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
  92 #include <asm/pgtable.h>
  93 #include <asm/pgalloc.h>
  94 #include <linux/uaccess.h>
  95 #include <asm/mmu_context.h>
  96 #include <asm/cacheflush.h>
  97 #include <asm/tlbflush.h>
  98
  99 #include <trace/events/sched.h>
 100
 101 #define CREATE_TRACE_POINTS
 102 #include <trace/events/task.h>
 103
 104 /*
 105  * Minimum number of threads to boot the kernel
 106  */
 107 #define MIN_THREADS 20
 108
 109 /*
 110  * Maximum number of threads
 111  */
 112 #define MAX_THREADS FUTEX_TID_MASK
 113
 114 /*
 115  * Protected counters by write_lock_irq(&tasklist_lock)
 116  */
 117 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 118 int nr_threads;                 /* The idle threads do not count.. */
 119
 120 int max_threads;                /* tunable limit on nr_threads */
 121
 122 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 123
 124 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 125
 126 #ifdef CONFIG_PROVE_RCU
 127 int lockdep_tasklist_lock_is_held(void)
 128 {
 129         return lockdep_is_held(&tasklist_lock);
 130 }
 131 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 132 #endif /* #ifdef CONFIG_PROVE_RCU */
 133
 134 int nr_processes(void)
 135 {
 136         int cpu;
 137         int total = 0;
 138
 139         for_each_possible_cpu(cpu)
 140                 total += per_cpu(process_counts, cpu);
 141
 142         return total;
 143 }
 144
 145 void __weak arch_release_task_struct(struct task_struct *tsk)
 146 {
 147 }
 148
 149 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 150 static struct kmem_cache *task_struct_cachep;
 151
 152 static inline struct task_struct *alloc_task_struct_node(int node)
 153 {
 154         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 155 }
 156
 157 static inline void free_task_struct(struct task_struct *tsk)
 158 {
 159         kmem_cache_free(task_struct_cachep, tsk);
 160 }
 161 #endif
 162
 163 void __weak arch_release_thread_stack(unsigned long *stack)
 164 {
 165 }
 166
 167 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 168
 169 /*
 170  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 171  * kmemcache based allocator.
 172  */
 173 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 174
 175 #ifdef CONFIG_VMAP_STACK
 176 /*
 177  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 178  * flush.  Try to minimize the number of calls by caching stacks.
 179  */
 180 #define NR_CACHED_STACKS 2
 181 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 182
 183 static int free_vm_stack_cache(unsigned int cpu)
 184 {
 185         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 186         int i;
 187
 188         for (i = 0; i < NR_CACHED_STACKS; i++) {
 189                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 190
 191                 if (!vm_stack)
 192                         continue;
 193
 194                 vfree(vm_stack->addr);
 195                 cached_vm_stacks[i] = NULL;
 196         }
 197
 198         return 0;
 199 }
 200 #endif
 201
 202 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 203 {
 204 #ifdef CONFIG_VMAP_STACK
 205         void *stack;
 206         int i;
 207
 208         local_irq_disable();
 209         for (i = 0; i < NR_CACHED_STACKS; i++) {
 210                 struct vm_struct *s = this_cpu_read(cached_stacks[i]);
 211
 212                 if (!s)
 213                         continue;
 214                 this_cpu_write(cached_stacks[i], NULL);
 215
 216                 tsk->stack_vm_area = s;
 217                 local_irq_enable();
 218                 return s->addr;
 219         }
 220         local_irq_enable();
 221
 222         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_SIZE,
 223                                      VMALLOC_START, VMALLOC_END,
 224                                      THREADINFO_GFP,
 225                                      PAGE_KERNEL,
 226                                      0, node, __builtin_return_address(0));
 227
 228         /*
 229          * We can't call find_vm_area() in interrupt context, and
 230          * free_thread_stack() can be called in interrupt context,
 231          * so cache the vm_struct.
 232          */
 233         if (stack)
 234                 tsk->stack_vm_area = find_vm_area(stack);
 235         return stack;
 236 #else
 237         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 238                                              THREAD_SIZE_ORDER);
 239
 240         return page ? page_address(page) : NULL;
 241 #endif
 242 }
 243
 244 static inline void free_thread_stack(struct task_struct *tsk)
 245 {
 246 #ifdef CONFIG_VMAP_STACK
 247         if (task_stack_vm_area(tsk)) {
 248                 unsigned long flags;
 249                 int i;
 250
 251                 local_irq_save(flags);
 252                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 253                         if (this_cpu_read(cached_stacks[i]))
 254                                 continue;
 255
 256                         this_cpu_write(cached_stacks[i], tsk->stack_vm_area);
 257                         local_irq_restore(flags);
 258                         return;
 259                 }
 260                 local_irq_restore(flags);
 261
 262                 vfree_atomic(tsk->stack);
 263                 return;
 264         }
 265 #endif
 266
 267         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 268 }
 269 # else
 270 static struct kmem_cache *thread_stack_cache;
 271
 272 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 273                                                   int node)
 274 {
 275         return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 276 }
 277
 278 static void free_thread_stack(struct task_struct *tsk)
 279 {
 280         kmem_cache_free(thread_stack_cache, tsk->stack);
 281 }
 282
 283 void thread_stack_cache_init(void)
 284 {
 285         thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
 286                                               THREAD_SIZE, 0, NULL);
 287         BUG_ON(thread_stack_cache == NULL);
 288 }
 289 # endif
 290 #endif
 291
 292 /* SLAB cache for signal_struct structures (tsk->signal) */
 293 static struct kmem_cache *signal_cachep;
 294
 295 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 296 struct kmem_cache *sighand_cachep;
 297
 298 /* SLAB cache for files_struct structures (tsk->files) */
 299 struct kmem_cache *files_cachep;
 300
 301 /* SLAB cache for fs_struct structures (tsk->fs) */
 302 struct kmem_cache *fs_cachep;
 303
 304 /* SLAB cache for vm_area_struct structures */
 305 struct kmem_cache *vm_area_cachep;
 306
 307 /* SLAB cache for mm_struct structures (tsk->mm) */
 308 static struct kmem_cache *mm_cachep;
 309
 310 static void account_kernel_stack(struct task_struct *tsk, int account)
 311 {
 312         void *stack = task_stack_page(tsk);
 313         struct vm_struct *vm = task_stack_vm_area(tsk);
 314
 315         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 316
 317         if (vm) {
 318                 int i;
 319
 320                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 321
 322                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 323                         mod_zone_page_state(page_zone(vm->pages[i]),
 324                                             NR_KERNEL_STACK_KB,
 325                                             PAGE_SIZE / 1024 * account);
 326                 }
 327
 328                 /* All stack pages belong to the same memcg. */
 329                 mod_memcg_page_state(vm->pages[0], MEMCG_KERNEL_STACK_KB,
 330                                      account * (THREAD_SIZE / 1024));
 331         } else {
 332                 /*
 333                  * All stack pages are in the same zone and belong to the
 334                  * same memcg.
 335                  */
 336                 struct page *first_page = virt_to_page(stack);
 337
 338                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 339                                     THREAD_SIZE / 1024 * account);
 340
 341                 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
 342                                      account * (THREAD_SIZE / 1024));
 343         }
 344 }
 345
 346 static void release_task_stack(struct task_struct *tsk)
 347 {
 348         if (WARN_ON(tsk->state != TASK_DEAD))
 349                 return;  /* Better to leak the stack than to free prematurely */
 350
 351         account_kernel_stack(tsk, -1);
 352         arch_release_thread_stack(tsk->stack);
 353         free_thread_stack(tsk);
 354         tsk->stack = NULL;
 355 #ifdef CONFIG_VMAP_STACK
 356         tsk->stack_vm_area = NULL;
 357 #endif
 358 }
 359
 360 #ifdef CONFIG_THREAD_INFO_IN_TASK
 361 void put_task_stack(struct task_struct *tsk)
 362 {
 363         if (atomic_dec_and_test(&tsk->stack_refcount))
 364                 release_task_stack(tsk);
 365 }
 366 #endif
 367
 368 void free_task(struct task_struct *tsk)
 369 {
 370 #ifndef CONFIG_THREAD_INFO_IN_TASK
 371         /*
 372          * The task is finally done with both the stack and thread_info,
 373          * so free both.
 374          */
 375         release_task_stack(tsk);
 376 #else
 377         /*
 378          * If the task had a separate stack allocation, it should be gone
 379          * by now.
 380          */
 381         WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
 382 #endif
 383         rt_mutex_debug_task_free(tsk);
 384         ftrace_graph_exit_task(tsk);
 385         put_seccomp_filter(tsk);
 386         arch_release_task_struct(tsk);
 387         if (tsk->flags & PF_KTHREAD)
 388                 free_kthread_struct(tsk);
 389         free_task_struct(tsk);
 390 }
 391 EXPORT_SYMBOL(free_task);
 392
 393 static inline void free_signal_struct(struct signal_struct *sig)
 394 {
 395         taskstats_tgid_free(sig);
 396         sched_autogroup_exit(sig);
 397         /*
 398          * __mmdrop is not safe to call from softirq context on x86 due to
 399          * pgd_dtor so postpone it to the async context
 400          */
 401         if (sig->oom_mm)
 402                 mmdrop_async(sig->oom_mm);
 403         kmem_cache_free(signal_cachep, sig);
 404 }
 405
 406 static inline void put_signal_struct(struct signal_struct *sig)
 407 {
 408         if (atomic_dec_and_test(&sig->sigcnt))
 409                 free_signal_struct(sig);
 410 }
 411
 412 void __put_task_struct(struct task_struct *tsk)
 413 {
 414         WARN_ON(!tsk->exit_state);
 415         WARN_ON(atomic_read(&tsk->usage));
 416         WARN_ON(tsk == current);
 417
 418         cgroup_free(tsk);
 419         task_numa_free(tsk);
 420         security_task_free(tsk);
 421         exit_creds(tsk);
 422         delayacct_tsk_free(tsk);
 423         put_signal_struct(tsk->signal);
 424
 425         if (!profile_handoff_task(tsk))
 426                 free_task(tsk);
 427 }
 428 EXPORT_SYMBOL_GPL(__put_task_struct);
 429
 430 void __init __weak arch_task_cache_init(void) { }
 431
 432 /*
 433  * set_max_threads
 434  */
 435 static void set_max_threads(unsigned int max_threads_suggested)
 436 {
 437         u64 threads;
 438
 439         /*
 440          * The number of threads shall be limited such that the thread
 441          * structures may only consume a small part of the available memory.
 442          */
 443         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 444                 threads = MAX_THREADS;
 445         else
 446                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 447                                     (u64) THREAD_SIZE * 8UL);
 448
 449         if (threads > max_threads_suggested)
 450                 threads = max_threads_suggested;
 451
 452         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 453 }
 454
 455 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 456 /* Initialized by the architecture: */
 457 int arch_task_struct_size __read_mostly;
 458 #endif
 459
 460 void __init fork_init(void)
 461 {
 462         int i;
 463 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 464 #ifndef ARCH_MIN_TASKALIGN
 465 #define ARCH_MIN_TASKALIGN      0
 466 #endif
 467         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 468
 469         /* create a slab on which task_structs can be allocated */
 470         task_struct_cachep = kmem_cache_create("task_struct",
 471                         arch_task_struct_size, align,
 472                         SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
 473 #endif
 474
 475         /* do the arch specific task caches init */
 476         arch_task_cache_init();
 477
 478         set_max_threads(MAX_THREADS);
 479
 480         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 481         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 482         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 483                 init_task.signal->rlim[RLIMIT_NPROC];
 484
 485         for (i = 0; i < UCOUNT_COUNTS; i++) {
 486                 init_user_ns.ucount_max[i] = max_threads/2;
 487         }
 488
 489 #ifdef CONFIG_VMAP_STACK
 490         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 491                           NULL, free_vm_stack_cache);
 492 #endif
 493 }
 494
 495 int __weak arch_dup_task_struct(struct task_struct *dst,
 496                                                struct task_struct *src)
 497 {
 498         *dst = *src;
 499         return 0;
 500 }
 501
 502 void set_task_stack_end_magic(struct task_struct *tsk)
 503 {
 504         unsigned long *stackend;
 505
 506         stackend = end_of_stack(tsk);
 507         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 508 }
 509
 510 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 511 {
 512         struct task_struct *tsk;
 513         unsigned long *stack;
 514         struct vm_struct *stack_vm_area;
 515         int err;
 516
 517         if (node == NUMA_NO_NODE)
 518                 node = tsk_fork_get_node(orig);
 519         tsk = alloc_task_struct_node(node);
 520         if (!tsk)
 521                 return NULL;
 522
 523         stack = alloc_thread_stack_node(tsk, node);
 524         if (!stack)
 525                 goto free_tsk;
 526
 527         stack_vm_area = task_stack_vm_area(tsk);
 528
 529         err = arch_dup_task_struct(tsk, orig);
 530
 531         /*
 532          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 533          * sure they're properly initialized before using any stack-related
 534          * functions again.
 535          */
 536         tsk->stack = stack;
 537 #ifdef CONFIG_VMAP_STACK
 538         tsk->stack_vm_area = stack_vm_area;
 539 #endif
 540 #ifdef CONFIG_THREAD_INFO_IN_TASK
 541         atomic_set(&tsk->stack_refcount, 1);
 542 #endif
 543
 544         if (err)
 545                 goto free_stack;
 546
 547 #ifdef CONFIG_SECCOMP
 548         /*
 549          * We must handle setting up seccomp filters once we're under
 550          * the sighand lock in case orig has changed between now and
 551          * then. Until then, filter must be NULL to avoid messing up
 552          * the usage counts on the error path calling free_task.
 553          */
 554         tsk->seccomp.filter = NULL;
 555 #endif
 556
 557         setup_thread_stack(tsk, orig);
 558         clear_user_return_notifier(tsk);
 559         clear_tsk_need_resched(tsk);
 560         set_task_stack_end_magic(tsk);
 561
 562 #ifdef CONFIG_CC_STACKPROTECTOR
 563         tsk->stack_canary = get_random_long();
 564 #endif
 565
 566         /*
 567          * One for us, one for whoever does the "release_task()" (usually
 568          * parent)
 569          */
 570         atomic_set(&tsk->usage, 2);
 571 #ifdef CONFIG_BLK_DEV_IO_TRACE
 572         tsk->btrace_seq = 0;
 573 #endif
 574         tsk->splice_pipe = NULL;
 575         tsk->task_frag.page = NULL;
 576         tsk->wake_q.next = NULL;
 577
 578         account_kernel_stack(tsk, 1);
 579
 580         kcov_task_init(tsk);
 581
 582         return tsk;
 583
 584 free_stack:
 585         free_thread_stack(tsk);
 586 free_tsk:
 587         free_task_struct(tsk);
 588         return NULL;
 589 }
 590
 591 #ifdef CONFIG_MMU
 592 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 593                                         struct mm_struct *oldmm)
 594 {
 595         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 596         struct rb_node **rb_link, *rb_parent;
 597         int retval;
 598         unsigned long charge;
 599         LIST_HEAD(uf);
 600
 601         uprobe_start_dup_mmap();
 602         if (down_write_killable(&oldmm->mmap_sem)) {
 603                 retval = -EINTR;
 604                 goto fail_uprobe_end;
 605         }
 606         flush_cache_dup_mm(oldmm);
 607         uprobe_dup_mmap(oldmm, mm);
 608         /*
 609          * Not linked in yet - no deadlock potential:
 610          */
 611         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 612
 613         /* No ordering required: file already has been exposed. */
 614         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 615
 616         mm->total_vm = oldmm->total_vm;
 617         mm->data_vm = oldmm->data_vm;
 618         mm->exec_vm = oldmm->exec_vm;
 619         mm->stack_vm = oldmm->stack_vm;
 620
 621         rb_link = &mm->mm_rb.rb_node;
 622         rb_parent = NULL;
 623         pprev = &mm->mmap;
 624         retval = ksm_fork(mm, oldmm);
 625         if (retval)
 626                 goto out;
 627         retval = khugepaged_fork(mm, oldmm);
 628         if (retval)
 629                 goto out;
 630
 631         prev = NULL;
 632         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 633                 struct file *file;
 634
 635                 if (mpnt->vm_flags & VM_DONTCOPY) {
 636                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 637                         continue;
 638                 }
 639                 charge = 0;
 640                 if (mpnt->vm_flags & VM_ACCOUNT) {
 641                         unsigned long len = vma_pages(mpnt);
 642
 643                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 644                                 goto fail_nomem;
 645                         charge = len;
 646                 }
 647                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 648                 if (!tmp)
 649                         goto fail_nomem;
 650                 *tmp = *mpnt;
 651                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 652                 retval = vma_dup_policy(mpnt, tmp);
 653                 if (retval)
 654                         goto fail_nomem_policy;
 655                 tmp->vm_mm = mm;
 656                 retval = dup_userfaultfd(tmp, &uf);
 657                 if (retval)
 658                         goto fail_nomem_anon_vma_fork;
 659                 if (anon_vma_fork(tmp, mpnt))
 660                         goto fail_nomem_anon_vma_fork;
 661                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 662                 tmp->vm_next = tmp->vm_prev = NULL;
 663                 file = tmp->vm_file;
 664                 if (file) {
 665                         struct inode *inode = file_inode(file);
 666                         struct address_space *mapping = file->f_mapping;
 667
 668                         get_file(file);
 669                         if (tmp->vm_flags & VM_DENYWRITE)
 670                                 atomic_dec(&inode->i_writecount);
 671                         i_mmap_lock_write(mapping);
 672                         if (tmp->vm_flags & VM_SHARED)
 673                                 atomic_inc(&mapping->i_mmap_writable);
 674                         flush_dcache_mmap_lock(mapping);
 675                         /* insert tmp into the share list, just after mpnt */
 676                         vma_interval_tree_insert_after(tmp, mpnt,
 677                                         &mapping->i_mmap);
 678                         flush_dcache_mmap_unlock(mapping);
 679                         i_mmap_unlock_write(mapping);
 680                 }
 681
 682                 /*
 683                  * Clear hugetlb-related page reserves for children. This only
 684                  * affects MAP_PRIVATE mappings. Faults generated by the child
 685                  * are not guaranteed to succeed, even if read-only
 686                  */
 687                 if (is_vm_hugetlb_page(tmp))
 688                         reset_vma_resv_huge_pages(tmp);
 689
 690                 /*
 691                  * Link in the new vma and copy the page table entries.
 692                  */
 693                 *pprev = tmp;
 694                 pprev = &tmp->vm_next;
 695                 tmp->vm_prev = prev;
 696                 prev = tmp;
 697
 698                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 699                 rb_link = &tmp->vm_rb.rb_right;
 700                 rb_parent = &tmp->vm_rb;
 701
 702                 mm->map_count++;
 703                 retval = copy_page_range(mm, oldmm, mpnt);
 704
 705                 if (tmp->vm_ops && tmp->vm_ops->open)
 706                         tmp->vm_ops->open(tmp);
 707
 708                 if (retval)
 709                         goto out;
 710         }
 711         /* a new mm has just been created */
 712         arch_dup_mmap(oldmm, mm);
 713         retval = 0;
 714 out:
 715         up_write(&mm->mmap_sem);
 716         flush_tlb_mm(oldmm);
 717         up_write(&oldmm->mmap_sem);
 718         dup_userfaultfd_complete(&uf);
 719 fail_uprobe_end:
 720         uprobe_end_dup_mmap();
 721         return retval;
 722 fail_nomem_anon_vma_fork:
 723         mpol_put(vma_policy(tmp));
 724 fail_nomem_policy:
 725         kmem_cache_free(vm_area_cachep, tmp);
 726 fail_nomem:
 727         retval = -ENOMEM;
 728         vm_unacct_memory(charge);
 729         goto out;
 730 }
 731
 732 static inline int mm_alloc_pgd(struct mm_struct *mm)
 733 {
 734         mm->pgd = pgd_alloc(mm);
 735         if (unlikely(!mm->pgd))
 736                 return -ENOMEM;
 737         return 0;
 738 }
 739
 740 static inline void mm_free_pgd(struct mm_struct *mm)
 741 {
 742         pgd_free(mm, mm->pgd);
 743 }
 744 #else
 745 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 746 {
 747         down_write(&oldmm->mmap_sem);
 748         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 749         up_write(&oldmm->mmap_sem);
 750         return 0;
 751 }
 752 #define mm_alloc_pgd(mm)        (0)
 753 #define mm_free_pgd(mm)
 754 #endif /* CONFIG_MMU */
 755
 756 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 757
 758 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 759 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 760
 761 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 762
 763 static int __init coredump_filter_setup(char *s)
 764 {
 765         default_dump_filter =
 766                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 767                 MMF_DUMP_FILTER_MASK;
 768         return 1;
 769 }
 770
 771 __setup("coredump_filter=", coredump_filter_setup);
 772
 773 #include <linux/init_task.h>
 774
 775 static void mm_init_aio(struct mm_struct *mm)
 776 {
 777 #ifdef CONFIG_AIO
 778         spin_lock_init(&mm->ioctx_lock);
 779         mm->ioctx_table = NULL;
 780 #endif
 781 }
 782
 783 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 784 {
 785 #ifdef CONFIG_MEMCG
 786         mm->owner = p;
 787 #endif
 788 }
 789
 790 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 791         struct user_namespace *user_ns)
 792 {
 793         mm->mmap = NULL;
 794         mm->mm_rb = RB_ROOT;
 795         mm->vmacache_seqnum = 0;
 796         atomic_set(&mm->mm_users, 1);
 797         atomic_set(&mm->mm_count, 1);
 798         init_rwsem(&mm->mmap_sem);
 799         INIT_LIST_HEAD(&mm->mmlist);
 800         mm->core_state = NULL;
 801         atomic_long_set(&mm->nr_ptes, 0);
 802         mm_nr_pmds_init(mm);
 803         mm->map_count = 0;
 804         mm->locked_vm = 0;
 805         mm->pinned_vm = 0;
 806         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 807         spin_lock_init(&mm->page_table_lock);
 808         mm_init_cpumask(mm);
 809         mm_init_aio(mm);
 810         mm_init_owner(mm, p);
 811         mmu_notifier_mm_init(mm);
 812         clear_tlb_flush_pending(mm);
 813 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 814         mm->pmd_huge_pte = NULL;
 815 #endif
 816
 817         if (current->mm) {
 818                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 819                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 820         } else {
 821                 mm->flags = default_dump_filter;
 822                 mm->def_flags = 0;
 823         }
 824
 825         if (mm_alloc_pgd(mm))
 826                 goto fail_nopgd;
 827
 828         if (init_new_context(p, mm))
 829                 goto fail_nocontext;
 830
 831         mm->user_ns = get_user_ns(user_ns);
 832         return mm;
 833
 834 fail_nocontext:
 835         mm_free_pgd(mm);
 836 fail_nopgd:
 837         free_mm(mm);
 838         return NULL;
 839 }
 840
 841 static void check_mm(struct mm_struct *mm)
 842 {
 843         int i;
 844
 845         for (i = 0; i < NR_MM_COUNTERS; i++) {
 846                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 847
 848                 if (unlikely(x))
 849                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 850                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 851         }
 852
 853         if (atomic_long_read(&mm->nr_ptes))
 854                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 855                                 atomic_long_read(&mm->nr_ptes));
 856         if (mm_nr_pmds(mm))
 857                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 858                                 mm_nr_pmds(mm));
 859
 860 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 861         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 862 #endif
 863 }
 864
 865 /*
 866  * Allocate and initialize an mm_struct.
 867  */
 868 struct mm_struct *mm_alloc(void)
 869 {
 870         struct mm_struct *mm;
 871
 872         mm = allocate_mm();
 873         if (!mm)
 874                 return NULL;
 875
 876         memset(mm, 0, sizeof(*mm));
 877         return mm_init(mm, current, current_user_ns());
 878 }
 879
 880 /*
 881  * Called when the last reference to the mm
 882  * is dropped: either by a lazy thread or by
 883  * mmput. Free the page directory and the mm.
 884  */
 885 void __mmdrop(struct mm_struct *mm)
 886 {
 887         BUG_ON(mm == &init_mm);
 888         mm_free_pgd(mm);
 889         destroy_context(mm);
 890         mmu_notifier_mm_destroy(mm);
 891         check_mm(mm);
 892         put_user_ns(mm->user_ns);
 893         free_mm(mm);
 894 }
 895 EXPORT_SYMBOL_GPL(__mmdrop);
 896
 897 static inline void __mmput(struct mm_struct *mm)
 898 {
 899         VM_BUG_ON(atomic_read(&mm->mm_users));
 900
 901         uprobe_clear_state(mm);
 902         exit_aio(mm);
 903         ksm_exit(mm);
 904         khugepaged_exit(mm); /* must run before exit_mmap */
 905         exit_mmap(mm);
 906         mm_put_huge_zero_page(mm);
 907         set_mm_exe_file(mm, NULL);
 908         if (!list_empty(&mm->mmlist)) {
 909                 spin_lock(&mmlist_lock);
 910                 list_del(&mm->mmlist);
 911                 spin_unlock(&mmlist_lock);
 912         }
 913         if (mm->binfmt)
 914                 module_put(mm->binfmt->module);
 915         set_bit(MMF_OOM_SKIP, &mm->flags);
 916         mmdrop(mm);
 917 }
 918
 919 /*
 920  * Decrement the use count and release all resources for an mm.
 921  */
 922 void mmput(struct mm_struct *mm)
 923 {
 924         might_sleep();
 925
 926         if (atomic_dec_and_test(&mm->mm_users))
 927                 __mmput(mm);
 928 }
 929 EXPORT_SYMBOL_GPL(mmput);
 930
 931 #ifdef CONFIG_MMU
 932 static void mmput_async_fn(struct work_struct *work)
 933 {
 934         struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
 935         __mmput(mm);
 936 }
 937
 938 void mmput_async(struct mm_struct *mm)
 939 {
 940         if (atomic_dec_and_test(&mm->mm_users)) {
 941                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
 942                 schedule_work(&mm->async_put_work);
 943         }
 944 }
 945 #endif
 946
 947 /**
 948  * set_mm_exe_file - change a reference to the mm's executable file
 949  *
 950  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 951  *
 952  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 953  * invocations: in mmput() nobody alive left, in execve task is single
 954  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 955  * mm->exe_file, but does so without using set_mm_exe_file() in order
 956  * to do avoid the need for any locks.
 957  */
 958 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 959 {
 960         struct file *old_exe_file;
 961
 962         /*
 963          * It is safe to dereference the exe_file without RCU as
 964          * this function is only called if nobody else can access
 965          * this mm -- see comment above for justification.
 966          */
 967         old_exe_file = rcu_dereference_raw(mm->exe_file);
 968
 969         if (new_exe_file)
 970                 get_file(new_exe_file);
 971         rcu_assign_pointer(mm->exe_file, new_exe_file);
 972         if (old_exe_file)
 973                 fput(old_exe_file);
 974 }
 975
 976 /**
 977  * get_mm_exe_file - acquire a reference to the mm's executable file
 978  *
 979  * Returns %NULL if mm has no associated executable file.
 980  * User must release file via fput().
 981  */
 982 struct file *get_mm_exe_file(struct mm_struct *mm)
 983 {
 984         struct file *exe_file;
 985
 986         rcu_read_lock();
 987         exe_file = rcu_dereference(mm->exe_file);
 988         if (exe_file && !get_file_rcu(exe_file))
 989                 exe_file = NULL;
 990         rcu_read_unlock();
 991         return exe_file;
 992 }
 993 EXPORT_SYMBOL(get_mm_exe_file);
 994
 995 /**
 996  * get_task_exe_file - acquire a reference to the task's executable file
 997  *
 998  * Returns %NULL if task's mm (if any) has no associated executable file or
 999  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1000  * User must release file via fput().
1001  */
1002 struct file *get_task_exe_file(struct task_struct *task)
1003 {
1004         struct file *exe_file = NULL;
1005         struct mm_struct *mm;
1006
1007         task_lock(task);
1008         mm = task->mm;
1009         if (mm) {
1010                 if (!(task->flags & PF_KTHREAD))
1011                         exe_file = get_mm_exe_file(mm);
1012         }
1013         task_unlock(task);
1014         return exe_file;
1015 }
1016 EXPORT_SYMBOL(get_task_exe_file);
1017
1018 /**
1019  * get_task_mm - acquire a reference to the task's mm
1020  *
1021  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1022  * this kernel workthread has transiently adopted a user mm with use_mm,
1023  * to do its AIO) is not set and if so returns a reference to it, after
1024  * bumping up the use count.  User must release the mm via mmput()
1025  * after use.  Typically used by /proc and ptrace.
1026  */
1027 struct mm_struct *get_task_mm(struct task_struct *task)
1028 {
1029         struct mm_struct *mm;
1030
1031         task_lock(task);
1032         mm = task->mm;
1033         if (mm) {
1034                 if (task->flags & PF_KTHREAD)
1035                         mm = NULL;
1036                 else
1037                         mmget(mm);
1038         }
1039         task_unlock(task);
1040         return mm;
1041 }
1042 EXPORT_SYMBOL_GPL(get_task_mm);
1043
1044 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1045 {
1046         struct mm_struct *mm;
1047         int err;
1048
1049         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1050         if (err)
1051                 return ERR_PTR(err);
1052
1053         mm = get_task_mm(task);
1054         if (mm && mm != current->mm &&
1055                         !ptrace_may_access(task, mode)) {
1056                 mmput(mm);
1057                 mm = ERR_PTR(-EACCES);
1058         }
1059         mutex_unlock(&task->signal->cred_guard_mutex);
1060
1061         return mm;
1062 }
1063
1064 static void complete_vfork_done(struct task_struct *tsk)
1065 {
1066         struct completion *vfork;
1067
1068         task_lock(tsk);
1069         vfork = tsk->vfork_done;
1070         if (likely(vfork)) {
1071                 tsk->vfork_done = NULL;
1072                 complete(vfork);
1073         }
1074         task_unlock(tsk);
1075 }
1076
1077 static int wait_for_vfork_done(struct task_struct *child,
1078                                 struct completion *vfork)
1079 {
1080         int killed;
1081
1082         freezer_do_not_count();
1083         killed = wait_for_completion_killable(vfork);
1084         freezer_count();
1085
1086         if (killed) {
1087                 task_lock(child);
1088                 child->vfork_done = NULL;
1089                 task_unlock(child);
1090         }
1091
1092         put_task_struct(child);
1093         return killed;
1094 }
1095
1096 /* Please note the differences between mmput and mm_release.
1097  * mmput is called whenever we stop holding onto a mm_struct,
1098  * error success whatever.
1099  *
1100  * mm_release is called after a mm_struct has been removed
1101  * from the current process.
1102  *
1103  * This difference is important for error handling, when we
1104  * only half set up a mm_struct for a new process and need to restore
1105  * the old one.  Because we mmput the new mm_struct before
1106  * restoring the old one. . .
1107  * Eric Biederman 10 January 1998
1108  */
1109 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1110 {
1111         /* Get rid of any futexes when releasing the mm */
1112 #ifdef CONFIG_FUTEX
1113         if (unlikely(tsk->robust_list)) {
1114                 exit_robust_list(tsk);
1115                 tsk->robust_list = NULL;
1116         }
1117 #ifdef CONFIG_COMPAT
1118         if (unlikely(tsk->compat_robust_list)) {
1119                 compat_exit_robust_list(tsk);
1120                 tsk->compat_robust_list = NULL;
1121         }
1122 #endif
1123         if (unlikely(!list_empty(&tsk->pi_state_list)))
1124                 exit_pi_state_list(tsk);
1125 #endif
1126
1127         uprobe_free_utask(tsk);
1128
1129         /* Get rid of any cached register state */
1130         deactivate_mm(tsk, mm);
1131
1132         /*
1133          * Signal userspace if we're not exiting with a core dump
1134          * because we want to leave the value intact for debugging
1135          * purposes.
1136          */
1137         if (tsk->clear_child_tid) {
1138                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1139                     atomic_read(&mm->mm_users) > 1) {
1140                         /*
1141                          * We don't check the error code - if userspace has
1142                          * not set up a proper pointer then tough luck.
1143                          */
1144                         put_user(0, tsk->clear_child_tid);
1145                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1146                                         1, NULL, NULL, 0);
1147                 }
1148                 tsk->clear_child_tid = NULL;
1149         }
1150
1151         /*
1152          * All done, finally we can wake up parent and return this mm to him.
1153          * Also kthread_stop() uses this completion for synchronization.
1154          */
1155         if (tsk->vfork_done)
1156                 complete_vfork_done(tsk);
1157 }
1158
1159 /*
1160  * Allocate a new mm structure and copy contents from the
1161  * mm structure of the passed in task structure.
1162  */
1163 static struct mm_struct *dup_mm(struct task_struct *tsk)
1164 {
1165         struct mm_struct *mm, *oldmm = current->mm;
1166         int err;
1167
1168         mm = allocate_mm();
1169         if (!mm)
1170                 goto fail_nomem;
1171
1172         memcpy(mm, oldmm, sizeof(*mm));
1173
1174         if (!mm_init(mm, tsk, mm->user_ns))
1175                 goto fail_nomem;
1176
1177         err = dup_mmap(mm, oldmm);
1178         if (err)
1179                 goto free_pt;
1180
1181         mm->hiwater_rss = get_mm_rss(mm);
1182         mm->hiwater_vm = mm->total_vm;
1183
1184         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1185                 goto free_pt;
1186
1187         return mm;
1188
1189 free_pt:
1190         /* don't put binfmt in mmput, we haven't got module yet */
1191         mm->binfmt = NULL;
1192         mmput(mm);
1193
1194 fail_nomem:
1195         return NULL;
1196 }
1197
1198 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1199 {
1200         struct mm_struct *mm, *oldmm;
1201         int retval;
1202
1203         tsk->min_flt = tsk->maj_flt = 0;
1204         tsk->nvcsw = tsk->nivcsw = 0;
1205 #ifdef CONFIG_DETECT_HUNG_TASK
1206         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1207 #endif
1208
1209         tsk->mm = NULL;
1210         tsk->active_mm = NULL;
1211
1212         /*
1213          * Are we cloning a kernel thread?
1214          *
1215          * We need to steal a active VM for that..
1216          */
1217         oldmm = current->mm;
1218         if (!oldmm)
1219                 return 0;
1220
1221         /* initialize the new vmacache entries */
1222         vmacache_flush(tsk);
1223
1224         if (clone_flags & CLONE_VM) {
1225                 mmget(oldmm);
1226                 mm = oldmm;
1227                 goto good_mm;
1228         }
1229
1230         retval = -ENOMEM;
1231         mm = dup_mm(tsk);
1232         if (!mm)
1233                 goto fail_nomem;
1234
1235 good_mm:
1236         tsk->mm = mm;
1237         tsk->active_mm = mm;
1238         return 0;
1239
1240 fail_nomem:
1241         return retval;
1242 }
1243
1244 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1245 {
1246         struct fs_struct *fs = current->fs;
1247         if (clone_flags & CLONE_FS) {
1248                 /* tsk->fs is already what we want */
1249                 spin_lock(&fs->lock);
1250                 if (fs->in_exec) {
1251                         spin_unlock(&fs->lock);
1252                         return -EAGAIN;
1253                 }
1254                 fs->users++;
1255                 spin_unlock(&fs->lock);
1256                 return 0;
1257         }
1258         tsk->fs = copy_fs_struct(fs);
1259         if (!tsk->fs)
1260                 return -ENOMEM;
1261         return 0;
1262 }
1263
1264 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1265 {
1266         struct files_struct *oldf, *newf;
1267         int error = 0;
1268
1269         /*
1270          * A background process may not have any files ...
1271          */
1272         oldf = current->files;
1273         if (!oldf)
1274                 goto out;
1275
1276         if (clone_flags & CLONE_FILES) {
1277                 atomic_inc(&oldf->count);
1278                 goto out;
1279         }
1280
1281         newf = dup_fd(oldf, &error);
1282         if (!newf)
1283                 goto out;
1284
1285         tsk->files = newf;
1286         error = 0;
1287 out:
1288         return error;
1289 }
1290
1291 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1292 {
1293 #ifdef CONFIG_BLOCK
1294         struct io_context *ioc = current->io_context;
1295         struct io_context *new_ioc;
1296
1297         if (!ioc)
1298                 return 0;
1299         /*
1300          * Share io context with parent, if CLONE_IO is set
1301          */
1302         if (clone_flags & CLONE_IO) {
1303                 ioc_task_link(ioc);
1304                 tsk->io_context = ioc;
1305         } else if (ioprio_valid(ioc->ioprio)) {
1306                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1307                 if (unlikely(!new_ioc))
1308                         return -ENOMEM;
1309
1310                 new_ioc->ioprio = ioc->ioprio;
1311                 put_io_context(new_ioc);
1312         }
1313 #endif
1314         return 0;
1315 }
1316
1317 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1318 {
1319         struct sighand_struct *sig;
1320
1321         if (clone_flags & CLONE_SIGHAND) {
1322                 atomic_inc(&current->sighand->count);
1323                 return 0;
1324         }
1325         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1326         rcu_assign_pointer(tsk->sighand, sig);
1327         if (!sig)
1328                 return -ENOMEM;
1329
1330         atomic_set(&sig->count, 1);
1331         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1332         return 0;
1333 }
1334
1335 void __cleanup_sighand(struct sighand_struct *sighand)
1336 {
1337         if (atomic_dec_and_test(&sighand->count)) {
1338                 signalfd_cleanup(sighand);
1339                 /*
1340                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1341                  * without an RCU grace period, see __lock_task_sighand().
1342                  */
1343                 kmem_cache_free(sighand_cachep, sighand);
1344         }
1345 }
1346
1347 #ifdef CONFIG_POSIX_TIMERS
1348 /*
1349  * Initialize POSIX timer handling for a thread group.
1350  */
1351 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1352 {
1353         unsigned long cpu_limit;
1354
1355         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1356         if (cpu_limit != RLIM_INFINITY) {
1357                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1358                 sig->cputimer.running = true;
1359         }
1360
1361         /* The timer lists. */
1362         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1363         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1364         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1365 }
1366 #else
1367 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1368 #endif
1369
1370 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1371 {
1372         struct signal_struct *sig;
1373
1374         if (clone_flags & CLONE_THREAD)
1375                 return 0;
1376
1377         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1378         tsk->signal = sig;
1379         if (!sig)
1380                 return -ENOMEM;
1381
1382         sig->nr_threads = 1;
1383         atomic_set(&sig->live, 1);
1384         atomic_set(&sig->sigcnt, 1);
1385
1386         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1387         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1388         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1389
1390         init_waitqueue_head(&sig->wait_chldexit);
1391         sig->curr_target = tsk;
1392         init_sigpending(&sig->shared_pending);
1393         seqlock_init(&sig->stats_lock);
1394         prev_cputime_init(&sig->prev_cputime);
1395
1396 #ifdef CONFIG_POSIX_TIMERS
1397         INIT_LIST_HEAD(&sig->posix_timers);
1398         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1399         sig->real_timer.function = it_real_fn;
1400 #endif
1401
1402         task_lock(current->group_leader);
1403         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1404         task_unlock(current->group_leader);
1405
1406         posix_cpu_timers_init_group(sig);
1407
1408         tty_audit_fork(sig);
1409         sched_autogroup_fork(sig);
1410
1411         sig->oom_score_adj = current->signal->oom_score_adj;
1412         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1413
1414         mutex_init(&sig->cred_guard_mutex);
1415
1416         return 0;
1417 }
1418
1419 static void copy_seccomp(struct task_struct *p)
1420 {
1421 #ifdef CONFIG_SECCOMP
1422         /*
1423          * Must be called with sighand->lock held, which is common to
1424          * all threads in the group. Holding cred_guard_mutex is not
1425          * needed because this new task is not yet running and cannot
1426          * be racing exec.
1427          */
1428         assert_spin_locked(&current->sighand->siglock);
1429
1430         /* Ref-count the new filter user, and assign it. */
1431         get_seccomp_filter(current);
1432         p->seccomp = current->seccomp;
1433
1434         /*
1435          * Explicitly enable no_new_privs here in case it got set
1436          * between the task_struct being duplicated and holding the
1437          * sighand lock. The seccomp state and nnp must be in sync.
1438          */
1439         if (task_no_new_privs(current))
1440                 task_set_no_new_privs(p);
1441
1442         /*
1443          * If the parent gained a seccomp mode after copying thread
1444          * flags and between before we held the sighand lock, we have
1445          * to manually enable the seccomp thread flag here.
1446          */
1447         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1448                 set_tsk_thread_flag(p, TIF_SECCOMP);
1449 #endif
1450 }
1451
1452 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1453 {
1454         current->clear_child_tid = tidptr;
1455
1456         return task_pid_vnr(current);
1457 }
1458
1459 static void rt_mutex_init_task(struct task_struct *p)
1460 {
1461         raw_spin_lock_init(&p->pi_lock);
1462 #ifdef CONFIG_RT_MUTEXES
1463         p->pi_waiters = RB_ROOT;
1464         p->pi_waiters_leftmost = NULL;
1465         p->pi_top_task = NULL;
1466         p->pi_blocked_on = NULL;
1467 #endif
1468 }
1469
1470 #ifdef CONFIG_POSIX_TIMERS
1471 /*
1472  * Initialize POSIX timer handling for a single task.
1473  */
1474 static void posix_cpu_timers_init(struct task_struct *tsk)
1475 {
1476         tsk->cputime_expires.prof_exp = 0;
1477         tsk->cputime_expires.virt_exp = 0;
1478         tsk->cputime_expires.sched_exp = 0;
1479         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1480         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1481         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1482 }
1483 #else
1484 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1485 #endif
1486
1487 static inline void
1488 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1489 {
1490          task->pids[type].pid = pid;
1491 }
1492
1493 static inline void rcu_copy_process(struct task_struct *p)
1494 {
1495 #ifdef CONFIG_PREEMPT_RCU
1496         p->rcu_read_lock_nesting = 0;
1497         p->rcu_read_unlock_special.s = 0;
1498         p->rcu_blocked_node = NULL;
1499         INIT_LIST_HEAD(&p->rcu_node_entry);
1500 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1501 #ifdef CONFIG_TASKS_RCU
1502         p->rcu_tasks_holdout = false;
1503         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1504         p->rcu_tasks_idle_cpu = -1;
1505 #endif /* #ifdef CONFIG_TASKS_RCU */
1506 }
1507
1508 /*
1509  * This creates a new process as a copy of the old one,
1510  * but does not actually start it yet.
1511  *
1512  * It copies the registers, and all the appropriate
1513  * parts of the process environment (as per the clone
1514  * flags). The actual kick-off is left to the caller.
1515  */
1516 static __latent_entropy struct task_struct *copy_process(
1517                                         unsigned long clone_flags,
1518                                         unsigned long stack_start,
1519                                         unsigned long stack_size,
1520                                         int __user *child_tidptr,
1521                                         struct pid *pid,
1522                                         int trace,
1523                                         unsigned long tls,
1524                                         int node)
1525 {
1526         int retval;
1527         struct task_struct *p;
1528
1529         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1530                 return ERR_PTR(-EINVAL);
1531
1532         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1533                 return ERR_PTR(-EINVAL);
1534
1535         /*
1536          * Thread groups must share signals as well, and detached threads
1537          * can only be started up within the thread group.
1538          */
1539         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1540                 return ERR_PTR(-EINVAL);
1541
1542         /*
1543          * Shared signal handlers imply shared VM. By way of the above,
1544          * thread groups also imply shared VM. Blocking this case allows
1545          * for various simplifications in other code.
1546          */
1547         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1548                 return ERR_PTR(-EINVAL);
1549
1550         /*
1551          * Siblings of global init remain as zombies on exit since they are
1552          * not reaped by their parent (swapper). To solve this and to avoid
1553          * multi-rooted process trees, prevent global and container-inits
1554          * from creating siblings.
1555          */
1556         if ((clone_flags & CLONE_PARENT) &&
1557                                 current->signal->flags & SIGNAL_UNKILLABLE)
1558                 return ERR_PTR(-EINVAL);
1559
1560         /*
1561          * If the new process will be in a different pid or user namespace
1562          * do not allow it to share a thread group with the forking task.
1563          */
1564         if (clone_flags & CLONE_THREAD) {
1565                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1566                     (task_active_pid_ns(current) !=
1567                                 current->nsproxy->pid_ns_for_children))
1568                         return ERR_PTR(-EINVAL);
1569         }
1570
1571         retval = security_task_create(clone_flags);
1572         if (retval)
1573                 goto fork_out;
1574
1575         retval = -ENOMEM;
1576         p = dup_task_struct(current, node);
1577         if (!p)
1578                 goto fork_out;
1579
1580         /*
1581          * This _must_ happen before we call free_task(), i.e. before we jump
1582          * to any of the bad_fork_* labels. This is to avoid freeing
1583          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1584          * kernel threads (PF_KTHREAD).
1585          */
1586         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1587         /*
1588          * Clear TID on mm_release()?
1589          */
1590         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1591
1592         ftrace_graph_init_task(p);
1593
1594         rt_mutex_init_task(p);
1595
1596 #ifdef CONFIG_PROVE_LOCKING
1597         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1598         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1599 #endif
1600         retval = -EAGAIN;
1601         if (atomic_read(&p->real_cred->user->processes) >=
1602                         task_rlimit(p, RLIMIT_NPROC)) {
1603                 if (p->real_cred->user != INIT_USER &&
1604                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1605                         goto bad_fork_free;
1606         }
1607         current->flags &= ~PF_NPROC_EXCEEDED;
1608
1609         retval = copy_creds(p, clone_flags);
1610         if (retval < 0)
1611                 goto bad_fork_free;
1612
1613         /*
1614          * If multiple threads are within copy_process(), then this check
1615          * triggers too late. This doesn't hurt, the check is only there
1616          * to stop root fork bombs.
1617          */
1618         retval = -EAGAIN;
1619         if (nr_threads >= max_threads)
1620                 goto bad_fork_cleanup_count;
1621
1622         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1623         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1624         p->flags |= PF_FORKNOEXEC;
1625         INIT_LIST_HEAD(&p->children);
1626         INIT_LIST_HEAD(&p->sibling);
1627         rcu_copy_process(p);
1628         p->vfork_done = NULL;
1629         spin_lock_init(&p->alloc_lock);
1630
1631         init_sigpending(&p->pending);
1632
1633         p->utime = p->stime = p->gtime = 0;
1634 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1635         p->utimescaled = p->stimescaled = 0;
1636 #endif
1637         prev_cputime_init(&p->prev_cputime);
1638
1639 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1640         seqcount_init(&p->vtime.seqcount);
1641         p->vtime.starttime = 0;
1642         p->vtime.state = VTIME_INACTIVE;
1643 #endif
1644
1645 #if defined(SPLIT_RSS_COUNTING)
1646         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1647 #endif
1648
1649         p->default_timer_slack_ns = current->timer_slack_ns;
1650
1651         task_io_accounting_init(&p->ioac);
1652         acct_clear_integrals(p);
1653
1654         posix_cpu_timers_init(p);
1655
1656         p->start_time = ktime_get_ns();
1657         p->real_start_time = ktime_get_boot_ns();
1658         p->io_context = NULL;
1659         p->audit_context = NULL;
1660         cgroup_fork(p);
1661 #ifdef CONFIG_NUMA
1662         p->mempolicy = mpol_dup(p->mempolicy);
1663         if (IS_ERR(p->mempolicy)) {
1664                 retval = PTR_ERR(p->mempolicy);
1665                 p->mempolicy = NULL;
1666                 goto bad_fork_cleanup_threadgroup_lock;
1667         }
1668 #endif
1669 #ifdef CONFIG_CPUSETS
1670         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1671         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1672         seqcount_init(&p->mems_allowed_seq);
1673 #endif
1674 #ifdef CONFIG_TRACE_IRQFLAGS
1675         p->irq_events = 0;
1676         p->hardirqs_enabled = 0;
1677         p->hardirq_enable_ip = 0;
1678         p->hardirq_enable_event = 0;
1679         p->hardirq_disable_ip = _THIS_IP_;
1680         p->hardirq_disable_event = 0;
1681         p->softirqs_enabled = 1;
1682         p->softirq_enable_ip = _THIS_IP_;
1683         p->softirq_enable_event = 0;
1684         p->softirq_disable_ip = 0;
1685         p->softirq_disable_event = 0;
1686         p->hardirq_context = 0;
1687         p->softirq_context = 0;
1688 #endif
1689
1690         p->pagefault_disabled = 0;
1691
1692 #ifdef CONFIG_LOCKDEP
1693         p->lockdep_depth = 0; /* no locks held yet */
1694         p->curr_chain_key = 0;
1695         p->lockdep_recursion = 0;
1696 #endif
1697
1698 #ifdef CONFIG_DEBUG_MUTEXES
1699         p->blocked_on = NULL; /* not blocked yet */
1700 #endif
1701 #ifdef CONFIG_BCACHE
1702         p->sequential_io        = 0;
1703         p->sequential_io_avg    = 0;
1704 #endif
1705
1706         /* Perform scheduler related setup. Assign this task to a CPU. */
1707         retval = sched_fork(clone_flags, p);
1708         if (retval)
1709                 goto bad_fork_cleanup_policy;
1710
1711         retval = perf_event_init_task(p);
1712         if (retval)
1713                 goto bad_fork_cleanup_policy;
1714         retval = audit_alloc(p);
1715         if (retval)
1716                 goto bad_fork_cleanup_perf;
1717         /* copy all the process information */
1718         shm_init_task(p);
1719         retval = security_task_alloc(p, clone_flags);
1720         if (retval)
1721                 goto bad_fork_cleanup_audit;
1722         retval = copy_semundo(clone_flags, p);
1723         if (retval)
1724                 goto bad_fork_cleanup_security;
1725         retval = copy_files(clone_flags, p);
1726         if (retval)
1727                 goto bad_fork_cleanup_semundo;
1728         retval = copy_fs(clone_flags, p);
1729         if (retval)
1730                 goto bad_fork_cleanup_files;
1731         retval = copy_sighand(clone_flags, p);
1732         if (retval)
1733                 goto bad_fork_cleanup_fs;
1734         retval = copy_signal(clone_flags, p);
1735         if (retval)
1736                 goto bad_fork_cleanup_sighand;
1737         retval = copy_mm(clone_flags, p);
1738         if (retval)
1739                 goto bad_fork_cleanup_signal;
1740         retval = copy_namespaces(clone_flags, p);
1741         if (retval)
1742                 goto bad_fork_cleanup_mm;
1743         retval = copy_io(clone_flags, p);
1744         if (retval)
1745                 goto bad_fork_cleanup_namespaces;
1746         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1747         if (retval)
1748                 goto bad_fork_cleanup_io;
1749
1750         if (pid != &init_struct_pid) {
1751                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1752                 if (IS_ERR(pid)) {
1753                         retval = PTR_ERR(pid);
1754                         goto bad_fork_cleanup_thread;
1755                 }
1756         }
1757
1758 #ifdef CONFIG_BLOCK
1759         p->plug = NULL;
1760 #endif
1761 #ifdef CONFIG_FUTEX
1762         p->robust_list = NULL;
1763 #ifdef CONFIG_COMPAT
1764         p->compat_robust_list = NULL;
1765 #endif
1766         INIT_LIST_HEAD(&p->pi_state_list);
1767         p->pi_state_cache = NULL;
1768 #endif
1769         /*
1770          * sigaltstack should be cleared when sharing the same VM
1771          */
1772         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1773                 sas_ss_reset(p);
1774
1775         /*
1776          * Syscall tracing and stepping should be turned off in the
1777          * child regardless of CLONE_PTRACE.
1778          */
1779         user_disable_single_step(p);
1780         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1781 #ifdef TIF_SYSCALL_EMU
1782         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1783 #endif
1784         clear_all_latency_tracing(p);
1785
1786         /* ok, now we should be set up.. */
1787         p->pid = pid_nr(pid);
1788         if (clone_flags & CLONE_THREAD) {
1789                 p->exit_signal = -1;
1790                 p->group_leader = current->group_leader;
1791                 p->tgid = current->tgid;
1792         } else {
1793                 if (clone_flags & CLONE_PARENT)
1794                         p->exit_signal = current->group_leader->exit_signal;
1795                 else
1796                         p->exit_signal = (clone_flags & CSIGNAL);
1797                 p->group_leader = p;
1798                 p->tgid = p->pid;
1799         }
1800
1801         p->nr_dirtied = 0;
1802         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1803         p->dirty_paused_when = 0;
1804
1805         p->pdeath_signal = 0;
1806         INIT_LIST_HEAD(&p->thread_group);
1807         p->task_works = NULL;
1808
1809         cgroup_threadgroup_change_begin(current);
1810         /*
1811          * Ensure that the cgroup subsystem policies allow the new process to be
1812          * forked. It should be noted the the new process's css_set can be changed
1813          * between here and cgroup_post_fork() if an organisation operation is in
1814          * progress.
1815          */
1816         retval = cgroup_can_fork(p);
1817         if (retval)
1818                 goto bad_fork_free_pid;
1819
1820         /*
1821          * Make it visible to the rest of the system, but dont wake it up yet.
1822          * Need tasklist lock for parent etc handling!
1823          */
1824         write_lock_irq(&tasklist_lock);
1825
1826         /* CLONE_PARENT re-uses the old parent */
1827         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1828                 p->real_parent = current->real_parent;
1829                 p->parent_exec_id = current->parent_exec_id;
1830         } else {
1831                 p->real_parent = current;
1832                 p->parent_exec_id = current->self_exec_id;
1833         }
1834
1835         klp_copy_process(p);
1836
1837         spin_lock(&current->sighand->siglock);
1838
1839         /*
1840          * Copy seccomp details explicitly here, in case they were changed
1841          * before holding sighand lock.
1842          */
1843         copy_seccomp(p);
1844
1845         /*
1846          * Process group and session signals need to be delivered to just the
1847          * parent before the fork or both the parent and the child after the
1848          * fork. Restart if a signal comes in before we add the new process to
1849          * it's process group.
1850          * A fatal signal pending means that current will exit, so the new
1851          * thread can't slip out of an OOM kill (or normal SIGKILL).
1852         */
1853         recalc_sigpending();
1854         if (signal_pending(current)) {
1855                 retval = -ERESTARTNOINTR;
1856                 goto bad_fork_cancel_cgroup;
1857         }
1858         if (unlikely(!(ns_of_pid(pid)->nr_hashed & PIDNS_HASH_ADDING))) {
1859                 retval = -ENOMEM;
1860                 goto bad_fork_cancel_cgroup;
1861         }
1862
1863         if (likely(p->pid)) {
1864                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1865
1866                 init_task_pid(p, PIDTYPE_PID, pid);
1867                 if (thread_group_leader(p)) {
1868                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1869                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1870
1871                         if (is_child_reaper(pid)) {
1872                                 ns_of_pid(pid)->child_reaper = p;
1873                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1874                         }
1875
1876                         p->signal->leader_pid = pid;
1877                         p->signal->tty = tty_kref_get(current->signal->tty);
1878                         /*
1879                          * Inherit has_child_subreaper flag under the same
1880                          * tasklist_lock with adding child to the process tree
1881                          * for propagate_has_child_subreaper optimization.
1882                          */
1883                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1884                                                          p->real_parent->signal->is_child_subreaper;
1885                         list_add_tail(&p->sibling, &p->real_parent->children);
1886                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1887                         attach_pid(p, PIDTYPE_PGID);
1888                         attach_pid(p, PIDTYPE_SID);
1889                         __this_cpu_inc(process_counts);
1890                 } else {
1891                         current->signal->nr_threads++;
1892                         atomic_inc(&current->signal->live);
1893                         atomic_inc(&current->signal->sigcnt);
1894                         list_add_tail_rcu(&p->thread_group,
1895                                           &p->group_leader->thread_group);
1896                         list_add_tail_rcu(&p->thread_node,
1897                                           &p->signal->thread_head);
1898                 }
1899                 attach_pid(p, PIDTYPE_PID);
1900                 nr_threads++;
1901         }
1902
1903         total_forks++;
1904         spin_unlock(&current->sighand->siglock);
1905         syscall_tracepoint_update(p);
1906         write_unlock_irq(&tasklist_lock);
1907
1908         proc_fork_connector(p);
1909         cgroup_post_fork(p);
1910         cgroup_threadgroup_change_end(current);
1911         perf_event_fork(p);
1912
1913         trace_task_newtask(p, clone_flags);
1914         uprobe_copy_process(p, clone_flags);
1915
1916         return p;
1917
1918 bad_fork_cancel_cgroup:
1919         spin_unlock(&current->sighand->siglock);
1920         write_unlock_irq(&tasklist_lock);
1921         cgroup_cancel_fork(p);
1922 bad_fork_free_pid:
1923         cgroup_threadgroup_change_end(current);
1924         if (pid != &init_struct_pid)
1925                 free_pid(pid);
1926 bad_fork_cleanup_thread:
1927         exit_thread(p);
1928 bad_fork_cleanup_io:
1929         if (p->io_context)
1930                 exit_io_context(p);
1931 bad_fork_cleanup_namespaces:
1932         exit_task_namespaces(p);
1933 bad_fork_cleanup_mm:
1934         if (p->mm)
1935                 mmput(p->mm);
1936 bad_fork_cleanup_signal:
1937         if (!(clone_flags & CLONE_THREAD))
1938                 free_signal_struct(p->signal);
1939 bad_fork_cleanup_sighand:
1940         __cleanup_sighand(p->sighand);
1941 bad_fork_cleanup_fs:
1942         exit_fs(p); /* blocking */
1943 bad_fork_cleanup_files:
1944         exit_files(p); /* blocking */
1945 bad_fork_cleanup_semundo:
1946         exit_sem(p);
1947 bad_fork_cleanup_security:
1948         security_task_free(p);
1949 bad_fork_cleanup_audit:
1950         audit_free(p);
1951 bad_fork_cleanup_perf:
1952         perf_event_free_task(p);
1953 bad_fork_cleanup_policy:
1954 #ifdef CONFIG_NUMA
1955         mpol_put(p->mempolicy);
1956 bad_fork_cleanup_threadgroup_lock:
1957 #endif
1958         delayacct_tsk_free(p);
1959 bad_fork_cleanup_count:
1960         atomic_dec(&p->cred->user->processes);
1961         exit_creds(p);
1962 bad_fork_free:
1963         p->state = TASK_DEAD;
1964         put_task_stack(p);
1965         free_task(p);
1966 fork_out:
1967         return ERR_PTR(retval);
1968 }
1969
1970 static inline void init_idle_pids(struct pid_link *links)
1971 {
1972         enum pid_type type;
1973
1974         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1975                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1976                 links[type].pid = &init_struct_pid;
1977         }
1978 }
1979
1980 struct task_struct *fork_idle(int cpu)
1981 {
1982         struct task_struct *task;
1983         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1984                             cpu_to_node(cpu));
1985         if (!IS_ERR(task)) {
1986                 init_idle_pids(task->pids);
1987                 init_idle(task, cpu);
1988         }
1989
1990         return task;
1991 }
1992
1993 /*
1994  *  Ok, this is the main fork-routine.
1995  *
1996  * It copies the process, and if successful kick-starts
1997  * it and waits for it to finish using the VM if required.
1998  */
1999 long _do_fork(unsigned long clone_flags,
2000               unsigned long stack_start,
2001               unsigned long stack_size,
2002               int __user *parent_tidptr,
2003               int __user *child_tidptr,
2004               unsigned long tls)
2005 {
2006         struct task_struct *p;
2007         int trace = 0;
2008         long nr;
2009
2010         /*
2011          * Determine whether and which event to report to ptracer.  When
2012          * called from kernel_thread or CLONE_UNTRACED is explicitly
2013          * requested, no event is reported; otherwise, report if the event
2014          * for the type of forking is enabled.
2015          */
2016         if (!(clone_flags & CLONE_UNTRACED)) {
2017                 if (clone_flags & CLONE_VFORK)
2018                         trace = PTRACE_EVENT_VFORK;
2019                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2020                         trace = PTRACE_EVENT_CLONE;
2021                 else
2022                         trace = PTRACE_EVENT_FORK;
2023
2024                 if (likely(!ptrace_event_enabled(current, trace)))
2025                         trace = 0;
2026         }
2027
2028         p = copy_process(clone_flags, stack_start, stack_size,
2029                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2030         add_latent_entropy();
2031         /*
2032          * Do this prior waking up the new thread - the thread pointer
2033          * might get invalid after that point, if the thread exits quickly.
2034          */
2035         if (!IS_ERR(p)) {
2036                 struct completion vfork;
2037                 struct pid *pid;
2038
2039                 trace_sched_process_fork(current, p);
2040
2041                 pid = get_task_pid(p, PIDTYPE_PID);
2042                 nr = pid_vnr(pid);
2043
2044                 if (clone_flags & CLONE_PARENT_SETTID)
2045                         put_user(nr, parent_tidptr);
2046
2047                 if (clone_flags & CLONE_VFORK) {
2048                         p->vfork_done = &vfork;
2049                         init_completion(&vfork);
2050                         get_task_struct(p);
2051                 }
2052
2053                 wake_up_new_task(p);
2054
2055                 /* forking complete and child started to run, tell ptracer */
2056                 if (unlikely(trace))
2057                         ptrace_event_pid(trace, pid);
2058
2059                 if (clone_flags & CLONE_VFORK) {
2060                         if (!wait_for_vfork_done(p, &vfork))
2061                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2062                 }
2063
2064                 put_pid(pid);
2065         } else {
2066                 nr = PTR_ERR(p);
2067         }
2068         return nr;
2069 }
2070
2071 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2072 /* For compatibility with architectures that call do_fork directly rather than
2073  * using the syscall entry points below. */
2074 long do_fork(unsigned long clone_flags,
2075               unsigned long stack_start,
2076               unsigned long stack_size,
2077               int __user *parent_tidptr,
2078               int __user *child_tidptr)
2079 {
2080         return _do_fork(clone_flags, stack_start, stack_size,
2081                         parent_tidptr, child_tidptr, 0);
2082 }
2083 #endif
2084
2085 /*
2086  * Create a kernel thread.
2087  */
2088 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2089 {
2090         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2091                 (unsigned long)arg, NULL, NULL, 0);
2092 }
2093
2094 #ifdef __ARCH_WANT_SYS_FORK
2095 SYSCALL_DEFINE0(fork)
2096 {
2097 #ifdef CONFIG_MMU
2098         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2099 #else
2100         /* can not support in nommu mode */
2101         return -EINVAL;
2102 #endif
2103 }
2104 #endif
2105
2106 #ifdef __ARCH_WANT_SYS_VFORK
2107 SYSCALL_DEFINE0(vfork)
2108 {
2109         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2110                         0, NULL, NULL, 0);
2111 }
2112 #endif
2113
2114 #ifdef __ARCH_WANT_SYS_CLONE
2115 #ifdef CONFIG_CLONE_BACKWARDS
2116 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2117                  int __user *, parent_tidptr,
2118                  unsigned long, tls,
2119                  int __user *, child_tidptr)
2120 #elif defined(CONFIG_CLONE_BACKWARDS2)
2121 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2122                  int __user *, parent_tidptr,
2123                  int __user *, child_tidptr,
2124                  unsigned long, tls)
2125 #elif defined(CONFIG_CLONE_BACKWARDS3)
2126 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2127                 int, stack_size,
2128                 int __user *, parent_tidptr,
2129                 int __user *, child_tidptr,
2130                 unsigned long, tls)
2131 #else
2132 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2133                  int __user *, parent_tidptr,
2134                  int __user *, child_tidptr,
2135                  unsigned long, tls)
2136 #endif
2137 {
2138         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2139 }
2140 #endif
2141
2142 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2143 {
2144         struct task_struct *leader, *parent, *child;
2145         int res;
2146
2147         read_lock(&tasklist_lock);
2148         leader = top = top->group_leader;
2149 down:
2150         for_each_thread(leader, parent) {
2151                 list_for_each_entry(child, &parent->children, sibling) {
2152                         res = visitor(child, data);
2153                         if (res) {
2154                                 if (res < 0)
2155                                         goto out;
2156                                 leader = child;
2157                                 goto down;
2158                         }
2159 up:
2160                         ;
2161                 }
2162         }
2163
2164         if (leader != top) {
2165                 child = leader;
2166                 parent = child->real_parent;
2167                 leader = parent->group_leader;
2168                 goto up;
2169         }
2170 out:
2171         read_unlock(&tasklist_lock);
2172 }
2173
2174 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2175 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2176 #endif
2177
2178 static void sighand_ctor(void *data)
2179 {
2180         struct sighand_struct *sighand = data;
2181
2182         spin_lock_init(&sighand->siglock);
2183         init_waitqueue_head(&sighand->signalfd_wqh);
2184 }
2185
2186 void __init proc_caches_init(void)
2187 {
2188         sighand_cachep = kmem_cache_create("sighand_cache",
2189                         sizeof(struct sighand_struct), 0,
2190                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2191                         SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2192         signal_cachep = kmem_cache_create("signal_cache",
2193                         sizeof(struct signal_struct), 0,
2194                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2195                         NULL);
2196         files_cachep = kmem_cache_create("files_cache",
2197                         sizeof(struct files_struct), 0,
2198                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2199                         NULL);
2200         fs_cachep = kmem_cache_create("fs_cache",
2201                         sizeof(struct fs_struct), 0,
2202                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2203                         NULL);
2204         /*
2205          * FIXME! The "sizeof(struct mm_struct)" currently includes the
2206          * whole struct cpumask for the OFFSTACK case. We could change
2207          * this to *only* allocate as much of it as required by the
2208          * maximum number of CPU's we can ever have.  The cpumask_allocation
2209          * is at the end of the structure, exactly for that reason.
2210          */
2211         mm_cachep = kmem_cache_create("mm_struct",
2212                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2213                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2214                         NULL);
2215         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2216         mmap_init();
2217         nsproxy_cache_init();
2218 }
2219
2220 /*
2221  * Check constraints on flags passed to the unshare system call.
2222  */
2223 static int check_unshare_flags(unsigned long unshare_flags)
2224 {
2225         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2226                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2227                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2228                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2229                 return -EINVAL;
2230         /*
2231          * Not implemented, but pretend it works if there is nothing
2232          * to unshare.  Note that unsharing the address space or the
2233          * signal handlers also need to unshare the signal queues (aka
2234          * CLONE_THREAD).
2235          */
2236         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2237                 if (!thread_group_empty(current))
2238                         return -EINVAL;
2239         }
2240         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2241                 if (atomic_read(&current->sighand->count) > 1)
2242                         return -EINVAL;
2243         }
2244         if (unshare_flags & CLONE_VM) {
2245                 if (!current_is_single_threaded())
2246                         return -EINVAL;
2247         }
2248
2249         return 0;
2250 }
2251
2252 /*
2253  * Unshare the filesystem structure if it is being shared
2254  */
2255 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2256 {
2257         struct fs_struct *fs = current->fs;
2258
2259         if (!(unshare_flags & CLONE_FS) || !fs)
2260                 return 0;
2261
2262         /* don't need lock here; in the worst case we'll do useless copy */
2263         if (fs->users == 1)
2264                 return 0;
2265
2266         *new_fsp = copy_fs_struct(fs);
2267         if (!*new_fsp)
2268                 return -ENOMEM;
2269
2270         return 0;
2271 }
2272
2273 /*
2274  * Unshare file descriptor table if it is being shared
2275  */
2276 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2277 {
2278         struct files_struct *fd = current->files;
2279         int error = 0;
2280
2281         if ((unshare_flags & CLONE_FILES) &&
2282             (fd && atomic_read(&fd->count) > 1)) {
2283                 *new_fdp = dup_fd(fd, &error);
2284                 if (!*new_fdp)
2285                         return error;
2286         }
2287
2288         return 0;
2289 }
2290
2291 /*
2292  * unshare allows a process to 'unshare' part of the process
2293  * context which was originally shared using clone.  copy_*
2294  * functions used by do_fork() cannot be used here directly
2295  * because they modify an inactive task_struct that is being
2296  * constructed. Here we are modifying the current, active,
2297  * task_struct.
2298  */
2299 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2300 {
2301         struct fs_struct *fs, *new_fs = NULL;
2302         struct files_struct *fd, *new_fd = NULL;
2303         struct cred *new_cred = NULL;
2304         struct nsproxy *new_nsproxy = NULL;
2305         int do_sysvsem = 0;
2306         int err;
2307
2308         /*
2309          * If unsharing a user namespace must also unshare the thread group
2310          * and unshare the filesystem root and working directories.
2311          */
2312         if (unshare_flags & CLONE_NEWUSER)
2313                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2314         /*
2315          * If unsharing vm, must also unshare signal handlers.
2316          */
2317         if (unshare_flags & CLONE_VM)
2318                 unshare_flags |= CLONE_SIGHAND;
2319         /*
2320          * If unsharing a signal handlers, must also unshare the signal queues.
2321          */
2322         if (unshare_flags & CLONE_SIGHAND)
2323                 unshare_flags |= CLONE_THREAD;
2324         /*
2325          * If unsharing namespace, must also unshare filesystem information.
2326          */
2327         if (unshare_flags & CLONE_NEWNS)
2328                 unshare_flags |= CLONE_FS;
2329
2330         err = check_unshare_flags(unshare_flags);
2331         if (err)
2332                 goto bad_unshare_out;
2333         /*
2334          * CLONE_NEWIPC must also detach from the undolist: after switching
2335          * to a new ipc namespace, the semaphore arrays from the old
2336          * namespace are unreachable.
2337          */
2338         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2339                 do_sysvsem = 1;
2340         err = unshare_fs(unshare_flags, &new_fs);
2341         if (err)
2342                 goto bad_unshare_out;
2343         err = unshare_fd(unshare_flags, &new_fd);
2344         if (err)
2345                 goto bad_unshare_cleanup_fs;
2346         err = unshare_userns(unshare_flags, &new_cred);
2347         if (err)
2348                 goto bad_unshare_cleanup_fd;
2349         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2350                                          new_cred, new_fs);
2351         if (err)
2352                 goto bad_unshare_cleanup_cred;
2353
2354         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2355                 if (do_sysvsem) {
2356                         /*
2357                          * CLONE_SYSVSEM is equivalent to sys_exit().
2358                          */
2359                         exit_sem(current);
2360                 }
2361                 if (unshare_flags & CLONE_NEWIPC) {
2362                         /* Orphan segments in old ns (see sem above). */
2363                         exit_shm(current);
2364                         shm_init_task(current);
2365                 }
2366
2367                 if (new_nsproxy)
2368                         switch_task_namespaces(current, new_nsproxy);
2369
2370                 task_lock(current);
2371
2372                 if (new_fs) {
2373                         fs = current->fs;
2374                         spin_lock(&fs->lock);
2375                         current->fs = new_fs;
2376                         if (--fs->users)
2377                                 new_fs = NULL;
2378                         else
2379                                 new_fs = fs;
2380                         spin_unlock(&fs->lock);
2381                 }
2382
2383                 if (new_fd) {
2384                         fd = current->files;
2385                         current->files = new_fd;
2386                         new_fd = fd;
2387                 }
2388
2389                 task_unlock(current);
2390
2391                 if (new_cred) {
2392                         /* Install the new user namespace */
2393                         commit_creds(new_cred);
2394                         new_cred = NULL;
2395                 }
2396         }
2397
2398         perf_event_namespaces(current);
2399
2400 bad_unshare_cleanup_cred:
2401         if (new_cred)
2402                 put_cred(new_cred);
2403 bad_unshare_cleanup_fd:
2404         if (new_fd)
2405                 put_files_struct(new_fd);
2406
2407 bad_unshare_cleanup_fs:
2408         if (new_fs)
2409                 free_fs_struct(new_fs);
2410
2411 bad_unshare_out:
2412         return err;
2413 }
2414
2415 /*
2416  *      Helper to unshare the files of the current task.
2417  *      We don't want to expose copy_files internals to
2418  *      the exec layer of the kernel.
2419  */
2420
2421 int unshare_files(struct files_struct **displaced)
2422 {
2423         struct task_struct *task = current;
2424         struct files_struct *copy = NULL;
2425         int error;
2426
2427         error = unshare_fd(CLONE_FILES, &copy);
2428         if (error || !copy) {
2429                 *displaced = NULL;
2430                 return error;
2431         }
2432         *displaced = task->files;
2433         task_lock(task);
2434         task->files = copy;
2435         task_unlock(task);
2436         return 0;
2437 }
2438
2439 int sysctl_max_threads(struct ctl_table *table, int write,
2440                        void __user *buffer, size_t *lenp, loff_t *ppos)
2441 {
2442         struct ctl_table t;
2443         int ret;
2444         int threads = max_threads;
2445         int min = MIN_THREADS;
2446         int max = MAX_THREADS;
2447
2448         t = *table;
2449         t.data = &threads;
2450         t.extra1 = &min;
2451         t.extra2 = &max;
2452
2453         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2454         if (ret || !write)
2455                 return ret;
2456
2457         set_max_threads(threads);
2458
2459         return 0;
2460 }