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 /*
 703  * Decrement the use count and release all resources for an mm.
 704  */
 705 void mmput(struct mm_struct *mm)
 706 {
 707         might_sleep();
 708
 709         if (atomic_dec_and_test(&mm->mm_users)) {
 710                 uprobe_clear_state(mm);
 711                 exit_aio(mm);
 712                 ksm_exit(mm);
 713                 khugepaged_exit(mm); /* must run before exit_mmap */
 714                 exit_mmap(mm);
 715                 set_mm_exe_file(mm, NULL);
 716                 if (!list_empty(&mm->mmlist)) {
 717                         spin_lock(&mmlist_lock);
 718                         list_del(&mm->mmlist);
 719                         spin_unlock(&mmlist_lock);
 720                 }
 721                 if (mm->binfmt)
 722                         module_put(mm->binfmt->module);
 723                 mmdrop(mm);
 724         }
 725 }
 726 EXPORT_SYMBOL_GPL(mmput);
 727
 728 /**
 729  * set_mm_exe_file - change a reference to the mm's executable file
 730  *
 731  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 732  *
 733  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 734  * invocations: in mmput() nobody alive left, in execve task is single
 735  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 736  * mm->exe_file, but does so without using set_mm_exe_file() in order
 737  * to do avoid the need for any locks.
 738  */
 739 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 740 {
 741         struct file *old_exe_file;
 742
 743         /*
 744          * It is safe to dereference the exe_file without RCU as
 745          * this function is only called if nobody else can access
 746          * this mm -- see comment above for justification.
 747          */
 748         old_exe_file = rcu_dereference_raw(mm->exe_file);
 749
 750         if (new_exe_file)
 751                 get_file(new_exe_file);
 752         rcu_assign_pointer(mm->exe_file, new_exe_file);
 753         if (old_exe_file)
 754                 fput(old_exe_file);
 755 }
 756
 757 /**
 758  * get_mm_exe_file - acquire a reference to the mm's executable file
 759  *
 760  * Returns %NULL if mm has no associated executable file.
 761  * User must release file via fput().
 762  */
 763 struct file *get_mm_exe_file(struct mm_struct *mm)
 764 {
 765         struct file *exe_file;
 766
 767         rcu_read_lock();
 768         exe_file = rcu_dereference(mm->exe_file);
 769         if (exe_file && !get_file_rcu(exe_file))
 770                 exe_file = NULL;
 771         rcu_read_unlock();
 772         return exe_file;
 773 }
 774 EXPORT_SYMBOL(get_mm_exe_file);
 775
 776 /**
 777  * get_task_exe_file - acquire a reference to the task's executable file
 778  *
 779  * Returns %NULL if task's mm (if any) has no associated executable file or
 780  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
 781  * User must release file via fput().
 782  */
 783 struct file *get_task_exe_file(struct task_struct *task)
 784 {
 785         struct file *exe_file = NULL;
 786         struct mm_struct *mm;
 787
 788         task_lock(task);
 789         mm = task->mm;
 790         if (mm) {
 791                 if (!(task->flags & PF_KTHREAD))
 792                         exe_file = get_mm_exe_file(mm);
 793         }
 794         task_unlock(task);
 795         return exe_file;
 796 }
 797 EXPORT_SYMBOL(get_task_exe_file);
 798
 799 /**
 800  * get_task_mm - acquire a reference to the task's mm
 801  *
 802  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 803  * this kernel workthread has transiently adopted a user mm with use_mm,
 804  * to do its AIO) is not set and if so returns a reference to it, after
 805  * bumping up the use count.  User must release the mm via mmput()
 806  * after use.  Typically used by /proc and ptrace.
 807  */
 808 struct mm_struct *get_task_mm(struct task_struct *task)
 809 {
 810         struct mm_struct *mm;
 811
 812         task_lock(task);
 813         mm = task->mm;
 814         if (mm) {
 815                 if (task->flags & PF_KTHREAD)
 816                         mm = NULL;
 817                 else
 818                         atomic_inc(&mm->mm_users);
 819         }
 820         task_unlock(task);
 821         return mm;
 822 }
 823 EXPORT_SYMBOL_GPL(get_task_mm);
 824
 825 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 826 {
 827         struct mm_struct *mm;
 828         int err;
 829
 830         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 831         if (err)
 832                 return ERR_PTR(err);
 833
 834         mm = get_task_mm(task);
 835         if (mm && mm != current->mm &&
 836                         !ptrace_may_access(task, mode)) {
 837                 mmput(mm);
 838                 mm = ERR_PTR(-EACCES);
 839         }
 840         mutex_unlock(&task->signal->cred_guard_mutex);
 841
 842         return mm;
 843 }
 844
 845 static void complete_vfork_done(struct task_struct *tsk)
 846 {
 847         struct completion *vfork;
 848
 849         task_lock(tsk);
 850         vfork = tsk->vfork_done;
 851         if (likely(vfork)) {
 852                 tsk->vfork_done = NULL;
 853                 complete(vfork);
 854         }
 855         task_unlock(tsk);
 856 }
 857
 858 static int wait_for_vfork_done(struct task_struct *child,
 859                                 struct completion *vfork)
 860 {
 861         int killed;
 862
 863         freezer_do_not_count();
 864         killed = wait_for_completion_killable(vfork);
 865         freezer_count();
 866
 867         if (killed) {
 868                 task_lock(child);
 869                 child->vfork_done = NULL;
 870                 task_unlock(child);
 871         }
 872
 873         put_task_struct(child);
 874         return killed;
 875 }
 876
 877 /* Please note the differences between mmput and mm_release.
 878  * mmput is called whenever we stop holding onto a mm_struct,
 879  * error success whatever.
 880  *
 881  * mm_release is called after a mm_struct has been removed
 882  * from the current process.
 883  *
 884  * This difference is important for error handling, when we
 885  * only half set up a mm_struct for a new process and need to restore
 886  * the old one.  Because we mmput the new mm_struct before
 887  * restoring the old one. . .
 888  * Eric Biederman 10 January 1998
 889  */
 890 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 891 {
 892         /* Get rid of any futexes when releasing the mm */
 893 #ifdef CONFIG_FUTEX
 894         if (unlikely(tsk->robust_list)) {
 895                 exit_robust_list(tsk);
 896                 tsk->robust_list = NULL;
 897         }
 898 #ifdef CONFIG_COMPAT
 899         if (unlikely(tsk->compat_robust_list)) {
 900                 compat_exit_robust_list(tsk);
 901                 tsk->compat_robust_list = NULL;
 902         }
 903 #endif
 904         if (unlikely(!list_empty(&tsk->pi_state_list)))
 905                 exit_pi_state_list(tsk);
 906 #endif
 907
 908         uprobe_free_utask(tsk);
 909
 910         /* Get rid of any cached register state */
 911         deactivate_mm(tsk, mm);
 912
 913         /*
 914          * If we're exiting normally, clear a user-space tid field if
 915          * requested.  We leave this alone when dying by signal, to leave
 916          * the value intact in a core dump, and to save the unnecessary
 917          * trouble, say, a killed vfork parent shouldn't touch this mm.
 918          * Userland only wants this done for a sys_exit.
 919          */
 920         if (tsk->clear_child_tid) {
 921                 if (!(tsk->flags & PF_SIGNALED) &&
 922                     atomic_read(&mm->mm_users) > 1) {
 923                         /*
 924                          * We don't check the error code - if userspace has
 925                          * not set up a proper pointer then tough luck.
 926                          */
 927                         put_user(0, tsk->clear_child_tid);
 928                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 929                                         1, NULL, NULL, 0);
 930                 }
 931                 tsk->clear_child_tid = NULL;
 932         }
 933
 934         /*
 935          * All done, finally we can wake up parent and return this mm to him.
 936          * Also kthread_stop() uses this completion for synchronization.
 937          */
 938         if (tsk->vfork_done)
 939                 complete_vfork_done(tsk);
 940 }
 941
 942 /*
 943  * Allocate a new mm structure and copy contents from the
 944  * mm structure of the passed in task structure.
 945  */
 946 static struct mm_struct *dup_mm(struct task_struct *tsk)
 947 {
 948         struct mm_struct *mm, *oldmm = current->mm;
 949         int err;
 950
 951         mm = allocate_mm();
 952         if (!mm)
 953                 goto fail_nomem;
 954
 955         memcpy(mm, oldmm, sizeof(*mm));
 956
 957         if (!mm_init(mm, tsk))
 958                 goto fail_nomem;
 959
 960         err = dup_mmap(mm, oldmm);
 961         if (err)
 962                 goto free_pt;
 963
 964         mm->hiwater_rss = get_mm_rss(mm);
 965         mm->hiwater_vm = mm->total_vm;
 966
 967         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 968                 goto free_pt;
 969
 970         return mm;
 971
 972 free_pt:
 973         /* don't put binfmt in mmput, we haven't got module yet */
 974         mm->binfmt = NULL;
 975         mmput(mm);
 976
 977 fail_nomem:
 978         return NULL;
 979 }
 980
 981 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 982 {
 983         struct mm_struct *mm, *oldmm;
 984         int retval;
 985
 986         tsk->min_flt = tsk->maj_flt = 0;
 987         tsk->nvcsw = tsk->nivcsw = 0;
 988 #ifdef CONFIG_DETECT_HUNG_TASK
 989         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 990 #endif
 991
 992         tsk->mm = NULL;
 993         tsk->active_mm = NULL;
 994
 995         /*
 996          * Are we cloning a kernel thread?
 997          *
 998          * We need to steal a active VM for that..
 999          */
1000         oldmm = current->mm;
1001         if (!oldmm)
1002                 return 0;
1003
1004         /* initialize the new vmacache entries */
1005         vmacache_flush(tsk);
1006
1007         if (clone_flags & CLONE_VM) {
1008                 atomic_inc(&oldmm->mm_users);
1009                 mm = oldmm;
1010                 goto good_mm;
1011         }
1012
1013         retval = -ENOMEM;
1014         mm = dup_mm(tsk);
1015         if (!mm)
1016                 goto fail_nomem;
1017
1018 good_mm:
1019         tsk->mm = mm;
1020         tsk->active_mm = mm;
1021         return 0;
1022
1023 fail_nomem:
1024         return retval;
1025 }
1026
1027 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1028 {
1029         struct fs_struct *fs = current->fs;
1030         if (clone_flags & CLONE_FS) {
1031                 /* tsk->fs is already what we want */
1032                 spin_lock(&fs->lock);
1033                 if (fs->in_exec) {
1034                         spin_unlock(&fs->lock);
1035                         return -EAGAIN;
1036                 }
1037                 fs->users++;
1038                 spin_unlock(&fs->lock);
1039                 return 0;
1040         }
1041         tsk->fs = copy_fs_struct(fs);
1042         if (!tsk->fs)
1043                 return -ENOMEM;
1044         return 0;
1045 }
1046
1047 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1048 {
1049         struct files_struct *oldf, *newf;
1050         int error = 0;
1051
1052         /*
1053          * A background process may not have any files ...
1054          */
1055         oldf = current->files;
1056         if (!oldf)
1057                 goto out;
1058
1059         if (clone_flags & CLONE_FILES) {
1060                 atomic_inc(&oldf->count);
1061                 goto out;
1062         }
1063
1064         newf = dup_fd(oldf, &error);
1065         if (!newf)
1066                 goto out;
1067
1068         tsk->files = newf;
1069         error = 0;
1070 out:
1071         return error;
1072 }
1073
1074 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1075 {
1076 #ifdef CONFIG_BLOCK
1077         struct io_context *ioc = current->io_context;
1078         struct io_context *new_ioc;
1079
1080         if (!ioc)
1081                 return 0;
1082         /*
1083          * Share io context with parent, if CLONE_IO is set
1084          */
1085         if (clone_flags & CLONE_IO) {
1086                 ioc_task_link(ioc);
1087                 tsk->io_context = ioc;
1088         } else if (ioprio_valid(ioc->ioprio)) {
1089                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1090                 if (unlikely(!new_ioc))
1091                         return -ENOMEM;
1092
1093                 new_ioc->ioprio = ioc->ioprio;
1094                 put_io_context(new_ioc);
1095         }
1096 #endif
1097         return 0;
1098 }
1099
1100 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1101 {
1102         struct sighand_struct *sig;
1103
1104         if (clone_flags & CLONE_SIGHAND) {
1105                 atomic_inc(&current->sighand->count);
1106                 return 0;
1107         }
1108         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1109         rcu_assign_pointer(tsk->sighand, sig);
1110         if (!sig)
1111                 return -ENOMEM;
1112
1113         atomic_set(&sig->count, 1);
1114         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1115         return 0;
1116 }
1117
1118 void __cleanup_sighand(struct sighand_struct *sighand)
1119 {
1120         if (atomic_dec_and_test(&sighand->count)) {
1121                 signalfd_cleanup(sighand);
1122                 /*
1123                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1124                  * without an RCU grace period, see __lock_task_sighand().
1125                  */
1126                 kmem_cache_free(sighand_cachep, sighand);
1127         }
1128 }
1129
1130 /*
1131  * Initialize POSIX timer handling for a thread group.
1132  */
1133 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1134 {
1135         unsigned long cpu_limit;
1136
1137         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1138         if (cpu_limit != RLIM_INFINITY) {
1139                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1140                 sig->cputimer.running = true;
1141         }
1142
1143         /* The timer lists. */
1144         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1145         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1146         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1147 }
1148
1149 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1150 {
1151         struct signal_struct *sig;
1152
1153         if (clone_flags & CLONE_THREAD)
1154                 return 0;
1155
1156         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1157         tsk->signal = sig;
1158         if (!sig)
1159                 return -ENOMEM;
1160
1161         sig->nr_threads = 1;
1162         atomic_set(&sig->live, 1);
1163         atomic_set(&sig->sigcnt, 1);
1164
1165         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1166         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1167         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1168
1169         init_waitqueue_head(&sig->wait_chldexit);
1170         sig->curr_target = tsk;
1171         init_sigpending(&sig->shared_pending);
1172         INIT_LIST_HEAD(&sig->posix_timers);
1173         seqlock_init(&sig->stats_lock);
1174         prev_cputime_init(&sig->prev_cputime);
1175
1176         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1177         sig->real_timer.function = it_real_fn;
1178
1179         task_lock(current->group_leader);
1180         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1181         task_unlock(current->group_leader);
1182
1183         posix_cpu_timers_init_group(sig);
1184
1185         tty_audit_fork(sig);
1186         sched_autogroup_fork(sig);
1187
1188         sig->oom_score_adj = current->signal->oom_score_adj;
1189         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1190
1191         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1192                                    current->signal->is_child_subreaper;
1193
1194         mutex_init(&sig->cred_guard_mutex);
1195
1196         return 0;
1197 }
1198
1199 static void copy_seccomp(struct task_struct *p)
1200 {
1201 #ifdef CONFIG_SECCOMP
1202         /*
1203          * Must be called with sighand->lock held, which is common to
1204          * all threads in the group. Holding cred_guard_mutex is not
1205          * needed because this new task is not yet running and cannot
1206          * be racing exec.
1207          */
1208         assert_spin_locked(&current->sighand->siglock);
1209
1210         /* Ref-count the new filter user, and assign it. */
1211         get_seccomp_filter(current);
1212         p->seccomp = current->seccomp;
1213
1214         /*
1215          * Explicitly enable no_new_privs here in case it got set
1216          * between the task_struct being duplicated and holding the
1217          * sighand lock. The seccomp state and nnp must be in sync.
1218          */
1219         if (task_no_new_privs(current))
1220                 task_set_no_new_privs(p);
1221
1222         /*
1223          * If the parent gained a seccomp mode after copying thread
1224          * flags and between before we held the sighand lock, we have
1225          * to manually enable the seccomp thread flag here.
1226          */
1227         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1228                 set_tsk_thread_flag(p, TIF_SECCOMP);
1229 #endif
1230 }
1231
1232 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1233 {
1234         current->clear_child_tid = tidptr;
1235
1236         return task_pid_vnr(current);
1237 }
1238
1239 static void rt_mutex_init_task(struct task_struct *p)
1240 {
1241         raw_spin_lock_init(&p->pi_lock);
1242 #ifdef CONFIG_RT_MUTEXES
1243         p->pi_waiters = RB_ROOT;
1244         p->pi_waiters_leftmost = NULL;
1245         p->pi_blocked_on = NULL;
1246 #endif
1247 }
1248
1249 /*
1250  * Initialize POSIX timer handling for a single task.
1251  */
1252 static void posix_cpu_timers_init(struct task_struct *tsk)
1253 {
1254         tsk->cputime_expires.prof_exp = 0;
1255         tsk->cputime_expires.virt_exp = 0;
1256         tsk->cputime_expires.sched_exp = 0;
1257         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1258         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1259         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1260 }
1261
1262 static inline void
1263 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1264 {
1265          task->pids[type].pid = pid;
1266 }
1267
1268 /*
1269  * This creates a new process as a copy of the old one,
1270  * but does not actually start it yet.
1271  *
1272  * It copies the registers, and all the appropriate
1273  * parts of the process environment (as per the clone
1274  * flags). The actual kick-off is left to the caller.
1275  */
1276 static struct task_struct *copy_process(unsigned long clone_flags,
1277                                         unsigned long stack_start,
1278                                         unsigned long stack_size,
1279                                         int __user *child_tidptr,
1280                                         struct pid *pid,
1281                                         int trace,
1282                                         unsigned long tls)
1283 {
1284         int retval;
1285         struct task_struct *p;
1286
1287         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1288                 return ERR_PTR(-EINVAL);
1289
1290         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1291                 return ERR_PTR(-EINVAL);
1292
1293         /*
1294          * Thread groups must share signals as well, and detached threads
1295          * can only be started up within the thread group.
1296          */
1297         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1298                 return ERR_PTR(-EINVAL);
1299
1300         /*
1301          * Shared signal handlers imply shared VM. By way of the above,
1302          * thread groups also imply shared VM. Blocking this case allows
1303          * for various simplifications in other code.
1304          */
1305         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1306                 return ERR_PTR(-EINVAL);
1307
1308         /*
1309          * Siblings of global init remain as zombies on exit since they are
1310          * not reaped by their parent (swapper). To solve this and to avoid
1311          * multi-rooted process trees, prevent global and container-inits
1312          * from creating siblings.
1313          */
1314         if ((clone_flags & CLONE_PARENT) &&
1315                                 current->signal->flags & SIGNAL_UNKILLABLE)
1316                 return ERR_PTR(-EINVAL);
1317
1318         /*
1319          * If the new process will be in a different pid or user namespace
1320          * do not allow it to share a thread group with the forking task.
1321          */
1322         if (clone_flags & CLONE_THREAD) {
1323                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1324                     (task_active_pid_ns(current) !=
1325                                 current->nsproxy->pid_ns_for_children))
1326                         return ERR_PTR(-EINVAL);
1327         }
1328
1329         retval = security_task_create(clone_flags);
1330         if (retval)
1331                 goto fork_out;
1332
1333         retval = -ENOMEM;
1334         p = dup_task_struct(current);
1335         if (!p)
1336                 goto fork_out;
1337
1338         ftrace_graph_init_task(p);
1339
1340         rt_mutex_init_task(p);
1341
1342 #ifdef CONFIG_PROVE_LOCKING
1343         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1344         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1345 #endif
1346         retval = -EAGAIN;
1347         if (atomic_read(&p->real_cred->user->processes) >=
1348                         task_rlimit(p, RLIMIT_NPROC)) {
1349                 if (p->real_cred->user != INIT_USER &&
1350                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1351                         goto bad_fork_free;
1352         }
1353         current->flags &= ~PF_NPROC_EXCEEDED;
1354
1355         retval = copy_creds(p, clone_flags);
1356         if (retval < 0)
1357                 goto bad_fork_free;
1358
1359         /*
1360          * If multiple threads are within copy_process(), then this check
1361          * triggers too late. This doesn't hurt, the check is only there
1362          * to stop root fork bombs.
1363          */
1364         retval = -EAGAIN;
1365         if (nr_threads >= max_threads)
1366                 goto bad_fork_cleanup_count;
1367
1368         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1369         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1370         p->flags |= PF_FORKNOEXEC;
1371         INIT_LIST_HEAD(&p->children);
1372         INIT_LIST_HEAD(&p->sibling);
1373         rcu_copy_process(p);
1374         p->vfork_done = NULL;
1375         spin_lock_init(&p->alloc_lock);
1376
1377         init_sigpending(&p->pending);
1378
1379         p->utime = p->stime = p->gtime = 0;
1380         p->utimescaled = p->stimescaled = 0;
1381         prev_cputime_init(&p->prev_cputime);
1382
1383 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1384         seqcount_init(&p->vtime_seqcount);
1385         p->vtime_snap = 0;
1386         p->vtime_snap_whence = VTIME_INACTIVE;
1387 #endif
1388
1389 #if defined(SPLIT_RSS_COUNTING)
1390         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1391 #endif
1392
1393         p->default_timer_slack_ns = current->timer_slack_ns;
1394
1395         task_io_accounting_init(&p->ioac);
1396         acct_clear_integrals(p);
1397
1398         posix_cpu_timers_init(p);
1399
1400         p->start_time = ktime_get_ns();
1401         p->real_start_time = ktime_get_boot_ns();
1402         p->io_context = NULL;
1403         p->audit_context = NULL;
1404         threadgroup_change_begin(current);
1405         cgroup_fork(p);
1406 #ifdef CONFIG_NUMA
1407         p->mempolicy = mpol_dup(p->mempolicy);
1408         if (IS_ERR(p->mempolicy)) {
1409                 retval = PTR_ERR(p->mempolicy);
1410                 p->mempolicy = NULL;
1411                 goto bad_fork_cleanup_threadgroup_lock;
1412         }
1413 #endif
1414 #ifdef CONFIG_CPUSETS
1415         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1416         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1417         seqcount_init(&p->mems_allowed_seq);
1418 #endif
1419 #ifdef CONFIG_TRACE_IRQFLAGS
1420         p->irq_events = 0;
1421         p->hardirqs_enabled = 0;
1422         p->hardirq_enable_ip = 0;
1423         p->hardirq_enable_event = 0;
1424         p->hardirq_disable_ip = _THIS_IP_;
1425         p->hardirq_disable_event = 0;
1426         p->softirqs_enabled = 1;
1427         p->softirq_enable_ip = _THIS_IP_;
1428         p->softirq_enable_event = 0;
1429         p->softirq_disable_ip = 0;
1430         p->softirq_disable_event = 0;
1431         p->hardirq_context = 0;
1432         p->softirq_context = 0;
1433 #endif
1434
1435         p->pagefault_disabled = 0;
1436
1437 #ifdef CONFIG_LOCKDEP
1438         p->lockdep_depth = 0; /* no locks held yet */
1439         p->curr_chain_key = 0;
1440         p->lockdep_recursion = 0;
1441 #endif
1442
1443 #ifdef CONFIG_DEBUG_MUTEXES
1444         p->blocked_on = NULL; /* not blocked yet */
1445 #endif
1446 #ifdef CONFIG_BCACHE
1447         p->sequential_io        = 0;
1448         p->sequential_io_avg    = 0;
1449 #endif
1450
1451         /* Perform scheduler related setup. Assign this task to a CPU. */
1452         retval = sched_fork(clone_flags, p);
1453         if (retval)
1454                 goto bad_fork_cleanup_policy;
1455
1456         retval = perf_event_init_task(p);
1457         if (retval)
1458                 goto bad_fork_cleanup_policy;
1459         retval = audit_alloc(p);
1460         if (retval)
1461                 goto bad_fork_cleanup_perf;
1462         /* copy all the process information */
1463         shm_init_task(p);
1464         retval = copy_semundo(clone_flags, p);
1465         if (retval)
1466                 goto bad_fork_cleanup_audit;
1467         retval = copy_files(clone_flags, p);
1468         if (retval)
1469                 goto bad_fork_cleanup_semundo;
1470         retval = copy_fs(clone_flags, p);
1471         if (retval)
1472                 goto bad_fork_cleanup_files;
1473         retval = copy_sighand(clone_flags, p);
1474         if (retval)
1475                 goto bad_fork_cleanup_fs;
1476         retval = copy_signal(clone_flags, p);
1477         if (retval)
1478                 goto bad_fork_cleanup_sighand;
1479         retval = copy_mm(clone_flags, p);
1480         if (retval)
1481                 goto bad_fork_cleanup_signal;
1482         retval = copy_namespaces(clone_flags, p);
1483         if (retval)
1484                 goto bad_fork_cleanup_mm;
1485         retval = copy_io(clone_flags, p);
1486         if (retval)
1487                 goto bad_fork_cleanup_namespaces;
1488         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1489         if (retval)
1490                 goto bad_fork_cleanup_io;
1491
1492         if (pid != &init_struct_pid) {
1493                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1494                 if (IS_ERR(pid)) {
1495                         retval = PTR_ERR(pid);
1496                         goto bad_fork_cleanup_io;
1497                 }
1498         }
1499
1500         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1501         /*
1502          * Clear TID on mm_release()?
1503          */
1504         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1505 #ifdef CONFIG_BLOCK
1506         p->plug = NULL;
1507 #endif
1508 #ifdef CONFIG_FUTEX
1509         p->robust_list = NULL;
1510 #ifdef CONFIG_COMPAT
1511         p->compat_robust_list = NULL;
1512 #endif
1513         INIT_LIST_HEAD(&p->pi_state_list);
1514         p->pi_state_cache = NULL;
1515 #endif
1516         /*
1517          * sigaltstack should be cleared when sharing the same VM
1518          */
1519         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1520                 p->sas_ss_sp = p->sas_ss_size = 0;
1521
1522         /*
1523          * Syscall tracing and stepping should be turned off in the
1524          * child regardless of CLONE_PTRACE.
1525          */
1526         user_disable_single_step(p);
1527         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1528 #ifdef TIF_SYSCALL_EMU
1529         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1530 #endif
1531         clear_all_latency_tracing(p);
1532
1533         /* ok, now we should be set up.. */
1534         p->pid = pid_nr(pid);
1535         if (clone_flags & CLONE_THREAD) {
1536                 p->exit_signal = -1;
1537                 p->group_leader = current->group_leader;
1538                 p->tgid = current->tgid;
1539         } else {
1540                 if (clone_flags & CLONE_PARENT)
1541                         p->exit_signal = current->group_leader->exit_signal;
1542                 else
1543                         p->exit_signal = (clone_flags & CSIGNAL);
1544                 p->group_leader = p;
1545                 p->tgid = p->pid;
1546         }
1547
1548         p->nr_dirtied = 0;
1549         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1550         p->dirty_paused_when = 0;
1551
1552         p->pdeath_signal = 0;
1553         INIT_LIST_HEAD(&p->thread_group);
1554         p->task_works = NULL;
1555
1556         /*
1557          * Ensure that the cgroup subsystem policies allow the new process to be
1558          * forked. It should be noted the the new process's css_set can be changed
1559          * between here and cgroup_post_fork() if an organisation operation is in
1560          * progress.
1561          */
1562         retval = cgroup_can_fork(p);
1563         if (retval)
1564                 goto bad_fork_free_pid;
1565
1566         /*
1567          * Make it visible to the rest of the system, but dont wake it up yet.
1568          * Need tasklist lock for parent etc handling!
1569          */
1570         write_lock_irq(&tasklist_lock);
1571
1572         /* CLONE_PARENT re-uses the old parent */
1573         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1574                 p->real_parent = current->real_parent;
1575                 p->parent_exec_id = current->parent_exec_id;
1576         } else {
1577                 p->real_parent = current;
1578                 p->parent_exec_id = current->self_exec_id;
1579         }
1580
1581         spin_lock(&current->sighand->siglock);
1582
1583         /*
1584          * Copy seccomp details explicitly here, in case they were changed
1585          * before holding sighand lock.
1586          */
1587         copy_seccomp(p);
1588
1589         /*
1590          * Process group and session signals need to be delivered to just the
1591          * parent before the fork or both the parent and the child after the
1592          * fork. Restart if a signal comes in before we add the new process to
1593          * it's process group.
1594          * A fatal signal pending means that current will exit, so the new
1595          * thread can't slip out of an OOM kill (or normal SIGKILL).
1596         */
1597         recalc_sigpending();
1598         if (signal_pending(current)) {
1599                 spin_unlock(&current->sighand->siglock);
1600                 write_unlock_irq(&tasklist_lock);
1601                 retval = -ERESTARTNOINTR;
1602                 goto bad_fork_cancel_cgroup;
1603         }
1604
1605         if (likely(p->pid)) {
1606                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1607
1608                 init_task_pid(p, PIDTYPE_PID, pid);
1609                 if (thread_group_leader(p)) {
1610                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1611                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1612
1613                         if (is_child_reaper(pid)) {
1614                                 ns_of_pid(pid)->child_reaper = p;
1615                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1616                         }
1617
1618                         p->signal->leader_pid = pid;
1619                         p->signal->tty = tty_kref_get(current->signal->tty);
1620                         list_add_tail(&p->sibling, &p->real_parent->children);
1621                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1622                         attach_pid(p, PIDTYPE_PGID);
1623                         attach_pid(p, PIDTYPE_SID);
1624                         __this_cpu_inc(process_counts);
1625                 } else {
1626                         current->signal->nr_threads++;
1627                         atomic_inc(&current->signal->live);
1628                         atomic_inc(&current->signal->sigcnt);
1629                         list_add_tail_rcu(&p->thread_group,
1630                                           &p->group_leader->thread_group);
1631                         list_add_tail_rcu(&p->thread_node,
1632                                           &p->signal->thread_head);
1633                 }
1634                 attach_pid(p, PIDTYPE_PID);
1635                 nr_threads++;
1636         }
1637
1638         total_forks++;
1639         spin_unlock(&current->sighand->siglock);
1640         syscall_tracepoint_update(p);
1641         write_unlock_irq(&tasklist_lock);
1642
1643         proc_fork_connector(p);
1644         cgroup_post_fork(p);
1645         threadgroup_change_end(current);
1646         perf_event_fork(p);
1647
1648         trace_task_newtask(p, clone_flags);
1649         uprobe_copy_process(p, clone_flags);
1650
1651         return p;
1652
1653 bad_fork_cancel_cgroup:
1654         cgroup_cancel_fork(p);
1655 bad_fork_free_pid:
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         threadgroup_change_end(current);
1687         delayacct_tsk_free(p);
1688 bad_fork_cleanup_count:
1689         atomic_dec(&p->cred->user->processes);
1690         exit_creds(p);
1691 bad_fork_free:
1692         free_task(p);
1693 fork_out:
1694         return ERR_PTR(retval);
1695 }
1696
1697 static inline void init_idle_pids(struct pid_link *links)
1698 {
1699         enum pid_type type;
1700
1701         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1702                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1703                 links[type].pid = &init_struct_pid;
1704         }
1705 }
1706
1707 struct task_struct *fork_idle(int cpu)
1708 {
1709         struct task_struct *task;
1710         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0);
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);
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|SLAB_ACCOUNT, sighand_ctor);
1885         signal_cachep = kmem_cache_create("signal_cache",
1886                         sizeof(struct signal_struct), 0,
1887                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1888                         NULL);
1889         files_cachep = kmem_cache_create("files_cache",
1890                         sizeof(struct files_struct), 0,
1891                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1892                         NULL);
1893         fs_cachep = kmem_cache_create("fs_cache",
1894                         sizeof(struct fs_struct), 0,
1895                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1896                         NULL);
1897         /*
1898          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1899          * whole struct cpumask for the OFFSTACK case. We could change
1900          * this to *only* allocate as much of it as required by the
1901          * maximum number of CPU's we can ever have.  The cpumask_allocation
1902          * is at the end of the structure, exactly for that reason.
1903          */
1904         mm_cachep = kmem_cache_create("mm_struct",
1905                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1906                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1907                         NULL);
1908         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
1909         mmap_init();
1910         nsproxy_cache_init();
1911 }
1912
1913 /*
1914  * Check constraints on flags passed to the unshare system call.
1915  */
1916 static int check_unshare_flags(unsigned long unshare_flags)
1917 {
1918         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1919                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1920                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1921                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
1922                 return -EINVAL;
1923         /*
1924          * Not implemented, but pretend it works if there is nothing
1925          * to unshare.  Note that unsharing the address space or the
1926          * signal handlers also need to unshare the signal queues (aka
1927          * CLONE_THREAD).
1928          */
1929         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1930                 if (!thread_group_empty(current))
1931                         return -EINVAL;
1932         }
1933         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
1934                 if (atomic_read(&current->sighand->count) > 1)
1935                         return -EINVAL;
1936         }
1937         if (unshare_flags & CLONE_VM) {
1938                 if (!current_is_single_threaded())
1939                         return -EINVAL;
1940         }
1941
1942         return 0;
1943 }
1944
1945 /*
1946  * Unshare the filesystem structure if it is being shared
1947  */
1948 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1949 {
1950         struct fs_struct *fs = current->fs;
1951
1952         if (!(unshare_flags & CLONE_FS) || !fs)
1953                 return 0;
1954
1955         /* don't need lock here; in the worst case we'll do useless copy */
1956         if (fs->users == 1)
1957                 return 0;
1958
1959         *new_fsp = copy_fs_struct(fs);
1960         if (!*new_fsp)
1961                 return -ENOMEM;
1962
1963         return 0;
1964 }
1965
1966 /*
1967  * Unshare file descriptor table if it is being shared
1968  */
1969 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1970 {
1971         struct files_struct *fd = current->files;
1972         int error = 0;
1973
1974         if ((unshare_flags & CLONE_FILES) &&
1975             (fd && atomic_read(&fd->count) > 1)) {
1976                 *new_fdp = dup_fd(fd, &error);
1977                 if (!*new_fdp)
1978                         return error;
1979         }
1980
1981         return 0;
1982 }
1983
1984 /*
1985  * unshare allows a process to 'unshare' part of the process
1986  * context which was originally shared using clone.  copy_*
1987  * functions used by do_fork() cannot be used here directly
1988  * because they modify an inactive task_struct that is being
1989  * constructed. Here we are modifying the current, active,
1990  * task_struct.
1991  */
1992 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1993 {
1994         struct fs_struct *fs, *new_fs = NULL;
1995         struct files_struct *fd, *new_fd = NULL;
1996         struct cred *new_cred = NULL;
1997         struct nsproxy *new_nsproxy = NULL;
1998         int do_sysvsem = 0;
1999         int err;
2000
2001         /*
2002          * If unsharing a user namespace must also unshare the thread group
2003          * and unshare the filesystem root and working directories.
2004          */
2005         if (unshare_flags & CLONE_NEWUSER)
2006                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2007         /*
2008          * If unsharing vm, must also unshare signal handlers.
2009          */
2010         if (unshare_flags & CLONE_VM)
2011                 unshare_flags |= CLONE_SIGHAND;
2012         /*
2013          * If unsharing a signal handlers, must also unshare the signal queues.
2014          */
2015         if (unshare_flags & CLONE_SIGHAND)
2016                 unshare_flags |= CLONE_THREAD;
2017         /*
2018          * If unsharing namespace, must also unshare filesystem information.
2019          */
2020         if (unshare_flags & CLONE_NEWNS)
2021                 unshare_flags |= CLONE_FS;
2022
2023         err = check_unshare_flags(unshare_flags);
2024         if (err)
2025                 goto bad_unshare_out;
2026         /*
2027          * CLONE_NEWIPC must also detach from the undolist: after switching
2028          * to a new ipc namespace, the semaphore arrays from the old
2029          * namespace are unreachable.
2030          */
2031         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2032                 do_sysvsem = 1;
2033         err = unshare_fs(unshare_flags, &new_fs);
2034         if (err)
2035                 goto bad_unshare_out;
2036         err = unshare_fd(unshare_flags, &new_fd);
2037         if (err)
2038                 goto bad_unshare_cleanup_fs;
2039         err = unshare_userns(unshare_flags, &new_cred);
2040         if (err)
2041                 goto bad_unshare_cleanup_fd;
2042         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2043                                          new_cred, new_fs);
2044         if (err)
2045                 goto bad_unshare_cleanup_cred;
2046
2047         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2048                 if (do_sysvsem) {
2049                         /*
2050                          * CLONE_SYSVSEM is equivalent to sys_exit().
2051                          */
2052                         exit_sem(current);
2053                 }
2054                 if (unshare_flags & CLONE_NEWIPC) {
2055                         /* Orphan segments in old ns (see sem above). */
2056                         exit_shm(current);
2057                         shm_init_task(current);
2058                 }
2059
2060                 if (new_nsproxy)
2061                         switch_task_namespaces(current, new_nsproxy);
2062
2063                 task_lock(current);
2064
2065                 if (new_fs) {
2066                         fs = current->fs;
2067                         spin_lock(&fs->lock);
2068                         current->fs = new_fs;
2069                         if (--fs->users)
2070                                 new_fs = NULL;
2071                         else
2072                                 new_fs = fs;
2073                         spin_unlock(&fs->lock);
2074                 }
2075
2076                 if (new_fd) {
2077                         fd = current->files;
2078                         current->files = new_fd;
2079                         new_fd = fd;
2080                 }
2081
2082                 task_unlock(current);
2083
2084                 if (new_cred) {
2085                         /* Install the new user namespace */
2086                         commit_creds(new_cred);
2087                         new_cred = NULL;
2088                 }
2089         }
2090
2091 bad_unshare_cleanup_cred:
2092         if (new_cred)
2093                 put_cred(new_cred);
2094 bad_unshare_cleanup_fd:
2095         if (new_fd)
2096                 put_files_struct(new_fd);
2097
2098 bad_unshare_cleanup_fs:
2099         if (new_fs)
2100                 free_fs_struct(new_fs);
2101
2102 bad_unshare_out:
2103         return err;
2104 }
2105
2106 /*
2107  *      Helper to unshare the files of the current task.
2108  *      We don't want to expose copy_files internals to
2109  *      the exec layer of the kernel.
2110  */
2111
2112 int unshare_files(struct files_struct **displaced)
2113 {
2114         struct task_struct *task = current;
2115         struct files_struct *copy = NULL;
2116         int error;
2117
2118         error = unshare_fd(CLONE_FILES, &copy);
2119         if (error || !copy) {
2120                 *displaced = NULL;
2121                 return error;
2122         }
2123         *displaced = task->files;
2124         task_lock(task);
2125         task->files = copy;
2126         task_unlock(task);
2127         return 0;
2128 }
2129
2130 int sysctl_max_threads(struct ctl_table *table, int write,
2131                        void __user *buffer, size_t *lenp, loff_t *ppos)
2132 {
2133         struct ctl_table t;
2134         int ret;
2135         int threads = max_threads;
2136         int min = MIN_THREADS;
2137         int max = MAX_THREADS;
2138
2139         t = *table;
2140         t.data = &threads;
2141         t.extra1 = &min;
2142         t.extra2 = &max;
2143
2144         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2145         if (ret || !write)
2146                 return ret;
2147
2148         set_max_threads(threads);
2149
2150         return 0;
2151 }