kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/mm.h>
  32 #include <linux/vmacache.h>
  33 #include <linux/nsproxy.h>
  34 #include <linux/capability.h>
  35 #include <linux/cpu.h>
  36 #include <linux/cgroup.h>
  37 #include <linux/security.h>
  38 #include <linux/hugetlb.h>
  39 #include <linux/seccomp.h>
  40 #include <linux/swap.h>
  41 #include <linux/syscalls.h>
  42 #include <linux/jiffies.h>
  43 #include <linux/futex.h>
  44 #include <linux/compat.h>
  45 #include <linux/kthread.h>
  46 #include <linux/task_io_accounting_ops.h>
  47 #include <linux/rcupdate.h>
  48 #include <linux/ptrace.h>
  49 #include <linux/mount.h>
  50 #include <linux/audit.h>
  51 #include <linux/memcontrol.h>
  52 #include <linux/ftrace.h>
  53 #include <linux/proc_fs.h>
  54 #include <linux/profile.h>
  55 #include <linux/rmap.h>
  56 #include <linux/ksm.h>
  57 #include <linux/acct.h>
  58 #include <linux/tsacct_kern.h>
  59 #include <linux/cn_proc.h>
  60 #include <linux/freezer.h>
  61 #include <linux/delayacct.h>
  62 #include <linux/taskstats_kern.h>
  63 #include <linux/random.h>
  64 #include <linux/tty.h>
  65 #include <linux/blkdev.h>
  66 #include <linux/fs_struct.h>
  67 #include <linux/magic.h>
  68 #include <linux/perf_event.h>
  69 #include <linux/posix-timers.h>
  70 #include <linux/user-return-notifier.h>
  71 #include <linux/oom.h>
  72 #include <linux/khugepaged.h>
  73 #include <linux/signalfd.h>
  74 #include <linux/uprobes.h>
  75 #include <linux/aio.h>
  76 #include <linux/compiler.h>
  77 #include <linux/sysctl.h>
  78
  79 #include <asm/pgtable.h>
  80 #include <asm/pgalloc.h>
  81 #include <asm/uaccess.h>
  82 #include <asm/mmu_context.h>
  83 #include <asm/cacheflush.h>
  84 #include <asm/tlbflush.h>
  85
  86 #include <trace/events/sched.h>
  87
  88 #define CREATE_TRACE_POINTS
  89 #include <trace/events/task.h>
  90
  91 /*
  92  * Minimum number of threads to boot the kernel
  93  */
  94 #define MIN_THREADS 20
  95
  96 /*
  97  * Maximum number of threads
  98  */
  99 #define MAX_THREADS FUTEX_TID_MASK
 100
 101 /*
 102  * Protected counters by write_lock_irq(&tasklist_lock)
 103  */
 104 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 105 int nr_threads;                 /* The idle threads do not count.. */
 106
 107 int max_threads;                /* tunable limit on nr_threads */
 108
 109 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 110
 111 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 112
 113 #ifdef CONFIG_PROVE_RCU
 114 int lockdep_tasklist_lock_is_held(void)
 115 {
 116         return lockdep_is_held(&tasklist_lock);
 117 }
 118 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 119 #endif /* #ifdef CONFIG_PROVE_RCU */
 120
 121 int nr_processes(void)
 122 {
 123         int cpu;
 124         int total = 0;
 125
 126         for_each_possible_cpu(cpu)
 127                 total += per_cpu(process_counts, cpu);
 128
 129         return total;
 130 }
 131
 132 void __weak arch_release_task_struct(struct task_struct *tsk)
 133 {
 134 }
 135
 136 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 137 static struct kmem_cache *task_struct_cachep;
 138
 139 static inline struct task_struct *alloc_task_struct_node(int node)
 140 {
 141         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 142 }
 143
 144 static inline void free_task_struct(struct task_struct *tsk)
 145 {
 146         kmem_cache_free(task_struct_cachep, tsk);
 147 }
 148 #endif
 149
 150 void __weak arch_release_thread_info(struct thread_info *ti)
 151 {
 152 }
 153
 154 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
 155
 156 /*
 157  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 158  * kmemcache based allocator.
 159  */
 160 # if THREAD_SIZE >= PAGE_SIZE
 161 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 162                                                   int node)
 163 {
 164         struct page *page = alloc_kmem_pages_node(node, THREADINFO_GFP,
 165                                                   THREAD_SIZE_ORDER);
 166
 167         return page ? page_address(page) : NULL;
 168 }
 169
 170 static inline void free_thread_info(struct thread_info *ti)
 171 {
 172         free_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 173 }
 174 # else
 175 static struct kmem_cache *thread_info_cache;
 176
 177 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 178                                                   int node)
 179 {
 180         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
 181 }
 182
 183 static void free_thread_info(struct thread_info *ti)
 184 {
 185         kmem_cache_free(thread_info_cache, ti);
 186 }
 187
 188 void thread_info_cache_init(void)
 189 {
 190         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
 191                                               THREAD_SIZE, 0, NULL);
 192         BUG_ON(thread_info_cache == NULL);
 193 }
 194 # endif
 195 #endif
 196
 197 /* SLAB cache for signal_struct structures (tsk->signal) */
 198 static struct kmem_cache *signal_cachep;
 199
 200 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 201 struct kmem_cache *sighand_cachep;
 202
 203 /* SLAB cache for files_struct structures (tsk->files) */
 204 struct kmem_cache *files_cachep;
 205
 206 /* SLAB cache for fs_struct structures (tsk->fs) */
 207 struct kmem_cache *fs_cachep;
 208
 209 /* SLAB cache for vm_area_struct structures */
 210 struct kmem_cache *vm_area_cachep;
 211
 212 /* SLAB cache for mm_struct structures (tsk->mm) */
 213 static struct kmem_cache *mm_cachep;
 214
 215 static void account_kernel_stack(struct thread_info *ti, int account)
 216 {
 217         struct zone *zone = page_zone(virt_to_page(ti));
 218
 219         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 220 }
 221
 222 void free_task(struct task_struct *tsk)
 223 {
 224         account_kernel_stack(tsk->stack, -1);
 225         arch_release_thread_info(tsk->stack);
 226         free_thread_info(tsk->stack);
 227         rt_mutex_debug_task_free(tsk);
 228         ftrace_graph_exit_task(tsk);
 229         put_seccomp_filter(tsk);
 230         arch_release_task_struct(tsk);
 231         free_task_struct(tsk);
 232 }
 233 EXPORT_SYMBOL(free_task);
 234
 235 static inline void free_signal_struct(struct signal_struct *sig)
 236 {
 237         taskstats_tgid_free(sig);
 238         sched_autogroup_exit(sig);
 239         kmem_cache_free(signal_cachep, sig);
 240 }
 241
 242 static inline void put_signal_struct(struct signal_struct *sig)
 243 {
 244         if (atomic_dec_and_test(&sig->sigcnt))
 245                 free_signal_struct(sig);
 246 }
 247
 248 void __put_task_struct(struct task_struct *tsk)
 249 {
 250         WARN_ON(!tsk->exit_state);
 251         WARN_ON(atomic_read(&tsk->usage));
 252         WARN_ON(tsk == current);
 253
 254         cgroup_free(tsk);
 255         task_numa_free(tsk);
 256         security_task_free(tsk);
 257         exit_creds(tsk);
 258         delayacct_tsk_free(tsk);
 259         put_signal_struct(tsk->signal);
 260
 261         if (!profile_handoff_task(tsk))
 262                 free_task(tsk);
 263 }
 264 EXPORT_SYMBOL_GPL(__put_task_struct);
 265
 266 void __init __weak arch_task_cache_init(void) { }
 267
 268 /*
 269  * set_max_threads
 270  */
 271 static void set_max_threads(unsigned int max_threads_suggested)
 272 {
 273         u64 threads;
 274
 275         /*
 276          * The number of threads shall be limited such that the thread
 277          * structures may only consume a small part of the available memory.
 278          */
 279         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 280                 threads = MAX_THREADS;
 281         else
 282                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 283                                     (u64) THREAD_SIZE * 8UL);
 284
 285         if (threads > max_threads_suggested)
 286                 threads = max_threads_suggested;
 287
 288         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 289 }
 290
 291 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 292 /* Initialized by the architecture: */
 293 int arch_task_struct_size __read_mostly;
 294 #endif
 295
 296 void __init fork_init(void)
 297 {
 298 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 299 #ifndef ARCH_MIN_TASKALIGN
 300 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 301 #endif
 302         /* create a slab on which task_structs can be allocated */
 303         task_struct_cachep =
 304                 kmem_cache_create("task_struct", arch_task_struct_size,
 305                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 306 #endif
 307
 308         /* do the arch specific task caches init */
 309         arch_task_cache_init();
 310
 311         set_max_threads(MAX_THREADS);
 312
 313         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 314         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 315         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 316                 init_task.signal->rlim[RLIMIT_NPROC];
 317 }
 318
 319 int __weak arch_dup_task_struct(struct task_struct *dst,
 320                                                struct task_struct *src)
 321 {
 322         *dst = *src;
 323         return 0;
 324 }
 325
 326 void set_task_stack_end_magic(struct task_struct *tsk)
 327 {
 328         unsigned long *stackend;
 329
 330         stackend = end_of_stack(tsk);
 331         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 332 }
 333
 334 static struct task_struct *dup_task_struct(struct task_struct *orig)
 335 {
 336         struct task_struct *tsk;
 337         struct thread_info *ti;
 338         int node = tsk_fork_get_node(orig);
 339         int err;
 340
 341         tsk = alloc_task_struct_node(node);
 342         if (!tsk)
 343                 return NULL;
 344
 345         ti = alloc_thread_info_node(tsk, node);
 346         if (!ti)
 347                 goto free_tsk;
 348
 349         err = arch_dup_task_struct(tsk, orig);
 350         if (err)
 351                 goto free_ti;
 352
 353         tsk->stack = ti;
 354 #ifdef CONFIG_SECCOMP
 355         /*
 356          * We must handle setting up seccomp filters once we're under
 357          * the sighand lock in case orig has changed between now and
 358          * then. Until then, filter must be NULL to avoid messing up
 359          * the usage counts on the error path calling free_task.
 360          */
 361         tsk->seccomp.filter = NULL;
 362 #endif
 363
 364         setup_thread_stack(tsk, orig);
 365         clear_user_return_notifier(tsk);
 366         clear_tsk_need_resched(tsk);
 367         set_task_stack_end_magic(tsk);
 368
 369 #ifdef CONFIG_CC_STACKPROTECTOR
 370         tsk->stack_canary = get_random_int();
 371 #endif
 372
 373         /*
 374          * One for us, one for whoever does the "release_task()" (usually
 375          * parent)
 376          */
 377         atomic_set(&tsk->usage, 2);
 378 #ifdef CONFIG_BLK_DEV_IO_TRACE
 379         tsk->btrace_seq = 0;
 380 #endif
 381         tsk->splice_pipe = NULL;
 382         tsk->task_frag.page = NULL;
 383         tsk->wake_q.next = NULL;
 384
 385         account_kernel_stack(ti, 1);
 386
 387         return tsk;
 388
 389 free_ti:
 390         free_thread_info(ti);
 391 free_tsk:
 392         free_task_struct(tsk);
 393         return NULL;
 394 }
 395
 396 #ifdef CONFIG_MMU
 397 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 398 {
 399         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 400         struct rb_node **rb_link, *rb_parent;
 401         int retval;
 402         unsigned long charge;
 403
 404         uprobe_start_dup_mmap();
 405         down_write(&oldmm->mmap_sem);
 406         flush_cache_dup_mm(oldmm);
 407         uprobe_dup_mmap(oldmm, mm);
 408         /*
 409          * Not linked in yet - no deadlock potential:
 410          */
 411         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 412
 413         /* No ordering required: file already has been exposed. */
 414         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 415
 416         mm->total_vm = oldmm->total_vm;
 417         mm->shared_vm = oldmm->shared_vm;
 418         mm->exec_vm = oldmm->exec_vm;
 419         mm->stack_vm = oldmm->stack_vm;
 420
 421         rb_link = &mm->mm_rb.rb_node;
 422         rb_parent = NULL;
 423         pprev = &mm->mmap;
 424         retval = ksm_fork(mm, oldmm);
 425         if (retval)
 426                 goto out;
 427         retval = khugepaged_fork(mm, oldmm);
 428         if (retval)
 429                 goto out;
 430
 431         prev = NULL;
 432         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 433                 struct file *file;
 434
 435                 if (mpnt->vm_flags & VM_DONTCOPY) {
 436                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 437                                                         -vma_pages(mpnt));
 438                         continue;
 439                 }
 440                 charge = 0;
 441                 if (mpnt->vm_flags & VM_ACCOUNT) {
 442                         unsigned long len = vma_pages(mpnt);
 443
 444                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 445                                 goto fail_nomem;
 446                         charge = len;
 447                 }
 448                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 449                 if (!tmp)
 450                         goto fail_nomem;
 451                 *tmp = *mpnt;
 452                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 453                 retval = vma_dup_policy(mpnt, tmp);
 454                 if (retval)
 455                         goto fail_nomem_policy;
 456                 tmp->vm_mm = mm;
 457                 if (anon_vma_fork(tmp, mpnt))
 458                         goto fail_nomem_anon_vma_fork;
 459                 tmp->vm_flags &=
 460                         ~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
 461                 tmp->vm_next = tmp->vm_prev = NULL;
 462                 tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
 463                 file = tmp->vm_file;
 464                 if (file) {
 465                         struct inode *inode = file_inode(file);
 466                         struct address_space *mapping = file->f_mapping;
 467
 468                         get_file(file);
 469                         if (tmp->vm_flags & VM_DENYWRITE)
 470                                 atomic_dec(&inode->i_writecount);
 471                         i_mmap_lock_write(mapping);
 472                         if (tmp->vm_flags & VM_SHARED)
 473                                 atomic_inc(&mapping->i_mmap_writable);
 474                         flush_dcache_mmap_lock(mapping);
 475                         /* insert tmp into the share list, just after mpnt */
 476                         vma_interval_tree_insert_after(tmp, mpnt,
 477                                         &mapping->i_mmap);
 478                         flush_dcache_mmap_unlock(mapping);
 479                         i_mmap_unlock_write(mapping);
 480                 }
 481
 482                 /*
 483                  * Clear hugetlb-related page reserves for children. This only
 484                  * affects MAP_PRIVATE mappings. Faults generated by the child
 485                  * are not guaranteed to succeed, even if read-only
 486                  */
 487                 if (is_vm_hugetlb_page(tmp))
 488                         reset_vma_resv_huge_pages(tmp);
 489
 490                 /*
 491                  * Link in the new vma and copy the page table entries.
 492                  */
 493                 *pprev = tmp;
 494                 pprev = &tmp->vm_next;
 495                 tmp->vm_prev = prev;
 496                 prev = tmp;
 497
 498                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 499                 rb_link = &tmp->vm_rb.rb_right;
 500                 rb_parent = &tmp->vm_rb;
 501
 502                 mm->map_count++;
 503                 retval = copy_page_range(mm, oldmm, mpnt);
 504
 505                 if (tmp->vm_ops && tmp->vm_ops->open)
 506                         tmp->vm_ops->open(tmp);
 507
 508                 if (retval)
 509                         goto out;
 510         }
 511         /* a new mm has just been created */
 512         arch_dup_mmap(oldmm, mm);
 513         retval = 0;
 514 out:
 515         up_write(&mm->mmap_sem);
 516         flush_tlb_mm(oldmm);
 517         up_write(&oldmm->mmap_sem);
 518         uprobe_end_dup_mmap();
 519         return retval;
 520 fail_nomem_anon_vma_fork:
 521         mpol_put(vma_policy(tmp));
 522 fail_nomem_policy:
 523         kmem_cache_free(vm_area_cachep, tmp);
 524 fail_nomem:
 525         retval = -ENOMEM;
 526         vm_unacct_memory(charge);
 527         goto out;
 528 }
 529
 530 static inline int mm_alloc_pgd(struct mm_struct *mm)
 531 {
 532         mm->pgd = pgd_alloc(mm);
 533         if (unlikely(!mm->pgd))
 534                 return -ENOMEM;
 535         return 0;
 536 }
 537
 538 static inline void mm_free_pgd(struct mm_struct *mm)
 539 {
 540         pgd_free(mm, mm->pgd);
 541 }
 542 #else
 543 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 544 {
 545         down_write(&oldmm->mmap_sem);
 546         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 547         up_write(&oldmm->mmap_sem);
 548         return 0;
 549 }
 550 #define mm_alloc_pgd(mm)        (0)
 551 #define mm_free_pgd(mm)
 552 #endif /* CONFIG_MMU */
 553
 554 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 555
 556 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 557 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 558
 559 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 560
 561 static int __init coredump_filter_setup(char *s)
 562 {
 563         default_dump_filter =
 564                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 565                 MMF_DUMP_FILTER_MASK;
 566         return 1;
 567 }
 568
 569 __setup("coredump_filter=", coredump_filter_setup);
 570
 571 #include <linux/init_task.h>
 572
 573 static void mm_init_aio(struct mm_struct *mm)
 574 {
 575 #ifdef CONFIG_AIO
 576         spin_lock_init(&mm->ioctx_lock);
 577         mm->ioctx_table = NULL;
 578 #endif
 579 }
 580
 581 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 582 {
 583 #ifdef CONFIG_MEMCG
 584         mm->owner = p;
 585 #endif
 586 }
 587
 588 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 589         struct user_namespace *user_ns)
 590 {
 591         mm->mmap = NULL;
 592         mm->mm_rb = RB_ROOT;
 593         mm->vmacache_seqnum = 0;
 594         atomic_set(&mm->mm_users, 1);
 595         atomic_set(&mm->mm_count, 1);
 596         init_rwsem(&mm->mmap_sem);
 597         INIT_LIST_HEAD(&mm->mmlist);
 598         mm->core_state = NULL;
 599         atomic_long_set(&mm->nr_ptes, 0);
 600         mm_nr_pmds_init(mm);
 601         mm->map_count = 0;
 602         mm->locked_vm = 0;
 603         mm->pinned_vm = 0;
 604         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 605         spin_lock_init(&mm->page_table_lock);
 606         mm_init_cpumask(mm);
 607         mm_init_aio(mm);
 608         mm_init_owner(mm, p);
 609         mmu_notifier_mm_init(mm);
 610         clear_tlb_flush_pending(mm);
 611 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 612         mm->pmd_huge_pte = NULL;
 613 #endif
 614
 615         if (current->mm) {
 616                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 617                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 618         } else {
 619                 mm->flags = default_dump_filter;
 620                 mm->def_flags = 0;
 621         }
 622
 623         if (mm_alloc_pgd(mm))
 624                 goto fail_nopgd;
 625
 626         if (init_new_context(p, mm))
 627                 goto fail_nocontext;
 628
 629         mm->user_ns = get_user_ns(user_ns);
 630         return mm;
 631
 632 fail_nocontext:
 633         mm_free_pgd(mm);
 634 fail_nopgd:
 635         free_mm(mm);
 636         return NULL;
 637 }
 638
 639 static void check_mm(struct mm_struct *mm)
 640 {
 641         int i;
 642
 643         for (i = 0; i < NR_MM_COUNTERS; i++) {
 644                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 645
 646                 if (unlikely(x))
 647                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 648                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 649         }
 650
 651         if (atomic_long_read(&mm->nr_ptes))
 652                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 653                                 atomic_long_read(&mm->nr_ptes));
 654         if (mm_nr_pmds(mm))
 655                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 656                                 mm_nr_pmds(mm));
 657
 658 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 659         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 660 #endif
 661 }
 662
 663 /*
 664  * Allocate and initialize an mm_struct.
 665  */
 666 struct mm_struct *mm_alloc(void)
 667 {
 668         struct mm_struct *mm;
 669
 670         mm = allocate_mm();
 671         if (!mm)
 672                 return NULL;
 673
 674         memset(mm, 0, sizeof(*mm));
 675         return mm_init(mm, current, current_user_ns());
 676 }
 677
 678 /*
 679  * Called when the last reference to the mm
 680  * is dropped: either by a lazy thread or by
 681  * mmput. Free the page directory and the mm.
 682  */
 683 void __mmdrop(struct mm_struct *mm)
 684 {
 685         BUG_ON(mm == &init_mm);
 686         mm_free_pgd(mm);
 687         destroy_context(mm);
 688         mmu_notifier_mm_destroy(mm);
 689         check_mm(mm);
 690         put_user_ns(mm->user_ns);
 691         free_mm(mm);
 692 }
 693 EXPORT_SYMBOL_GPL(__mmdrop);
 694
 695 /*
 696  * Decrement the use count and release all resources for an mm.
 697  */
 698 void mmput(struct mm_struct *mm)
 699 {
 700         might_sleep();
 701
 702         if (atomic_dec_and_test(&mm->mm_users)) {
 703                 uprobe_clear_state(mm);
 704                 exit_aio(mm);
 705                 ksm_exit(mm);
 706                 khugepaged_exit(mm); /* must run before exit_mmap */
 707                 exit_mmap(mm);
 708                 set_mm_exe_file(mm, NULL);
 709                 if (!list_empty(&mm->mmlist)) {
 710                         spin_lock(&mmlist_lock);
 711                         list_del(&mm->mmlist);
 712                         spin_unlock(&mmlist_lock);
 713                 }
 714                 if (mm->binfmt)
 715                         module_put(mm->binfmt->module);
 716                 mmdrop(mm);
 717         }
 718 }
 719 EXPORT_SYMBOL_GPL(mmput);
 720
 721 /**
 722  * set_mm_exe_file - change a reference to the mm's executable file
 723  *
 724  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 725  *
 726  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 727  * invocations: in mmput() nobody alive left, in execve task is single
 728  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 729  * mm->exe_file, but does so without using set_mm_exe_file() in order
 730  * to do avoid the need for any locks.
 731  */
 732 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 733 {
 734         struct file *old_exe_file;
 735
 736         /*
 737          * It is safe to dereference the exe_file without RCU as
 738          * this function is only called if nobody else can access
 739          * this mm -- see comment above for justification.
 740          */
 741         old_exe_file = rcu_dereference_raw(mm->exe_file);
 742
 743         if (new_exe_file)
 744                 get_file(new_exe_file);
 745         rcu_assign_pointer(mm->exe_file, new_exe_file);
 746         if (old_exe_file)
 747                 fput(old_exe_file);
 748 }
 749
 750 /**
 751  * get_mm_exe_file - acquire a reference to the mm's executable file
 752  *
 753  * Returns %NULL if mm has no associated executable file.
 754  * User must release file via fput().
 755  */
 756 struct file *get_mm_exe_file(struct mm_struct *mm)
 757 {
 758         struct file *exe_file;
 759
 760         rcu_read_lock();
 761         exe_file = rcu_dereference(mm->exe_file);
 762         if (exe_file && !get_file_rcu(exe_file))
 763                 exe_file = NULL;
 764         rcu_read_unlock();
 765         return exe_file;
 766 }
 767 EXPORT_SYMBOL(get_mm_exe_file);
 768
 769 /**
 770  * get_task_exe_file - acquire a reference to the task's executable file
 771  *
 772  * Returns %NULL if task's mm (if any) has no associated executable file or
 773  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
 774  * User must release file via fput().
 775  */
 776 struct file *get_task_exe_file(struct task_struct *task)
 777 {
 778         struct file *exe_file = NULL;
 779         struct mm_struct *mm;
 780
 781         task_lock(task);
 782         mm = task->mm;
 783         if (mm) {
 784                 if (!(task->flags & PF_KTHREAD))
 785                         exe_file = get_mm_exe_file(mm);
 786         }
 787         task_unlock(task);
 788         return exe_file;
 789 }
 790 EXPORT_SYMBOL(get_task_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                         !capable(CAP_SYS_RESOURCE)) {
 831                 mmput(mm);
 832                 mm = ERR_PTR(-EACCES);
 833         }
 834         mutex_unlock(&task->signal->cred_guard_mutex);
 835
 836         return mm;
 837 }
 838
 839 static void complete_vfork_done(struct task_struct *tsk)
 840 {
 841         struct completion *vfork;
 842
 843         task_lock(tsk);
 844         vfork = tsk->vfork_done;
 845         if (likely(vfork)) {
 846                 tsk->vfork_done = NULL;
 847                 complete(vfork);
 848         }
 849         task_unlock(tsk);
 850 }
 851
 852 static int wait_for_vfork_done(struct task_struct *child,
 853                                 struct completion *vfork)
 854 {
 855         int killed;
 856
 857         freezer_do_not_count();
 858         killed = wait_for_completion_killable(vfork);
 859         freezer_count();
 860
 861         if (killed) {
 862                 task_lock(child);
 863                 child->vfork_done = NULL;
 864                 task_unlock(child);
 865         }
 866
 867         put_task_struct(child);
 868         return killed;
 869 }
 870
 871 /* Please note the differences between mmput and mm_release.
 872  * mmput is called whenever we stop holding onto a mm_struct,
 873  * error success whatever.
 874  *
 875  * mm_release is called after a mm_struct has been removed
 876  * from the current process.
 877  *
 878  * This difference is important for error handling, when we
 879  * only half set up a mm_struct for a new process and need to restore
 880  * the old one.  Because we mmput the new mm_struct before
 881  * restoring the old one. . .
 882  * Eric Biederman 10 January 1998
 883  */
 884 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 885 {
 886         /* Get rid of any futexes when releasing the mm */
 887 #ifdef CONFIG_FUTEX
 888         if (unlikely(tsk->robust_list)) {
 889                 exit_robust_list(tsk);
 890                 tsk->robust_list = NULL;
 891         }
 892 #ifdef CONFIG_COMPAT
 893         if (unlikely(tsk->compat_robust_list)) {
 894                 compat_exit_robust_list(tsk);
 895                 tsk->compat_robust_list = NULL;
 896         }
 897 #endif
 898         if (unlikely(!list_empty(&tsk->pi_state_list)))
 899                 exit_pi_state_list(tsk);
 900 #endif
 901
 902         uprobe_free_utask(tsk);
 903
 904         /* Get rid of any cached register state */
 905         deactivate_mm(tsk, mm);
 906
 907         /*
 908          * Signal userspace if we're not exiting with a core dump
 909          * because we want to leave the value intact for debugging
 910          * purposes.
 911          */
 912         if (tsk->clear_child_tid) {
 913                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
 914                     atomic_read(&mm->mm_users) > 1) {
 915                         /*
 916                          * We don't check the error code - if userspace has
 917                          * not set up a proper pointer then tough luck.
 918                          */
 919                         put_user(0, tsk->clear_child_tid);
 920                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 921                                         1, NULL, NULL, 0);
 922                 }
 923                 tsk->clear_child_tid = NULL;
 924         }
 925
 926         /*
 927          * All done, finally we can wake up parent and return this mm to him.
 928          * Also kthread_stop() uses this completion for synchronization.
 929          */
 930         if (tsk->vfork_done)
 931                 complete_vfork_done(tsk);
 932 }
 933
 934 /*
 935  * Allocate a new mm structure and copy contents from the
 936  * mm structure of the passed in task structure.
 937  */
 938 static struct mm_struct *dup_mm(struct task_struct *tsk)
 939 {
 940         struct mm_struct *mm, *oldmm = current->mm;
 941         int err;
 942
 943         mm = allocate_mm();
 944         if (!mm)
 945                 goto fail_nomem;
 946
 947         memcpy(mm, oldmm, sizeof(*mm));
 948
 949         if (!mm_init(mm, tsk, mm->user_ns))
 950                 goto fail_nomem;
 951
 952         err = dup_mmap(mm, oldmm);
 953         if (err)
 954                 goto free_pt;
 955
 956         mm->hiwater_rss = get_mm_rss(mm);
 957         mm->hiwater_vm = mm->total_vm;
 958
 959         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 960                 goto free_pt;
 961
 962         return mm;
 963
 964 free_pt:
 965         /* don't put binfmt in mmput, we haven't got module yet */
 966         mm->binfmt = NULL;
 967         mmput(mm);
 968
 969 fail_nomem:
 970         return NULL;
 971 }
 972
 973 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 974 {
 975         struct mm_struct *mm, *oldmm;
 976         int retval;
 977
 978         tsk->min_flt = tsk->maj_flt = 0;
 979         tsk->nvcsw = tsk->nivcsw = 0;
 980 #ifdef CONFIG_DETECT_HUNG_TASK
 981         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 982 #endif
 983
 984         tsk->mm = NULL;
 985         tsk->active_mm = NULL;
 986
 987         /*
 988          * Are we cloning a kernel thread?
 989          *
 990          * We need to steal a active VM for that..
 991          */
 992         oldmm = current->mm;
 993         if (!oldmm)
 994                 return 0;
 995
 996         /* initialize the new vmacache entries */
 997         vmacache_flush(tsk);
 998
 999         if (clone_flags & CLONE_VM) {
1000                 atomic_inc(&oldmm->mm_users);
1001                 mm = oldmm;
1002                 goto good_mm;
1003         }
1004
1005         retval = -ENOMEM;
1006         mm = dup_mm(tsk);
1007         if (!mm)
1008                 goto fail_nomem;
1009
1010 good_mm:
1011         tsk->mm = mm;
1012         tsk->active_mm = mm;
1013         return 0;
1014
1015 fail_nomem:
1016         return retval;
1017 }
1018
1019 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1020 {
1021         struct fs_struct *fs = current->fs;
1022         if (clone_flags & CLONE_FS) {
1023                 /* tsk->fs is already what we want */
1024                 spin_lock(&fs->lock);
1025                 if (fs->in_exec) {
1026                         spin_unlock(&fs->lock);
1027                         return -EAGAIN;
1028                 }
1029                 fs->users++;
1030                 spin_unlock(&fs->lock);
1031                 return 0;
1032         }
1033         tsk->fs = copy_fs_struct(fs);
1034         if (!tsk->fs)
1035                 return -ENOMEM;
1036         return 0;
1037 }
1038
1039 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1040 {
1041         struct files_struct *oldf, *newf;
1042         int error = 0;
1043
1044         /*
1045          * A background process may not have any files ...
1046          */
1047         oldf = current->files;
1048         if (!oldf)
1049                 goto out;
1050
1051         if (clone_flags & CLONE_FILES) {
1052                 atomic_inc(&oldf->count);
1053                 goto out;
1054         }
1055
1056         newf = dup_fd(oldf, &error);
1057         if (!newf)
1058                 goto out;
1059
1060         tsk->files = newf;
1061         error = 0;
1062 out:
1063         return error;
1064 }
1065
1066 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1067 {
1068 #ifdef CONFIG_BLOCK
1069         struct io_context *ioc = current->io_context;
1070         struct io_context *new_ioc;
1071
1072         if (!ioc)
1073                 return 0;
1074         /*
1075          * Share io context with parent, if CLONE_IO is set
1076          */
1077         if (clone_flags & CLONE_IO) {
1078                 ioc_task_link(ioc);
1079                 tsk->io_context = ioc;
1080         } else if (ioprio_valid(ioc->ioprio)) {
1081                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1082                 if (unlikely(!new_ioc))
1083                         return -ENOMEM;
1084
1085                 new_ioc->ioprio = ioc->ioprio;
1086                 put_io_context(new_ioc);
1087         }
1088 #endif
1089         return 0;
1090 }
1091
1092 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1093 {
1094         struct sighand_struct *sig;
1095
1096         if (clone_flags & CLONE_SIGHAND) {
1097                 atomic_inc(&current->sighand->count);
1098                 return 0;
1099         }
1100         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1101         rcu_assign_pointer(tsk->sighand, sig);
1102         if (!sig)
1103                 return -ENOMEM;
1104
1105         atomic_set(&sig->count, 1);
1106         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1107         return 0;
1108 }
1109
1110 void __cleanup_sighand(struct sighand_struct *sighand)
1111 {
1112         if (atomic_dec_and_test(&sighand->count)) {
1113                 signalfd_cleanup(sighand);
1114                 /*
1115                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1116                  * without an RCU grace period, see __lock_task_sighand().
1117                  */
1118                 kmem_cache_free(sighand_cachep, sighand);
1119         }
1120 }
1121
1122 /*
1123  * Initialize POSIX timer handling for a thread group.
1124  */
1125 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1126 {
1127         unsigned long cpu_limit;
1128
1129         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1130         if (cpu_limit != RLIM_INFINITY) {
1131                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1132                 sig->cputimer.running = true;
1133         }
1134
1135         /* The timer lists. */
1136         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1137         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1138         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1139 }
1140
1141 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1142 {
1143         struct signal_struct *sig;
1144
1145         if (clone_flags & CLONE_THREAD)
1146                 return 0;
1147
1148         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1149         tsk->signal = sig;
1150         if (!sig)
1151                 return -ENOMEM;
1152
1153         sig->nr_threads = 1;
1154         atomic_set(&sig->live, 1);
1155         atomic_set(&sig->sigcnt, 1);
1156
1157         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1158         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1159         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1160
1161         init_waitqueue_head(&sig->wait_chldexit);
1162         sig->curr_target = tsk;
1163         init_sigpending(&sig->shared_pending);
1164         INIT_LIST_HEAD(&sig->posix_timers);
1165         seqlock_init(&sig->stats_lock);
1166         prev_cputime_init(&sig->prev_cputime);
1167
1168         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1169         sig->real_timer.function = it_real_fn;
1170
1171         task_lock(current->group_leader);
1172         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1173         task_unlock(current->group_leader);
1174
1175         posix_cpu_timers_init_group(sig);
1176
1177         tty_audit_fork(sig);
1178         sched_autogroup_fork(sig);
1179
1180         sig->oom_score_adj = current->signal->oom_score_adj;
1181         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1182
1183         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1184                                    current->signal->is_child_subreaper;
1185
1186         mutex_init(&sig->cred_guard_mutex);
1187
1188         return 0;
1189 }
1190
1191 static void copy_seccomp(struct task_struct *p)
1192 {
1193 #ifdef CONFIG_SECCOMP
1194         /*
1195          * Must be called with sighand->lock held, which is common to
1196          * all threads in the group. Holding cred_guard_mutex is not
1197          * needed because this new task is not yet running and cannot
1198          * be racing exec.
1199          */
1200         assert_spin_locked(&current->sighand->siglock);
1201
1202         /* Ref-count the new filter user, and assign it. */
1203         get_seccomp_filter(current);
1204         p->seccomp = current->seccomp;
1205
1206         /*
1207          * Explicitly enable no_new_privs here in case it got set
1208          * between the task_struct being duplicated and holding the
1209          * sighand lock. The seccomp state and nnp must be in sync.
1210          */
1211         if (task_no_new_privs(current))
1212                 task_set_no_new_privs(p);
1213
1214         /*
1215          * If the parent gained a seccomp mode after copying thread
1216          * flags and between before we held the sighand lock, we have
1217          * to manually enable the seccomp thread flag here.
1218          */
1219         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1220                 set_tsk_thread_flag(p, TIF_SECCOMP);
1221 #endif
1222 }
1223
1224 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1225 {
1226         current->clear_child_tid = tidptr;
1227
1228         return task_pid_vnr(current);
1229 }
1230
1231 static void rt_mutex_init_task(struct task_struct *p)
1232 {
1233         raw_spin_lock_init(&p->pi_lock);
1234 #ifdef CONFIG_RT_MUTEXES
1235         p->pi_waiters = RB_ROOT;
1236         p->pi_waiters_leftmost = NULL;
1237         p->pi_blocked_on = NULL;
1238 #endif
1239 }
1240
1241 /*
1242  * Initialize POSIX timer handling for a single task.
1243  */
1244 static void posix_cpu_timers_init(struct task_struct *tsk)
1245 {
1246         tsk->cputime_expires.prof_exp = 0;
1247         tsk->cputime_expires.virt_exp = 0;
1248         tsk->cputime_expires.sched_exp = 0;
1249         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1250         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1251         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1252 }
1253
1254 static inline void
1255 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1256 {
1257          task->pids[type].pid = pid;
1258 }
1259
1260 /*
1261  * This creates a new process as a copy of the old one,
1262  * but does not actually start it yet.
1263  *
1264  * It copies the registers, and all the appropriate
1265  * parts of the process environment (as per the clone
1266  * flags). The actual kick-off is left to the caller.
1267  */
1268 static struct task_struct *copy_process(unsigned long clone_flags,
1269                                         unsigned long stack_start,
1270                                         unsigned long stack_size,
1271                                         int __user *child_tidptr,
1272                                         struct pid *pid,
1273                                         int trace,
1274                                         unsigned long tls)
1275 {
1276         int retval;
1277         struct task_struct *p;
1278         void *cgrp_ss_priv[CGROUP_CANFORK_COUNT] = {};
1279
1280         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1281                 return ERR_PTR(-EINVAL);
1282
1283         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1284                 return ERR_PTR(-EINVAL);
1285
1286         /*
1287          * Thread groups must share signals as well, and detached threads
1288          * can only be started up within the thread group.
1289          */
1290         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1291                 return ERR_PTR(-EINVAL);
1292
1293         /*
1294          * Shared signal handlers imply shared VM. By way of the above,
1295          * thread groups also imply shared VM. Blocking this case allows
1296          * for various simplifications in other code.
1297          */
1298         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1299                 return ERR_PTR(-EINVAL);
1300
1301         /*
1302          * Siblings of global init remain as zombies on exit since they are
1303          * not reaped by their parent (swapper). To solve this and to avoid
1304          * multi-rooted process trees, prevent global and container-inits
1305          * from creating siblings.
1306          */
1307         if ((clone_flags & CLONE_PARENT) &&
1308                                 current->signal->flags & SIGNAL_UNKILLABLE)
1309                 return ERR_PTR(-EINVAL);
1310
1311         /*
1312          * If the new process will be in a different pid or user namespace
1313          * do not allow it to share a thread group with the forking task.
1314          */
1315         if (clone_flags & CLONE_THREAD) {
1316                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1317                     (task_active_pid_ns(current) !=
1318                                 current->nsproxy->pid_ns_for_children))
1319                         return ERR_PTR(-EINVAL);
1320         }
1321
1322         retval = security_task_create(clone_flags);
1323         if (retval)
1324                 goto fork_out;
1325
1326         retval = -ENOMEM;
1327         p = dup_task_struct(current);
1328         if (!p)
1329                 goto fork_out;
1330
1331         ftrace_graph_init_task(p);
1332
1333         rt_mutex_init_task(p);
1334
1335 #ifdef CONFIG_PROVE_LOCKING
1336         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1337         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1338 #endif
1339         retval = -EAGAIN;
1340         if (atomic_read(&p->real_cred->user->processes) >=
1341                         task_rlimit(p, RLIMIT_NPROC)) {
1342                 if (p->real_cred->user != INIT_USER &&
1343                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1344                         goto bad_fork_free;
1345         }
1346         current->flags &= ~PF_NPROC_EXCEEDED;
1347
1348         retval = copy_creds(p, clone_flags);
1349         if (retval < 0)
1350                 goto bad_fork_free;
1351
1352         /*
1353          * If multiple threads are within copy_process(), then this check
1354          * triggers too late. This doesn't hurt, the check is only there
1355          * to stop root fork bombs.
1356          */
1357         retval = -EAGAIN;
1358         if (nr_threads >= max_threads)
1359                 goto bad_fork_cleanup_count;
1360
1361         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1362         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1363         p->flags |= PF_FORKNOEXEC;
1364         INIT_LIST_HEAD(&p->children);
1365         INIT_LIST_HEAD(&p->sibling);
1366         rcu_copy_process(p);
1367         p->vfork_done = NULL;
1368         spin_lock_init(&p->alloc_lock);
1369
1370         init_sigpending(&p->pending);
1371
1372         p->utime = p->stime = p->gtime = 0;
1373         p->utimescaled = p->stimescaled = 0;
1374         prev_cputime_init(&p->prev_cputime);
1375
1376 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1377         seqlock_init(&p->vtime_seqlock);
1378         p->vtime_snap = 0;
1379         p->vtime_snap_whence = VTIME_SLEEPING;
1380 #endif
1381
1382 #if defined(SPLIT_RSS_COUNTING)
1383         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1384 #endif
1385
1386         p->default_timer_slack_ns = current->timer_slack_ns;
1387
1388         task_io_accounting_init(&p->ioac);
1389         acct_clear_integrals(p);
1390
1391         posix_cpu_timers_init(p);
1392
1393         p->start_time = ktime_get_ns();
1394         p->real_start_time = ktime_get_boot_ns();
1395         p->io_context = NULL;
1396         p->audit_context = NULL;
1397         cgroup_fork(p);
1398 #ifdef CONFIG_NUMA
1399         p->mempolicy = mpol_dup(p->mempolicy);
1400         if (IS_ERR(p->mempolicy)) {
1401                 retval = PTR_ERR(p->mempolicy);
1402                 p->mempolicy = NULL;
1403                 goto bad_fork_cleanup_threadgroup_lock;
1404         }
1405 #endif
1406 #ifdef CONFIG_CPUSETS
1407         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1408         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1409         seqcount_init(&p->mems_allowed_seq);
1410 #endif
1411 #ifdef CONFIG_TRACE_IRQFLAGS
1412         p->irq_events = 0;
1413         p->hardirqs_enabled = 0;
1414         p->hardirq_enable_ip = 0;
1415         p->hardirq_enable_event = 0;
1416         p->hardirq_disable_ip = _THIS_IP_;
1417         p->hardirq_disable_event = 0;
1418         p->softirqs_enabled = 1;
1419         p->softirq_enable_ip = _THIS_IP_;
1420         p->softirq_enable_event = 0;
1421         p->softirq_disable_ip = 0;
1422         p->softirq_disable_event = 0;
1423         p->hardirq_context = 0;
1424         p->softirq_context = 0;
1425 #endif
1426
1427         p->pagefault_disabled = 0;
1428
1429 #ifdef CONFIG_LOCKDEP
1430         p->lockdep_depth = 0; /* no locks held yet */
1431         p->curr_chain_key = 0;
1432         p->lockdep_recursion = 0;
1433 #endif
1434
1435 #ifdef CONFIG_DEBUG_MUTEXES
1436         p->blocked_on = NULL; /* not blocked yet */
1437 #endif
1438 #ifdef CONFIG_BCACHE
1439         p->sequential_io        = 0;
1440         p->sequential_io_avg    = 0;
1441 #endif
1442
1443         /* Perform scheduler related setup. Assign this task to a CPU. */
1444         retval = sched_fork(clone_flags, p);
1445         if (retval)
1446                 goto bad_fork_cleanup_policy;
1447
1448         retval = perf_event_init_task(p);
1449         if (retval)
1450                 goto bad_fork_cleanup_policy;
1451         retval = audit_alloc(p);
1452         if (retval)
1453                 goto bad_fork_cleanup_perf;
1454         /* copy all the process information */
1455         shm_init_task(p);
1456         retval = copy_semundo(clone_flags, p);
1457         if (retval)
1458                 goto bad_fork_cleanup_audit;
1459         retval = copy_files(clone_flags, p);
1460         if (retval)
1461                 goto bad_fork_cleanup_semundo;
1462         retval = copy_fs(clone_flags, p);
1463         if (retval)
1464                 goto bad_fork_cleanup_files;
1465         retval = copy_sighand(clone_flags, p);
1466         if (retval)
1467                 goto bad_fork_cleanup_fs;
1468         retval = copy_signal(clone_flags, p);
1469         if (retval)
1470                 goto bad_fork_cleanup_sighand;
1471         retval = copy_mm(clone_flags, p);
1472         if (retval)
1473                 goto bad_fork_cleanup_signal;
1474         retval = copy_namespaces(clone_flags, p);
1475         if (retval)
1476                 goto bad_fork_cleanup_mm;
1477         retval = copy_io(clone_flags, p);
1478         if (retval)
1479                 goto bad_fork_cleanup_namespaces;
1480         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1481         if (retval)
1482                 goto bad_fork_cleanup_io;
1483
1484         if (pid != &init_struct_pid) {
1485                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1486                 if (IS_ERR(pid)) {
1487                         retval = PTR_ERR(pid);
1488                         goto bad_fork_cleanup_io;
1489                 }
1490         }
1491
1492         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1493         /*
1494          * Clear TID on mm_release()?
1495          */
1496         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1497 #ifdef CONFIG_BLOCK
1498         p->plug = NULL;
1499 #endif
1500 #ifdef CONFIG_FUTEX
1501         p->robust_list = NULL;
1502 #ifdef CONFIG_COMPAT
1503         p->compat_robust_list = NULL;
1504 #endif
1505         INIT_LIST_HEAD(&p->pi_state_list);
1506         p->pi_state_cache = NULL;
1507 #endif
1508         /*
1509          * sigaltstack should be cleared when sharing the same VM
1510          */
1511         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1512                 p->sas_ss_sp = p->sas_ss_size = 0;
1513
1514         /*
1515          * Syscall tracing and stepping should be turned off in the
1516          * child regardless of CLONE_PTRACE.
1517          */
1518         user_disable_single_step(p);
1519         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1520 #ifdef TIF_SYSCALL_EMU
1521         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1522 #endif
1523         clear_all_latency_tracing(p);
1524
1525         /* ok, now we should be set up.. */
1526         p->pid = pid_nr(pid);
1527         if (clone_flags & CLONE_THREAD) {
1528                 p->exit_signal = -1;
1529                 p->group_leader = current->group_leader;
1530                 p->tgid = current->tgid;
1531         } else {
1532                 if (clone_flags & CLONE_PARENT)
1533                         p->exit_signal = current->group_leader->exit_signal;
1534                 else
1535                         p->exit_signal = (clone_flags & CSIGNAL);
1536                 p->group_leader = p;
1537                 p->tgid = p->pid;
1538         }
1539
1540         p->nr_dirtied = 0;
1541         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1542         p->dirty_paused_when = 0;
1543
1544         p->pdeath_signal = 0;
1545         INIT_LIST_HEAD(&p->thread_group);
1546         p->task_works = NULL;
1547
1548         threadgroup_change_begin(current);
1549         /*
1550          * Ensure that the cgroup subsystem policies allow the new process to be
1551          * forked. It should be noted the the new process's css_set can be changed
1552          * between here and cgroup_post_fork() if an organisation operation is in
1553          * progress.
1554          */
1555         retval = cgroup_can_fork(p, cgrp_ss_priv);
1556         if (retval)
1557                 goto bad_fork_free_pid;
1558
1559         /*
1560          * Make it visible to the rest of the system, but dont wake it up yet.
1561          * Need tasklist lock for parent etc handling!
1562          */
1563         write_lock_irq(&tasklist_lock);
1564
1565         /* CLONE_PARENT re-uses the old parent */
1566         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1567                 p->real_parent = current->real_parent;
1568                 p->parent_exec_id = current->parent_exec_id;
1569         } else {
1570                 p->real_parent = current;
1571                 p->parent_exec_id = current->self_exec_id;
1572         }
1573
1574         spin_lock(&current->sighand->siglock);
1575
1576         /*
1577          * Copy seccomp details explicitly here, in case they were changed
1578          * before holding sighand lock.
1579          */
1580         copy_seccomp(p);
1581
1582         /*
1583          * Process group and session signals need to be delivered to just the
1584          * parent before the fork or both the parent and the child after the
1585          * fork. Restart if a signal comes in before we add the new process to
1586          * it's process group.
1587          * A fatal signal pending means that current will exit, so the new
1588          * thread can't slip out of an OOM kill (or normal SIGKILL).
1589         */
1590         recalc_sigpending();
1591         if (signal_pending(current)) {
1592                 spin_unlock(&current->sighand->siglock);
1593                 write_unlock_irq(&tasklist_lock);
1594                 retval = -ERESTARTNOINTR;
1595                 goto bad_fork_cancel_cgroup;
1596         }
1597
1598         if (likely(p->pid)) {
1599                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1600
1601                 init_task_pid(p, PIDTYPE_PID, pid);
1602                 if (thread_group_leader(p)) {
1603                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1604                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1605
1606                         if (is_child_reaper(pid)) {
1607                                 ns_of_pid(pid)->child_reaper = p;
1608                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1609                         }
1610
1611                         p->signal->leader_pid = pid;
1612                         p->signal->tty = tty_kref_get(current->signal->tty);
1613                         list_add_tail(&p->sibling, &p->real_parent->children);
1614                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1615                         attach_pid(p, PIDTYPE_PGID);
1616                         attach_pid(p, PIDTYPE_SID);
1617                         __this_cpu_inc(process_counts);
1618                 } else {
1619                         current->signal->nr_threads++;
1620                         atomic_inc(&current->signal->live);
1621                         atomic_inc(&current->signal->sigcnt);
1622                         list_add_tail_rcu(&p->thread_group,
1623                                           &p->group_leader->thread_group);
1624                         list_add_tail_rcu(&p->thread_node,
1625                                           &p->signal->thread_head);
1626                 }
1627                 attach_pid(p, PIDTYPE_PID);
1628                 nr_threads++;
1629         }
1630
1631         total_forks++;
1632         spin_unlock(&current->sighand->siglock);
1633         syscall_tracepoint_update(p);
1634         write_unlock_irq(&tasklist_lock);
1635
1636         proc_fork_connector(p);
1637         cgroup_post_fork(p, cgrp_ss_priv);
1638         threadgroup_change_end(current);
1639         perf_event_fork(p);
1640
1641         trace_task_newtask(p, clone_flags);
1642         uprobe_copy_process(p, clone_flags);
1643
1644         return p;
1645
1646 bad_fork_cancel_cgroup:
1647         cgroup_cancel_fork(p, cgrp_ss_priv);
1648 bad_fork_free_pid:
1649         threadgroup_change_end(current);
1650         if (pid != &init_struct_pid)
1651                 free_pid(pid);
1652 bad_fork_cleanup_io:
1653         if (p->io_context)
1654                 exit_io_context(p);
1655 bad_fork_cleanup_namespaces:
1656         exit_task_namespaces(p);
1657 bad_fork_cleanup_mm:
1658         if (p->mm)
1659                 mmput(p->mm);
1660 bad_fork_cleanup_signal:
1661         if (!(clone_flags & CLONE_THREAD))
1662                 free_signal_struct(p->signal);
1663 bad_fork_cleanup_sighand:
1664         __cleanup_sighand(p->sighand);
1665 bad_fork_cleanup_fs:
1666         exit_fs(p); /* blocking */
1667 bad_fork_cleanup_files:
1668         exit_files(p); /* blocking */
1669 bad_fork_cleanup_semundo:
1670         exit_sem(p);
1671 bad_fork_cleanup_audit:
1672         audit_free(p);
1673 bad_fork_cleanup_perf:
1674         perf_event_free_task(p);
1675 bad_fork_cleanup_policy:
1676 #ifdef CONFIG_NUMA
1677         mpol_put(p->mempolicy);
1678 bad_fork_cleanup_threadgroup_lock:
1679 #endif
1680         delayacct_tsk_free(p);
1681 bad_fork_cleanup_count:
1682         atomic_dec(&p->cred->user->processes);
1683         exit_creds(p);
1684 bad_fork_free:
1685         free_task(p);
1686 fork_out:
1687         return ERR_PTR(retval);
1688 }
1689
1690 static inline void init_idle_pids(struct pid_link *links)
1691 {
1692         enum pid_type type;
1693
1694         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1695                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1696                 links[type].pid = &init_struct_pid;
1697         }
1698 }
1699
1700 struct task_struct *fork_idle(int cpu)
1701 {
1702         struct task_struct *task;
1703         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0);
1704         if (!IS_ERR(task)) {
1705                 init_idle_pids(task->pids);
1706                 init_idle(task, cpu);
1707         }
1708
1709         return task;
1710 }
1711
1712 /*
1713  *  Ok, this is the main fork-routine.
1714  *
1715  * It copies the process, and if successful kick-starts
1716  * it and waits for it to finish using the VM if required.
1717  */
1718 long _do_fork(unsigned long clone_flags,
1719               unsigned long stack_start,
1720               unsigned long stack_size,
1721               int __user *parent_tidptr,
1722               int __user *child_tidptr,
1723               unsigned long tls)
1724 {
1725         struct task_struct *p;
1726         int trace = 0;
1727         long nr;
1728
1729         /*
1730          * Determine whether and which event to report to ptracer.  When
1731          * called from kernel_thread or CLONE_UNTRACED is explicitly
1732          * requested, no event is reported; otherwise, report if the event
1733          * for the type of forking is enabled.
1734          */
1735         if (!(clone_flags & CLONE_UNTRACED)) {
1736                 if (clone_flags & CLONE_VFORK)
1737                         trace = PTRACE_EVENT_VFORK;
1738                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1739                         trace = PTRACE_EVENT_CLONE;
1740                 else
1741                         trace = PTRACE_EVENT_FORK;
1742
1743                 if (likely(!ptrace_event_enabled(current, trace)))
1744                         trace = 0;
1745         }
1746
1747         p = copy_process(clone_flags, stack_start, stack_size,
1748                          child_tidptr, NULL, trace, tls);
1749         /*
1750          * Do this prior waking up the new thread - the thread pointer
1751          * might get invalid after that point, if the thread exits quickly.
1752          */
1753         if (!IS_ERR(p)) {
1754                 struct completion vfork;
1755                 struct pid *pid;
1756
1757                 trace_sched_process_fork(current, p);
1758
1759                 pid = get_task_pid(p, PIDTYPE_PID);
1760                 nr = pid_vnr(pid);
1761
1762                 if (clone_flags & CLONE_PARENT_SETTID)
1763                         put_user(nr, parent_tidptr);
1764
1765                 if (clone_flags & CLONE_VFORK) {
1766                         p->vfork_done = &vfork;
1767                         init_completion(&vfork);
1768                         get_task_struct(p);
1769                 }
1770
1771                 wake_up_new_task(p);
1772
1773                 /* forking complete and child started to run, tell ptracer */
1774                 if (unlikely(trace))
1775                         ptrace_event_pid(trace, pid);
1776
1777                 if (clone_flags & CLONE_VFORK) {
1778                         if (!wait_for_vfork_done(p, &vfork))
1779                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1780                 }
1781
1782                 put_pid(pid);
1783         } else {
1784                 nr = PTR_ERR(p);
1785         }
1786         return nr;
1787 }
1788
1789 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1790 /* For compatibility with architectures that call do_fork directly rather than
1791  * using the syscall entry points below. */
1792 long do_fork(unsigned long clone_flags,
1793               unsigned long stack_start,
1794               unsigned long stack_size,
1795               int __user *parent_tidptr,
1796               int __user *child_tidptr)
1797 {
1798         return _do_fork(clone_flags, stack_start, stack_size,
1799                         parent_tidptr, child_tidptr, 0);
1800 }
1801 #endif
1802
1803 /*
1804  * Create a kernel thread.
1805  */
1806 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1807 {
1808         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1809                 (unsigned long)arg, NULL, NULL, 0);
1810 }
1811
1812 #ifdef __ARCH_WANT_SYS_FORK
1813 SYSCALL_DEFINE0(fork)
1814 {
1815 #ifdef CONFIG_MMU
1816         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1817 #else
1818         /* can not support in nommu mode */
1819         return -EINVAL;
1820 #endif
1821 }
1822 #endif
1823
1824 #ifdef __ARCH_WANT_SYS_VFORK
1825 SYSCALL_DEFINE0(vfork)
1826 {
1827         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1828                         0, NULL, NULL, 0);
1829 }
1830 #endif
1831
1832 #ifdef __ARCH_WANT_SYS_CLONE
1833 #ifdef CONFIG_CLONE_BACKWARDS
1834 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1835                  int __user *, parent_tidptr,
1836                  unsigned long, tls,
1837                  int __user *, child_tidptr)
1838 #elif defined(CONFIG_CLONE_BACKWARDS2)
1839 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1840                  int __user *, parent_tidptr,
1841                  int __user *, child_tidptr,
1842                  unsigned long, tls)
1843 #elif defined(CONFIG_CLONE_BACKWARDS3)
1844 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1845                 int, stack_size,
1846                 int __user *, parent_tidptr,
1847                 int __user *, child_tidptr,
1848                 unsigned long, tls)
1849 #else
1850 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1851                  int __user *, parent_tidptr,
1852                  int __user *, child_tidptr,
1853                  unsigned long, tls)
1854 #endif
1855 {
1856         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
1857 }
1858 #endif
1859
1860 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1861 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1862 #endif
1863
1864 static void sighand_ctor(void *data)
1865 {
1866         struct sighand_struct *sighand = data;
1867
1868         spin_lock_init(&sighand->siglock);
1869         init_waitqueue_head(&sighand->signalfd_wqh);
1870 }
1871
1872 void __init proc_caches_init(void)
1873 {
1874         sighand_cachep = kmem_cache_create("sighand_cache",
1875                         sizeof(struct sighand_struct), 0,
1876                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1877                         SLAB_NOTRACK, sighand_ctor);
1878         signal_cachep = kmem_cache_create("signal_cache",
1879                         sizeof(struct signal_struct), 0,
1880                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1881         files_cachep = kmem_cache_create("files_cache",
1882                         sizeof(struct files_struct), 0,
1883                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1884         fs_cachep = kmem_cache_create("fs_cache",
1885                         sizeof(struct fs_struct), 0,
1886                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1887         /*
1888          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1889          * whole struct cpumask for the OFFSTACK case. We could change
1890          * this to *only* allocate as much of it as required by the
1891          * maximum number of CPU's we can ever have.  The cpumask_allocation
1892          * is at the end of the structure, exactly for that reason.
1893          */
1894         mm_cachep = kmem_cache_create("mm_struct",
1895                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1896                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1897         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1898         mmap_init();
1899         nsproxy_cache_init();
1900 }
1901
1902 /*
1903  * Check constraints on flags passed to the unshare system call.
1904  */
1905 static int check_unshare_flags(unsigned long unshare_flags)
1906 {
1907         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1908                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1909                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1910                                 CLONE_NEWUSER|CLONE_NEWPID))
1911                 return -EINVAL;
1912         /*
1913          * Not implemented, but pretend it works if there is nothing
1914          * to unshare.  Note that unsharing the address space or the
1915          * signal handlers also need to unshare the signal queues (aka
1916          * CLONE_THREAD).
1917          */
1918         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1919                 if (!thread_group_empty(current))
1920                         return -EINVAL;
1921         }
1922         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
1923                 if (atomic_read(&current->sighand->count) > 1)
1924                         return -EINVAL;
1925         }
1926         if (unshare_flags & CLONE_VM) {
1927                 if (!current_is_single_threaded())
1928                         return -EINVAL;
1929         }
1930
1931         return 0;
1932 }
1933
1934 /*
1935  * Unshare the filesystem structure if it is being shared
1936  */
1937 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1938 {
1939         struct fs_struct *fs = current->fs;
1940
1941         if (!(unshare_flags & CLONE_FS) || !fs)
1942                 return 0;
1943
1944         /* don't need lock here; in the worst case we'll do useless copy */
1945         if (fs->users == 1)
1946                 return 0;
1947
1948         *new_fsp = copy_fs_struct(fs);
1949         if (!*new_fsp)
1950                 return -ENOMEM;
1951
1952         return 0;
1953 }
1954
1955 /*
1956  * Unshare file descriptor table if it is being shared
1957  */
1958 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1959 {
1960         struct files_struct *fd = current->files;
1961         int error = 0;
1962
1963         if ((unshare_flags & CLONE_FILES) &&
1964             (fd && atomic_read(&fd->count) > 1)) {
1965                 *new_fdp = dup_fd(fd, &error);
1966                 if (!*new_fdp)
1967                         return error;
1968         }
1969
1970         return 0;
1971 }
1972
1973 /*
1974  * unshare allows a process to 'unshare' part of the process
1975  * context which was originally shared using clone.  copy_*
1976  * functions used by do_fork() cannot be used here directly
1977  * because they modify an inactive task_struct that is being
1978  * constructed. Here we are modifying the current, active,
1979  * task_struct.
1980  */
1981 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1982 {
1983         struct fs_struct *fs, *new_fs = NULL;
1984         struct files_struct *fd, *new_fd = NULL;
1985         struct cred *new_cred = NULL;
1986         struct nsproxy *new_nsproxy = NULL;
1987         int do_sysvsem = 0;
1988         int err;
1989
1990         /*
1991          * If unsharing a user namespace must also unshare the thread group
1992          * and unshare the filesystem root and working directories.
1993          */
1994         if (unshare_flags & CLONE_NEWUSER)
1995                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1996         /*
1997          * If unsharing vm, must also unshare signal handlers.
1998          */
1999         if (unshare_flags & CLONE_VM)
2000                 unshare_flags |= CLONE_SIGHAND;
2001         /*
2002          * If unsharing a signal handlers, must also unshare the signal queues.
2003          */
2004         if (unshare_flags & CLONE_SIGHAND)
2005                 unshare_flags |= CLONE_THREAD;
2006         /*
2007          * If unsharing namespace, must also unshare filesystem information.
2008          */
2009         if (unshare_flags & CLONE_NEWNS)
2010                 unshare_flags |= CLONE_FS;
2011
2012         err = check_unshare_flags(unshare_flags);
2013         if (err)
2014                 goto bad_unshare_out;
2015         /*
2016          * CLONE_NEWIPC must also detach from the undolist: after switching
2017          * to a new ipc namespace, the semaphore arrays from the old
2018          * namespace are unreachable.
2019          */
2020         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2021                 do_sysvsem = 1;
2022         err = unshare_fs(unshare_flags, &new_fs);
2023         if (err)
2024                 goto bad_unshare_out;
2025         err = unshare_fd(unshare_flags, &new_fd);
2026         if (err)
2027                 goto bad_unshare_cleanup_fs;
2028         err = unshare_userns(unshare_flags, &new_cred);
2029         if (err)
2030                 goto bad_unshare_cleanup_fd;
2031         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2032                                          new_cred, new_fs);
2033         if (err)
2034                 goto bad_unshare_cleanup_cred;
2035
2036         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2037                 if (do_sysvsem) {
2038                         /*
2039                          * CLONE_SYSVSEM is equivalent to sys_exit().
2040                          */
2041                         exit_sem(current);
2042                 }
2043                 if (unshare_flags & CLONE_NEWIPC) {
2044                         /* Orphan segments in old ns (see sem above). */
2045                         exit_shm(current);
2046                         shm_init_task(current);
2047                 }
2048
2049                 if (new_nsproxy)
2050                         switch_task_namespaces(current, new_nsproxy);
2051
2052                 task_lock(current);
2053
2054                 if (new_fs) {
2055                         fs = current->fs;
2056                         spin_lock(&fs->lock);
2057                         current->fs = new_fs;
2058                         if (--fs->users)
2059                                 new_fs = NULL;
2060                         else
2061                                 new_fs = fs;
2062                         spin_unlock(&fs->lock);
2063                 }
2064
2065                 if (new_fd) {
2066                         fd = current->files;
2067                         current->files = new_fd;
2068                         new_fd = fd;
2069                 }
2070
2071                 task_unlock(current);
2072
2073                 if (new_cred) {
2074                         /* Install the new user namespace */
2075                         commit_creds(new_cred);
2076                         new_cred = NULL;
2077                 }
2078         }
2079
2080 bad_unshare_cleanup_cred:
2081         if (new_cred)
2082                 put_cred(new_cred);
2083 bad_unshare_cleanup_fd:
2084         if (new_fd)
2085                 put_files_struct(new_fd);
2086
2087 bad_unshare_cleanup_fs:
2088         if (new_fs)
2089                 free_fs_struct(new_fs);
2090
2091 bad_unshare_out:
2092         return err;
2093 }
2094
2095 /*
2096  *      Helper to unshare the files of the current task.
2097  *      We don't want to expose copy_files internals to
2098  *      the exec layer of the kernel.
2099  */
2100
2101 int unshare_files(struct files_struct **displaced)
2102 {
2103         struct task_struct *task = current;
2104         struct files_struct *copy = NULL;
2105         int error;
2106
2107         error = unshare_fd(CLONE_FILES, &copy);
2108         if (error || !copy) {
2109                 *displaced = NULL;
2110                 return error;
2111         }
2112         *displaced = task->files;
2113         task_lock(task);
2114         task->files = copy;
2115         task_unlock(task);
2116         return 0;
2117 }
2118
2119 int sysctl_max_threads(struct ctl_table *table, int write,
2120                        void __user *buffer, size_t *lenp, loff_t *ppos)
2121 {
2122         struct ctl_table t;
2123         int ret;
2124         int threads = max_threads;
2125         int min = MIN_THREADS;
2126         int max = MAX_THREADS;
2127
2128         t = *table;
2129         t.data = &threads;
2130         t.extra1 = &min;
2131         t.extra2 = &max;
2132
2133         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2134         if (ret || !write)
2135                 return ret;
2136
2137         set_max_threads(threads);
2138
2139         return 0;
2140 }