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 #include <linux/kcov.h>
  79
  80 #include <asm/pgtable.h>
  81 #include <asm/pgalloc.h>
  82 #include <asm/uaccess.h>
  83 #include <asm/mmu_context.h>
  84 #include <asm/cacheflush.h>
  85 #include <asm/tlbflush.h>
  86
  87 #include <trace/events/sched.h>
  88
  89 #define CREATE_TRACE_POINTS
  90 #include <trace/events/task.h>
  91
  92 /*
  93  * Minimum number of threads to boot the kernel
  94  */
  95 #define MIN_THREADS 20
  96
  97 /*
  98  * Maximum number of threads
  99  */
 100 #define MAX_THREADS FUTEX_TID_MASK
 101
 102 /*
 103  * Protected counters by write_lock_irq(&tasklist_lock)
 104  */
 105 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 106 int nr_threads;                 /* The idle threads do not count.. */
 107
 108 int max_threads;                /* tunable limit on nr_threads */
 109
 110 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 111
 112 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 113
 114 #ifdef CONFIG_PROVE_RCU
 115 int lockdep_tasklist_lock_is_held(void)
 116 {
 117         return lockdep_is_held(&tasklist_lock);
 118 }
 119 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 120 #endif /* #ifdef CONFIG_PROVE_RCU */
 121
 122 int nr_processes(void)
 123 {
 124         int cpu;
 125         int total = 0;
 126
 127         for_each_possible_cpu(cpu)
 128                 total += per_cpu(process_counts, cpu);
 129
 130         return total;
 131 }
 132
 133 void __weak arch_release_task_struct(struct task_struct *tsk)
 134 {
 135 }
 136
 137 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 138 static struct kmem_cache *task_struct_cachep;
 139
 140 static inline struct task_struct *alloc_task_struct_node(int node)
 141 {
 142         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 143 }
 144
 145 static inline void free_task_struct(struct task_struct *tsk)
 146 {
 147         kmem_cache_free(task_struct_cachep, tsk);
 148 }
 149 #endif
 150
 151 void __weak arch_release_thread_info(struct thread_info *ti)
 152 {
 153 }
 154
 155 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
 156
 157 /*
 158  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 159  * kmemcache based allocator.
 160  */
 161 # if THREAD_SIZE >= PAGE_SIZE
 162 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 163                                                   int node)
 164 {
 165         struct page *page = alloc_kmem_pages_node(node, THREADINFO_GFP,
 166                                                   THREAD_SIZE_ORDER);
 167
 168         if (page)
 169                 memcg_kmem_update_page_stat(page, MEMCG_KERNEL_STACK,
 170                                             1 << THREAD_SIZE_ORDER);
 171
 172         return page ? page_address(page) : NULL;
 173 }
 174
 175 static inline void free_thread_info(struct thread_info *ti)
 176 {
 177         struct page *page = virt_to_page(ti);
 178
 179         memcg_kmem_update_page_stat(page, MEMCG_KERNEL_STACK,
 180                                     -(1 << THREAD_SIZE_ORDER));
 181         __free_kmem_pages(page, THREAD_SIZE_ORDER);
 182 }
 183 # else
 184 static struct kmem_cache *thread_info_cache;
 185
 186 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 187                                                   int node)
 188 {
 189         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
 190 }
 191
 192 static void free_thread_info(struct thread_info *ti)
 193 {
 194         kmem_cache_free(thread_info_cache, ti);
 195 }
 196
 197 void thread_info_cache_init(void)
 198 {
 199         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
 200                                               THREAD_SIZE, 0, NULL);
 201         BUG_ON(thread_info_cache == NULL);
 202 }
 203 # endif
 204 #endif
 205
 206 /* SLAB cache for signal_struct structures (tsk->signal) */
 207 static struct kmem_cache *signal_cachep;
 208
 209 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 210 struct kmem_cache *sighand_cachep;
 211
 212 /* SLAB cache for files_struct structures (tsk->files) */
 213 struct kmem_cache *files_cachep;
 214
 215 /* SLAB cache for fs_struct structures (tsk->fs) */
 216 struct kmem_cache *fs_cachep;
 217
 218 /* SLAB cache for vm_area_struct structures */
 219 struct kmem_cache *vm_area_cachep;
 220
 221 /* SLAB cache for mm_struct structures (tsk->mm) */
 222 static struct kmem_cache *mm_cachep;
 223
 224 static void account_kernel_stack(struct thread_info *ti, int account)
 225 {
 226         struct zone *zone = page_zone(virt_to_page(ti));
 227
 228         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 229 }
 230
 231 void free_task(struct task_struct *tsk)
 232 {
 233         account_kernel_stack(tsk->stack, -1);
 234         arch_release_thread_info(tsk->stack);
 235         free_thread_info(tsk->stack);
 236         rt_mutex_debug_task_free(tsk);
 237         ftrace_graph_exit_task(tsk);
 238         put_seccomp_filter(tsk);
 239         arch_release_task_struct(tsk);
 240         free_task_struct(tsk);
 241 }
 242 EXPORT_SYMBOL(free_task);
 243
 244 static inline void free_signal_struct(struct signal_struct *sig)
 245 {
 246         taskstats_tgid_free(sig);
 247         sched_autogroup_exit(sig);
 248         kmem_cache_free(signal_cachep, sig);
 249 }
 250
 251 static inline void put_signal_struct(struct signal_struct *sig)
 252 {
 253         if (atomic_dec_and_test(&sig->sigcnt))
 254                 free_signal_struct(sig);
 255 }
 256
 257 void __put_task_struct(struct task_struct *tsk)
 258 {
 259         WARN_ON(!tsk->exit_state);
 260         WARN_ON(atomic_read(&tsk->usage));
 261         WARN_ON(tsk == current);
 262
 263         cgroup_free(tsk);
 264         task_numa_free(tsk);
 265         security_task_free(tsk);
 266         exit_creds(tsk);
 267         delayacct_tsk_free(tsk);
 268         put_signal_struct(tsk->signal);
 269
 270         if (!profile_handoff_task(tsk))
 271                 free_task(tsk);
 272 }
 273 EXPORT_SYMBOL_GPL(__put_task_struct);
 274
 275 void __init __weak arch_task_cache_init(void) { }
 276
 277 /*
 278  * set_max_threads
 279  */
 280 static void set_max_threads(unsigned int max_threads_suggested)
 281 {
 282         u64 threads;
 283
 284         /*
 285          * The number of threads shall be limited such that the thread
 286          * structures may only consume a small part of the available memory.
 287          */
 288         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 289                 threads = MAX_THREADS;
 290         else
 291                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 292                                     (u64) THREAD_SIZE * 8UL);
 293
 294         if (threads > max_threads_suggested)
 295                 threads = max_threads_suggested;
 296
 297         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 298 }
 299
 300 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 301 /* Initialized by the architecture: */
 302 int arch_task_struct_size __read_mostly;
 303 #endif
 304
 305 void __init fork_init(void)
 306 {
 307 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 308 #ifndef ARCH_MIN_TASKALIGN
 309 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 310 #endif
 311         /* create a slab on which task_structs can be allocated */
 312         task_struct_cachep = kmem_cache_create("task_struct",
 313                         arch_task_struct_size, ARCH_MIN_TASKALIGN,
 314                         SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
 315 #endif
 316
 317         /* do the arch specific task caches init */
 318         arch_task_cache_init();
 319
 320         set_max_threads(MAX_THREADS);
 321
 322         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 323         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 324         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 325                 init_task.signal->rlim[RLIMIT_NPROC];
 326 }
 327
 328 int __weak arch_dup_task_struct(struct task_struct *dst,
 329                                                struct task_struct *src)
 330 {
 331         *dst = *src;
 332         return 0;
 333 }
 334
 335 void set_task_stack_end_magic(struct task_struct *tsk)
 336 {
 337         unsigned long *stackend;
 338
 339         stackend = end_of_stack(tsk);
 340         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 341 }
 342
 343 static struct task_struct *dup_task_struct(struct task_struct *orig)
 344 {
 345         struct task_struct *tsk;
 346         struct thread_info *ti;
 347         int node = tsk_fork_get_node(orig);
 348         int err;
 349
 350         tsk = alloc_task_struct_node(node);
 351         if (!tsk)
 352                 return NULL;
 353
 354         ti = alloc_thread_info_node(tsk, node);
 355         if (!ti)
 356                 goto free_tsk;
 357
 358         err = arch_dup_task_struct(tsk, orig);
 359         if (err)
 360                 goto free_ti;
 361
 362         tsk->stack = ti;
 363 #ifdef CONFIG_SECCOMP
 364         /*
 365          * We must handle setting up seccomp filters once we're under
 366          * the sighand lock in case orig has changed between now and
 367          * then. Until then, filter must be NULL to avoid messing up
 368          * the usage counts on the error path calling free_task.
 369          */
 370         tsk->seccomp.filter = NULL;
 371 #endif
 372
 373         setup_thread_stack(tsk, orig);
 374         clear_user_return_notifier(tsk);
 375         clear_tsk_need_resched(tsk);
 376         set_task_stack_end_magic(tsk);
 377
 378 #ifdef CONFIG_CC_STACKPROTECTOR
 379         tsk->stack_canary = get_random_int();
 380 #endif
 381
 382         /*
 383          * One for us, one for whoever does the "release_task()" (usually
 384          * parent)
 385          */
 386         atomic_set(&tsk->usage, 2);
 387 #ifdef CONFIG_BLK_DEV_IO_TRACE
 388         tsk->btrace_seq = 0;
 389 #endif
 390         tsk->splice_pipe = NULL;
 391         tsk->task_frag.page = NULL;
 392         tsk->wake_q.next = NULL;
 393
 394         account_kernel_stack(ti, 1);
 395
 396         kcov_task_init(tsk);
 397
 398         return tsk;
 399
 400 free_ti:
 401         free_thread_info(ti);
 402 free_tsk:
 403         free_task_struct(tsk);
 404         return NULL;
 405 }
 406
 407 #ifdef CONFIG_MMU
 408 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 409 {
 410         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 411         struct rb_node **rb_link, *rb_parent;
 412         int retval;
 413         unsigned long charge;
 414
 415         uprobe_start_dup_mmap();
 416         down_write(&oldmm->mmap_sem);
 417         flush_cache_dup_mm(oldmm);
 418         uprobe_dup_mmap(oldmm, mm);
 419         /*
 420          * Not linked in yet - no deadlock potential:
 421          */
 422         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 423
 424         /* No ordering required: file already has been exposed. */
 425         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 426
 427         mm->total_vm = oldmm->total_vm;
 428         mm->data_vm = oldmm->data_vm;
 429         mm->exec_vm = oldmm->exec_vm;
 430         mm->stack_vm = oldmm->stack_vm;
 431
 432         rb_link = &mm->mm_rb.rb_node;
 433         rb_parent = NULL;
 434         pprev = &mm->mmap;
 435         retval = ksm_fork(mm, oldmm);
 436         if (retval)
 437                 goto out;
 438         retval = khugepaged_fork(mm, oldmm);
 439         if (retval)
 440                 goto out;
 441
 442         prev = NULL;
 443         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 444                 struct file *file;
 445
 446                 if (mpnt->vm_flags & VM_DONTCOPY) {
 447                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 448                         continue;
 449                 }
 450                 charge = 0;
 451                 if (mpnt->vm_flags & VM_ACCOUNT) {
 452                         unsigned long len = vma_pages(mpnt);
 453
 454                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 455                                 goto fail_nomem;
 456                         charge = len;
 457                 }
 458                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 459                 if (!tmp)
 460                         goto fail_nomem;
 461                 *tmp = *mpnt;
 462                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 463                 retval = vma_dup_policy(mpnt, tmp);
 464                 if (retval)
 465                         goto fail_nomem_policy;
 466                 tmp->vm_mm = mm;
 467                 if (anon_vma_fork(tmp, mpnt))
 468                         goto fail_nomem_anon_vma_fork;
 469                 tmp->vm_flags &=
 470                         ~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
 471                 tmp->vm_next = tmp->vm_prev = NULL;
 472                 tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
 473                 file = tmp->vm_file;
 474                 if (file) {
 475                         struct inode *inode = file_inode(file);
 476                         struct address_space *mapping = file->f_mapping;
 477
 478                         get_file(file);
 479                         if (tmp->vm_flags & VM_DENYWRITE)
 480                                 atomic_dec(&inode->i_writecount);
 481                         i_mmap_lock_write(mapping);
 482                         if (tmp->vm_flags & VM_SHARED)
 483                                 atomic_inc(&mapping->i_mmap_writable);
 484                         flush_dcache_mmap_lock(mapping);
 485                         /* insert tmp into the share list, just after mpnt */
 486                         vma_interval_tree_insert_after(tmp, mpnt,
 487                                         &mapping->i_mmap);
 488                         flush_dcache_mmap_unlock(mapping);
 489                         i_mmap_unlock_write(mapping);
 490                 }
 491
 492                 /*
 493                  * Clear hugetlb-related page reserves for children. This only
 494                  * affects MAP_PRIVATE mappings. Faults generated by the child
 495                  * are not guaranteed to succeed, even if read-only
 496                  */
 497                 if (is_vm_hugetlb_page(tmp))
 498                         reset_vma_resv_huge_pages(tmp);
 499
 500                 /*
 501                  * Link in the new vma and copy the page table entries.
 502                  */
 503                 *pprev = tmp;
 504                 pprev = &tmp->vm_next;
 505                 tmp->vm_prev = prev;
 506                 prev = tmp;
 507
 508                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 509                 rb_link = &tmp->vm_rb.rb_right;
 510                 rb_parent = &tmp->vm_rb;
 511
 512                 mm->map_count++;
 513                 retval = copy_page_range(mm, oldmm, mpnt);
 514
 515                 if (tmp->vm_ops && tmp->vm_ops->open)
 516                         tmp->vm_ops->open(tmp);
 517
 518                 if (retval)
 519                         goto out;
 520         }
 521         /* a new mm has just been created */
 522         arch_dup_mmap(oldmm, mm);
 523         retval = 0;
 524 out:
 525         up_write(&mm->mmap_sem);
 526         flush_tlb_mm(oldmm);
 527         up_write(&oldmm->mmap_sem);
 528         uprobe_end_dup_mmap();
 529         return retval;
 530 fail_nomem_anon_vma_fork:
 531         mpol_put(vma_policy(tmp));
 532 fail_nomem_policy:
 533         kmem_cache_free(vm_area_cachep, tmp);
 534 fail_nomem:
 535         retval = -ENOMEM;
 536         vm_unacct_memory(charge);
 537         goto out;
 538 }
 539
 540 static inline int mm_alloc_pgd(struct mm_struct *mm)
 541 {
 542         mm->pgd = pgd_alloc(mm);
 543         if (unlikely(!mm->pgd))
 544                 return -ENOMEM;
 545         return 0;
 546 }
 547
 548 static inline void mm_free_pgd(struct mm_struct *mm)
 549 {
 550         pgd_free(mm, mm->pgd);
 551 }
 552 #else
 553 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 554 {
 555         down_write(&oldmm->mmap_sem);
 556         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 557         up_write(&oldmm->mmap_sem);
 558         return 0;
 559 }
 560 #define mm_alloc_pgd(mm)        (0)
 561 #define mm_free_pgd(mm)
 562 #endif /* CONFIG_MMU */
 563
 564 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 565
 566 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 567 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 568
 569 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 570
 571 static int __init coredump_filter_setup(char *s)
 572 {
 573         default_dump_filter =
 574                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 575                 MMF_DUMP_FILTER_MASK;
 576         return 1;
 577 }
 578
 579 __setup("coredump_filter=", coredump_filter_setup);
 580
 581 #include <linux/init_task.h>
 582
 583 static void mm_init_aio(struct mm_struct *mm)
 584 {
 585 #ifdef CONFIG_AIO
 586         spin_lock_init(&mm->ioctx_lock);
 587         mm->ioctx_table = NULL;
 588 #endif
 589 }
 590
 591 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 592 {
 593 #ifdef CONFIG_MEMCG
 594         mm->owner = p;
 595 #endif
 596 }
 597
 598 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 599 {
 600         mm->mmap = NULL;
 601         mm->mm_rb = RB_ROOT;
 602         mm->vmacache_seqnum = 0;
 603         atomic_set(&mm->mm_users, 1);
 604         atomic_set(&mm->mm_count, 1);
 605         init_rwsem(&mm->mmap_sem);
 606         INIT_LIST_HEAD(&mm->mmlist);
 607         mm->core_state = NULL;
 608         atomic_long_set(&mm->nr_ptes, 0);
 609         mm_nr_pmds_init(mm);
 610         mm->map_count = 0;
 611         mm->locked_vm = 0;
 612         mm->pinned_vm = 0;
 613         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 614         spin_lock_init(&mm->page_table_lock);
 615         mm_init_cpumask(mm);
 616         mm_init_aio(mm);
 617         mm_init_owner(mm, p);
 618         mmu_notifier_mm_init(mm);
 619         clear_tlb_flush_pending(mm);
 620 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 621         mm->pmd_huge_pte = NULL;
 622 #endif
 623
 624         if (current->mm) {
 625                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 626                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 627         } else {
 628                 mm->flags = default_dump_filter;
 629                 mm->def_flags = 0;
 630         }
 631
 632         if (mm_alloc_pgd(mm))
 633                 goto fail_nopgd;
 634
 635         if (init_new_context(p, mm))
 636                 goto fail_nocontext;
 637
 638         return mm;
 639
 640 fail_nocontext:
 641         mm_free_pgd(mm);
 642 fail_nopgd:
 643         free_mm(mm);
 644         return NULL;
 645 }
 646
 647 static void check_mm(struct mm_struct *mm)
 648 {
 649         int i;
 650
 651         for (i = 0; i < NR_MM_COUNTERS; i++) {
 652                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 653
 654                 if (unlikely(x))
 655                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 656                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 657         }
 658
 659         if (atomic_long_read(&mm->nr_ptes))
 660                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 661                                 atomic_long_read(&mm->nr_ptes));
 662         if (mm_nr_pmds(mm))
 663                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 664                                 mm_nr_pmds(mm));
 665
 666 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 667         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 668 #endif
 669 }
 670
 671 /*
 672  * Allocate and initialize an mm_struct.
 673  */
 674 struct mm_struct *mm_alloc(void)
 675 {
 676         struct mm_struct *mm;
 677
 678         mm = allocate_mm();
 679         if (!mm)
 680                 return NULL;
 681
 682         memset(mm, 0, sizeof(*mm));
 683         return mm_init(mm, current);
 684 }
 685
 686 /*
 687  * Called when the last reference to the mm
 688  * is dropped: either by a lazy thread or by
 689  * mmput. Free the page directory and the mm.
 690  */
 691 void __mmdrop(struct mm_struct *mm)
 692 {
 693         BUG_ON(mm == &init_mm);
 694         mm_free_pgd(mm);
 695         destroy_context(mm);
 696         mmu_notifier_mm_destroy(mm);
 697         check_mm(mm);
 698         free_mm(mm);
 699 }
 700 EXPORT_SYMBOL_GPL(__mmdrop);
 701
 702 static inline void __mmput(struct mm_struct *mm)
 703 {
 704         VM_BUG_ON(atomic_read(&mm->mm_users));
 705
 706         uprobe_clear_state(mm);
 707         exit_aio(mm);
 708         ksm_exit(mm);
 709         khugepaged_exit(mm); /* must run before exit_mmap */
 710         exit_mmap(mm);
 711         set_mm_exe_file(mm, NULL);
 712         if (!list_empty(&mm->mmlist)) {
 713                 spin_lock(&mmlist_lock);
 714                 list_del(&mm->mmlist);
 715                 spin_unlock(&mmlist_lock);
 716         }
 717         if (mm->binfmt)
 718                 module_put(mm->binfmt->module);
 719         mmdrop(mm);
 720 }
 721
 722 /*
 723  * Decrement the use count and release all resources for an mm.
 724  */
 725 void mmput(struct mm_struct *mm)
 726 {
 727         might_sleep();
 728
 729         if (atomic_dec_and_test(&mm->mm_users))
 730                 __mmput(mm);
 731 }
 732 EXPORT_SYMBOL_GPL(mmput);
 733
 734 static void mmput_async_fn(struct work_struct *work)
 735 {
 736         struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
 737         __mmput(mm);
 738 }
 739
 740 void mmput_async(struct mm_struct *mm)
 741 {
 742         if (atomic_dec_and_test(&mm->mm_users)) {
 743                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
 744                 schedule_work(&mm->async_put_work);
 745         }
 746 }
 747
 748 /**
 749  * set_mm_exe_file - change a reference to the mm's executable file
 750  *
 751  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 752  *
 753  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 754  * invocations: in mmput() nobody alive left, in execve task is single
 755  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 756  * mm->exe_file, but does so without using set_mm_exe_file() in order
 757  * to do avoid the need for any locks.
 758  */
 759 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 760 {
 761         struct file *old_exe_file;
 762
 763         /*
 764          * It is safe to dereference the exe_file without RCU as
 765          * this function is only called if nobody else can access
 766          * this mm -- see comment above for justification.
 767          */
 768         old_exe_file = rcu_dereference_raw(mm->exe_file);
 769
 770         if (new_exe_file)
 771                 get_file(new_exe_file);
 772         rcu_assign_pointer(mm->exe_file, new_exe_file);
 773         if (old_exe_file)
 774                 fput(old_exe_file);
 775 }
 776
 777 /**
 778  * get_mm_exe_file - acquire a reference to the mm's executable file
 779  *
 780  * Returns %NULL if mm has no associated executable file.
 781  * User must release file via fput().
 782  */
 783 struct file *get_mm_exe_file(struct mm_struct *mm)
 784 {
 785         struct file *exe_file;
 786
 787         rcu_read_lock();
 788         exe_file = rcu_dereference(mm->exe_file);
 789         if (exe_file && !get_file_rcu(exe_file))
 790                 exe_file = NULL;
 791         rcu_read_unlock();
 792         return exe_file;
 793 }
 794 EXPORT_SYMBOL(get_mm_exe_file);
 795
 796 /**
 797  * get_task_mm - acquire a reference to the task's mm
 798  *
 799  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 800  * this kernel workthread has transiently adopted a user mm with use_mm,
 801  * to do its AIO) is not set and if so returns a reference to it, after
 802  * bumping up the use count.  User must release the mm via mmput()
 803  * after use.  Typically used by /proc and ptrace.
 804  */
 805 struct mm_struct *get_task_mm(struct task_struct *task)
 806 {
 807         struct mm_struct *mm;
 808
 809         task_lock(task);
 810         mm = task->mm;
 811         if (mm) {
 812                 if (task->flags & PF_KTHREAD)
 813                         mm = NULL;
 814                 else
 815                         atomic_inc(&mm->mm_users);
 816         }
 817         task_unlock(task);
 818         return mm;
 819 }
 820 EXPORT_SYMBOL_GPL(get_task_mm);
 821
 822 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 823 {
 824         struct mm_struct *mm;
 825         int err;
 826
 827         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 828         if (err)
 829                 return ERR_PTR(err);
 830
 831         mm = get_task_mm(task);
 832         if (mm && mm != current->mm &&
 833                         !ptrace_may_access(task, mode)) {
 834                 mmput(mm);
 835                 mm = ERR_PTR(-EACCES);
 836         }
 837         mutex_unlock(&task->signal->cred_guard_mutex);
 838
 839         return mm;
 840 }
 841
 842 static void complete_vfork_done(struct task_struct *tsk)
 843 {
 844         struct completion *vfork;
 845
 846         task_lock(tsk);
 847         vfork = tsk->vfork_done;
 848         if (likely(vfork)) {
 849                 tsk->vfork_done = NULL;
 850                 complete(vfork);
 851         }
 852         task_unlock(tsk);
 853 }
 854
 855 static int wait_for_vfork_done(struct task_struct *child,
 856                                 struct completion *vfork)
 857 {
 858         int killed;
 859
 860         freezer_do_not_count();
 861         killed = wait_for_completion_killable(vfork);
 862         freezer_count();
 863
 864         if (killed) {
 865                 task_lock(child);
 866                 child->vfork_done = NULL;
 867                 task_unlock(child);
 868         }
 869
 870         put_task_struct(child);
 871         return killed;
 872 }
 873
 874 /* Please note the differences between mmput and mm_release.
 875  * mmput is called whenever we stop holding onto a mm_struct,
 876  * error success whatever.
 877  *
 878  * mm_release is called after a mm_struct has been removed
 879  * from the current process.
 880  *
 881  * This difference is important for error handling, when we
 882  * only half set up a mm_struct for a new process and need to restore
 883  * the old one.  Because we mmput the new mm_struct before
 884  * restoring the old one. . .
 885  * Eric Biederman 10 January 1998
 886  */
 887 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 888 {
 889         /* Get rid of any futexes when releasing the mm */
 890 #ifdef CONFIG_FUTEX
 891         if (unlikely(tsk->robust_list)) {
 892                 exit_robust_list(tsk);
 893                 tsk->robust_list = NULL;
 894         }
 895 #ifdef CONFIG_COMPAT
 896         if (unlikely(tsk->compat_robust_list)) {
 897                 compat_exit_robust_list(tsk);
 898                 tsk->compat_robust_list = NULL;
 899         }
 900 #endif
 901         if (unlikely(!list_empty(&tsk->pi_state_list)))
 902                 exit_pi_state_list(tsk);
 903 #endif
 904
 905         uprobe_free_utask(tsk);
 906
 907         /* Get rid of any cached register state */
 908         deactivate_mm(tsk, mm);
 909
 910         /*
 911          * If we're exiting normally, clear a user-space tid field if
 912          * requested.  We leave this alone when dying by signal, to leave
 913          * the value intact in a core dump, and to save the unnecessary
 914          * trouble, say, a killed vfork parent shouldn't touch this mm.
 915          * Userland only wants this done for a sys_exit.
 916          */
 917         if (tsk->clear_child_tid) {
 918                 if (!(tsk->flags & PF_SIGNALED) &&
 919                     atomic_read(&mm->mm_users) > 1) {
 920                         /*
 921                          * We don't check the error code - if userspace has
 922                          * not set up a proper pointer then tough luck.
 923                          */
 924                         put_user(0, tsk->clear_child_tid);
 925                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 926                                         1, NULL, NULL, 0);
 927                 }
 928                 tsk->clear_child_tid = NULL;
 929         }
 930
 931         /*
 932          * All done, finally we can wake up parent and return this mm to him.
 933          * Also kthread_stop() uses this completion for synchronization.
 934          */
 935         if (tsk->vfork_done)
 936                 complete_vfork_done(tsk);
 937 }
 938
 939 /*
 940  * Allocate a new mm structure and copy contents from the
 941  * mm structure of the passed in task structure.
 942  */
 943 static struct mm_struct *dup_mm(struct task_struct *tsk)
 944 {
 945         struct mm_struct *mm, *oldmm = current->mm;
 946         int err;
 947
 948         mm = allocate_mm();
 949         if (!mm)
 950                 goto fail_nomem;
 951
 952         memcpy(mm, oldmm, sizeof(*mm));
 953
 954         if (!mm_init(mm, tsk))
 955                 goto fail_nomem;
 956
 957         err = dup_mmap(mm, oldmm);
 958         if (err)
 959                 goto free_pt;
 960
 961         mm->hiwater_rss = get_mm_rss(mm);
 962         mm->hiwater_vm = mm->total_vm;
 963
 964         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 965                 goto free_pt;
 966
 967         return mm;
 968
 969 free_pt:
 970         /* don't put binfmt in mmput, we haven't got module yet */
 971         mm->binfmt = NULL;
 972         mmput(mm);
 973
 974 fail_nomem:
 975         return NULL;
 976 }
 977
 978 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 979 {
 980         struct mm_struct *mm, *oldmm;
 981         int retval;
 982
 983         tsk->min_flt = tsk->maj_flt = 0;
 984         tsk->nvcsw = tsk->nivcsw = 0;
 985 #ifdef CONFIG_DETECT_HUNG_TASK
 986         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 987 #endif
 988
 989         tsk->mm = NULL;
 990         tsk->active_mm = NULL;
 991
 992         /*
 993          * Are we cloning a kernel thread?
 994          *
 995          * We need to steal a active VM for that..
 996          */
 997         oldmm = current->mm;
 998         if (!oldmm)
 999                 return 0;
1000
1001         /* initialize the new vmacache entries */
1002         vmacache_flush(tsk);
1003
1004         if (clone_flags & CLONE_VM) {
1005                 atomic_inc(&oldmm->mm_users);
1006                 mm = oldmm;
1007                 goto good_mm;
1008         }
1009
1010         retval = -ENOMEM;
1011         mm = dup_mm(tsk);
1012         if (!mm)
1013                 goto fail_nomem;
1014
1015 good_mm:
1016         tsk->mm = mm;
1017         tsk->active_mm = mm;
1018         return 0;
1019
1020 fail_nomem:
1021         return retval;
1022 }
1023
1024 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1025 {
1026         struct fs_struct *fs = current->fs;
1027         if (clone_flags & CLONE_FS) {
1028                 /* tsk->fs is already what we want */
1029                 spin_lock(&fs->lock);
1030                 if (fs->in_exec) {
1031                         spin_unlock(&fs->lock);
1032                         return -EAGAIN;
1033                 }
1034                 fs->users++;
1035                 spin_unlock(&fs->lock);
1036                 return 0;
1037         }
1038         tsk->fs = copy_fs_struct(fs);
1039         if (!tsk->fs)
1040                 return -ENOMEM;
1041         return 0;
1042 }
1043
1044 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1045 {
1046         struct files_struct *oldf, *newf;
1047         int error = 0;
1048
1049         /*
1050          * A background process may not have any files ...
1051          */
1052         oldf = current->files;
1053         if (!oldf)
1054                 goto out;
1055
1056         if (clone_flags & CLONE_FILES) {
1057                 atomic_inc(&oldf->count);
1058                 goto out;
1059         }
1060
1061         newf = dup_fd(oldf, &error);
1062         if (!newf)
1063                 goto out;
1064
1065         tsk->files = newf;
1066         error = 0;
1067 out:
1068         return error;
1069 }
1070
1071 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1072 {
1073 #ifdef CONFIG_BLOCK
1074         struct io_context *ioc = current->io_context;
1075         struct io_context *new_ioc;
1076
1077         if (!ioc)
1078                 return 0;
1079         /*
1080          * Share io context with parent, if CLONE_IO is set
1081          */
1082         if (clone_flags & CLONE_IO) {
1083                 ioc_task_link(ioc);
1084                 tsk->io_context = ioc;
1085         } else if (ioprio_valid(ioc->ioprio)) {
1086                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1087                 if (unlikely(!new_ioc))
1088                         return -ENOMEM;
1089
1090                 new_ioc->ioprio = ioc->ioprio;
1091                 put_io_context(new_ioc);
1092         }
1093 #endif
1094         return 0;
1095 }
1096
1097 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1098 {
1099         struct sighand_struct *sig;
1100
1101         if (clone_flags & CLONE_SIGHAND) {
1102                 atomic_inc(&current->sighand->count);
1103                 return 0;
1104         }
1105         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1106         rcu_assign_pointer(tsk->sighand, sig);
1107         if (!sig)
1108                 return -ENOMEM;
1109
1110         atomic_set(&sig->count, 1);
1111         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1112         return 0;
1113 }
1114
1115 void __cleanup_sighand(struct sighand_struct *sighand)
1116 {
1117         if (atomic_dec_and_test(&sighand->count)) {
1118                 signalfd_cleanup(sighand);
1119                 /*
1120                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1121                  * without an RCU grace period, see __lock_task_sighand().
1122                  */
1123                 kmem_cache_free(sighand_cachep, sighand);
1124         }
1125 }
1126
1127 /*
1128  * Initialize POSIX timer handling for a thread group.
1129  */
1130 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1131 {
1132         unsigned long cpu_limit;
1133
1134         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1135         if (cpu_limit != RLIM_INFINITY) {
1136                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1137                 sig->cputimer.running = true;
1138         }
1139
1140         /* The timer lists. */
1141         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1142         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1143         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1144 }
1145
1146 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1147 {
1148         struct signal_struct *sig;
1149
1150         if (clone_flags & CLONE_THREAD)
1151                 return 0;
1152
1153         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1154         tsk->signal = sig;
1155         if (!sig)
1156                 return -ENOMEM;
1157
1158         sig->nr_threads = 1;
1159         atomic_set(&sig->live, 1);
1160         atomic_set(&sig->sigcnt, 1);
1161
1162         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1163         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1164         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1165
1166         init_waitqueue_head(&sig->wait_chldexit);
1167         sig->curr_target = tsk;
1168         init_sigpending(&sig->shared_pending);
1169         INIT_LIST_HEAD(&sig->posix_timers);
1170         seqlock_init(&sig->stats_lock);
1171         prev_cputime_init(&sig->prev_cputime);
1172
1173         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1174         sig->real_timer.function = it_real_fn;
1175
1176         task_lock(current->group_leader);
1177         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1178         task_unlock(current->group_leader);
1179
1180         posix_cpu_timers_init_group(sig);
1181
1182         tty_audit_fork(sig);
1183         sched_autogroup_fork(sig);
1184
1185         sig->oom_score_adj = current->signal->oom_score_adj;
1186         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1187
1188         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1189                                    current->signal->is_child_subreaper;
1190
1191         mutex_init(&sig->cred_guard_mutex);
1192
1193         return 0;
1194 }
1195
1196 static void copy_seccomp(struct task_struct *p)
1197 {
1198 #ifdef CONFIG_SECCOMP
1199         /*
1200          * Must be called with sighand->lock held, which is common to
1201          * all threads in the group. Holding cred_guard_mutex is not
1202          * needed because this new task is not yet running and cannot
1203          * be racing exec.
1204          */
1205         assert_spin_locked(&current->sighand->siglock);
1206
1207         /* Ref-count the new filter user, and assign it. */
1208         get_seccomp_filter(current);
1209         p->seccomp = current->seccomp;
1210
1211         /*
1212          * Explicitly enable no_new_privs here in case it got set
1213          * between the task_struct being duplicated and holding the
1214          * sighand lock. The seccomp state and nnp must be in sync.
1215          */
1216         if (task_no_new_privs(current))
1217                 task_set_no_new_privs(p);
1218
1219         /*
1220          * If the parent gained a seccomp mode after copying thread
1221          * flags and between before we held the sighand lock, we have
1222          * to manually enable the seccomp thread flag here.
1223          */
1224         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1225                 set_tsk_thread_flag(p, TIF_SECCOMP);
1226 #endif
1227 }
1228
1229 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1230 {
1231         current->clear_child_tid = tidptr;
1232
1233         return task_pid_vnr(current);
1234 }
1235
1236 static void rt_mutex_init_task(struct task_struct *p)
1237 {
1238         raw_spin_lock_init(&p->pi_lock);
1239 #ifdef CONFIG_RT_MUTEXES
1240         p->pi_waiters = RB_ROOT;
1241         p->pi_waiters_leftmost = NULL;
1242         p->pi_blocked_on = NULL;
1243 #endif
1244 }
1245
1246 /*
1247  * Initialize POSIX timer handling for a single task.
1248  */
1249 static void posix_cpu_timers_init(struct task_struct *tsk)
1250 {
1251         tsk->cputime_expires.prof_exp = 0;
1252         tsk->cputime_expires.virt_exp = 0;
1253         tsk->cputime_expires.sched_exp = 0;
1254         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1255         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1256         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1257 }
1258
1259 static inline void
1260 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1261 {
1262          task->pids[type].pid = pid;
1263 }
1264
1265 /*
1266  * This creates a new process as a copy of the old one,
1267  * but does not actually start it yet.
1268  *
1269  * It copies the registers, and all the appropriate
1270  * parts of the process environment (as per the clone
1271  * flags). The actual kick-off is left to the caller.
1272  */
1273 static struct task_struct *copy_process(unsigned long clone_flags,
1274                                         unsigned long stack_start,
1275                                         unsigned long stack_size,
1276                                         int __user *child_tidptr,
1277                                         struct pid *pid,
1278                                         int trace,
1279                                         unsigned long tls)
1280 {
1281         int retval;
1282         struct task_struct *p;
1283
1284         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1285                 return ERR_PTR(-EINVAL);
1286
1287         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1288                 return ERR_PTR(-EINVAL);
1289
1290         /*
1291          * Thread groups must share signals as well, and detached threads
1292          * can only be started up within the thread group.
1293          */
1294         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1295                 return ERR_PTR(-EINVAL);
1296
1297         /*
1298          * Shared signal handlers imply shared VM. By way of the above,
1299          * thread groups also imply shared VM. Blocking this case allows
1300          * for various simplifications in other code.
1301          */
1302         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1303                 return ERR_PTR(-EINVAL);
1304
1305         /*
1306          * Siblings of global init remain as zombies on exit since they are
1307          * not reaped by their parent (swapper). To solve this and to avoid
1308          * multi-rooted process trees, prevent global and container-inits
1309          * from creating siblings.
1310          */
1311         if ((clone_flags & CLONE_PARENT) &&
1312                                 current->signal->flags & SIGNAL_UNKILLABLE)
1313                 return ERR_PTR(-EINVAL);
1314
1315         /*
1316          * If the new process will be in a different pid or user namespace
1317          * do not allow it to share a thread group with the forking task.
1318          */
1319         if (clone_flags & CLONE_THREAD) {
1320                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1321                     (task_active_pid_ns(current) !=
1322                                 current->nsproxy->pid_ns_for_children))
1323                         return ERR_PTR(-EINVAL);
1324         }
1325
1326         retval = security_task_create(clone_flags);
1327         if (retval)
1328                 goto fork_out;
1329
1330         retval = -ENOMEM;
1331         p = dup_task_struct(current);
1332         if (!p)
1333                 goto fork_out;
1334
1335         ftrace_graph_init_task(p);
1336
1337         rt_mutex_init_task(p);
1338
1339 #ifdef CONFIG_PROVE_LOCKING
1340         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1341         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1342 #endif
1343         retval = -EAGAIN;
1344         if (atomic_read(&p->real_cred->user->processes) >=
1345                         task_rlimit(p, RLIMIT_NPROC)) {
1346                 if (p->real_cred->user != INIT_USER &&
1347                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1348                         goto bad_fork_free;
1349         }
1350         current->flags &= ~PF_NPROC_EXCEEDED;
1351
1352         retval = copy_creds(p, clone_flags);
1353         if (retval < 0)
1354                 goto bad_fork_free;
1355
1356         /*
1357          * If multiple threads are within copy_process(), then this check
1358          * triggers too late. This doesn't hurt, the check is only there
1359          * to stop root fork bombs.
1360          */
1361         retval = -EAGAIN;
1362         if (nr_threads >= max_threads)
1363                 goto bad_fork_cleanup_count;
1364
1365         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1366         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1367         p->flags |= PF_FORKNOEXEC;
1368         INIT_LIST_HEAD(&p->children);
1369         INIT_LIST_HEAD(&p->sibling);
1370         rcu_copy_process(p);
1371         p->vfork_done = NULL;
1372         spin_lock_init(&p->alloc_lock);
1373
1374         init_sigpending(&p->pending);
1375
1376         p->utime = p->stime = p->gtime = 0;
1377         p->utimescaled = p->stimescaled = 0;
1378         prev_cputime_init(&p->prev_cputime);
1379
1380 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1381         seqcount_init(&p->vtime_seqcount);
1382         p->vtime_snap = 0;
1383         p->vtime_snap_whence = VTIME_INACTIVE;
1384 #endif
1385
1386 #if defined(SPLIT_RSS_COUNTING)
1387         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1388 #endif
1389
1390         p->default_timer_slack_ns = current->timer_slack_ns;
1391
1392         task_io_accounting_init(&p->ioac);
1393         acct_clear_integrals(p);
1394
1395         posix_cpu_timers_init(p);
1396
1397         p->start_time = ktime_get_ns();
1398         p->real_start_time = ktime_get_boot_ns();
1399         p->io_context = NULL;
1400         p->audit_context = NULL;
1401         threadgroup_change_begin(current);
1402         cgroup_fork(p);
1403 #ifdef CONFIG_NUMA
1404         p->mempolicy = mpol_dup(p->mempolicy);
1405         if (IS_ERR(p->mempolicy)) {
1406                 retval = PTR_ERR(p->mempolicy);
1407                 p->mempolicy = NULL;
1408                 goto bad_fork_cleanup_threadgroup_lock;
1409         }
1410 #endif
1411 #ifdef CONFIG_CPUSETS
1412         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1413         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1414         seqcount_init(&p->mems_allowed_seq);
1415 #endif
1416 #ifdef CONFIG_TRACE_IRQFLAGS
1417         p->irq_events = 0;
1418         p->hardirqs_enabled = 0;
1419         p->hardirq_enable_ip = 0;
1420         p->hardirq_enable_event = 0;
1421         p->hardirq_disable_ip = _THIS_IP_;
1422         p->hardirq_disable_event = 0;
1423         p->softirqs_enabled = 1;
1424         p->softirq_enable_ip = _THIS_IP_;
1425         p->softirq_enable_event = 0;
1426         p->softirq_disable_ip = 0;
1427         p->softirq_disable_event = 0;
1428         p->hardirq_context = 0;
1429         p->softirq_context = 0;
1430 #endif
1431
1432         p->pagefault_disabled = 0;
1433
1434 #ifdef CONFIG_LOCKDEP
1435         p->lockdep_depth = 0; /* no locks held yet */
1436         p->curr_chain_key = 0;
1437         p->lockdep_recursion = 0;
1438 #endif
1439
1440 #ifdef CONFIG_DEBUG_MUTEXES
1441         p->blocked_on = NULL; /* not blocked yet */
1442 #endif
1443 #ifdef CONFIG_BCACHE
1444         p->sequential_io        = 0;
1445         p->sequential_io_avg    = 0;
1446 #endif
1447
1448         /* Perform scheduler related setup. Assign this task to a CPU. */
1449         retval = sched_fork(clone_flags, p);
1450         if (retval)
1451                 goto bad_fork_cleanup_policy;
1452
1453         retval = perf_event_init_task(p);
1454         if (retval)
1455                 goto bad_fork_cleanup_policy;
1456         retval = audit_alloc(p);
1457         if (retval)
1458                 goto bad_fork_cleanup_perf;
1459         /* copy all the process information */
1460         shm_init_task(p);
1461         retval = copy_semundo(clone_flags, p);
1462         if (retval)
1463                 goto bad_fork_cleanup_audit;
1464         retval = copy_files(clone_flags, p);
1465         if (retval)
1466                 goto bad_fork_cleanup_semundo;
1467         retval = copy_fs(clone_flags, p);
1468         if (retval)
1469                 goto bad_fork_cleanup_files;
1470         retval = copy_sighand(clone_flags, p);
1471         if (retval)
1472                 goto bad_fork_cleanup_fs;
1473         retval = copy_signal(clone_flags, p);
1474         if (retval)
1475                 goto bad_fork_cleanup_sighand;
1476         retval = copy_mm(clone_flags, p);
1477         if (retval)
1478                 goto bad_fork_cleanup_signal;
1479         retval = copy_namespaces(clone_flags, p);
1480         if (retval)
1481                 goto bad_fork_cleanup_mm;
1482         retval = copy_io(clone_flags, p);
1483         if (retval)
1484                 goto bad_fork_cleanup_namespaces;
1485         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1486         if (retval)
1487                 goto bad_fork_cleanup_io;
1488
1489         if (pid != &init_struct_pid) {
1490                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1491                 if (IS_ERR(pid)) {
1492                         retval = PTR_ERR(pid);
1493                         goto bad_fork_cleanup_thread;
1494                 }
1495         }
1496
1497         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1498         /*
1499          * Clear TID on mm_release()?
1500          */
1501         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1502 #ifdef CONFIG_BLOCK
1503         p->plug = NULL;
1504 #endif
1505 #ifdef CONFIG_FUTEX
1506         p->robust_list = NULL;
1507 #ifdef CONFIG_COMPAT
1508         p->compat_robust_list = NULL;
1509 #endif
1510         INIT_LIST_HEAD(&p->pi_state_list);
1511         p->pi_state_cache = NULL;
1512 #endif
1513         /*
1514          * sigaltstack should be cleared when sharing the same VM
1515          */
1516         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1517                 sas_ss_reset(p);
1518
1519         /*
1520          * Syscall tracing and stepping should be turned off in the
1521          * child regardless of CLONE_PTRACE.
1522          */
1523         user_disable_single_step(p);
1524         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1525 #ifdef TIF_SYSCALL_EMU
1526         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1527 #endif
1528         clear_all_latency_tracing(p);
1529
1530         /* ok, now we should be set up.. */
1531         p->pid = pid_nr(pid);
1532         if (clone_flags & CLONE_THREAD) {
1533                 p->exit_signal = -1;
1534                 p->group_leader = current->group_leader;
1535                 p->tgid = current->tgid;
1536         } else {
1537                 if (clone_flags & CLONE_PARENT)
1538                         p->exit_signal = current->group_leader->exit_signal;
1539                 else
1540                         p->exit_signal = (clone_flags & CSIGNAL);
1541                 p->group_leader = p;
1542                 p->tgid = p->pid;
1543         }
1544
1545         p->nr_dirtied = 0;
1546         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1547         p->dirty_paused_when = 0;
1548
1549         p->pdeath_signal = 0;
1550         INIT_LIST_HEAD(&p->thread_group);
1551         p->task_works = NULL;
1552
1553         /*
1554          * Ensure that the cgroup subsystem policies allow the new process to be
1555          * forked. It should be noted the the new process's css_set can be changed
1556          * between here and cgroup_post_fork() if an organisation operation is in
1557          * progress.
1558          */
1559         retval = cgroup_can_fork(p);
1560         if (retval)
1561                 goto bad_fork_free_pid;
1562
1563         /*
1564          * Make it visible to the rest of the system, but dont wake it up yet.
1565          * Need tasklist lock for parent etc handling!
1566          */
1567         write_lock_irq(&tasklist_lock);
1568
1569         /* CLONE_PARENT re-uses the old parent */
1570         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1571                 p->real_parent = current->real_parent;
1572                 p->parent_exec_id = current->parent_exec_id;
1573         } else {
1574                 p->real_parent = current;
1575                 p->parent_exec_id = current->self_exec_id;
1576         }
1577
1578         spin_lock(&current->sighand->siglock);
1579
1580         /*
1581          * Copy seccomp details explicitly here, in case they were changed
1582          * before holding sighand lock.
1583          */
1584         copy_seccomp(p);
1585
1586         /*
1587          * Process group and session signals need to be delivered to just the
1588          * parent before the fork or both the parent and the child after the
1589          * fork. Restart if a signal comes in before we add the new process to
1590          * it's process group.
1591          * A fatal signal pending means that current will exit, so the new
1592          * thread can't slip out of an OOM kill (or normal SIGKILL).
1593         */
1594         recalc_sigpending();
1595         if (signal_pending(current)) {
1596                 spin_unlock(&current->sighand->siglock);
1597                 write_unlock_irq(&tasklist_lock);
1598                 retval = -ERESTARTNOINTR;
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);
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         cgroup_cancel_fork(p);
1652 bad_fork_free_pid:
1653         if (pid != &init_struct_pid)
1654                 free_pid(pid);
1655 bad_fork_cleanup_thread:
1656         exit_thread(p);
1657 bad_fork_cleanup_io:
1658         if (p->io_context)
1659                 exit_io_context(p);
1660 bad_fork_cleanup_namespaces:
1661         exit_task_namespaces(p);
1662 bad_fork_cleanup_mm:
1663         if (p->mm)
1664                 mmput(p->mm);
1665 bad_fork_cleanup_signal:
1666         if (!(clone_flags & CLONE_THREAD))
1667                 free_signal_struct(p->signal);
1668 bad_fork_cleanup_sighand:
1669         __cleanup_sighand(p->sighand);
1670 bad_fork_cleanup_fs:
1671         exit_fs(p); /* blocking */
1672 bad_fork_cleanup_files:
1673         exit_files(p); /* blocking */
1674 bad_fork_cleanup_semundo:
1675         exit_sem(p);
1676 bad_fork_cleanup_audit:
1677         audit_free(p);
1678 bad_fork_cleanup_perf:
1679         perf_event_free_task(p);
1680 bad_fork_cleanup_policy:
1681 #ifdef CONFIG_NUMA
1682         mpol_put(p->mempolicy);
1683 bad_fork_cleanup_threadgroup_lock:
1684 #endif
1685         threadgroup_change_end(current);
1686         delayacct_tsk_free(p);
1687 bad_fork_cleanup_count:
1688         atomic_dec(&p->cred->user->processes);
1689         exit_creds(p);
1690 bad_fork_free:
1691         free_task(p);
1692 fork_out:
1693         return ERR_PTR(retval);
1694 }
1695
1696 static inline void init_idle_pids(struct pid_link *links)
1697 {
1698         enum pid_type type;
1699
1700         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1701                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1702                 links[type].pid = &init_struct_pid;
1703         }
1704 }
1705
1706 struct task_struct *fork_idle(int cpu)
1707 {
1708         struct task_struct *task;
1709         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0);
1710         if (!IS_ERR(task)) {
1711                 init_idle_pids(task->pids);
1712                 init_idle(task, cpu);
1713         }
1714
1715         return task;
1716 }
1717
1718 /*
1719  *  Ok, this is the main fork-routine.
1720  *
1721  * It copies the process, and if successful kick-starts
1722  * it and waits for it to finish using the VM if required.
1723  */
1724 long _do_fork(unsigned long clone_flags,
1725               unsigned long stack_start,
1726               unsigned long stack_size,
1727               int __user *parent_tidptr,
1728               int __user *child_tidptr,
1729               unsigned long tls)
1730 {
1731         struct task_struct *p;
1732         int trace = 0;
1733         long nr;
1734
1735         /*
1736          * Determine whether and which event to report to ptracer.  When
1737          * called from kernel_thread or CLONE_UNTRACED is explicitly
1738          * requested, no event is reported; otherwise, report if the event
1739          * for the type of forking is enabled.
1740          */
1741         if (!(clone_flags & CLONE_UNTRACED)) {
1742                 if (clone_flags & CLONE_VFORK)
1743                         trace = PTRACE_EVENT_VFORK;
1744                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1745                         trace = PTRACE_EVENT_CLONE;
1746                 else
1747                         trace = PTRACE_EVENT_FORK;
1748
1749                 if (likely(!ptrace_event_enabled(current, trace)))
1750                         trace = 0;
1751         }
1752
1753         p = copy_process(clone_flags, stack_start, stack_size,
1754                          child_tidptr, NULL, trace, tls);
1755         /*
1756          * Do this prior waking up the new thread - the thread pointer
1757          * might get invalid after that point, if the thread exits quickly.
1758          */
1759         if (!IS_ERR(p)) {
1760                 struct completion vfork;
1761                 struct pid *pid;
1762
1763                 trace_sched_process_fork(current, p);
1764
1765                 pid = get_task_pid(p, PIDTYPE_PID);
1766                 nr = pid_vnr(pid);
1767
1768                 if (clone_flags & CLONE_PARENT_SETTID)
1769                         put_user(nr, parent_tidptr);
1770
1771                 if (clone_flags & CLONE_VFORK) {
1772                         p->vfork_done = &vfork;
1773                         init_completion(&vfork);
1774                         get_task_struct(p);
1775                 }
1776
1777                 wake_up_new_task(p);
1778
1779                 /* forking complete and child started to run, tell ptracer */
1780                 if (unlikely(trace))
1781                         ptrace_event_pid(trace, pid);
1782
1783                 if (clone_flags & CLONE_VFORK) {
1784                         if (!wait_for_vfork_done(p, &vfork))
1785                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1786                 }
1787
1788                 put_pid(pid);
1789         } else {
1790                 nr = PTR_ERR(p);
1791         }
1792         return nr;
1793 }
1794
1795 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1796 /* For compatibility with architectures that call do_fork directly rather than
1797  * using the syscall entry points below. */
1798 long do_fork(unsigned long clone_flags,
1799               unsigned long stack_start,
1800               unsigned long stack_size,
1801               int __user *parent_tidptr,
1802               int __user *child_tidptr)
1803 {
1804         return _do_fork(clone_flags, stack_start, stack_size,
1805                         parent_tidptr, child_tidptr, 0);
1806 }
1807 #endif
1808
1809 /*
1810  * Create a kernel thread.
1811  */
1812 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1813 {
1814         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1815                 (unsigned long)arg, NULL, NULL, 0);
1816 }
1817
1818 #ifdef __ARCH_WANT_SYS_FORK
1819 SYSCALL_DEFINE0(fork)
1820 {
1821 #ifdef CONFIG_MMU
1822         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1823 #else
1824         /* can not support in nommu mode */
1825         return -EINVAL;
1826 #endif
1827 }
1828 #endif
1829
1830 #ifdef __ARCH_WANT_SYS_VFORK
1831 SYSCALL_DEFINE0(vfork)
1832 {
1833         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1834                         0, NULL, NULL, 0);
1835 }
1836 #endif
1837
1838 #ifdef __ARCH_WANT_SYS_CLONE
1839 #ifdef CONFIG_CLONE_BACKWARDS
1840 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1841                  int __user *, parent_tidptr,
1842                  unsigned long, tls,
1843                  int __user *, child_tidptr)
1844 #elif defined(CONFIG_CLONE_BACKWARDS2)
1845 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1846                  int __user *, parent_tidptr,
1847                  int __user *, child_tidptr,
1848                  unsigned long, tls)
1849 #elif defined(CONFIG_CLONE_BACKWARDS3)
1850 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1851                 int, stack_size,
1852                 int __user *, parent_tidptr,
1853                 int __user *, child_tidptr,
1854                 unsigned long, tls)
1855 #else
1856 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1857                  int __user *, parent_tidptr,
1858                  int __user *, child_tidptr,
1859                  unsigned long, tls)
1860 #endif
1861 {
1862         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
1863 }
1864 #endif
1865
1866 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1867 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1868 #endif
1869
1870 static void sighand_ctor(void *data)
1871 {
1872         struct sighand_struct *sighand = data;
1873
1874         spin_lock_init(&sighand->siglock);
1875         init_waitqueue_head(&sighand->signalfd_wqh);
1876 }
1877
1878 void __init proc_caches_init(void)
1879 {
1880         sighand_cachep = kmem_cache_create("sighand_cache",
1881                         sizeof(struct sighand_struct), 0,
1882                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1883                         SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
1884         signal_cachep = kmem_cache_create("signal_cache",
1885                         sizeof(struct signal_struct), 0,
1886                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1887                         NULL);
1888         files_cachep = kmem_cache_create("files_cache",
1889                         sizeof(struct files_struct), 0,
1890                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1891                         NULL);
1892         fs_cachep = kmem_cache_create("fs_cache",
1893                         sizeof(struct fs_struct), 0,
1894                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1895                         NULL);
1896         /*
1897          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1898          * whole struct cpumask for the OFFSTACK case. We could change
1899          * this to *only* allocate as much of it as required by the
1900          * maximum number of CPU's we can ever have.  The cpumask_allocation
1901          * is at the end of the structure, exactly for that reason.
1902          */
1903         mm_cachep = kmem_cache_create("mm_struct",
1904                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1905                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1906                         NULL);
1907         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
1908         mmap_init();
1909         nsproxy_cache_init();
1910 }
1911
1912 /*
1913  * Check constraints on flags passed to the unshare system call.
1914  */
1915 static int check_unshare_flags(unsigned long unshare_flags)
1916 {
1917         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1918                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1919                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1920                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
1921                 return -EINVAL;
1922         /*
1923          * Not implemented, but pretend it works if there is nothing
1924          * to unshare.  Note that unsharing the address space or the
1925          * signal handlers also need to unshare the signal queues (aka
1926          * CLONE_THREAD).
1927          */
1928         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1929                 if (!thread_group_empty(current))
1930                         return -EINVAL;
1931         }
1932         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
1933                 if (atomic_read(&current->sighand->count) > 1)
1934                         return -EINVAL;
1935         }
1936         if (unshare_flags & CLONE_VM) {
1937                 if (!current_is_single_threaded())
1938                         return -EINVAL;
1939         }
1940
1941         return 0;
1942 }
1943
1944 /*
1945  * Unshare the filesystem structure if it is being shared
1946  */
1947 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1948 {
1949         struct fs_struct *fs = current->fs;
1950
1951         if (!(unshare_flags & CLONE_FS) || !fs)
1952                 return 0;
1953
1954         /* don't need lock here; in the worst case we'll do useless copy */
1955         if (fs->users == 1)
1956                 return 0;
1957
1958         *new_fsp = copy_fs_struct(fs);
1959         if (!*new_fsp)
1960                 return -ENOMEM;
1961
1962         return 0;
1963 }
1964
1965 /*
1966  * Unshare file descriptor table if it is being shared
1967  */
1968 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1969 {
1970         struct files_struct *fd = current->files;
1971         int error = 0;
1972
1973         if ((unshare_flags & CLONE_FILES) &&
1974             (fd && atomic_read(&fd->count) > 1)) {
1975                 *new_fdp = dup_fd(fd, &error);
1976                 if (!*new_fdp)
1977                         return error;
1978         }
1979
1980         return 0;
1981 }
1982
1983 /*
1984  * unshare allows a process to 'unshare' part of the process
1985  * context which was originally shared using clone.  copy_*
1986  * functions used by do_fork() cannot be used here directly
1987  * because they modify an inactive task_struct that is being
1988  * constructed. Here we are modifying the current, active,
1989  * task_struct.
1990  */
1991 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1992 {
1993         struct fs_struct *fs, *new_fs = NULL;
1994         struct files_struct *fd, *new_fd = NULL;
1995         struct cred *new_cred = NULL;
1996         struct nsproxy *new_nsproxy = NULL;
1997         int do_sysvsem = 0;
1998         int err;
1999
2000         /*
2001          * If unsharing a user namespace must also unshare the thread group
2002          * and unshare the filesystem root and working directories.
2003          */
2004         if (unshare_flags & CLONE_NEWUSER)
2005                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2006         /*
2007          * If unsharing vm, must also unshare signal handlers.
2008          */
2009         if (unshare_flags & CLONE_VM)
2010                 unshare_flags |= CLONE_SIGHAND;
2011         /*
2012          * If unsharing a signal handlers, must also unshare the signal queues.
2013          */
2014         if (unshare_flags & CLONE_SIGHAND)
2015                 unshare_flags |= CLONE_THREAD;
2016         /*
2017          * If unsharing namespace, must also unshare filesystem information.
2018          */
2019         if (unshare_flags & CLONE_NEWNS)
2020                 unshare_flags |= CLONE_FS;
2021
2022         err = check_unshare_flags(unshare_flags);
2023         if (err)
2024                 goto bad_unshare_out;
2025         /*
2026          * CLONE_NEWIPC must also detach from the undolist: after switching
2027          * to a new ipc namespace, the semaphore arrays from the old
2028          * namespace are unreachable.
2029          */
2030         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2031                 do_sysvsem = 1;
2032         err = unshare_fs(unshare_flags, &new_fs);
2033         if (err)
2034                 goto bad_unshare_out;
2035         err = unshare_fd(unshare_flags, &new_fd);
2036         if (err)
2037                 goto bad_unshare_cleanup_fs;
2038         err = unshare_userns(unshare_flags, &new_cred);
2039         if (err)
2040                 goto bad_unshare_cleanup_fd;
2041         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2042                                          new_cred, new_fs);
2043         if (err)
2044                 goto bad_unshare_cleanup_cred;
2045
2046         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2047                 if (do_sysvsem) {
2048                         /*
2049                          * CLONE_SYSVSEM is equivalent to sys_exit().
2050                          */
2051                         exit_sem(current);
2052                 }
2053                 if (unshare_flags & CLONE_NEWIPC) {
2054                         /* Orphan segments in old ns (see sem above). */
2055                         exit_shm(current);
2056                         shm_init_task(current);
2057                 }
2058
2059                 if (new_nsproxy)
2060                         switch_task_namespaces(current, new_nsproxy);
2061
2062                 task_lock(current);
2063
2064                 if (new_fs) {
2065                         fs = current->fs;
2066                         spin_lock(&fs->lock);
2067                         current->fs = new_fs;
2068                         if (--fs->users)
2069                                 new_fs = NULL;
2070                         else
2071                                 new_fs = fs;
2072                         spin_unlock(&fs->lock);
2073                 }
2074
2075                 if (new_fd) {
2076                         fd = current->files;
2077                         current->files = new_fd;
2078                         new_fd = fd;
2079                 }
2080
2081                 task_unlock(current);
2082
2083                 if (new_cred) {
2084                         /* Install the new user namespace */
2085                         commit_creds(new_cred);
2086                         new_cred = NULL;
2087                 }
2088         }
2089
2090 bad_unshare_cleanup_cred:
2091         if (new_cred)
2092                 put_cred(new_cred);
2093 bad_unshare_cleanup_fd:
2094         if (new_fd)
2095                 put_files_struct(new_fd);
2096
2097 bad_unshare_cleanup_fs:
2098         if (new_fs)
2099                 free_fs_struct(new_fs);
2100
2101 bad_unshare_out:
2102         return err;
2103 }
2104
2105 /*
2106  *      Helper to unshare the files of the current task.
2107  *      We don't want to expose copy_files internals to
2108  *      the exec layer of the kernel.
2109  */
2110
2111 int unshare_files(struct files_struct **displaced)
2112 {
2113         struct task_struct *task = current;
2114         struct files_struct *copy = NULL;
2115         int error;
2116
2117         error = unshare_fd(CLONE_FILES, &copy);
2118         if (error || !copy) {
2119                 *displaced = NULL;
2120                 return error;
2121         }
2122         *displaced = task->files;
2123         task_lock(task);
2124         task->files = copy;
2125         task_unlock(task);
2126         return 0;
2127 }
2128
2129 int sysctl_max_threads(struct ctl_table *table, int write,
2130                        void __user *buffer, size_t *lenp, loff_t *ppos)
2131 {
2132         struct ctl_table t;
2133         int ret;
2134         int threads = max_threads;
2135         int min = MIN_THREADS;
2136         int max = MAX_THREADS;
2137
2138         t = *table;
2139         t.data = &threads;
2140         t.extra1 = &min;
2141         t.extra2 = &max;
2142
2143         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2144         if (ret || !write)
2145                 return ret;
2146
2147         set_max_threads(threads);
2148
2149         return 0;
2150 }