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_stack(unsigned long *stack)
 152 {
 153 }
 154
 155 #ifndef CONFIG_ARCH_THREAD_STACK_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 unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 163                                                   int node)
 164 {
 165         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 166                                              THREAD_SIZE_ORDER);
 167
 168         return page ? page_address(page) : NULL;
 169 }
 170
 171 static inline void free_thread_stack(unsigned long *stack)
 172 {
 173         __free_pages(virt_to_page(stack), THREAD_SIZE_ORDER);
 174 }
 175 # else
 176 static struct kmem_cache *thread_stack_cache;
 177
 178 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 179                                                   int node)
 180 {
 181         return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 182 }
 183
 184 static void free_thread_stack(unsigned long *stack)
 185 {
 186         kmem_cache_free(thread_stack_cache, stack);
 187 }
 188
 189 void thread_stack_cache_init(void)
 190 {
 191         thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
 192                                               THREAD_SIZE, 0, NULL);
 193         BUG_ON(thread_stack_cache == NULL);
 194 }
 195 # endif
 196 #endif
 197
 198 /* SLAB cache for signal_struct structures (tsk->signal) */
 199 static struct kmem_cache *signal_cachep;
 200
 201 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 202 struct kmem_cache *sighand_cachep;
 203
 204 /* SLAB cache for files_struct structures (tsk->files) */
 205 struct kmem_cache *files_cachep;
 206
 207 /* SLAB cache for fs_struct structures (tsk->fs) */
 208 struct kmem_cache *fs_cachep;
 209
 210 /* SLAB cache for vm_area_struct structures */
 211 struct kmem_cache *vm_area_cachep;
 212
 213 /* SLAB cache for mm_struct structures (tsk->mm) */
 214 static struct kmem_cache *mm_cachep;
 215
 216 static void account_kernel_stack(unsigned long *stack, int account)
 217 {
 218         /* All stack pages are in the same zone and belong to the same memcg. */
 219         struct page *first_page = virt_to_page(stack);
 220
 221         mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 222                             THREAD_SIZE / 1024 * account);
 223
 224         memcg_kmem_update_page_stat(
 225                 first_page, MEMCG_KERNEL_STACK_KB,
 226                 account * (THREAD_SIZE / 1024));
 227 }
 228
 229 void free_task(struct task_struct *tsk)
 230 {
 231         account_kernel_stack(tsk->stack, -1);
 232         arch_release_thread_stack(tsk->stack);
 233         free_thread_stack(tsk->stack);
 234         rt_mutex_debug_task_free(tsk);
 235         ftrace_graph_exit_task(tsk);
 236         put_seccomp_filter(tsk);
 237         arch_release_task_struct(tsk);
 238         free_task_struct(tsk);
 239 }
 240 EXPORT_SYMBOL(free_task);
 241
 242 static inline void free_signal_struct(struct signal_struct *sig)
 243 {
 244         taskstats_tgid_free(sig);
 245         sched_autogroup_exit(sig);
 246         kmem_cache_free(signal_cachep, sig);
 247 }
 248
 249 static inline void put_signal_struct(struct signal_struct *sig)
 250 {
 251         if (atomic_dec_and_test(&sig->sigcnt))
 252                 free_signal_struct(sig);
 253 }
 254
 255 void __put_task_struct(struct task_struct *tsk)
 256 {
 257         WARN_ON(!tsk->exit_state);
 258         WARN_ON(atomic_read(&tsk->usage));
 259         WARN_ON(tsk == current);
 260
 261         cgroup_free(tsk);
 262         task_numa_free(tsk);
 263         security_task_free(tsk);
 264         exit_creds(tsk);
 265         delayacct_tsk_free(tsk);
 266         put_signal_struct(tsk->signal);
 267
 268         if (!profile_handoff_task(tsk))
 269                 free_task(tsk);
 270 }
 271 EXPORT_SYMBOL_GPL(__put_task_struct);
 272
 273 void __init __weak arch_task_cache_init(void) { }
 274
 275 /*
 276  * set_max_threads
 277  */
 278 static void set_max_threads(unsigned int max_threads_suggested)
 279 {
 280         u64 threads;
 281
 282         /*
 283          * The number of threads shall be limited such that the thread
 284          * structures may only consume a small part of the available memory.
 285          */
 286         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 287                 threads = MAX_THREADS;
 288         else
 289                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 290                                     (u64) THREAD_SIZE * 8UL);
 291
 292         if (threads > max_threads_suggested)
 293                 threads = max_threads_suggested;
 294
 295         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 296 }
 297
 298 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 299 /* Initialized by the architecture: */
 300 int arch_task_struct_size __read_mostly;
 301 #endif
 302
 303 void __init fork_init(void)
 304 {
 305 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 306 #ifndef ARCH_MIN_TASKALIGN
 307 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 308 #endif
 309         /* create a slab on which task_structs can be allocated */
 310         task_struct_cachep = kmem_cache_create("task_struct",
 311                         arch_task_struct_size, ARCH_MIN_TASKALIGN,
 312                         SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
 313 #endif
 314
 315         /* do the arch specific task caches init */
 316         arch_task_cache_init();
 317
 318         set_max_threads(MAX_THREADS);
 319
 320         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 321         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 322         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 323                 init_task.signal->rlim[RLIMIT_NPROC];
 324 }
 325
 326 int __weak arch_dup_task_struct(struct task_struct *dst,
 327                                                struct task_struct *src)
 328 {
 329         *dst = *src;
 330         return 0;
 331 }
 332
 333 void set_task_stack_end_magic(struct task_struct *tsk)
 334 {
 335         unsigned long *stackend;
 336
 337         stackend = end_of_stack(tsk);
 338         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 339 }
 340
 341 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 342 {
 343         struct task_struct *tsk;
 344         unsigned long *stack;
 345         int err;
 346
 347         if (node == NUMA_NO_NODE)
 348                 node = tsk_fork_get_node(orig);
 349         tsk = alloc_task_struct_node(node);
 350         if (!tsk)
 351                 return NULL;
 352
 353         stack = alloc_thread_stack_node(tsk, node);
 354         if (!stack)
 355                 goto free_tsk;
 356
 357         err = arch_dup_task_struct(tsk, orig);
 358         if (err)
 359                 goto free_stack;
 360
 361         tsk->stack = stack;
 362 #ifdef CONFIG_SECCOMP
 363         /*
 364          * We must handle setting up seccomp filters once we're under
 365          * the sighand lock in case orig has changed between now and
 366          * then. Until then, filter must be NULL to avoid messing up
 367          * the usage counts on the error path calling free_task.
 368          */
 369         tsk->seccomp.filter = NULL;
 370 #endif
 371
 372         setup_thread_stack(tsk, orig);
 373         clear_user_return_notifier(tsk);
 374         clear_tsk_need_resched(tsk);
 375         set_task_stack_end_magic(tsk);
 376
 377 #ifdef CONFIG_CC_STACKPROTECTOR
 378         tsk->stack_canary = get_random_int();
 379 #endif
 380
 381         /*
 382          * One for us, one for whoever does the "release_task()" (usually
 383          * parent)
 384          */
 385         atomic_set(&tsk->usage, 2);
 386 #ifdef CONFIG_BLK_DEV_IO_TRACE
 387         tsk->btrace_seq = 0;
 388 #endif
 389         tsk->splice_pipe = NULL;
 390         tsk->task_frag.page = NULL;
 391         tsk->wake_q.next = NULL;
 392
 393         account_kernel_stack(stack, 1);
 394
 395         kcov_task_init(tsk);
 396
 397         return tsk;
 398
 399 free_stack:
 400         free_thread_stack(stack);
 401 free_tsk:
 402         free_task_struct(tsk);
 403         return NULL;
 404 }
 405
 406 #ifdef CONFIG_MMU
 407 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 408 {
 409         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 410         struct rb_node **rb_link, *rb_parent;
 411         int retval;
 412         unsigned long charge;
 413
 414         uprobe_start_dup_mmap();
 415         if (down_write_killable(&oldmm->mmap_sem)) {
 416                 retval = -EINTR;
 417                 goto fail_uprobe_end;
 418         }
 419         flush_cache_dup_mm(oldmm);
 420         uprobe_dup_mmap(oldmm, mm);
 421         /*
 422          * Not linked in yet - no deadlock potential:
 423          */
 424         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 425
 426         /* No ordering required: file already has been exposed. */
 427         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 428
 429         mm->total_vm = oldmm->total_vm;
 430         mm->data_vm = oldmm->data_vm;
 431         mm->exec_vm = oldmm->exec_vm;
 432         mm->stack_vm = oldmm->stack_vm;
 433
 434         rb_link = &mm->mm_rb.rb_node;
 435         rb_parent = NULL;
 436         pprev = &mm->mmap;
 437         retval = ksm_fork(mm, oldmm);
 438         if (retval)
 439                 goto out;
 440         retval = khugepaged_fork(mm, oldmm);
 441         if (retval)
 442                 goto out;
 443
 444         prev = NULL;
 445         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 446                 struct file *file;
 447
 448                 if (mpnt->vm_flags & VM_DONTCOPY) {
 449                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 450                         continue;
 451                 }
 452                 charge = 0;
 453                 if (mpnt->vm_flags & VM_ACCOUNT) {
 454                         unsigned long len = vma_pages(mpnt);
 455
 456                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 457                                 goto fail_nomem;
 458                         charge = len;
 459                 }
 460                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 461                 if (!tmp)
 462                         goto fail_nomem;
 463                 *tmp = *mpnt;
 464                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 465                 retval = vma_dup_policy(mpnt, tmp);
 466                 if (retval)
 467                         goto fail_nomem_policy;
 468                 tmp->vm_mm = mm;
 469                 if (anon_vma_fork(tmp, mpnt))
 470                         goto fail_nomem_anon_vma_fork;
 471                 tmp->vm_flags &=
 472                         ~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
 473                 tmp->vm_next = tmp->vm_prev = NULL;
 474                 tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
 475                 file = tmp->vm_file;
 476                 if (file) {
 477                         struct inode *inode = file_inode(file);
 478                         struct address_space *mapping = file->f_mapping;
 479
 480                         get_file(file);
 481                         if (tmp->vm_flags & VM_DENYWRITE)
 482                                 atomic_dec(&inode->i_writecount);
 483                         i_mmap_lock_write(mapping);
 484                         if (tmp->vm_flags & VM_SHARED)
 485                                 atomic_inc(&mapping->i_mmap_writable);
 486                         flush_dcache_mmap_lock(mapping);
 487                         /* insert tmp into the share list, just after mpnt */
 488                         vma_interval_tree_insert_after(tmp, mpnt,
 489                                         &mapping->i_mmap);
 490                         flush_dcache_mmap_unlock(mapping);
 491                         i_mmap_unlock_write(mapping);
 492                 }
 493
 494                 /*
 495                  * Clear hugetlb-related page reserves for children. This only
 496                  * affects MAP_PRIVATE mappings. Faults generated by the child
 497                  * are not guaranteed to succeed, even if read-only
 498                  */
 499                 if (is_vm_hugetlb_page(tmp))
 500                         reset_vma_resv_huge_pages(tmp);
 501
 502                 /*
 503                  * Link in the new vma and copy the page table entries.
 504                  */
 505                 *pprev = tmp;
 506                 pprev = &tmp->vm_next;
 507                 tmp->vm_prev = prev;
 508                 prev = tmp;
 509
 510                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 511                 rb_link = &tmp->vm_rb.rb_right;
 512                 rb_parent = &tmp->vm_rb;
 513
 514                 mm->map_count++;
 515                 retval = copy_page_range(mm, oldmm, mpnt);
 516
 517                 if (tmp->vm_ops && tmp->vm_ops->open)
 518                         tmp->vm_ops->open(tmp);
 519
 520                 if (retval)
 521                         goto out;
 522         }
 523         /* a new mm has just been created */
 524         arch_dup_mmap(oldmm, mm);
 525         retval = 0;
 526 out:
 527         up_write(&mm->mmap_sem);
 528         flush_tlb_mm(oldmm);
 529         up_write(&oldmm->mmap_sem);
 530 fail_uprobe_end:
 531         uprobe_end_dup_mmap();
 532         return retval;
 533 fail_nomem_anon_vma_fork:
 534         mpol_put(vma_policy(tmp));
 535 fail_nomem_policy:
 536         kmem_cache_free(vm_area_cachep, tmp);
 537 fail_nomem:
 538         retval = -ENOMEM;
 539         vm_unacct_memory(charge);
 540         goto out;
 541 }
 542
 543 static inline int mm_alloc_pgd(struct mm_struct *mm)
 544 {
 545         mm->pgd = pgd_alloc(mm);
 546         if (unlikely(!mm->pgd))
 547                 return -ENOMEM;
 548         return 0;
 549 }
 550
 551 static inline void mm_free_pgd(struct mm_struct *mm)
 552 {
 553         pgd_free(mm, mm->pgd);
 554 }
 555 #else
 556 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 557 {
 558         down_write(&oldmm->mmap_sem);
 559         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 560         up_write(&oldmm->mmap_sem);
 561         return 0;
 562 }
 563 #define mm_alloc_pgd(mm)        (0)
 564 #define mm_free_pgd(mm)
 565 #endif /* CONFIG_MMU */
 566
 567 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 568
 569 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 570 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 571
 572 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 573
 574 static int __init coredump_filter_setup(char *s)
 575 {
 576         default_dump_filter =
 577                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 578                 MMF_DUMP_FILTER_MASK;
 579         return 1;
 580 }
 581
 582 __setup("coredump_filter=", coredump_filter_setup);
 583
 584 #include <linux/init_task.h>
 585
 586 static void mm_init_aio(struct mm_struct *mm)
 587 {
 588 #ifdef CONFIG_AIO
 589         spin_lock_init(&mm->ioctx_lock);
 590         mm->ioctx_table = NULL;
 591 #endif
 592 }
 593
 594 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 595 {
 596 #ifdef CONFIG_MEMCG
 597         mm->owner = p;
 598 #endif
 599 }
 600
 601 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 602 {
 603         mm->mmap = NULL;
 604         mm->mm_rb = RB_ROOT;
 605         mm->vmacache_seqnum = 0;
 606         atomic_set(&mm->mm_users, 1);
 607         atomic_set(&mm->mm_count, 1);
 608         init_rwsem(&mm->mmap_sem);
 609         INIT_LIST_HEAD(&mm->mmlist);
 610         mm->core_state = NULL;
 611         atomic_long_set(&mm->nr_ptes, 0);
 612         mm_nr_pmds_init(mm);
 613         mm->map_count = 0;
 614         mm->locked_vm = 0;
 615         mm->pinned_vm = 0;
 616         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 617         spin_lock_init(&mm->page_table_lock);
 618         mm_init_cpumask(mm);
 619         mm_init_aio(mm);
 620         mm_init_owner(mm, p);
 621         mmu_notifier_mm_init(mm);
 622         clear_tlb_flush_pending(mm);
 623 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 624         mm->pmd_huge_pte = NULL;
 625 #endif
 626
 627         if (current->mm) {
 628                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 629                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 630         } else {
 631                 mm->flags = default_dump_filter;
 632                 mm->def_flags = 0;
 633         }
 634
 635         if (mm_alloc_pgd(mm))
 636                 goto fail_nopgd;
 637
 638         if (init_new_context(p, mm))
 639                 goto fail_nocontext;
 640
 641         return mm;
 642
 643 fail_nocontext:
 644         mm_free_pgd(mm);
 645 fail_nopgd:
 646         free_mm(mm);
 647         return NULL;
 648 }
 649
 650 static void check_mm(struct mm_struct *mm)
 651 {
 652         int i;
 653
 654         for (i = 0; i < NR_MM_COUNTERS; i++) {
 655                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 656
 657                 if (unlikely(x))
 658                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 659                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 660         }
 661
 662         if (atomic_long_read(&mm->nr_ptes))
 663                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 664                                 atomic_long_read(&mm->nr_ptes));
 665         if (mm_nr_pmds(mm))
 666                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 667                                 mm_nr_pmds(mm));
 668
 669 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 670         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 671 #endif
 672 }
 673
 674 /*
 675  * Allocate and initialize an mm_struct.
 676  */
 677 struct mm_struct *mm_alloc(void)
 678 {
 679         struct mm_struct *mm;
 680
 681         mm = allocate_mm();
 682         if (!mm)
 683                 return NULL;
 684
 685         memset(mm, 0, sizeof(*mm));
 686         return mm_init(mm, current);
 687 }
 688
 689 /*
 690  * Called when the last reference to the mm
 691  * is dropped: either by a lazy thread or by
 692  * mmput. Free the page directory and the mm.
 693  */
 694 void __mmdrop(struct mm_struct *mm)
 695 {
 696         BUG_ON(mm == &init_mm);
 697         mm_free_pgd(mm);
 698         destroy_context(mm);
 699         mmu_notifier_mm_destroy(mm);
 700         check_mm(mm);
 701         free_mm(mm);
 702 }
 703 EXPORT_SYMBOL_GPL(__mmdrop);
 704
 705 static inline void __mmput(struct mm_struct *mm)
 706 {
 707         VM_BUG_ON(atomic_read(&mm->mm_users));
 708
 709         uprobe_clear_state(mm);
 710         exit_aio(mm);
 711         ksm_exit(mm);
 712         khugepaged_exit(mm); /* must run before exit_mmap */
 713         exit_mmap(mm);
 714         set_mm_exe_file(mm, NULL);
 715         if (!list_empty(&mm->mmlist)) {
 716                 spin_lock(&mmlist_lock);
 717                 list_del(&mm->mmlist);
 718                 spin_unlock(&mmlist_lock);
 719         }
 720         if (mm->binfmt)
 721                 module_put(mm->binfmt->module);
 722         mmdrop(mm);
 723 }
 724
 725 /*
 726  * Decrement the use count and release all resources for an mm.
 727  */
 728 void mmput(struct mm_struct *mm)
 729 {
 730         might_sleep();
 731
 732         if (atomic_dec_and_test(&mm->mm_users))
 733                 __mmput(mm);
 734 }
 735 EXPORT_SYMBOL_GPL(mmput);
 736
 737 #ifdef CONFIG_MMU
 738 static void mmput_async_fn(struct work_struct *work)
 739 {
 740         struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
 741         __mmput(mm);
 742 }
 743
 744 void mmput_async(struct mm_struct *mm)
 745 {
 746         if (atomic_dec_and_test(&mm->mm_users)) {
 747                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
 748                 schedule_work(&mm->async_put_work);
 749         }
 750 }
 751 #endif
 752
 753 /**
 754  * set_mm_exe_file - change a reference to the mm's executable file
 755  *
 756  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 757  *
 758  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 759  * invocations: in mmput() nobody alive left, in execve task is single
 760  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 761  * mm->exe_file, but does so without using set_mm_exe_file() in order
 762  * to do avoid the need for any locks.
 763  */
 764 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 765 {
 766         struct file *old_exe_file;
 767
 768         /*
 769          * It is safe to dereference the exe_file without RCU as
 770          * this function is only called if nobody else can access
 771          * this mm -- see comment above for justification.
 772          */
 773         old_exe_file = rcu_dereference_raw(mm->exe_file);
 774
 775         if (new_exe_file)
 776                 get_file(new_exe_file);
 777         rcu_assign_pointer(mm->exe_file, new_exe_file);
 778         if (old_exe_file)
 779                 fput(old_exe_file);
 780 }
 781
 782 /**
 783  * get_mm_exe_file - acquire a reference to the mm's executable file
 784  *
 785  * Returns %NULL if mm has no associated executable file.
 786  * User must release file via fput().
 787  */
 788 struct file *get_mm_exe_file(struct mm_struct *mm)
 789 {
 790         struct file *exe_file;
 791
 792         rcu_read_lock();
 793         exe_file = rcu_dereference(mm->exe_file);
 794         if (exe_file && !get_file_rcu(exe_file))
 795                 exe_file = NULL;
 796         rcu_read_unlock();
 797         return exe_file;
 798 }
 799 EXPORT_SYMBOL(get_mm_exe_file);
 800
 801 /**
 802  * get_task_exe_file - acquire a reference to the task's executable file
 803  *
 804  * Returns %NULL if task's mm (if any) has no associated executable file or
 805  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
 806  * User must release file via fput().
 807  */
 808 struct file *get_task_exe_file(struct task_struct *task)
 809 {
 810         struct file *exe_file = NULL;
 811         struct mm_struct *mm;
 812
 813         task_lock(task);
 814         mm = task->mm;
 815         if (mm) {
 816                 if (!(task->flags & PF_KTHREAD))
 817                         exe_file = get_mm_exe_file(mm);
 818         }
 819         task_unlock(task);
 820         return exe_file;
 821 }
 822 EXPORT_SYMBOL(get_task_exe_file);
 823
 824 /**
 825  * get_task_mm - acquire a reference to the task's mm
 826  *
 827  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 828  * this kernel workthread has transiently adopted a user mm with use_mm,
 829  * to do its AIO) is not set and if so returns a reference to it, after
 830  * bumping up the use count.  User must release the mm via mmput()
 831  * after use.  Typically used by /proc and ptrace.
 832  */
 833 struct mm_struct *get_task_mm(struct task_struct *task)
 834 {
 835         struct mm_struct *mm;
 836
 837         task_lock(task);
 838         mm = task->mm;
 839         if (mm) {
 840                 if (task->flags & PF_KTHREAD)
 841                         mm = NULL;
 842                 else
 843                         atomic_inc(&mm->mm_users);
 844         }
 845         task_unlock(task);
 846         return mm;
 847 }
 848 EXPORT_SYMBOL_GPL(get_task_mm);
 849
 850 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 851 {
 852         struct mm_struct *mm;
 853         int err;
 854
 855         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 856         if (err)
 857                 return ERR_PTR(err);
 858
 859         mm = get_task_mm(task);
 860         if (mm && mm != current->mm &&
 861                         !ptrace_may_access(task, mode)) {
 862                 mmput(mm);
 863                 mm = ERR_PTR(-EACCES);
 864         }
 865         mutex_unlock(&task->signal->cred_guard_mutex);
 866
 867         return mm;
 868 }
 869
 870 static void complete_vfork_done(struct task_struct *tsk)
 871 {
 872         struct completion *vfork;
 873
 874         task_lock(tsk);
 875         vfork = tsk->vfork_done;
 876         if (likely(vfork)) {
 877                 tsk->vfork_done = NULL;
 878                 complete(vfork);
 879         }
 880         task_unlock(tsk);
 881 }
 882
 883 static int wait_for_vfork_done(struct task_struct *child,
 884                                 struct completion *vfork)
 885 {
 886         int killed;
 887
 888         freezer_do_not_count();
 889         killed = wait_for_completion_killable(vfork);
 890         freezer_count();
 891
 892         if (killed) {
 893                 task_lock(child);
 894                 child->vfork_done = NULL;
 895                 task_unlock(child);
 896         }
 897
 898         put_task_struct(child);
 899         return killed;
 900 }
 901
 902 /* Please note the differences between mmput and mm_release.
 903  * mmput is called whenever we stop holding onto a mm_struct,
 904  * error success whatever.
 905  *
 906  * mm_release is called after a mm_struct has been removed
 907  * from the current process.
 908  *
 909  * This difference is important for error handling, when we
 910  * only half set up a mm_struct for a new process and need to restore
 911  * the old one.  Because we mmput the new mm_struct before
 912  * restoring the old one. . .
 913  * Eric Biederman 10 January 1998
 914  */
 915 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 916 {
 917         /* Get rid of any futexes when releasing the mm */
 918 #ifdef CONFIG_FUTEX
 919         if (unlikely(tsk->robust_list)) {
 920                 exit_robust_list(tsk);
 921                 tsk->robust_list = NULL;
 922         }
 923 #ifdef CONFIG_COMPAT
 924         if (unlikely(tsk->compat_robust_list)) {
 925                 compat_exit_robust_list(tsk);
 926                 tsk->compat_robust_list = NULL;
 927         }
 928 #endif
 929         if (unlikely(!list_empty(&tsk->pi_state_list)))
 930                 exit_pi_state_list(tsk);
 931 #endif
 932
 933         uprobe_free_utask(tsk);
 934
 935         /* Get rid of any cached register state */
 936         deactivate_mm(tsk, mm);
 937
 938         /*
 939          * If we're exiting normally, clear a user-space tid field if
 940          * requested.  We leave this alone when dying by signal, to leave
 941          * the value intact in a core dump, and to save the unnecessary
 942          * trouble, say, a killed vfork parent shouldn't touch this mm.
 943          * Userland only wants this done for a sys_exit.
 944          */
 945         if (tsk->clear_child_tid) {
 946                 if (!(tsk->flags & PF_SIGNALED) &&
 947                     atomic_read(&mm->mm_users) > 1) {
 948                         /*
 949                          * We don't check the error code - if userspace has
 950                          * not set up a proper pointer then tough luck.
 951                          */
 952                         put_user(0, tsk->clear_child_tid);
 953                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 954                                         1, NULL, NULL, 0);
 955                 }
 956                 tsk->clear_child_tid = NULL;
 957         }
 958
 959         /*
 960          * All done, finally we can wake up parent and return this mm to him.
 961          * Also kthread_stop() uses this completion for synchronization.
 962          */
 963         if (tsk->vfork_done)
 964                 complete_vfork_done(tsk);
 965 }
 966
 967 /*
 968  * Allocate a new mm structure and copy contents from the
 969  * mm structure of the passed in task structure.
 970  */
 971 static struct mm_struct *dup_mm(struct task_struct *tsk)
 972 {
 973         struct mm_struct *mm, *oldmm = current->mm;
 974         int err;
 975
 976         mm = allocate_mm();
 977         if (!mm)
 978                 goto fail_nomem;
 979
 980         memcpy(mm, oldmm, sizeof(*mm));
 981
 982         if (!mm_init(mm, tsk))
 983                 goto fail_nomem;
 984
 985         err = dup_mmap(mm, oldmm);
 986         if (err)
 987                 goto free_pt;
 988
 989         mm->hiwater_rss = get_mm_rss(mm);
 990         mm->hiwater_vm = mm->total_vm;
 991
 992         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 993                 goto free_pt;
 994
 995         return mm;
 996
 997 free_pt:
 998         /* don't put binfmt in mmput, we haven't got module yet */
 999         mm->binfmt = NULL;
1000         mmput(mm);
1001
1002 fail_nomem:
1003         return NULL;
1004 }
1005
1006 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1007 {
1008         struct mm_struct *mm, *oldmm;
1009         int retval;
1010
1011         tsk->min_flt = tsk->maj_flt = 0;
1012         tsk->nvcsw = tsk->nivcsw = 0;
1013 #ifdef CONFIG_DETECT_HUNG_TASK
1014         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1015 #endif
1016
1017         tsk->mm = NULL;
1018         tsk->active_mm = NULL;
1019
1020         /*
1021          * Are we cloning a kernel thread?
1022          *
1023          * We need to steal a active VM for that..
1024          */
1025         oldmm = current->mm;
1026         if (!oldmm)
1027                 return 0;
1028
1029         /* initialize the new vmacache entries */
1030         vmacache_flush(tsk);
1031
1032         if (clone_flags & CLONE_VM) {
1033                 atomic_inc(&oldmm->mm_users);
1034                 mm = oldmm;
1035                 goto good_mm;
1036         }
1037
1038         retval = -ENOMEM;
1039         mm = dup_mm(tsk);
1040         if (!mm)
1041                 goto fail_nomem;
1042
1043 good_mm:
1044         tsk->mm = mm;
1045         tsk->active_mm = mm;
1046         return 0;
1047
1048 fail_nomem:
1049         return retval;
1050 }
1051
1052 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1053 {
1054         struct fs_struct *fs = current->fs;
1055         if (clone_flags & CLONE_FS) {
1056                 /* tsk->fs is already what we want */
1057                 spin_lock(&fs->lock);
1058                 if (fs->in_exec) {
1059                         spin_unlock(&fs->lock);
1060                         return -EAGAIN;
1061                 }
1062                 fs->users++;
1063                 spin_unlock(&fs->lock);
1064                 return 0;
1065         }
1066         tsk->fs = copy_fs_struct(fs);
1067         if (!tsk->fs)
1068                 return -ENOMEM;
1069         return 0;
1070 }
1071
1072 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1073 {
1074         struct files_struct *oldf, *newf;
1075         int error = 0;
1076
1077         /*
1078          * A background process may not have any files ...
1079          */
1080         oldf = current->files;
1081         if (!oldf)
1082                 goto out;
1083
1084         if (clone_flags & CLONE_FILES) {
1085                 atomic_inc(&oldf->count);
1086                 goto out;
1087         }
1088
1089         newf = dup_fd(oldf, &error);
1090         if (!newf)
1091                 goto out;
1092
1093         tsk->files = newf;
1094         error = 0;
1095 out:
1096         return error;
1097 }
1098
1099 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1100 {
1101 #ifdef CONFIG_BLOCK
1102         struct io_context *ioc = current->io_context;
1103         struct io_context *new_ioc;
1104
1105         if (!ioc)
1106                 return 0;
1107         /*
1108          * Share io context with parent, if CLONE_IO is set
1109          */
1110         if (clone_flags & CLONE_IO) {
1111                 ioc_task_link(ioc);
1112                 tsk->io_context = ioc;
1113         } else if (ioprio_valid(ioc->ioprio)) {
1114                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1115                 if (unlikely(!new_ioc))
1116                         return -ENOMEM;
1117
1118                 new_ioc->ioprio = ioc->ioprio;
1119                 put_io_context(new_ioc);
1120         }
1121 #endif
1122         return 0;
1123 }
1124
1125 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1126 {
1127         struct sighand_struct *sig;
1128
1129         if (clone_flags & CLONE_SIGHAND) {
1130                 atomic_inc(&current->sighand->count);
1131                 return 0;
1132         }
1133         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1134         rcu_assign_pointer(tsk->sighand, sig);
1135         if (!sig)
1136                 return -ENOMEM;
1137
1138         atomic_set(&sig->count, 1);
1139         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1140         return 0;
1141 }
1142
1143 void __cleanup_sighand(struct sighand_struct *sighand)
1144 {
1145         if (atomic_dec_and_test(&sighand->count)) {
1146                 signalfd_cleanup(sighand);
1147                 /*
1148                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1149                  * without an RCU grace period, see __lock_task_sighand().
1150                  */
1151                 kmem_cache_free(sighand_cachep, sighand);
1152         }
1153 }
1154
1155 /*
1156  * Initialize POSIX timer handling for a thread group.
1157  */
1158 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1159 {
1160         unsigned long cpu_limit;
1161
1162         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1163         if (cpu_limit != RLIM_INFINITY) {
1164                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1165                 sig->cputimer.running = true;
1166         }
1167
1168         /* The timer lists. */
1169         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1170         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1171         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1172 }
1173
1174 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1175 {
1176         struct signal_struct *sig;
1177
1178         if (clone_flags & CLONE_THREAD)
1179                 return 0;
1180
1181         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1182         tsk->signal = sig;
1183         if (!sig)
1184                 return -ENOMEM;
1185
1186         sig->nr_threads = 1;
1187         atomic_set(&sig->live, 1);
1188         atomic_set(&sig->sigcnt, 1);
1189
1190         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1191         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1192         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1193
1194         init_waitqueue_head(&sig->wait_chldexit);
1195         sig->curr_target = tsk;
1196         init_sigpending(&sig->shared_pending);
1197         INIT_LIST_HEAD(&sig->posix_timers);
1198         seqlock_init(&sig->stats_lock);
1199         prev_cputime_init(&sig->prev_cputime);
1200
1201         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1202         sig->real_timer.function = it_real_fn;
1203
1204         task_lock(current->group_leader);
1205         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1206         task_unlock(current->group_leader);
1207
1208         posix_cpu_timers_init_group(sig);
1209
1210         tty_audit_fork(sig);
1211         sched_autogroup_fork(sig);
1212
1213         sig->oom_score_adj = current->signal->oom_score_adj;
1214         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1215
1216         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1217                                    current->signal->is_child_subreaper;
1218
1219         mutex_init(&sig->cred_guard_mutex);
1220
1221         return 0;
1222 }
1223
1224 static void copy_seccomp(struct task_struct *p)
1225 {
1226 #ifdef CONFIG_SECCOMP
1227         /*
1228          * Must be called with sighand->lock held, which is common to
1229          * all threads in the group. Holding cred_guard_mutex is not
1230          * needed because this new task is not yet running and cannot
1231          * be racing exec.
1232          */
1233         assert_spin_locked(&current->sighand->siglock);
1234
1235         /* Ref-count the new filter user, and assign it. */
1236         get_seccomp_filter(current);
1237         p->seccomp = current->seccomp;
1238
1239         /*
1240          * Explicitly enable no_new_privs here in case it got set
1241          * between the task_struct being duplicated and holding the
1242          * sighand lock. The seccomp state and nnp must be in sync.
1243          */
1244         if (task_no_new_privs(current))
1245                 task_set_no_new_privs(p);
1246
1247         /*
1248          * If the parent gained a seccomp mode after copying thread
1249          * flags and between before we held the sighand lock, we have
1250          * to manually enable the seccomp thread flag here.
1251          */
1252         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1253                 set_tsk_thread_flag(p, TIF_SECCOMP);
1254 #endif
1255 }
1256
1257 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1258 {
1259         current->clear_child_tid = tidptr;
1260
1261         return task_pid_vnr(current);
1262 }
1263
1264 static void rt_mutex_init_task(struct task_struct *p)
1265 {
1266         raw_spin_lock_init(&p->pi_lock);
1267 #ifdef CONFIG_RT_MUTEXES
1268         p->pi_waiters = RB_ROOT;
1269         p->pi_waiters_leftmost = NULL;
1270         p->pi_blocked_on = NULL;
1271 #endif
1272 }
1273
1274 /*
1275  * Initialize POSIX timer handling for a single task.
1276  */
1277 static void posix_cpu_timers_init(struct task_struct *tsk)
1278 {
1279         tsk->cputime_expires.prof_exp = 0;
1280         tsk->cputime_expires.virt_exp = 0;
1281         tsk->cputime_expires.sched_exp = 0;
1282         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1283         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1284         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1285 }
1286
1287 static inline void
1288 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1289 {
1290          task->pids[type].pid = pid;
1291 }
1292
1293 /*
1294  * This creates a new process as a copy of the old one,
1295  * but does not actually start it yet.
1296  *
1297  * It copies the registers, and all the appropriate
1298  * parts of the process environment (as per the clone
1299  * flags). The actual kick-off is left to the caller.
1300  */
1301 static struct task_struct *copy_process(unsigned long clone_flags,
1302                                         unsigned long stack_start,
1303                                         unsigned long stack_size,
1304                                         int __user *child_tidptr,
1305                                         struct pid *pid,
1306                                         int trace,
1307                                         unsigned long tls,
1308                                         int node)
1309 {
1310         int retval;
1311         struct task_struct *p;
1312
1313         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1314                 return ERR_PTR(-EINVAL);
1315
1316         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1317                 return ERR_PTR(-EINVAL);
1318
1319         /*
1320          * Thread groups must share signals as well, and detached threads
1321          * can only be started up within the thread group.
1322          */
1323         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1324                 return ERR_PTR(-EINVAL);
1325
1326         /*
1327          * Shared signal handlers imply shared VM. By way of the above,
1328          * thread groups also imply shared VM. Blocking this case allows
1329          * for various simplifications in other code.
1330          */
1331         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1332                 return ERR_PTR(-EINVAL);
1333
1334         /*
1335          * Siblings of global init remain as zombies on exit since they are
1336          * not reaped by their parent (swapper). To solve this and to avoid
1337          * multi-rooted process trees, prevent global and container-inits
1338          * from creating siblings.
1339          */
1340         if ((clone_flags & CLONE_PARENT) &&
1341                                 current->signal->flags & SIGNAL_UNKILLABLE)
1342                 return ERR_PTR(-EINVAL);
1343
1344         /*
1345          * If the new process will be in a different pid or user namespace
1346          * do not allow it to share a thread group with the forking task.
1347          */
1348         if (clone_flags & CLONE_THREAD) {
1349                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1350                     (task_active_pid_ns(current) !=
1351                                 current->nsproxy->pid_ns_for_children))
1352                         return ERR_PTR(-EINVAL);
1353         }
1354
1355         retval = security_task_create(clone_flags);
1356         if (retval)
1357                 goto fork_out;
1358
1359         retval = -ENOMEM;
1360         p = dup_task_struct(current, node);
1361         if (!p)
1362                 goto fork_out;
1363
1364         ftrace_graph_init_task(p);
1365
1366         rt_mutex_init_task(p);
1367
1368 #ifdef CONFIG_PROVE_LOCKING
1369         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1370         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1371 #endif
1372         retval = -EAGAIN;
1373         if (atomic_read(&p->real_cred->user->processes) >=
1374                         task_rlimit(p, RLIMIT_NPROC)) {
1375                 if (p->real_cred->user != INIT_USER &&
1376                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1377                         goto bad_fork_free;
1378         }
1379         current->flags &= ~PF_NPROC_EXCEEDED;
1380
1381         retval = copy_creds(p, clone_flags);
1382         if (retval < 0)
1383                 goto bad_fork_free;
1384
1385         /*
1386          * If multiple threads are within copy_process(), then this check
1387          * triggers too late. This doesn't hurt, the check is only there
1388          * to stop root fork bombs.
1389          */
1390         retval = -EAGAIN;
1391         if (nr_threads >= max_threads)
1392                 goto bad_fork_cleanup_count;
1393
1394         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1395         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1396         p->flags |= PF_FORKNOEXEC;
1397         INIT_LIST_HEAD(&p->children);
1398         INIT_LIST_HEAD(&p->sibling);
1399         rcu_copy_process(p);
1400         p->vfork_done = NULL;
1401         spin_lock_init(&p->alloc_lock);
1402
1403         init_sigpending(&p->pending);
1404
1405         p->utime = p->stime = p->gtime = 0;
1406         p->utimescaled = p->stimescaled = 0;
1407         prev_cputime_init(&p->prev_cputime);
1408
1409 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1410         seqcount_init(&p->vtime_seqcount);
1411         p->vtime_snap = 0;
1412         p->vtime_snap_whence = VTIME_INACTIVE;
1413 #endif
1414
1415 #if defined(SPLIT_RSS_COUNTING)
1416         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1417 #endif
1418
1419         p->default_timer_slack_ns = current->timer_slack_ns;
1420
1421         task_io_accounting_init(&p->ioac);
1422         acct_clear_integrals(p);
1423
1424         posix_cpu_timers_init(p);
1425
1426         p->start_time = ktime_get_ns();
1427         p->real_start_time = ktime_get_boot_ns();
1428         p->io_context = NULL;
1429         p->audit_context = NULL;
1430         cgroup_fork(p);
1431 #ifdef CONFIG_NUMA
1432         p->mempolicy = mpol_dup(p->mempolicy);
1433         if (IS_ERR(p->mempolicy)) {
1434                 retval = PTR_ERR(p->mempolicy);
1435                 p->mempolicy = NULL;
1436                 goto bad_fork_cleanup_threadgroup_lock;
1437         }
1438 #endif
1439 #ifdef CONFIG_CPUSETS
1440         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1441         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1442         seqcount_init(&p->mems_allowed_seq);
1443 #endif
1444 #ifdef CONFIG_TRACE_IRQFLAGS
1445         p->irq_events = 0;
1446         p->hardirqs_enabled = 0;
1447         p->hardirq_enable_ip = 0;
1448         p->hardirq_enable_event = 0;
1449         p->hardirq_disable_ip = _THIS_IP_;
1450         p->hardirq_disable_event = 0;
1451         p->softirqs_enabled = 1;
1452         p->softirq_enable_ip = _THIS_IP_;
1453         p->softirq_enable_event = 0;
1454         p->softirq_disable_ip = 0;
1455         p->softirq_disable_event = 0;
1456         p->hardirq_context = 0;
1457         p->softirq_context = 0;
1458 #endif
1459
1460         p->pagefault_disabled = 0;
1461
1462 #ifdef CONFIG_LOCKDEP
1463         p->lockdep_depth = 0; /* no locks held yet */
1464         p->curr_chain_key = 0;
1465         p->lockdep_recursion = 0;
1466 #endif
1467
1468 #ifdef CONFIG_DEBUG_MUTEXES
1469         p->blocked_on = NULL; /* not blocked yet */
1470 #endif
1471 #ifdef CONFIG_BCACHE
1472         p->sequential_io        = 0;
1473         p->sequential_io_avg    = 0;
1474 #endif
1475
1476         /* Perform scheduler related setup. Assign this task to a CPU. */
1477         retval = sched_fork(clone_flags, p);
1478         if (retval)
1479                 goto bad_fork_cleanup_policy;
1480
1481         retval = perf_event_init_task(p);
1482         if (retval)
1483                 goto bad_fork_cleanup_policy;
1484         retval = audit_alloc(p);
1485         if (retval)
1486                 goto bad_fork_cleanup_perf;
1487         /* copy all the process information */
1488         shm_init_task(p);
1489         retval = copy_semundo(clone_flags, p);
1490         if (retval)
1491                 goto bad_fork_cleanup_audit;
1492         retval = copy_files(clone_flags, p);
1493         if (retval)
1494                 goto bad_fork_cleanup_semundo;
1495         retval = copy_fs(clone_flags, p);
1496         if (retval)
1497                 goto bad_fork_cleanup_files;
1498         retval = copy_sighand(clone_flags, p);
1499         if (retval)
1500                 goto bad_fork_cleanup_fs;
1501         retval = copy_signal(clone_flags, p);
1502         if (retval)
1503                 goto bad_fork_cleanup_sighand;
1504         retval = copy_mm(clone_flags, p);
1505         if (retval)
1506                 goto bad_fork_cleanup_signal;
1507         retval = copy_namespaces(clone_flags, p);
1508         if (retval)
1509                 goto bad_fork_cleanup_mm;
1510         retval = copy_io(clone_flags, p);
1511         if (retval)
1512                 goto bad_fork_cleanup_namespaces;
1513         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1514         if (retval)
1515                 goto bad_fork_cleanup_io;
1516
1517         if (pid != &init_struct_pid) {
1518                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1519                 if (IS_ERR(pid)) {
1520                         retval = PTR_ERR(pid);
1521                         goto bad_fork_cleanup_thread;
1522                 }
1523         }
1524
1525         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1526         /*
1527          * Clear TID on mm_release()?
1528          */
1529         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1530 #ifdef CONFIG_BLOCK
1531         p->plug = NULL;
1532 #endif
1533 #ifdef CONFIG_FUTEX
1534         p->robust_list = NULL;
1535 #ifdef CONFIG_COMPAT
1536         p->compat_robust_list = NULL;
1537 #endif
1538         INIT_LIST_HEAD(&p->pi_state_list);
1539         p->pi_state_cache = NULL;
1540 #endif
1541         /*
1542          * sigaltstack should be cleared when sharing the same VM
1543          */
1544         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1545                 sas_ss_reset(p);
1546
1547         /*
1548          * Syscall tracing and stepping should be turned off in the
1549          * child regardless of CLONE_PTRACE.
1550          */
1551         user_disable_single_step(p);
1552         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1553 #ifdef TIF_SYSCALL_EMU
1554         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1555 #endif
1556         clear_all_latency_tracing(p);
1557
1558         /* ok, now we should be set up.. */
1559         p->pid = pid_nr(pid);
1560         if (clone_flags & CLONE_THREAD) {
1561                 p->exit_signal = -1;
1562                 p->group_leader = current->group_leader;
1563                 p->tgid = current->tgid;
1564         } else {
1565                 if (clone_flags & CLONE_PARENT)
1566                         p->exit_signal = current->group_leader->exit_signal;
1567                 else
1568                         p->exit_signal = (clone_flags & CSIGNAL);
1569                 p->group_leader = p;
1570                 p->tgid = p->pid;
1571         }
1572
1573         p->nr_dirtied = 0;
1574         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1575         p->dirty_paused_when = 0;
1576
1577         p->pdeath_signal = 0;
1578         INIT_LIST_HEAD(&p->thread_group);
1579         p->task_works = NULL;
1580
1581         threadgroup_change_begin(current);
1582         /*
1583          * Ensure that the cgroup subsystem policies allow the new process to be
1584          * forked. It should be noted the the new process's css_set can be changed
1585          * between here and cgroup_post_fork() if an organisation operation is in
1586          * progress.
1587          */
1588         retval = cgroup_can_fork(p);
1589         if (retval)
1590                 goto bad_fork_free_pid;
1591
1592         /*
1593          * Make it visible to the rest of the system, but dont wake it up yet.
1594          * Need tasklist lock for parent etc handling!
1595          */
1596         write_lock_irq(&tasklist_lock);
1597
1598         /* CLONE_PARENT re-uses the old parent */
1599         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1600                 p->real_parent = current->real_parent;
1601                 p->parent_exec_id = current->parent_exec_id;
1602         } else {
1603                 p->real_parent = current;
1604                 p->parent_exec_id = current->self_exec_id;
1605         }
1606
1607         spin_lock(&current->sighand->siglock);
1608
1609         /*
1610          * Copy seccomp details explicitly here, in case they were changed
1611          * before holding sighand lock.
1612          */
1613         copy_seccomp(p);
1614
1615         /*
1616          * Process group and session signals need to be delivered to just the
1617          * parent before the fork or both the parent and the child after the
1618          * fork. Restart if a signal comes in before we add the new process to
1619          * it's process group.
1620          * A fatal signal pending means that current will exit, so the new
1621          * thread can't slip out of an OOM kill (or normal SIGKILL).
1622         */
1623         recalc_sigpending();
1624         if (signal_pending(current)) {
1625                 spin_unlock(&current->sighand->siglock);
1626                 write_unlock_irq(&tasklist_lock);
1627                 retval = -ERESTARTNOINTR;
1628                 goto bad_fork_cancel_cgroup;
1629         }
1630
1631         if (likely(p->pid)) {
1632                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1633
1634                 init_task_pid(p, PIDTYPE_PID, pid);
1635                 if (thread_group_leader(p)) {
1636                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1637                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1638
1639                         if (is_child_reaper(pid)) {
1640                                 ns_of_pid(pid)->child_reaper = p;
1641                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1642                         }
1643
1644                         p->signal->leader_pid = pid;
1645                         p->signal->tty = tty_kref_get(current->signal->tty);
1646                         list_add_tail(&p->sibling, &p->real_parent->children);
1647                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1648                         attach_pid(p, PIDTYPE_PGID);
1649                         attach_pid(p, PIDTYPE_SID);
1650                         __this_cpu_inc(process_counts);
1651                 } else {
1652                         current->signal->nr_threads++;
1653                         atomic_inc(&current->signal->live);
1654                         atomic_inc(&current->signal->sigcnt);
1655                         list_add_tail_rcu(&p->thread_group,
1656                                           &p->group_leader->thread_group);
1657                         list_add_tail_rcu(&p->thread_node,
1658                                           &p->signal->thread_head);
1659                 }
1660                 attach_pid(p, PIDTYPE_PID);
1661                 nr_threads++;
1662         }
1663
1664         total_forks++;
1665         spin_unlock(&current->sighand->siglock);
1666         syscall_tracepoint_update(p);
1667         write_unlock_irq(&tasklist_lock);
1668
1669         proc_fork_connector(p);
1670         cgroup_post_fork(p);
1671         threadgroup_change_end(current);
1672         perf_event_fork(p);
1673
1674         trace_task_newtask(p, clone_flags);
1675         uprobe_copy_process(p, clone_flags);
1676
1677         return p;
1678
1679 bad_fork_cancel_cgroup:
1680         cgroup_cancel_fork(p);
1681 bad_fork_free_pid:
1682         threadgroup_change_end(current);
1683         if (pid != &init_struct_pid)
1684                 free_pid(pid);
1685 bad_fork_cleanup_thread:
1686         exit_thread(p);
1687 bad_fork_cleanup_io:
1688         if (p->io_context)
1689                 exit_io_context(p);
1690 bad_fork_cleanup_namespaces:
1691         exit_task_namespaces(p);
1692 bad_fork_cleanup_mm:
1693         if (p->mm)
1694                 mmput(p->mm);
1695 bad_fork_cleanup_signal:
1696         if (!(clone_flags & CLONE_THREAD))
1697                 free_signal_struct(p->signal);
1698 bad_fork_cleanup_sighand:
1699         __cleanup_sighand(p->sighand);
1700 bad_fork_cleanup_fs:
1701         exit_fs(p); /* blocking */
1702 bad_fork_cleanup_files:
1703         exit_files(p); /* blocking */
1704 bad_fork_cleanup_semundo:
1705         exit_sem(p);
1706 bad_fork_cleanup_audit:
1707         audit_free(p);
1708 bad_fork_cleanup_perf:
1709         perf_event_free_task(p);
1710 bad_fork_cleanup_policy:
1711 #ifdef CONFIG_NUMA
1712         mpol_put(p->mempolicy);
1713 bad_fork_cleanup_threadgroup_lock:
1714 #endif
1715         delayacct_tsk_free(p);
1716 bad_fork_cleanup_count:
1717         atomic_dec(&p->cred->user->processes);
1718         exit_creds(p);
1719 bad_fork_free:
1720         free_task(p);
1721 fork_out:
1722         return ERR_PTR(retval);
1723 }
1724
1725 static inline void init_idle_pids(struct pid_link *links)
1726 {
1727         enum pid_type type;
1728
1729         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1730                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1731                 links[type].pid = &init_struct_pid;
1732         }
1733 }
1734
1735 struct task_struct *fork_idle(int cpu)
1736 {
1737         struct task_struct *task;
1738         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1739                             cpu_to_node(cpu));
1740         if (!IS_ERR(task)) {
1741                 init_idle_pids(task->pids);
1742                 init_idle(task, cpu);
1743         }
1744
1745         return task;
1746 }
1747
1748 /*
1749  *  Ok, this is the main fork-routine.
1750  *
1751  * It copies the process, and if successful kick-starts
1752  * it and waits for it to finish using the VM if required.
1753  */
1754 long _do_fork(unsigned long clone_flags,
1755               unsigned long stack_start,
1756               unsigned long stack_size,
1757               int __user *parent_tidptr,
1758               int __user *child_tidptr,
1759               unsigned long tls)
1760 {
1761         struct task_struct *p;
1762         int trace = 0;
1763         long nr;
1764
1765         /*
1766          * Determine whether and which event to report to ptracer.  When
1767          * called from kernel_thread or CLONE_UNTRACED is explicitly
1768          * requested, no event is reported; otherwise, report if the event
1769          * for the type of forking is enabled.
1770          */
1771         if (!(clone_flags & CLONE_UNTRACED)) {
1772                 if (clone_flags & CLONE_VFORK)
1773                         trace = PTRACE_EVENT_VFORK;
1774                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1775                         trace = PTRACE_EVENT_CLONE;
1776                 else
1777                         trace = PTRACE_EVENT_FORK;
1778
1779                 if (likely(!ptrace_event_enabled(current, trace)))
1780                         trace = 0;
1781         }
1782
1783         p = copy_process(clone_flags, stack_start, stack_size,
1784                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
1785         /*
1786          * Do this prior waking up the new thread - the thread pointer
1787          * might get invalid after that point, if the thread exits quickly.
1788          */
1789         if (!IS_ERR(p)) {
1790                 struct completion vfork;
1791                 struct pid *pid;
1792
1793                 trace_sched_process_fork(current, p);
1794
1795                 pid = get_task_pid(p, PIDTYPE_PID);
1796                 nr = pid_vnr(pid);
1797
1798                 if (clone_flags & CLONE_PARENT_SETTID)
1799                         put_user(nr, parent_tidptr);
1800
1801                 if (clone_flags & CLONE_VFORK) {
1802                         p->vfork_done = &vfork;
1803                         init_completion(&vfork);
1804                         get_task_struct(p);
1805                 }
1806
1807                 wake_up_new_task(p);
1808
1809                 /* forking complete and child started to run, tell ptracer */
1810                 if (unlikely(trace))
1811                         ptrace_event_pid(trace, pid);
1812
1813                 if (clone_flags & CLONE_VFORK) {
1814                         if (!wait_for_vfork_done(p, &vfork))
1815                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1816                 }
1817
1818                 put_pid(pid);
1819         } else {
1820                 nr = PTR_ERR(p);
1821         }
1822         return nr;
1823 }
1824
1825 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1826 /* For compatibility with architectures that call do_fork directly rather than
1827  * using the syscall entry points below. */
1828 long do_fork(unsigned long clone_flags,
1829               unsigned long stack_start,
1830               unsigned long stack_size,
1831               int __user *parent_tidptr,
1832               int __user *child_tidptr)
1833 {
1834         return _do_fork(clone_flags, stack_start, stack_size,
1835                         parent_tidptr, child_tidptr, 0);
1836 }
1837 #endif
1838
1839 /*
1840  * Create a kernel thread.
1841  */
1842 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1843 {
1844         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1845                 (unsigned long)arg, NULL, NULL, 0);
1846 }
1847
1848 #ifdef __ARCH_WANT_SYS_FORK
1849 SYSCALL_DEFINE0(fork)
1850 {
1851 #ifdef CONFIG_MMU
1852         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1853 #else
1854         /* can not support in nommu mode */
1855         return -EINVAL;
1856 #endif
1857 }
1858 #endif
1859
1860 #ifdef __ARCH_WANT_SYS_VFORK
1861 SYSCALL_DEFINE0(vfork)
1862 {
1863         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1864                         0, NULL, NULL, 0);
1865 }
1866 #endif
1867
1868 #ifdef __ARCH_WANT_SYS_CLONE
1869 #ifdef CONFIG_CLONE_BACKWARDS
1870 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1871                  int __user *, parent_tidptr,
1872                  unsigned long, tls,
1873                  int __user *, child_tidptr)
1874 #elif defined(CONFIG_CLONE_BACKWARDS2)
1875 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1876                  int __user *, parent_tidptr,
1877                  int __user *, child_tidptr,
1878                  unsigned long, tls)
1879 #elif defined(CONFIG_CLONE_BACKWARDS3)
1880 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1881                 int, stack_size,
1882                 int __user *, parent_tidptr,
1883                 int __user *, child_tidptr,
1884                 unsigned long, tls)
1885 #else
1886 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1887                  int __user *, parent_tidptr,
1888                  int __user *, child_tidptr,
1889                  unsigned long, tls)
1890 #endif
1891 {
1892         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
1893 }
1894 #endif
1895
1896 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1897 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1898 #endif
1899
1900 static void sighand_ctor(void *data)
1901 {
1902         struct sighand_struct *sighand = data;
1903
1904         spin_lock_init(&sighand->siglock);
1905         init_waitqueue_head(&sighand->signalfd_wqh);
1906 }
1907
1908 void __init proc_caches_init(void)
1909 {
1910         sighand_cachep = kmem_cache_create("sighand_cache",
1911                         sizeof(struct sighand_struct), 0,
1912                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1913                         SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
1914         signal_cachep = kmem_cache_create("signal_cache",
1915                         sizeof(struct signal_struct), 0,
1916                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1917                         NULL);
1918         files_cachep = kmem_cache_create("files_cache",
1919                         sizeof(struct files_struct), 0,
1920                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1921                         NULL);
1922         fs_cachep = kmem_cache_create("fs_cache",
1923                         sizeof(struct fs_struct), 0,
1924                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1925                         NULL);
1926         /*
1927          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1928          * whole struct cpumask for the OFFSTACK case. We could change
1929          * this to *only* allocate as much of it as required by the
1930          * maximum number of CPU's we can ever have.  The cpumask_allocation
1931          * is at the end of the structure, exactly for that reason.
1932          */
1933         mm_cachep = kmem_cache_create("mm_struct",
1934                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1935                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1936                         NULL);
1937         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
1938         mmap_init();
1939         nsproxy_cache_init();
1940 }
1941
1942 /*
1943  * Check constraints on flags passed to the unshare system call.
1944  */
1945 static int check_unshare_flags(unsigned long unshare_flags)
1946 {
1947         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1948                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1949                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1950                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
1951                 return -EINVAL;
1952         /*
1953          * Not implemented, but pretend it works if there is nothing
1954          * to unshare.  Note that unsharing the address space or the
1955          * signal handlers also need to unshare the signal queues (aka
1956          * CLONE_THREAD).
1957          */
1958         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1959                 if (!thread_group_empty(current))
1960                         return -EINVAL;
1961         }
1962         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
1963                 if (atomic_read(&current->sighand->count) > 1)
1964                         return -EINVAL;
1965         }
1966         if (unshare_flags & CLONE_VM) {
1967                 if (!current_is_single_threaded())
1968                         return -EINVAL;
1969         }
1970
1971         return 0;
1972 }
1973
1974 /*
1975  * Unshare the filesystem structure if it is being shared
1976  */
1977 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1978 {
1979         struct fs_struct *fs = current->fs;
1980
1981         if (!(unshare_flags & CLONE_FS) || !fs)
1982                 return 0;
1983
1984         /* don't need lock here; in the worst case we'll do useless copy */
1985         if (fs->users == 1)
1986                 return 0;
1987
1988         *new_fsp = copy_fs_struct(fs);
1989         if (!*new_fsp)
1990                 return -ENOMEM;
1991
1992         return 0;
1993 }
1994
1995 /*
1996  * Unshare file descriptor table if it is being shared
1997  */
1998 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1999 {
2000         struct files_struct *fd = current->files;
2001         int error = 0;
2002
2003         if ((unshare_flags & CLONE_FILES) &&
2004             (fd && atomic_read(&fd->count) > 1)) {
2005                 *new_fdp = dup_fd(fd, &error);
2006                 if (!*new_fdp)
2007                         return error;
2008         }
2009
2010         return 0;
2011 }
2012
2013 /*
2014  * unshare allows a process to 'unshare' part of the process
2015  * context which was originally shared using clone.  copy_*
2016  * functions used by do_fork() cannot be used here directly
2017  * because they modify an inactive task_struct that is being
2018  * constructed. Here we are modifying the current, active,
2019  * task_struct.
2020  */
2021 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2022 {
2023         struct fs_struct *fs, *new_fs = NULL;
2024         struct files_struct *fd, *new_fd = NULL;
2025         struct cred *new_cred = NULL;
2026         struct nsproxy *new_nsproxy = NULL;
2027         int do_sysvsem = 0;
2028         int err;
2029
2030         /*
2031          * If unsharing a user namespace must also unshare the thread group
2032          * and unshare the filesystem root and working directories.
2033          */
2034         if (unshare_flags & CLONE_NEWUSER)
2035                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2036         /*
2037          * If unsharing vm, must also unshare signal handlers.
2038          */
2039         if (unshare_flags & CLONE_VM)
2040                 unshare_flags |= CLONE_SIGHAND;
2041         /*
2042          * If unsharing a signal handlers, must also unshare the signal queues.
2043          */
2044         if (unshare_flags & CLONE_SIGHAND)
2045                 unshare_flags |= CLONE_THREAD;
2046         /*
2047          * If unsharing namespace, must also unshare filesystem information.
2048          */
2049         if (unshare_flags & CLONE_NEWNS)
2050                 unshare_flags |= CLONE_FS;
2051
2052         err = check_unshare_flags(unshare_flags);
2053         if (err)
2054                 goto bad_unshare_out;
2055         /*
2056          * CLONE_NEWIPC must also detach from the undolist: after switching
2057          * to a new ipc namespace, the semaphore arrays from the old
2058          * namespace are unreachable.
2059          */
2060         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2061                 do_sysvsem = 1;
2062         err = unshare_fs(unshare_flags, &new_fs);
2063         if (err)
2064                 goto bad_unshare_out;
2065         err = unshare_fd(unshare_flags, &new_fd);
2066         if (err)
2067                 goto bad_unshare_cleanup_fs;
2068         err = unshare_userns(unshare_flags, &new_cred);
2069         if (err)
2070                 goto bad_unshare_cleanup_fd;
2071         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2072                                          new_cred, new_fs);
2073         if (err)
2074                 goto bad_unshare_cleanup_cred;
2075
2076         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2077                 if (do_sysvsem) {
2078                         /*
2079                          * CLONE_SYSVSEM is equivalent to sys_exit().
2080                          */
2081                         exit_sem(current);
2082                 }
2083                 if (unshare_flags & CLONE_NEWIPC) {
2084                         /* Orphan segments in old ns (see sem above). */
2085                         exit_shm(current);
2086                         shm_init_task(current);
2087                 }
2088
2089                 if (new_nsproxy)
2090                         switch_task_namespaces(current, new_nsproxy);
2091
2092                 task_lock(current);
2093
2094                 if (new_fs) {
2095                         fs = current->fs;
2096                         spin_lock(&fs->lock);
2097                         current->fs = new_fs;
2098                         if (--fs->users)
2099                                 new_fs = NULL;
2100                         else
2101                                 new_fs = fs;
2102                         spin_unlock(&fs->lock);
2103                 }
2104
2105                 if (new_fd) {
2106                         fd = current->files;
2107                         current->files = new_fd;
2108                         new_fd = fd;
2109                 }
2110
2111                 task_unlock(current);
2112
2113                 if (new_cred) {
2114                         /* Install the new user namespace */
2115                         commit_creds(new_cred);
2116                         new_cred = NULL;
2117                 }
2118         }
2119
2120 bad_unshare_cleanup_cred:
2121         if (new_cred)
2122                 put_cred(new_cred);
2123 bad_unshare_cleanup_fd:
2124         if (new_fd)
2125                 put_files_struct(new_fd);
2126
2127 bad_unshare_cleanup_fs:
2128         if (new_fs)
2129                 free_fs_struct(new_fs);
2130
2131 bad_unshare_out:
2132         return err;
2133 }
2134
2135 /*
2136  *      Helper to unshare the files of the current task.
2137  *      We don't want to expose copy_files internals to
2138  *      the exec layer of the kernel.
2139  */
2140
2141 int unshare_files(struct files_struct **displaced)
2142 {
2143         struct task_struct *task = current;
2144         struct files_struct *copy = NULL;
2145         int error;
2146
2147         error = unshare_fd(CLONE_FILES, &copy);
2148         if (error || !copy) {
2149                 *displaced = NULL;
2150                 return error;
2151         }
2152         *displaced = task->files;
2153         task_lock(task);
2154         task->files = copy;
2155         task_unlock(task);
2156         return 0;
2157 }
2158
2159 int sysctl_max_threads(struct ctl_table *table, int write,
2160                        void __user *buffer, size_t *lenp, loff_t *ppos)
2161 {
2162         struct ctl_table t;
2163         int ret;
2164         int threads = max_threads;
2165         int min = MIN_THREADS;
2166         int max = MAX_THREADS;
2167
2168         t = *table;
2169         t.data = &threads;
2170         t.extra1 = &min;
2171         t.extra2 = &max;
2172
2173         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2174         if (ret || !write)
2175                 return ret;
2176
2177         set_max_threads(threads);
2178
2179         return 0;
2180 }