kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/mm.h>
  32 #include <linux/vmacache.h>
  33 #include <linux/nsproxy.h>
  34 #include <linux/capability.h>
  35 #include <linux/cpu.h>
  36 #include <linux/cgroup.h>
  37 #include <linux/security.h>
  38 #include <linux/hugetlb.h>
  39 #include <linux/seccomp.h>
  40 #include <linux/swap.h>
  41 #include <linux/syscalls.h>
  42 #include <linux/jiffies.h>
  43 #include <linux/futex.h>
  44 #include <linux/compat.h>
  45 #include <linux/kthread.h>
  46 #include <linux/task_io_accounting_ops.h>
  47 #include <linux/rcupdate.h>
  48 #include <linux/ptrace.h>
  49 #include <linux/mount.h>
  50 #include <linux/audit.h>
  51 #include <linux/memcontrol.h>
  52 #include <linux/ftrace.h>
  53 #include <linux/proc_fs.h>
  54 #include <linux/profile.h>
  55 #include <linux/rmap.h>
  56 #include <linux/ksm.h>
  57 #include <linux/acct.h>
  58 #include <linux/tsacct_kern.h>
  59 #include <linux/cn_proc.h>
  60 #include <linux/freezer.h>
  61 #include <linux/delayacct.h>
  62 #include <linux/taskstats_kern.h>
  63 #include <linux/random.h>
  64 #include <linux/tty.h>
  65 #include <linux/blkdev.h>
  66 #include <linux/fs_struct.h>
  67 #include <linux/magic.h>
  68 #include <linux/perf_event.h>
  69 #include <linux/posix-timers.h>
  70 #include <linux/user-return-notifier.h>
  71 #include <linux/oom.h>
  72 #include <linux/khugepaged.h>
  73 #include <linux/signalfd.h>
  74 #include <linux/uprobes.h>
  75 #include <linux/aio.h>
  76 #include <linux/compiler.h>
  77 #include <linux/sysctl.h>
  78 #include <linux/kcov.h>
  79
  80 #include <asm/pgtable.h>
  81 #include <asm/pgalloc.h>
  82 #include <asm/uaccess.h>
  83 #include <asm/mmu_context.h>
  84 #include <asm/cacheflush.h>
  85 #include <asm/tlbflush.h>
  86
  87 #include <trace/events/sched.h>
  88
  89 #define CREATE_TRACE_POINTS
  90 #include <trace/events/task.h>
  91
  92 /*
  93  * Minimum number of threads to boot the kernel
  94  */
  95 #define MIN_THREADS 20
  96
  97 /*
  98  * Maximum number of threads
  99  */
 100 #define MAX_THREADS FUTEX_TID_MASK
 101
 102 /*
 103  * Protected counters by write_lock_irq(&tasklist_lock)
 104  */
 105 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 106 int nr_threads;                 /* The idle threads do not count.. */
 107
 108 int max_threads;                /* tunable limit on nr_threads */
 109
 110 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 111
 112 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 113
 114 #ifdef CONFIG_PROVE_RCU
 115 int lockdep_tasklist_lock_is_held(void)
 116 {
 117         return lockdep_is_held(&tasklist_lock);
 118 }
 119 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 120 #endif /* #ifdef CONFIG_PROVE_RCU */
 121
 122 int nr_processes(void)
 123 {
 124         int cpu;
 125         int total = 0;
 126
 127         for_each_possible_cpu(cpu)
 128                 total += per_cpu(process_counts, cpu);
 129
 130         return total;
 131 }
 132
 133 void __weak arch_release_task_struct(struct task_struct *tsk)
 134 {
 135 }
 136
 137 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 138 static struct kmem_cache *task_struct_cachep;
 139
 140 static inline struct task_struct *alloc_task_struct_node(int node)
 141 {
 142         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 143 }
 144
 145 static inline void free_task_struct(struct task_struct *tsk)
 146 {
 147         kmem_cache_free(task_struct_cachep, tsk);
 148 }
 149 #endif
 150
 151 void __weak arch_release_thread_info(struct thread_info *ti)
 152 {
 153 }
 154
 155 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
 156
 157 /*
 158  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 159  * kmemcache based allocator.
 160  */
 161 # if THREAD_SIZE >= PAGE_SIZE
 162 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 163                                                   int node)
 164 {
 165         struct page *page = alloc_kmem_pages_node(node, THREADINFO_GFP,
 166                                                   THREAD_SIZE_ORDER);
 167
 168         if (page)
 169                 memcg_kmem_update_page_stat(page, MEMCG_KERNEL_STACK,
 170                                             1 << THREAD_SIZE_ORDER);
 171
 172         return page ? page_address(page) : NULL;
 173 }
 174
 175 static inline void free_thread_info(struct thread_info *ti)
 176 {
 177         struct page *page = virt_to_page(ti);
 178
 179         memcg_kmem_update_page_stat(page, MEMCG_KERNEL_STACK,
 180                                     -(1 << THREAD_SIZE_ORDER));
 181         __free_kmem_pages(page, THREAD_SIZE_ORDER);
 182 }
 183 # else
 184 static struct kmem_cache *thread_info_cache;
 185
 186 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
 187                                                   int node)
 188 {
 189         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
 190 }
 191
 192 static void free_thread_info(struct thread_info *ti)
 193 {
 194         kmem_cache_free(thread_info_cache, ti);
 195 }
 196
 197 void thread_info_cache_init(void)
 198 {
 199         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
 200                                               THREAD_SIZE, 0, NULL);
 201         BUG_ON(thread_info_cache == NULL);
 202 }
 203 # endif
 204 #endif
 205
 206 /* SLAB cache for signal_struct structures (tsk->signal) */
 207 static struct kmem_cache *signal_cachep;
 208
 209 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 210 struct kmem_cache *sighand_cachep;
 211
 212 /* SLAB cache for files_struct structures (tsk->files) */
 213 struct kmem_cache *files_cachep;
 214
 215 /* SLAB cache for fs_struct structures (tsk->fs) */
 216 struct kmem_cache *fs_cachep;
 217
 218 /* SLAB cache for vm_area_struct structures */
 219 struct kmem_cache *vm_area_cachep;
 220
 221 /* SLAB cache for mm_struct structures (tsk->mm) */
 222 static struct kmem_cache *mm_cachep;
 223
 224 static void account_kernel_stack(struct thread_info *ti, int account)
 225 {
 226         struct zone *zone = page_zone(virt_to_page(ti));
 227
 228         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
 229 }
 230
 231 void free_task(struct task_struct *tsk)
 232 {
 233         account_kernel_stack(tsk->stack, -1);
 234         arch_release_thread_info(tsk->stack);
 235         free_thread_info(tsk->stack);
 236         rt_mutex_debug_task_free(tsk);
 237         ftrace_graph_exit_task(tsk);
 238         put_seccomp_filter(tsk);
 239         arch_release_task_struct(tsk);
 240         free_task_struct(tsk);
 241 }
 242 EXPORT_SYMBOL(free_task);
 243
 244 static inline void free_signal_struct(struct signal_struct *sig)
 245 {
 246         taskstats_tgid_free(sig);
 247         sched_autogroup_exit(sig);
 248         kmem_cache_free(signal_cachep, sig);
 249 }
 250
 251 static inline void put_signal_struct(struct signal_struct *sig)
 252 {
 253         if (atomic_dec_and_test(&sig->sigcnt))
 254                 free_signal_struct(sig);
 255 }
 256
 257 void __put_task_struct(struct task_struct *tsk)
 258 {
 259         WARN_ON(!tsk->exit_state);
 260         WARN_ON(atomic_read(&tsk->usage));
 261         WARN_ON(tsk == current);
 262
 263         cgroup_free(tsk);
 264         task_numa_free(tsk);
 265         security_task_free(tsk);
 266         exit_creds(tsk);
 267         delayacct_tsk_free(tsk);
 268         put_signal_struct(tsk->signal);
 269
 270         if (!profile_handoff_task(tsk))
 271                 free_task(tsk);
 272 }
 273 EXPORT_SYMBOL_GPL(__put_task_struct);
 274
 275 void __init __weak arch_task_cache_init(void) { }
 276
 277 /*
 278  * set_max_threads
 279  */
 280 static void set_max_threads(unsigned int max_threads_suggested)
 281 {
 282         u64 threads;
 283
 284         /*
 285          * The number of threads shall be limited such that the thread
 286          * structures may only consume a small part of the available memory.
 287          */
 288         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 289                 threads = MAX_THREADS;
 290         else
 291                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 292                                     (u64) THREAD_SIZE * 8UL);
 293
 294         if (threads > max_threads_suggested)
 295                 threads = max_threads_suggested;
 296
 297         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 298 }
 299
 300 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 301 /* Initialized by the architecture: */
 302 int arch_task_struct_size __read_mostly;
 303 #endif
 304
 305 void __init fork_init(void)
 306 {
 307 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 308 #ifndef ARCH_MIN_TASKALIGN
 309 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 310 #endif
 311         /* create a slab on which task_structs can be allocated */
 312         task_struct_cachep = kmem_cache_create("task_struct",
 313                         arch_task_struct_size, ARCH_MIN_TASKALIGN,
 314                         SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
 315 #endif
 316
 317         /* do the arch specific task caches init */
 318         arch_task_cache_init();
 319
 320         set_max_threads(MAX_THREADS);
 321
 322         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 323         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 324         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 325                 init_task.signal->rlim[RLIMIT_NPROC];
 326 }
 327
 328 int __weak arch_dup_task_struct(struct task_struct *dst,
 329                                                struct task_struct *src)
 330 {
 331         *dst = *src;
 332         return 0;
 333 }
 334
 335 void set_task_stack_end_magic(struct task_struct *tsk)
 336 {
 337         unsigned long *stackend;
 338
 339         stackend = end_of_stack(tsk);
 340         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 341 }
 342
 343 static struct task_struct *dup_task_struct(struct task_struct *orig)
 344 {
 345         struct task_struct *tsk;
 346         struct thread_info *ti;
 347         int node = tsk_fork_get_node(orig);
 348         int err;
 349
 350         tsk = alloc_task_struct_node(node);
 351         if (!tsk)
 352                 return NULL;
 353
 354         ti = alloc_thread_info_node(tsk, node);
 355         if (!ti)
 356                 goto free_tsk;
 357
 358         err = arch_dup_task_struct(tsk, orig);
 359         if (err)
 360                 goto free_ti;
 361
 362         tsk->stack = ti;
 363 #ifdef CONFIG_SECCOMP
 364         /*
 365          * We must handle setting up seccomp filters once we're under
 366          * the sighand lock in case orig has changed between now and
 367          * then. Until then, filter must be NULL to avoid messing up
 368          * the usage counts on the error path calling free_task.
 369          */
 370         tsk->seccomp.filter = NULL;
 371 #endif
 372
 373         setup_thread_stack(tsk, orig);
 374         clear_user_return_notifier(tsk);
 375         clear_tsk_need_resched(tsk);
 376         set_task_stack_end_magic(tsk);
 377
 378 #ifdef CONFIG_CC_STACKPROTECTOR
 379         tsk->stack_canary = get_random_int();
 380 #endif
 381
 382         /*
 383          * One for us, one for whoever does the "release_task()" (usually
 384          * parent)
 385          */
 386         atomic_set(&tsk->usage, 2);
 387 #ifdef CONFIG_BLK_DEV_IO_TRACE
 388         tsk->btrace_seq = 0;
 389 #endif
 390         tsk->splice_pipe = NULL;
 391         tsk->task_frag.page = NULL;
 392         tsk->wake_q.next = NULL;
 393
 394         account_kernel_stack(ti, 1);
 395
 396         kcov_task_init(tsk);
 397
 398         return tsk;
 399
 400 free_ti:
 401         free_thread_info(ti);
 402 free_tsk:
 403         free_task_struct(tsk);
 404         return NULL;
 405 }
 406
 407 #ifdef CONFIG_MMU
 408 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 409 {
 410         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 411         struct rb_node **rb_link, *rb_parent;
 412         int retval;
 413         unsigned long charge;
 414
 415         uprobe_start_dup_mmap();
 416         down_write(&oldmm->mmap_sem);
 417         flush_cache_dup_mm(oldmm);
 418         uprobe_dup_mmap(oldmm, mm);
 419         /*
 420          * Not linked in yet - no deadlock potential:
 421          */
 422         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 423
 424         /* No ordering required: file already has been exposed. */
 425         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 426
 427         mm->total_vm = oldmm->total_vm;
 428         mm->data_vm = oldmm->data_vm;
 429         mm->exec_vm = oldmm->exec_vm;
 430         mm->stack_vm = oldmm->stack_vm;
 431
 432         rb_link = &mm->mm_rb.rb_node;
 433         rb_parent = NULL;
 434         pprev = &mm->mmap;
 435         retval = ksm_fork(mm, oldmm);
 436         if (retval)
 437                 goto out;
 438         retval = khugepaged_fork(mm, oldmm);
 439         if (retval)
 440                 goto out;
 441
 442         prev = NULL;
 443         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 444                 struct file *file;
 445
 446                 if (mpnt->vm_flags & VM_DONTCOPY) {
 447                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 448                         continue;
 449                 }
 450                 charge = 0;
 451                 if (mpnt->vm_flags & VM_ACCOUNT) {
 452                         unsigned long len = vma_pages(mpnt);
 453
 454                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 455                                 goto fail_nomem;
 456                         charge = len;
 457                 }
 458                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 459                 if (!tmp)
 460                         goto fail_nomem;
 461                 *tmp = *mpnt;
 462                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 463                 retval = vma_dup_policy(mpnt, tmp);
 464                 if (retval)
 465                         goto fail_nomem_policy;
 466                 tmp->vm_mm = mm;
 467                 if (anon_vma_fork(tmp, mpnt))
 468                         goto fail_nomem_anon_vma_fork;
 469                 tmp->vm_flags &=
 470                         ~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
 471                 tmp->vm_next = tmp->vm_prev = NULL;
 472                 tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
 473                 file = tmp->vm_file;
 474                 if (file) {
 475                         struct inode *inode = file_inode(file);
 476                         struct address_space *mapping = file->f_mapping;
 477
 478                         get_file(file);
 479                         if (tmp->vm_flags & VM_DENYWRITE)
 480                                 atomic_dec(&inode->i_writecount);
 481                         i_mmap_lock_write(mapping);
 482                         if (tmp->vm_flags & VM_SHARED)
 483                                 atomic_inc(&mapping->i_mmap_writable);
 484                         flush_dcache_mmap_lock(mapping);
 485                         /* insert tmp into the share list, just after mpnt */
 486                         vma_interval_tree_insert_after(tmp, mpnt,
 487                                         &mapping->i_mmap);
 488                         flush_dcache_mmap_unlock(mapping);
 489                         i_mmap_unlock_write(mapping);
 490                 }
 491
 492                 /*
 493                  * Clear hugetlb-related page reserves for children. This only
 494                  * affects MAP_PRIVATE mappings. Faults generated by the child
 495                  * are not guaranteed to succeed, even if read-only
 496                  */
 497                 if (is_vm_hugetlb_page(tmp))
 498                         reset_vma_resv_huge_pages(tmp);
 499
 500                 /*
 501                  * Link in the new vma and copy the page table entries.
 502                  */
 503                 *pprev = tmp;
 504                 pprev = &tmp->vm_next;
 505                 tmp->vm_prev = prev;
 506                 prev = tmp;
 507
 508                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 509                 rb_link = &tmp->vm_rb.rb_right;
 510                 rb_parent = &tmp->vm_rb;
 511
 512                 mm->map_count++;
 513                 retval = copy_page_range(mm, oldmm, mpnt);
 514
 515                 if (tmp->vm_ops && tmp->vm_ops->open)
 516                         tmp->vm_ops->open(tmp);
 517
 518                 if (retval)
 519                         goto out;
 520         }
 521         /* a new mm has just been created */
 522         arch_dup_mmap(oldmm, mm);
 523         retval = 0;
 524 out:
 525         up_write(&mm->mmap_sem);
 526         flush_tlb_mm(oldmm);
 527         up_write(&oldmm->mmap_sem);
 528         uprobe_end_dup_mmap();
 529         return retval;
 530 fail_nomem_anon_vma_fork:
 531         mpol_put(vma_policy(tmp));
 532 fail_nomem_policy:
 533         kmem_cache_free(vm_area_cachep, tmp);
 534 fail_nomem:
 535         retval = -ENOMEM;
 536         vm_unacct_memory(charge);
 537         goto out;
 538 }
 539
 540 static inline int mm_alloc_pgd(struct mm_struct *mm)
 541 {
 542         mm->pgd = pgd_alloc(mm);
 543         if (unlikely(!mm->pgd))
 544                 return -ENOMEM;
 545         return 0;
 546 }
 547
 548 static inline void mm_free_pgd(struct mm_struct *mm)
 549 {
 550         pgd_free(mm, mm->pgd);
 551 }
 552 #else
 553 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 554 {
 555         down_write(&oldmm->mmap_sem);
 556         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 557         up_write(&oldmm->mmap_sem);
 558         return 0;
 559 }
 560 #define mm_alloc_pgd(mm)        (0)
 561 #define mm_free_pgd(mm)
 562 #endif /* CONFIG_MMU */
 563
 564 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 565
 566 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 567 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 568
 569 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 570
 571 static int __init coredump_filter_setup(char *s)
 572 {
 573         default_dump_filter =
 574                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 575                 MMF_DUMP_FILTER_MASK;
 576         return 1;
 577 }
 578
 579 __setup("coredump_filter=", coredump_filter_setup);
 580
 581 #include <linux/init_task.h>
 582
 583 static void mm_init_aio(struct mm_struct *mm)
 584 {
 585 #ifdef CONFIG_AIO
 586         spin_lock_init(&mm->ioctx_lock);
 587         mm->ioctx_table = NULL;
 588 #endif
 589 }
 590
 591 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 592 {
 593 #ifdef CONFIG_MEMCG
 594         mm->owner = p;
 595 #endif
 596 }
 597
 598 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 599 {
 600         mm->mmap = NULL;
 601         mm->mm_rb = RB_ROOT;
 602         mm->vmacache_seqnum = 0;
 603         atomic_set(&mm->mm_users, 1);
 604         atomic_set(&mm->mm_count, 1);
 605         init_rwsem(&mm->mmap_sem);
 606         INIT_LIST_HEAD(&mm->mmlist);
 607         mm->core_state = NULL;
 608         atomic_long_set(&mm->nr_ptes, 0);
 609         mm_nr_pmds_init(mm);
 610         mm->map_count = 0;
 611         mm->locked_vm = 0;
 612         mm->pinned_vm = 0;
 613         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 614         spin_lock_init(&mm->page_table_lock);
 615         mm_init_cpumask(mm);
 616         mm_init_aio(mm);
 617         mm_init_owner(mm, p);
 618         mmu_notifier_mm_init(mm);
 619         clear_tlb_flush_pending(mm);
 620 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 621         mm->pmd_huge_pte = NULL;
 622 #endif
 623
 624         if (current->mm) {
 625                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 626                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 627         } else {
 628                 mm->flags = default_dump_filter;
 629                 mm->def_flags = 0;
 630         }
 631
 632         if (mm_alloc_pgd(mm))
 633                 goto fail_nopgd;
 634
 635         if (init_new_context(p, mm))
 636                 goto fail_nocontext;
 637
 638         return mm;
 639
 640 fail_nocontext:
 641         mm_free_pgd(mm);
 642 fail_nopgd:
 643         free_mm(mm);
 644         return NULL;
 645 }
 646
 647 static void check_mm(struct mm_struct *mm)
 648 {
 649         int i;
 650
 651         for (i = 0; i < NR_MM_COUNTERS; i++) {
 652                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 653
 654                 if (unlikely(x))
 655                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 656                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 657         }
 658
 659         if (atomic_long_read(&mm->nr_ptes))
 660                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 661                                 atomic_long_read(&mm->nr_ptes));
 662         if (mm_nr_pmds(mm))
 663                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 664                                 mm_nr_pmds(mm));
 665
 666 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 667         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 668 #endif
 669 }
 670
 671 /*
 672  * Allocate and initialize an mm_struct.
 673  */
 674 struct mm_struct *mm_alloc(void)
 675 {
 676         struct mm_struct *mm;
 677
 678         mm = allocate_mm();
 679         if (!mm)
 680                 return NULL;
 681
 682         memset(mm, 0, sizeof(*mm));
 683         return mm_init(mm, current);
 684 }
 685
 686 /*
 687  * Called when the last reference to the mm
 688  * is dropped: either by a lazy thread or by
 689  * mmput. Free the page directory and the mm.
 690  */
 691 void __mmdrop(struct mm_struct *mm)
 692 {
 693         BUG_ON(mm == &init_mm);
 694         mm_free_pgd(mm);
 695         destroy_context(mm);
 696         mmu_notifier_mm_destroy(mm);
 697         check_mm(mm);
 698         free_mm(mm);
 699 }
 700 EXPORT_SYMBOL_GPL(__mmdrop);
 701
 702 /*
 703  * Decrement the use count and release all resources for an mm.
 704  */
 705 void mmput(struct mm_struct *mm)
 706 {
 707         might_sleep();
 708
 709         if (atomic_dec_and_test(&mm->mm_users)) {
 710                 uprobe_clear_state(mm);
 711                 exit_aio(mm);
 712                 ksm_exit(mm);
 713                 khugepaged_exit(mm); /* must run before exit_mmap */
 714                 exit_mmap(mm);
 715                 set_mm_exe_file(mm, NULL);
 716                 if (!list_empty(&mm->mmlist)) {
 717                         spin_lock(&mmlist_lock);
 718                         list_del(&mm->mmlist);
 719                         spin_unlock(&mmlist_lock);
 720                 }
 721                 if (mm->binfmt)
 722                         module_put(mm->binfmt->module);
 723                 mmdrop(mm);
 724         }
 725 }
 726 EXPORT_SYMBOL_GPL(mmput);
 727
 728 /**
 729  * set_mm_exe_file - change a reference to the mm's executable file
 730  *
 731  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 732  *
 733  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 734  * invocations: in mmput() nobody alive left, in execve task is single
 735  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 736  * mm->exe_file, but does so without using set_mm_exe_file() in order
 737  * to do avoid the need for any locks.
 738  */
 739 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 740 {
 741         struct file *old_exe_file;
 742
 743         /*
 744          * It is safe to dereference the exe_file without RCU as
 745          * this function is only called if nobody else can access
 746          * this mm -- see comment above for justification.
 747          */
 748         old_exe_file = rcu_dereference_raw(mm->exe_file);
 749
 750         if (new_exe_file)
 751                 get_file(new_exe_file);
 752         rcu_assign_pointer(mm->exe_file, new_exe_file);
 753         if (old_exe_file)
 754                 fput(old_exe_file);
 755 }
 756
 757 /**
 758  * get_mm_exe_file - acquire a reference to the mm's executable file
 759  *
 760  * Returns %NULL if mm has no associated executable file.
 761  * User must release file via fput().
 762  */
 763 struct file *get_mm_exe_file(struct mm_struct *mm)
 764 {
 765         struct file *exe_file;
 766
 767         rcu_read_lock();
 768         exe_file = rcu_dereference(mm->exe_file);
 769         if (exe_file && !get_file_rcu(exe_file))
 770                 exe_file = NULL;
 771         rcu_read_unlock();
 772         return exe_file;
 773 }
 774 EXPORT_SYMBOL(get_mm_exe_file);
 775
 776 /**
 777  * get_task_mm - acquire a reference to the task's mm
 778  *
 779  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 780  * this kernel workthread has transiently adopted a user mm with use_mm,
 781  * to do its AIO) is not set and if so returns a reference to it, after
 782  * bumping up the use count.  User must release the mm via mmput()
 783  * after use.  Typically used by /proc and ptrace.
 784  */
 785 struct mm_struct *get_task_mm(struct task_struct *task)
 786 {
 787         struct mm_struct *mm;
 788
 789         task_lock(task);
 790         mm = task->mm;
 791         if (mm) {
 792                 if (task->flags & PF_KTHREAD)
 793                         mm = NULL;
 794                 else
 795                         atomic_inc(&mm->mm_users);
 796         }
 797         task_unlock(task);
 798         return mm;
 799 }
 800 EXPORT_SYMBOL_GPL(get_task_mm);
 801
 802 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 803 {
 804         struct mm_struct *mm;
 805         int err;
 806
 807         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 808         if (err)
 809                 return ERR_PTR(err);
 810
 811         mm = get_task_mm(task);
 812         if (mm && mm != current->mm &&
 813                         !ptrace_may_access(task, mode)) {
 814                 mmput(mm);
 815                 mm = ERR_PTR(-EACCES);
 816         }
 817         mutex_unlock(&task->signal->cred_guard_mutex);
 818
 819         return mm;
 820 }
 821
 822 static void complete_vfork_done(struct task_struct *tsk)
 823 {
 824         struct completion *vfork;
 825
 826         task_lock(tsk);
 827         vfork = tsk->vfork_done;
 828         if (likely(vfork)) {
 829                 tsk->vfork_done = NULL;
 830                 complete(vfork);
 831         }
 832         task_unlock(tsk);
 833 }
 834
 835 static int wait_for_vfork_done(struct task_struct *child,
 836                                 struct completion *vfork)
 837 {
 838         int killed;
 839
 840         freezer_do_not_count();
 841         killed = wait_for_completion_killable(vfork);
 842         freezer_count();
 843
 844         if (killed) {
 845                 task_lock(child);
 846                 child->vfork_done = NULL;
 847                 task_unlock(child);
 848         }
 849
 850         put_task_struct(child);
 851         return killed;
 852 }
 853
 854 /* Please note the differences between mmput and mm_release.
 855  * mmput is called whenever we stop holding onto a mm_struct,
 856  * error success whatever.
 857  *
 858  * mm_release is called after a mm_struct has been removed
 859  * from the current process.
 860  *
 861  * This difference is important for error handling, when we
 862  * only half set up a mm_struct for a new process and need to restore
 863  * the old one.  Because we mmput the new mm_struct before
 864  * restoring the old one. . .
 865  * Eric Biederman 10 January 1998
 866  */
 867 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 868 {
 869         /* Get rid of any futexes when releasing the mm */
 870 #ifdef CONFIG_FUTEX
 871         if (unlikely(tsk->robust_list)) {
 872                 exit_robust_list(tsk);
 873                 tsk->robust_list = NULL;
 874         }
 875 #ifdef CONFIG_COMPAT
 876         if (unlikely(tsk->compat_robust_list)) {
 877                 compat_exit_robust_list(tsk);
 878                 tsk->compat_robust_list = NULL;
 879         }
 880 #endif
 881         if (unlikely(!list_empty(&tsk->pi_state_list)))
 882                 exit_pi_state_list(tsk);
 883 #endif
 884
 885         uprobe_free_utask(tsk);
 886
 887         /* Get rid of any cached register state */
 888         deactivate_mm(tsk, mm);
 889
 890         /*
 891          * If we're exiting normally, clear a user-space tid field if
 892          * requested.  We leave this alone when dying by signal, to leave
 893          * the value intact in a core dump, and to save the unnecessary
 894          * trouble, say, a killed vfork parent shouldn't touch this mm.
 895          * Userland only wants this done for a sys_exit.
 896          */
 897         if (tsk->clear_child_tid) {
 898                 if (!(tsk->flags & PF_SIGNALED) &&
 899                     atomic_read(&mm->mm_users) > 1) {
 900                         /*
 901                          * We don't check the error code - if userspace has
 902                          * not set up a proper pointer then tough luck.
 903                          */
 904                         put_user(0, tsk->clear_child_tid);
 905                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 906                                         1, NULL, NULL, 0);
 907                 }
 908                 tsk->clear_child_tid = NULL;
 909         }
 910
 911         /*
 912          * All done, finally we can wake up parent and return this mm to him.
 913          * Also kthread_stop() uses this completion for synchronization.
 914          */
 915         if (tsk->vfork_done)
 916                 complete_vfork_done(tsk);
 917 }
 918
 919 /*
 920  * Allocate a new mm structure and copy contents from the
 921  * mm structure of the passed in task structure.
 922  */
 923 static struct mm_struct *dup_mm(struct task_struct *tsk)
 924 {
 925         struct mm_struct *mm, *oldmm = current->mm;
 926         int err;
 927
 928         mm = allocate_mm();
 929         if (!mm)
 930                 goto fail_nomem;
 931
 932         memcpy(mm, oldmm, sizeof(*mm));
 933
 934         if (!mm_init(mm, tsk))
 935                 goto fail_nomem;
 936
 937         err = dup_mmap(mm, oldmm);
 938         if (err)
 939                 goto free_pt;
 940
 941         mm->hiwater_rss = get_mm_rss(mm);
 942         mm->hiwater_vm = mm->total_vm;
 943
 944         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 945                 goto free_pt;
 946
 947         return mm;
 948
 949 free_pt:
 950         /* don't put binfmt in mmput, we haven't got module yet */
 951         mm->binfmt = NULL;
 952         mmput(mm);
 953
 954 fail_nomem:
 955         return NULL;
 956 }
 957
 958 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 959 {
 960         struct mm_struct *mm, *oldmm;
 961         int retval;
 962
 963         tsk->min_flt = tsk->maj_flt = 0;
 964         tsk->nvcsw = tsk->nivcsw = 0;
 965 #ifdef CONFIG_DETECT_HUNG_TASK
 966         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 967 #endif
 968
 969         tsk->mm = NULL;
 970         tsk->active_mm = NULL;
 971
 972         /*
 973          * Are we cloning a kernel thread?
 974          *
 975          * We need to steal a active VM for that..
 976          */
 977         oldmm = current->mm;
 978         if (!oldmm)
 979                 return 0;
 980
 981         /* initialize the new vmacache entries */
 982         vmacache_flush(tsk);
 983
 984         if (clone_flags & CLONE_VM) {
 985                 atomic_inc(&oldmm->mm_users);
 986                 mm = oldmm;
 987                 goto good_mm;
 988         }
 989
 990         retval = -ENOMEM;
 991         mm = dup_mm(tsk);
 992         if (!mm)
 993                 goto fail_nomem;
 994
 995 good_mm:
 996         tsk->mm = mm;
 997         tsk->active_mm = mm;
 998         return 0;
 999
1000 fail_nomem:
1001         return retval;
1002 }
1003
1004 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1005 {
1006         struct fs_struct *fs = current->fs;
1007         if (clone_flags & CLONE_FS) {
1008                 /* tsk->fs is already what we want */
1009                 spin_lock(&fs->lock);
1010                 if (fs->in_exec) {
1011                         spin_unlock(&fs->lock);
1012                         return -EAGAIN;
1013                 }
1014                 fs->users++;
1015                 spin_unlock(&fs->lock);
1016                 return 0;
1017         }
1018         tsk->fs = copy_fs_struct(fs);
1019         if (!tsk->fs)
1020                 return -ENOMEM;
1021         return 0;
1022 }
1023
1024 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1025 {
1026         struct files_struct *oldf, *newf;
1027         int error = 0;
1028
1029         /*
1030          * A background process may not have any files ...
1031          */
1032         oldf = current->files;
1033         if (!oldf)
1034                 goto out;
1035
1036         if (clone_flags & CLONE_FILES) {
1037                 atomic_inc(&oldf->count);
1038                 goto out;
1039         }
1040
1041         newf = dup_fd(oldf, &error);
1042         if (!newf)
1043                 goto out;
1044
1045         tsk->files = newf;
1046         error = 0;
1047 out:
1048         return error;
1049 }
1050
1051 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1052 {
1053 #ifdef CONFIG_BLOCK
1054         struct io_context *ioc = current->io_context;
1055         struct io_context *new_ioc;
1056
1057         if (!ioc)
1058                 return 0;
1059         /*
1060          * Share io context with parent, if CLONE_IO is set
1061          */
1062         if (clone_flags & CLONE_IO) {
1063                 ioc_task_link(ioc);
1064                 tsk->io_context = ioc;
1065         } else if (ioprio_valid(ioc->ioprio)) {
1066                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1067                 if (unlikely(!new_ioc))
1068                         return -ENOMEM;
1069
1070                 new_ioc->ioprio = ioc->ioprio;
1071                 put_io_context(new_ioc);
1072         }
1073 #endif
1074         return 0;
1075 }
1076
1077 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1078 {
1079         struct sighand_struct *sig;
1080
1081         if (clone_flags & CLONE_SIGHAND) {
1082                 atomic_inc(&current->sighand->count);
1083                 return 0;
1084         }
1085         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1086         rcu_assign_pointer(tsk->sighand, sig);
1087         if (!sig)
1088                 return -ENOMEM;
1089
1090         atomic_set(&sig->count, 1);
1091         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1092         return 0;
1093 }
1094
1095 void __cleanup_sighand(struct sighand_struct *sighand)
1096 {
1097         if (atomic_dec_and_test(&sighand->count)) {
1098                 signalfd_cleanup(sighand);
1099                 /*
1100                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1101                  * without an RCU grace period, see __lock_task_sighand().
1102                  */
1103                 kmem_cache_free(sighand_cachep, sighand);
1104         }
1105 }
1106
1107 /*
1108  * Initialize POSIX timer handling for a thread group.
1109  */
1110 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1111 {
1112         unsigned long cpu_limit;
1113
1114         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1115         if (cpu_limit != RLIM_INFINITY) {
1116                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1117                 sig->cputimer.running = true;
1118         }
1119
1120         /* The timer lists. */
1121         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1122         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1123         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1124 }
1125
1126 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1127 {
1128         struct signal_struct *sig;
1129
1130         if (clone_flags & CLONE_THREAD)
1131                 return 0;
1132
1133         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1134         tsk->signal = sig;
1135         if (!sig)
1136                 return -ENOMEM;
1137
1138         sig->nr_threads = 1;
1139         atomic_set(&sig->live, 1);
1140         atomic_set(&sig->sigcnt, 1);
1141
1142         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1143         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1144         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1145
1146         init_waitqueue_head(&sig->wait_chldexit);
1147         sig->curr_target = tsk;
1148         init_sigpending(&sig->shared_pending);
1149         INIT_LIST_HEAD(&sig->posix_timers);
1150         seqlock_init(&sig->stats_lock);
1151         prev_cputime_init(&sig->prev_cputime);
1152
1153         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1154         sig->real_timer.function = it_real_fn;
1155
1156         task_lock(current->group_leader);
1157         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1158         task_unlock(current->group_leader);
1159
1160         posix_cpu_timers_init_group(sig);
1161
1162         tty_audit_fork(sig);
1163         sched_autogroup_fork(sig);
1164
1165         sig->oom_score_adj = current->signal->oom_score_adj;
1166         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1167
1168         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1169                                    current->signal->is_child_subreaper;
1170
1171         mutex_init(&sig->cred_guard_mutex);
1172
1173         return 0;
1174 }
1175
1176 static void copy_seccomp(struct task_struct *p)
1177 {
1178 #ifdef CONFIG_SECCOMP
1179         /*
1180          * Must be called with sighand->lock held, which is common to
1181          * all threads in the group. Holding cred_guard_mutex is not
1182          * needed because this new task is not yet running and cannot
1183          * be racing exec.
1184          */
1185         assert_spin_locked(&current->sighand->siglock);
1186
1187         /* Ref-count the new filter user, and assign it. */
1188         get_seccomp_filter(current);
1189         p->seccomp = current->seccomp;
1190
1191         /*
1192          * Explicitly enable no_new_privs here in case it got set
1193          * between the task_struct being duplicated and holding the
1194          * sighand lock. The seccomp state and nnp must be in sync.
1195          */
1196         if (task_no_new_privs(current))
1197                 task_set_no_new_privs(p);
1198
1199         /*
1200          * If the parent gained a seccomp mode after copying thread
1201          * flags and between before we held the sighand lock, we have
1202          * to manually enable the seccomp thread flag here.
1203          */
1204         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1205                 set_tsk_thread_flag(p, TIF_SECCOMP);
1206 #endif
1207 }
1208
1209 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1210 {
1211         current->clear_child_tid = tidptr;
1212
1213         return task_pid_vnr(current);
1214 }
1215
1216 static void rt_mutex_init_task(struct task_struct *p)
1217 {
1218         raw_spin_lock_init(&p->pi_lock);
1219 #ifdef CONFIG_RT_MUTEXES
1220         p->pi_waiters = RB_ROOT;
1221         p->pi_waiters_leftmost = NULL;
1222         p->pi_blocked_on = NULL;
1223 #endif
1224 }
1225
1226 /*
1227  * Initialize POSIX timer handling for a single task.
1228  */
1229 static void posix_cpu_timers_init(struct task_struct *tsk)
1230 {
1231         tsk->cputime_expires.prof_exp = 0;
1232         tsk->cputime_expires.virt_exp = 0;
1233         tsk->cputime_expires.sched_exp = 0;
1234         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1235         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1236         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1237 }
1238
1239 static inline void
1240 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1241 {
1242          task->pids[type].pid = pid;
1243 }
1244
1245 /*
1246  * This creates a new process as a copy of the old one,
1247  * but does not actually start it yet.
1248  *
1249  * It copies the registers, and all the appropriate
1250  * parts of the process environment (as per the clone
1251  * flags). The actual kick-off is left to the caller.
1252  */
1253 static struct task_struct *copy_process(unsigned long clone_flags,
1254                                         unsigned long stack_start,
1255                                         unsigned long stack_size,
1256                                         int __user *child_tidptr,
1257                                         struct pid *pid,
1258                                         int trace,
1259                                         unsigned long tls)
1260 {
1261         int retval;
1262         struct task_struct *p;
1263
1264         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1265                 return ERR_PTR(-EINVAL);
1266
1267         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1268                 return ERR_PTR(-EINVAL);
1269
1270         /*
1271          * Thread groups must share signals as well, and detached threads
1272          * can only be started up within the thread group.
1273          */
1274         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1275                 return ERR_PTR(-EINVAL);
1276
1277         /*
1278          * Shared signal handlers imply shared VM. By way of the above,
1279          * thread groups also imply shared VM. Blocking this case allows
1280          * for various simplifications in other code.
1281          */
1282         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1283                 return ERR_PTR(-EINVAL);
1284
1285         /*
1286          * Siblings of global init remain as zombies on exit since they are
1287          * not reaped by their parent (swapper). To solve this and to avoid
1288          * multi-rooted process trees, prevent global and container-inits
1289          * from creating siblings.
1290          */
1291         if ((clone_flags & CLONE_PARENT) &&
1292                                 current->signal->flags & SIGNAL_UNKILLABLE)
1293                 return ERR_PTR(-EINVAL);
1294
1295         /*
1296          * If the new process will be in a different pid or user namespace
1297          * do not allow it to share a thread group with the forking task.
1298          */
1299         if (clone_flags & CLONE_THREAD) {
1300                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1301                     (task_active_pid_ns(current) !=
1302                                 current->nsproxy->pid_ns_for_children))
1303                         return ERR_PTR(-EINVAL);
1304         }
1305
1306         retval = security_task_create(clone_flags);
1307         if (retval)
1308                 goto fork_out;
1309
1310         retval = -ENOMEM;
1311         p = dup_task_struct(current);
1312         if (!p)
1313                 goto fork_out;
1314
1315         ftrace_graph_init_task(p);
1316
1317         rt_mutex_init_task(p);
1318
1319 #ifdef CONFIG_PROVE_LOCKING
1320         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1321         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1322 #endif
1323         retval = -EAGAIN;
1324         if (atomic_read(&p->real_cred->user->processes) >=
1325                         task_rlimit(p, RLIMIT_NPROC)) {
1326                 if (p->real_cred->user != INIT_USER &&
1327                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1328                         goto bad_fork_free;
1329         }
1330         current->flags &= ~PF_NPROC_EXCEEDED;
1331
1332         retval = copy_creds(p, clone_flags);
1333         if (retval < 0)
1334                 goto bad_fork_free;
1335
1336         /*
1337          * If multiple threads are within copy_process(), then this check
1338          * triggers too late. This doesn't hurt, the check is only there
1339          * to stop root fork bombs.
1340          */
1341         retval = -EAGAIN;
1342         if (nr_threads >= max_threads)
1343                 goto bad_fork_cleanup_count;
1344
1345         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1346         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1347         p->flags |= PF_FORKNOEXEC;
1348         INIT_LIST_HEAD(&p->children);
1349         INIT_LIST_HEAD(&p->sibling);
1350         rcu_copy_process(p);
1351         p->vfork_done = NULL;
1352         spin_lock_init(&p->alloc_lock);
1353
1354         init_sigpending(&p->pending);
1355
1356         p->utime = p->stime = p->gtime = 0;
1357         p->utimescaled = p->stimescaled = 0;
1358         prev_cputime_init(&p->prev_cputime);
1359
1360 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1361         seqcount_init(&p->vtime_seqcount);
1362         p->vtime_snap = 0;
1363         p->vtime_snap_whence = VTIME_INACTIVE;
1364 #endif
1365
1366 #if defined(SPLIT_RSS_COUNTING)
1367         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1368 #endif
1369
1370         p->default_timer_slack_ns = current->timer_slack_ns;
1371
1372         task_io_accounting_init(&p->ioac);
1373         acct_clear_integrals(p);
1374
1375         posix_cpu_timers_init(p);
1376
1377         p->start_time = ktime_get_ns();
1378         p->real_start_time = ktime_get_boot_ns();
1379         p->io_context = NULL;
1380         p->audit_context = NULL;
1381         threadgroup_change_begin(current);
1382         cgroup_fork(p);
1383 #ifdef CONFIG_NUMA
1384         p->mempolicy = mpol_dup(p->mempolicy);
1385         if (IS_ERR(p->mempolicy)) {
1386                 retval = PTR_ERR(p->mempolicy);
1387                 p->mempolicy = NULL;
1388                 goto bad_fork_cleanup_threadgroup_lock;
1389         }
1390 #endif
1391 #ifdef CONFIG_CPUSETS
1392         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1393         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1394         seqcount_init(&p->mems_allowed_seq);
1395 #endif
1396 #ifdef CONFIG_TRACE_IRQFLAGS
1397         p->irq_events = 0;
1398         p->hardirqs_enabled = 0;
1399         p->hardirq_enable_ip = 0;
1400         p->hardirq_enable_event = 0;
1401         p->hardirq_disable_ip = _THIS_IP_;
1402         p->hardirq_disable_event = 0;
1403         p->softirqs_enabled = 1;
1404         p->softirq_enable_ip = _THIS_IP_;
1405         p->softirq_enable_event = 0;
1406         p->softirq_disable_ip = 0;
1407         p->softirq_disable_event = 0;
1408         p->hardirq_context = 0;
1409         p->softirq_context = 0;
1410 #endif
1411
1412         p->pagefault_disabled = 0;
1413
1414 #ifdef CONFIG_LOCKDEP
1415         p->lockdep_depth = 0; /* no locks held yet */
1416         p->curr_chain_key = 0;
1417         p->lockdep_recursion = 0;
1418 #endif
1419
1420 #ifdef CONFIG_DEBUG_MUTEXES
1421         p->blocked_on = NULL; /* not blocked yet */
1422 #endif
1423 #ifdef CONFIG_BCACHE
1424         p->sequential_io        = 0;
1425         p->sequential_io_avg    = 0;
1426 #endif
1427
1428         /* Perform scheduler related setup. Assign this task to a CPU. */
1429         retval = sched_fork(clone_flags, p);
1430         if (retval)
1431                 goto bad_fork_cleanup_policy;
1432
1433         retval = perf_event_init_task(p);
1434         if (retval)
1435                 goto bad_fork_cleanup_policy;
1436         retval = audit_alloc(p);
1437         if (retval)
1438                 goto bad_fork_cleanup_perf;
1439         /* copy all the process information */
1440         shm_init_task(p);
1441         retval = copy_semundo(clone_flags, p);
1442         if (retval)
1443                 goto bad_fork_cleanup_audit;
1444         retval = copy_files(clone_flags, p);
1445         if (retval)
1446                 goto bad_fork_cleanup_semundo;
1447         retval = copy_fs(clone_flags, p);
1448         if (retval)
1449                 goto bad_fork_cleanup_files;
1450         retval = copy_sighand(clone_flags, p);
1451         if (retval)
1452                 goto bad_fork_cleanup_fs;
1453         retval = copy_signal(clone_flags, p);
1454         if (retval)
1455                 goto bad_fork_cleanup_sighand;
1456         retval = copy_mm(clone_flags, p);
1457         if (retval)
1458                 goto bad_fork_cleanup_signal;
1459         retval = copy_namespaces(clone_flags, p);
1460         if (retval)
1461                 goto bad_fork_cleanup_mm;
1462         retval = copy_io(clone_flags, p);
1463         if (retval)
1464                 goto bad_fork_cleanup_namespaces;
1465         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1466         if (retval)
1467                 goto bad_fork_cleanup_io;
1468
1469         if (pid != &init_struct_pid) {
1470                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1471                 if (IS_ERR(pid)) {
1472                         retval = PTR_ERR(pid);
1473                         goto bad_fork_cleanup_io;
1474                 }
1475         }
1476
1477         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1478         /*
1479          * Clear TID on mm_release()?
1480          */
1481         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1482 #ifdef CONFIG_BLOCK
1483         p->plug = NULL;
1484 #endif
1485 #ifdef CONFIG_FUTEX
1486         p->robust_list = NULL;
1487 #ifdef CONFIG_COMPAT
1488         p->compat_robust_list = NULL;
1489 #endif
1490         INIT_LIST_HEAD(&p->pi_state_list);
1491         p->pi_state_cache = NULL;
1492 #endif
1493         /*
1494          * sigaltstack should be cleared when sharing the same VM
1495          */
1496         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1497                 p->sas_ss_sp = p->sas_ss_size = 0;
1498
1499         /*
1500          * Syscall tracing and stepping should be turned off in the
1501          * child regardless of CLONE_PTRACE.
1502          */
1503         user_disable_single_step(p);
1504         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1505 #ifdef TIF_SYSCALL_EMU
1506         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1507 #endif
1508         clear_all_latency_tracing(p);
1509
1510         /* ok, now we should be set up.. */
1511         p->pid = pid_nr(pid);
1512         if (clone_flags & CLONE_THREAD) {
1513                 p->exit_signal = -1;
1514                 p->group_leader = current->group_leader;
1515                 p->tgid = current->tgid;
1516         } else {
1517                 if (clone_flags & CLONE_PARENT)
1518                         p->exit_signal = current->group_leader->exit_signal;
1519                 else
1520                         p->exit_signal = (clone_flags & CSIGNAL);
1521                 p->group_leader = p;
1522                 p->tgid = p->pid;
1523         }
1524
1525         p->nr_dirtied = 0;
1526         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1527         p->dirty_paused_when = 0;
1528
1529         p->pdeath_signal = 0;
1530         INIT_LIST_HEAD(&p->thread_group);
1531         p->task_works = NULL;
1532
1533         /*
1534          * Ensure that the cgroup subsystem policies allow the new process to be
1535          * forked. It should be noted the the new process's css_set can be changed
1536          * between here and cgroup_post_fork() if an organisation operation is in
1537          * progress.
1538          */
1539         retval = cgroup_can_fork(p);
1540         if (retval)
1541                 goto bad_fork_free_pid;
1542
1543         /*
1544          * Make it visible to the rest of the system, but dont wake it up yet.
1545          * Need tasklist lock for parent etc handling!
1546          */
1547         write_lock_irq(&tasklist_lock);
1548
1549         /* CLONE_PARENT re-uses the old parent */
1550         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1551                 p->real_parent = current->real_parent;
1552                 p->parent_exec_id = current->parent_exec_id;
1553         } else {
1554                 p->real_parent = current;
1555                 p->parent_exec_id = current->self_exec_id;
1556         }
1557
1558         spin_lock(&current->sighand->siglock);
1559
1560         /*
1561          * Copy seccomp details explicitly here, in case they were changed
1562          * before holding sighand lock.
1563          */
1564         copy_seccomp(p);
1565
1566         /*
1567          * Process group and session signals need to be delivered to just the
1568          * parent before the fork or both the parent and the child after the
1569          * fork. Restart if a signal comes in before we add the new process to
1570          * it's process group.
1571          * A fatal signal pending means that current will exit, so the new
1572          * thread can't slip out of an OOM kill (or normal SIGKILL).
1573         */
1574         recalc_sigpending();
1575         if (signal_pending(current)) {
1576                 spin_unlock(&current->sighand->siglock);
1577                 write_unlock_irq(&tasklist_lock);
1578                 retval = -ERESTARTNOINTR;
1579                 goto bad_fork_cancel_cgroup;
1580         }
1581
1582         if (likely(p->pid)) {
1583                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1584
1585                 init_task_pid(p, PIDTYPE_PID, pid);
1586                 if (thread_group_leader(p)) {
1587                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1588                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1589
1590                         if (is_child_reaper(pid)) {
1591                                 ns_of_pid(pid)->child_reaper = p;
1592                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1593                         }
1594
1595                         p->signal->leader_pid = pid;
1596                         p->signal->tty = tty_kref_get(current->signal->tty);
1597                         list_add_tail(&p->sibling, &p->real_parent->children);
1598                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1599                         attach_pid(p, PIDTYPE_PGID);
1600                         attach_pid(p, PIDTYPE_SID);
1601                         __this_cpu_inc(process_counts);
1602                 } else {
1603                         current->signal->nr_threads++;
1604                         atomic_inc(&current->signal->live);
1605                         atomic_inc(&current->signal->sigcnt);
1606                         list_add_tail_rcu(&p->thread_group,
1607                                           &p->group_leader->thread_group);
1608                         list_add_tail_rcu(&p->thread_node,
1609                                           &p->signal->thread_head);
1610                 }
1611                 attach_pid(p, PIDTYPE_PID);
1612                 nr_threads++;
1613         }
1614
1615         total_forks++;
1616         spin_unlock(&current->sighand->siglock);
1617         syscall_tracepoint_update(p);
1618         write_unlock_irq(&tasklist_lock);
1619
1620         proc_fork_connector(p);
1621         cgroup_post_fork(p);
1622         threadgroup_change_end(current);
1623         perf_event_fork(p);
1624
1625         trace_task_newtask(p, clone_flags);
1626         uprobe_copy_process(p, clone_flags);
1627
1628         return p;
1629
1630 bad_fork_cancel_cgroup:
1631         cgroup_cancel_fork(p);
1632 bad_fork_free_pid:
1633         if (pid != &init_struct_pid)
1634                 free_pid(pid);
1635 bad_fork_cleanup_io:
1636         if (p->io_context)
1637                 exit_io_context(p);
1638 bad_fork_cleanup_namespaces:
1639         exit_task_namespaces(p);
1640 bad_fork_cleanup_mm:
1641         if (p->mm)
1642                 mmput(p->mm);
1643 bad_fork_cleanup_signal:
1644         if (!(clone_flags & CLONE_THREAD))
1645                 free_signal_struct(p->signal);
1646 bad_fork_cleanup_sighand:
1647         __cleanup_sighand(p->sighand);
1648 bad_fork_cleanup_fs:
1649         exit_fs(p); /* blocking */
1650 bad_fork_cleanup_files:
1651         exit_files(p); /* blocking */
1652 bad_fork_cleanup_semundo:
1653         exit_sem(p);
1654 bad_fork_cleanup_audit:
1655         audit_free(p);
1656 bad_fork_cleanup_perf:
1657         perf_event_free_task(p);
1658 bad_fork_cleanup_policy:
1659 #ifdef CONFIG_NUMA
1660         mpol_put(p->mempolicy);
1661 bad_fork_cleanup_threadgroup_lock:
1662 #endif
1663         threadgroup_change_end(current);
1664         delayacct_tsk_free(p);
1665 bad_fork_cleanup_count:
1666         atomic_dec(&p->cred->user->processes);
1667         exit_creds(p);
1668 bad_fork_free:
1669         free_task(p);
1670 fork_out:
1671         return ERR_PTR(retval);
1672 }
1673
1674 static inline void init_idle_pids(struct pid_link *links)
1675 {
1676         enum pid_type type;
1677
1678         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1679                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1680                 links[type].pid = &init_struct_pid;
1681         }
1682 }
1683
1684 struct task_struct *fork_idle(int cpu)
1685 {
1686         struct task_struct *task;
1687         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0);
1688         if (!IS_ERR(task)) {
1689                 init_idle_pids(task->pids);
1690                 init_idle(task, cpu);
1691         }
1692
1693         return task;
1694 }
1695
1696 /*
1697  *  Ok, this is the main fork-routine.
1698  *
1699  * It copies the process, and if successful kick-starts
1700  * it and waits for it to finish using the VM if required.
1701  */
1702 long _do_fork(unsigned long clone_flags,
1703               unsigned long stack_start,
1704               unsigned long stack_size,
1705               int __user *parent_tidptr,
1706               int __user *child_tidptr,
1707               unsigned long tls)
1708 {
1709         struct task_struct *p;
1710         int trace = 0;
1711         long nr;
1712
1713         /*
1714          * Determine whether and which event to report to ptracer.  When
1715          * called from kernel_thread or CLONE_UNTRACED is explicitly
1716          * requested, no event is reported; otherwise, report if the event
1717          * for the type of forking is enabled.
1718          */
1719         if (!(clone_flags & CLONE_UNTRACED)) {
1720                 if (clone_flags & CLONE_VFORK)
1721                         trace = PTRACE_EVENT_VFORK;
1722                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1723                         trace = PTRACE_EVENT_CLONE;
1724                 else
1725                         trace = PTRACE_EVENT_FORK;
1726
1727                 if (likely(!ptrace_event_enabled(current, trace)))
1728                         trace = 0;
1729         }
1730
1731         p = copy_process(clone_flags, stack_start, stack_size,
1732                          child_tidptr, NULL, trace, tls);
1733         /*
1734          * Do this prior waking up the new thread - the thread pointer
1735          * might get invalid after that point, if the thread exits quickly.
1736          */
1737         if (!IS_ERR(p)) {
1738                 struct completion vfork;
1739                 struct pid *pid;
1740
1741                 trace_sched_process_fork(current, p);
1742
1743                 pid = get_task_pid(p, PIDTYPE_PID);
1744                 nr = pid_vnr(pid);
1745
1746                 if (clone_flags & CLONE_PARENT_SETTID)
1747                         put_user(nr, parent_tidptr);
1748
1749                 if (clone_flags & CLONE_VFORK) {
1750                         p->vfork_done = &vfork;
1751                         init_completion(&vfork);
1752                         get_task_struct(p);
1753                 }
1754
1755                 wake_up_new_task(p);
1756
1757                 /* forking complete and child started to run, tell ptracer */
1758                 if (unlikely(trace))
1759                         ptrace_event_pid(trace, pid);
1760
1761                 if (clone_flags & CLONE_VFORK) {
1762                         if (!wait_for_vfork_done(p, &vfork))
1763                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1764                 }
1765
1766                 put_pid(pid);
1767         } else {
1768                 nr = PTR_ERR(p);
1769         }
1770         return nr;
1771 }
1772
1773 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1774 /* For compatibility with architectures that call do_fork directly rather than
1775  * using the syscall entry points below. */
1776 long do_fork(unsigned long clone_flags,
1777               unsigned long stack_start,
1778               unsigned long stack_size,
1779               int __user *parent_tidptr,
1780               int __user *child_tidptr)
1781 {
1782         return _do_fork(clone_flags, stack_start, stack_size,
1783                         parent_tidptr, child_tidptr, 0);
1784 }
1785 #endif
1786
1787 /*
1788  * Create a kernel thread.
1789  */
1790 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1791 {
1792         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1793                 (unsigned long)arg, NULL, NULL, 0);
1794 }
1795
1796 #ifdef __ARCH_WANT_SYS_FORK
1797 SYSCALL_DEFINE0(fork)
1798 {
1799 #ifdef CONFIG_MMU
1800         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1801 #else
1802         /* can not support in nommu mode */
1803         return -EINVAL;
1804 #endif
1805 }
1806 #endif
1807
1808 #ifdef __ARCH_WANT_SYS_VFORK
1809 SYSCALL_DEFINE0(vfork)
1810 {
1811         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1812                         0, NULL, NULL, 0);
1813 }
1814 #endif
1815
1816 #ifdef __ARCH_WANT_SYS_CLONE
1817 #ifdef CONFIG_CLONE_BACKWARDS
1818 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1819                  int __user *, parent_tidptr,
1820                  unsigned long, tls,
1821                  int __user *, child_tidptr)
1822 #elif defined(CONFIG_CLONE_BACKWARDS2)
1823 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1824                  int __user *, parent_tidptr,
1825                  int __user *, child_tidptr,
1826                  unsigned long, tls)
1827 #elif defined(CONFIG_CLONE_BACKWARDS3)
1828 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1829                 int, stack_size,
1830                 int __user *, parent_tidptr,
1831                 int __user *, child_tidptr,
1832                 unsigned long, tls)
1833 #else
1834 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1835                  int __user *, parent_tidptr,
1836                  int __user *, child_tidptr,
1837                  unsigned long, tls)
1838 #endif
1839 {
1840         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
1841 }
1842 #endif
1843
1844 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1845 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1846 #endif
1847
1848 static void sighand_ctor(void *data)
1849 {
1850         struct sighand_struct *sighand = data;
1851
1852         spin_lock_init(&sighand->siglock);
1853         init_waitqueue_head(&sighand->signalfd_wqh);
1854 }
1855
1856 void __init proc_caches_init(void)
1857 {
1858         sighand_cachep = kmem_cache_create("sighand_cache",
1859                         sizeof(struct sighand_struct), 0,
1860                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1861                         SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
1862         signal_cachep = kmem_cache_create("signal_cache",
1863                         sizeof(struct signal_struct), 0,
1864                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1865                         NULL);
1866         files_cachep = kmem_cache_create("files_cache",
1867                         sizeof(struct files_struct), 0,
1868                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1869                         NULL);
1870         fs_cachep = kmem_cache_create("fs_cache",
1871                         sizeof(struct fs_struct), 0,
1872                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1873                         NULL);
1874         /*
1875          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1876          * whole struct cpumask for the OFFSTACK case. We could change
1877          * this to *only* allocate as much of it as required by the
1878          * maximum number of CPU's we can ever have.  The cpumask_allocation
1879          * is at the end of the structure, exactly for that reason.
1880          */
1881         mm_cachep = kmem_cache_create("mm_struct",
1882                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1883                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1884                         NULL);
1885         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
1886         mmap_init();
1887         nsproxy_cache_init();
1888 }
1889
1890 /*
1891  * Check constraints on flags passed to the unshare system call.
1892  */
1893 static int check_unshare_flags(unsigned long unshare_flags)
1894 {
1895         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1896                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1897                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1898                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
1899                 return -EINVAL;
1900         /*
1901          * Not implemented, but pretend it works if there is nothing
1902          * to unshare.  Note that unsharing the address space or the
1903          * signal handlers also need to unshare the signal queues (aka
1904          * CLONE_THREAD).
1905          */
1906         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1907                 if (!thread_group_empty(current))
1908                         return -EINVAL;
1909         }
1910         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
1911                 if (atomic_read(&current->sighand->count) > 1)
1912                         return -EINVAL;
1913         }
1914         if (unshare_flags & CLONE_VM) {
1915                 if (!current_is_single_threaded())
1916                         return -EINVAL;
1917         }
1918
1919         return 0;
1920 }
1921
1922 /*
1923  * Unshare the filesystem structure if it is being shared
1924  */
1925 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1926 {
1927         struct fs_struct *fs = current->fs;
1928
1929         if (!(unshare_flags & CLONE_FS) || !fs)
1930                 return 0;
1931
1932         /* don't need lock here; in the worst case we'll do useless copy */
1933         if (fs->users == 1)
1934                 return 0;
1935
1936         *new_fsp = copy_fs_struct(fs);
1937         if (!*new_fsp)
1938                 return -ENOMEM;
1939
1940         return 0;
1941 }
1942
1943 /*
1944  * Unshare file descriptor table if it is being shared
1945  */
1946 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1947 {
1948         struct files_struct *fd = current->files;
1949         int error = 0;
1950
1951         if ((unshare_flags & CLONE_FILES) &&
1952             (fd && atomic_read(&fd->count) > 1)) {
1953                 *new_fdp = dup_fd(fd, &error);
1954                 if (!*new_fdp)
1955                         return error;
1956         }
1957
1958         return 0;
1959 }
1960
1961 /*
1962  * unshare allows a process to 'unshare' part of the process
1963  * context which was originally shared using clone.  copy_*
1964  * functions used by do_fork() cannot be used here directly
1965  * because they modify an inactive task_struct that is being
1966  * constructed. Here we are modifying the current, active,
1967  * task_struct.
1968  */
1969 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1970 {
1971         struct fs_struct *fs, *new_fs = NULL;
1972         struct files_struct *fd, *new_fd = NULL;
1973         struct cred *new_cred = NULL;
1974         struct nsproxy *new_nsproxy = NULL;
1975         int do_sysvsem = 0;
1976         int err;
1977
1978         /*
1979          * If unsharing a user namespace must also unshare the thread group
1980          * and unshare the filesystem root and working directories.
1981          */
1982         if (unshare_flags & CLONE_NEWUSER)
1983                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1984         /*
1985          * If unsharing vm, must also unshare signal handlers.
1986          */
1987         if (unshare_flags & CLONE_VM)
1988                 unshare_flags |= CLONE_SIGHAND;
1989         /*
1990          * If unsharing a signal handlers, must also unshare the signal queues.
1991          */
1992         if (unshare_flags & CLONE_SIGHAND)
1993                 unshare_flags |= CLONE_THREAD;
1994         /*
1995          * If unsharing namespace, must also unshare filesystem information.
1996          */
1997         if (unshare_flags & CLONE_NEWNS)
1998                 unshare_flags |= CLONE_FS;
1999
2000         err = check_unshare_flags(unshare_flags);
2001         if (err)
2002                 goto bad_unshare_out;
2003         /*
2004          * CLONE_NEWIPC must also detach from the undolist: after switching
2005          * to a new ipc namespace, the semaphore arrays from the old
2006          * namespace are unreachable.
2007          */
2008         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2009                 do_sysvsem = 1;
2010         err = unshare_fs(unshare_flags, &new_fs);
2011         if (err)
2012                 goto bad_unshare_out;
2013         err = unshare_fd(unshare_flags, &new_fd);
2014         if (err)
2015                 goto bad_unshare_cleanup_fs;
2016         err = unshare_userns(unshare_flags, &new_cred);
2017         if (err)
2018                 goto bad_unshare_cleanup_fd;
2019         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2020                                          new_cred, new_fs);
2021         if (err)
2022                 goto bad_unshare_cleanup_cred;
2023
2024         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2025                 if (do_sysvsem) {
2026                         /*
2027                          * CLONE_SYSVSEM is equivalent to sys_exit().
2028                          */
2029                         exit_sem(current);
2030                 }
2031                 if (unshare_flags & CLONE_NEWIPC) {
2032                         /* Orphan segments in old ns (see sem above). */
2033                         exit_shm(current);
2034                         shm_init_task(current);
2035                 }
2036
2037                 if (new_nsproxy)
2038                         switch_task_namespaces(current, new_nsproxy);
2039
2040                 task_lock(current);
2041
2042                 if (new_fs) {
2043                         fs = current->fs;
2044                         spin_lock(&fs->lock);
2045                         current->fs = new_fs;
2046                         if (--fs->users)
2047                                 new_fs = NULL;
2048                         else
2049                                 new_fs = fs;
2050                         spin_unlock(&fs->lock);
2051                 }
2052
2053                 if (new_fd) {
2054                         fd = current->files;
2055                         current->files = new_fd;
2056                         new_fd = fd;
2057                 }
2058
2059                 task_unlock(current);
2060
2061                 if (new_cred) {
2062                         /* Install the new user namespace */
2063                         commit_creds(new_cred);
2064                         new_cred = NULL;
2065                 }
2066         }
2067
2068 bad_unshare_cleanup_cred:
2069         if (new_cred)
2070                 put_cred(new_cred);
2071 bad_unshare_cleanup_fd:
2072         if (new_fd)
2073                 put_files_struct(new_fd);
2074
2075 bad_unshare_cleanup_fs:
2076         if (new_fs)
2077                 free_fs_struct(new_fs);
2078
2079 bad_unshare_out:
2080         return err;
2081 }
2082
2083 /*
2084  *      Helper to unshare the files of the current task.
2085  *      We don't want to expose copy_files internals to
2086  *      the exec layer of the kernel.
2087  */
2088
2089 int unshare_files(struct files_struct **displaced)
2090 {
2091         struct task_struct *task = current;
2092         struct files_struct *copy = NULL;
2093         int error;
2094
2095         error = unshare_fd(CLONE_FILES, &copy);
2096         if (error || !copy) {
2097                 *displaced = NULL;
2098                 return error;
2099         }
2100         *displaced = task->files;
2101         task_lock(task);
2102         task->files = copy;
2103         task_unlock(task);
2104         return 0;
2105 }
2106
2107 int sysctl_max_threads(struct ctl_table *table, int write,
2108                        void __user *buffer, size_t *lenp, loff_t *ppos)
2109 {
2110         struct ctl_table t;
2111         int ret;
2112         int threads = max_threads;
2113         int min = MIN_THREADS;
2114         int max = MAX_THREADS;
2115
2116         t = *table;
2117         t.data = &threads;
2118         t.extra1 = &min;
2119         t.extra2 = &max;
2120
2121         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2122         if (ret || !write)
2123                 return ret;
2124
2125         set_max_threads(threads);
2126
2127         return 0;
2128 }