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