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/sched/autogroup.h>
  16 #include <linux/init.h>
  17 #include <linux/unistd.h>
  18 #include <linux/module.h>
  19 #include <linux/vmalloc.h>
  20 #include <linux/completion.h>
  21 #include <linux/personality.h>
  22 #include <linux/mempolicy.h>
  23 #include <linux/sem.h>
  24 #include <linux/file.h>
  25 #include <linux/fdtable.h>
  26 #include <linux/iocontext.h>
  27 #include <linux/key.h>
  28 #include <linux/binfmts.h>
  29 #include <linux/mman.h>
  30 #include <linux/mmu_notifier.h>
  31 #include <linux/fs.h>
  32 #include <linux/mm.h>
  33 #include <linux/vmacache.h>
  34 #include <linux/nsproxy.h>
  35 #include <linux/capability.h>
  36 #include <linux/cpu.h>
  37 #include <linux/cgroup.h>
  38 #include <linux/security.h>
  39 #include <linux/hugetlb.h>
  40 #include <linux/seccomp.h>
  41 #include <linux/swap.h>
  42 #include <linux/syscalls.h>
  43 #include <linux/jiffies.h>
  44 #include <linux/futex.h>
  45 #include <linux/compat.h>
  46 #include <linux/kthread.h>
  47 #include <linux/task_io_accounting_ops.h>
  48 #include <linux/rcupdate.h>
  49 #include <linux/ptrace.h>
  50 #include <linux/mount.h>
  51 #include <linux/audit.h>
  52 #include <linux/memcontrol.h>
  53 #include <linux/ftrace.h>
  54 #include <linux/proc_fs.h>
  55 #include <linux/profile.h>
  56 #include <linux/rmap.h>
  57 #include <linux/ksm.h>
  58 #include <linux/acct.h>
  59 #include <linux/userfaultfd_k.h>
  60 #include <linux/tsacct_kern.h>
  61 #include <linux/cn_proc.h>
  62 #include <linux/freezer.h>
  63 #include <linux/delayacct.h>
  64 #include <linux/taskstats_kern.h>
  65 #include <linux/random.h>
  66 #include <linux/tty.h>
  67 #include <linux/blkdev.h>
  68 #include <linux/fs_struct.h>
  69 #include <linux/magic.h>
  70 #include <linux/perf_event.h>
  71 #include <linux/posix-timers.h>
  72 #include <linux/user-return-notifier.h>
  73 #include <linux/oom.h>
  74 #include <linux/khugepaged.h>
  75 #include <linux/signalfd.h>
  76 #include <linux/uprobes.h>
  77 #include <linux/aio.h>
  78 #include <linux/compiler.h>
  79 #include <linux/sysctl.h>
  80 #include <linux/kcov.h>
  81
  82 #include <asm/pgtable.h>
  83 #include <asm/pgalloc.h>
  84 #include <linux/uaccess.h>
  85 #include <asm/mmu_context.h>
  86 #include <asm/cacheflush.h>
  87 #include <asm/tlbflush.h>
  88
  89 #include <trace/events/sched.h>
  90
  91 #define CREATE_TRACE_POINTS
  92 #include <trace/events/task.h>
  93
  94 /*
  95  * Minimum number of threads to boot the kernel
  96  */
  97 #define MIN_THREADS 20
  98
  99 /*
 100  * Maximum number of threads
 101  */
 102 #define MAX_THREADS FUTEX_TID_MASK
 103
 104 /*
 105  * Protected counters by write_lock_irq(&tasklist_lock)
 106  */
 107 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 108 int nr_threads;                 /* The idle threads do not count.. */
 109
 110 int max_threads;                /* tunable limit on nr_threads */
 111
 112 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 113
 114 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 115
 116 #ifdef CONFIG_PROVE_RCU
 117 int lockdep_tasklist_lock_is_held(void)
 118 {
 119         return lockdep_is_held(&tasklist_lock);
 120 }
 121 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 122 #endif /* #ifdef CONFIG_PROVE_RCU */
 123
 124 int nr_processes(void)
 125 {
 126         int cpu;
 127         int total = 0;
 128
 129         for_each_possible_cpu(cpu)
 130                 total += per_cpu(process_counts, cpu);
 131
 132         return total;
 133 }
 134
 135 void __weak arch_release_task_struct(struct task_struct *tsk)
 136 {
 137 }
 138
 139 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 140 static struct kmem_cache *task_struct_cachep;
 141
 142 static inline struct task_struct *alloc_task_struct_node(int node)
 143 {
 144         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 145 }
 146
 147 static inline void free_task_struct(struct task_struct *tsk)
 148 {
 149         kmem_cache_free(task_struct_cachep, tsk);
 150 }
 151 #endif
 152
 153 void __weak arch_release_thread_stack(unsigned long *stack)
 154 {
 155 }
 156
 157 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 158
 159 /*
 160  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 161  * kmemcache based allocator.
 162  */
 163 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 164
 165 #ifdef CONFIG_VMAP_STACK
 166 /*
 167  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 168  * flush.  Try to minimize the number of calls by caching stacks.
 169  */
 170 #define NR_CACHED_STACKS 2
 171 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 172 #endif
 173
 174 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 175 {
 176 #ifdef CONFIG_VMAP_STACK
 177         void *stack;
 178         int i;
 179
 180         local_irq_disable();
 181         for (i = 0; i < NR_CACHED_STACKS; i++) {
 182                 struct vm_struct *s = this_cpu_read(cached_stacks[i]);
 183
 184                 if (!s)
 185                         continue;
 186                 this_cpu_write(cached_stacks[i], NULL);
 187
 188                 tsk->stack_vm_area = s;
 189                 local_irq_enable();
 190                 return s->addr;
 191         }
 192         local_irq_enable();
 193
 194         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_SIZE,
 195                                      VMALLOC_START, VMALLOC_END,
 196                                      THREADINFO_GFP | __GFP_HIGHMEM,
 197                                      PAGE_KERNEL,
 198                                      0, node, __builtin_return_address(0));
 199
 200         /*
 201          * We can't call find_vm_area() in interrupt context, and
 202          * free_thread_stack() can be called in interrupt context,
 203          * so cache the vm_struct.
 204          */
 205         if (stack)
 206                 tsk->stack_vm_area = find_vm_area(stack);
 207         return stack;
 208 #else
 209         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 210                                              THREAD_SIZE_ORDER);
 211
 212         return page ? page_address(page) : NULL;
 213 #endif
 214 }
 215
 216 static inline void free_thread_stack(struct task_struct *tsk)
 217 {
 218 #ifdef CONFIG_VMAP_STACK
 219         if (task_stack_vm_area(tsk)) {
 220                 unsigned long flags;
 221                 int i;
 222
 223                 local_irq_save(flags);
 224                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 225                         if (this_cpu_read(cached_stacks[i]))
 226                                 continue;
 227
 228                         this_cpu_write(cached_stacks[i], tsk->stack_vm_area);
 229                         local_irq_restore(flags);
 230                         return;
 231                 }
 232                 local_irq_restore(flags);
 233
 234                 vfree_atomic(tsk->stack);
 235                 return;
 236         }
 237 #endif
 238
 239         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 240 }
 241 # else
 242 static struct kmem_cache *thread_stack_cache;
 243
 244 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 245                                                   int node)
 246 {
 247         return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 248 }
 249
 250 static void free_thread_stack(struct task_struct *tsk)
 251 {
 252         kmem_cache_free(thread_stack_cache, tsk->stack);
 253 }
 254
 255 void thread_stack_cache_init(void)
 256 {
 257         thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
 258                                               THREAD_SIZE, 0, NULL);
 259         BUG_ON(thread_stack_cache == NULL);
 260 }
 261 # endif
 262 #endif
 263
 264 /* SLAB cache for signal_struct structures (tsk->signal) */
 265 static struct kmem_cache *signal_cachep;
 266
 267 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 268 struct kmem_cache *sighand_cachep;
 269
 270 /* SLAB cache for files_struct structures (tsk->files) */
 271 struct kmem_cache *files_cachep;
 272
 273 /* SLAB cache for fs_struct structures (tsk->fs) */
 274 struct kmem_cache *fs_cachep;
 275
 276 /* SLAB cache for vm_area_struct structures */
 277 struct kmem_cache *vm_area_cachep;
 278
 279 /* SLAB cache for mm_struct structures (tsk->mm) */
 280 static struct kmem_cache *mm_cachep;
 281
 282 static void account_kernel_stack(struct task_struct *tsk, int account)
 283 {
 284         void *stack = task_stack_page(tsk);
 285         struct vm_struct *vm = task_stack_vm_area(tsk);
 286
 287         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 288
 289         if (vm) {
 290                 int i;
 291
 292                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 293
 294                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 295                         mod_zone_page_state(page_zone(vm->pages[i]),
 296                                             NR_KERNEL_STACK_KB,
 297                                             PAGE_SIZE / 1024 * account);
 298                 }
 299
 300                 /* All stack pages belong to the same memcg. */
 301                 memcg_kmem_update_page_stat(vm->pages[0], MEMCG_KERNEL_STACK_KB,
 302                                             account * (THREAD_SIZE / 1024));
 303         } else {
 304                 /*
 305                  * All stack pages are in the same zone and belong to the
 306                  * same memcg.
 307                  */
 308                 struct page *first_page = virt_to_page(stack);
 309
 310                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 311                                     THREAD_SIZE / 1024 * account);
 312
 313                 memcg_kmem_update_page_stat(first_page, MEMCG_KERNEL_STACK_KB,
 314                                             account * (THREAD_SIZE / 1024));
 315         }
 316 }
 317
 318 static void release_task_stack(struct task_struct *tsk)
 319 {
 320         if (WARN_ON(tsk->state != TASK_DEAD))
 321                 return;  /* Better to leak the stack than to free prematurely */
 322
 323         account_kernel_stack(tsk, -1);
 324         arch_release_thread_stack(tsk->stack);
 325         free_thread_stack(tsk);
 326         tsk->stack = NULL;
 327 #ifdef CONFIG_VMAP_STACK
 328         tsk->stack_vm_area = NULL;
 329 #endif
 330 }
 331
 332 #ifdef CONFIG_THREAD_INFO_IN_TASK
 333 void put_task_stack(struct task_struct *tsk)
 334 {
 335         if (atomic_dec_and_test(&tsk->stack_refcount))
 336                 release_task_stack(tsk);
 337 }
 338 #endif
 339
 340 void free_task(struct task_struct *tsk)
 341 {
 342 #ifndef CONFIG_THREAD_INFO_IN_TASK
 343         /*
 344          * The task is finally done with both the stack and thread_info,
 345          * so free both.
 346          */
 347         release_task_stack(tsk);
 348 #else
 349         /*
 350          * If the task had a separate stack allocation, it should be gone
 351          * by now.
 352          */
 353         WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
 354 #endif
 355         rt_mutex_debug_task_free(tsk);
 356         ftrace_graph_exit_task(tsk);
 357         put_seccomp_filter(tsk);
 358         arch_release_task_struct(tsk);
 359         if (tsk->flags & PF_KTHREAD)
 360                 free_kthread_struct(tsk);
 361         free_task_struct(tsk);
 362 }
 363 EXPORT_SYMBOL(free_task);
 364
 365 static inline void free_signal_struct(struct signal_struct *sig)
 366 {
 367         taskstats_tgid_free(sig);
 368         sched_autogroup_exit(sig);
 369         /*
 370          * __mmdrop is not safe to call from softirq context on x86 due to
 371          * pgd_dtor so postpone it to the async context
 372          */
 373         if (sig->oom_mm)
 374                 mmdrop_async(sig->oom_mm);
 375         kmem_cache_free(signal_cachep, sig);
 376 }
 377
 378 static inline void put_signal_struct(struct signal_struct *sig)
 379 {
 380         if (atomic_dec_and_test(&sig->sigcnt))
 381                 free_signal_struct(sig);
 382 }
 383
 384 void __put_task_struct(struct task_struct *tsk)
 385 {
 386         WARN_ON(!tsk->exit_state);
 387         WARN_ON(atomic_read(&tsk->usage));
 388         WARN_ON(tsk == current);
 389
 390         cgroup_free(tsk);
 391         task_numa_free(tsk);
 392         security_task_free(tsk);
 393         exit_creds(tsk);
 394         delayacct_tsk_free(tsk);
 395         put_signal_struct(tsk->signal);
 396
 397         if (!profile_handoff_task(tsk))
 398                 free_task(tsk);
 399 }
 400 EXPORT_SYMBOL_GPL(__put_task_struct);
 401
 402 void __init __weak arch_task_cache_init(void) { }
 403
 404 /*
 405  * set_max_threads
 406  */
 407 static void set_max_threads(unsigned int max_threads_suggested)
 408 {
 409         u64 threads;
 410
 411         /*
 412          * The number of threads shall be limited such that the thread
 413          * structures may only consume a small part of the available memory.
 414          */
 415         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 416                 threads = MAX_THREADS;
 417         else
 418                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 419                                     (u64) THREAD_SIZE * 8UL);
 420
 421         if (threads > max_threads_suggested)
 422                 threads = max_threads_suggested;
 423
 424         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 425 }
 426
 427 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 428 /* Initialized by the architecture: */
 429 int arch_task_struct_size __read_mostly;
 430 #endif
 431
 432 void __init fork_init(void)
 433 {
 434         int i;
 435 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 436 #ifndef ARCH_MIN_TASKALIGN
 437 #define ARCH_MIN_TASKALIGN      0
 438 #endif
 439         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 440
 441         /* create a slab on which task_structs can be allocated */
 442         task_struct_cachep = kmem_cache_create("task_struct",
 443                         arch_task_struct_size, align,
 444                         SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
 445 #endif
 446
 447         /* do the arch specific task caches init */
 448         arch_task_cache_init();
 449
 450         set_max_threads(MAX_THREADS);
 451
 452         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 453         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 454         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 455                 init_task.signal->rlim[RLIMIT_NPROC];
 456
 457         for (i = 0; i < UCOUNT_COUNTS; i++) {
 458                 init_user_ns.ucount_max[i] = max_threads/2;
 459         }
 460 }
 461
 462 int __weak arch_dup_task_struct(struct task_struct *dst,
 463                                                struct task_struct *src)
 464 {
 465         *dst = *src;
 466         return 0;
 467 }
 468
 469 void set_task_stack_end_magic(struct task_struct *tsk)
 470 {
 471         unsigned long *stackend;
 472
 473         stackend = end_of_stack(tsk);
 474         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 475 }
 476
 477 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 478 {
 479         struct task_struct *tsk;
 480         unsigned long *stack;
 481         struct vm_struct *stack_vm_area;
 482         int err;
 483
 484         if (node == NUMA_NO_NODE)
 485                 node = tsk_fork_get_node(orig);
 486         tsk = alloc_task_struct_node(node);
 487         if (!tsk)
 488                 return NULL;
 489
 490         stack = alloc_thread_stack_node(tsk, node);
 491         if (!stack)
 492                 goto free_tsk;
 493
 494         stack_vm_area = task_stack_vm_area(tsk);
 495
 496         err = arch_dup_task_struct(tsk, orig);
 497
 498         /*
 499          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 500          * sure they're properly initialized before using any stack-related
 501          * functions again.
 502          */
 503         tsk->stack = stack;
 504 #ifdef CONFIG_VMAP_STACK
 505         tsk->stack_vm_area = stack_vm_area;
 506 #endif
 507 #ifdef CONFIG_THREAD_INFO_IN_TASK
 508         atomic_set(&tsk->stack_refcount, 1);
 509 #endif
 510
 511         if (err)
 512                 goto free_stack;
 513
 514 #ifdef CONFIG_SECCOMP
 515         /*
 516          * We must handle setting up seccomp filters once we're under
 517          * the sighand lock in case orig has changed between now and
 518          * then. Until then, filter must be NULL to avoid messing up
 519          * the usage counts on the error path calling free_task.
 520          */
 521         tsk->seccomp.filter = NULL;
 522 #endif
 523
 524         setup_thread_stack(tsk, orig);
 525         clear_user_return_notifier(tsk);
 526         clear_tsk_need_resched(tsk);
 527         set_task_stack_end_magic(tsk);
 528
 529 #ifdef CONFIG_CC_STACKPROTECTOR
 530         tsk->stack_canary = get_random_int();
 531 #endif
 532
 533         /*
 534          * One for us, one for whoever does the "release_task()" (usually
 535          * parent)
 536          */
 537         atomic_set(&tsk->usage, 2);
 538 #ifdef CONFIG_BLK_DEV_IO_TRACE
 539         tsk->btrace_seq = 0;
 540 #endif
 541         tsk->splice_pipe = NULL;
 542         tsk->task_frag.page = NULL;
 543         tsk->wake_q.next = NULL;
 544
 545         account_kernel_stack(tsk, 1);
 546
 547         kcov_task_init(tsk);
 548
 549         return tsk;
 550
 551 free_stack:
 552         free_thread_stack(tsk);
 553 free_tsk:
 554         free_task_struct(tsk);
 555         return NULL;
 556 }
 557
 558 #ifdef CONFIG_MMU
 559 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 560                                         struct mm_struct *oldmm)
 561 {
 562         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 563         struct rb_node **rb_link, *rb_parent;
 564         int retval;
 565         unsigned long charge;
 566         LIST_HEAD(uf);
 567
 568         uprobe_start_dup_mmap();
 569         if (down_write_killable(&oldmm->mmap_sem)) {
 570                 retval = -EINTR;
 571                 goto fail_uprobe_end;
 572         }
 573         flush_cache_dup_mm(oldmm);
 574         uprobe_dup_mmap(oldmm, mm);
 575         /*
 576          * Not linked in yet - no deadlock potential:
 577          */
 578         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 579
 580         /* No ordering required: file already has been exposed. */
 581         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 582
 583         mm->total_vm = oldmm->total_vm;
 584         mm->data_vm = oldmm->data_vm;
 585         mm->exec_vm = oldmm->exec_vm;
 586         mm->stack_vm = oldmm->stack_vm;
 587
 588         rb_link = &mm->mm_rb.rb_node;
 589         rb_parent = NULL;
 590         pprev = &mm->mmap;
 591         retval = ksm_fork(mm, oldmm);
 592         if (retval)
 593                 goto out;
 594         retval = khugepaged_fork(mm, oldmm);
 595         if (retval)
 596                 goto out;
 597
 598         prev = NULL;
 599         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 600                 struct file *file;
 601
 602                 if (mpnt->vm_flags & VM_DONTCOPY) {
 603                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 604                         continue;
 605                 }
 606                 charge = 0;
 607                 if (mpnt->vm_flags & VM_ACCOUNT) {
 608                         unsigned long len = vma_pages(mpnt);
 609
 610                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 611                                 goto fail_nomem;
 612                         charge = len;
 613                 }
 614                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 615                 if (!tmp)
 616                         goto fail_nomem;
 617                 *tmp = *mpnt;
 618                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 619                 retval = vma_dup_policy(mpnt, tmp);
 620                 if (retval)
 621                         goto fail_nomem_policy;
 622                 tmp->vm_mm = mm;
 623                 retval = dup_userfaultfd(tmp, &uf);
 624                 if (retval)
 625                         goto fail_nomem_anon_vma_fork;
 626                 if (anon_vma_fork(tmp, mpnt))
 627                         goto fail_nomem_anon_vma_fork;
 628                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 629                 tmp->vm_next = tmp->vm_prev = NULL;
 630                 file = tmp->vm_file;
 631                 if (file) {
 632                         struct inode *inode = file_inode(file);
 633                         struct address_space *mapping = file->f_mapping;
 634
 635                         get_file(file);
 636                         if (tmp->vm_flags & VM_DENYWRITE)
 637                                 atomic_dec(&inode->i_writecount);
 638                         i_mmap_lock_write(mapping);
 639                         if (tmp->vm_flags & VM_SHARED)
 640                                 atomic_inc(&mapping->i_mmap_writable);
 641                         flush_dcache_mmap_lock(mapping);
 642                         /* insert tmp into the share list, just after mpnt */
 643                         vma_interval_tree_insert_after(tmp, mpnt,
 644                                         &mapping->i_mmap);
 645                         flush_dcache_mmap_unlock(mapping);
 646                         i_mmap_unlock_write(mapping);
 647                 }
 648
 649                 /*
 650                  * Clear hugetlb-related page reserves for children. This only
 651                  * affects MAP_PRIVATE mappings. Faults generated by the child
 652                  * are not guaranteed to succeed, even if read-only
 653                  */
 654                 if (is_vm_hugetlb_page(tmp))
 655                         reset_vma_resv_huge_pages(tmp);
 656
 657                 /*
 658                  * Link in the new vma and copy the page table entries.
 659                  */
 660                 *pprev = tmp;
 661                 pprev = &tmp->vm_next;
 662                 tmp->vm_prev = prev;
 663                 prev = tmp;
 664
 665                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 666                 rb_link = &tmp->vm_rb.rb_right;
 667                 rb_parent = &tmp->vm_rb;
 668
 669                 mm->map_count++;
 670                 retval = copy_page_range(mm, oldmm, mpnt);
 671
 672                 if (tmp->vm_ops && tmp->vm_ops->open)
 673                         tmp->vm_ops->open(tmp);
 674
 675                 if (retval)
 676                         goto out;
 677         }
 678         /* a new mm has just been created */
 679         arch_dup_mmap(oldmm, mm);
 680         retval = 0;
 681 out:
 682         up_write(&mm->mmap_sem);
 683         flush_tlb_mm(oldmm);
 684         up_write(&oldmm->mmap_sem);
 685         dup_userfaultfd_complete(&uf);
 686 fail_uprobe_end:
 687         uprobe_end_dup_mmap();
 688         return retval;
 689 fail_nomem_anon_vma_fork:
 690         mpol_put(vma_policy(tmp));
 691 fail_nomem_policy:
 692         kmem_cache_free(vm_area_cachep, tmp);
 693 fail_nomem:
 694         retval = -ENOMEM;
 695         vm_unacct_memory(charge);
 696         goto out;
 697 }
 698
 699 static inline int mm_alloc_pgd(struct mm_struct *mm)
 700 {
 701         mm->pgd = pgd_alloc(mm);
 702         if (unlikely(!mm->pgd))
 703                 return -ENOMEM;
 704         return 0;
 705 }
 706
 707 static inline void mm_free_pgd(struct mm_struct *mm)
 708 {
 709         pgd_free(mm, mm->pgd);
 710 }
 711 #else
 712 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 713 {
 714         down_write(&oldmm->mmap_sem);
 715         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 716         up_write(&oldmm->mmap_sem);
 717         return 0;
 718 }
 719 #define mm_alloc_pgd(mm)        (0)
 720 #define mm_free_pgd(mm)
 721 #endif /* CONFIG_MMU */
 722
 723 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 724
 725 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 726 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 727
 728 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 729
 730 static int __init coredump_filter_setup(char *s)
 731 {
 732         default_dump_filter =
 733                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 734                 MMF_DUMP_FILTER_MASK;
 735         return 1;
 736 }
 737
 738 __setup("coredump_filter=", coredump_filter_setup);
 739
 740 #include <linux/init_task.h>
 741
 742 static void mm_init_aio(struct mm_struct *mm)
 743 {
 744 #ifdef CONFIG_AIO
 745         spin_lock_init(&mm->ioctx_lock);
 746         mm->ioctx_table = NULL;
 747 #endif
 748 }
 749
 750 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 751 {
 752 #ifdef CONFIG_MEMCG
 753         mm->owner = p;
 754 #endif
 755 }
 756
 757 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 758         struct user_namespace *user_ns)
 759 {
 760         mm->mmap = NULL;
 761         mm->mm_rb = RB_ROOT;
 762         mm->vmacache_seqnum = 0;
 763         atomic_set(&mm->mm_users, 1);
 764         atomic_set(&mm->mm_count, 1);
 765         init_rwsem(&mm->mmap_sem);
 766         INIT_LIST_HEAD(&mm->mmlist);
 767         mm->core_state = NULL;
 768         atomic_long_set(&mm->nr_ptes, 0);
 769         mm_nr_pmds_init(mm);
 770         mm->map_count = 0;
 771         mm->locked_vm = 0;
 772         mm->pinned_vm = 0;
 773         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 774         spin_lock_init(&mm->page_table_lock);
 775         mm_init_cpumask(mm);
 776         mm_init_aio(mm);
 777         mm_init_owner(mm, p);
 778         mmu_notifier_mm_init(mm);
 779         clear_tlb_flush_pending(mm);
 780 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 781         mm->pmd_huge_pte = NULL;
 782 #endif
 783
 784         if (current->mm) {
 785                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 786                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 787         } else {
 788                 mm->flags = default_dump_filter;
 789                 mm->def_flags = 0;
 790         }
 791
 792         if (mm_alloc_pgd(mm))
 793                 goto fail_nopgd;
 794
 795         if (init_new_context(p, mm))
 796                 goto fail_nocontext;
 797
 798         mm->user_ns = get_user_ns(user_ns);
 799         return mm;
 800
 801 fail_nocontext:
 802         mm_free_pgd(mm);
 803 fail_nopgd:
 804         free_mm(mm);
 805         return NULL;
 806 }
 807
 808 static void check_mm(struct mm_struct *mm)
 809 {
 810         int i;
 811
 812         for (i = 0; i < NR_MM_COUNTERS; i++) {
 813                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 814
 815                 if (unlikely(x))
 816                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 817                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 818         }
 819
 820         if (atomic_long_read(&mm->nr_ptes))
 821                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 822                                 atomic_long_read(&mm->nr_ptes));
 823         if (mm_nr_pmds(mm))
 824                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 825                                 mm_nr_pmds(mm));
 826
 827 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 828         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 829 #endif
 830 }
 831
 832 /*
 833  * Allocate and initialize an mm_struct.
 834  */
 835 struct mm_struct *mm_alloc(void)
 836 {
 837         struct mm_struct *mm;
 838
 839         mm = allocate_mm();
 840         if (!mm)
 841                 return NULL;
 842
 843         memset(mm, 0, sizeof(*mm));
 844         return mm_init(mm, current, current_user_ns());
 845 }
 846
 847 /*
 848  * Called when the last reference to the mm
 849  * is dropped: either by a lazy thread or by
 850  * mmput. Free the page directory and the mm.
 851  */
 852 void __mmdrop(struct mm_struct *mm)
 853 {
 854         BUG_ON(mm == &init_mm);
 855         mm_free_pgd(mm);
 856         destroy_context(mm);
 857         mmu_notifier_mm_destroy(mm);
 858         check_mm(mm);
 859         put_user_ns(mm->user_ns);
 860         free_mm(mm);
 861 }
 862 EXPORT_SYMBOL_GPL(__mmdrop);
 863
 864 static inline void __mmput(struct mm_struct *mm)
 865 {
 866         VM_BUG_ON(atomic_read(&mm->mm_users));
 867
 868         uprobe_clear_state(mm);
 869         exit_aio(mm);
 870         ksm_exit(mm);
 871         khugepaged_exit(mm); /* must run before exit_mmap */
 872         exit_mmap(mm);
 873         mm_put_huge_zero_page(mm);
 874         set_mm_exe_file(mm, NULL);
 875         if (!list_empty(&mm->mmlist)) {
 876                 spin_lock(&mmlist_lock);
 877                 list_del(&mm->mmlist);
 878                 spin_unlock(&mmlist_lock);
 879         }
 880         if (mm->binfmt)
 881                 module_put(mm->binfmt->module);
 882         set_bit(MMF_OOM_SKIP, &mm->flags);
 883         mmdrop(mm);
 884 }
 885
 886 /*
 887  * Decrement the use count and release all resources for an mm.
 888  */
 889 void mmput(struct mm_struct *mm)
 890 {
 891         might_sleep();
 892
 893         if (atomic_dec_and_test(&mm->mm_users))
 894                 __mmput(mm);
 895 }
 896 EXPORT_SYMBOL_GPL(mmput);
 897
 898 #ifdef CONFIG_MMU
 899 static void mmput_async_fn(struct work_struct *work)
 900 {
 901         struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
 902         __mmput(mm);
 903 }
 904
 905 void mmput_async(struct mm_struct *mm)
 906 {
 907         if (atomic_dec_and_test(&mm->mm_users)) {
 908                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
 909                 schedule_work(&mm->async_put_work);
 910         }
 911 }
 912 #endif
 913
 914 /**
 915  * set_mm_exe_file - change a reference to the mm's executable file
 916  *
 917  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 918  *
 919  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 920  * invocations: in mmput() nobody alive left, in execve task is single
 921  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 922  * mm->exe_file, but does so without using set_mm_exe_file() in order
 923  * to do avoid the need for any locks.
 924  */
 925 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 926 {
 927         struct file *old_exe_file;
 928
 929         /*
 930          * It is safe to dereference the exe_file without RCU as
 931          * this function is only called if nobody else can access
 932          * this mm -- see comment above for justification.
 933          */
 934         old_exe_file = rcu_dereference_raw(mm->exe_file);
 935
 936         if (new_exe_file)
 937                 get_file(new_exe_file);
 938         rcu_assign_pointer(mm->exe_file, new_exe_file);
 939         if (old_exe_file)
 940                 fput(old_exe_file);
 941 }
 942
 943 /**
 944  * get_mm_exe_file - acquire a reference to the mm's executable file
 945  *
 946  * Returns %NULL if mm has no associated executable file.
 947  * User must release file via fput().
 948  */
 949 struct file *get_mm_exe_file(struct mm_struct *mm)
 950 {
 951         struct file *exe_file;
 952
 953         rcu_read_lock();
 954         exe_file = rcu_dereference(mm->exe_file);
 955         if (exe_file && !get_file_rcu(exe_file))
 956                 exe_file = NULL;
 957         rcu_read_unlock();
 958         return exe_file;
 959 }
 960 EXPORT_SYMBOL(get_mm_exe_file);
 961
 962 /**
 963  * get_task_exe_file - acquire a reference to the task's executable file
 964  *
 965  * Returns %NULL if task's mm (if any) has no associated executable file or
 966  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
 967  * User must release file via fput().
 968  */
 969 struct file *get_task_exe_file(struct task_struct *task)
 970 {
 971         struct file *exe_file = NULL;
 972         struct mm_struct *mm;
 973
 974         task_lock(task);
 975         mm = task->mm;
 976         if (mm) {
 977                 if (!(task->flags & PF_KTHREAD))
 978                         exe_file = get_mm_exe_file(mm);
 979         }
 980         task_unlock(task);
 981         return exe_file;
 982 }
 983 EXPORT_SYMBOL(get_task_exe_file);
 984
 985 /**
 986  * get_task_mm - acquire a reference to the task's mm
 987  *
 988  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 989  * this kernel workthread has transiently adopted a user mm with use_mm,
 990  * to do its AIO) is not set and if so returns a reference to it, after
 991  * bumping up the use count.  User must release the mm via mmput()
 992  * after use.  Typically used by /proc and ptrace.
 993  */
 994 struct mm_struct *get_task_mm(struct task_struct *task)
 995 {
 996         struct mm_struct *mm;
 997
 998         task_lock(task);
 999         mm = task->mm;
1000         if (mm) {
1001                 if (task->flags & PF_KTHREAD)
1002                         mm = NULL;
1003                 else
1004                         mmget(mm);
1005         }
1006         task_unlock(task);
1007         return mm;
1008 }
1009 EXPORT_SYMBOL_GPL(get_task_mm);
1010
1011 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1012 {
1013         struct mm_struct *mm;
1014         int err;
1015
1016         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1017         if (err)
1018                 return ERR_PTR(err);
1019
1020         mm = get_task_mm(task);
1021         if (mm && mm != current->mm &&
1022                         !ptrace_may_access(task, mode)) {
1023                 mmput(mm);
1024                 mm = ERR_PTR(-EACCES);
1025         }
1026         mutex_unlock(&task->signal->cred_guard_mutex);
1027
1028         return mm;
1029 }
1030
1031 static void complete_vfork_done(struct task_struct *tsk)
1032 {
1033         struct completion *vfork;
1034
1035         task_lock(tsk);
1036         vfork = tsk->vfork_done;
1037         if (likely(vfork)) {
1038                 tsk->vfork_done = NULL;
1039                 complete(vfork);
1040         }
1041         task_unlock(tsk);
1042 }
1043
1044 static int wait_for_vfork_done(struct task_struct *child,
1045                                 struct completion *vfork)
1046 {
1047         int killed;
1048
1049         freezer_do_not_count();
1050         killed = wait_for_completion_killable(vfork);
1051         freezer_count();
1052
1053         if (killed) {
1054                 task_lock(child);
1055                 child->vfork_done = NULL;
1056                 task_unlock(child);
1057         }
1058
1059         put_task_struct(child);
1060         return killed;
1061 }
1062
1063 /* Please note the differences between mmput and mm_release.
1064  * mmput is called whenever we stop holding onto a mm_struct,
1065  * error success whatever.
1066  *
1067  * mm_release is called after a mm_struct has been removed
1068  * from the current process.
1069  *
1070  * This difference is important for error handling, when we
1071  * only half set up a mm_struct for a new process and need to restore
1072  * the old one.  Because we mmput the new mm_struct before
1073  * restoring the old one. . .
1074  * Eric Biederman 10 January 1998
1075  */
1076 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1077 {
1078         /* Get rid of any futexes when releasing the mm */
1079 #ifdef CONFIG_FUTEX
1080         if (unlikely(tsk->robust_list)) {
1081                 exit_robust_list(tsk);
1082                 tsk->robust_list = NULL;
1083         }
1084 #ifdef CONFIG_COMPAT
1085         if (unlikely(tsk->compat_robust_list)) {
1086                 compat_exit_robust_list(tsk);
1087                 tsk->compat_robust_list = NULL;
1088         }
1089 #endif
1090         if (unlikely(!list_empty(&tsk->pi_state_list)))
1091                 exit_pi_state_list(tsk);
1092 #endif
1093
1094         uprobe_free_utask(tsk);
1095
1096         /* Get rid of any cached register state */
1097         deactivate_mm(tsk, mm);
1098
1099         /*
1100          * Signal userspace if we're not exiting with a core dump
1101          * because we want to leave the value intact for debugging
1102          * purposes.
1103          */
1104         if (tsk->clear_child_tid) {
1105                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1106                     atomic_read(&mm->mm_users) > 1) {
1107                         /*
1108                          * We don't check the error code - if userspace has
1109                          * not set up a proper pointer then tough luck.
1110                          */
1111                         put_user(0, tsk->clear_child_tid);
1112                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1113                                         1, NULL, NULL, 0);
1114                 }
1115                 tsk->clear_child_tid = NULL;
1116         }
1117
1118         /*
1119          * All done, finally we can wake up parent and return this mm to him.
1120          * Also kthread_stop() uses this completion for synchronization.
1121          */
1122         if (tsk->vfork_done)
1123                 complete_vfork_done(tsk);
1124 }
1125
1126 /*
1127  * Allocate a new mm structure and copy contents from the
1128  * mm structure of the passed in task structure.
1129  */
1130 static struct mm_struct *dup_mm(struct task_struct *tsk)
1131 {
1132         struct mm_struct *mm, *oldmm = current->mm;
1133         int err;
1134
1135         mm = allocate_mm();
1136         if (!mm)
1137                 goto fail_nomem;
1138
1139         memcpy(mm, oldmm, sizeof(*mm));
1140
1141         if (!mm_init(mm, tsk, mm->user_ns))
1142                 goto fail_nomem;
1143
1144         err = dup_mmap(mm, oldmm);
1145         if (err)
1146                 goto free_pt;
1147
1148         mm->hiwater_rss = get_mm_rss(mm);
1149         mm->hiwater_vm = mm->total_vm;
1150
1151         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1152                 goto free_pt;
1153
1154         return mm;
1155
1156 free_pt:
1157         /* don't put binfmt in mmput, we haven't got module yet */
1158         mm->binfmt = NULL;
1159         mmput(mm);
1160
1161 fail_nomem:
1162         return NULL;
1163 }
1164
1165 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1166 {
1167         struct mm_struct *mm, *oldmm;
1168         int retval;
1169
1170         tsk->min_flt = tsk->maj_flt = 0;
1171         tsk->nvcsw = tsk->nivcsw = 0;
1172 #ifdef CONFIG_DETECT_HUNG_TASK
1173         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1174 #endif
1175
1176         tsk->mm = NULL;
1177         tsk->active_mm = NULL;
1178
1179         /*
1180          * Are we cloning a kernel thread?
1181          *
1182          * We need to steal a active VM for that..
1183          */
1184         oldmm = current->mm;
1185         if (!oldmm)
1186                 return 0;
1187
1188         /* initialize the new vmacache entries */
1189         vmacache_flush(tsk);
1190
1191         if (clone_flags & CLONE_VM) {
1192                 mmget(oldmm);
1193                 mm = oldmm;
1194                 goto good_mm;
1195         }
1196
1197         retval = -ENOMEM;
1198         mm = dup_mm(tsk);
1199         if (!mm)
1200                 goto fail_nomem;
1201
1202 good_mm:
1203         tsk->mm = mm;
1204         tsk->active_mm = mm;
1205         return 0;
1206
1207 fail_nomem:
1208         return retval;
1209 }
1210
1211 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1212 {
1213         struct fs_struct *fs = current->fs;
1214         if (clone_flags & CLONE_FS) {
1215                 /* tsk->fs is already what we want */
1216                 spin_lock(&fs->lock);
1217                 if (fs->in_exec) {
1218                         spin_unlock(&fs->lock);
1219                         return -EAGAIN;
1220                 }
1221                 fs->users++;
1222                 spin_unlock(&fs->lock);
1223                 return 0;
1224         }
1225         tsk->fs = copy_fs_struct(fs);
1226         if (!tsk->fs)
1227                 return -ENOMEM;
1228         return 0;
1229 }
1230
1231 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1232 {
1233         struct files_struct *oldf, *newf;
1234         int error = 0;
1235
1236         /*
1237          * A background process may not have any files ...
1238          */
1239         oldf = current->files;
1240         if (!oldf)
1241                 goto out;
1242
1243         if (clone_flags & CLONE_FILES) {
1244                 atomic_inc(&oldf->count);
1245                 goto out;
1246         }
1247
1248         newf = dup_fd(oldf, &error);
1249         if (!newf)
1250                 goto out;
1251
1252         tsk->files = newf;
1253         error = 0;
1254 out:
1255         return error;
1256 }
1257
1258 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1259 {
1260 #ifdef CONFIG_BLOCK
1261         struct io_context *ioc = current->io_context;
1262         struct io_context *new_ioc;
1263
1264         if (!ioc)
1265                 return 0;
1266         /*
1267          * Share io context with parent, if CLONE_IO is set
1268          */
1269         if (clone_flags & CLONE_IO) {
1270                 ioc_task_link(ioc);
1271                 tsk->io_context = ioc;
1272         } else if (ioprio_valid(ioc->ioprio)) {
1273                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1274                 if (unlikely(!new_ioc))
1275                         return -ENOMEM;
1276
1277                 new_ioc->ioprio = ioc->ioprio;
1278                 put_io_context(new_ioc);
1279         }
1280 #endif
1281         return 0;
1282 }
1283
1284 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1285 {
1286         struct sighand_struct *sig;
1287
1288         if (clone_flags & CLONE_SIGHAND) {
1289                 atomic_inc(&current->sighand->count);
1290                 return 0;
1291         }
1292         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1293         rcu_assign_pointer(tsk->sighand, sig);
1294         if (!sig)
1295                 return -ENOMEM;
1296
1297         atomic_set(&sig->count, 1);
1298         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1299         return 0;
1300 }
1301
1302 void __cleanup_sighand(struct sighand_struct *sighand)
1303 {
1304         if (atomic_dec_and_test(&sighand->count)) {
1305                 signalfd_cleanup(sighand);
1306                 /*
1307                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1308                  * without an RCU grace period, see __lock_task_sighand().
1309                  */
1310                 kmem_cache_free(sighand_cachep, sighand);
1311         }
1312 }
1313
1314 #ifdef CONFIG_POSIX_TIMERS
1315 /*
1316  * Initialize POSIX timer handling for a thread group.
1317  */
1318 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1319 {
1320         unsigned long cpu_limit;
1321
1322         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1323         if (cpu_limit != RLIM_INFINITY) {
1324                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1325                 sig->cputimer.running = true;
1326         }
1327
1328         /* The timer lists. */
1329         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1330         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1331         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1332 }
1333 #else
1334 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1335 #endif
1336
1337 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1338 {
1339         struct signal_struct *sig;
1340
1341         if (clone_flags & CLONE_THREAD)
1342                 return 0;
1343
1344         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1345         tsk->signal = sig;
1346         if (!sig)
1347                 return -ENOMEM;
1348
1349         sig->nr_threads = 1;
1350         atomic_set(&sig->live, 1);
1351         atomic_set(&sig->sigcnt, 1);
1352
1353         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1354         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1355         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1356
1357         init_waitqueue_head(&sig->wait_chldexit);
1358         sig->curr_target = tsk;
1359         init_sigpending(&sig->shared_pending);
1360         seqlock_init(&sig->stats_lock);
1361         prev_cputime_init(&sig->prev_cputime);
1362
1363 #ifdef CONFIG_POSIX_TIMERS
1364         INIT_LIST_HEAD(&sig->posix_timers);
1365         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1366         sig->real_timer.function = it_real_fn;
1367 #endif
1368
1369         task_lock(current->group_leader);
1370         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1371         task_unlock(current->group_leader);
1372
1373         posix_cpu_timers_init_group(sig);
1374
1375         tty_audit_fork(sig);
1376         sched_autogroup_fork(sig);
1377
1378         sig->oom_score_adj = current->signal->oom_score_adj;
1379         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1380
1381         mutex_init(&sig->cred_guard_mutex);
1382
1383         return 0;
1384 }
1385
1386 static void copy_seccomp(struct task_struct *p)
1387 {
1388 #ifdef CONFIG_SECCOMP
1389         /*
1390          * Must be called with sighand->lock held, which is common to
1391          * all threads in the group. Holding cred_guard_mutex is not
1392          * needed because this new task is not yet running and cannot
1393          * be racing exec.
1394          */
1395         assert_spin_locked(&current->sighand->siglock);
1396
1397         /* Ref-count the new filter user, and assign it. */
1398         get_seccomp_filter(current);
1399         p->seccomp = current->seccomp;
1400
1401         /*
1402          * Explicitly enable no_new_privs here in case it got set
1403          * between the task_struct being duplicated and holding the
1404          * sighand lock. The seccomp state and nnp must be in sync.
1405          */
1406         if (task_no_new_privs(current))
1407                 task_set_no_new_privs(p);
1408
1409         /*
1410          * If the parent gained a seccomp mode after copying thread
1411          * flags and between before we held the sighand lock, we have
1412          * to manually enable the seccomp thread flag here.
1413          */
1414         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1415                 set_tsk_thread_flag(p, TIF_SECCOMP);
1416 #endif
1417 }
1418
1419 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1420 {
1421         current->clear_child_tid = tidptr;
1422
1423         return task_pid_vnr(current);
1424 }
1425
1426 static void rt_mutex_init_task(struct task_struct *p)
1427 {
1428         raw_spin_lock_init(&p->pi_lock);
1429 #ifdef CONFIG_RT_MUTEXES
1430         p->pi_waiters = RB_ROOT;
1431         p->pi_waiters_leftmost = NULL;
1432         p->pi_blocked_on = NULL;
1433 #endif
1434 }
1435
1436 #ifdef CONFIG_POSIX_TIMERS
1437 /*
1438  * Initialize POSIX timer handling for a single task.
1439  */
1440 static void posix_cpu_timers_init(struct task_struct *tsk)
1441 {
1442         tsk->cputime_expires.prof_exp = 0;
1443         tsk->cputime_expires.virt_exp = 0;
1444         tsk->cputime_expires.sched_exp = 0;
1445         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1446         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1447         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1448 }
1449 #else
1450 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1451 #endif
1452
1453 static inline void
1454 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1455 {
1456          task->pids[type].pid = pid;
1457 }
1458
1459 /*
1460  * This creates a new process as a copy of the old one,
1461  * but does not actually start it yet.
1462  *
1463  * It copies the registers, and all the appropriate
1464  * parts of the process environment (as per the clone
1465  * flags). The actual kick-off is left to the caller.
1466  */
1467 static __latent_entropy struct task_struct *copy_process(
1468                                         unsigned long clone_flags,
1469                                         unsigned long stack_start,
1470                                         unsigned long stack_size,
1471                                         int __user *child_tidptr,
1472                                         struct pid *pid,
1473                                         int trace,
1474                                         unsigned long tls,
1475                                         int node)
1476 {
1477         int retval;
1478         struct task_struct *p;
1479
1480         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1481                 return ERR_PTR(-EINVAL);
1482
1483         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1484                 return ERR_PTR(-EINVAL);
1485
1486         /*
1487          * Thread groups must share signals as well, and detached threads
1488          * can only be started up within the thread group.
1489          */
1490         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1491                 return ERR_PTR(-EINVAL);
1492
1493         /*
1494          * Shared signal handlers imply shared VM. By way of the above,
1495          * thread groups also imply shared VM. Blocking this case allows
1496          * for various simplifications in other code.
1497          */
1498         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1499                 return ERR_PTR(-EINVAL);
1500
1501         /*
1502          * Siblings of global init remain as zombies on exit since they are
1503          * not reaped by their parent (swapper). To solve this and to avoid
1504          * multi-rooted process trees, prevent global and container-inits
1505          * from creating siblings.
1506          */
1507         if ((clone_flags & CLONE_PARENT) &&
1508                                 current->signal->flags & SIGNAL_UNKILLABLE)
1509                 return ERR_PTR(-EINVAL);
1510
1511         /*
1512          * If the new process will be in a different pid or user namespace
1513          * do not allow it to share a thread group with the forking task.
1514          */
1515         if (clone_flags & CLONE_THREAD) {
1516                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1517                     (task_active_pid_ns(current) !=
1518                                 current->nsproxy->pid_ns_for_children))
1519                         return ERR_PTR(-EINVAL);
1520         }
1521
1522         retval = security_task_create(clone_flags);
1523         if (retval)
1524                 goto fork_out;
1525
1526         retval = -ENOMEM;
1527         p = dup_task_struct(current, node);
1528         if (!p)
1529                 goto fork_out;
1530
1531         ftrace_graph_init_task(p);
1532
1533         rt_mutex_init_task(p);
1534
1535 #ifdef CONFIG_PROVE_LOCKING
1536         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1537         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1538 #endif
1539         retval = -EAGAIN;
1540         if (atomic_read(&p->real_cred->user->processes) >=
1541                         task_rlimit(p, RLIMIT_NPROC)) {
1542                 if (p->real_cred->user != INIT_USER &&
1543                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1544                         goto bad_fork_free;
1545         }
1546         current->flags &= ~PF_NPROC_EXCEEDED;
1547
1548         retval = copy_creds(p, clone_flags);
1549         if (retval < 0)
1550                 goto bad_fork_free;
1551
1552         /*
1553          * If multiple threads are within copy_process(), then this check
1554          * triggers too late. This doesn't hurt, the check is only there
1555          * to stop root fork bombs.
1556          */
1557         retval = -EAGAIN;
1558         if (nr_threads >= max_threads)
1559                 goto bad_fork_cleanup_count;
1560
1561         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1562         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1563         p->flags |= PF_FORKNOEXEC;
1564         INIT_LIST_HEAD(&p->children);
1565         INIT_LIST_HEAD(&p->sibling);
1566         rcu_copy_process(p);
1567         p->vfork_done = NULL;
1568         spin_lock_init(&p->alloc_lock);
1569
1570         init_sigpending(&p->pending);
1571
1572         p->utime = p->stime = p->gtime = 0;
1573 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1574         p->utimescaled = p->stimescaled = 0;
1575 #endif
1576         prev_cputime_init(&p->prev_cputime);
1577
1578 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1579         seqcount_init(&p->vtime_seqcount);
1580         p->vtime_snap = 0;
1581         p->vtime_snap_whence = VTIME_INACTIVE;
1582 #endif
1583
1584 #if defined(SPLIT_RSS_COUNTING)
1585         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1586 #endif
1587
1588         p->default_timer_slack_ns = current->timer_slack_ns;
1589
1590         task_io_accounting_init(&p->ioac);
1591         acct_clear_integrals(p);
1592
1593         posix_cpu_timers_init(p);
1594
1595         p->start_time = ktime_get_ns();
1596         p->real_start_time = ktime_get_boot_ns();
1597         p->io_context = NULL;
1598         p->audit_context = NULL;
1599         cgroup_fork(p);
1600 #ifdef CONFIG_NUMA
1601         p->mempolicy = mpol_dup(p->mempolicy);
1602         if (IS_ERR(p->mempolicy)) {
1603                 retval = PTR_ERR(p->mempolicy);
1604                 p->mempolicy = NULL;
1605                 goto bad_fork_cleanup_threadgroup_lock;
1606         }
1607 #endif
1608 #ifdef CONFIG_CPUSETS
1609         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1610         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1611         seqcount_init(&p->mems_allowed_seq);
1612 #endif
1613 #ifdef CONFIG_TRACE_IRQFLAGS
1614         p->irq_events = 0;
1615         p->hardirqs_enabled = 0;
1616         p->hardirq_enable_ip = 0;
1617         p->hardirq_enable_event = 0;
1618         p->hardirq_disable_ip = _THIS_IP_;
1619         p->hardirq_disable_event = 0;
1620         p->softirqs_enabled = 1;
1621         p->softirq_enable_ip = _THIS_IP_;
1622         p->softirq_enable_event = 0;
1623         p->softirq_disable_ip = 0;
1624         p->softirq_disable_event = 0;
1625         p->hardirq_context = 0;
1626         p->softirq_context = 0;
1627 #endif
1628
1629         p->pagefault_disabled = 0;
1630
1631 #ifdef CONFIG_LOCKDEP
1632         p->lockdep_depth = 0; /* no locks held yet */
1633         p->curr_chain_key = 0;
1634         p->lockdep_recursion = 0;
1635 #endif
1636
1637 #ifdef CONFIG_DEBUG_MUTEXES
1638         p->blocked_on = NULL; /* not blocked yet */
1639 #endif
1640 #ifdef CONFIG_BCACHE
1641         p->sequential_io        = 0;
1642         p->sequential_io_avg    = 0;
1643 #endif
1644
1645         /* Perform scheduler related setup. Assign this task to a CPU. */
1646         retval = sched_fork(clone_flags, p);
1647         if (retval)
1648                 goto bad_fork_cleanup_policy;
1649
1650         retval = perf_event_init_task(p);
1651         if (retval)
1652                 goto bad_fork_cleanup_policy;
1653         retval = audit_alloc(p);
1654         if (retval)
1655                 goto bad_fork_cleanup_perf;
1656         /* copy all the process information */
1657         shm_init_task(p);
1658         retval = copy_semundo(clone_flags, p);
1659         if (retval)
1660                 goto bad_fork_cleanup_audit;
1661         retval = copy_files(clone_flags, p);
1662         if (retval)
1663                 goto bad_fork_cleanup_semundo;
1664         retval = copy_fs(clone_flags, p);
1665         if (retval)
1666                 goto bad_fork_cleanup_files;
1667         retval = copy_sighand(clone_flags, p);
1668         if (retval)
1669                 goto bad_fork_cleanup_fs;
1670         retval = copy_signal(clone_flags, p);
1671         if (retval)
1672                 goto bad_fork_cleanup_sighand;
1673         retval = copy_mm(clone_flags, p);
1674         if (retval)
1675                 goto bad_fork_cleanup_signal;
1676         retval = copy_namespaces(clone_flags, p);
1677         if (retval)
1678                 goto bad_fork_cleanup_mm;
1679         retval = copy_io(clone_flags, p);
1680         if (retval)
1681                 goto bad_fork_cleanup_namespaces;
1682         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1683         if (retval)
1684                 goto bad_fork_cleanup_io;
1685
1686         if (pid != &init_struct_pid) {
1687                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1688                 if (IS_ERR(pid)) {
1689                         retval = PTR_ERR(pid);
1690                         goto bad_fork_cleanup_thread;
1691                 }
1692         }
1693
1694         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1695         /*
1696          * Clear TID on mm_release()?
1697          */
1698         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1699 #ifdef CONFIG_BLOCK
1700         p->plug = NULL;
1701 #endif
1702 #ifdef CONFIG_FUTEX
1703         p->robust_list = NULL;
1704 #ifdef CONFIG_COMPAT
1705         p->compat_robust_list = NULL;
1706 #endif
1707         INIT_LIST_HEAD(&p->pi_state_list);
1708         p->pi_state_cache = NULL;
1709 #endif
1710         /*
1711          * sigaltstack should be cleared when sharing the same VM
1712          */
1713         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1714                 sas_ss_reset(p);
1715
1716         /*
1717          * Syscall tracing and stepping should be turned off in the
1718          * child regardless of CLONE_PTRACE.
1719          */
1720         user_disable_single_step(p);
1721         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1722 #ifdef TIF_SYSCALL_EMU
1723         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1724 #endif
1725         clear_all_latency_tracing(p);
1726
1727         /* ok, now we should be set up.. */
1728         p->pid = pid_nr(pid);
1729         if (clone_flags & CLONE_THREAD) {
1730                 p->exit_signal = -1;
1731                 p->group_leader = current->group_leader;
1732                 p->tgid = current->tgid;
1733         } else {
1734                 if (clone_flags & CLONE_PARENT)
1735                         p->exit_signal = current->group_leader->exit_signal;
1736                 else
1737                         p->exit_signal = (clone_flags & CSIGNAL);
1738                 p->group_leader = p;
1739                 p->tgid = p->pid;
1740         }
1741
1742         p->nr_dirtied = 0;
1743         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1744         p->dirty_paused_when = 0;
1745
1746         p->pdeath_signal = 0;
1747         INIT_LIST_HEAD(&p->thread_group);
1748         p->task_works = NULL;
1749
1750         cgroup_threadgroup_change_begin(current);
1751         /*
1752          * Ensure that the cgroup subsystem policies allow the new process to be
1753          * forked. It should be noted the the new process's css_set can be changed
1754          * between here and cgroup_post_fork() if an organisation operation is in
1755          * progress.
1756          */
1757         retval = cgroup_can_fork(p);
1758         if (retval)
1759                 goto bad_fork_free_pid;
1760
1761         /*
1762          * Make it visible to the rest of the system, but dont wake it up yet.
1763          * Need tasklist lock for parent etc handling!
1764          */
1765         write_lock_irq(&tasklist_lock);
1766
1767         /* CLONE_PARENT re-uses the old parent */
1768         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1769                 p->real_parent = current->real_parent;
1770                 p->parent_exec_id = current->parent_exec_id;
1771         } else {
1772                 p->real_parent = current;
1773                 p->parent_exec_id = current->self_exec_id;
1774         }
1775
1776         spin_lock(&current->sighand->siglock);
1777
1778         /*
1779          * Copy seccomp details explicitly here, in case they were changed
1780          * before holding sighand lock.
1781          */
1782         copy_seccomp(p);
1783
1784         /*
1785          * Process group and session signals need to be delivered to just the
1786          * parent before the fork or both the parent and the child after the
1787          * fork. Restart if a signal comes in before we add the new process to
1788          * it's process group.
1789          * A fatal signal pending means that current will exit, so the new
1790          * thread can't slip out of an OOM kill (or normal SIGKILL).
1791         */
1792         recalc_sigpending();
1793         if (signal_pending(current)) {
1794                 spin_unlock(&current->sighand->siglock);
1795                 write_unlock_irq(&tasklist_lock);
1796                 retval = -ERESTARTNOINTR;
1797                 goto bad_fork_cancel_cgroup;
1798         }
1799
1800         if (likely(p->pid)) {
1801                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1802
1803                 init_task_pid(p, PIDTYPE_PID, pid);
1804                 if (thread_group_leader(p)) {
1805                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1806                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1807
1808                         if (is_child_reaper(pid)) {
1809                                 ns_of_pid(pid)->child_reaper = p;
1810                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1811                         }
1812
1813                         p->signal->leader_pid = pid;
1814                         p->signal->tty = tty_kref_get(current->signal->tty);
1815                         /*
1816                          * Inherit has_child_subreaper flag under the same
1817                          * tasklist_lock with adding child to the process tree
1818                          * for propagate_has_child_subreaper optimization.
1819                          */
1820                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1821                                                          p->real_parent->signal->is_child_subreaper;
1822                         list_add_tail(&p->sibling, &p->real_parent->children);
1823                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1824                         attach_pid(p, PIDTYPE_PGID);
1825                         attach_pid(p, PIDTYPE_SID);
1826                         __this_cpu_inc(process_counts);
1827                 } else {
1828                         current->signal->nr_threads++;
1829                         atomic_inc(&current->signal->live);
1830                         atomic_inc(&current->signal->sigcnt);
1831                         list_add_tail_rcu(&p->thread_group,
1832                                           &p->group_leader->thread_group);
1833                         list_add_tail_rcu(&p->thread_node,
1834                                           &p->signal->thread_head);
1835                 }
1836                 attach_pid(p, PIDTYPE_PID);
1837                 nr_threads++;
1838         }
1839
1840         total_forks++;
1841         spin_unlock(&current->sighand->siglock);
1842         syscall_tracepoint_update(p);
1843         write_unlock_irq(&tasklist_lock);
1844
1845         proc_fork_connector(p);
1846         cgroup_post_fork(p);
1847         cgroup_threadgroup_change_end(current);
1848         perf_event_fork(p);
1849
1850         trace_task_newtask(p, clone_flags);
1851         uprobe_copy_process(p, clone_flags);
1852
1853         return p;
1854
1855 bad_fork_cancel_cgroup:
1856         cgroup_cancel_fork(p);
1857 bad_fork_free_pid:
1858         cgroup_threadgroup_change_end(current);
1859         if (pid != &init_struct_pid)
1860                 free_pid(pid);
1861 bad_fork_cleanup_thread:
1862         exit_thread(p);
1863 bad_fork_cleanup_io:
1864         if (p->io_context)
1865                 exit_io_context(p);
1866 bad_fork_cleanup_namespaces:
1867         exit_task_namespaces(p);
1868 bad_fork_cleanup_mm:
1869         if (p->mm)
1870                 mmput(p->mm);
1871 bad_fork_cleanup_signal:
1872         if (!(clone_flags & CLONE_THREAD))
1873                 free_signal_struct(p->signal);
1874 bad_fork_cleanup_sighand:
1875         __cleanup_sighand(p->sighand);
1876 bad_fork_cleanup_fs:
1877         exit_fs(p); /* blocking */
1878 bad_fork_cleanup_files:
1879         exit_files(p); /* blocking */
1880 bad_fork_cleanup_semundo:
1881         exit_sem(p);
1882 bad_fork_cleanup_audit:
1883         audit_free(p);
1884 bad_fork_cleanup_perf:
1885         perf_event_free_task(p);
1886 bad_fork_cleanup_policy:
1887 #ifdef CONFIG_NUMA
1888         mpol_put(p->mempolicy);
1889 bad_fork_cleanup_threadgroup_lock:
1890 #endif
1891         delayacct_tsk_free(p);
1892 bad_fork_cleanup_count:
1893         atomic_dec(&p->cred->user->processes);
1894         exit_creds(p);
1895 bad_fork_free:
1896         p->state = TASK_DEAD;
1897         put_task_stack(p);
1898         free_task(p);
1899 fork_out:
1900         return ERR_PTR(retval);
1901 }
1902
1903 static inline void init_idle_pids(struct pid_link *links)
1904 {
1905         enum pid_type type;
1906
1907         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1908                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1909                 links[type].pid = &init_struct_pid;
1910         }
1911 }
1912
1913 struct task_struct *fork_idle(int cpu)
1914 {
1915         struct task_struct *task;
1916         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1917                             cpu_to_node(cpu));
1918         if (!IS_ERR(task)) {
1919                 init_idle_pids(task->pids);
1920                 init_idle(task, cpu);
1921         }
1922
1923         return task;
1924 }
1925
1926 /*
1927  *  Ok, this is the main fork-routine.
1928  *
1929  * It copies the process, and if successful kick-starts
1930  * it and waits for it to finish using the VM if required.
1931  */
1932 long _do_fork(unsigned long clone_flags,
1933               unsigned long stack_start,
1934               unsigned long stack_size,
1935               int __user *parent_tidptr,
1936               int __user *child_tidptr,
1937               unsigned long tls)
1938 {
1939         struct task_struct *p;
1940         int trace = 0;
1941         long nr;
1942
1943         /*
1944          * Determine whether and which event to report to ptracer.  When
1945          * called from kernel_thread or CLONE_UNTRACED is explicitly
1946          * requested, no event is reported; otherwise, report if the event
1947          * for the type of forking is enabled.
1948          */
1949         if (!(clone_flags & CLONE_UNTRACED)) {
1950                 if (clone_flags & CLONE_VFORK)
1951                         trace = PTRACE_EVENT_VFORK;
1952                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1953                         trace = PTRACE_EVENT_CLONE;
1954                 else
1955                         trace = PTRACE_EVENT_FORK;
1956
1957                 if (likely(!ptrace_event_enabled(current, trace)))
1958                         trace = 0;
1959         }
1960
1961         p = copy_process(clone_flags, stack_start, stack_size,
1962                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
1963         add_latent_entropy();
1964         /*
1965          * Do this prior waking up the new thread - the thread pointer
1966          * might get invalid after that point, if the thread exits quickly.
1967          */
1968         if (!IS_ERR(p)) {
1969                 struct completion vfork;
1970                 struct pid *pid;
1971
1972                 trace_sched_process_fork(current, p);
1973
1974                 pid = get_task_pid(p, PIDTYPE_PID);
1975                 nr = pid_vnr(pid);
1976
1977                 if (clone_flags & CLONE_PARENT_SETTID)
1978                         put_user(nr, parent_tidptr);
1979
1980                 if (clone_flags & CLONE_VFORK) {
1981                         p->vfork_done = &vfork;
1982                         init_completion(&vfork);
1983                         get_task_struct(p);
1984                 }
1985
1986                 wake_up_new_task(p);
1987
1988                 /* forking complete and child started to run, tell ptracer */
1989                 if (unlikely(trace))
1990                         ptrace_event_pid(trace, pid);
1991
1992                 if (clone_flags & CLONE_VFORK) {
1993                         if (!wait_for_vfork_done(p, &vfork))
1994                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1995                 }
1996
1997                 put_pid(pid);
1998         } else {
1999                 nr = PTR_ERR(p);
2000         }
2001         return nr;
2002 }
2003
2004 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2005 /* For compatibility with architectures that call do_fork directly rather than
2006  * using the syscall entry points below. */
2007 long do_fork(unsigned long clone_flags,
2008               unsigned long stack_start,
2009               unsigned long stack_size,
2010               int __user *parent_tidptr,
2011               int __user *child_tidptr)
2012 {
2013         return _do_fork(clone_flags, stack_start, stack_size,
2014                         parent_tidptr, child_tidptr, 0);
2015 }
2016 #endif
2017
2018 /*
2019  * Create a kernel thread.
2020  */
2021 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2022 {
2023         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2024                 (unsigned long)arg, NULL, NULL, 0);
2025 }
2026
2027 #ifdef __ARCH_WANT_SYS_FORK
2028 SYSCALL_DEFINE0(fork)
2029 {
2030 #ifdef CONFIG_MMU
2031         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2032 #else
2033         /* can not support in nommu mode */
2034         return -EINVAL;
2035 #endif
2036 }
2037 #endif
2038
2039 #ifdef __ARCH_WANT_SYS_VFORK
2040 SYSCALL_DEFINE0(vfork)
2041 {
2042         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2043                         0, NULL, NULL, 0);
2044 }
2045 #endif
2046
2047 #ifdef __ARCH_WANT_SYS_CLONE
2048 #ifdef CONFIG_CLONE_BACKWARDS
2049 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2050                  int __user *, parent_tidptr,
2051                  unsigned long, tls,
2052                  int __user *, child_tidptr)
2053 #elif defined(CONFIG_CLONE_BACKWARDS2)
2054 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2055                  int __user *, parent_tidptr,
2056                  int __user *, child_tidptr,
2057                  unsigned long, tls)
2058 #elif defined(CONFIG_CLONE_BACKWARDS3)
2059 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2060                 int, stack_size,
2061                 int __user *, parent_tidptr,
2062                 int __user *, child_tidptr,
2063                 unsigned long, tls)
2064 #else
2065 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2066                  int __user *, parent_tidptr,
2067                  int __user *, child_tidptr,
2068                  unsigned long, tls)
2069 #endif
2070 {
2071         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2072 }
2073 #endif
2074
2075 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2076 {
2077         struct task_struct *leader, *parent, *child;
2078         int res;
2079
2080         read_lock(&tasklist_lock);
2081         leader = top = top->group_leader;
2082 down:
2083         for_each_thread(leader, parent) {
2084                 list_for_each_entry(child, &parent->children, sibling) {
2085                         res = visitor(child, data);
2086                         if (res) {
2087                                 if (res < 0)
2088                                         goto out;
2089                                 leader = child;
2090                                 goto down;
2091                         }
2092 up:
2093                         ;
2094                 }
2095         }
2096
2097         if (leader != top) {
2098                 child = leader;
2099                 parent = child->real_parent;
2100                 leader = parent->group_leader;
2101                 goto up;
2102         }
2103 out:
2104         read_unlock(&tasklist_lock);
2105 }
2106
2107 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2108 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2109 #endif
2110
2111 static void sighand_ctor(void *data)
2112 {
2113         struct sighand_struct *sighand = data;
2114
2115         spin_lock_init(&sighand->siglock);
2116         init_waitqueue_head(&sighand->signalfd_wqh);
2117 }
2118
2119 void __init proc_caches_init(void)
2120 {
2121         sighand_cachep = kmem_cache_create("sighand_cache",
2122                         sizeof(struct sighand_struct), 0,
2123                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
2124                         SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2125         signal_cachep = kmem_cache_create("signal_cache",
2126                         sizeof(struct signal_struct), 0,
2127                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2128                         NULL);
2129         files_cachep = kmem_cache_create("files_cache",
2130                         sizeof(struct files_struct), 0,
2131                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2132                         NULL);
2133         fs_cachep = kmem_cache_create("fs_cache",
2134                         sizeof(struct fs_struct), 0,
2135                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2136                         NULL);
2137         /*
2138          * FIXME! The "sizeof(struct mm_struct)" currently includes the
2139          * whole struct cpumask for the OFFSTACK case. We could change
2140          * this to *only* allocate as much of it as required by the
2141          * maximum number of CPU's we can ever have.  The cpumask_allocation
2142          * is at the end of the structure, exactly for that reason.
2143          */
2144         mm_cachep = kmem_cache_create("mm_struct",
2145                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2146                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2147                         NULL);
2148         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2149         mmap_init();
2150         nsproxy_cache_init();
2151 }
2152
2153 /*
2154  * Check constraints on flags passed to the unshare system call.
2155  */
2156 static int check_unshare_flags(unsigned long unshare_flags)
2157 {
2158         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2159                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2160                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2161                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2162                 return -EINVAL;
2163         /*
2164          * Not implemented, but pretend it works if there is nothing
2165          * to unshare.  Note that unsharing the address space or the
2166          * signal handlers also need to unshare the signal queues (aka
2167          * CLONE_THREAD).
2168          */
2169         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2170                 if (!thread_group_empty(current))
2171                         return -EINVAL;
2172         }
2173         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2174                 if (atomic_read(&current->sighand->count) > 1)
2175                         return -EINVAL;
2176         }
2177         if (unshare_flags & CLONE_VM) {
2178                 if (!current_is_single_threaded())
2179                         return -EINVAL;
2180         }
2181
2182         return 0;
2183 }
2184
2185 /*
2186  * Unshare the filesystem structure if it is being shared
2187  */
2188 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2189 {
2190         struct fs_struct *fs = current->fs;
2191
2192         if (!(unshare_flags & CLONE_FS) || !fs)
2193                 return 0;
2194
2195         /* don't need lock here; in the worst case we'll do useless copy */
2196         if (fs->users == 1)
2197                 return 0;
2198
2199         *new_fsp = copy_fs_struct(fs);
2200         if (!*new_fsp)
2201                 return -ENOMEM;
2202
2203         return 0;
2204 }
2205
2206 /*
2207  * Unshare file descriptor table if it is being shared
2208  */
2209 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2210 {
2211         struct files_struct *fd = current->files;
2212         int error = 0;
2213
2214         if ((unshare_flags & CLONE_FILES) &&
2215             (fd && atomic_read(&fd->count) > 1)) {
2216                 *new_fdp = dup_fd(fd, &error);
2217                 if (!*new_fdp)
2218                         return error;
2219         }
2220
2221         return 0;
2222 }
2223
2224 /*
2225  * unshare allows a process to 'unshare' part of the process
2226  * context which was originally shared using clone.  copy_*
2227  * functions used by do_fork() cannot be used here directly
2228  * because they modify an inactive task_struct that is being
2229  * constructed. Here we are modifying the current, active,
2230  * task_struct.
2231  */
2232 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2233 {
2234         struct fs_struct *fs, *new_fs = NULL;
2235         struct files_struct *fd, *new_fd = NULL;
2236         struct cred *new_cred = NULL;
2237         struct nsproxy *new_nsproxy = NULL;
2238         int do_sysvsem = 0;
2239         int err;
2240
2241         /*
2242          * If unsharing a user namespace must also unshare the thread group
2243          * and unshare the filesystem root and working directories.
2244          */
2245         if (unshare_flags & CLONE_NEWUSER)
2246                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2247         /*
2248          * If unsharing vm, must also unshare signal handlers.
2249          */
2250         if (unshare_flags & CLONE_VM)
2251                 unshare_flags |= CLONE_SIGHAND;
2252         /*
2253          * If unsharing a signal handlers, must also unshare the signal queues.
2254          */
2255         if (unshare_flags & CLONE_SIGHAND)
2256                 unshare_flags |= CLONE_THREAD;
2257         /*
2258          * If unsharing namespace, must also unshare filesystem information.
2259          */
2260         if (unshare_flags & CLONE_NEWNS)
2261                 unshare_flags |= CLONE_FS;
2262
2263         err = check_unshare_flags(unshare_flags);
2264         if (err)
2265                 goto bad_unshare_out;
2266         /*
2267          * CLONE_NEWIPC must also detach from the undolist: after switching
2268          * to a new ipc namespace, the semaphore arrays from the old
2269          * namespace are unreachable.
2270          */
2271         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2272                 do_sysvsem = 1;
2273         err = unshare_fs(unshare_flags, &new_fs);
2274         if (err)
2275                 goto bad_unshare_out;
2276         err = unshare_fd(unshare_flags, &new_fd);
2277         if (err)
2278                 goto bad_unshare_cleanup_fs;
2279         err = unshare_userns(unshare_flags, &new_cred);
2280         if (err)
2281                 goto bad_unshare_cleanup_fd;
2282         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2283                                          new_cred, new_fs);
2284         if (err)
2285                 goto bad_unshare_cleanup_cred;
2286
2287         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2288                 if (do_sysvsem) {
2289                         /*
2290                          * CLONE_SYSVSEM is equivalent to sys_exit().
2291                          */
2292                         exit_sem(current);
2293                 }
2294                 if (unshare_flags & CLONE_NEWIPC) {
2295                         /* Orphan segments in old ns (see sem above). */
2296                         exit_shm(current);
2297                         shm_init_task(current);
2298                 }
2299
2300                 if (new_nsproxy)
2301                         switch_task_namespaces(current, new_nsproxy);
2302
2303                 task_lock(current);
2304
2305                 if (new_fs) {
2306                         fs = current->fs;
2307                         spin_lock(&fs->lock);
2308                         current->fs = new_fs;
2309                         if (--fs->users)
2310                                 new_fs = NULL;
2311                         else
2312                                 new_fs = fs;
2313                         spin_unlock(&fs->lock);
2314                 }
2315
2316                 if (new_fd) {
2317                         fd = current->files;
2318                         current->files = new_fd;
2319                         new_fd = fd;
2320                 }
2321
2322                 task_unlock(current);
2323
2324                 if (new_cred) {
2325                         /* Install the new user namespace */
2326                         commit_creds(new_cred);
2327                         new_cred = NULL;
2328                 }
2329         }
2330
2331 bad_unshare_cleanup_cred:
2332         if (new_cred)
2333                 put_cred(new_cred);
2334 bad_unshare_cleanup_fd:
2335         if (new_fd)
2336                 put_files_struct(new_fd);
2337
2338 bad_unshare_cleanup_fs:
2339         if (new_fs)
2340                 free_fs_struct(new_fs);
2341
2342 bad_unshare_out:
2343         return err;
2344 }
2345
2346 /*
2347  *      Helper to unshare the files of the current task.
2348  *      We don't want to expose copy_files internals to
2349  *      the exec layer of the kernel.
2350  */
2351
2352 int unshare_files(struct files_struct **displaced)
2353 {
2354         struct task_struct *task = current;
2355         struct files_struct *copy = NULL;
2356         int error;
2357
2358         error = unshare_fd(CLONE_FILES, &copy);
2359         if (error || !copy) {
2360                 *displaced = NULL;
2361                 return error;
2362         }
2363         *displaced = task->files;
2364         task_lock(task);
2365         task->files = copy;
2366         task_unlock(task);
2367         return 0;
2368 }
2369
2370 int sysctl_max_threads(struct ctl_table *table, int write,
2371                        void __user *buffer, size_t *lenp, loff_t *ppos)
2372 {
2373         struct ctl_table t;
2374         int ret;
2375         int threads = max_threads;
2376         int min = MIN_THREADS;
2377         int max = MAX_THREADS;
2378
2379         t = *table;
2380         t.data = &threads;
2381         t.extra1 = &min;
2382         t.extra2 = &max;
2383
2384         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2385         if (ret || !write)
2386                 return ret;
2387
2388         set_max_threads(threads);
2389
2390         return 0;
2391 }