kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/mm.h>
  32 #include <linux/vmacache.h>
  33 #include <linux/nsproxy.h>
  34 #include <linux/capability.h>
  35 #include <linux/cpu.h>
  36 #include <linux/cgroup.h>
  37 #include <linux/security.h>
  38 #include <linux/hugetlb.h>
  39 #include <linux/seccomp.h>
  40 #include <linux/swap.h>
  41 #include <linux/syscalls.h>
  42 #include <linux/jiffies.h>
  43 #include <linux/futex.h>
  44 #include <linux/compat.h>
  45 #include <linux/kthread.h>
  46 #include <linux/task_io_accounting_ops.h>
  47 #include <linux/rcupdate.h>
  48 #include <linux/ptrace.h>
  49 #include <linux/mount.h>
  50 #include <linux/audit.h>
  51 #include <linux/memcontrol.h>
  52 #include <linux/ftrace.h>
  53 #include <linux/proc_fs.h>
  54 #include <linux/profile.h>
  55 #include <linux/rmap.h>
  56 #include <linux/ksm.h>
  57 #include <linux/acct.h>
  58 #include <linux/tsacct_kern.h>
  59 #include <linux/cn_proc.h>
  60 #include <linux/freezer.h>
  61 #include <linux/delayacct.h>
  62 #include <linux/taskstats_kern.h>
  63 #include <linux/random.h>
  64 #include <linux/tty.h>
  65 #include <linux/blkdev.h>
  66 #include <linux/fs_struct.h>
  67 #include <linux/magic.h>
  68 #include <linux/perf_event.h>
  69 #include <linux/posix-timers.h>
  70 #include <linux/user-return-notifier.h>
  71 #include <linux/oom.h>
  72 #include <linux/khugepaged.h>
  73 #include <linux/signalfd.h>
  74 #include <linux/uprobes.h>
  75 #include <linux/aio.h>
  76 #include <linux/compiler.h>
  77 #include <linux/sysctl.h>
  78 #include <linux/kcov.h>
  79
  80 #include <asm/pgtable.h>
  81 #include <asm/pgalloc.h>
  82 #include <asm/uaccess.h>
  83 #include <asm/mmu_context.h>
  84 #include <asm/cacheflush.h>
  85 #include <asm/tlbflush.h>
  86
  87 #include <trace/events/sched.h>
  88
  89 #define CREATE_TRACE_POINTS
  90 #include <trace/events/task.h>
  91
  92 /*
  93  * Minimum number of threads to boot the kernel
  94  */
  95 #define MIN_THREADS 20
  96
  97 /*
  98  * Maximum number of threads
  99  */
 100 #define MAX_THREADS FUTEX_TID_MASK
 101
 102 /*
 103  * Protected counters by write_lock_irq(&tasklist_lock)
 104  */
 105 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 106 int nr_threads;                 /* The idle threads do not count.. */
 107
 108 int max_threads;                /* tunable limit on nr_threads */
 109
 110 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 111
 112 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 113
 114 #ifdef CONFIG_PROVE_RCU
 115 int lockdep_tasklist_lock_is_held(void)
 116 {
 117         return lockdep_is_held(&tasklist_lock);
 118 }
 119 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 120 #endif /* #ifdef CONFIG_PROVE_RCU */
 121
 122 int nr_processes(void)
 123 {
 124         int cpu;
 125         int total = 0;
 126
 127         for_each_possible_cpu(cpu)
 128                 total += per_cpu(process_counts, cpu);
 129
 130         return total;
 131 }
 132
 133 void __weak arch_release_task_struct(struct task_struct *tsk)
 134 {
 135 }
 136
 137 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 138 static struct kmem_cache *task_struct_cachep;
 139
 140 static inline struct task_struct *alloc_task_struct_node(int node)
 141 {
 142         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 143 }
 144
 145 static inline void free_task_struct(struct task_struct *tsk)
 146 {
 147         kmem_cache_free(task_struct_cachep, tsk);
 148 }
 149 #endif
 150
 151 void __weak arch_release_thread_stack(unsigned long *stack)
 152 {
 153 }
 154
 155 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 156
 157 /*
 158  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 159  * kmemcache based allocator.
 160  */
 161 # if THREAD_SIZE >= PAGE_SIZE
 162 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 163                                                   int node)
 164 {
 165         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 166                                              THREAD_SIZE_ORDER);
 167
 168         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_stack(unsigned long *stack)
 176 {
 177         struct page *page = virt_to_page(stack);
 178
 179         memcg_kmem_update_page_stat(page, MEMCG_KERNEL_STACK,
 180                                     -(1 << THREAD_SIZE_ORDER));
 181         __free_pages(page, THREAD_SIZE_ORDER);
 182 }
 183 # else
 184 static struct kmem_cache *thread_stack_cache;
 185
 186 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 187                                                   int node)
 188 {
 189         return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 190 }
 191
 192 static void free_thread_stack(unsigned long *stack)
 193 {
 194         kmem_cache_free(thread_stack_cache, stack);
 195 }
 196
 197 void thread_stack_cache_init(void)
 198 {
 199         thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
 200                                               THREAD_SIZE, 0, NULL);
 201         BUG_ON(thread_stack_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(unsigned long *stack, int account)
 225 {
 226         struct zone *zone = page_zone(virt_to_page(stack));
 227
 228         mod_zone_page_state(zone, NR_KERNEL_STACK_KB,
 229                             THREAD_SIZE / 1024 * account);
 230 }
 231
 232 void free_task(struct task_struct *tsk)
 233 {
 234         account_kernel_stack(tsk->stack, -1);
 235         arch_release_thread_stack(tsk->stack);
 236         free_thread_stack(tsk->stack);
 237         rt_mutex_debug_task_free(tsk);
 238         ftrace_graph_exit_task(tsk);
 239         put_seccomp_filter(tsk);
 240         arch_release_task_struct(tsk);
 241         free_task_struct(tsk);
 242 }
 243 EXPORT_SYMBOL(free_task);
 244
 245 static inline void free_signal_struct(struct signal_struct *sig)
 246 {
 247         taskstats_tgid_free(sig);
 248         sched_autogroup_exit(sig);
 249         kmem_cache_free(signal_cachep, sig);
 250 }
 251
 252 static inline void put_signal_struct(struct signal_struct *sig)
 253 {
 254         if (atomic_dec_and_test(&sig->sigcnt))
 255                 free_signal_struct(sig);
 256 }
 257
 258 void __put_task_struct(struct task_struct *tsk)
 259 {
 260         WARN_ON(!tsk->exit_state);
 261         WARN_ON(atomic_read(&tsk->usage));
 262         WARN_ON(tsk == current);
 263
 264         cgroup_free(tsk);
 265         task_numa_free(tsk);
 266         security_task_free(tsk);
 267         exit_creds(tsk);
 268         delayacct_tsk_free(tsk);
 269         put_signal_struct(tsk->signal);
 270
 271         if (!profile_handoff_task(tsk))
 272                 free_task(tsk);
 273 }
 274 EXPORT_SYMBOL_GPL(__put_task_struct);
 275
 276 void __init __weak arch_task_cache_init(void) { }
 277
 278 /*
 279  * set_max_threads
 280  */
 281 static void set_max_threads(unsigned int max_threads_suggested)
 282 {
 283         u64 threads;
 284
 285         /*
 286          * The number of threads shall be limited such that the thread
 287          * structures may only consume a small part of the available memory.
 288          */
 289         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 290                 threads = MAX_THREADS;
 291         else
 292                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 293                                     (u64) THREAD_SIZE * 8UL);
 294
 295         if (threads > max_threads_suggested)
 296                 threads = max_threads_suggested;
 297
 298         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 299 }
 300
 301 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 302 /* Initialized by the architecture: */
 303 int arch_task_struct_size __read_mostly;
 304 #endif
 305
 306 void __init fork_init(void)
 307 {
 308 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 309 #ifndef ARCH_MIN_TASKALIGN
 310 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 311 #endif
 312         /* create a slab on which task_structs can be allocated */
 313         task_struct_cachep = kmem_cache_create("task_struct",
 314                         arch_task_struct_size, ARCH_MIN_TASKALIGN,
 315                         SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
 316 #endif
 317
 318         /* do the arch specific task caches init */
 319         arch_task_cache_init();
 320
 321         set_max_threads(MAX_THREADS);
 322
 323         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 324         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 325         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 326                 init_task.signal->rlim[RLIMIT_NPROC];
 327 }
 328
 329 int __weak arch_dup_task_struct(struct task_struct *dst,
 330                                                struct task_struct *src)
 331 {
 332         *dst = *src;
 333         return 0;
 334 }
 335
 336 void set_task_stack_end_magic(struct task_struct *tsk)
 337 {
 338         unsigned long *stackend;
 339
 340         stackend = end_of_stack(tsk);
 341         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 342 }
 343
 344 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 345 {
 346         struct task_struct *tsk;
 347         unsigned long *stack;
 348         int err;
 349
 350         if (node == NUMA_NO_NODE)
 351                 node = tsk_fork_get_node(orig);
 352         tsk = alloc_task_struct_node(node);
 353         if (!tsk)
 354                 return NULL;
 355
 356         stack = alloc_thread_stack_node(tsk, node);
 357         if (!stack)
 358                 goto free_tsk;
 359
 360         err = arch_dup_task_struct(tsk, orig);
 361         if (err)
 362                 goto free_stack;
 363
 364         tsk->stack = stack;
 365 #ifdef CONFIG_SECCOMP
 366         /*
 367          * We must handle setting up seccomp filters once we're under
 368          * the sighand lock in case orig has changed between now and
 369          * then. Until then, filter must be NULL to avoid messing up
 370          * the usage counts on the error path calling free_task.
 371          */
 372         tsk->seccomp.filter = NULL;
 373 #endif
 374
 375         setup_thread_stack(tsk, orig);
 376         clear_user_return_notifier(tsk);
 377         clear_tsk_need_resched(tsk);
 378         set_task_stack_end_magic(tsk);
 379
 380 #ifdef CONFIG_CC_STACKPROTECTOR
 381         tsk->stack_canary = get_random_int();
 382 #endif
 383
 384         /*
 385          * One for us, one for whoever does the "release_task()" (usually
 386          * parent)
 387          */
 388         atomic_set(&tsk->usage, 2);
 389 #ifdef CONFIG_BLK_DEV_IO_TRACE
 390         tsk->btrace_seq = 0;
 391 #endif
 392         tsk->splice_pipe = NULL;
 393         tsk->task_frag.page = NULL;
 394         tsk->wake_q.next = NULL;
 395
 396         account_kernel_stack(stack, 1);
 397
 398         kcov_task_init(tsk);
 399
 400         return tsk;
 401
 402 free_stack:
 403         free_thread_stack(stack);
 404 free_tsk:
 405         free_task_struct(tsk);
 406         return NULL;
 407 }
 408
 409 #ifdef CONFIG_MMU
 410 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 411 {
 412         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 413         struct rb_node **rb_link, *rb_parent;
 414         int retval;
 415         unsigned long charge;
 416
 417         uprobe_start_dup_mmap();
 418         if (down_write_killable(&oldmm->mmap_sem)) {
 419                 retval = -EINTR;
 420                 goto fail_uprobe_end;
 421         }
 422         flush_cache_dup_mm(oldmm);
 423         uprobe_dup_mmap(oldmm, mm);
 424         /*
 425          * Not linked in yet - no deadlock potential:
 426          */
 427         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 428
 429         /* No ordering required: file already has been exposed. */
 430         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 431
 432         mm->total_vm = oldmm->total_vm;
 433         mm->data_vm = oldmm->data_vm;
 434         mm->exec_vm = oldmm->exec_vm;
 435         mm->stack_vm = oldmm->stack_vm;
 436
 437         rb_link = &mm->mm_rb.rb_node;
 438         rb_parent = NULL;
 439         pprev = &mm->mmap;
 440         retval = ksm_fork(mm, oldmm);
 441         if (retval)
 442                 goto out;
 443         retval = khugepaged_fork(mm, oldmm);
 444         if (retval)
 445                 goto out;
 446
 447         prev = NULL;
 448         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 449                 struct file *file;
 450
 451                 if (mpnt->vm_flags & VM_DONTCOPY) {
 452                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 453                         continue;
 454                 }
 455                 charge = 0;
 456                 if (mpnt->vm_flags & VM_ACCOUNT) {
 457                         unsigned long len = vma_pages(mpnt);
 458
 459                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 460                                 goto fail_nomem;
 461                         charge = len;
 462                 }
 463                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 464                 if (!tmp)
 465                         goto fail_nomem;
 466                 *tmp = *mpnt;
 467                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 468                 retval = vma_dup_policy(mpnt, tmp);
 469                 if (retval)
 470                         goto fail_nomem_policy;
 471                 tmp->vm_mm = mm;
 472                 if (anon_vma_fork(tmp, mpnt))
 473                         goto fail_nomem_anon_vma_fork;
 474                 tmp->vm_flags &=
 475                         ~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
 476                 tmp->vm_next = tmp->vm_prev = NULL;
 477                 tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
 478                 file = tmp->vm_file;
 479                 if (file) {
 480                         struct inode *inode = file_inode(file);
 481                         struct address_space *mapping = file->f_mapping;
 482
 483                         get_file(file);
 484                         if (tmp->vm_flags & VM_DENYWRITE)
 485                                 atomic_dec(&inode->i_writecount);
 486                         i_mmap_lock_write(mapping);
 487                         if (tmp->vm_flags & VM_SHARED)
 488                                 atomic_inc(&mapping->i_mmap_writable);
 489                         flush_dcache_mmap_lock(mapping);
 490                         /* insert tmp into the share list, just after mpnt */
 491                         vma_interval_tree_insert_after(tmp, mpnt,
 492                                         &mapping->i_mmap);
 493                         flush_dcache_mmap_unlock(mapping);
 494                         i_mmap_unlock_write(mapping);
 495                 }
 496
 497                 /*
 498                  * Clear hugetlb-related page reserves for children. This only
 499                  * affects MAP_PRIVATE mappings. Faults generated by the child
 500                  * are not guaranteed to succeed, even if read-only
 501                  */
 502                 if (is_vm_hugetlb_page(tmp))
 503                         reset_vma_resv_huge_pages(tmp);
 504
 505                 /*
 506                  * Link in the new vma and copy the page table entries.
 507                  */
 508                 *pprev = tmp;
 509                 pprev = &tmp->vm_next;
 510                 tmp->vm_prev = prev;
 511                 prev = tmp;
 512
 513                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 514                 rb_link = &tmp->vm_rb.rb_right;
 515                 rb_parent = &tmp->vm_rb;
 516
 517                 mm->map_count++;
 518                 retval = copy_page_range(mm, oldmm, mpnt);
 519
 520                 if (tmp->vm_ops && tmp->vm_ops->open)
 521                         tmp->vm_ops->open(tmp);
 522
 523                 if (retval)
 524                         goto out;
 525         }
 526         /* a new mm has just been created */
 527         arch_dup_mmap(oldmm, mm);
 528         retval = 0;
 529 out:
 530         up_write(&mm->mmap_sem);
 531         flush_tlb_mm(oldmm);
 532         up_write(&oldmm->mmap_sem);
 533 fail_uprobe_end:
 534         uprobe_end_dup_mmap();
 535         return retval;
 536 fail_nomem_anon_vma_fork:
 537         mpol_put(vma_policy(tmp));
 538 fail_nomem_policy:
 539         kmem_cache_free(vm_area_cachep, tmp);
 540 fail_nomem:
 541         retval = -ENOMEM;
 542         vm_unacct_memory(charge);
 543         goto out;
 544 }
 545
 546 static inline int mm_alloc_pgd(struct mm_struct *mm)
 547 {
 548         mm->pgd = pgd_alloc(mm);
 549         if (unlikely(!mm->pgd))
 550                 return -ENOMEM;
 551         return 0;
 552 }
 553
 554 static inline void mm_free_pgd(struct mm_struct *mm)
 555 {
 556         pgd_free(mm, mm->pgd);
 557 }
 558 #else
 559 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 560 {
 561         down_write(&oldmm->mmap_sem);
 562         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 563         up_write(&oldmm->mmap_sem);
 564         return 0;
 565 }
 566 #define mm_alloc_pgd(mm)        (0)
 567 #define mm_free_pgd(mm)
 568 #endif /* CONFIG_MMU */
 569
 570 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 571
 572 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 573 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 574
 575 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 576
 577 static int __init coredump_filter_setup(char *s)
 578 {
 579         default_dump_filter =
 580                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 581                 MMF_DUMP_FILTER_MASK;
 582         return 1;
 583 }
 584
 585 __setup("coredump_filter=", coredump_filter_setup);
 586
 587 #include <linux/init_task.h>
 588
 589 static void mm_init_aio(struct mm_struct *mm)
 590 {
 591 #ifdef CONFIG_AIO
 592         spin_lock_init(&mm->ioctx_lock);
 593         mm->ioctx_table = NULL;
 594 #endif
 595 }
 596
 597 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 598 {
 599 #ifdef CONFIG_MEMCG
 600         mm->owner = p;
 601 #endif
 602 }
 603
 604 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 605 {
 606         mm->mmap = NULL;
 607         mm->mm_rb = RB_ROOT;
 608         mm->vmacache_seqnum = 0;
 609         atomic_set(&mm->mm_users, 1);
 610         atomic_set(&mm->mm_count, 1);
 611         init_rwsem(&mm->mmap_sem);
 612         INIT_LIST_HEAD(&mm->mmlist);
 613         mm->core_state = NULL;
 614         atomic_long_set(&mm->nr_ptes, 0);
 615         mm_nr_pmds_init(mm);
 616         mm->map_count = 0;
 617         mm->locked_vm = 0;
 618         mm->pinned_vm = 0;
 619         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 620         spin_lock_init(&mm->page_table_lock);
 621         mm_init_cpumask(mm);
 622         mm_init_aio(mm);
 623         mm_init_owner(mm, p);
 624         mmu_notifier_mm_init(mm);
 625         clear_tlb_flush_pending(mm);
 626 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 627         mm->pmd_huge_pte = NULL;
 628 #endif
 629
 630         if (current->mm) {
 631                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 632                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 633         } else {
 634                 mm->flags = default_dump_filter;
 635                 mm->def_flags = 0;
 636         }
 637
 638         if (mm_alloc_pgd(mm))
 639                 goto fail_nopgd;
 640
 641         if (init_new_context(p, mm))
 642                 goto fail_nocontext;
 643
 644         return mm;
 645
 646 fail_nocontext:
 647         mm_free_pgd(mm);
 648 fail_nopgd:
 649         free_mm(mm);
 650         return NULL;
 651 }
 652
 653 static void check_mm(struct mm_struct *mm)
 654 {
 655         int i;
 656
 657         for (i = 0; i < NR_MM_COUNTERS; i++) {
 658                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 659
 660                 if (unlikely(x))
 661                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 662                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 663         }
 664
 665         if (atomic_long_read(&mm->nr_ptes))
 666                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 667                                 atomic_long_read(&mm->nr_ptes));
 668         if (mm_nr_pmds(mm))
 669                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 670                                 mm_nr_pmds(mm));
 671
 672 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 673         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 674 #endif
 675 }
 676
 677 /*
 678  * Allocate and initialize an mm_struct.
 679  */
 680 struct mm_struct *mm_alloc(void)
 681 {
 682         struct mm_struct *mm;
 683
 684         mm = allocate_mm();
 685         if (!mm)
 686                 return NULL;
 687
 688         memset(mm, 0, sizeof(*mm));
 689         return mm_init(mm, current);
 690 }
 691
 692 /*
 693  * Called when the last reference to the mm
 694  * is dropped: either by a lazy thread or by
 695  * mmput. Free the page directory and the mm.
 696  */
 697 void __mmdrop(struct mm_struct *mm)
 698 {
 699         BUG_ON(mm == &init_mm);
 700         mm_free_pgd(mm);
 701         destroy_context(mm);
 702         mmu_notifier_mm_destroy(mm);
 703         check_mm(mm);
 704         free_mm(mm);
 705 }
 706 EXPORT_SYMBOL_GPL(__mmdrop);
 707
 708 static inline void __mmput(struct mm_struct *mm)
 709 {
 710         VM_BUG_ON(atomic_read(&mm->mm_users));
 711
 712         uprobe_clear_state(mm);
 713         exit_aio(mm);
 714         ksm_exit(mm);
 715         khugepaged_exit(mm); /* must run before exit_mmap */
 716         exit_mmap(mm);
 717         set_mm_exe_file(mm, NULL);
 718         if (!list_empty(&mm->mmlist)) {
 719                 spin_lock(&mmlist_lock);
 720                 list_del(&mm->mmlist);
 721                 spin_unlock(&mmlist_lock);
 722         }
 723         if (mm->binfmt)
 724                 module_put(mm->binfmt->module);
 725         mmdrop(mm);
 726 }
 727
 728 /*
 729  * Decrement the use count and release all resources for an mm.
 730  */
 731 void mmput(struct mm_struct *mm)
 732 {
 733         might_sleep();
 734
 735         if (atomic_dec_and_test(&mm->mm_users))
 736                 __mmput(mm);
 737 }
 738 EXPORT_SYMBOL_GPL(mmput);
 739
 740 #ifdef CONFIG_MMU
 741 static void mmput_async_fn(struct work_struct *work)
 742 {
 743         struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
 744         __mmput(mm);
 745 }
 746
 747 void mmput_async(struct mm_struct *mm)
 748 {
 749         if (atomic_dec_and_test(&mm->mm_users)) {
 750                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
 751                 schedule_work(&mm->async_put_work);
 752         }
 753 }
 754 #endif
 755
 756 /**
 757  * set_mm_exe_file - change a reference to the mm's executable file
 758  *
 759  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 760  *
 761  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 762  * invocations: in mmput() nobody alive left, in execve task is single
 763  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 764  * mm->exe_file, but does so without using set_mm_exe_file() in order
 765  * to do avoid the need for any locks.
 766  */
 767 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 768 {
 769         struct file *old_exe_file;
 770
 771         /*
 772          * It is safe to dereference the exe_file without RCU as
 773          * this function is only called if nobody else can access
 774          * this mm -- see comment above for justification.
 775          */
 776         old_exe_file = rcu_dereference_raw(mm->exe_file);
 777
 778         if (new_exe_file)
 779                 get_file(new_exe_file);
 780         rcu_assign_pointer(mm->exe_file, new_exe_file);
 781         if (old_exe_file)
 782                 fput(old_exe_file);
 783 }
 784
 785 /**
 786  * get_mm_exe_file - acquire a reference to the mm's executable file
 787  *
 788  * Returns %NULL if mm has no associated executable file.
 789  * User must release file via fput().
 790  */
 791 struct file *get_mm_exe_file(struct mm_struct *mm)
 792 {
 793         struct file *exe_file;
 794
 795         rcu_read_lock();
 796         exe_file = rcu_dereference(mm->exe_file);
 797         if (exe_file && !get_file_rcu(exe_file))
 798                 exe_file = NULL;
 799         rcu_read_unlock();
 800         return exe_file;
 801 }
 802 EXPORT_SYMBOL(get_mm_exe_file);
 803
 804 /**
 805  * get_task_mm - acquire a reference to the task's mm
 806  *
 807  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 808  * this kernel workthread has transiently adopted a user mm with use_mm,
 809  * to do its AIO) is not set and if so returns a reference to it, after
 810  * bumping up the use count.  User must release the mm via mmput()
 811  * after use.  Typically used by /proc and ptrace.
 812  */
 813 struct mm_struct *get_task_mm(struct task_struct *task)
 814 {
 815         struct mm_struct *mm;
 816
 817         task_lock(task);
 818         mm = task->mm;
 819         if (mm) {
 820                 if (task->flags & PF_KTHREAD)
 821                         mm = NULL;
 822                 else
 823                         atomic_inc(&mm->mm_users);
 824         }
 825         task_unlock(task);
 826         return mm;
 827 }
 828 EXPORT_SYMBOL_GPL(get_task_mm);
 829
 830 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 831 {
 832         struct mm_struct *mm;
 833         int err;
 834
 835         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 836         if (err)
 837                 return ERR_PTR(err);
 838
 839         mm = get_task_mm(task);
 840         if (mm && mm != current->mm &&
 841                         !ptrace_may_access(task, mode)) {
 842                 mmput(mm);
 843                 mm = ERR_PTR(-EACCES);
 844         }
 845         mutex_unlock(&task->signal->cred_guard_mutex);
 846
 847         return mm;
 848 }
 849
 850 static void complete_vfork_done(struct task_struct *tsk)
 851 {
 852         struct completion *vfork;
 853
 854         task_lock(tsk);
 855         vfork = tsk->vfork_done;
 856         if (likely(vfork)) {
 857                 tsk->vfork_done = NULL;
 858                 complete(vfork);
 859         }
 860         task_unlock(tsk);
 861 }
 862
 863 static int wait_for_vfork_done(struct task_struct *child,
 864                                 struct completion *vfork)
 865 {
 866         int killed;
 867
 868         freezer_do_not_count();
 869         killed = wait_for_completion_killable(vfork);
 870         freezer_count();
 871
 872         if (killed) {
 873                 task_lock(child);
 874                 child->vfork_done = NULL;
 875                 task_unlock(child);
 876         }
 877
 878         put_task_struct(child);
 879         return killed;
 880 }
 881
 882 /* Please note the differences between mmput and mm_release.
 883  * mmput is called whenever we stop holding onto a mm_struct,
 884  * error success whatever.
 885  *
 886  * mm_release is called after a mm_struct has been removed
 887  * from the current process.
 888  *
 889  * This difference is important for error handling, when we
 890  * only half set up a mm_struct for a new process and need to restore
 891  * the old one.  Because we mmput the new mm_struct before
 892  * restoring the old one. . .
 893  * Eric Biederman 10 January 1998
 894  */
 895 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
 896 {
 897         /* Get rid of any futexes when releasing the mm */
 898 #ifdef CONFIG_FUTEX
 899         if (unlikely(tsk->robust_list)) {
 900                 exit_robust_list(tsk);
 901                 tsk->robust_list = NULL;
 902         }
 903 #ifdef CONFIG_COMPAT
 904         if (unlikely(tsk->compat_robust_list)) {
 905                 compat_exit_robust_list(tsk);
 906                 tsk->compat_robust_list = NULL;
 907         }
 908 #endif
 909         if (unlikely(!list_empty(&tsk->pi_state_list)))
 910                 exit_pi_state_list(tsk);
 911 #endif
 912
 913         uprobe_free_utask(tsk);
 914
 915         /* Get rid of any cached register state */
 916         deactivate_mm(tsk, mm);
 917
 918         /*
 919          * If we're exiting normally, clear a user-space tid field if
 920          * requested.  We leave this alone when dying by signal, to leave
 921          * the value intact in a core dump, and to save the unnecessary
 922          * trouble, say, a killed vfork parent shouldn't touch this mm.
 923          * Userland only wants this done for a sys_exit.
 924          */
 925         if (tsk->clear_child_tid) {
 926                 if (!(tsk->flags & PF_SIGNALED) &&
 927                     atomic_read(&mm->mm_users) > 1) {
 928                         /*
 929                          * We don't check the error code - if userspace has
 930                          * not set up a proper pointer then tough luck.
 931                          */
 932                         put_user(0, tsk->clear_child_tid);
 933                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
 934                                         1, NULL, NULL, 0);
 935                 }
 936                 tsk->clear_child_tid = NULL;
 937         }
 938
 939         /*
 940          * All done, finally we can wake up parent and return this mm to him.
 941          * Also kthread_stop() uses this completion for synchronization.
 942          */
 943         if (tsk->vfork_done)
 944                 complete_vfork_done(tsk);
 945 }
 946
 947 /*
 948  * Allocate a new mm structure and copy contents from the
 949  * mm structure of the passed in task structure.
 950  */
 951 static struct mm_struct *dup_mm(struct task_struct *tsk)
 952 {
 953         struct mm_struct *mm, *oldmm = current->mm;
 954         int err;
 955
 956         mm = allocate_mm();
 957         if (!mm)
 958                 goto fail_nomem;
 959
 960         memcpy(mm, oldmm, sizeof(*mm));
 961
 962         if (!mm_init(mm, tsk))
 963                 goto fail_nomem;
 964
 965         err = dup_mmap(mm, oldmm);
 966         if (err)
 967                 goto free_pt;
 968
 969         mm->hiwater_rss = get_mm_rss(mm);
 970         mm->hiwater_vm = mm->total_vm;
 971
 972         if (mm->binfmt && !try_module_get(mm->binfmt->module))
 973                 goto free_pt;
 974
 975         return mm;
 976
 977 free_pt:
 978         /* don't put binfmt in mmput, we haven't got module yet */
 979         mm->binfmt = NULL;
 980         mmput(mm);
 981
 982 fail_nomem:
 983         return NULL;
 984 }
 985
 986 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
 987 {
 988         struct mm_struct *mm, *oldmm;
 989         int retval;
 990
 991         tsk->min_flt = tsk->maj_flt = 0;
 992         tsk->nvcsw = tsk->nivcsw = 0;
 993 #ifdef CONFIG_DETECT_HUNG_TASK
 994         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
 995 #endif
 996
 997         tsk->mm = NULL;
 998         tsk->active_mm = NULL;
 999
1000         /*
1001          * Are we cloning a kernel thread?
1002          *
1003          * We need to steal a active VM for that..
1004          */
1005         oldmm = current->mm;
1006         if (!oldmm)
1007                 return 0;
1008
1009         /* initialize the new vmacache entries */
1010         vmacache_flush(tsk);
1011
1012         if (clone_flags & CLONE_VM) {
1013                 atomic_inc(&oldmm->mm_users);
1014                 mm = oldmm;
1015                 goto good_mm;
1016         }
1017
1018         retval = -ENOMEM;
1019         mm = dup_mm(tsk);
1020         if (!mm)
1021                 goto fail_nomem;
1022
1023 good_mm:
1024         tsk->mm = mm;
1025         tsk->active_mm = mm;
1026         return 0;
1027
1028 fail_nomem:
1029         return retval;
1030 }
1031
1032 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1033 {
1034         struct fs_struct *fs = current->fs;
1035         if (clone_flags & CLONE_FS) {
1036                 /* tsk->fs is already what we want */
1037                 spin_lock(&fs->lock);
1038                 if (fs->in_exec) {
1039                         spin_unlock(&fs->lock);
1040                         return -EAGAIN;
1041                 }
1042                 fs->users++;
1043                 spin_unlock(&fs->lock);
1044                 return 0;
1045         }
1046         tsk->fs = copy_fs_struct(fs);
1047         if (!tsk->fs)
1048                 return -ENOMEM;
1049         return 0;
1050 }
1051
1052 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1053 {
1054         struct files_struct *oldf, *newf;
1055         int error = 0;
1056
1057         /*
1058          * A background process may not have any files ...
1059          */
1060         oldf = current->files;
1061         if (!oldf)
1062                 goto out;
1063
1064         if (clone_flags & CLONE_FILES) {
1065                 atomic_inc(&oldf->count);
1066                 goto out;
1067         }
1068
1069         newf = dup_fd(oldf, &error);
1070         if (!newf)
1071                 goto out;
1072
1073         tsk->files = newf;
1074         error = 0;
1075 out:
1076         return error;
1077 }
1078
1079 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1080 {
1081 #ifdef CONFIG_BLOCK
1082         struct io_context *ioc = current->io_context;
1083         struct io_context *new_ioc;
1084
1085         if (!ioc)
1086                 return 0;
1087         /*
1088          * Share io context with parent, if CLONE_IO is set
1089          */
1090         if (clone_flags & CLONE_IO) {
1091                 ioc_task_link(ioc);
1092                 tsk->io_context = ioc;
1093         } else if (ioprio_valid(ioc->ioprio)) {
1094                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1095                 if (unlikely(!new_ioc))
1096                         return -ENOMEM;
1097
1098                 new_ioc->ioprio = ioc->ioprio;
1099                 put_io_context(new_ioc);
1100         }
1101 #endif
1102         return 0;
1103 }
1104
1105 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1106 {
1107         struct sighand_struct *sig;
1108
1109         if (clone_flags & CLONE_SIGHAND) {
1110                 atomic_inc(&current->sighand->count);
1111                 return 0;
1112         }
1113         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1114         rcu_assign_pointer(tsk->sighand, sig);
1115         if (!sig)
1116                 return -ENOMEM;
1117
1118         atomic_set(&sig->count, 1);
1119         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1120         return 0;
1121 }
1122
1123 void __cleanup_sighand(struct sighand_struct *sighand)
1124 {
1125         if (atomic_dec_and_test(&sighand->count)) {
1126                 signalfd_cleanup(sighand);
1127                 /*
1128                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1129                  * without an RCU grace period, see __lock_task_sighand().
1130                  */
1131                 kmem_cache_free(sighand_cachep, sighand);
1132         }
1133 }
1134
1135 /*
1136  * Initialize POSIX timer handling for a thread group.
1137  */
1138 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1139 {
1140         unsigned long cpu_limit;
1141
1142         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1143         if (cpu_limit != RLIM_INFINITY) {
1144                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1145                 sig->cputimer.running = true;
1146         }
1147
1148         /* The timer lists. */
1149         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1150         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1151         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1152 }
1153
1154 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1155 {
1156         struct signal_struct *sig;
1157
1158         if (clone_flags & CLONE_THREAD)
1159                 return 0;
1160
1161         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1162         tsk->signal = sig;
1163         if (!sig)
1164                 return -ENOMEM;
1165
1166         sig->nr_threads = 1;
1167         atomic_set(&sig->live, 1);
1168         atomic_set(&sig->sigcnt, 1);
1169
1170         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1171         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1172         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1173
1174         init_waitqueue_head(&sig->wait_chldexit);
1175         sig->curr_target = tsk;
1176         init_sigpending(&sig->shared_pending);
1177         INIT_LIST_HEAD(&sig->posix_timers);
1178         seqlock_init(&sig->stats_lock);
1179         prev_cputime_init(&sig->prev_cputime);
1180
1181         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1182         sig->real_timer.function = it_real_fn;
1183
1184         task_lock(current->group_leader);
1185         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1186         task_unlock(current->group_leader);
1187
1188         posix_cpu_timers_init_group(sig);
1189
1190         tty_audit_fork(sig);
1191         sched_autogroup_fork(sig);
1192
1193         sig->oom_score_adj = current->signal->oom_score_adj;
1194         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1195
1196         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1197                                    current->signal->is_child_subreaper;
1198
1199         mutex_init(&sig->cred_guard_mutex);
1200
1201         return 0;
1202 }
1203
1204 static void copy_seccomp(struct task_struct *p)
1205 {
1206 #ifdef CONFIG_SECCOMP
1207         /*
1208          * Must be called with sighand->lock held, which is common to
1209          * all threads in the group. Holding cred_guard_mutex is not
1210          * needed because this new task is not yet running and cannot
1211          * be racing exec.
1212          */
1213         assert_spin_locked(&current->sighand->siglock);
1214
1215         /* Ref-count the new filter user, and assign it. */
1216         get_seccomp_filter(current);
1217         p->seccomp = current->seccomp;
1218
1219         /*
1220          * Explicitly enable no_new_privs here in case it got set
1221          * between the task_struct being duplicated and holding the
1222          * sighand lock. The seccomp state and nnp must be in sync.
1223          */
1224         if (task_no_new_privs(current))
1225                 task_set_no_new_privs(p);
1226
1227         /*
1228          * If the parent gained a seccomp mode after copying thread
1229          * flags and between before we held the sighand lock, we have
1230          * to manually enable the seccomp thread flag here.
1231          */
1232         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1233                 set_tsk_thread_flag(p, TIF_SECCOMP);
1234 #endif
1235 }
1236
1237 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1238 {
1239         current->clear_child_tid = tidptr;
1240
1241         return task_pid_vnr(current);
1242 }
1243
1244 static void rt_mutex_init_task(struct task_struct *p)
1245 {
1246         raw_spin_lock_init(&p->pi_lock);
1247 #ifdef CONFIG_RT_MUTEXES
1248         p->pi_waiters = RB_ROOT;
1249         p->pi_waiters_leftmost = NULL;
1250         p->pi_blocked_on = NULL;
1251 #endif
1252 }
1253
1254 /*
1255  * Initialize POSIX timer handling for a single task.
1256  */
1257 static void posix_cpu_timers_init(struct task_struct *tsk)
1258 {
1259         tsk->cputime_expires.prof_exp = 0;
1260         tsk->cputime_expires.virt_exp = 0;
1261         tsk->cputime_expires.sched_exp = 0;
1262         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1263         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1264         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1265 }
1266
1267 static inline void
1268 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1269 {
1270          task->pids[type].pid = pid;
1271 }
1272
1273 /*
1274  * This creates a new process as a copy of the old one,
1275  * but does not actually start it yet.
1276  *
1277  * It copies the registers, and all the appropriate
1278  * parts of the process environment (as per the clone
1279  * flags). The actual kick-off is left to the caller.
1280  */
1281 static struct task_struct *copy_process(unsigned long clone_flags,
1282                                         unsigned long stack_start,
1283                                         unsigned long stack_size,
1284                                         int __user *child_tidptr,
1285                                         struct pid *pid,
1286                                         int trace,
1287                                         unsigned long tls,
1288                                         int node)
1289 {
1290         int retval;
1291         struct task_struct *p;
1292
1293         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1294                 return ERR_PTR(-EINVAL);
1295
1296         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1297                 return ERR_PTR(-EINVAL);
1298
1299         /*
1300          * Thread groups must share signals as well, and detached threads
1301          * can only be started up within the thread group.
1302          */
1303         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1304                 return ERR_PTR(-EINVAL);
1305
1306         /*
1307          * Shared signal handlers imply shared VM. By way of the above,
1308          * thread groups also imply shared VM. Blocking this case allows
1309          * for various simplifications in other code.
1310          */
1311         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1312                 return ERR_PTR(-EINVAL);
1313
1314         /*
1315          * Siblings of global init remain as zombies on exit since they are
1316          * not reaped by their parent (swapper). To solve this and to avoid
1317          * multi-rooted process trees, prevent global and container-inits
1318          * from creating siblings.
1319          */
1320         if ((clone_flags & CLONE_PARENT) &&
1321                                 current->signal->flags & SIGNAL_UNKILLABLE)
1322                 return ERR_PTR(-EINVAL);
1323
1324         /*
1325          * If the new process will be in a different pid or user namespace
1326          * do not allow it to share a thread group with the forking task.
1327          */
1328         if (clone_flags & CLONE_THREAD) {
1329                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1330                     (task_active_pid_ns(current) !=
1331                                 current->nsproxy->pid_ns_for_children))
1332                         return ERR_PTR(-EINVAL);
1333         }
1334
1335         retval = security_task_create(clone_flags);
1336         if (retval)
1337                 goto fork_out;
1338
1339         retval = -ENOMEM;
1340         p = dup_task_struct(current, node);
1341         if (!p)
1342                 goto fork_out;
1343
1344         ftrace_graph_init_task(p);
1345
1346         rt_mutex_init_task(p);
1347
1348 #ifdef CONFIG_PROVE_LOCKING
1349         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1350         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1351 #endif
1352         retval = -EAGAIN;
1353         if (atomic_read(&p->real_cred->user->processes) >=
1354                         task_rlimit(p, RLIMIT_NPROC)) {
1355                 if (p->real_cred->user != INIT_USER &&
1356                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1357                         goto bad_fork_free;
1358         }
1359         current->flags &= ~PF_NPROC_EXCEEDED;
1360
1361         retval = copy_creds(p, clone_flags);
1362         if (retval < 0)
1363                 goto bad_fork_free;
1364
1365         /*
1366          * If multiple threads are within copy_process(), then this check
1367          * triggers too late. This doesn't hurt, the check is only there
1368          * to stop root fork bombs.
1369          */
1370         retval = -EAGAIN;
1371         if (nr_threads >= max_threads)
1372                 goto bad_fork_cleanup_count;
1373
1374         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1375         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1376         p->flags |= PF_FORKNOEXEC;
1377         INIT_LIST_HEAD(&p->children);
1378         INIT_LIST_HEAD(&p->sibling);
1379         rcu_copy_process(p);
1380         p->vfork_done = NULL;
1381         spin_lock_init(&p->alloc_lock);
1382
1383         init_sigpending(&p->pending);
1384
1385         p->utime = p->stime = p->gtime = 0;
1386         p->utimescaled = p->stimescaled = 0;
1387         prev_cputime_init(&p->prev_cputime);
1388
1389 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1390         seqcount_init(&p->vtime_seqcount);
1391         p->vtime_snap = 0;
1392         p->vtime_snap_whence = VTIME_INACTIVE;
1393 #endif
1394
1395 #if defined(SPLIT_RSS_COUNTING)
1396         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1397 #endif
1398
1399         p->default_timer_slack_ns = current->timer_slack_ns;
1400
1401         task_io_accounting_init(&p->ioac);
1402         acct_clear_integrals(p);
1403
1404         posix_cpu_timers_init(p);
1405
1406         p->start_time = ktime_get_ns();
1407         p->real_start_time = ktime_get_boot_ns();
1408         p->io_context = NULL;
1409         p->audit_context = NULL;
1410         threadgroup_change_begin(current);
1411         cgroup_fork(p);
1412 #ifdef CONFIG_NUMA
1413         p->mempolicy = mpol_dup(p->mempolicy);
1414         if (IS_ERR(p->mempolicy)) {
1415                 retval = PTR_ERR(p->mempolicy);
1416                 p->mempolicy = NULL;
1417                 goto bad_fork_cleanup_threadgroup_lock;
1418         }
1419 #endif
1420 #ifdef CONFIG_CPUSETS
1421         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1422         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1423         seqcount_init(&p->mems_allowed_seq);
1424 #endif
1425 #ifdef CONFIG_TRACE_IRQFLAGS
1426         p->irq_events = 0;
1427         p->hardirqs_enabled = 0;
1428         p->hardirq_enable_ip = 0;
1429         p->hardirq_enable_event = 0;
1430         p->hardirq_disable_ip = _THIS_IP_;
1431         p->hardirq_disable_event = 0;
1432         p->softirqs_enabled = 1;
1433         p->softirq_enable_ip = _THIS_IP_;
1434         p->softirq_enable_event = 0;
1435         p->softirq_disable_ip = 0;
1436         p->softirq_disable_event = 0;
1437         p->hardirq_context = 0;
1438         p->softirq_context = 0;
1439 #endif
1440
1441         p->pagefault_disabled = 0;
1442
1443 #ifdef CONFIG_LOCKDEP
1444         p->lockdep_depth = 0; /* no locks held yet */
1445         p->curr_chain_key = 0;
1446         p->lockdep_recursion = 0;
1447 #endif
1448
1449 #ifdef CONFIG_DEBUG_MUTEXES
1450         p->blocked_on = NULL; /* not blocked yet */
1451 #endif
1452 #ifdef CONFIG_BCACHE
1453         p->sequential_io        = 0;
1454         p->sequential_io_avg    = 0;
1455 #endif
1456
1457         /* Perform scheduler related setup. Assign this task to a CPU. */
1458         retval = sched_fork(clone_flags, p);
1459         if (retval)
1460                 goto bad_fork_cleanup_policy;
1461
1462         retval = perf_event_init_task(p);
1463         if (retval)
1464                 goto bad_fork_cleanup_policy;
1465         retval = audit_alloc(p);
1466         if (retval)
1467                 goto bad_fork_cleanup_perf;
1468         /* copy all the process information */
1469         shm_init_task(p);
1470         retval = copy_semundo(clone_flags, p);
1471         if (retval)
1472                 goto bad_fork_cleanup_audit;
1473         retval = copy_files(clone_flags, p);
1474         if (retval)
1475                 goto bad_fork_cleanup_semundo;
1476         retval = copy_fs(clone_flags, p);
1477         if (retval)
1478                 goto bad_fork_cleanup_files;
1479         retval = copy_sighand(clone_flags, p);
1480         if (retval)
1481                 goto bad_fork_cleanup_fs;
1482         retval = copy_signal(clone_flags, p);
1483         if (retval)
1484                 goto bad_fork_cleanup_sighand;
1485         retval = copy_mm(clone_flags, p);
1486         if (retval)
1487                 goto bad_fork_cleanup_signal;
1488         retval = copy_namespaces(clone_flags, p);
1489         if (retval)
1490                 goto bad_fork_cleanup_mm;
1491         retval = copy_io(clone_flags, p);
1492         if (retval)
1493                 goto bad_fork_cleanup_namespaces;
1494         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1495         if (retval)
1496                 goto bad_fork_cleanup_io;
1497
1498         if (pid != &init_struct_pid) {
1499                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1500                 if (IS_ERR(pid)) {
1501                         retval = PTR_ERR(pid);
1502                         goto bad_fork_cleanup_thread;
1503                 }
1504         }
1505
1506         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1507         /*
1508          * Clear TID on mm_release()?
1509          */
1510         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1511 #ifdef CONFIG_BLOCK
1512         p->plug = NULL;
1513 #endif
1514 #ifdef CONFIG_FUTEX
1515         p->robust_list = NULL;
1516 #ifdef CONFIG_COMPAT
1517         p->compat_robust_list = NULL;
1518 #endif
1519         INIT_LIST_HEAD(&p->pi_state_list);
1520         p->pi_state_cache = NULL;
1521 #endif
1522         /*
1523          * sigaltstack should be cleared when sharing the same VM
1524          */
1525         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1526                 sas_ss_reset(p);
1527
1528         /*
1529          * Syscall tracing and stepping should be turned off in the
1530          * child regardless of CLONE_PTRACE.
1531          */
1532         user_disable_single_step(p);
1533         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1534 #ifdef TIF_SYSCALL_EMU
1535         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1536 #endif
1537         clear_all_latency_tracing(p);
1538
1539         /* ok, now we should be set up.. */
1540         p->pid = pid_nr(pid);
1541         if (clone_flags & CLONE_THREAD) {
1542                 p->exit_signal = -1;
1543                 p->group_leader = current->group_leader;
1544                 p->tgid = current->tgid;
1545         } else {
1546                 if (clone_flags & CLONE_PARENT)
1547                         p->exit_signal = current->group_leader->exit_signal;
1548                 else
1549                         p->exit_signal = (clone_flags & CSIGNAL);
1550                 p->group_leader = p;
1551                 p->tgid = p->pid;
1552         }
1553
1554         p->nr_dirtied = 0;
1555         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1556         p->dirty_paused_when = 0;
1557
1558         p->pdeath_signal = 0;
1559         INIT_LIST_HEAD(&p->thread_group);
1560         p->task_works = NULL;
1561
1562         /*
1563          * Ensure that the cgroup subsystem policies allow the new process to be
1564          * forked. It should be noted the the new process's css_set can be changed
1565          * between here and cgroup_post_fork() if an organisation operation is in
1566          * progress.
1567          */
1568         retval = cgroup_can_fork(p);
1569         if (retval)
1570                 goto bad_fork_free_pid;
1571
1572         /*
1573          * Make it visible to the rest of the system, but dont wake it up yet.
1574          * Need tasklist lock for parent etc handling!
1575          */
1576         write_lock_irq(&tasklist_lock);
1577
1578         /* CLONE_PARENT re-uses the old parent */
1579         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1580                 p->real_parent = current->real_parent;
1581                 p->parent_exec_id = current->parent_exec_id;
1582         } else {
1583                 p->real_parent = current;
1584                 p->parent_exec_id = current->self_exec_id;
1585         }
1586
1587         spin_lock(&current->sighand->siglock);
1588
1589         /*
1590          * Copy seccomp details explicitly here, in case they were changed
1591          * before holding sighand lock.
1592          */
1593         copy_seccomp(p);
1594
1595         /*
1596          * Process group and session signals need to be delivered to just the
1597          * parent before the fork or both the parent and the child after the
1598          * fork. Restart if a signal comes in before we add the new process to
1599          * it's process group.
1600          * A fatal signal pending means that current will exit, so the new
1601          * thread can't slip out of an OOM kill (or normal SIGKILL).
1602         */
1603         recalc_sigpending();
1604         if (signal_pending(current)) {
1605                 spin_unlock(&current->sighand->siglock);
1606                 write_unlock_irq(&tasklist_lock);
1607                 retval = -ERESTARTNOINTR;
1608                 goto bad_fork_cancel_cgroup;
1609         }
1610
1611         if (likely(p->pid)) {
1612                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1613
1614                 init_task_pid(p, PIDTYPE_PID, pid);
1615                 if (thread_group_leader(p)) {
1616                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1617                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1618
1619                         if (is_child_reaper(pid)) {
1620                                 ns_of_pid(pid)->child_reaper = p;
1621                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1622                         }
1623
1624                         p->signal->leader_pid = pid;
1625                         p->signal->tty = tty_kref_get(current->signal->tty);
1626                         list_add_tail(&p->sibling, &p->real_parent->children);
1627                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1628                         attach_pid(p, PIDTYPE_PGID);
1629                         attach_pid(p, PIDTYPE_SID);
1630                         __this_cpu_inc(process_counts);
1631                 } else {
1632                         current->signal->nr_threads++;
1633                         atomic_inc(&current->signal->live);
1634                         atomic_inc(&current->signal->sigcnt);
1635                         list_add_tail_rcu(&p->thread_group,
1636                                           &p->group_leader->thread_group);
1637                         list_add_tail_rcu(&p->thread_node,
1638                                           &p->signal->thread_head);
1639                 }
1640                 attach_pid(p, PIDTYPE_PID);
1641                 nr_threads++;
1642         }
1643
1644         total_forks++;
1645         spin_unlock(&current->sighand->siglock);
1646         syscall_tracepoint_update(p);
1647         write_unlock_irq(&tasklist_lock);
1648
1649         proc_fork_connector(p);
1650         cgroup_post_fork(p);
1651         threadgroup_change_end(current);
1652         perf_event_fork(p);
1653
1654         trace_task_newtask(p, clone_flags);
1655         uprobe_copy_process(p, clone_flags);
1656
1657         return p;
1658
1659 bad_fork_cancel_cgroup:
1660         cgroup_cancel_fork(p);
1661 bad_fork_free_pid:
1662         if (pid != &init_struct_pid)
1663                 free_pid(pid);
1664 bad_fork_cleanup_thread:
1665         exit_thread(p);
1666 bad_fork_cleanup_io:
1667         if (p->io_context)
1668                 exit_io_context(p);
1669 bad_fork_cleanup_namespaces:
1670         exit_task_namespaces(p);
1671 bad_fork_cleanup_mm:
1672         if (p->mm)
1673                 mmput(p->mm);
1674 bad_fork_cleanup_signal:
1675         if (!(clone_flags & CLONE_THREAD))
1676                 free_signal_struct(p->signal);
1677 bad_fork_cleanup_sighand:
1678         __cleanup_sighand(p->sighand);
1679 bad_fork_cleanup_fs:
1680         exit_fs(p); /* blocking */
1681 bad_fork_cleanup_files:
1682         exit_files(p); /* blocking */
1683 bad_fork_cleanup_semundo:
1684         exit_sem(p);
1685 bad_fork_cleanup_audit:
1686         audit_free(p);
1687 bad_fork_cleanup_perf:
1688         perf_event_free_task(p);
1689 bad_fork_cleanup_policy:
1690 #ifdef CONFIG_NUMA
1691         mpol_put(p->mempolicy);
1692 bad_fork_cleanup_threadgroup_lock:
1693 #endif
1694         threadgroup_change_end(current);
1695         delayacct_tsk_free(p);
1696 bad_fork_cleanup_count:
1697         atomic_dec(&p->cred->user->processes);
1698         exit_creds(p);
1699 bad_fork_free:
1700         free_task(p);
1701 fork_out:
1702         return ERR_PTR(retval);
1703 }
1704
1705 static inline void init_idle_pids(struct pid_link *links)
1706 {
1707         enum pid_type type;
1708
1709         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1710                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1711                 links[type].pid = &init_struct_pid;
1712         }
1713 }
1714
1715 struct task_struct *fork_idle(int cpu)
1716 {
1717         struct task_struct *task;
1718         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1719                             cpu_to_node(cpu));
1720         if (!IS_ERR(task)) {
1721                 init_idle_pids(task->pids);
1722                 init_idle(task, cpu);
1723         }
1724
1725         return task;
1726 }
1727
1728 /*
1729  *  Ok, this is the main fork-routine.
1730  *
1731  * It copies the process, and if successful kick-starts
1732  * it and waits for it to finish using the VM if required.
1733  */
1734 long _do_fork(unsigned long clone_flags,
1735               unsigned long stack_start,
1736               unsigned long stack_size,
1737               int __user *parent_tidptr,
1738               int __user *child_tidptr,
1739               unsigned long tls)
1740 {
1741         struct task_struct *p;
1742         int trace = 0;
1743         long nr;
1744
1745         /*
1746          * Determine whether and which event to report to ptracer.  When
1747          * called from kernel_thread or CLONE_UNTRACED is explicitly
1748          * requested, no event is reported; otherwise, report if the event
1749          * for the type of forking is enabled.
1750          */
1751         if (!(clone_flags & CLONE_UNTRACED)) {
1752                 if (clone_flags & CLONE_VFORK)
1753                         trace = PTRACE_EVENT_VFORK;
1754                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1755                         trace = PTRACE_EVENT_CLONE;
1756                 else
1757                         trace = PTRACE_EVENT_FORK;
1758
1759                 if (likely(!ptrace_event_enabled(current, trace)))
1760                         trace = 0;
1761         }
1762
1763         p = copy_process(clone_flags, stack_start, stack_size,
1764                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
1765         /*
1766          * Do this prior waking up the new thread - the thread pointer
1767          * might get invalid after that point, if the thread exits quickly.
1768          */
1769         if (!IS_ERR(p)) {
1770                 struct completion vfork;
1771                 struct pid *pid;
1772
1773                 trace_sched_process_fork(current, p);
1774
1775                 pid = get_task_pid(p, PIDTYPE_PID);
1776                 nr = pid_vnr(pid);
1777
1778                 if (clone_flags & CLONE_PARENT_SETTID)
1779                         put_user(nr, parent_tidptr);
1780
1781                 if (clone_flags & CLONE_VFORK) {
1782                         p->vfork_done = &vfork;
1783                         init_completion(&vfork);
1784                         get_task_struct(p);
1785                 }
1786
1787                 wake_up_new_task(p);
1788
1789                 /* forking complete and child started to run, tell ptracer */
1790                 if (unlikely(trace))
1791                         ptrace_event_pid(trace, pid);
1792
1793                 if (clone_flags & CLONE_VFORK) {
1794                         if (!wait_for_vfork_done(p, &vfork))
1795                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1796                 }
1797
1798                 put_pid(pid);
1799         } else {
1800                 nr = PTR_ERR(p);
1801         }
1802         return nr;
1803 }
1804
1805 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1806 /* For compatibility with architectures that call do_fork directly rather than
1807  * using the syscall entry points below. */
1808 long do_fork(unsigned long clone_flags,
1809               unsigned long stack_start,
1810               unsigned long stack_size,
1811               int __user *parent_tidptr,
1812               int __user *child_tidptr)
1813 {
1814         return _do_fork(clone_flags, stack_start, stack_size,
1815                         parent_tidptr, child_tidptr, 0);
1816 }
1817 #endif
1818
1819 /*
1820  * Create a kernel thread.
1821  */
1822 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1823 {
1824         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1825                 (unsigned long)arg, NULL, NULL, 0);
1826 }
1827
1828 #ifdef __ARCH_WANT_SYS_FORK
1829 SYSCALL_DEFINE0(fork)
1830 {
1831 #ifdef CONFIG_MMU
1832         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1833 #else
1834         /* can not support in nommu mode */
1835         return -EINVAL;
1836 #endif
1837 }
1838 #endif
1839
1840 #ifdef __ARCH_WANT_SYS_VFORK
1841 SYSCALL_DEFINE0(vfork)
1842 {
1843         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1844                         0, NULL, NULL, 0);
1845 }
1846 #endif
1847
1848 #ifdef __ARCH_WANT_SYS_CLONE
1849 #ifdef CONFIG_CLONE_BACKWARDS
1850 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1851                  int __user *, parent_tidptr,
1852                  unsigned long, tls,
1853                  int __user *, child_tidptr)
1854 #elif defined(CONFIG_CLONE_BACKWARDS2)
1855 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1856                  int __user *, parent_tidptr,
1857                  int __user *, child_tidptr,
1858                  unsigned long, tls)
1859 #elif defined(CONFIG_CLONE_BACKWARDS3)
1860 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1861                 int, stack_size,
1862                 int __user *, parent_tidptr,
1863                 int __user *, child_tidptr,
1864                 unsigned long, tls)
1865 #else
1866 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1867                  int __user *, parent_tidptr,
1868                  int __user *, child_tidptr,
1869                  unsigned long, tls)
1870 #endif
1871 {
1872         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
1873 }
1874 #endif
1875
1876 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1877 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1878 #endif
1879
1880 static void sighand_ctor(void *data)
1881 {
1882         struct sighand_struct *sighand = data;
1883
1884         spin_lock_init(&sighand->siglock);
1885         init_waitqueue_head(&sighand->signalfd_wqh);
1886 }
1887
1888 void __init proc_caches_init(void)
1889 {
1890         sighand_cachep = kmem_cache_create("sighand_cache",
1891                         sizeof(struct sighand_struct), 0,
1892                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1893                         SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
1894         signal_cachep = kmem_cache_create("signal_cache",
1895                         sizeof(struct signal_struct), 0,
1896                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1897                         NULL);
1898         files_cachep = kmem_cache_create("files_cache",
1899                         sizeof(struct files_struct), 0,
1900                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1901                         NULL);
1902         fs_cachep = kmem_cache_create("fs_cache",
1903                         sizeof(struct fs_struct), 0,
1904                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1905                         NULL);
1906         /*
1907          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1908          * whole struct cpumask for the OFFSTACK case. We could change
1909          * this to *only* allocate as much of it as required by the
1910          * maximum number of CPU's we can ever have.  The cpumask_allocation
1911          * is at the end of the structure, exactly for that reason.
1912          */
1913         mm_cachep = kmem_cache_create("mm_struct",
1914                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1915                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1916                         NULL);
1917         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
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|CLONE_NEWCGROUP))
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 = MIN_THREADS;
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         set_max_threads(threads);
2158
2159         return 0;
2160 }