kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/mm.h>
  32 #include <linux/vmacache.h>
  33 #include <linux/nsproxy.h>
  34 #include <linux/capability.h>
  35 #include <linux/cpu.h>
  36 #include <linux/cgroup.h>
  37 #include <linux/security.h>
  38 #include <linux/hugetlb.h>
  39 #include <linux/seccomp.h>
  40 #include <linux/swap.h>
  41 #include <linux/syscalls.h>
  42 #include <linux/jiffies.h>
  43 #include <linux/futex.h>
  44 #include <linux/compat.h>
  45 #include <linux/kthread.h>
  46 #include <linux/task_io_accounting_ops.h>
  47 #include <linux/rcupdate.h>
  48 #include <linux/ptrace.h>
  49 #include <linux/mount.h>
  50 #include <linux/audit.h>
  51 #include <linux/memcontrol.h>
  52 #include <linux/ftrace.h>
  53 #include <linux/proc_fs.h>
  54 #include <linux/profile.h>
  55 #include <linux/rmap.h>
  56 #include <linux/ksm.h>
  57 #include <linux/acct.h>
  58 #include <linux/tsacct_kern.h>
  59 #include <linux/cn_proc.h>
  60 #include <linux/freezer.h>
  61 #include <linux/delayacct.h>
  62 #include <linux/taskstats_kern.h>
  63 #include <linux/random.h>
  64 #include <linux/tty.h>
  65 #include <linux/blkdev.h>
  66 #include <linux/fs_struct.h>
  67 #include <linux/magic.h>
  68 #include <linux/perf_event.h>
  69 #include <linux/posix-timers.h>
  70 #include <linux/user-return-notifier.h>
  71 #include <linux/oom.h>
  72 #include <linux/khugepaged.h>
  73 #include <linux/signalfd.h>
  74 #include <linux/uprobes.h>
  75 #include <linux/aio.h>
  76 #include <linux/compiler.h>
  77 #include <linux/sysctl.h>
  78
  79 #include <asm/pgtable.h>
  80 #include <asm/pgalloc.h>
  81 #include <asm/uaccess.h>
  82 #include <asm/mmu_context.h>
  83 #include <asm/cacheflush.h>
  84 #include <asm/tlbflush.h>
  85
  86 #include <trace/events/sched.h>
  87
  88 #define CREATE_TRACE_POINTS
  89 #include <trace/events/task.h>
  90
  91 /*
  92  * Minimum number of threads to boot the kernel
  93  */
  94 #define MIN_THREADS 20
  95
  96 /*
  97  * Maximum number of threads
  98  */
  99 #define MAX_THREADS FUTEX_TID_MASK
 100
 101 /*
 102  * Protected counters by write_lock_irq(&tasklist_lock)
 103  */
 104 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 105 int nr_threads;                 /* The idle threads do not count.. */
 106
 107 int max_threads;                /* tunable limit on nr_threads */
 108
 109 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 110
 111 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 112
 113 #ifdef CONFIG_PROVE_RCU
 114 int lockdep_tasklist_lock_is_held(void)
 115 {
 116         return lockdep_is_held(&tasklist_lock);
 117 }
 118 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 119 #endif /* #ifdef CONFIG_PROVE_RCU */
 120
 121 int nr_processes(void)
 122 {
 123         int cpu;
 124         int total = 0;
 125
 126         for_each_possible_cpu(cpu)
 127                 total += per_cpu(process_counts, cpu);
 128
 129         return total;
 130 }
 131
 132 void __weak arch_release_task_struct(struct task_struct *tsk)
 133 {
 134 }
 135
 136 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 137 static struct kmem_cache *task_struct_cachep;
 138
 139 static inline struct task_struct *alloc_task_struct_node(int node)
 140 {
 141         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 142 }
 143
 144 static inline void free_task_struct(struct task_struct *tsk)
 145 {
 146         kmem_cache_free(task_struct_cachep, tsk);
 147 }
 148 #endif
 149
 150 void __weak arch_release_thread_info(struct thread_info *ti)
 151 {
 152 }
 153
 154 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
 155
 156 /*
 157  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 158  * kmemcache based allocator.
 159  */
 160 # if THREAD_SIZE >= PAGE_SIZE
 161 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 162                                                   int node)
 163 {
 164         struct page *page = alloc_kmem_pages_node(node, THREADINFO_GFP,
 165                                                   THREAD_SIZE_ORDER);
 166
 167         return page ? page_address(page) : NULL;
 168 }
 169
 170 static inline void free_thread_info(struct thread_info *ti)
 171 {
 172         free_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
 173 }
 174 # else
 175 static struct kmem_cache *thread_info_cache;
 176
 177 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 178                                                   int node)
 179 {
 180         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
 181 }
 182
 183 static void free_thread_info(struct thread_info *ti)
 184 {
 185         kmem_cache_free(thread_info_cache, ti);
 186 }
 187
 188 void thread_info_cache_init(void)
 189 {
 190         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
 191                                               THREAD_SIZE, 0, NULL);
 192         BUG_ON(thread_info_cache == NULL);
 193 }
 194 # endif
 195 #endif
 196
 197 /* SLAB cache for signal_struct structures (tsk->signal) */
 198 static struct kmem_cache *signal_cachep;
 199
 200 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 201 struct kmem_cache *sighand_cachep;
 202
 203 /* SLAB cache for files_struct structures (tsk->files) */
 204 struct kmem_cache *files_cachep;
 205
 206 /* SLAB cache for fs_struct structures (tsk->fs) */
 207 struct kmem_cache *fs_cachep;
 208
 209 /* SLAB cache for vm_area_struct structures */
 210 struct kmem_cache *vm_area_cachep;
 211
 212 /* SLAB cache for mm_struct structures (tsk->mm) */
 213 static struct kmem_cache *mm_cachep;
 214
 215 static void account_kernel_stack(struct thread_info *ti, int account)
 216 {
 217         struct zone *zone = page_zone(virt_to_page(ti));
 218
 219         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 220 }
 221
 222 void free_task(struct task_struct *tsk)
 223 {
 224         account_kernel_stack(tsk->stack, -1);
 225         arch_release_thread_info(tsk->stack);
 226         free_thread_info(tsk->stack);
 227         rt_mutex_debug_task_free(tsk);
 228         ftrace_graph_exit_task(tsk);
 229         put_seccomp_filter(tsk);
 230         arch_release_task_struct(tsk);
 231         free_task_struct(tsk);
 232 }
 233 EXPORT_SYMBOL(free_task);
 234
 235 static inline void free_signal_struct(struct signal_struct *sig)
 236 {
 237         taskstats_tgid_free(sig);
 238         sched_autogroup_exit(sig);
 239         kmem_cache_free(signal_cachep, sig);
 240 }
 241
 242 static inline void put_signal_struct(struct signal_struct *sig)
 243 {
 244         if (atomic_dec_and_test(&sig->sigcnt))
 245                 free_signal_struct(sig);
 246 }
 247
 248 void __put_task_struct(struct task_struct *tsk)
 249 {
 250         WARN_ON(!tsk->exit_state);
 251         WARN_ON(atomic_read(&tsk->usage));
 252         WARN_ON(tsk == current);
 253
 254         task_numa_free(tsk);
 255         security_task_free(tsk);
 256         exit_creds(tsk);
 257         delayacct_tsk_free(tsk);
 258         put_signal_struct(tsk->signal);
 259
 260         if (!profile_handoff_task(tsk))
 261                 free_task(tsk);
 262 }
 263 EXPORT_SYMBOL_GPL(__put_task_struct);
 264
 265 void __init __weak arch_task_cache_init(void) { }
 266
 267 /*
 268  * set_max_threads
 269  */
 270 static void set_max_threads(unsigned int max_threads_suggested)
 271 {
 272         u64 threads;
 273
 274         /*
 275          * The number of threads shall be limited such that the thread
 276          * structures may only consume a small part of the available memory.
 277          */
 278         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 279                 threads = MAX_THREADS;
 280         else
 281                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 282                                     (u64) THREAD_SIZE * 8UL);
 283
 284         if (threads > max_threads_suggested)
 285                 threads = max_threads_suggested;
 286
 287         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 288 }
 289
 290 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 291 /* Initialized by the architecture: */
 292 int arch_task_struct_size __read_mostly;
 293 #endif
 294
 295 void __init fork_init(void)
 296 {
 297 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 298 #ifndef ARCH_MIN_TASKALIGN
 299 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 300 #endif
 301         /* create a slab on which task_structs can be allocated */
 302         task_struct_cachep =
 303                 kmem_cache_create("task_struct", arch_task_struct_size,
 304                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 305 #endif
 306
 307         /* do the arch specific task caches init */
 308         arch_task_cache_init();
 309
 310         set_max_threads(MAX_THREADS);
 311
 312         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 313         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 314         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 315                 init_task.signal->rlim[RLIMIT_NPROC];
 316 }
 317
 318 int __weak arch_dup_task_struct(struct task_struct *dst,
 319                                                struct task_struct *src)
 320 {
 321         *dst = *src;
 322         return 0;
 323 }
 324
 325 void set_task_stack_end_magic(struct task_struct *tsk)
 326 {
 327         unsigned long *stackend;
 328
 329         stackend = end_of_stack(tsk);
 330         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 331 }
 332
 333 static struct task_struct *dup_task_struct(struct task_struct *orig)
 334 {
 335         struct task_struct *tsk;
 336         struct thread_info *ti;
 337         int node = tsk_fork_get_node(orig);
 338         int err;
 339
 340         tsk = alloc_task_struct_node(node);
 341         if (!tsk)
 342                 return NULL;
 343
 344         ti = alloc_thread_info_node(tsk, node);
 345         if (!ti)
 346                 goto free_tsk;
 347
 348         err = arch_dup_task_struct(tsk, orig);
 349         if (err)
 350                 goto free_ti;
 351
 352         tsk->stack = ti;
 353 #ifdef CONFIG_SECCOMP
 354         /*
 355          * We must handle setting up seccomp filters once we're under
 356          * the sighand lock in case orig has changed between now and
 357          * then. Until then, filter must be NULL to avoid messing up
 358          * the usage counts on the error path calling free_task.
 359          */
 360         tsk->seccomp.filter = NULL;
 361 #endif
 362
 363         setup_thread_stack(tsk, orig);
 364         clear_user_return_notifier(tsk);
 365         clear_tsk_need_resched(tsk);
 366         set_task_stack_end_magic(tsk);
 367
 368 #ifdef CONFIG_CC_STACKPROTECTOR
 369         tsk->stack_canary = get_random_int();
 370 #endif
 371
 372         /*
 373          * One for us, one for whoever does the "release_task()" (usually
 374          * parent)
 375          */
 376         atomic_set(&tsk->usage, 2);
 377 #ifdef CONFIG_BLK_DEV_IO_TRACE
 378         tsk->btrace_seq = 0;
 379 #endif
 380         tsk->splice_pipe = NULL;
 381         tsk->task_frag.page = NULL;
 382
 383         account_kernel_stack(ti, 1);
 384
 385         return tsk;
 386
 387 free_ti:
 388         free_thread_info(ti);
 389 free_tsk:
 390         free_task_struct(tsk);
 391         return NULL;
 392 }
 393
 394 #ifdef CONFIG_MMU
 395 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 396 {
 397         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 398         struct rb_node **rb_link, *rb_parent;
 399         int retval;
 400         unsigned long charge;
 401
 402         uprobe_start_dup_mmap();
 403         down_write(&oldmm->mmap_sem);
 404         flush_cache_dup_mm(oldmm);
 405         uprobe_dup_mmap(oldmm, mm);
 406         /*
 407          * Not linked in yet - no deadlock potential:
 408          */
 409         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 410
 411         /* No ordering required: file already has been exposed. */
 412         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 413
 414         mm->total_vm = oldmm->total_vm;
 415         mm->shared_vm = oldmm->shared_vm;
 416         mm->exec_vm = oldmm->exec_vm;
 417         mm->stack_vm = oldmm->stack_vm;
 418
 419         rb_link = &mm->mm_rb.rb_node;
 420         rb_parent = NULL;
 421         pprev = &mm->mmap;
 422         retval = ksm_fork(mm, oldmm);
 423         if (retval)
 424                 goto out;
 425         retval = khugepaged_fork(mm, oldmm);
 426         if (retval)
 427                 goto out;
 428
 429         prev = NULL;
 430         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 431                 struct file *file;
 432
 433                 if (mpnt->vm_flags & VM_DONTCOPY) {
 434                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
 435                                                         -vma_pages(mpnt));
 436                         continue;
 437                 }
 438                 charge = 0;
 439                 if (mpnt->vm_flags & VM_ACCOUNT) {
 440                         unsigned long len = vma_pages(mpnt);
 441
 442                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 443                                 goto fail_nomem;
 444                         charge = len;
 445                 }
 446                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 447                 if (!tmp)
 448                         goto fail_nomem;
 449                 *tmp = *mpnt;
 450                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 451                 retval = vma_dup_policy(mpnt, tmp);
 452                 if (retval)
 453                         goto fail_nomem_policy;
 454                 tmp->vm_mm = mm;
 455                 if (anon_vma_fork(tmp, mpnt))
 456                         goto fail_nomem_anon_vma_fork;
 457                 tmp->vm_flags &= ~VM_LOCKED;
 458                 tmp->vm_next = tmp->vm_prev = NULL;
 459                 file = tmp->vm_file;
 460                 if (file) {
 461                         struct inode *inode = file_inode(file);
 462                         struct address_space *mapping = file->f_mapping;
 463
 464                         get_file(file);
 465                         if (tmp->vm_flags & VM_DENYWRITE)
 466                                 atomic_dec(&inode->i_writecount);
 467                         i_mmap_lock_write(mapping);
 468                         if (tmp->vm_flags & VM_SHARED)
 469                                 atomic_inc(&mapping->i_mmap_writable);
 470                         flush_dcache_mmap_lock(mapping);
 471                         /* insert tmp into the share list, just after mpnt */
 472                         vma_interval_tree_insert_after(tmp, mpnt,
 473                                         &mapping->i_mmap);
 474                         flush_dcache_mmap_unlock(mapping);
 475                         i_mmap_unlock_write(mapping);
 476                 }
 477
 478                 /*
 479                  * Clear hugetlb-related page reserves for children. This only
 480                  * affects MAP_PRIVATE mappings. Faults generated by the child
 481                  * are not guaranteed to succeed, even if read-only
 482                  */
 483                 if (is_vm_hugetlb_page(tmp))
 484                         reset_vma_resv_huge_pages(tmp);
 485
 486                 /*
 487                  * Link in the new vma and copy the page table entries.
 488                  */
 489                 *pprev = tmp;
 490                 pprev = &tmp->vm_next;
 491                 tmp->vm_prev = prev;
 492                 prev = tmp;
 493
 494                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 495                 rb_link = &tmp->vm_rb.rb_right;
 496                 rb_parent = &tmp->vm_rb;
 497
 498                 mm->map_count++;
 499                 retval = copy_page_range(mm, oldmm, mpnt);
 500
 501                 if (tmp->vm_ops && tmp->vm_ops->open)
 502                         tmp->vm_ops->open(tmp);
 503
 504                 if (retval)
 505                         goto out;
 506         }
 507         /* a new mm has just been created */
 508         arch_dup_mmap(oldmm, mm);
 509         retval = 0;
 510 out:
 511         up_write(&mm->mmap_sem);
 512         flush_tlb_mm(oldmm);
 513         up_write(&oldmm->mmap_sem);
 514         uprobe_end_dup_mmap();
 515         return retval;
 516 fail_nomem_anon_vma_fork:
 517         mpol_put(vma_policy(tmp));
 518 fail_nomem_policy:
 519         kmem_cache_free(vm_area_cachep, tmp);
 520 fail_nomem:
 521         retval = -ENOMEM;
 522         vm_unacct_memory(charge);
 523         goto out;
 524 }
 525
 526 static inline int mm_alloc_pgd(struct mm_struct *mm)
 527 {
 528         mm->pgd = pgd_alloc(mm);
 529         if (unlikely(!mm->pgd))
 530                 return -ENOMEM;
 531         return 0;
 532 }
 533
 534 static inline void mm_free_pgd(struct mm_struct *mm)
 535 {
 536         pgd_free(mm, mm->pgd);
 537 }
 538 #else
 539 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 540 {
 541         down_write(&oldmm->mmap_sem);
 542         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 543         up_write(&oldmm->mmap_sem);
 544         return 0;
 545 }
 546 #define mm_alloc_pgd(mm)        (0)
 547 #define mm_free_pgd(mm)
 548 #endif /* CONFIG_MMU */
 549
 550 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 551
 552 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 553 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 554
 555 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 556
 557 static int __init coredump_filter_setup(char *s)
 558 {
 559         default_dump_filter =
 560                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 561                 MMF_DUMP_FILTER_MASK;
 562         return 1;
 563 }
 564
 565 __setup("coredump_filter=", coredump_filter_setup);
 566
 567 #include <linux/init_task.h>
 568
 569 static void mm_init_aio(struct mm_struct *mm)
 570 {
 571 #ifdef CONFIG_AIO
 572         spin_lock_init(&mm->ioctx_lock);
 573         mm->ioctx_table = NULL;
 574 #endif
 575 }
 576
 577 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 578 {
 579 #ifdef CONFIG_MEMCG
 580         mm->owner = p;
 581 #endif
 582 }
 583
 584 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 585 {
 586         mm->mmap = NULL;
 587         mm->mm_rb = RB_ROOT;
 588         mm->vmacache_seqnum = 0;
 589         atomic_set(&mm->mm_users, 1);
 590         atomic_set(&mm->mm_count, 1);
 591         init_rwsem(&mm->mmap_sem);
 592         INIT_LIST_HEAD(&mm->mmlist);
 593         mm->core_state = NULL;
 594         atomic_long_set(&mm->nr_ptes, 0);
 595         mm_nr_pmds_init(mm);
 596         mm->map_count = 0;
 597         mm->locked_vm = 0;
 598         mm->pinned_vm = 0;
 599         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 600         spin_lock_init(&mm->page_table_lock);
 601         mm_init_cpumask(mm);
 602         mm_init_aio(mm);
 603         mm_init_owner(mm, p);
 604         mmu_notifier_mm_init(mm);
 605         clear_tlb_flush_pending(mm);
 606 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 607         mm->pmd_huge_pte = NULL;
 608 #endif
 609
 610         if (current->mm) {
 611                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 612                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 613         } else {
 614                 mm->flags = default_dump_filter;
 615                 mm->def_flags = 0;
 616         }
 617
 618         if (mm_alloc_pgd(mm))
 619                 goto fail_nopgd;
 620
 621         if (init_new_context(p, mm))
 622                 goto fail_nocontext;
 623
 624         return mm;
 625
 626 fail_nocontext:
 627         mm_free_pgd(mm);
 628 fail_nopgd:
 629         free_mm(mm);
 630         return NULL;
 631 }
 632
 633 static void check_mm(struct mm_struct *mm)
 634 {
 635         int i;
 636
 637         for (i = 0; i < NR_MM_COUNTERS; i++) {
 638                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 639
 640                 if (unlikely(x))
 641                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 642                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 643         }
 644
 645         if (atomic_long_read(&mm->nr_ptes))
 646                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 647                                 atomic_long_read(&mm->nr_ptes));
 648         if (mm_nr_pmds(mm))
 649                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 650                                 mm_nr_pmds(mm));
 651
 652 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 653         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 654 #endif
 655 }
 656
 657 /*
 658  * Allocate and initialize an mm_struct.
 659  */
 660 struct mm_struct *mm_alloc(void)
 661 {
 662         struct mm_struct *mm;
 663
 664         mm = allocate_mm();
 665         if (!mm)
 666                 return NULL;
 667
 668         memset(mm, 0, sizeof(*mm));
 669         return mm_init(mm, current);
 670 }
 671
 672 /*
 673  * Called when the last reference to the mm
 674  * is dropped: either by a lazy thread or by
 675  * mmput. Free the page directory and the mm.
 676  */
 677 void __mmdrop(struct mm_struct *mm)
 678 {
 679         BUG_ON(mm == &init_mm);
 680         mm_free_pgd(mm);
 681         destroy_context(mm);
 682         mmu_notifier_mm_destroy(mm);
 683         check_mm(mm);
 684         free_mm(mm);
 685 }
 686 EXPORT_SYMBOL_GPL(__mmdrop);
 687
 688 /*
 689  * Decrement the use count and release all resources for an mm.
 690  */
 691 void mmput(struct mm_struct *mm)
 692 {
 693         might_sleep();
 694
 695         if (atomic_dec_and_test(&mm->mm_users)) {
 696                 uprobe_clear_state(mm);
 697                 exit_aio(mm);
 698                 ksm_exit(mm);
 699                 khugepaged_exit(mm); /* must run before exit_mmap */
 700                 exit_mmap(mm);
 701                 set_mm_exe_file(mm, NULL);
 702                 if (!list_empty(&mm->mmlist)) {
 703                         spin_lock(&mmlist_lock);
 704                         list_del(&mm->mmlist);
 705                         spin_unlock(&mmlist_lock);
 706                 }
 707                 if (mm->binfmt)
 708                         module_put(mm->binfmt->module);
 709                 mmdrop(mm);
 710         }
 711 }
 712 EXPORT_SYMBOL_GPL(mmput);
 713
 714 /**
 715  * set_mm_exe_file - change a reference to the mm's executable file
 716  *
 717  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 718  *
 719  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 720  * invocations: in mmput() nobody alive left, in execve task is single
 721  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 722  * mm->exe_file, but does so without using set_mm_exe_file() in order
 723  * to do avoid the need for any locks.
 724  */
 725 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 726 {
 727         struct file *old_exe_file;
 728
 729         /*
 730          * It is safe to dereference the exe_file without RCU as
 731          * this function is only called if nobody else can access
 732          * this mm -- see comment above for justification.
 733          */
 734         old_exe_file = rcu_dereference_raw(mm->exe_file);
 735
 736         if (new_exe_file)
 737                 get_file(new_exe_file);
 738         rcu_assign_pointer(mm->exe_file, new_exe_file);
 739         if (old_exe_file)
 740                 fput(old_exe_file);
 741 }
 742
 743 /**
 744  * get_mm_exe_file - acquire a reference to the mm's executable file
 745  *
 746  * Returns %NULL if mm has no associated executable file.
 747  * User must release file via fput().
 748  */
 749 struct file *get_mm_exe_file(struct mm_struct *mm)
 750 {
 751         struct file *exe_file;
 752
 753         rcu_read_lock();
 754         exe_file = rcu_dereference(mm->exe_file);
 755         if (exe_file && !get_file_rcu(exe_file))
 756                 exe_file = NULL;
 757         rcu_read_unlock();
 758         return exe_file;
 759 }
 760 EXPORT_SYMBOL(get_mm_exe_file);
 761
 762 /**
 763  * get_task_mm - acquire a reference to the task's mm
 764  *
 765  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 766  * this kernel workthread has transiently adopted a user mm with use_mm,
 767  * to do its AIO) is not set and if so returns a reference to it, after
 768  * bumping up the use count.  User must release the mm via mmput()
 769  * after use.  Typically used by /proc and ptrace.
 770  */
 771 struct mm_struct *get_task_mm(struct task_struct *task)
 772 {
 773         struct mm_struct *mm;
 774
 775         task_lock(task);
 776         mm = task->mm;
 777         if (mm) {
 778                 if (task->flags & PF_KTHREAD)
 779                         mm = NULL;
 780                 else
 781                         atomic_inc(&mm->mm_users);
 782         }
 783         task_unlock(task);
 784         return mm;
 785 }
 786 EXPORT_SYMBOL_GPL(get_task_mm);
 787
 788 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 789 {
 790         struct mm_struct *mm;
 791         int err;
 792
 793         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 794         if (err)
 795                 return ERR_PTR(err);
 796
 797         mm = get_task_mm(task);
 798         if (mm && mm != current->mm &&
 799                         !ptrace_may_access(task, mode)) {
 800                 mmput(mm);
 801                 mm = ERR_PTR(-EACCES);
 802         }
 803         mutex_unlock(&task->signal->cred_guard_mutex);
 804
 805         return mm;
 806 }
 807
 808 static void complete_vfork_done(struct task_struct *tsk)
 809 {
 810         struct completion *vfork;
 811
 812         task_lock(tsk);
 813         vfork = tsk->vfork_done;
 814         if (likely(vfork)) {
 815                 tsk->vfork_done = NULL;
 816                 complete(vfork);
 817         }
 818         task_unlock(tsk);
 819 }
 820
 821 static int wait_for_vfork_done(struct task_struct *child,
 822                                 struct completion *vfork)
 823 {
 824         int killed;
 825
 826         freezer_do_not_count();
 827         killed = wait_for_completion_killable(vfork);
 828         freezer_count();
 829
 830         if (killed) {
 831                 task_lock(child);
 832                 child->vfork_done = NULL;
 833                 task_unlock(child);
 834         }
 835
 836         put_task_struct(child);
 837         return killed;
 838 }
 839
 840 /* Please note the differences between mmput and mm_release.
 841  * mmput is called whenever we stop holding onto a mm_struct,
 842  * error success whatever.
 843  *
 844  * mm_release is called after a mm_struct has been removed
 845  * from the current process.
 846  *
 847  * This difference is important for error handling, when we
 848  * only half set up a mm_struct for a new process and need to restore
 849  * the old one.  Because we mmput the new mm_struct before
 850  * restoring the old one. . .
 851  * Eric Biederman 10 January 1998
 852  */
 853 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 854 {
 855         /* Get rid of any futexes when releasing the mm */
 856 #ifdef CONFIG_FUTEX
 857         if (unlikely(tsk->robust_list)) {
 858                 exit_robust_list(tsk);
 859                 tsk->robust_list = NULL;
 860         }
 861 #ifdef CONFIG_COMPAT
 862         if (unlikely(tsk->compat_robust_list)) {
 863                 compat_exit_robust_list(tsk);
 864                 tsk->compat_robust_list = NULL;
 865         }
 866 #endif
 867         if (unlikely(!list_empty(&tsk->pi_state_list)))
 868                 exit_pi_state_list(tsk);
 869 #endif
 870
 871         uprobe_free_utask(tsk);
 872
 873         /* Get rid of any cached register state */
 874         deactivate_mm(tsk, mm);
 875
 876         /*
 877          * If we're exiting normally, clear a user-space tid field if
 878          * requested.  We leave this alone when dying by signal, to leave
 879          * the value intact in a core dump, and to save the unnecessary
 880          * trouble, say, a killed vfork parent shouldn't touch this mm.
 881          * Userland only wants this done for a sys_exit.
 882          */
 883         if (tsk->clear_child_tid) {
 884                 if (!(tsk->flags & PF_SIGNALED) &&
 885                     atomic_read(&mm->mm_users) > 1) {
 886                         /*
 887                          * We don't check the error code - if userspace has
 888                          * not set up a proper pointer then tough luck.
 889                          */
 890                         put_user(0, tsk->clear_child_tid);
 891                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 892                                         1, NULL, NULL, 0);
 893                 }
 894                 tsk->clear_child_tid = NULL;
 895         }
 896
 897         /*
 898          * All done, finally we can wake up parent and return this mm to him.
 899          * Also kthread_stop() uses this completion for synchronization.
 900          */
 901         if (tsk->vfork_done)
 902                 complete_vfork_done(tsk);
 903 }
 904
 905 /*
 906  * Allocate a new mm structure and copy contents from the
 907  * mm structure of the passed in task structure.
 908  */
 909 static struct mm_struct *dup_mm(struct task_struct *tsk)
 910 {
 911         struct mm_struct *mm, *oldmm = current->mm;
 912         int err;
 913
 914         mm = allocate_mm();
 915         if (!mm)
 916                 goto fail_nomem;
 917
 918         memcpy(mm, oldmm, sizeof(*mm));
 919
 920         if (!mm_init(mm, tsk))
 921                 goto fail_nomem;
 922
 923         err = dup_mmap(mm, oldmm);
 924         if (err)
 925                 goto free_pt;
 926
 927         mm->hiwater_rss = get_mm_rss(mm);
 928         mm->hiwater_vm = mm->total_vm;
 929
 930         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 931                 goto free_pt;
 932
 933         return mm;
 934
 935 free_pt:
 936         /* don't put binfmt in mmput, we haven't got module yet */
 937         mm->binfmt = NULL;
 938         mmput(mm);
 939
 940 fail_nomem:
 941         return NULL;
 942 }
 943
 944 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 945 {
 946         struct mm_struct *mm, *oldmm;
 947         int retval;
 948
 949         tsk->min_flt = tsk->maj_flt = 0;
 950         tsk->nvcsw = tsk->nivcsw = 0;
 951 #ifdef CONFIG_DETECT_HUNG_TASK
 952         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 953 #endif
 954
 955         tsk->mm = NULL;
 956         tsk->active_mm = NULL;
 957
 958         /*
 959          * Are we cloning a kernel thread?
 960          *
 961          * We need to steal a active VM for that..
 962          */
 963         oldmm = current->mm;
 964         if (!oldmm)
 965                 return 0;
 966
 967         /* initialize the new vmacache entries */
 968         vmacache_flush(tsk);
 969
 970         if (clone_flags & CLONE_VM) {
 971                 atomic_inc(&oldmm->mm_users);
 972                 mm = oldmm;
 973                 goto good_mm;
 974         }
 975
 976         retval = -ENOMEM;
 977         mm = dup_mm(tsk);
 978         if (!mm)
 979                 goto fail_nomem;
 980
 981 good_mm:
 982         tsk->mm = mm;
 983         tsk->active_mm = mm;
 984         return 0;
 985
 986 fail_nomem:
 987         return retval;
 988 }
 989
 990 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 991 {
 992         struct fs_struct *fs = current->fs;
 993         if (clone_flags & CLONE_FS) {
 994                 /* tsk->fs is already what we want */
 995                 spin_lock(&fs->lock);
 996                 if (fs->in_exec) {
 997                         spin_unlock(&fs->lock);
 998                         return -EAGAIN;
 999                 }
1000                 fs->users++;
1001                 spin_unlock(&fs->lock);
1002                 return 0;
1003         }
1004         tsk->fs = copy_fs_struct(fs);
1005         if (!tsk->fs)
1006                 return -ENOMEM;
1007         return 0;
1008 }
1009
1010 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1011 {
1012         struct files_struct *oldf, *newf;
1013         int error = 0;
1014
1015         /*
1016          * A background process may not have any files ...
1017          */
1018         oldf = current->files;
1019         if (!oldf)
1020                 goto out;
1021
1022         if (clone_flags & CLONE_FILES) {
1023                 atomic_inc(&oldf->count);
1024                 goto out;
1025         }
1026
1027         newf = dup_fd(oldf, &error);
1028         if (!newf)
1029                 goto out;
1030
1031         tsk->files = newf;
1032         error = 0;
1033 out:
1034         return error;
1035 }
1036
1037 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1038 {
1039 #ifdef CONFIG_BLOCK
1040         struct io_context *ioc = current->io_context;
1041         struct io_context *new_ioc;
1042
1043         if (!ioc)
1044                 return 0;
1045         /*
1046          * Share io context with parent, if CLONE_IO is set
1047          */
1048         if (clone_flags & CLONE_IO) {
1049                 ioc_task_link(ioc);
1050                 tsk->io_context = ioc;
1051         } else if (ioprio_valid(ioc->ioprio)) {
1052                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1053                 if (unlikely(!new_ioc))
1054                         return -ENOMEM;
1055
1056                 new_ioc->ioprio = ioc->ioprio;
1057                 put_io_context(new_ioc);
1058         }
1059 #endif
1060         return 0;
1061 }
1062
1063 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1064 {
1065         struct sighand_struct *sig;
1066
1067         if (clone_flags & CLONE_SIGHAND) {
1068                 atomic_inc(&current->sighand->count);
1069                 return 0;
1070         }
1071         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1072         rcu_assign_pointer(tsk->sighand, sig);
1073         if (!sig)
1074                 return -ENOMEM;
1075
1076         atomic_set(&sig->count, 1);
1077         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1078         return 0;
1079 }
1080
1081 void __cleanup_sighand(struct sighand_struct *sighand)
1082 {
1083         if (atomic_dec_and_test(&sighand->count)) {
1084                 signalfd_cleanup(sighand);
1085                 /*
1086                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1087                  * without an RCU grace period, see __lock_task_sighand().
1088                  */
1089                 kmem_cache_free(sighand_cachep, sighand);
1090         }
1091 }
1092
1093 /*
1094  * Initialize POSIX timer handling for a thread group.
1095  */
1096 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1097 {
1098         unsigned long cpu_limit;
1099
1100         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1101         if (cpu_limit != RLIM_INFINITY) {
1102                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1103                 sig->cputimer.running = 1;
1104         }
1105
1106         /* The timer lists. */
1107         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1108         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1109         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1110 }
1111
1112 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1113 {
1114         struct signal_struct *sig;
1115
1116         if (clone_flags & CLONE_THREAD)
1117                 return 0;
1118
1119         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1120         tsk->signal = sig;
1121         if (!sig)
1122                 return -ENOMEM;
1123
1124         sig->nr_threads = 1;
1125         atomic_set(&sig->live, 1);
1126         atomic_set(&sig->sigcnt, 1);
1127
1128         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1129         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1130         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1131
1132         init_waitqueue_head(&sig->wait_chldexit);
1133         sig->curr_target = tsk;
1134         init_sigpending(&sig->shared_pending);
1135         INIT_LIST_HEAD(&sig->posix_timers);
1136         seqlock_init(&sig->stats_lock);
1137         prev_cputime_init(&sig->prev_cputime);
1138
1139         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1140         sig->real_timer.function = it_real_fn;
1141
1142         task_lock(current->group_leader);
1143         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1144         task_unlock(current->group_leader);
1145
1146         posix_cpu_timers_init_group(sig);
1147
1148         tty_audit_fork(sig);
1149         sched_autogroup_fork(sig);
1150
1151         sig->oom_score_adj = current->signal->oom_score_adj;
1152         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1153
1154         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1155                                    current->signal->is_child_subreaper;
1156
1157         mutex_init(&sig->cred_guard_mutex);
1158
1159         return 0;
1160 }
1161
1162 static void copy_seccomp(struct task_struct *p)
1163 {
1164 #ifdef CONFIG_SECCOMP
1165         /*
1166          * Must be called with sighand->lock held, which is common to
1167          * all threads in the group. Holding cred_guard_mutex is not
1168          * needed because this new task is not yet running and cannot
1169          * be racing exec.
1170          */
1171         assert_spin_locked(&current->sighand->siglock);
1172
1173         /* Ref-count the new filter user, and assign it. */
1174         get_seccomp_filter(current);
1175         p->seccomp = current->seccomp;
1176
1177         /*
1178          * Explicitly enable no_new_privs here in case it got set
1179          * between the task_struct being duplicated and holding the
1180          * sighand lock. The seccomp state and nnp must be in sync.
1181          */
1182         if (task_no_new_privs(current))
1183                 task_set_no_new_privs(p);
1184
1185         /*
1186          * If the parent gained a seccomp mode after copying thread
1187          * flags and between before we held the sighand lock, we have
1188          * to manually enable the seccomp thread flag here.
1189          */
1190         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1191                 set_tsk_thread_flag(p, TIF_SECCOMP);
1192 #endif
1193 }
1194
1195 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1196 {
1197         current->clear_child_tid = tidptr;
1198
1199         return task_pid_vnr(current);
1200 }
1201
1202 static void rt_mutex_init_task(struct task_struct *p)
1203 {
1204         raw_spin_lock_init(&p->pi_lock);
1205 #ifdef CONFIG_RT_MUTEXES
1206         p->pi_waiters = RB_ROOT;
1207         p->pi_waiters_leftmost = NULL;
1208         p->pi_blocked_on = NULL;
1209 #endif
1210 }
1211
1212 /*
1213  * Initialize POSIX timer handling for a single task.
1214  */
1215 static void posix_cpu_timers_init(struct task_struct *tsk)
1216 {
1217         tsk->cputime_expires.prof_exp = 0;
1218         tsk->cputime_expires.virt_exp = 0;
1219         tsk->cputime_expires.sched_exp = 0;
1220         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1221         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1222         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1223 }
1224
1225 static inline void
1226 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1227 {
1228          task->pids[type].pid = pid;
1229 }
1230
1231 /*
1232  * This creates a new process as a copy of the old one,
1233  * but does not actually start it yet.
1234  *
1235  * It copies the registers, and all the appropriate
1236  * parts of the process environment (as per the clone
1237  * flags). The actual kick-off is left to the caller.
1238  */
1239 static struct task_struct *copy_process(unsigned long clone_flags,
1240                                         unsigned long stack_start,
1241                                         unsigned long stack_size,
1242                                         int __user *child_tidptr,
1243                                         struct pid *pid,
1244                                         int trace,
1245                                         unsigned long tls)
1246 {
1247         int retval;
1248         struct task_struct *p;
1249
1250         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1251                 return ERR_PTR(-EINVAL);
1252
1253         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1254                 return ERR_PTR(-EINVAL);
1255
1256         /*
1257          * Thread groups must share signals as well, and detached threads
1258          * can only be started up within the thread group.
1259          */
1260         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1261                 return ERR_PTR(-EINVAL);
1262
1263         /*
1264          * Shared signal handlers imply shared VM. By way of the above,
1265          * thread groups also imply shared VM. Blocking this case allows
1266          * for various simplifications in other code.
1267          */
1268         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1269                 return ERR_PTR(-EINVAL);
1270
1271         /*
1272          * Siblings of global init remain as zombies on exit since they are
1273          * not reaped by their parent (swapper). To solve this and to avoid
1274          * multi-rooted process trees, prevent global and container-inits
1275          * from creating siblings.
1276          */
1277         if ((clone_flags & CLONE_PARENT) &&
1278                                 current->signal->flags & SIGNAL_UNKILLABLE)
1279                 return ERR_PTR(-EINVAL);
1280
1281         /*
1282          * If the new process will be in a different pid or user namespace
1283          * do not allow it to share a thread group or signal handlers or
1284          * parent with the forking task.
1285          */
1286         if (clone_flags & CLONE_SIGHAND) {
1287                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1288                     (task_active_pid_ns(current) !=
1289                                 current->nsproxy->pid_ns_for_children))
1290                         return ERR_PTR(-EINVAL);
1291         }
1292
1293         retval = security_task_create(clone_flags);
1294         if (retval)
1295                 goto fork_out;
1296
1297         retval = -ENOMEM;
1298         p = dup_task_struct(current);
1299         if (!p)
1300                 goto fork_out;
1301
1302         ftrace_graph_init_task(p);
1303
1304         rt_mutex_init_task(p);
1305
1306 #ifdef CONFIG_PROVE_LOCKING
1307         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1308         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1309 #endif
1310         retval = -EAGAIN;
1311         if (atomic_read(&p->real_cred->user->processes) >=
1312                         task_rlimit(p, RLIMIT_NPROC)) {
1313                 if (p->real_cred->user != INIT_USER &&
1314                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1315                         goto bad_fork_free;
1316         }
1317         current->flags &= ~PF_NPROC_EXCEEDED;
1318
1319         retval = copy_creds(p, clone_flags);
1320         if (retval < 0)
1321                 goto bad_fork_free;
1322
1323         /*
1324          * If multiple threads are within copy_process(), then this check
1325          * triggers too late. This doesn't hurt, the check is only there
1326          * to stop root fork bombs.
1327          */
1328         retval = -EAGAIN;
1329         if (nr_threads >= max_threads)
1330                 goto bad_fork_cleanup_count;
1331
1332         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1333         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1334         p->flags |= PF_FORKNOEXEC;
1335         INIT_LIST_HEAD(&p->children);
1336         INIT_LIST_HEAD(&p->sibling);
1337         rcu_copy_process(p);
1338         p->vfork_done = NULL;
1339         spin_lock_init(&p->alloc_lock);
1340
1341         init_sigpending(&p->pending);
1342
1343         p->utime = p->stime = p->gtime = 0;
1344         p->utimescaled = p->stimescaled = 0;
1345         prev_cputime_init(&p->prev_cputime);
1346
1347 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1348         seqlock_init(&p->vtime_seqlock);
1349         p->vtime_snap = 0;
1350         p->vtime_snap_whence = VTIME_SLEEPING;
1351 #endif
1352
1353 #if defined(SPLIT_RSS_COUNTING)
1354         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1355 #endif
1356
1357         p->default_timer_slack_ns = current->timer_slack_ns;
1358
1359         task_io_accounting_init(&p->ioac);
1360         acct_clear_integrals(p);
1361
1362         posix_cpu_timers_init(p);
1363
1364         p->start_time = ktime_get_ns();
1365         p->real_start_time = ktime_get_boot_ns();
1366         p->io_context = NULL;
1367         p->audit_context = NULL;
1368         if (clone_flags & CLONE_THREAD)
1369                 threadgroup_change_begin(current);
1370         cgroup_fork(p);
1371 #ifdef CONFIG_NUMA
1372         p->mempolicy = mpol_dup(p->mempolicy);
1373         if (IS_ERR(p->mempolicy)) {
1374                 retval = PTR_ERR(p->mempolicy);
1375                 p->mempolicy = NULL;
1376                 goto bad_fork_cleanup_threadgroup_lock;
1377         }
1378 #endif
1379 #ifdef CONFIG_CPUSETS
1380         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1381         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1382         seqcount_init(&p->mems_allowed_seq);
1383 #endif
1384 #ifdef CONFIG_TRACE_IRQFLAGS
1385         p->irq_events = 0;
1386         p->hardirqs_enabled = 0;
1387         p->hardirq_enable_ip = 0;
1388         p->hardirq_enable_event = 0;
1389         p->hardirq_disable_ip = _THIS_IP_;
1390         p->hardirq_disable_event = 0;
1391         p->softirqs_enabled = 1;
1392         p->softirq_enable_ip = _THIS_IP_;
1393         p->softirq_enable_event = 0;
1394         p->softirq_disable_ip = 0;
1395         p->softirq_disable_event = 0;
1396         p->hardirq_context = 0;
1397         p->softirq_context = 0;
1398 #endif
1399
1400         p->pagefault_disabled = 0;
1401
1402 #ifdef CONFIG_LOCKDEP
1403         p->lockdep_depth = 0; /* no locks held yet */
1404         p->curr_chain_key = 0;
1405         p->lockdep_recursion = 0;
1406 #endif
1407
1408 #ifdef CONFIG_DEBUG_MUTEXES
1409         p->blocked_on = NULL; /* not blocked yet */
1410 #endif
1411 #ifdef CONFIG_BCACHE
1412         p->sequential_io        = 0;
1413         p->sequential_io_avg    = 0;
1414 #endif
1415
1416         /* Perform scheduler related setup. Assign this task to a CPU. */
1417         retval = sched_fork(clone_flags, p);
1418         if (retval)
1419                 goto bad_fork_cleanup_policy;
1420
1421         retval = perf_event_init_task(p);
1422         if (retval)
1423                 goto bad_fork_cleanup_policy;
1424         retval = audit_alloc(p);
1425         if (retval)
1426                 goto bad_fork_cleanup_perf;
1427         /* copy all the process information */
1428         shm_init_task(p);
1429         retval = copy_semundo(clone_flags, p);
1430         if (retval)
1431                 goto bad_fork_cleanup_audit;
1432         retval = copy_files(clone_flags, p);
1433         if (retval)
1434                 goto bad_fork_cleanup_semundo;
1435         retval = copy_fs(clone_flags, p);
1436         if (retval)
1437                 goto bad_fork_cleanup_files;
1438         retval = copy_sighand(clone_flags, p);
1439         if (retval)
1440                 goto bad_fork_cleanup_fs;
1441         retval = copy_signal(clone_flags, p);
1442         if (retval)
1443                 goto bad_fork_cleanup_sighand;
1444         retval = copy_mm(clone_flags, p);
1445         if (retval)
1446                 goto bad_fork_cleanup_signal;
1447         retval = copy_namespaces(clone_flags, p);
1448         if (retval)
1449                 goto bad_fork_cleanup_mm;
1450         retval = copy_io(clone_flags, p);
1451         if (retval)
1452                 goto bad_fork_cleanup_namespaces;
1453         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1454         if (retval)
1455                 goto bad_fork_cleanup_io;
1456
1457         if (pid != &init_struct_pid) {
1458                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1459                 if (IS_ERR(pid)) {
1460                         retval = PTR_ERR(pid);
1461                         goto bad_fork_cleanup_io;
1462                 }
1463         }
1464
1465         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1466         /*
1467          * Clear TID on mm_release()?
1468          */
1469         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1470 #ifdef CONFIG_BLOCK
1471         p->plug = NULL;
1472 #endif
1473 #ifdef CONFIG_FUTEX
1474         p->robust_list = NULL;
1475 #ifdef CONFIG_COMPAT
1476         p->compat_robust_list = NULL;
1477 #endif
1478         INIT_LIST_HEAD(&p->pi_state_list);
1479         p->pi_state_cache = NULL;
1480 #endif
1481         /*
1482          * sigaltstack should be cleared when sharing the same VM
1483          */
1484         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1485                 p->sas_ss_sp = p->sas_ss_size = 0;
1486
1487         /*
1488          * Syscall tracing and stepping should be turned off in the
1489          * child regardless of CLONE_PTRACE.
1490          */
1491         user_disable_single_step(p);
1492         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1493 #ifdef TIF_SYSCALL_EMU
1494         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1495 #endif
1496         clear_all_latency_tracing(p);
1497
1498         /* ok, now we should be set up.. */
1499         p->pid = pid_nr(pid);
1500         if (clone_flags & CLONE_THREAD) {
1501                 p->exit_signal = -1;
1502                 p->group_leader = current->group_leader;
1503                 p->tgid = current->tgid;
1504         } else {
1505                 if (clone_flags & CLONE_PARENT)
1506                         p->exit_signal = current->group_leader->exit_signal;
1507                 else
1508                         p->exit_signal = (clone_flags & CSIGNAL);
1509                 p->group_leader = p;
1510                 p->tgid = p->pid;
1511         }
1512
1513         p->nr_dirtied = 0;
1514         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1515         p->dirty_paused_when = 0;
1516
1517         p->pdeath_signal = 0;
1518         INIT_LIST_HEAD(&p->thread_group);
1519         p->task_works = NULL;
1520
1521         /*
1522          * Make it visible to the rest of the system, but dont wake it up yet.
1523          * Need tasklist lock for parent etc handling!
1524          */
1525         write_lock_irq(&tasklist_lock);
1526
1527         /* CLONE_PARENT re-uses the old parent */
1528         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1529                 p->real_parent = current->real_parent;
1530                 p->parent_exec_id = current->parent_exec_id;
1531         } else {
1532                 p->real_parent = current;
1533                 p->parent_exec_id = current->self_exec_id;
1534         }
1535
1536         spin_lock(&current->sighand->siglock);
1537
1538         /*
1539          * Copy seccomp details explicitly here, in case they were changed
1540          * before holding sighand lock.
1541          */
1542         copy_seccomp(p);
1543
1544         /*
1545          * Process group and session signals need to be delivered to just the
1546          * parent before the fork or both the parent and the child after the
1547          * fork. Restart if a signal comes in before we add the new process to
1548          * it's process group.
1549          * A fatal signal pending means that current will exit, so the new
1550          * thread can't slip out of an OOM kill (or normal SIGKILL).
1551         */
1552         recalc_sigpending();
1553         if (signal_pending(current)) {
1554                 spin_unlock(&current->sighand->siglock);
1555                 write_unlock_irq(&tasklist_lock);
1556                 retval = -ERESTARTNOINTR;
1557                 goto bad_fork_free_pid;
1558         }
1559
1560         if (likely(p->pid)) {
1561                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1562
1563                 init_task_pid(p, PIDTYPE_PID, pid);
1564                 if (thread_group_leader(p)) {
1565                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1566                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1567
1568                         if (is_child_reaper(pid)) {
1569                                 ns_of_pid(pid)->child_reaper = p;
1570                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1571                         }
1572
1573                         p->signal->leader_pid = pid;
1574                         p->signal->tty = tty_kref_get(current->signal->tty);
1575                         list_add_tail(&p->sibling, &p->real_parent->children);
1576                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1577                         attach_pid(p, PIDTYPE_PGID);
1578                         attach_pid(p, PIDTYPE_SID);
1579                         __this_cpu_inc(process_counts);
1580                 } else {
1581                         current->signal->nr_threads++;
1582                         atomic_inc(&current->signal->live);
1583                         atomic_inc(&current->signal->sigcnt);
1584                         list_add_tail_rcu(&p->thread_group,
1585                                           &p->group_leader->thread_group);
1586                         list_add_tail_rcu(&p->thread_node,
1587                                           &p->signal->thread_head);
1588                 }
1589                 attach_pid(p, PIDTYPE_PID);
1590                 nr_threads++;
1591         }
1592
1593         total_forks++;
1594         spin_unlock(&current->sighand->siglock);
1595         syscall_tracepoint_update(p);
1596         write_unlock_irq(&tasklist_lock);
1597
1598         proc_fork_connector(p);
1599         cgroup_post_fork(p);
1600         if (clone_flags & CLONE_THREAD)
1601                 threadgroup_change_end(current);
1602         perf_event_fork(p);
1603
1604         trace_task_newtask(p, clone_flags);
1605         uprobe_copy_process(p, clone_flags);
1606
1607         return p;
1608
1609 bad_fork_free_pid:
1610         if (pid != &init_struct_pid)
1611                 free_pid(pid);
1612 bad_fork_cleanup_io:
1613         if (p->io_context)
1614                 exit_io_context(p);
1615 bad_fork_cleanup_namespaces:
1616         exit_task_namespaces(p);
1617 bad_fork_cleanup_mm:
1618         if (p->mm)
1619                 mmput(p->mm);
1620 bad_fork_cleanup_signal:
1621         if (!(clone_flags & CLONE_THREAD))
1622                 free_signal_struct(p->signal);
1623 bad_fork_cleanup_sighand:
1624         __cleanup_sighand(p->sighand);
1625 bad_fork_cleanup_fs:
1626         exit_fs(p); /* blocking */
1627 bad_fork_cleanup_files:
1628         exit_files(p); /* blocking */
1629 bad_fork_cleanup_semundo:
1630         exit_sem(p);
1631 bad_fork_cleanup_audit:
1632         audit_free(p);
1633 bad_fork_cleanup_perf:
1634         perf_event_free_task(p);
1635 bad_fork_cleanup_policy:
1636 #ifdef CONFIG_NUMA
1637         mpol_put(p->mempolicy);
1638 bad_fork_cleanup_threadgroup_lock:
1639 #endif
1640         if (clone_flags & CLONE_THREAD)
1641                 threadgroup_change_end(current);
1642         delayacct_tsk_free(p);
1643 bad_fork_cleanup_count:
1644         atomic_dec(&p->cred->user->processes);
1645         exit_creds(p);
1646 bad_fork_free:
1647         free_task(p);
1648 fork_out:
1649         return ERR_PTR(retval);
1650 }
1651
1652 static inline void init_idle_pids(struct pid_link *links)
1653 {
1654         enum pid_type type;
1655
1656         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1657                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1658                 links[type].pid = &init_struct_pid;
1659         }
1660 }
1661
1662 struct task_struct *fork_idle(int cpu)
1663 {
1664         struct task_struct *task;
1665         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0);
1666         if (!IS_ERR(task)) {
1667                 init_idle_pids(task->pids);
1668                 init_idle(task, cpu);
1669         }
1670
1671         return task;
1672 }
1673
1674 /*
1675  *  Ok, this is the main fork-routine.
1676  *
1677  * It copies the process, and if successful kick-starts
1678  * it and waits for it to finish using the VM if required.
1679  */
1680 long _do_fork(unsigned long clone_flags,
1681               unsigned long stack_start,
1682               unsigned long stack_size,
1683               int __user *parent_tidptr,
1684               int __user *child_tidptr,
1685               unsigned long tls)
1686 {
1687         struct task_struct *p;
1688         int trace = 0;
1689         long nr;
1690
1691         /*
1692          * Determine whether and which event to report to ptracer.  When
1693          * called from kernel_thread or CLONE_UNTRACED is explicitly
1694          * requested, no event is reported; otherwise, report if the event
1695          * for the type of forking is enabled.
1696          */
1697         if (!(clone_flags & CLONE_UNTRACED)) {
1698                 if (clone_flags & CLONE_VFORK)
1699                         trace = PTRACE_EVENT_VFORK;
1700                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1701                         trace = PTRACE_EVENT_CLONE;
1702                 else
1703                         trace = PTRACE_EVENT_FORK;
1704
1705                 if (likely(!ptrace_event_enabled(current, trace)))
1706                         trace = 0;
1707         }
1708
1709         p = copy_process(clone_flags, stack_start, stack_size,
1710                          child_tidptr, NULL, trace, tls);
1711         /*
1712          * Do this prior waking up the new thread - the thread pointer
1713          * might get invalid after that point, if the thread exits quickly.
1714          */
1715         if (!IS_ERR(p)) {
1716                 struct completion vfork;
1717                 struct pid *pid;
1718
1719                 trace_sched_process_fork(current, p);
1720
1721                 pid = get_task_pid(p, PIDTYPE_PID);
1722                 nr = pid_vnr(pid);
1723
1724                 if (clone_flags & CLONE_PARENT_SETTID)
1725                         put_user(nr, parent_tidptr);
1726
1727                 if (clone_flags & CLONE_VFORK) {
1728                         p->vfork_done = &vfork;
1729                         init_completion(&vfork);
1730                         get_task_struct(p);
1731                 }
1732
1733                 wake_up_new_task(p);
1734
1735                 /* forking complete and child started to run, tell ptracer */
1736                 if (unlikely(trace))
1737                         ptrace_event_pid(trace, pid);
1738
1739                 if (clone_flags & CLONE_VFORK) {
1740                         if (!wait_for_vfork_done(p, &vfork))
1741                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1742                 }
1743
1744                 put_pid(pid);
1745         } else {
1746                 nr = PTR_ERR(p);
1747         }
1748         return nr;
1749 }
1750
1751 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1752 /* For compatibility with architectures that call do_fork directly rather than
1753  * using the syscall entry points below. */
1754 long do_fork(unsigned long clone_flags,
1755               unsigned long stack_start,
1756               unsigned long stack_size,
1757               int __user *parent_tidptr,
1758               int __user *child_tidptr)
1759 {
1760         return _do_fork(clone_flags, stack_start, stack_size,
1761                         parent_tidptr, child_tidptr, 0);
1762 }
1763 #endif
1764
1765 /*
1766  * Create a kernel thread.
1767  */
1768 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1769 {
1770         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1771                 (unsigned long)arg, NULL, NULL, 0);
1772 }
1773
1774 #ifdef __ARCH_WANT_SYS_FORK
1775 SYSCALL_DEFINE0(fork)
1776 {
1777 #ifdef CONFIG_MMU
1778         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1779 #else
1780         /* can not support in nommu mode */
1781         return -EINVAL;
1782 #endif
1783 }
1784 #endif
1785
1786 #ifdef __ARCH_WANT_SYS_VFORK
1787 SYSCALL_DEFINE0(vfork)
1788 {
1789         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1790                         0, NULL, NULL, 0);
1791 }
1792 #endif
1793
1794 #ifdef __ARCH_WANT_SYS_CLONE
1795 #ifdef CONFIG_CLONE_BACKWARDS
1796 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1797                  int __user *, parent_tidptr,
1798                  unsigned long, tls,
1799                  int __user *, child_tidptr)
1800 #elif defined(CONFIG_CLONE_BACKWARDS2)
1801 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1802                  int __user *, parent_tidptr,
1803                  int __user *, child_tidptr,
1804                  unsigned long, tls)
1805 #elif defined(CONFIG_CLONE_BACKWARDS3)
1806 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1807                 int, stack_size,
1808                 int __user *, parent_tidptr,
1809                 int __user *, child_tidptr,
1810                 unsigned long, tls)
1811 #else
1812 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1813                  int __user *, parent_tidptr,
1814                  int __user *, child_tidptr,
1815                  unsigned long, tls)
1816 #endif
1817 {
1818         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
1819 }
1820 #endif
1821
1822 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1823 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1824 #endif
1825
1826 static void sighand_ctor(void *data)
1827 {
1828         struct sighand_struct *sighand = data;
1829
1830         spin_lock_init(&sighand->siglock);
1831         init_waitqueue_head(&sighand->signalfd_wqh);
1832 }
1833
1834 void __init proc_caches_init(void)
1835 {
1836         sighand_cachep = kmem_cache_create("sighand_cache",
1837                         sizeof(struct sighand_struct), 0,
1838                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1839                         SLAB_NOTRACK, sighand_ctor);
1840         signal_cachep = kmem_cache_create("signal_cache",
1841                         sizeof(struct signal_struct), 0,
1842                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1843         files_cachep = kmem_cache_create("files_cache",
1844                         sizeof(struct files_struct), 0,
1845                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1846         fs_cachep = kmem_cache_create("fs_cache",
1847                         sizeof(struct fs_struct), 0,
1848                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1849         /*
1850          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1851          * whole struct cpumask for the OFFSTACK case. We could change
1852          * this to *only* allocate as much of it as required by the
1853          * maximum number of CPU's we can ever have.  The cpumask_allocation
1854          * is at the end of the structure, exactly for that reason.
1855          */
1856         mm_cachep = kmem_cache_create("mm_struct",
1857                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1858                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1859         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1860         mmap_init();
1861         nsproxy_cache_init();
1862 }
1863
1864 /*
1865  * Check constraints on flags passed to the unshare system call.
1866  */
1867 static int check_unshare_flags(unsigned long unshare_flags)
1868 {
1869         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1870                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1871                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1872                                 CLONE_NEWUSER|CLONE_NEWPID))
1873                 return -EINVAL;
1874         /*
1875          * Not implemented, but pretend it works if there is nothing to
1876          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1877          * needs to unshare vm.
1878          */
1879         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1880                 /* FIXME: get_task_mm() increments ->mm_users */
1881                 if (atomic_read(&current->mm->mm_users) > 1)
1882                         return -EINVAL;
1883         }
1884
1885         return 0;
1886 }
1887
1888 /*
1889  * Unshare the filesystem structure if it is being shared
1890  */
1891 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1892 {
1893         struct fs_struct *fs = current->fs;
1894
1895         if (!(unshare_flags & CLONE_FS) || !fs)
1896                 return 0;
1897
1898         /* don't need lock here; in the worst case we'll do useless copy */
1899         if (fs->users == 1)
1900                 return 0;
1901
1902         *new_fsp = copy_fs_struct(fs);
1903         if (!*new_fsp)
1904                 return -ENOMEM;
1905
1906         return 0;
1907 }
1908
1909 /*
1910  * Unshare file descriptor table if it is being shared
1911  */
1912 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1913 {
1914         struct files_struct *fd = current->files;
1915         int error = 0;
1916
1917         if ((unshare_flags & CLONE_FILES) &&
1918             (fd && atomic_read(&fd->count) > 1)) {
1919                 *new_fdp = dup_fd(fd, &error);
1920                 if (!*new_fdp)
1921                         return error;
1922         }
1923
1924         return 0;
1925 }
1926
1927 /*
1928  * unshare allows a process to 'unshare' part of the process
1929  * context which was originally shared using clone.  copy_*
1930  * functions used by do_fork() cannot be used here directly
1931  * because they modify an inactive task_struct that is being
1932  * constructed. Here we are modifying the current, active,
1933  * task_struct.
1934  */
1935 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1936 {
1937         struct fs_struct *fs, *new_fs = NULL;
1938         struct files_struct *fd, *new_fd = NULL;
1939         struct cred *new_cred = NULL;
1940         struct nsproxy *new_nsproxy = NULL;
1941         int do_sysvsem = 0;
1942         int err;
1943
1944         /*
1945          * If unsharing a user namespace must also unshare the thread.
1946          */
1947         if (unshare_flags & CLONE_NEWUSER)
1948                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1949         /*
1950          * If unsharing a thread from a thread group, must also unshare vm.
1951          */
1952         if (unshare_flags & CLONE_THREAD)
1953                 unshare_flags |= CLONE_VM;
1954         /*
1955          * If unsharing vm, must also unshare signal handlers.
1956          */
1957         if (unshare_flags & CLONE_VM)
1958                 unshare_flags |= CLONE_SIGHAND;
1959         /*
1960          * If unsharing namespace, must also unshare filesystem information.
1961          */
1962         if (unshare_flags & CLONE_NEWNS)
1963                 unshare_flags |= CLONE_FS;
1964
1965         err = check_unshare_flags(unshare_flags);
1966         if (err)
1967                 goto bad_unshare_out;
1968         /*
1969          * CLONE_NEWIPC must also detach from the undolist: after switching
1970          * to a new ipc namespace, the semaphore arrays from the old
1971          * namespace are unreachable.
1972          */
1973         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1974                 do_sysvsem = 1;
1975         err = unshare_fs(unshare_flags, &new_fs);
1976         if (err)
1977                 goto bad_unshare_out;
1978         err = unshare_fd(unshare_flags, &new_fd);
1979         if (err)
1980                 goto bad_unshare_cleanup_fs;
1981         err = unshare_userns(unshare_flags, &new_cred);
1982         if (err)
1983                 goto bad_unshare_cleanup_fd;
1984         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1985                                          new_cred, new_fs);
1986         if (err)
1987                 goto bad_unshare_cleanup_cred;
1988
1989         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1990                 if (do_sysvsem) {
1991                         /*
1992                          * CLONE_SYSVSEM is equivalent to sys_exit().
1993                          */
1994                         exit_sem(current);
1995                 }
1996                 if (unshare_flags & CLONE_NEWIPC) {
1997                         /* Orphan segments in old ns (see sem above). */
1998                         exit_shm(current);
1999                         shm_init_task(current);
2000                 }
2001
2002                 if (new_nsproxy)
2003                         switch_task_namespaces(current, new_nsproxy);
2004
2005                 task_lock(current);
2006
2007                 if (new_fs) {
2008                         fs = current->fs;
2009                         spin_lock(&fs->lock);
2010                         current->fs = new_fs;
2011                         if (--fs->users)
2012                                 new_fs = NULL;
2013                         else
2014                                 new_fs = fs;
2015                         spin_unlock(&fs->lock);
2016                 }
2017
2018                 if (new_fd) {
2019                         fd = current->files;
2020                         current->files = new_fd;
2021                         new_fd = fd;
2022                 }
2023
2024                 task_unlock(current);
2025
2026                 if (new_cred) {
2027                         /* Install the new user namespace */
2028                         commit_creds(new_cred);
2029                         new_cred = NULL;
2030                 }
2031         }
2032
2033 bad_unshare_cleanup_cred:
2034         if (new_cred)
2035                 put_cred(new_cred);
2036 bad_unshare_cleanup_fd:
2037         if (new_fd)
2038                 put_files_struct(new_fd);
2039
2040 bad_unshare_cleanup_fs:
2041         if (new_fs)
2042                 free_fs_struct(new_fs);
2043
2044 bad_unshare_out:
2045         return err;
2046 }
2047
2048 /*
2049  *      Helper to unshare the files of the current task.
2050  *      We don't want to expose copy_files internals to
2051  *      the exec layer of the kernel.
2052  */
2053
2054 int unshare_files(struct files_struct **displaced)
2055 {
2056         struct task_struct *task = current;
2057         struct files_struct *copy = NULL;
2058         int error;
2059
2060         error = unshare_fd(CLONE_FILES, &copy);
2061         if (error || !copy) {
2062                 *displaced = NULL;
2063                 return error;
2064         }
2065         *displaced = task->files;
2066         task_lock(task);
2067         task->files = copy;
2068         task_unlock(task);
2069         return 0;
2070 }
2071
2072 int sysctl_max_threads(struct ctl_table *table, int write,
2073                        void __user *buffer, size_t *lenp, loff_t *ppos)
2074 {
2075         struct ctl_table t;
2076         int ret;
2077         int threads = max_threads;
2078         int min = MIN_THREADS;
2079         int max = MAX_THREADS;
2080
2081         t = *table;
2082         t.data = &threads;
2083         t.extra1 = &min;
2084         t.extra2 = &max;
2085
2086         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2087         if (ret || !write)
2088                 return ret;
2089
2090         set_max_threads(threads);
2091
2092         return 0;
2093 }