr,
1300                                         struct pid *pid,
1301                                         int trace,
1302                                         unsigned long tls)
1303 {
1304         int retval;
1305         struct task_struct *p;
1306         void *cgrp_ss_priv[CGROUP_CANFORK_COUNT] = {};
1307
1308         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1309                 return ERR_PTR(-EINVAL);
1310
1311         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1312                 return ERR_PTR(-EINVAL);
1313
1314         /*
1315          * Thread groups must share signals as well, and detached threads
1316          * can only be started up within the thread group.
1317          */
1318         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1319                 return ERR_PTR(-EINVAL);
1320
1321         /*
1322          * Shared signal handlers imply shared VM. By way of the above,
1323          * thread groups also imply shared VM. Blocking this case allows
1324          * for various simplifications in other code.
1325          */
1326         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1327                 return ERR_PTR(-EINVAL);
1328
1329         /*
1330          * Siblings of global init remain as zombies on exit since they are
1331          * not reaped by their parent (swapper). To solve this and to avoid
1332          * multi-rooted process trees, prevent global and container-inits
1333          * from creating siblings.
1334          */
1335         if ((clone_flags & CLONE_PARENT) &&
1336                                 current->signal->flags & SIGNAL_UNKILLABLE)
1337                 return ERR_PTR(-EINVAL);
1338
1339         /*
1340          * If the new process will be in a different pid or user namespace
1341          * do not allow it to share a thread group with the forking task.
1342          */
1343         if (clone_flags & CLONE_THREAD) {
1344                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1345                     (task_active_pid_ns(current) !=
1346                                 current->nsproxy->pid_ns_for_children))
1347                         return ERR_PTR(-EINVAL);
1348         }
1349
1350         retval = security_task_create(clone_flags);
1351         if (retval)
1352                 goto fork_out;
1353
1354         retval = -ENOMEM;
1355         p = dup_task_struct(current);
1356         if (!p)
1357                 goto fork_out;
1358
1359         ftrace_graph_init_task(p);
1360
1361         rt_mutex_init_task(p);
1362
1363 #ifdef CONFIG_PROVE_LOCKING
1364         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1365         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1366 #endif
1367         retval = -EAGAIN;
1368         if (atomic_read(&p->real_cred->user->processes) >=
1369                         task_rlimit(p, RLIMIT_NPROC)) {
1370                 if (p->real_cred->user != INIT_USER &&
1371                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1372                         goto bad_fork_free;
1373         }
1374         current->flags &= ~PF_NPROC_EXCEEDED;
1375
1376         retval = copy_creds(p, clone_flags);
1377         if (retval < 0)
1378                 goto bad_fork_free;
1379
1380         /*
1381          * If multiple threads are within copy_process(), then this check
1382          * triggers too late. This doesn't hurt, the check is only there
1383          * to stop root fork bombs.
1384          */
1385         retval = -EAGAIN;
1386         if (nr_threads >= max_threads)
1387                 goto bad_fork_cleanup_count;
1388
1389         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1390         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1391         p->flags |= PF_FORKNOEXEC;
1392         INIT_LIST_HEAD(&p->children);
1393         INIT_LIST_HEAD(&p->sibling);
1394         rcu_copy_process(p);
1395         p->vfork_done = NULL;
1396         spin_lock_init(&p->alloc_lock);
1397
1398         init_sigpending(&p->pending);
1399         p->sigqueue_cache = NULL;
1400
1401         p->utime = p->stime = p->gtime = 0;
1402         p->utimescaled = p->stimescaled = 0;
1403         prev_cputime_init(&p->prev_cputime);
1404
1405 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1406         seqcount_init(&p->vtime_seqcount);
1407         p->vtime_snap = 0;
1408         p->vtime_snap_whence = VTIME_INACTIVE;
1409 #endif
1410
1411 #if defined(SPLIT_RSS_COUNTING)
1412         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1413 #endif
1414
1415         p->default_timer_slack_ns = current->timer_slack_ns;
1416
1417         task_io_accounting_init(&p->ioac);
1418         acct_clear_integrals(p);
1419
1420         posix_cpu_timers_init(p);
1421
1422         p->start_time = ktime_get_ns();
1423         p->real_start_time = ktime_get_boot_ns();
1424         p->io_context = NULL;
1425         p->audit_context = NULL;
1426         cgroup_fork(p);
1427 #ifdef CONFIG_NUMA
1428         p->mempolicy = mpol_dup(p->mempolicy);
1429         if (IS_ERR(p->mempolicy)) {
1430                 retval = PTR_ERR(p->mempolicy);
1431                 p->mempolicy = NULL;
1432                 goto bad_fork_cleanup_threadgroup_lock;
1433         }
1434 #endif
1435 #ifdef CONFIG_CPUSETS
1436         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1437         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1438         seqcount_init(&p->mems_allowed_seq);
1439 #endif
1440 #ifdef CONFIG_TRACE_IRQFLAGS
1441         p->irq_events = 0;
1442         p->hardirqs_enabled = 0;
1443         p->hardirq_enable_ip = 0;
1444         p->hardirq_enable_event = 0;
1445         p->hardirq_disable_ip = _THIS_IP_;
1446         p->hardirq_disable_event = 0;
1447         p->softirqs_enabled = 1;
1448         p->softirq_enable_ip = _THIS_IP_;
1449         p->softirq_enable_event = 0;
1450         p->softirq_disable_ip = 0;
1451         p->softirq_disable_event = 0;
1452         p->hardirq_context = 0;
1453         p->softirq_context = 0;
1454 #endif
1455
1456         p->pagefault_disabled = 0;
1457
1458 #ifdef CONFIG_LOCKDEP
1459         p->lockdep_depth = 0; /* no locks held yet */
1460         p->curr_chain_key = 0;
1461         p->lockdep_recursion = 0;
1462 #endif
1463
1464 #ifdef CONFIG_DEBUG_MUTEXES
1465         p->blocked_on = NULL; /* not blocked yet */
1466 #endif
1467 #ifdef CONFIG_BCACHE
1468         p->sequential_io        = 0;
1469         p->sequential_io_avg    = 0;
1470 #endif
1471
1472         /* Perform scheduler related setup. Assign this task to a CPU. */
1473         retval = sched_fork(clone_flags, p);
1474         if (retval)
1475                 goto bad_fork_cleanup_policy;
1476
1477         retval = perf_event_init_task(p);
1478         if (retval)
1479                 goto bad_fork_cleanup_policy;
1480         retval = audit_alloc(p);
1481         if (retval)
1482                 goto bad_fork_cleanup_perf;
1483         /* copy all the process information */
1484         shm_init_task(p);
1485         retval = copy_semundo(clone_flags, p);
1486         if (retval)
1487                 goto bad_fork_cleanup_audit;
1488         retval = copy_files(clone_flags, p);
1489         if (retval)
1490                 goto bad_fork_cleanup_semundo;
1491         retval = copy_fs(clone_flags, p);
1492         if (retval)
1493                 goto bad_fork_cleanup_files;
1494         retval = copy_sighand(clone_flags, p);
1495         if (retval)
1496                 goto bad_fork_cleanup_fs;
1497         retval = copy_signal(clone_flags, p);
1498         if (retval)
1499                 goto bad_fork_cleanup_sighand;
1500         retval = copy_mm(clone_flags, p);
1501         if (retval)
1502                 goto bad_fork_cleanup_signal;
1503         retval = copy_namespaces(clone_flags, p);
1504         if (retval)
1505                 goto bad_fork_cleanup_mm;
1506         retval = copy_io(clone_flags, p);
1507         if (retval)
1508                 goto bad_fork_cleanup_namespaces;
1509         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1510         if (retval)
1511                 goto bad_fork_cleanup_io;
1512
1513         if (pid != &init_struct_pid) {
1514                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1515                 if (IS_ERR(pid)) {
1516                         retval = PTR_ERR(pid);
1517                         goto bad_fork_cleanup_io;
1518                 }
1519         }
1520
1521         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1522         /*
1523          * Clear TID on mm_release()?
1524          */
1525         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1526 #ifdef CONFIG_BLOCK
1527         p->plug = NULL;
1528 #endif
1529 #ifdef CONFIG_FUTEX
1530         p->robust_list = NULL;
1531 #ifdef CONFIG_COMPAT
1532         p->compat_robust_list = NULL;
1533 #endif
1534         INIT_LIST_HEAD(&p->pi_state_list);
1535         p->pi_state_cache = NULL;
1536 #endif
1537         /*
1538          * sigaltstack should be cleared when sharing the same VM
1539          */
1540         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1541                 p->sas_ss_sp = p->sas_ss_size = 0;
1542
1543         /*
1544          * Syscall tracing and stepping should be turned off in the
1545          * child regardless of CLONE_PTRACE.
1546          */
1547         user_disable_single_step(p);
1548         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1549 #ifdef TIF_SYSCALL_EMU
1550         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1551 #endif
1552         clear_all_latency_tracing(p);
1553
1554         /* ok, now we should be set up.. */
1555         p->pid = pid_nr(pid);
1556         if (clone_flags & CLONE_THREAD) {
1557                 p->exit_signal = -1;
1558                 p->group_leader = current->group_leader;
1559                 p->tgid = current->tgid;
1560         } else {
1561                 if (clone_flags & CLONE_PARENT)
1562                         p->exit_signal = current->group_leader->exit_signal;
1563                 else
1564                         p->exit_signal = (clone_flags & CSIGNAL);
1565                 p->group_leader = p;
1566                 p->tgid = p->pid;
1567         }
1568
1569         p->nr_dirtied = 0;
1570         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1571         p->dirty_paused_when = 0;
1572
1573         p->pdeath_signal = 0;
1574         INIT_LIST_HEAD(&p->thread_group);
1575         p->task_works = NULL;
1576
1577         threadgroup_change_begin(current);
1578         /*
1579          * Ensure that the cgroup subsystem policies allow the new process to be
1580          * forked. It should be noted the the new process's css_set can be changed
1581          * between here and cgroup_post_fork() if an organisation operation is in
1582          * progress.
1583          */
1584         retval = cgroup_can_fork(p, cgrp_ss_priv);
1585         if (retval)
1586                 goto bad_fork_free_pid;
1587
1588         /*
1589          * Make it visible to the rest of the system, but dont wake it up yet.
1590          * Need tasklist lock for parent etc handling!
1591          */
1592         write_lock_irq(&tasklist_lock);
1593
1594         /* CLONE_PARENT re-uses the old parent */
1595         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1596                 p->real_parent = current->real_parent;
1597                 p->parent_exec_id = current->parent_exec_id;
1598         } else {
1599                 p->real_parent = current;
1600                 p->parent_exec_id = current->self_exec_id;
1601         }
1602
1603         spin_lock(&current->sighand->siglock);
1604
1605         /*
1606          * Copy seccomp details explicitly here, in case they were changed
1607          * before holding sighand lock.
1608          */
1609         copy_seccomp(p);
1610
1611         /*
1612          * Process group and session signals need to be delivered to just the
1613          * parent before the fork or both the parent and the child after the
1614          * fork. Restart if a signal comes in before we add the new process to
1615          * it's process group.
1616          * A fatal signal pending means that current will exit, so the new
1617          * thread can't slip out of an OOM kill (or normal SIGKILL).
1618         */
1619         recalc_sigpending();
1620         if (signal_pending(current)) {
1621                 spin_unlock(&current->sighand->siglock);
1622                 write_unlock_irq(&tasklist_lock);
1623                 retval = -ERESTARTNOINTR;
1624                 goto bad_fork_cancel_cgroup;
1625         }
1626
1627         if (likely(p->pid)) {
1628                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1629
1630                 init_task_pid(p, PIDTYPE_PID, pid);
1631                 if (thread_group_leader(p)) {
1632                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1633                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1634
1635                         if (is_child_reaper(pid)) {
1636                                 ns_of_pid(pid)->child_reaper = p;
1637                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1638                         }
1639
1640                         p->signal->leader_pid = pid;
1641                         p->signal->tty = tty_kref_get(current->signal->tty);
1642                         list_add_tail(&p->sibling, &p->real_parent->children);
1643                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1644                         attach_pid(p, PIDTYPE_PGID);
1645                         attach_pid(p, PIDTYPE_SID);
1646                         __this_cpu_inc(process_counts);
1647                 } else {
1648                         current->signal->nr_threads++;
1649                         atomic_inc(&current->signal->live);
1650                         atomic_inc(&current->signal->sigcnt);
1651                         list_add_tail_rcu(&p->thread_group,
1652                                           &p->group_leader->thread_group);
1653                         list_add_tail_rcu(&p->thread_node,
1654                                           &p->signal->thread_head);
1655                 }
1656                 attach_pid(p, PIDTYPE_PID);
1657                 nr_threads++;
1658         }
1659
1660         total_forks++;
1661         spin_unlock(&current->sighand->siglock);
1662         syscall_tracepoint_update(p);
1663         write_unlock_irq(&tasklist_lock);
1664
1665         proc_fork_connector(p);
1666         cgroup_post_fork(p, cgrp_ss_priv);
1667         threadgroup_change_end(current);
1668         perf_event_fork(p);
1669
1670         trace_task_newtask(p, clone_flags);
1671         uprobe_copy_process(p, clone_flags);
1672
1673         return p;
1674
1675 bad_fork_cancel_cgroup:
1676         cgroup_cancel_fork(p, cgrp_ss_priv);
1677 bad_fork_free_pid:
1678         threadgroup_change_end(current);
1679         if (pid != &init_struct_pid)
1680                 free_pid(pid);
1681 bad_fork_cleanup_io:
1682         if (p->io_context)
1683                 exit_io_context(p);
1684 bad_fork_cleanup_namespaces:
1685         exit_task_namespaces(p);
1686 bad_fork_cleanup_mm:
1687         if (p->mm)
1688                 mmput(p->mm);
1689 bad_fork_cleanup_signal:
1690         if (!(clone_flags & CLONE_THREAD))
1691                 free_signal_struct(p->signal);
1692 bad_fork_cleanup_sighand:
1693         __cleanup_sighand(p->sighand);
1694 bad_fork_cleanup_fs:
1695         exit_fs(p); /* blocking */
1696 bad_fork_cleanup_files:
1697         exit_files(p); /* blocking */
1698 bad_fork_cleanup_semundo:
1699         exit_sem(p);
1700 bad_fork_cleanup_audit:
1701         audit_free(p);
1702 bad_fork_cleanup_perf:
1703         perf_event_free_task(p);
1704 bad_fork_cleanup_policy:
1705 #ifdef CONFIG_NUMA
1706         mpol_put(p->mempolicy);
1707 bad_fork_cleanup_threadgroup_lock:
1708 #endif
1709         delayacct_tsk_free(p);
1710 bad_fork_cleanup_count:
1711         atomic_dec(&p->cred->user->processes);
1712         exit_creds(p);
1713 bad_fork_free:
1714         free_task(p);
1715 fork_out:
1716         return ERR_PTR(retval);
1717 }
1718
1719 static inline void init_idle_pids(struct pid_link *links)
1720 {
1721         enum pid_type type;
1722
1723         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1724                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1725                 links[type].pid = &init_struct_pid;
1726         }
1727 }
1728
1729 struct task_struct *fork_idle(int cpu)
1730 {
1731         struct task_struct *task;
1732         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0);
1733         if (!IS_ERR(task)) {
1734                 init_idle_pids(task->pids);
1735                 init_idle(task, cpu);
1736         }
1737
1738         return task;
1739 }
1740
1741 /*
1742  *  Ok, this is the main fork-routine.
1743  *
1744  * It copies the process, and if successful kick-starts
1745  * it and waits for it to finish using the VM if required.
1746  */
1747 long _do_fork(unsigned long clone_flags,
1748               unsigned long stack_start,
1749               unsigned long stack_size,
1750               int __user *parent_tidptr,
1751               int __user *child_tidptr,
1752               unsigned long tls)
1753 {
1754         struct task_struct *p;
1755         int trace = 0;
1756         long nr;
1757
1758         /*
1759          * Determine whether and which event to report to ptracer.  When
1760          * called from kernel_thread or CLONE_UNTRACED is explicitly
1761          * requested, no event is reported; otherwise, report if the event
1762          * for the type of forking is enabled.
1763          */
1764         if (!(clone_flags & CLONE_UNTRACED)) {
1765                 if (clone_flags & CLONE_VFORK)
1766                         trace = PTRACE_EVENT_VFORK;
1767                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1768                         trace = PTRACE_EVENT_CLONE;
1769                 else
1770                         trace = PTRACE_EVENT_FORK;
1771
1772                 if (likely(!ptrace_event_enabled(current, trace)))
1773                         trace = 0;
1774         }
1775
1776         p = copy_process(clone_flags, stack_start, stack_size,
1777                          child_tidptr, NULL, trace, tls);
1778         /*
1779          * Do this prior waking up the new thread - the thread pointer
1780          * might get invalid after that point, if the thread exits quickly.
1781          */
1782         if (!IS_ERR(p)) {
1783                 struct completion vfork;
1784                 struct pid *pid;
1785
1786                 trace_sched_process_fork(current, p);
1787
1788                 pid = get_task_pid(p, PIDTYPE_PID);
1789                 nr = pid_vnr(pid);
1790
1791                 if (clone_flags & CLONE_PARENT_SETTID)
1792                         put_user(nr, parent_tidptr);
1793
1794                 if (clone_flags & CLONE_VFORK) {
1795                         p->vfork_done = &vfork;
1796                         init_completion(&vfork);
1797                         get_task_struct(p);
1798                 }
1799
1800                 wake_up_new_task(p);
1801
1802                 /* forking complete and child started to run, tell ptracer */
1803                 if (unlikely(trace))
1804                         ptrace_event_pid(trace, pid);
1805
1806                 if (clone_flags & CLONE_VFORK) {
1807                         if (!wait_for_vfork_done(p, &vfork))
1808                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1809                 }
1810
1811                 put_pid(pid);
1812         } else {
1813                 nr = PTR_ERR(p);
1814         }
1815         return nr;
1816 }
1817
1818 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1819 /* For compatibility with architectures that call do_fork directly rather than
1820  * using the syscall entry points below. */
1821 long do_fork(unsigned long clone_flags,
1822               unsigned long stack_start,
1823               unsigned long stack_size,
1824               int __user *parent_tidptr,
1825               int __user *child_tidptr)
1826 {
1827         return _do_fork(clone_flags, stack_start, stack_size,
1828                         parent_tidptr, child_tidptr, 0);
1829 }
1830 #endif
1831
1832 /*
1833  * Create a kernel thread.
1834  */
1835 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1836 {
1837         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1838                 (unsigned long)arg, NULL, NULL, 0);
1839 }
1840
1841 #ifdef __ARCH_WANT_SYS_FORK
1842 SYSCALL_DEFINE0(fork)
1843 {
1844 #ifdef CONFIG_MMU
1845         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1846 #else
1847         /* can not support in nommu mode */
1848         return -EINVAL;
1849 #endif
1850 }
1851 #endif
1852
1853 #ifdef __ARCH_WANT_SYS_VFORK
1854 SYSCALL_DEFINE0(vfork)
1855 {
1856         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1857                         0, NULL, NULL, 0);
1858 }
1859 #endif
1860
1861 #ifdef __ARCH_WANT_SYS_CLONE
1862 #ifdef CONFIG_CLONE_BACKWARDS
1863 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1864                  int __user *, parent_tidptr,
1865                  unsigned long, tls,
1866                  int __user *, child_tidptr)
1867 #elif defined(CONFIG_CLONE_BACKWARDS2)
1868 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1869                  int __user *, parent_tidptr,
1870                  int __user *, child_tidptr,
1871                  unsigned long, tls)
1872 #elif defined(CONFIG_CLONE_BACKWARDS3)
1873 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1874                 int, stack_size,
1875                 int __user *, parent_tidptr,
1876                 int __user *, child_tidptr,
1877                 unsigned long, tls)
1878 #else
1879 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1880                  int __user *, parent_tidptr,
1881                  int __user *, child_tidptr,
1882                  unsigned long, tls)
1883 #endif
1884 {
1885         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
1886 }
1887 #endif
1888
1889 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1890 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1891 #endif
1892
1893 static void sighand_ctor(void *data)
1894 {
1895         struct sighand_struct *sighand = data;
1896
1897         spin_lock_init(&sighand->siglock);
1898         init_waitqueue_head(&sighand->signalfd_wqh);
1899 }
1900
1901 void __init proc_caches_init(void)
1902 {
1903         sighand_cachep = kmem_cache_create("sighand_cache",
1904                         sizeof(struct sighand_struct), 0,
1905                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1906                         SLAB_NOTRACK, sighand_ctor);
1907         signal_cachep = kmem_cache_create("signal_cache",
1908                         sizeof(struct signal_struct), 0,
1909                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1910         files_cachep = kmem_cache_create("files_cache",
1911                         sizeof(struct files_struct), 0,
1912                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1913         fs_cachep = kmem_cache_create("fs_cache",
1914                         sizeof(struct fs_struct), 0,
1915                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1916         /*
1917          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1918          * whole struct cpumask for the OFFSTACK case. We could change
1919          * this to *only* allocate as much of it as required by the
1920          * maximum number of CPU's we can ever have.  The cpumask_allocation
1921          * is at the end of the structure, exactly for that reason.
1922          */
1923         mm_cachep = kmem_cache_create("mm_struct",
1924                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1925                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1926         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1927         mmap_init();
1928         nsproxy_cache_init();
1929 }
1930
1931 /*
1932  * Check constraints on flags passed to the unshare system call.
1933  */
1934 static int check_unshare_flags(unsigned long unshare_flags)
1935 {
1936         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1937                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1938                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1939                                 CLONE_NEWUSER|CLONE_NEWPID))
1940                 return -EINVAL;
1941         /*
1942          * Not implemented, but pretend it works if there is nothing
1943          * to unshare.  Note that unsharing the address space or the
1944          * signal handlers also need to unshare the signal queues (aka
1945          * CLONE_THREAD).
1946          */
1947         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1948                 if (!thread_group_empty(current))
1949                         return -EINVAL;
1950         }
1951         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
1952                 if (atomic_read(&current->sighand->count) > 1)
1953                         return -EINVAL;
1954         }
1955         if (unshare_flags & CLONE_VM) {
1956                 if (!current_is_single_threaded())
1957                         return -EINVAL;
1958         }
1959
1960         return 0;
1961 }
1962
1963 /*
1964  * Unshare the filesystem structure if it is being shared
1965  */
1966 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1967 {
1968         struct fs_struct *fs = current->fs;
1969
1970         if (!(unshare_flags & CLONE_FS) || !fs)
1971                 return 0;
1972
1973         /* don't need lock here; in the worst case we'll do useless copy */
1974         if (fs->users == 1)
1975                 return 0;
1976
1977         *new_fsp = copy_fs_struct(fs);
1978         if (!*new_fsp)
1979                 return -ENOMEM;
1980
1981         return 0;
1982 }
1983
1984 /*
1985  * Unshare file descriptor table if it is being shared
1986  */
1987 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1988 {
1989         struct files_struct *fd = current->files;
1990         int error = 0;
1991
1992         if ((unshare_flags & CLONE_FILES) &&
1993             (fd && atomic_read(&fd->count) > 1)) {
1994                 *new_fdp = dup_fd(fd, &error);
1995                 if (!*new_fdp)
1996                         return error;
1997         }
1998
1999         return 0;
2000 }
2001
2002 /*
2003  * unshare allows a process to 'unshare' part of the process
2004  * context which was originally shared using clone.  copy_*
2005  * functions used by do_fork() cannot be used here directly
2006  * because they modify an inactive task_struct that is being
2007  * constructed. Here we are modifying the current, active,
2008  * task_struct.
2009  */
2010 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2011 {
2012         struct fs_struct *fs, *new_fs = NULL;
2013         struct files_struct *fd, *new_fd = NULL;
2014         struct cred *new_cred = NULL;
2015         struct nsproxy *new_nsproxy = NULL;
2016         int do_sysvsem = 0;
2017         int err;
2018
2019         /*
2020          * If unsharing a user namespace must also unshare the thread group
2021          * and unshare the filesystem root and working directories.
2022          */
2023         if (unshare_flags & CLONE_NEWUSER)
2024                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2025         /*
2026          * If unsharing vm, must also unshare signal handlers.
2027          */
2028         if (unshare_flags & CLONE_VM)
2029                 unshare_flags |= CLONE_SIGHAND;
2030         /*
2031          * If unsharing a signal handlers, must also unshare the signal queues.
2032          */
2033         if (unshare_flags & CLONE_SIGHAND)
2034                 unshare_flags |= CLONE_THREAD;
2035         /*
2036          * If unsharing namespace, must also unshare filesystem information.
2037          */
2038         if (unshare_flags & CLONE_NEWNS)
2039                 unshare_flags |= CLONE_FS;
2040
2041         err = check_unshare_flags(unshare_flags);
2042         if (err)
2043                 goto bad_unshare_out;
2044         /*
2045          * CLONE_NEWIPC must also detach from the undolist: after switching
2046          * to a new ipc namespace, the semaphore arrays from the old
2047          * namespace are unreachable.
2048          */
2049         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2050                 do_sysvsem = 1;
2051         err = unshare_fs(unshare_flags, &new_fs);
2052         if (err)
2053                 goto bad_unshare_out;
2054         err = unshare_fd(unshare_flags, &new_fd);
2055         if (err)
2056                 goto bad_unshare_cleanup_fs;
2057         err = unshare_userns(unshare_flags, &new_cred);
2058         if (err)
2059                 goto bad_unshare_cleanup_fd;
2060         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2061                                          new_cred, new_fs);
2062         if (err)
2063                 goto bad_unshare_cleanup_cred;
2064
2065         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2066                 if (do_sysvsem) {
2067                         /*
2068                          * CLONE_SYSVSEM is equivalent to sys_exit().
2069                          */
2070                         exit_sem(current);
2071                 }
2072                 if (unshare_flags & CLONE_NEWIPC) {
2073                         /* Orphan segments in old ns (see sem above). */
2074                         exit_shm(current);
2075                         shm_init_task(current);
2076                 }
2077
2078                 if (new_nsproxy)
2079                         switch_task_namespaces(current, new_nsproxy);
2080
2081                 task_lock(current);
2082
2083                 if (new_fs) {
2084                         fs = current->fs;
2085                         spin_lock(&fs->lock);
2086                         current->fs = new_fs;
2087                         if (--fs->users)
2088                                 new_fs = NULL;
2089                         else
2090                                 new_fs = fs;
2091                         spin_unlock(&fs->lock);
2092                 }
2093
2094                 if (new_fd) {
2095                         fd = current->files;
2096                         current->files = new_fd;
2097                         new_fd = fd;
2098                 }
2099
2100                 task_unlock(current);
2101
2102                 if (new_cred) {
2103                         /* Install the new user namespace */
2104                         commit_creds(new_cred);
2105                         new_cred = NULL;
2106                 }
2107         }
2108
2109 bad_unshare_cleanup_cred:
2110         if (new_cred)
2111                 put_cred(new_cred);
2112 bad_unshare_cleanup_fd:
2113         if (new_fd)
2114                 put_files_struct(new_fd);
2115
2116 bad_unshare_cleanup_fs:
2117         if (new_fs)
2118                 free_fs_struct(new_fs);
2119
2120 bad_unshare_out:
2121         return err;
2122 }
2123
2124 /*
2125  *      Helper to unshare the files of the current task.
2126  *      We don't want to expose copy_files internals to
2127  *      the exec layer of the kernel.
2128  */
2129
2130 int unshare_files(struct files_struct **displaced)
2131 {
2132         struct task_struct *task = current;
2133         struct files_struct *copy = NULL;
2134         int error;
2135
2136         error = unshare_fd(CLONE_FILES, &copy);
2137         if (error || !copy) {
2138                 *displaced = NULL;
2139                 return error;
2140         }
2141         *displaced = task->files;
2142         task_lock(task);
2143         task->files = copy;
2144         task_unlock(task);
2145         return 0;
2146 }
2147
2148 int sysctl_max_threads(struct ctl_table *table, int write,
2149                        void __user *buffer, size_t *lenp, loff_t *ppos)
2150 {
2151         struct ctl_table t;
2152         int ret;
2153         int threads = max_threads;
2154         int min = MIN_THREADS;
2155         int max = MAX_THREADS;
2156
2157         t = *table;
2158         t.data = &threads;
2159         t.extra1 = &min;
2160         t.extra2 = &max;
2161
2162         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2163         if (ret || !write)
2164                 return ret;
2165
2166         set_max_threads(threads);
2167
2168         return 0;
2169 }