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/sched/mm.h>
  17 #include <linux/sched/coredump.h>
  18 #include <linux/sched/user.h>
  19 #include <linux/sched/numa_balancing.h>
  20 #include <linux/sched/stat.h>
  21 #include <linux/sched/task.h>
  22 #include <linux/sched/task_stack.h>
  23 #include <linux/sched/cputime.h>
  24 #include <linux/rtmutex.h>
  25 #include <linux/init.h>
  26 #include <linux/unistd.h>
  27 #include <linux/module.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/completion.h>
  30 #include <linux/personality.h>
  31 #include <linux/mempolicy.h>
  32 #include <linux/sem.h>
  33 #include <linux/file.h>
  34 #include <linux/fdtable.h>
  35 #include <linux/iocontext.h>
  36 #include <linux/key.h>
  37 #include <linux/binfmts.h>
  38 #include <linux/mman.h>
  39 #include <linux/mmu_notifier.h>
  40 #include <linux/fs.h>
  41 #include <linux/mm.h>
  42 #include <linux/vmacache.h>
  43 #include <linux/nsproxy.h>
  44 #include <linux/capability.h>
  45 #include <linux/cpu.h>
  46 #include <linux/cgroup.h>
  47 #include <linux/security.h>
  48 #include <linux/hugetlb.h>
  49 #include <linux/seccomp.h>
  50 #include <linux/swap.h>
  51 #include <linux/syscalls.h>
  52 #include <linux/jiffies.h>
  53 #include <linux/futex.h>
  54 #include <linux/compat.h>
  55 #include <linux/kthread.h>
  56 #include <linux/task_io_accounting_ops.h>
  57 #include <linux/rcupdate.h>
  58 #include <linux/ptrace.h>
  59 #include <linux/mount.h>
  60 #include <linux/audit.h>
  61 #include <linux/memcontrol.h>
  62 #include <linux/ftrace.h>
  63 #include <linux/proc_fs.h>
  64 #include <linux/profile.h>
  65 #include <linux/rmap.h>
  66 #include <linux/ksm.h>
  67 #include <linux/acct.h>
  68 #include <linux/userfaultfd_k.h>
  69 #include <linux/tsacct_kern.h>
  70 #include <linux/cn_proc.h>
  71 #include <linux/freezer.h>
  72 #include <linux/delayacct.h>
  73 #include <linux/taskstats_kern.h>
  74 #include <linux/random.h>
  75 #include <linux/tty.h>
  76 #include <linux/blkdev.h>
  77 #include <linux/fs_struct.h>
  78 #include <linux/magic.h>
  79 #include <linux/perf_event.h>
  80 #include <linux/posix-timers.h>
  81 #include <linux/user-return-notifier.h>
  82 #include <linux/oom.h>
  83 #include <linux/khugepaged.h>
  84 #include <linux/signalfd.h>
  85 #include <linux/uprobes.h>
  86 #include <linux/aio.h>
  87 #include <linux/compiler.h>
  88 #include <linux/sysctl.h>
  89 #include <linux/kcov.h>
  90 #include <linux/livepatch.h>
  91 #include <linux/thread_info.h>
  92
  93 #include <asm/pgtable.h>
  94 #include <asm/pgalloc.h>
  95 #include <linux/uaccess.h>
  96 #include <asm/mmu_context.h>
  97 #include <asm/cacheflush.h>
  98 #include <asm/tlbflush.h>
  99
 100 #include <trace/events/sched.h>
 101
 102 #define CREATE_TRACE_POINTS
 103 #include <trace/events/task.h>
 104
 105 /*
 106  * Minimum number of threads to boot the kernel
 107  */
 108 #define MIN_THREADS 20
 109
 110 /*
 111  * Maximum number of threads
 112  */
 113 #define MAX_THREADS FUTEX_TID_MASK
 114
 115 /*
 116  * Protected counters by write_lock_irq(&tasklist_lock)
 117  */
 118 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 119 int nr_threads;                 /* The idle threads do not count.. */
 120
 121 int max_threads;                /* tunable limit on nr_threads */
 122
 123 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 124
 125 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 126
 127 #ifdef CONFIG_PROVE_RCU
 128 int lockdep_tasklist_lock_is_held(void)
 129 {
 130         return lockdep_is_held(&tasklist_lock);
 131 }
 132 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 133 #endif /* #ifdef CONFIG_PROVE_RCU */
 134
 135 int nr_processes(void)
 136 {
 137         int cpu;
 138         int total = 0;
 139
 140         for_each_possible_cpu(cpu)
 141                 total += per_cpu(process_counts, cpu);
 142
 143         return total;
 144 }
 145
 146 void __weak arch_release_task_struct(struct task_struct *tsk)
 147 {
 148 }
 149
 150 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 151 static struct kmem_cache *task_struct_cachep;
 152
 153 static inline struct task_struct *alloc_task_struct_node(int node)
 154 {
 155         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 156 }
 157
 158 static inline void free_task_struct(struct task_struct *tsk)
 159 {
 160         kmem_cache_free(task_struct_cachep, tsk);
 161 }
 162 #endif
 163
 164 void __weak arch_release_thread_stack(unsigned long *stack)
 165 {
 166 }
 167
 168 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 169
 170 /*
 171  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 172  * kmemcache based allocator.
 173  */
 174 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 175
 176 #ifdef CONFIG_VMAP_STACK
 177 /*
 178  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 179  * flush.  Try to minimize the number of calls by caching stacks.
 180  */
 181 #define NR_CACHED_STACKS 2
 182 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 183
 184 static int free_vm_stack_cache(unsigned int cpu)
 185 {
 186         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 187         int i;
 188
 189         for (i = 0; i < NR_CACHED_STACKS; i++) {
 190                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 191
 192                 if (!vm_stack)
 193                         continue;
 194
 195                 vfree(vm_stack->addr);
 196                 cached_vm_stacks[i] = NULL;
 197         }
 198
 199         return 0;
 200 }
 201 #endif
 202
 203 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 204 {
 205 #ifdef CONFIG_VMAP_STACK
 206         void *stack;
 207         int i;
 208
 209         for (i = 0; i < NR_CACHED_STACKS; i++) {
 210                 struct vm_struct *s;
 211
 212                 s = this_cpu_xchg(cached_stacks[i], NULL);
 213
 214                 if (!s)
 215                         continue;
 216
 217                 tsk->stack_vm_area = s;
 218                 return s->addr;
 219         }
 220
 221         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 222                                      VMALLOC_START, VMALLOC_END,
 223                                      THREADINFO_GFP,
 224                                      PAGE_KERNEL,
 225                                      0, node, __builtin_return_address(0));
 226
 227         /*
 228          * We can't call find_vm_area() in interrupt context, and
 229          * free_thread_stack() can be called in interrupt context,
 230          * so cache the vm_struct.
 231          */
 232         if (stack)
 233                 tsk->stack_vm_area = find_vm_area(stack);
 234         return stack;
 235 #else
 236         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 237                                              THREAD_SIZE_ORDER);
 238
 239         return page ? page_address(page) : NULL;
 240 #endif
 241 }
 242
 243 static inline void free_thread_stack(struct task_struct *tsk)
 244 {
 245 #ifdef CONFIG_VMAP_STACK
 246         if (task_stack_vm_area(tsk)) {
 247                 int i;
 248
 249                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 250                         if (this_cpu_cmpxchg(cached_stacks[i],
 251                                         NULL, tsk->stack_vm_area) != NULL)
 252                                 continue;
 253
 254                         return;
 255                 }
 256
 257                 vfree_atomic(tsk->stack);
 258                 return;
 259         }
 260 #endif
 261
 262         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 263 }
 264 # else
 265 static struct kmem_cache *thread_stack_cache;
 266
 267 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 268                                                   int node)
 269 {
 270         return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 271 }
 272
 273 static void free_thread_stack(struct task_struct *tsk)
 274 {
 275         kmem_cache_free(thread_stack_cache, tsk->stack);
 276 }
 277
 278 void thread_stack_cache_init(void)
 279 {
 280         thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
 281                                               THREAD_SIZE, 0, NULL);
 282         BUG_ON(thread_stack_cache == NULL);
 283 }
 284 # endif
 285 #endif
 286
 287 /* SLAB cache for signal_struct structures (tsk->signal) */
 288 static struct kmem_cache *signal_cachep;
 289
 290 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 291 struct kmem_cache *sighand_cachep;
 292
 293 /* SLAB cache for files_struct structures (tsk->files) */
 294 struct kmem_cache *files_cachep;
 295
 296 /* SLAB cache for fs_struct structures (tsk->fs) */
 297 struct kmem_cache *fs_cachep;
 298
 299 /* SLAB cache for vm_area_struct structures */
 300 struct kmem_cache *vm_area_cachep;
 301
 302 /* SLAB cache for mm_struct structures (tsk->mm) */
 303 static struct kmem_cache *mm_cachep;
 304
 305 static void account_kernel_stack(struct task_struct *tsk, int account)
 306 {
 307         void *stack = task_stack_page(tsk);
 308         struct vm_struct *vm = task_stack_vm_area(tsk);
 309
 310         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 311
 312         if (vm) {
 313                 int i;
 314
 315                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 316
 317                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 318                         mod_zone_page_state(page_zone(vm->pages[i]),
 319                                             NR_KERNEL_STACK_KB,
 320                                             PAGE_SIZE / 1024 * account);
 321                 }
 322
 323                 /* All stack pages belong to the same memcg. */
 324                 mod_memcg_page_state(vm->pages[0], MEMCG_KERNEL_STACK_KB,
 325                                      account * (THREAD_SIZE / 1024));
 326         } else {
 327                 /*
 328                  * All stack pages are in the same zone and belong to the
 329                  * same memcg.
 330                  */
 331                 struct page *first_page = virt_to_page(stack);
 332
 333                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 334                                     THREAD_SIZE / 1024 * account);
 335
 336                 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
 337                                      account * (THREAD_SIZE / 1024));
 338         }
 339 }
 340
 341 static void release_task_stack(struct task_struct *tsk)
 342 {
 343         if (WARN_ON(tsk->state != TASK_DEAD))
 344                 return;  /* Better to leak the stack than to free prematurely */
 345
 346         account_kernel_stack(tsk, -1);
 347         arch_release_thread_stack(tsk->stack);
 348         free_thread_stack(tsk);
 349         tsk->stack = NULL;
 350 #ifdef CONFIG_VMAP_STACK
 351         tsk->stack_vm_area = NULL;
 352 #endif
 353 }
 354
 355 #ifdef CONFIG_THREAD_INFO_IN_TASK
 356 void put_task_stack(struct task_struct *tsk)
 357 {
 358         if (atomic_dec_and_test(&tsk->stack_refcount))
 359                 release_task_stack(tsk);
 360 }
 361 #endif
 362
 363 void free_task(struct task_struct *tsk)
 364 {
 365 #ifndef CONFIG_THREAD_INFO_IN_TASK
 366         /*
 367          * The task is finally done with both the stack and thread_info,
 368          * so free both.
 369          */
 370         release_task_stack(tsk);
 371 #else
 372         /*
 373          * If the task had a separate stack allocation, it should be gone
 374          * by now.
 375          */
 376         WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
 377 #endif
 378         rt_mutex_debug_task_free(tsk);
 379         ftrace_graph_exit_task(tsk);
 380         put_seccomp_filter(tsk);
 381         arch_release_task_struct(tsk);
 382         if (tsk->flags & PF_KTHREAD)
 383                 free_kthread_struct(tsk);
 384         free_task_struct(tsk);
 385 }
 386 EXPORT_SYMBOL(free_task);
 387
 388 static inline void free_signal_struct(struct signal_struct *sig)
 389 {
 390         taskstats_tgid_free(sig);
 391         sched_autogroup_exit(sig);
 392         /*
 393          * __mmdrop is not safe to call from softirq context on x86 due to
 394          * pgd_dtor so postpone it to the async context
 395          */
 396         if (sig->oom_mm)
 397                 mmdrop_async(sig->oom_mm);
 398         kmem_cache_free(signal_cachep, sig);
 399 }
 400
 401 static inline void put_signal_struct(struct signal_struct *sig)
 402 {
 403         if (atomic_dec_and_test(&sig->sigcnt))
 404                 free_signal_struct(sig);
 405 }
 406
 407 void __put_task_struct(struct task_struct *tsk)
 408 {
 409         WARN_ON(!tsk->exit_state);
 410         WARN_ON(atomic_read(&tsk->usage));
 411         WARN_ON(tsk == current);
 412
 413         cgroup_free(tsk);
 414         task_numa_free(tsk);
 415         security_task_free(tsk);
 416         exit_creds(tsk);
 417         delayacct_tsk_free(tsk);
 418         put_signal_struct(tsk->signal);
 419
 420         if (!profile_handoff_task(tsk))
 421                 free_task(tsk);
 422 }
 423 EXPORT_SYMBOL_GPL(__put_task_struct);
 424
 425 void __init __weak arch_task_cache_init(void) { }
 426
 427 /*
 428  * set_max_threads
 429  */
 430 static void set_max_threads(unsigned int max_threads_suggested)
 431 {
 432         u64 threads;
 433
 434         /*
 435          * The number of threads shall be limited such that the thread
 436          * structures may only consume a small part of the available memory.
 437          */
 438         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 439                 threads = MAX_THREADS;
 440         else
 441                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 442                                     (u64) THREAD_SIZE * 8UL);
 443
 444         if (threads > max_threads_suggested)
 445                 threads = max_threads_suggested;
 446
 447         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 448 }
 449
 450 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 451 /* Initialized by the architecture: */
 452 int arch_task_struct_size __read_mostly;
 453 #endif
 454
 455 void __init fork_init(void)
 456 {
 457         int i;
 458 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 459 #ifndef ARCH_MIN_TASKALIGN
 460 #define ARCH_MIN_TASKALIGN      0
 461 #endif
 462         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 463
 464         /* create a slab on which task_structs can be allocated */
 465         task_struct_cachep = kmem_cache_create("task_struct",
 466                         arch_task_struct_size, align,
 467                         SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
 468 #endif
 469
 470         /* do the arch specific task caches init */
 471         arch_task_cache_init();
 472
 473         set_max_threads(MAX_THREADS);
 474
 475         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 476         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 477         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 478                 init_task.signal->rlim[RLIMIT_NPROC];
 479
 480         for (i = 0; i < UCOUNT_COUNTS; i++) {
 481                 init_user_ns.ucount_max[i] = max_threads/2;
 482         }
 483
 484 #ifdef CONFIG_VMAP_STACK
 485         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 486                           NULL, free_vm_stack_cache);
 487 #endif
 488
 489         lockdep_init_task(&init_task);
 490 }
 491
 492 int __weak arch_dup_task_struct(struct task_struct *dst,
 493                                                struct task_struct *src)
 494 {
 495         *dst = *src;
 496         return 0;
 497 }
 498
 499 void set_task_stack_end_magic(struct task_struct *tsk)
 500 {
 501         unsigned long *stackend;
 502
 503         stackend = end_of_stack(tsk);
 504         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 505 }
 506
 507 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 508 {
 509         struct task_struct *tsk;
 510         unsigned long *stack;
 511         struct vm_struct *stack_vm_area;
 512         int err;
 513
 514         if (node == NUMA_NO_NODE)
 515                 node = tsk_fork_get_node(orig);
 516         tsk = alloc_task_struct_node(node);
 517         if (!tsk)
 518                 return NULL;
 519
 520         stack = alloc_thread_stack_node(tsk, node);
 521         if (!stack)
 522                 goto free_tsk;
 523
 524         stack_vm_area = task_stack_vm_area(tsk);
 525
 526         err = arch_dup_task_struct(tsk, orig);
 527
 528         /*
 529          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 530          * sure they're properly initialized before using any stack-related
 531          * functions again.
 532          */
 533         tsk->stack = stack;
 534 #ifdef CONFIG_VMAP_STACK
 535         tsk->stack_vm_area = stack_vm_area;
 536 #endif
 537 #ifdef CONFIG_THREAD_INFO_IN_TASK
 538         atomic_set(&tsk->stack_refcount, 1);
 539 #endif
 540
 541         if (err)
 542                 goto free_stack;
 543
 544 #ifdef CONFIG_SECCOMP
 545         /*
 546          * We must handle setting up seccomp filters once we're under
 547          * the sighand lock in case orig has changed between now and
 548          * then. Until then, filter must be NULL to avoid messing up
 549          * the usage counts on the error path calling free_task.
 550          */
 551         tsk->seccomp.filter = NULL;
 552 #endif
 553
 554         setup_thread_stack(tsk, orig);
 555         clear_user_return_notifier(tsk);
 556         clear_tsk_need_resched(tsk);
 557         set_task_stack_end_magic(tsk);
 558
 559 #ifdef CONFIG_CC_STACKPROTECTOR
 560         tsk->stack_canary = get_random_canary();
 561 #endif
 562
 563         /*
 564          * One for us, one for whoever does the "release_task()" (usually
 565          * parent)
 566          */
 567         atomic_set(&tsk->usage, 2);
 568 #ifdef CONFIG_BLK_DEV_IO_TRACE
 569         tsk->btrace_seq = 0;
 570 #endif
 571         tsk->splice_pipe = NULL;
 572         tsk->task_frag.page = NULL;
 573         tsk->wake_q.next = NULL;
 574
 575         account_kernel_stack(tsk, 1);
 576
 577         kcov_task_init(tsk);
 578
 579 #ifdef CONFIG_FAULT_INJECTION
 580         tsk->fail_nth = 0;
 581 #endif
 582
 583         return tsk;
 584
 585 free_stack:
 586         free_thread_stack(tsk);
 587 free_tsk:
 588         free_task_struct(tsk);
 589         return NULL;
 590 }
 591
 592 #ifdef CONFIG_MMU
 593 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 594                                         struct mm_struct *oldmm)
 595 {
 596         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 597         struct rb_node **rb_link, *rb_parent;
 598         int retval;
 599         unsigned long charge;
 600         LIST_HEAD(uf);
 601
 602         uprobe_start_dup_mmap();
 603         if (down_write_killable(&oldmm->mmap_sem)) {
 604                 retval = -EINTR;
 605                 goto fail_uprobe_end;
 606         }
 607         flush_cache_dup_mm(oldmm);
 608         uprobe_dup_mmap(oldmm, mm);
 609         /*
 610          * Not linked in yet - no deadlock potential:
 611          */
 612         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 613
 614         /* No ordering required: file already has been exposed. */
 615         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 616
 617         mm->total_vm = oldmm->total_vm;
 618         mm->data_vm = oldmm->data_vm;
 619         mm->exec_vm = oldmm->exec_vm;
 620         mm->stack_vm = oldmm->stack_vm;
 621
 622         rb_link = &mm->mm_rb.rb_node;
 623         rb_parent = NULL;
 624         pprev = &mm->mmap;
 625         retval = ksm_fork(mm, oldmm);
 626         if (retval)
 627                 goto out;
 628         retval = khugepaged_fork(mm, oldmm);
 629         if (retval)
 630                 goto out;
 631
 632         prev = NULL;
 633         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 634                 struct file *file;
 635
 636                 if (mpnt->vm_flags & VM_DONTCOPY) {
 637                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 638                         continue;
 639                 }
 640                 charge = 0;
 641                 if (mpnt->vm_flags & VM_ACCOUNT) {
 642                         unsigned long len = vma_pages(mpnt);
 643
 644                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 645                                 goto fail_nomem;
 646                         charge = len;
 647                 }
 648                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 649                 if (!tmp)
 650                         goto fail_nomem;
 651                 *tmp = *mpnt;
 652                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 653                 retval = vma_dup_policy(mpnt, tmp);
 654                 if (retval)
 655                         goto fail_nomem_policy;
 656                 tmp->vm_mm = mm;
 657                 retval = dup_userfaultfd(tmp, &uf);
 658                 if (retval)
 659                         goto fail_nomem_anon_vma_fork;
 660                 if (anon_vma_fork(tmp, mpnt))
 661                         goto fail_nomem_anon_vma_fork;
 662                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 663                 tmp->vm_next = tmp->vm_prev = NULL;
 664                 file = tmp->vm_file;
 665                 if (file) {
 666                         struct inode *inode = file_inode(file);
 667                         struct address_space *mapping = file->f_mapping;
 668
 669                         get_file(file);
 670                         if (tmp->vm_flags & VM_DENYWRITE)
 671                                 atomic_dec(&inode->i_writecount);
 672                         i_mmap_lock_write(mapping);
 673                         if (tmp->vm_flags & VM_SHARED)
 674                                 atomic_inc(&mapping->i_mmap_writable);
 675                         flush_dcache_mmap_lock(mapping);
 676                         /* insert tmp into the share list, just after mpnt */
 677                         vma_interval_tree_insert_after(tmp, mpnt,
 678                                         &mapping->i_mmap);
 679                         flush_dcache_mmap_unlock(mapping);
 680                         i_mmap_unlock_write(mapping);
 681                 }
 682
 683                 /*
 684                  * Clear hugetlb-related page reserves for children. This only
 685                  * affects MAP_PRIVATE mappings. Faults generated by the child
 686                  * are not guaranteed to succeed, even if read-only
 687                  */
 688                 if (is_vm_hugetlb_page(tmp))
 689                         reset_vma_resv_huge_pages(tmp);
 690
 691                 /*
 692                  * Link in the new vma and copy the page table entries.
 693                  */
 694                 *pprev = tmp;
 695                 pprev = &tmp->vm_next;
 696                 tmp->vm_prev = prev;
 697                 prev = tmp;
 698
 699                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 700                 rb_link = &tmp->vm_rb.rb_right;
 701                 rb_parent = &tmp->vm_rb;
 702
 703                 mm->map_count++;
 704                 retval = copy_page_range(mm, oldmm, mpnt);
 705
 706                 if (tmp->vm_ops && tmp->vm_ops->open)
 707                         tmp->vm_ops->open(tmp);
 708
 709                 if (retval)
 710                         goto out;
 711         }
 712         /* a new mm has just been created */
 713         arch_dup_mmap(oldmm, mm);
 714         retval = 0;
 715 out:
 716         up_write(&mm->mmap_sem);
 717         flush_tlb_mm(oldmm);
 718         up_write(&oldmm->mmap_sem);
 719         dup_userfaultfd_complete(&uf);
 720 fail_uprobe_end:
 721         uprobe_end_dup_mmap();
 722         return retval;
 723 fail_nomem_anon_vma_fork:
 724         mpol_put(vma_policy(tmp));
 725 fail_nomem_policy:
 726         kmem_cache_free(vm_area_cachep, tmp);
 727 fail_nomem:
 728         retval = -ENOMEM;
 729         vm_unacct_memory(charge);
 730         goto out;
 731 }
 732
 733 static inline int mm_alloc_pgd(struct mm_struct *mm)
 734 {
 735         mm->pgd = pgd_alloc(mm);
 736         if (unlikely(!mm->pgd))
 737                 return -ENOMEM;
 738         return 0;
 739 }
 740
 741 static inline void mm_free_pgd(struct mm_struct *mm)
 742 {
 743         pgd_free(mm, mm->pgd);
 744 }
 745 #else
 746 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 747 {
 748         down_write(&oldmm->mmap_sem);
 749         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 750         up_write(&oldmm->mmap_sem);
 751         return 0;
 752 }
 753 #define mm_alloc_pgd(mm)        (0)
 754 #define mm_free_pgd(mm)
 755 #endif /* CONFIG_MMU */
 756
 757 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 758
 759 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 760 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 761
 762 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 763
 764 static int __init coredump_filter_setup(char *s)
 765 {
 766         default_dump_filter =
 767                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 768                 MMF_DUMP_FILTER_MASK;
 769         return 1;
 770 }
 771
 772 __setup("coredump_filter=", coredump_filter_setup);
 773
 774 #include <linux/init_task.h>
 775
 776 static void mm_init_aio(struct mm_struct *mm)
 777 {
 778 #ifdef CONFIG_AIO
 779         spin_lock_init(&mm->ioctx_lock);
 780         mm->ioctx_table = NULL;
 781 #endif
 782 }
 783
 784 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 785 {
 786 #ifdef CONFIG_MEMCG
 787         mm->owner = p;
 788 #endif
 789 }
 790
 791 static void mm_init_uprobes_state(struct mm_struct *mm)
 792 {
 793 #ifdef CONFIG_UPROBES
 794         mm->uprobes_state.xol_area = NULL;
 795 #endif
 796 }
 797
 798 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 799         struct user_namespace *user_ns)
 800 {
 801         mm->mmap = NULL;
 802         mm->mm_rb = RB_ROOT;
 803         mm->vmacache_seqnum = 0;
 804         atomic_set(&mm->mm_users, 1);
 805         atomic_set(&mm->mm_count, 1);
 806         init_rwsem(&mm->mmap_sem);
 807         INIT_LIST_HEAD(&mm->mmlist);
 808         mm->core_state = NULL;
 809         atomic_long_set(&mm->nr_ptes, 0);
 810         mm_nr_pmds_init(mm);
 811         mm->map_count = 0;
 812         mm->locked_vm = 0;
 813         mm->pinned_vm = 0;
 814         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 815         spin_lock_init(&mm->page_table_lock);
 816         mm_init_cpumask(mm);
 817         mm_init_aio(mm);
 818         mm_init_owner(mm, p);
 819         RCU_INIT_POINTER(mm->exe_file, NULL);
 820         mmu_notifier_mm_init(mm);
 821         init_tlb_flush_pending(mm);
 822 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 823         mm->pmd_huge_pte = NULL;
 824 #endif
 825         mm_init_uprobes_state(mm);
 826
 827         if (current->mm) {
 828                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 829                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 830         } else {
 831                 mm->flags = default_dump_filter;
 832                 mm->def_flags = 0;
 833         }
 834
 835         if (mm_alloc_pgd(mm))
 836                 goto fail_nopgd;
 837
 838         if (init_new_context(p, mm))
 839                 goto fail_nocontext;
 840
 841         mm->user_ns = get_user_ns(user_ns);
 842         return mm;
 843
 844 fail_nocontext:
 845         mm_free_pgd(mm);
 846 fail_nopgd:
 847         free_mm(mm);
 848         return NULL;
 849 }
 850
 851 static void check_mm(struct mm_struct *mm)
 852 {
 853         int i;
 854
 855         for (i = 0; i < NR_MM_COUNTERS; i++) {
 856                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 857
 858                 if (unlikely(x))
 859                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 860                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 861         }
 862
 863         if (atomic_long_read(&mm->nr_ptes))
 864                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 865                                 atomic_long_read(&mm->nr_ptes));
 866         if (mm_nr_pmds(mm))
 867                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 868                                 mm_nr_pmds(mm));
 869
 870 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 871         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 872 #endif
 873 }
 874
 875 /*
 876  * Allocate and initialize an mm_struct.
 877  */
 878 struct mm_struct *mm_alloc(void)
 879 {
 880         struct mm_struct *mm;
 881
 882         mm = allocate_mm();
 883         if (!mm)
 884                 return NULL;
 885
 886         memset(mm, 0, sizeof(*mm));
 887         return mm_init(mm, current, current_user_ns());
 888 }
 889
 890 /*
 891  * Called when the last reference to the mm
 892  * is dropped: either by a lazy thread or by
 893  * mmput. Free the page directory and the mm.
 894  */
 895 void __mmdrop(struct mm_struct *mm)
 896 {
 897         BUG_ON(mm == &init_mm);
 898         mm_free_pgd(mm);
 899         destroy_context(mm);
 900         mmu_notifier_mm_destroy(mm);
 901         check_mm(mm);
 902         put_user_ns(mm->user_ns);
 903         free_mm(mm);
 904 }
 905 EXPORT_SYMBOL_GPL(__mmdrop);
 906
 907 static inline void __mmput(struct mm_struct *mm)
 908 {
 909         VM_BUG_ON(atomic_read(&mm->mm_users));
 910
 911         uprobe_clear_state(mm);
 912         exit_aio(mm);
 913         ksm_exit(mm);
 914         khugepaged_exit(mm); /* must run before exit_mmap */
 915         exit_mmap(mm);
 916         mm_put_huge_zero_page(mm);
 917         set_mm_exe_file(mm, NULL);
 918         if (!list_empty(&mm->mmlist)) {
 919                 spin_lock(&mmlist_lock);
 920                 list_del(&mm->mmlist);
 921                 spin_unlock(&mmlist_lock);
 922         }
 923         if (mm->binfmt)
 924                 module_put(mm->binfmt->module);
 925         mmdrop(mm);
 926 }
 927
 928 /*
 929  * Decrement the use count and release all resources for an mm.
 930  */
 931 void mmput(struct mm_struct *mm)
 932 {
 933         might_sleep();
 934
 935         if (atomic_dec_and_test(&mm->mm_users))
 936                 __mmput(mm);
 937 }
 938 EXPORT_SYMBOL_GPL(mmput);
 939
 940 /**
 941  * set_mm_exe_file - change a reference to the mm's executable file
 942  *
 943  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 944  *
 945  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 946  * invocations: in mmput() nobody alive left, in execve task is single
 947  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 948  * mm->exe_file, but does so without using set_mm_exe_file() in order
 949  * to do avoid the need for any locks.
 950  */
 951 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 952 {
 953         struct file *old_exe_file;
 954
 955         /*
 956          * It is safe to dereference the exe_file without RCU as
 957          * this function is only called if nobody else can access
 958          * this mm -- see comment above for justification.
 959          */
 960         old_exe_file = rcu_dereference_raw(mm->exe_file);
 961
 962         if (new_exe_file)
 963                 get_file(new_exe_file);
 964         rcu_assign_pointer(mm->exe_file, new_exe_file);
 965         if (old_exe_file)
 966                 fput(old_exe_file);
 967 }
 968
 969 /**
 970  * get_mm_exe_file - acquire a reference to the mm's executable file
 971  *
 972  * Returns %NULL if mm has no associated executable file.
 973  * User must release file via fput().
 974  */
 975 struct file *get_mm_exe_file(struct mm_struct *mm)
 976 {
 977         struct file *exe_file;
 978
 979         rcu_read_lock();
 980         exe_file = rcu_dereference(mm->exe_file);
 981         if (exe_file && !get_file_rcu(exe_file))
 982                 exe_file = NULL;
 983         rcu_read_unlock();
 984         return exe_file;
 985 }
 986 EXPORT_SYMBOL(get_mm_exe_file);
 987
 988 /**
 989  * get_task_exe_file - acquire a reference to the task's executable file
 990  *
 991  * Returns %NULL if task's mm (if any) has no associated executable file or
 992  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
 993  * User must release file via fput().
 994  */
 995 struct file *get_task_exe_file(struct task_struct *task)
 996 {
 997         struct file *exe_file = NULL;
 998         struct mm_struct *mm;
 999
1000         task_lock(task);
1001         mm = task->mm;
1002         if (mm) {
1003                 if (!(task->flags & PF_KTHREAD))
1004                         exe_file = get_mm_exe_file(mm);
1005         }
1006         task_unlock(task);
1007         return exe_file;
1008 }
1009 EXPORT_SYMBOL(get_task_exe_file);
1010
1011 /**
1012  * get_task_mm - acquire a reference to the task's mm
1013  *
1014  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1015  * this kernel workthread has transiently adopted a user mm with use_mm,
1016  * to do its AIO) is not set and if so returns a reference to it, after
1017  * bumping up the use count.  User must release the mm via mmput()
1018  * after use.  Typically used by /proc and ptrace.
1019  */
1020 struct mm_struct *get_task_mm(struct task_struct *task)
1021 {
1022         struct mm_struct *mm;
1023
1024         task_lock(task);
1025         mm = task->mm;
1026         if (mm) {
1027                 if (task->flags & PF_KTHREAD)
1028                         mm = NULL;
1029                 else
1030                         mmget(mm);
1031         }
1032         task_unlock(task);
1033         return mm;
1034 }
1035 EXPORT_SYMBOL_GPL(get_task_mm);
1036
1037 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1038 {
1039         struct mm_struct *mm;
1040         int err;
1041
1042         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1043         if (err)
1044                 return ERR_PTR(err);
1045
1046         mm = get_task_mm(task);
1047         if (mm && mm != current->mm &&
1048                         !ptrace_may_access(task, mode)) {
1049                 mmput(mm);
1050                 mm = ERR_PTR(-EACCES);
1051         }
1052         mutex_unlock(&task->signal->cred_guard_mutex);
1053
1054         return mm;
1055 }
1056
1057 static void complete_vfork_done(struct task_struct *tsk)
1058 {
1059         struct completion *vfork;
1060
1061         task_lock(tsk);
1062         vfork = tsk->vfork_done;
1063         if (likely(vfork)) {
1064                 tsk->vfork_done = NULL;
1065                 complete(vfork);
1066         }
1067         task_unlock(tsk);
1068 }
1069
1070 static int wait_for_vfork_done(struct task_struct *child,
1071                                 struct completion *vfork)
1072 {
1073         int killed;
1074
1075         freezer_do_not_count();
1076         killed = wait_for_completion_killable(vfork);
1077         freezer_count();
1078
1079         if (killed) {
1080                 task_lock(child);
1081                 child->vfork_done = NULL;
1082                 task_unlock(child);
1083         }
1084
1085         put_task_struct(child);
1086         return killed;
1087 }
1088
1089 /* Please note the differences between mmput and mm_release.
1090  * mmput is called whenever we stop holding onto a mm_struct,
1091  * error success whatever.
1092  *
1093  * mm_release is called after a mm_struct has been removed
1094  * from the current process.
1095  *
1096  * This difference is important for error handling, when we
1097  * only half set up a mm_struct for a new process and need to restore
1098  * the old one.  Because we mmput the new mm_struct before
1099  * restoring the old one. . .
1100  * Eric Biederman 10 January 1998
1101  */
1102 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1103 {
1104         /* Get rid of any futexes when releasing the mm */
1105 #ifdef CONFIG_FUTEX
1106         if (unlikely(tsk->robust_list)) {
1107                 exit_robust_list(tsk);
1108                 tsk->robust_list = NULL;
1109         }
1110 #ifdef CONFIG_COMPAT
1111         if (unlikely(tsk->compat_robust_list)) {
1112                 compat_exit_robust_list(tsk);
1113                 tsk->compat_robust_list = NULL;
1114         }
1115 #endif
1116         if (unlikely(!list_empty(&tsk->pi_state_list)))
1117                 exit_pi_state_list(tsk);
1118 #endif
1119
1120         uprobe_free_utask(tsk);
1121
1122         /* Get rid of any cached register state */
1123         deactivate_mm(tsk, mm);
1124
1125         /*
1126          * Signal userspace if we're not exiting with a core dump
1127          * because we want to leave the value intact for debugging
1128          * purposes.
1129          */
1130         if (tsk->clear_child_tid) {
1131                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1132                     atomic_read(&mm->mm_users) > 1) {
1133                         /*
1134                          * We don't check the error code - if userspace has
1135                          * not set up a proper pointer then tough luck.
1136                          */
1137                         put_user(0, tsk->clear_child_tid);
1138                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1139                                         1, NULL, NULL, 0);
1140                 }
1141                 tsk->clear_child_tid = NULL;
1142         }
1143
1144         /*
1145          * All done, finally we can wake up parent and return this mm to him.
1146          * Also kthread_stop() uses this completion for synchronization.
1147          */
1148         if (tsk->vfork_done)
1149                 complete_vfork_done(tsk);
1150 }
1151
1152 /*
1153  * Allocate a new mm structure and copy contents from the
1154  * mm structure of the passed in task structure.
1155  */
1156 static struct mm_struct *dup_mm(struct task_struct *tsk)
1157 {
1158         struct mm_struct *mm, *oldmm = current->mm;
1159         int err;
1160
1161         mm = allocate_mm();
1162         if (!mm)
1163                 goto fail_nomem;
1164
1165         memcpy(mm, oldmm, sizeof(*mm));
1166
1167         if (!mm_init(mm, tsk, mm->user_ns))
1168                 goto fail_nomem;
1169
1170         err = dup_mmap(mm, oldmm);
1171         if (err)
1172                 goto free_pt;
1173
1174         mm->hiwater_rss = get_mm_rss(mm);
1175         mm->hiwater_vm = mm->total_vm;
1176
1177         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1178                 goto free_pt;
1179
1180         return mm;
1181
1182 free_pt:
1183         /* don't put binfmt in mmput, we haven't got module yet */
1184         mm->binfmt = NULL;
1185         mmput(mm);
1186
1187 fail_nomem:
1188         return NULL;
1189 }
1190
1191 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1192 {
1193         struct mm_struct *mm, *oldmm;
1194         int retval;
1195
1196         tsk->min_flt = tsk->maj_flt = 0;
1197         tsk->nvcsw = tsk->nivcsw = 0;
1198 #ifdef CONFIG_DETECT_HUNG_TASK
1199         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1200 #endif
1201
1202         tsk->mm = NULL;
1203         tsk->active_mm = NULL;
1204
1205         /*
1206          * Are we cloning a kernel thread?
1207          *
1208          * We need to steal a active VM for that..
1209          */
1210         oldmm = current->mm;
1211         if (!oldmm)
1212                 return 0;
1213
1214         /* initialize the new vmacache entries */
1215         vmacache_flush(tsk);
1216
1217         if (clone_flags & CLONE_VM) {
1218                 mmget(oldmm);
1219                 mm = oldmm;
1220                 goto good_mm;
1221         }
1222
1223         retval = -ENOMEM;
1224         mm = dup_mm(tsk);
1225         if (!mm)
1226                 goto fail_nomem;
1227
1228 good_mm:
1229         tsk->mm = mm;
1230         tsk->active_mm = mm;
1231         return 0;
1232
1233 fail_nomem:
1234         return retval;
1235 }
1236
1237 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1238 {
1239         struct fs_struct *fs = current->fs;
1240         if (clone_flags & CLONE_FS) {
1241                 /* tsk->fs is already what we want */
1242                 spin_lock(&fs->lock);
1243                 if (fs->in_exec) {
1244                         spin_unlock(&fs->lock);
1245                         return -EAGAIN;
1246                 }
1247                 fs->users++;
1248                 spin_unlock(&fs->lock);
1249                 return 0;
1250         }
1251         tsk->fs = copy_fs_struct(fs);
1252         if (!tsk->fs)
1253                 return -ENOMEM;
1254         return 0;
1255 }
1256
1257 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1258 {
1259         struct files_struct *oldf, *newf;
1260         int error = 0;
1261
1262         /*
1263          * A background process may not have any files ...
1264          */
1265         oldf = current->files;
1266         if (!oldf)
1267                 goto out;
1268
1269         if (clone_flags & CLONE_FILES) {
1270                 atomic_inc(&oldf->count);
1271                 goto out;
1272         }
1273
1274         newf = dup_fd(oldf, &error);
1275         if (!newf)
1276                 goto out;
1277
1278         tsk->files = newf;
1279         error = 0;
1280 out:
1281         return error;
1282 }
1283
1284 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1285 {
1286 #ifdef CONFIG_BLOCK
1287         struct io_context *ioc = current->io_context;
1288         struct io_context *new_ioc;
1289
1290         if (!ioc)
1291                 return 0;
1292         /*
1293          * Share io context with parent, if CLONE_IO is set
1294          */
1295         if (clone_flags & CLONE_IO) {
1296                 ioc_task_link(ioc);
1297                 tsk->io_context = ioc;
1298         } else if (ioprio_valid(ioc->ioprio)) {
1299                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1300                 if (unlikely(!new_ioc))
1301                         return -ENOMEM;
1302
1303                 new_ioc->ioprio = ioc->ioprio;
1304                 put_io_context(new_ioc);
1305         }
1306 #endif
1307         return 0;
1308 }
1309
1310 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1311 {
1312         struct sighand_struct *sig;
1313
1314         if (clone_flags & CLONE_SIGHAND) {
1315                 atomic_inc(&current->sighand->count);
1316                 return 0;
1317         }
1318         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1319         rcu_assign_pointer(tsk->sighand, sig);
1320         if (!sig)
1321                 return -ENOMEM;
1322
1323         atomic_set(&sig->count, 1);
1324         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1325         return 0;
1326 }
1327
1328 void __cleanup_sighand(struct sighand_struct *sighand)
1329 {
1330         if (atomic_dec_and_test(&sighand->count)) {
1331                 signalfd_cleanup(sighand);
1332                 /*
1333                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1334                  * without an RCU grace period, see __lock_task_sighand().
1335                  */
1336                 kmem_cache_free(sighand_cachep, sighand);
1337         }
1338 }
1339
1340 #ifdef CONFIG_POSIX_TIMERS
1341 /*
1342  * Initialize POSIX timer handling for a thread group.
1343  */
1344 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1345 {
1346         unsigned long cpu_limit;
1347
1348         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1349         if (cpu_limit != RLIM_INFINITY) {
1350                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1351                 sig->cputimer.running = true;
1352         }
1353
1354         /* The timer lists. */
1355         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1356         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1357         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1358 }
1359 #else
1360 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1361 #endif
1362
1363 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1364 {
1365         struct signal_struct *sig;
1366
1367         if (clone_flags & CLONE_THREAD)
1368                 return 0;
1369
1370         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1371         tsk->signal = sig;
1372         if (!sig)
1373                 return -ENOMEM;
1374
1375         sig->nr_threads = 1;
1376         atomic_set(&sig->live, 1);
1377         atomic_set(&sig->sigcnt, 1);
1378
1379         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1380         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1381         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1382
1383         init_waitqueue_head(&sig->wait_chldexit);
1384         sig->curr_target = tsk;
1385         init_sigpending(&sig->shared_pending);
1386         seqlock_init(&sig->stats_lock);
1387         prev_cputime_init(&sig->prev_cputime);
1388
1389 #ifdef CONFIG_POSIX_TIMERS
1390         INIT_LIST_HEAD(&sig->posix_timers);
1391         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1392         sig->real_timer.function = it_real_fn;
1393 #endif
1394
1395         task_lock(current->group_leader);
1396         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1397         task_unlock(current->group_leader);
1398
1399         posix_cpu_timers_init_group(sig);
1400
1401         tty_audit_fork(sig);
1402         sched_autogroup_fork(sig);
1403
1404         sig->oom_score_adj = current->signal->oom_score_adj;
1405         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1406
1407         mutex_init(&sig->cred_guard_mutex);
1408
1409         return 0;
1410 }
1411
1412 static void copy_seccomp(struct task_struct *p)
1413 {
1414 #ifdef CONFIG_SECCOMP
1415         /*
1416          * Must be called with sighand->lock held, which is common to
1417          * all threads in the group. Holding cred_guard_mutex is not
1418          * needed because this new task is not yet running and cannot
1419          * be racing exec.
1420          */
1421         assert_spin_locked(&current->sighand->siglock);
1422
1423         /* Ref-count the new filter user, and assign it. */
1424         get_seccomp_filter(current);
1425         p->seccomp = current->seccomp;
1426
1427         /*
1428          * Explicitly enable no_new_privs here in case it got set
1429          * between the task_struct being duplicated and holding the
1430          * sighand lock. The seccomp state and nnp must be in sync.
1431          */
1432         if (task_no_new_privs(current))
1433                 task_set_no_new_privs(p);
1434
1435         /*
1436          * If the parent gained a seccomp mode after copying thread
1437          * flags and between before we held the sighand lock, we have
1438          * to manually enable the seccomp thread flag here.
1439          */
1440         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1441                 set_tsk_thread_flag(p, TIF_SECCOMP);
1442 #endif
1443 }
1444
1445 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1446 {
1447         current->clear_child_tid = tidptr;
1448
1449         return task_pid_vnr(current);
1450 }
1451
1452 static void rt_mutex_init_task(struct task_struct *p)
1453 {
1454         raw_spin_lock_init(&p->pi_lock);
1455 #ifdef CONFIG_RT_MUTEXES
1456         p->pi_waiters = RB_ROOT;
1457         p->pi_waiters_leftmost = NULL;
1458         p->pi_top_task = NULL;
1459         p->pi_blocked_on = NULL;
1460 #endif
1461 }
1462
1463 #ifdef CONFIG_POSIX_TIMERS
1464 /*
1465  * Initialize POSIX timer handling for a single task.
1466  */
1467 static void posix_cpu_timers_init(struct task_struct *tsk)
1468 {
1469         tsk->cputime_expires.prof_exp = 0;
1470         tsk->cputime_expires.virt_exp = 0;
1471         tsk->cputime_expires.sched_exp = 0;
1472         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1473         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1474         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1475 }
1476 #else
1477 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1478 #endif
1479
1480 static inline void
1481 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1482 {
1483          task->pids[type].pid = pid;
1484 }
1485
1486 static inline void rcu_copy_process(struct task_struct *p)
1487 {
1488 #ifdef CONFIG_PREEMPT_RCU
1489         p->rcu_read_lock_nesting = 0;
1490         p->rcu_read_unlock_special.s = 0;
1491         p->rcu_blocked_node = NULL;
1492         INIT_LIST_HEAD(&p->rcu_node_entry);
1493 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1494 #ifdef CONFIG_TASKS_RCU
1495         p->rcu_tasks_holdout = false;
1496         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1497         p->rcu_tasks_idle_cpu = -1;
1498 #endif /* #ifdef CONFIG_TASKS_RCU */
1499 }
1500
1501 /*
1502  * This creates a new process as a copy of the old one,
1503  * but does not actually start it yet.
1504  *
1505  * It copies the registers, and all the appropriate
1506  * parts of the process environment (as per the clone
1507  * flags). The actual kick-off is left to the caller.
1508  */
1509 static __latent_entropy struct task_struct *copy_process(
1510                                         unsigned long clone_flags,
1511                                         unsigned long stack_start,
1512                                         unsigned long stack_size,
1513                                         int __user *child_tidptr,
1514                                         struct pid *pid,
1515                                         int trace,
1516                                         unsigned long tls,
1517                                         int node)
1518 {
1519         int retval;
1520         struct task_struct *p;
1521
1522         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1523                 return ERR_PTR(-EINVAL);
1524
1525         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1526                 return ERR_PTR(-EINVAL);
1527
1528         /*
1529          * Thread groups must share signals as well, and detached threads
1530          * can only be started up within the thread group.
1531          */
1532         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1533                 return ERR_PTR(-EINVAL);
1534
1535         /*
1536          * Shared signal handlers imply shared VM. By way of the above,
1537          * thread groups also imply shared VM. Blocking this case allows
1538          * for various simplifications in other code.
1539          */
1540         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1541                 return ERR_PTR(-EINVAL);
1542
1543         /*
1544          * Siblings of global init remain as zombies on exit since they are
1545          * not reaped by their parent (swapper). To solve this and to avoid
1546          * multi-rooted process trees, prevent global and container-inits
1547          * from creating siblings.
1548          */
1549         if ((clone_flags & CLONE_PARENT) &&
1550                                 current->signal->flags & SIGNAL_UNKILLABLE)
1551                 return ERR_PTR(-EINVAL);
1552
1553         /*
1554          * If the new process will be in a different pid or user namespace
1555          * do not allow it to share a thread group with the forking task.
1556          */
1557         if (clone_flags & CLONE_THREAD) {
1558                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1559                     (task_active_pid_ns(current) !=
1560                                 current->nsproxy->pid_ns_for_children))
1561                         return ERR_PTR(-EINVAL);
1562         }
1563
1564         retval = security_task_create(clone_flags);
1565         if (retval)
1566                 goto fork_out;
1567
1568         retval = -ENOMEM;
1569         p = dup_task_struct(current, node);
1570         if (!p)
1571                 goto fork_out;
1572
1573         /*
1574          * This _must_ happen before we call free_task(), i.e. before we jump
1575          * to any of the bad_fork_* labels. This is to avoid freeing
1576          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1577          * kernel threads (PF_KTHREAD).
1578          */
1579         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1580         /*
1581          * Clear TID on mm_release()?
1582          */
1583         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1584
1585         ftrace_graph_init_task(p);
1586
1587         rt_mutex_init_task(p);
1588
1589 #ifdef CONFIG_PROVE_LOCKING
1590         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1591         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1592 #endif
1593         retval = -EAGAIN;
1594         if (atomic_read(&p->real_cred->user->processes) >=
1595                         task_rlimit(p, RLIMIT_NPROC)) {
1596                 if (p->real_cred->user != INIT_USER &&
1597                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1598                         goto bad_fork_free;
1599         }
1600         current->flags &= ~PF_NPROC_EXCEEDED;
1601
1602         retval = copy_creds(p, clone_flags);
1603         if (retval < 0)
1604                 goto bad_fork_free;
1605
1606         /*
1607          * If multiple threads are within copy_process(), then this check
1608          * triggers too late. This doesn't hurt, the check is only there
1609          * to stop root fork bombs.
1610          */
1611         retval = -EAGAIN;
1612         if (nr_threads >= max_threads)
1613                 goto bad_fork_cleanup_count;
1614
1615         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1616         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1617         p->flags |= PF_FORKNOEXEC;
1618         INIT_LIST_HEAD(&p->children);
1619         INIT_LIST_HEAD(&p->sibling);
1620         rcu_copy_process(p);
1621         p->vfork_done = NULL;
1622         spin_lock_init(&p->alloc_lock);
1623
1624         init_sigpending(&p->pending);
1625
1626         p->utime = p->stime = p->gtime = 0;
1627 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1628         p->utimescaled = p->stimescaled = 0;
1629 #endif
1630         prev_cputime_init(&p->prev_cputime);
1631
1632 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1633         seqcount_init(&p->vtime.seqcount);
1634         p->vtime.starttime = 0;
1635         p->vtime.state = VTIME_INACTIVE;
1636 #endif
1637
1638 #if defined(SPLIT_RSS_COUNTING)
1639         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1640 #endif
1641
1642         p->default_timer_slack_ns = current->timer_slack_ns;
1643
1644         task_io_accounting_init(&p->ioac);
1645         acct_clear_integrals(p);
1646
1647         posix_cpu_timers_init(p);
1648
1649         p->start_time = ktime_get_ns();
1650         p->real_start_time = ktime_get_boot_ns();
1651         p->io_context = NULL;
1652         p->audit_context = NULL;
1653         cgroup_fork(p);
1654 #ifdef CONFIG_NUMA
1655         p->mempolicy = mpol_dup(p->mempolicy);
1656         if (IS_ERR(p->mempolicy)) {
1657                 retval = PTR_ERR(p->mempolicy);
1658                 p->mempolicy = NULL;
1659                 goto bad_fork_cleanup_threadgroup_lock;
1660         }
1661 #endif
1662 #ifdef CONFIG_CPUSETS
1663         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1664         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1665         seqcount_init(&p->mems_allowed_seq);
1666 #endif
1667 #ifdef CONFIG_TRACE_IRQFLAGS
1668         p->irq_events = 0;
1669         p->hardirqs_enabled = 0;
1670         p->hardirq_enable_ip = 0;
1671         p->hardirq_enable_event = 0;
1672         p->hardirq_disable_ip = _THIS_IP_;
1673         p->hardirq_disable_event = 0;
1674         p->softirqs_enabled = 1;
1675         p->softirq_enable_ip = _THIS_IP_;
1676         p->softirq_enable_event = 0;
1677         p->softirq_disable_ip = 0;
1678         p->softirq_disable_event = 0;
1679         p->hardirq_context = 0;
1680         p->softirq_context = 0;
1681 #endif
1682
1683         p->pagefault_disabled = 0;
1684
1685 #ifdef CONFIG_LOCKDEP
1686         p->lockdep_depth = 0; /* no locks held yet */
1687         p->curr_chain_key = 0;
1688         p->lockdep_recursion = 0;
1689         lockdep_init_task(p);
1690 #endif
1691
1692 #ifdef CONFIG_DEBUG_MUTEXES
1693         p->blocked_on = NULL; /* not blocked yet */
1694 #endif
1695 #ifdef CONFIG_BCACHE
1696         p->sequential_io        = 0;
1697         p->sequential_io_avg    = 0;
1698 #endif
1699
1700         /* Perform scheduler related setup. Assign this task to a CPU. */
1701         retval = sched_fork(clone_flags, p);
1702         if (retval)
1703                 goto bad_fork_cleanup_policy;
1704
1705         retval = perf_event_init_task(p);
1706         if (retval)
1707                 goto bad_fork_cleanup_policy;
1708         retval = audit_alloc(p);
1709         if (retval)
1710                 goto bad_fork_cleanup_perf;
1711         /* copy all the process information */
1712         shm_init_task(p);
1713         retval = security_task_alloc(p, clone_flags);
1714         if (retval)
1715                 goto bad_fork_cleanup_audit;
1716         retval = copy_semundo(clone_flags, p);
1717         if (retval)
1718                 goto bad_fork_cleanup_security;
1719         retval = copy_files(clone_flags, p);
1720         if (retval)
1721                 goto bad_fork_cleanup_semundo;
1722         retval = copy_fs(clone_flags, p);
1723         if (retval)
1724                 goto bad_fork_cleanup_files;
1725         retval = copy_sighand(clone_flags, p);
1726         if (retval)
1727                 goto bad_fork_cleanup_fs;
1728         retval = copy_signal(clone_flags, p);
1729         if (retval)
1730                 goto bad_fork_cleanup_sighand;
1731         retval = copy_mm(clone_flags, p);
1732         if (retval)
1733                 goto bad_fork_cleanup_signal;
1734         retval = copy_namespaces(clone_flags, p);
1735         if (retval)
1736                 goto bad_fork_cleanup_mm;
1737         retval = copy_io(clone_flags, p);
1738         if (retval)
1739                 goto bad_fork_cleanup_namespaces;
1740         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1741         if (retval)
1742                 goto bad_fork_cleanup_io;
1743
1744         if (pid != &init_struct_pid) {
1745                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1746                 if (IS_ERR(pid)) {
1747                         retval = PTR_ERR(pid);
1748                         goto bad_fork_cleanup_thread;
1749                 }
1750         }
1751
1752 #ifdef CONFIG_BLOCK
1753         p->plug = NULL;
1754 #endif
1755 #ifdef CONFIG_FUTEX
1756         p->robust_list = NULL;
1757 #ifdef CONFIG_COMPAT
1758         p->compat_robust_list = NULL;
1759 #endif
1760         INIT_LIST_HEAD(&p->pi_state_list);
1761         p->pi_state_cache = NULL;
1762 #endif
1763         /*
1764          * sigaltstack should be cleared when sharing the same VM
1765          */
1766         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1767                 sas_ss_reset(p);
1768
1769         /*
1770          * Syscall tracing and stepping should be turned off in the
1771          * child regardless of CLONE_PTRACE.
1772          */
1773         user_disable_single_step(p);
1774         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1775 #ifdef TIF_SYSCALL_EMU
1776         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1777 #endif
1778         clear_all_latency_tracing(p);
1779
1780         /* ok, now we should be set up.. */
1781         p->pid = pid_nr(pid);
1782         if (clone_flags & CLONE_THREAD) {
1783                 p->exit_signal = -1;
1784                 p->group_leader = current->group_leader;
1785                 p->tgid = current->tgid;
1786         } else {
1787                 if (clone_flags & CLONE_PARENT)
1788                         p->exit_signal = current->group_leader->exit_signal;
1789                 else
1790                         p->exit_signal = (clone_flags & CSIGNAL);
1791                 p->group_leader = p;
1792                 p->tgid = p->pid;
1793         }
1794
1795         p->nr_dirtied = 0;
1796         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1797         p->dirty_paused_when = 0;
1798
1799         p->pdeath_signal = 0;
1800         INIT_LIST_HEAD(&p->thread_group);
1801         p->task_works = NULL;
1802
1803         cgroup_threadgroup_change_begin(current);
1804         /*
1805          * Ensure that the cgroup subsystem policies allow the new process to be
1806          * forked. It should be noted the the new process's css_set can be changed
1807          * between here and cgroup_post_fork() if an organisation operation is in
1808          * progress.
1809          */
1810         retval = cgroup_can_fork(p);
1811         if (retval)
1812                 goto bad_fork_free_pid;
1813
1814         /*
1815          * Make it visible to the rest of the system, but dont wake it up yet.
1816          * Need tasklist lock for parent etc handling!
1817          */
1818         write_lock_irq(&tasklist_lock);
1819
1820         /* CLONE_PARENT re-uses the old parent */
1821         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1822                 p->real_parent = current->real_parent;
1823                 p->parent_exec_id = current->parent_exec_id;
1824         } else {
1825                 p->real_parent = current;
1826                 p->parent_exec_id = current->self_exec_id;
1827         }
1828
1829         klp_copy_process(p);
1830
1831         spin_lock(&current->sighand->siglock);
1832
1833         /*
1834          * Copy seccomp details explicitly here, in case they were changed
1835          * before holding sighand lock.
1836          */
1837         copy_seccomp(p);
1838
1839         /*
1840          * Process group and session signals need to be delivered to just the
1841          * parent before the fork or both the parent and the child after the
1842          * fork. Restart if a signal comes in before we add the new process to
1843          * it's process group.
1844          * A fatal signal pending means that current will exit, so the new
1845          * thread can't slip out of an OOM kill (or normal SIGKILL).
1846         */
1847         recalc_sigpending();
1848         if (signal_pending(current)) {
1849                 retval = -ERESTARTNOINTR;
1850                 goto bad_fork_cancel_cgroup;
1851         }
1852         if (unlikely(!(ns_of_pid(pid)->nr_hashed & PIDNS_HASH_ADDING))) {
1853                 retval = -ENOMEM;
1854                 goto bad_fork_cancel_cgroup;
1855         }
1856
1857         if (likely(p->pid)) {
1858                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1859
1860                 init_task_pid(p, PIDTYPE_PID, pid);
1861                 if (thread_group_leader(p)) {
1862                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1863                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1864
1865                         if (is_child_reaper(pid)) {
1866                                 ns_of_pid(pid)->child_reaper = p;
1867                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1868                         }
1869
1870                         p->signal->leader_pid = pid;
1871                         p->signal->tty = tty_kref_get(current->signal->tty);
1872                         /*
1873                          * Inherit has_child_subreaper flag under the same
1874                          * tasklist_lock with adding child to the process tree
1875                          * for propagate_has_child_subreaper optimization.
1876                          */
1877                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1878                                                          p->real_parent->signal->is_child_subreaper;
1879                         list_add_tail(&p->sibling, &p->real_parent->children);
1880                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1881                         attach_pid(p, PIDTYPE_PGID);
1882                         attach_pid(p, PIDTYPE_SID);
1883                         __this_cpu_inc(process_counts);
1884                 } else {
1885                         current->signal->nr_threads++;
1886                         atomic_inc(&current->signal->live);
1887                         atomic_inc(&current->signal->sigcnt);
1888                         list_add_tail_rcu(&p->thread_group,
1889                                           &p->group_leader->thread_group);
1890                         list_add_tail_rcu(&p->thread_node,
1891                                           &p->signal->thread_head);
1892                 }
1893                 attach_pid(p, PIDTYPE_PID);
1894                 nr_threads++;
1895         }
1896
1897         total_forks++;
1898         spin_unlock(&current->sighand->siglock);
1899         syscall_tracepoint_update(p);
1900         write_unlock_irq(&tasklist_lock);
1901
1902         proc_fork_connector(p);
1903         cgroup_post_fork(p);
1904         cgroup_threadgroup_change_end(current);
1905         perf_event_fork(p);
1906
1907         trace_task_newtask(p, clone_flags);
1908         uprobe_copy_process(p, clone_flags);
1909
1910         return p;
1911
1912 bad_fork_cancel_cgroup:
1913         spin_unlock(&current->sighand->siglock);
1914         write_unlock_irq(&tasklist_lock);
1915         cgroup_cancel_fork(p);
1916 bad_fork_free_pid:
1917         cgroup_threadgroup_change_end(current);
1918         if (pid != &init_struct_pid)
1919                 free_pid(pid);
1920 bad_fork_cleanup_thread:
1921         exit_thread(p);
1922 bad_fork_cleanup_io:
1923         if (p->io_context)
1924                 exit_io_context(p);
1925 bad_fork_cleanup_namespaces:
1926         exit_task_namespaces(p);
1927 bad_fork_cleanup_mm:
1928         if (p->mm)
1929                 mmput(p->mm);
1930 bad_fork_cleanup_signal:
1931         if (!(clone_flags & CLONE_THREAD))
1932                 free_signal_struct(p->signal);
1933 bad_fork_cleanup_sighand:
1934         __cleanup_sighand(p->sighand);
1935 bad_fork_cleanup_fs:
1936         exit_fs(p); /* blocking */
1937 bad_fork_cleanup_files:
1938         exit_files(p); /* blocking */
1939 bad_fork_cleanup_semundo:
1940         exit_sem(p);
1941 bad_fork_cleanup_security:
1942         security_task_free(p);
1943 bad_fork_cleanup_audit:
1944         audit_free(p);
1945 bad_fork_cleanup_perf:
1946         perf_event_free_task(p);
1947 bad_fork_cleanup_policy:
1948         lockdep_free_task(p);
1949 #ifdef CONFIG_NUMA
1950         mpol_put(p->mempolicy);
1951 bad_fork_cleanup_threadgroup_lock:
1952 #endif
1953         delayacct_tsk_free(p);
1954 bad_fork_cleanup_count:
1955         atomic_dec(&p->cred->user->processes);
1956         exit_creds(p);
1957 bad_fork_free:
1958         p->state = TASK_DEAD;
1959         put_task_stack(p);
1960         free_task(p);
1961 fork_out:
1962         return ERR_PTR(retval);
1963 }
1964
1965 static inline void init_idle_pids(struct pid_link *links)
1966 {
1967         enum pid_type type;
1968
1969         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1970                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1971                 links[type].pid = &init_struct_pid;
1972         }
1973 }
1974
1975 struct task_struct *fork_idle(int cpu)
1976 {
1977         struct task_struct *task;
1978         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1979                             cpu_to_node(cpu));
1980         if (!IS_ERR(task)) {
1981                 init_idle_pids(task->pids);
1982                 init_idle(task, cpu);
1983         }
1984
1985         return task;
1986 }
1987
1988 /*
1989  *  Ok, this is the main fork-routine.
1990  *
1991  * It copies the process, and if successful kick-starts
1992  * it and waits for it to finish using the VM if required.
1993  */
1994 long _do_fork(unsigned long clone_flags,
1995               unsigned long stack_start,
1996               unsigned long stack_size,
1997               int __user *parent_tidptr,
1998               int __user *child_tidptr,
1999               unsigned long tls)
2000 {
2001         struct task_struct *p;
2002         int trace = 0;
2003         long nr;
2004
2005         /*
2006          * Determine whether and which event to report to ptracer.  When
2007          * called from kernel_thread or CLONE_UNTRACED is explicitly
2008          * requested, no event is reported; otherwise, report if the event
2009          * for the type of forking is enabled.
2010          */
2011         if (!(clone_flags & CLONE_UNTRACED)) {
2012                 if (clone_flags & CLONE_VFORK)
2013                         trace = PTRACE_EVENT_VFORK;
2014                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2015                         trace = PTRACE_EVENT_CLONE;
2016                 else
2017                         trace = PTRACE_EVENT_FORK;
2018
2019                 if (likely(!ptrace_event_enabled(current, trace)))
2020                         trace = 0;
2021         }
2022
2023         p = copy_process(clone_flags, stack_start, stack_size,
2024                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2025         add_latent_entropy();
2026         /*
2027          * Do this prior waking up the new thread - the thread pointer
2028          * might get invalid after that point, if the thread exits quickly.
2029          */
2030         if (!IS_ERR(p)) {
2031                 struct completion vfork;
2032                 struct pid *pid;
2033
2034                 trace_sched_process_fork(current, p);
2035
2036                 pid = get_task_pid(p, PIDTYPE_PID);
2037                 nr = pid_vnr(pid);
2038
2039                 if (clone_flags & CLONE_PARENT_SETTID)
2040                         put_user(nr, parent_tidptr);
2041
2042                 if (clone_flags & CLONE_VFORK) {
2043                         p->vfork_done = &vfork;
2044                         init_completion(&vfork);
2045                         get_task_struct(p);
2046                 }
2047
2048                 wake_up_new_task(p);
2049
2050                 /* forking complete and child started to run, tell ptracer */
2051                 if (unlikely(trace))
2052                         ptrace_event_pid(trace, pid);
2053
2054                 if (clone_flags & CLONE_VFORK) {
2055                         if (!wait_for_vfork_done(p, &vfork))
2056                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2057                 }
2058
2059                 put_pid(pid);
2060         } else {
2061                 nr = PTR_ERR(p);
2062         }
2063         return nr;
2064 }
2065
2066 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2067 /* For compatibility with architectures that call do_fork directly rather than
2068  * using the syscall entry points below. */
2069 long do_fork(unsigned long clone_flags,
2070               unsigned long stack_start,
2071               unsigned long stack_size,
2072               int __user *parent_tidptr,
2073               int __user *child_tidptr)
2074 {
2075         return _do_fork(clone_flags, stack_start, stack_size,
2076                         parent_tidptr, child_tidptr, 0);
2077 }
2078 #endif
2079
2080 /*
2081  * Create a kernel thread.
2082  */
2083 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2084 {
2085         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2086                 (unsigned long)arg, NULL, NULL, 0);
2087 }
2088
2089 #ifdef __ARCH_WANT_SYS_FORK
2090 SYSCALL_DEFINE0(fork)
2091 {
2092 #ifdef CONFIG_MMU
2093         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2094 #else
2095         /* can not support in nommu mode */
2096         return -EINVAL;
2097 #endif
2098 }
2099 #endif
2100
2101 #ifdef __ARCH_WANT_SYS_VFORK
2102 SYSCALL_DEFINE0(vfork)
2103 {
2104         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2105                         0, NULL, NULL, 0);
2106 }
2107 #endif
2108
2109 #ifdef __ARCH_WANT_SYS_CLONE
2110 #ifdef CONFIG_CLONE_BACKWARDS
2111 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2112                  int __user *, parent_tidptr,
2113                  unsigned long, tls,
2114                  int __user *, child_tidptr)
2115 #elif defined(CONFIG_CLONE_BACKWARDS2)
2116 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2117                  int __user *, parent_tidptr,
2118                  int __user *, child_tidptr,
2119                  unsigned long, tls)
2120 #elif defined(CONFIG_CLONE_BACKWARDS3)
2121 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2122                 int, stack_size,
2123                 int __user *, parent_tidptr,
2124                 int __user *, child_tidptr,
2125                 unsigned long, tls)
2126 #else
2127 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2128                  int __user *, parent_tidptr,
2129                  int __user *, child_tidptr,
2130                  unsigned long, tls)
2131 #endif
2132 {
2133         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2134 }
2135 #endif
2136
2137 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2138 {
2139         struct task_struct *leader, *parent, *child;
2140         int res;
2141
2142         read_lock(&tasklist_lock);
2143         leader = top = top->group_leader;
2144 down:
2145         for_each_thread(leader, parent) {
2146                 list_for_each_entry(child, &parent->children, sibling) {
2147                         res = visitor(child, data);
2148                         if (res) {
2149                                 if (res < 0)
2150                                         goto out;
2151                                 leader = child;
2152                                 goto down;
2153                         }
2154 up:
2155                         ;
2156                 }
2157         }
2158
2159         if (leader != top) {
2160                 child = leader;
2161                 parent = child->real_parent;
2162                 leader = parent->group_leader;
2163                 goto up;
2164         }
2165 out:
2166         read_unlock(&tasklist_lock);
2167 }
2168
2169 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2170 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2171 #endif
2172
2173 static void sighand_ctor(void *data)
2174 {
2175         struct sighand_struct *sighand = data;
2176
2177         spin_lock_init(&sighand->siglock);
2178         init_waitqueue_head(&sighand->signalfd_wqh);
2179 }
2180
2181 void __init proc_caches_init(void)
2182 {
2183         sighand_cachep = kmem_cache_create("sighand_cache",
2184                         sizeof(struct sighand_struct), 0,
2185                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2186                         SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2187         signal_cachep = kmem_cache_create("signal_cache",
2188                         sizeof(struct signal_struct), 0,
2189                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2190                         NULL);
2191         files_cachep = kmem_cache_create("files_cache",
2192                         sizeof(struct files_struct), 0,
2193                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2194                         NULL);
2195         fs_cachep = kmem_cache_create("fs_cache",
2196                         sizeof(struct fs_struct), 0,
2197                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2198                         NULL);
2199         /*
2200          * FIXME! The "sizeof(struct mm_struct)" currently includes the
2201          * whole struct cpumask for the OFFSTACK case. We could change
2202          * this to *only* allocate as much of it as required by the
2203          * maximum number of CPU's we can ever have.  The cpumask_allocation
2204          * is at the end of the structure, exactly for that reason.
2205          */
2206         mm_cachep = kmem_cache_create("mm_struct",
2207                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2208                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2209                         NULL);
2210         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2211         mmap_init();
2212         nsproxy_cache_init();
2213 }
2214
2215 /*
2216  * Check constraints on flags passed to the unshare system call.
2217  */
2218 static int check_unshare_flags(unsigned long unshare_flags)
2219 {
2220         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2221                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2222                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2223                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2224                 return -EINVAL;
2225         /*
2226          * Not implemented, but pretend it works if there is nothing
2227          * to unshare.  Note that unsharing the address space or the
2228          * signal handlers also need to unshare the signal queues (aka
2229          * CLONE_THREAD).
2230          */
2231         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2232                 if (!thread_group_empty(current))
2233                         return -EINVAL;
2234         }
2235         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2236                 if (atomic_read(&current->sighand->count) > 1)
2237                         return -EINVAL;
2238         }
2239         if (unshare_flags & CLONE_VM) {
2240                 if (!current_is_single_threaded())
2241                         return -EINVAL;
2242         }
2243
2244         return 0;
2245 }
2246
2247 /*
2248  * Unshare the filesystem structure if it is being shared
2249  */
2250 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2251 {
2252         struct fs_struct *fs = current->fs;
2253
2254         if (!(unshare_flags & CLONE_FS) || !fs)
2255                 return 0;
2256
2257         /* don't need lock here; in the worst case we'll do useless copy */
2258         if (fs->users == 1)
2259                 return 0;
2260
2261         *new_fsp = copy_fs_struct(fs);
2262         if (!*new_fsp)
2263                 return -ENOMEM;
2264
2265         return 0;
2266 }
2267
2268 /*
2269  * Unshare file descriptor table if it is being shared
2270  */
2271 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2272 {
2273         struct files_struct *fd = current->files;
2274         int error = 0;
2275
2276         if ((unshare_flags & CLONE_FILES) &&
2277             (fd && atomic_read(&fd->count) > 1)) {
2278                 *new_fdp = dup_fd(fd, &error);
2279                 if (!*new_fdp)
2280                         return error;
2281         }
2282
2283         return 0;
2284 }
2285
2286 /*
2287  * unshare allows a process to 'unshare' part of the process
2288  * context which was originally shared using clone.  copy_*
2289  * functions used by do_fork() cannot be used here directly
2290  * because they modify an inactive task_struct that is being
2291  * constructed. Here we are modifying the current, active,
2292  * task_struct.
2293  */
2294 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2295 {
2296         struct fs_struct *fs, *new_fs = NULL;
2297         struct files_struct *fd, *new_fd = NULL;
2298         struct cred *new_cred = NULL;
2299         struct nsproxy *new_nsproxy = NULL;
2300         int do_sysvsem = 0;
2301         int err;
2302
2303         /*
2304          * If unsharing a user namespace must also unshare the thread group
2305          * and unshare the filesystem root and working directories.
2306          */
2307         if (unshare_flags & CLONE_NEWUSER)
2308                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2309         /*
2310          * If unsharing vm, must also unshare signal handlers.
2311          */
2312         if (unshare_flags & CLONE_VM)
2313                 unshare_flags |= CLONE_SIGHAND;
2314         /*
2315          * If unsharing a signal handlers, must also unshare the signal queues.
2316          */
2317         if (unshare_flags & CLONE_SIGHAND)
2318                 unshare_flags |= CLONE_THREAD;
2319         /*
2320          * If unsharing namespace, must also unshare filesystem information.
2321          */
2322         if (unshare_flags & CLONE_NEWNS)
2323                 unshare_flags |= CLONE_FS;
2324
2325         err = check_unshare_flags(unshare_flags);
2326         if (err)
2327                 goto bad_unshare_out;
2328         /*
2329          * CLONE_NEWIPC must also detach from the undolist: after switching
2330          * to a new ipc namespace, the semaphore arrays from the old
2331          * namespace are unreachable.
2332          */
2333         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2334                 do_sysvsem = 1;
2335         err = unshare_fs(unshare_flags, &new_fs);
2336         if (err)
2337                 goto bad_unshare_out;
2338         err = unshare_fd(unshare_flags, &new_fd);
2339         if (err)
2340                 goto bad_unshare_cleanup_fs;
2341         err = unshare_userns(unshare_flags, &new_cred);
2342         if (err)
2343                 goto bad_unshare_cleanup_fd;
2344         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2345                                          new_cred, new_fs);
2346         if (err)
2347                 goto bad_unshare_cleanup_cred;
2348
2349         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2350                 if (do_sysvsem) {
2351                         /*
2352                          * CLONE_SYSVSEM is equivalent to sys_exit().
2353                          */
2354                         exit_sem(current);
2355                 }
2356                 if (unshare_flags & CLONE_NEWIPC) {
2357                         /* Orphan segments in old ns (see sem above). */
2358                         exit_shm(current);
2359                         shm_init_task(current);
2360                 }
2361
2362                 if (new_nsproxy)
2363                         switch_task_namespaces(current, new_nsproxy);
2364
2365                 task_lock(current);
2366
2367                 if (new_fs) {
2368                         fs = current->fs;
2369                         spin_lock(&fs->lock);
2370                         current->fs = new_fs;
2371                         if (--fs->users)
2372                                 new_fs = NULL;
2373                         else
2374                                 new_fs = fs;
2375                         spin_unlock(&fs->lock);
2376                 }
2377
2378                 if (new_fd) {
2379                         fd = current->files;
2380                         current->files = new_fd;
2381                         new_fd = fd;
2382                 }
2383
2384                 task_unlock(current);
2385
2386                 if (new_cred) {
2387                         /* Install the new user namespace */
2388                         commit_creds(new_cred);
2389                         new_cred = NULL;
2390                 }
2391         }
2392
2393         perf_event_namespaces(current);
2394
2395 bad_unshare_cleanup_cred:
2396         if (new_cred)
2397                 put_cred(new_cred);
2398 bad_unshare_cleanup_fd:
2399         if (new_fd)
2400                 put_files_struct(new_fd);
2401
2402 bad_unshare_cleanup_fs:
2403         if (new_fs)
2404                 free_fs_struct(new_fs);
2405
2406 bad_unshare_out:
2407         return err;
2408 }
2409
2410 /*
2411  *      Helper to unshare the files of the current task.
2412  *      We don't want to expose copy_files internals to
2413  *      the exec layer of the kernel.
2414  */
2415
2416 int unshare_files(struct files_struct **displaced)
2417 {
2418         struct task_struct *task = current;
2419         struct files_struct *copy = NULL;
2420         int error;
2421
2422         error = unshare_fd(CLONE_FILES, &copy);
2423         if (error || !copy) {
2424                 *displaced = NULL;
2425                 return error;
2426         }
2427         *displaced = task->files;
2428         task_lock(task);
2429         task->files = copy;
2430         task_unlock(task);
2431         return 0;
2432 }
2433
2434 int sysctl_max_threads(struct ctl_table *table, int write,
2435                        void __user *buffer, size_t *lenp, loff_t *ppos)
2436 {
2437         struct ctl_table t;
2438         int ret;
2439         int threads = max_threads;
2440         int min = MIN_THREADS;
2441         int max = MAX_THREADS;
2442
2443         t = *table;
2444         t.data = &threads;
2445         t.extra1 = &min;
2446         t.extra2 = &max;
2447
2448         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2449         if (ret || !write)
2450                 return ret;
2451
2452         set_max_threads(threads);
2453
2454         return 0;
2455 }