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