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