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