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         for (i = 0; i < NR_CACHED_STACKS; i++) {
 209                 struct vm_struct *s;
 210
 211                 s = this_cpu_xchg(cached_stacks[i], NULL);
 212
 213                 if (!s)
 214                         continue;
 215
 216                 tsk->stack_vm_area = s;
 217                 return s->addr;
 218         }
 219
 220         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_SIZE,
 221                                      VMALLOC_START, VMALLOC_END,
 222                                      THREADINFO_GFP,
 223                                      PAGE_KERNEL,
 224                                      0, node, __builtin_return_address(0));
 225
 226         /*
 227          * We can't call find_vm_area() in interrupt context, and
 228          * free_thread_stack() can be called in interrupt context,
 229          * so cache the vm_struct.
 230          */
 231         if (stack)
 232                 tsk->stack_vm_area = find_vm_area(stack);
 233         return stack;
 234 #else
 235         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 236                                              THREAD_SIZE_ORDER);
 237
 238         return page ? page_address(page) : NULL;
 239 #endif
 240 }
 241
 242 static inline void free_thread_stack(struct task_struct *tsk)
 243 {
 244 #ifdef CONFIG_VMAP_STACK
 245         if (task_stack_vm_area(tsk)) {
 246                 int i;
 247
 248                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 249                         if (this_cpu_cmpxchg(cached_stacks[i],
 250                                         NULL, tsk->stack_vm_area) != NULL)
 251                                 continue;
 252
 253                         return;
 254                 }
 255
 256                 vfree_atomic(tsk->stack);
 257                 return;
 258         }
 259 #endif
 260
 261         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 262 }
 263 # else
 264 static struct kmem_cache *thread_stack_cache;
 265
 266 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 267                                                   int node)
 268 {
 269         return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 270 }
 271
 272 static void free_thread_stack(struct task_struct *tsk)
 273 {
 274         kmem_cache_free(thread_stack_cache, tsk->stack);
 275 }
 276
 277 void thread_stack_cache_init(void)
 278 {
 279         thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
 280                                               THREAD_SIZE, 0, NULL);
 281         BUG_ON(thread_stack_cache == NULL);
 282 }
 283 # endif
 284 #endif
 285
 286 /* SLAB cache for signal_struct structures (tsk->signal) */
 287 static struct kmem_cache *signal_cachep;
 288
 289 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 290 struct kmem_cache *sighand_cachep;
 291
 292 /* SLAB cache for files_struct structures (tsk->files) */
 293 struct kmem_cache *files_cachep;
 294
 295 /* SLAB cache for fs_struct structures (tsk->fs) */
 296 struct kmem_cache *fs_cachep;
 297
 298 /* SLAB cache for vm_area_struct structures */
 299 struct kmem_cache *vm_area_cachep;
 300
 301 /* SLAB cache for mm_struct structures (tsk->mm) */
 302 static struct kmem_cache *mm_cachep;
 303
 304 static void account_kernel_stack(struct task_struct *tsk, int account)
 305 {
 306         void *stack = task_stack_page(tsk);
 307         struct vm_struct *vm = task_stack_vm_area(tsk);
 308
 309         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 310
 311         if (vm) {
 312                 int i;
 313
 314                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 315
 316                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 317                         mod_zone_page_state(page_zone(vm->pages[i]),
 318                                             NR_KERNEL_STACK_KB,
 319                                             PAGE_SIZE / 1024 * account);
 320                 }
 321
 322                 /* All stack pages belong to the same memcg. */
 323                 mod_memcg_page_state(vm->pages[0], MEMCG_KERNEL_STACK_KB,
 324                                      account * (THREAD_SIZE / 1024));
 325         } else {
 326                 /*
 327                  * All stack pages are in the same zone and belong to the
 328                  * same memcg.
 329                  */
 330                 struct page *first_page = virt_to_page(stack);
 331
 332                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 333                                     THREAD_SIZE / 1024 * account);
 334
 335                 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
 336                                      account * (THREAD_SIZE / 1024));
 337         }
 338 }
 339
 340 static void release_task_stack(struct task_struct *tsk)
 341 {
 342         if (WARN_ON(tsk->state != TASK_DEAD))
 343                 return;  /* Better to leak the stack than to free prematurely */
 344
 345         account_kernel_stack(tsk, -1);
 346         arch_release_thread_stack(tsk->stack);
 347         free_thread_stack(tsk);
 348         tsk->stack = NULL;
 349 #ifdef CONFIG_VMAP_STACK
 350         tsk->stack_vm_area = NULL;
 351 #endif
 352 }
 353
 354 #ifdef CONFIG_THREAD_INFO_IN_TASK
 355 void put_task_stack(struct task_struct *tsk)
 356 {
 357         if (atomic_dec_and_test(&tsk->stack_refcount))
 358                 release_task_stack(tsk);
 359 }
 360 #endif
 361
 362 void free_task(struct task_struct *tsk)
 363 {
 364 #ifndef CONFIG_THREAD_INFO_IN_TASK
 365         /*
 366          * The task is finally done with both the stack and thread_info,
 367          * so free both.
 368          */
 369         release_task_stack(tsk);
 370 #else
 371         /*
 372          * If the task had a separate stack allocation, it should be gone
 373          * by now.
 374          */
 375         WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
 376 #endif
 377         rt_mutex_debug_task_free(tsk);
 378         ftrace_graph_exit_task(tsk);
 379         put_seccomp_filter(tsk);
 380         arch_release_task_struct(tsk);
 381         if (tsk->flags & PF_KTHREAD)
 382                 free_kthread_struct(tsk);
 383         free_task_struct(tsk);
 384 }
 385 EXPORT_SYMBOL(free_task);
 386
 387 static inline void free_signal_struct(struct signal_struct *sig)
 388 {
 389         taskstats_tgid_free(sig);
 390         sched_autogroup_exit(sig);
 391         /*
 392          * __mmdrop is not safe to call from softirq context on x86 due to
 393          * pgd_dtor so postpone it to the async context
 394          */
 395         if (sig->oom_mm)
 396                 mmdrop_async(sig->oom_mm);
 397         kmem_cache_free(signal_cachep, sig);
 398 }
 399
 400 static inline void put_signal_struct(struct signal_struct *sig)
 401 {
 402         if (atomic_dec_and_test(&sig->sigcnt))
 403                 free_signal_struct(sig);
 404 }
 405
 406 void __put_task_struct(struct task_struct *tsk)
 407 {
 408         WARN_ON(!tsk->exit_state);
 409         WARN_ON(atomic_read(&tsk->usage));
 410         WARN_ON(tsk == current);
 411
 412         cgroup_free(tsk);
 413         task_numa_free(tsk);
 414         security_task_free(tsk);
 415         exit_creds(tsk);
 416         delayacct_tsk_free(tsk);
 417         put_signal_struct(tsk->signal);
 418
 419         if (!profile_handoff_task(tsk))
 420                 free_task(tsk);
 421 }
 422 EXPORT_SYMBOL_GPL(__put_task_struct);
 423
 424 void __init __weak arch_task_cache_init(void) { }
 425
 426 /*
 427  * set_max_threads
 428  */
 429 static void set_max_threads(unsigned int max_threads_suggested)
 430 {
 431         u64 threads;
 432
 433         /*
 434          * The number of threads shall be limited such that the thread
 435          * structures may only consume a small part of the available memory.
 436          */
 437         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 438                 threads = MAX_THREADS;
 439         else
 440                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 441                                     (u64) THREAD_SIZE * 8UL);
 442
 443         if (threads > max_threads_suggested)
 444                 threads = max_threads_suggested;
 445
 446         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 447 }
 448
 449 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 450 /* Initialized by the architecture: */
 451 int arch_task_struct_size __read_mostly;
 452 #endif
 453
 454 void __init fork_init(void)
 455 {
 456         int i;
 457 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 458 #ifndef ARCH_MIN_TASKALIGN
 459 #define ARCH_MIN_TASKALIGN      0
 460 #endif
 461         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 462
 463         /* create a slab on which task_structs can be allocated */
 464         task_struct_cachep = kmem_cache_create("task_struct",
 465                         arch_task_struct_size, align,
 466                         SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
 467 #endif
 468
 469         /* do the arch specific task caches init */
 470         arch_task_cache_init();
 471
 472         set_max_threads(MAX_THREADS);
 473
 474         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 475         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 476         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 477                 init_task.signal->rlim[RLIMIT_NPROC];
 478
 479         for (i = 0; i < UCOUNT_COUNTS; i++) {
 480                 init_user_ns.ucount_max[i] = max_threads/2;
 481         }
 482
 483 #ifdef CONFIG_VMAP_STACK
 484         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 485                           NULL, free_vm_stack_cache);
 486 #endif
 487
 488         lockdep_init_task(&init_task);
 489 }
 490
 491 int __weak arch_dup_task_struct(struct task_struct *dst,
 492                                                struct task_struct *src)
 493 {
 494         *dst = *src;
 495         return 0;
 496 }
 497
 498 void set_task_stack_end_magic(struct task_struct *tsk)
 499 {
 500         unsigned long *stackend;
 501
 502         stackend = end_of_stack(tsk);
 503         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 504 }
 505
 506 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 507 {
 508         struct task_struct *tsk;
 509         unsigned long *stack;
 510         struct vm_struct *stack_vm_area;
 511         int err;
 512
 513         if (node == NUMA_NO_NODE)
 514                 node = tsk_fork_get_node(orig);
 515         tsk = alloc_task_struct_node(node);
 516         if (!tsk)
 517                 return NULL;
 518
 519         stack = alloc_thread_stack_node(tsk, node);
 520         if (!stack)
 521                 goto free_tsk;
 522
 523         stack_vm_area = task_stack_vm_area(tsk);
 524
 525         err = arch_dup_task_struct(tsk, orig);
 526
 527         /*
 528          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 529          * sure they're properly initialized before using any stack-related
 530          * functions again.
 531          */
 532         tsk->stack = stack;
 533 #ifdef CONFIG_VMAP_STACK
 534         tsk->stack_vm_area = stack_vm_area;
 535 #endif
 536 #ifdef CONFIG_THREAD_INFO_IN_TASK
 537         atomic_set(&tsk->stack_refcount, 1);
 538 #endif
 539
 540         if (err)
 541                 goto free_stack;
 542
 543 #ifdef CONFIG_SECCOMP
 544         /*
 545          * We must handle setting up seccomp filters once we're under
 546          * the sighand lock in case orig has changed between now and
 547          * then. Until then, filter must be NULL to avoid messing up
 548          * the usage counts on the error path calling free_task.
 549          */
 550         tsk->seccomp.filter = NULL;
 551 #endif
 552
 553         setup_thread_stack(tsk, orig);
 554         clear_user_return_notifier(tsk);
 555         clear_tsk_need_resched(tsk);
 556         set_task_stack_end_magic(tsk);
 557
 558 #ifdef CONFIG_CC_STACKPROTECTOR
 559         tsk->stack_canary = get_random_canary();
 560 #endif
 561
 562         /*
 563          * One for us, one for whoever does the "release_task()" (usually
 564          * parent)
 565          */
 566         atomic_set(&tsk->usage, 2);
 567 #ifdef CONFIG_BLK_DEV_IO_TRACE
 568         tsk->btrace_seq = 0;
 569 #endif
 570         tsk->splice_pipe = NULL;
 571         tsk->task_frag.page = NULL;
 572         tsk->wake_q.next = NULL;
 573
 574         account_kernel_stack(tsk, 1);
 575
 576         kcov_task_init(tsk);
 577
 578 #ifdef CONFIG_FAULT_INJECTION
 579         tsk->fail_nth = 0;
 580 #endif
 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 void mm_init_uprobes_state(struct mm_struct *mm)
 791 {
 792 #ifdef CONFIG_UPROBES
 793         mm->uprobes_state.xol_area = NULL;
 794 #endif
 795 }
 796
 797 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 798         struct user_namespace *user_ns)
 799 {
 800         mm->mmap = NULL;
 801         mm->mm_rb = RB_ROOT;
 802         mm->vmacache_seqnum = 0;
 803         atomic_set(&mm->mm_users, 1);
 804         atomic_set(&mm->mm_count, 1);
 805         init_rwsem(&mm->mmap_sem);
 806         INIT_LIST_HEAD(&mm->mmlist);
 807         mm->core_state = NULL;
 808         atomic_long_set(&mm->nr_ptes, 0);
 809         mm_nr_pmds_init(mm);
 810         mm->map_count = 0;
 811         mm->locked_vm = 0;
 812         mm->pinned_vm = 0;
 813         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 814         spin_lock_init(&mm->page_table_lock);
 815         mm_init_cpumask(mm);
 816         mm_init_aio(mm);
 817         mm_init_owner(mm, p);
 818         RCU_INIT_POINTER(mm->exe_file, NULL);
 819         mmu_notifier_mm_init(mm);
 820         init_tlb_flush_pending(mm);
 821 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 822         mm->pmd_huge_pte = NULL;
 823 #endif
 824         mm_init_uprobes_state(mm);
 825
 826         if (current->mm) {
 827                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 828                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 829         } else {
 830                 mm->flags = default_dump_filter;
 831                 mm->def_flags = 0;
 832         }
 833
 834         if (mm_alloc_pgd(mm))
 835                 goto fail_nopgd;
 836
 837         if (init_new_context(p, mm))
 838                 goto fail_nocontext;
 839
 840         mm->user_ns = get_user_ns(user_ns);
 841         return mm;
 842
 843 fail_nocontext:
 844         mm_free_pgd(mm);
 845 fail_nopgd:
 846         free_mm(mm);
 847         return NULL;
 848 }
 849
 850 static void check_mm(struct mm_struct *mm)
 851 {
 852         int i;
 853
 854         for (i = 0; i < NR_MM_COUNTERS; i++) {
 855                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 856
 857                 if (unlikely(x))
 858                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 859                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 860         }
 861
 862         if (atomic_long_read(&mm->nr_ptes))
 863                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 864                                 atomic_long_read(&mm->nr_ptes));
 865         if (mm_nr_pmds(mm))
 866                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 867                                 mm_nr_pmds(mm));
 868
 869 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 870         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 871 #endif
 872 }
 873
 874 /*
 875  * Allocate and initialize an mm_struct.
 876  */
 877 struct mm_struct *mm_alloc(void)
 878 {
 879         struct mm_struct *mm;
 880
 881         mm = allocate_mm();
 882         if (!mm)
 883                 return NULL;
 884
 885         memset(mm, 0, sizeof(*mm));
 886         return mm_init(mm, current, current_user_ns());
 887 }
 888
 889 /*
 890  * Called when the last reference to the mm
 891  * is dropped: either by a lazy thread or by
 892  * mmput. Free the page directory and the mm.
 893  */
 894 void __mmdrop(struct mm_struct *mm)
 895 {
 896         BUG_ON(mm == &init_mm);
 897         mm_free_pgd(mm);
 898         destroy_context(mm);
 899         mmu_notifier_mm_destroy(mm);
 900         check_mm(mm);
 901         put_user_ns(mm->user_ns);
 902         free_mm(mm);
 903 }
 904 EXPORT_SYMBOL_GPL(__mmdrop);
 905
 906 static inline void __mmput(struct mm_struct *mm)
 907 {
 908         VM_BUG_ON(atomic_read(&mm->mm_users));
 909
 910         uprobe_clear_state(mm);
 911         exit_aio(mm);
 912         ksm_exit(mm);
 913         khugepaged_exit(mm); /* must run before exit_mmap */
 914         exit_mmap(mm);
 915         mm_put_huge_zero_page(mm);
 916         set_mm_exe_file(mm, NULL);
 917         if (!list_empty(&mm->mmlist)) {
 918                 spin_lock(&mmlist_lock);
 919                 list_del(&mm->mmlist);
 920                 spin_unlock(&mmlist_lock);
 921         }
 922         if (mm->binfmt)
 923                 module_put(mm->binfmt->module);
 924         set_bit(MMF_OOM_SKIP, &mm->flags);
 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 #ifdef CONFIG_MMU
 941 static void mmput_async_fn(struct work_struct *work)
 942 {
 943         struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
 944         __mmput(mm);
 945 }
 946
 947 void mmput_async(struct mm_struct *mm)
 948 {
 949         if (atomic_dec_and_test(&mm->mm_users)) {
 950                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
 951                 schedule_work(&mm->async_put_work);
 952         }
 953 }
 954 #endif
 955
 956 /**
 957  * set_mm_exe_file - change a reference to the mm's executable file
 958  *
 959  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 960  *
 961  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 962  * invocations: in mmput() nobody alive left, in execve task is single
 963  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 964  * mm->exe_file, but does so without using set_mm_exe_file() in order
 965  * to do avoid the need for any locks.
 966  */
 967 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 968 {
 969         struct file *old_exe_file;
 970
 971         /*
 972          * It is safe to dereference the exe_file without RCU as
 973          * this function is only called if nobody else can access
 974          * this mm -- see comment above for justification.
 975          */
 976         old_exe_file = rcu_dereference_raw(mm->exe_file);
 977
 978         if (new_exe_file)
 979                 get_file(new_exe_file);
 980         rcu_assign_pointer(mm->exe_file, new_exe_file);
 981         if (old_exe_file)
 982                 fput(old_exe_file);
 983 }
 984
 985 /**
 986  * get_mm_exe_file - acquire a reference to the mm's executable file
 987  *
 988  * Returns %NULL if mm has no associated executable file.
 989  * User must release file via fput().
 990  */
 991 struct file *get_mm_exe_file(struct mm_struct *mm)
 992 {
 993         struct file *exe_file;
 994
 995         rcu_read_lock();
 996         exe_file = rcu_dereference(mm->exe_file);
 997         if (exe_file && !get_file_rcu(exe_file))
 998                 exe_file = NULL;
 999         rcu_read_unlock();
1000         return exe_file;
1001 }
1002 EXPORT_SYMBOL(get_mm_exe_file);
1003
1004 /**
1005  * get_task_exe_file - acquire a reference to the task's executable file
1006  *
1007  * Returns %NULL if task's mm (if any) has no associated executable file or
1008  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1009  * User must release file via fput().
1010  */
1011 struct file *get_task_exe_file(struct task_struct *task)
1012 {
1013         struct file *exe_file = NULL;
1014         struct mm_struct *mm;
1015
1016         task_lock(task);
1017         mm = task->mm;
1018         if (mm) {
1019                 if (!(task->flags & PF_KTHREAD))
1020                         exe_file = get_mm_exe_file(mm);
1021         }
1022         task_unlock(task);
1023         return exe_file;
1024 }
1025 EXPORT_SYMBOL(get_task_exe_file);
1026
1027 /**
1028  * get_task_mm - acquire a reference to the task's mm
1029  *
1030  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1031  * this kernel workthread has transiently adopted a user mm with use_mm,
1032  * to do its AIO) is not set and if so returns a reference to it, after
1033  * bumping up the use count.  User must release the mm via mmput()
1034  * after use.  Typically used by /proc and ptrace.
1035  */
1036 struct mm_struct *get_task_mm(struct task_struct *task)
1037 {
1038         struct mm_struct *mm;
1039
1040         task_lock(task);
1041         mm = task->mm;
1042         if (mm) {
1043                 if (task->flags & PF_KTHREAD)
1044                         mm = NULL;
1045                 else
1046                         mmget(mm);
1047         }
1048         task_unlock(task);
1049         return mm;
1050 }
1051 EXPORT_SYMBOL_GPL(get_task_mm);
1052
1053 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1054 {
1055         struct mm_struct *mm;
1056         int err;
1057
1058         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1059         if (err)
1060                 return ERR_PTR(err);
1061
1062         mm = get_task_mm(task);
1063         if (mm && mm != current->mm &&
1064                         !ptrace_may_access(task, mode)) {
1065                 mmput(mm);
1066                 mm = ERR_PTR(-EACCES);
1067         }
1068         mutex_unlock(&task->signal->cred_guard_mutex);
1069
1070         return mm;
1071 }
1072
1073 static void complete_vfork_done(struct task_struct *tsk)
1074 {
1075         struct completion *vfork;
1076
1077         task_lock(tsk);
1078         vfork = tsk->vfork_done;
1079         if (likely(vfork)) {
1080                 tsk->vfork_done = NULL;
1081                 complete(vfork);
1082         }
1083         task_unlock(tsk);
1084 }
1085
1086 static int wait_for_vfork_done(struct task_struct *child,
1087                                 struct completion *vfork)
1088 {
1089         int killed;
1090
1091         freezer_do_not_count();
1092         killed = wait_for_completion_killable(vfork);
1093         freezer_count();
1094
1095         if (killed) {
1096                 task_lock(child);
1097                 child->vfork_done = NULL;
1098                 task_unlock(child);
1099         }
1100
1101         put_task_struct(child);
1102         return killed;
1103 }
1104
1105 /* Please note the differences between mmput and mm_release.
1106  * mmput is called whenever we stop holding onto a mm_struct,
1107  * error success whatever.
1108  *
1109  * mm_release is called after a mm_struct has been removed
1110  * from the current process.
1111  *
1112  * This difference is important for error handling, when we
1113  * only half set up a mm_struct for a new process and need to restore
1114  * the old one.  Because we mmput the new mm_struct before
1115  * restoring the old one. . .
1116  * Eric Biederman 10 January 1998
1117  */
1118 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1119 {
1120         /* Get rid of any futexes when releasing the mm */
1121 #ifdef CONFIG_FUTEX
1122         if (unlikely(tsk->robust_list)) {
1123                 exit_robust_list(tsk);
1124                 tsk->robust_list = NULL;
1125         }
1126 #ifdef CONFIG_COMPAT
1127         if (unlikely(tsk->compat_robust_list)) {
1128                 compat_exit_robust_list(tsk);
1129                 tsk->compat_robust_list = NULL;
1130         }
1131 #endif
1132         if (unlikely(!list_empty(&tsk->pi_state_list)))
1133                 exit_pi_state_list(tsk);
1134 #endif
1135
1136         uprobe_free_utask(tsk);
1137
1138         /* Get rid of any cached register state */
1139         deactivate_mm(tsk, mm);
1140
1141         /*
1142          * Signal userspace if we're not exiting with a core dump
1143          * because we want to leave the value intact for debugging
1144          * purposes.
1145          */
1146         if (tsk->clear_child_tid) {
1147                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1148                     atomic_read(&mm->mm_users) > 1) {
1149                         /*
1150                          * We don't check the error code - if userspace has
1151                          * not set up a proper pointer then tough luck.
1152                          */
1153                         put_user(0, tsk->clear_child_tid);
1154                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1155                                         1, NULL, NULL, 0);
1156                 }
1157                 tsk->clear_child_tid = NULL;
1158         }
1159
1160         /*
1161          * All done, finally we can wake up parent and return this mm to him.
1162          * Also kthread_stop() uses this completion for synchronization.
1163          */
1164         if (tsk->vfork_done)
1165                 complete_vfork_done(tsk);
1166 }
1167
1168 /*
1169  * Allocate a new mm structure and copy contents from the
1170  * mm structure of the passed in task structure.
1171  */
1172 static struct mm_struct *dup_mm(struct task_struct *tsk)
1173 {
1174         struct mm_struct *mm, *oldmm = current->mm;
1175         int err;
1176
1177         mm = allocate_mm();
1178         if (!mm)
1179                 goto fail_nomem;
1180
1181         memcpy(mm, oldmm, sizeof(*mm));
1182
1183         if (!mm_init(mm, tsk, mm->user_ns))
1184                 goto fail_nomem;
1185
1186         err = dup_mmap(mm, oldmm);
1187         if (err)
1188                 goto free_pt;
1189
1190         mm->hiwater_rss = get_mm_rss(mm);
1191         mm->hiwater_vm = mm->total_vm;
1192
1193         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1194                 goto free_pt;
1195
1196         return mm;
1197
1198 free_pt:
1199         /* don't put binfmt in mmput, we haven't got module yet */
1200         mm->binfmt = NULL;
1201         mmput(mm);
1202
1203 fail_nomem:
1204         return NULL;
1205 }
1206
1207 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1208 {
1209         struct mm_struct *mm, *oldmm;
1210         int retval;
1211
1212         tsk->min_flt = tsk->maj_flt = 0;
1213         tsk->nvcsw = tsk->nivcsw = 0;
1214 #ifdef CONFIG_DETECT_HUNG_TASK
1215         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1216 #endif
1217
1218         tsk->mm = NULL;
1219         tsk->active_mm = NULL;
1220
1221         /*
1222          * Are we cloning a kernel thread?
1223          *
1224          * We need to steal a active VM for that..
1225          */
1226         oldmm = current->mm;
1227         if (!oldmm)
1228                 return 0;
1229
1230         /* initialize the new vmacache entries */
1231         vmacache_flush(tsk);
1232
1233         if (clone_flags & CLONE_VM) {
1234                 mmget(oldmm);
1235                 mm = oldmm;
1236                 goto good_mm;
1237         }
1238
1239         retval = -ENOMEM;
1240         mm = dup_mm(tsk);
1241         if (!mm)
1242                 goto fail_nomem;
1243
1244 good_mm:
1245         tsk->mm = mm;
1246         tsk->active_mm = mm;
1247         return 0;
1248
1249 fail_nomem:
1250         return retval;
1251 }
1252
1253 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1254 {
1255         struct fs_struct *fs = current->fs;
1256         if (clone_flags & CLONE_FS) {
1257                 /* tsk->fs is already what we want */
1258                 spin_lock(&fs->lock);
1259                 if (fs->in_exec) {
1260                         spin_unlock(&fs->lock);
1261                         return -EAGAIN;
1262                 }
1263                 fs->users++;
1264                 spin_unlock(&fs->lock);
1265                 return 0;
1266         }
1267         tsk->fs = copy_fs_struct(fs);
1268         if (!tsk->fs)
1269                 return -ENOMEM;
1270         return 0;
1271 }
1272
1273 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1274 {
1275         struct files_struct *oldf, *newf;
1276         int error = 0;
1277
1278         /*
1279          * A background process may not have any files ...
1280          */
1281         oldf = current->files;
1282         if (!oldf)
1283                 goto out;
1284
1285         if (clone_flags & CLONE_FILES) {
1286                 atomic_inc(&oldf->count);
1287                 goto out;
1288         }
1289
1290         newf = dup_fd(oldf, &error);
1291         if (!newf)
1292                 goto out;
1293
1294         tsk->files = newf;
1295         error = 0;
1296 out:
1297         return error;
1298 }
1299
1300 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1301 {
1302 #ifdef CONFIG_BLOCK
1303         struct io_context *ioc = current->io_context;
1304         struct io_context *new_ioc;
1305
1306         if (!ioc)
1307                 return 0;
1308         /*
1309          * Share io context with parent, if CLONE_IO is set
1310          */
1311         if (clone_flags & CLONE_IO) {
1312                 ioc_task_link(ioc);
1313                 tsk->io_context = ioc;
1314         } else if (ioprio_valid(ioc->ioprio)) {
1315                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1316                 if (unlikely(!new_ioc))
1317                         return -ENOMEM;
1318
1319                 new_ioc->ioprio = ioc->ioprio;
1320                 put_io_context(new_ioc);
1321         }
1322 #endif
1323         return 0;
1324 }
1325
1326 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1327 {
1328         struct sighand_struct *sig;
1329
1330         if (clone_flags & CLONE_SIGHAND) {
1331                 atomic_inc(&current->sighand->count);
1332                 return 0;
1333         }
1334         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1335         rcu_assign_pointer(tsk->sighand, sig);
1336         if (!sig)
1337                 return -ENOMEM;
1338
1339         atomic_set(&sig->count, 1);
1340         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1341         return 0;
1342 }
1343
1344 void __cleanup_sighand(struct sighand_struct *sighand)
1345 {
1346         if (atomic_dec_and_test(&sighand->count)) {
1347                 signalfd_cleanup(sighand);
1348                 /*
1349                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1350                  * without an RCU grace period, see __lock_task_sighand().
1351                  */
1352                 kmem_cache_free(sighand_cachep, sighand);
1353         }
1354 }
1355
1356 #ifdef CONFIG_POSIX_TIMERS
1357 /*
1358  * Initialize POSIX timer handling for a thread group.
1359  */
1360 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1361 {
1362         unsigned long cpu_limit;
1363
1364         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1365         if (cpu_limit != RLIM_INFINITY) {
1366                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1367                 sig->cputimer.running = true;
1368         }
1369
1370         /* The timer lists. */
1371         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1372         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1373         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1374 }
1375 #else
1376 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1377 #endif
1378
1379 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1380 {
1381         struct signal_struct *sig;
1382
1383         if (clone_flags & CLONE_THREAD)
1384                 return 0;
1385
1386         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1387         tsk->signal = sig;
1388         if (!sig)
1389                 return -ENOMEM;
1390
1391         sig->nr_threads = 1;
1392         atomic_set(&sig->live, 1);
1393         atomic_set(&sig->sigcnt, 1);
1394
1395         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1396         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1397         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1398
1399         init_waitqueue_head(&sig->wait_chldexit);
1400         sig->curr_target = tsk;
1401         init_sigpending(&sig->shared_pending);
1402         seqlock_init(&sig->stats_lock);
1403         prev_cputime_init(&sig->prev_cputime);
1404
1405 #ifdef CONFIG_POSIX_TIMERS
1406         INIT_LIST_HEAD(&sig->posix_timers);
1407         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1408         sig->real_timer.function = it_real_fn;
1409 #endif
1410
1411         task_lock(current->group_leader);
1412         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1413         task_unlock(current->group_leader);
1414
1415         posix_cpu_timers_init_group(sig);
1416
1417         tty_audit_fork(sig);
1418         sched_autogroup_fork(sig);
1419
1420         sig->oom_score_adj = current->signal->oom_score_adj;
1421         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1422
1423         mutex_init(&sig->cred_guard_mutex);
1424
1425         return 0;
1426 }
1427
1428 static void copy_seccomp(struct task_struct *p)
1429 {
1430 #ifdef CONFIG_SECCOMP
1431         /*
1432          * Must be called with sighand->lock held, which is common to
1433          * all threads in the group. Holding cred_guard_mutex is not
1434          * needed because this new task is not yet running and cannot
1435          * be racing exec.
1436          */
1437         assert_spin_locked(&current->sighand->siglock);
1438
1439         /* Ref-count the new filter user, and assign it. */
1440         get_seccomp_filter(current);
1441         p->seccomp = current->seccomp;
1442
1443         /*
1444          * Explicitly enable no_new_privs here in case it got set
1445          * between the task_struct being duplicated and holding the
1446          * sighand lock. The seccomp state and nnp must be in sync.
1447          */
1448         if (task_no_new_privs(current))
1449                 task_set_no_new_privs(p);
1450
1451         /*
1452          * If the parent gained a seccomp mode after copying thread
1453          * flags and between before we held the sighand lock, we have
1454          * to manually enable the seccomp thread flag here.
1455          */
1456         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1457                 set_tsk_thread_flag(p, TIF_SECCOMP);
1458 #endif
1459 }
1460
1461 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1462 {
1463         current->clear_child_tid = tidptr;
1464
1465         return task_pid_vnr(current);
1466 }
1467
1468 static void rt_mutex_init_task(struct task_struct *p)
1469 {
1470         raw_spin_lock_init(&p->pi_lock);
1471 #ifdef CONFIG_RT_MUTEXES
1472         p->pi_waiters = RB_ROOT;
1473         p->pi_waiters_leftmost = NULL;
1474         p->pi_top_task = NULL;
1475         p->pi_blocked_on = NULL;
1476 #endif
1477 }
1478
1479 #ifdef CONFIG_POSIX_TIMERS
1480 /*
1481  * Initialize POSIX timer handling for a single task.
1482  */
1483 static void posix_cpu_timers_init(struct task_struct *tsk)
1484 {
1485         tsk->cputime_expires.prof_exp = 0;
1486         tsk->cputime_expires.virt_exp = 0;
1487         tsk->cputime_expires.sched_exp = 0;
1488         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1489         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1490         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1491 }
1492 #else
1493 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1494 #endif
1495
1496 static inline void
1497 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1498 {
1499          task->pids[type].pid = pid;
1500 }
1501
1502 static inline void rcu_copy_process(struct task_struct *p)
1503 {
1504 #ifdef CONFIG_PREEMPT_RCU
1505         p->rcu_read_lock_nesting = 0;
1506         p->rcu_read_unlock_special.s = 0;
1507         p->rcu_blocked_node = NULL;
1508         INIT_LIST_HEAD(&p->rcu_node_entry);
1509 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1510 #ifdef CONFIG_TASKS_RCU
1511         p->rcu_tasks_holdout = false;
1512         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1513         p->rcu_tasks_idle_cpu = -1;
1514 #endif /* #ifdef CONFIG_TASKS_RCU */
1515 }
1516
1517 /*
1518  * This creates a new process as a copy of the old one,
1519  * but does not actually start it yet.
1520  *
1521  * It copies the registers, and all the appropriate
1522  * parts of the process environment (as per the clone
1523  * flags). The actual kick-off is left to the caller.
1524  */
1525 static __latent_entropy struct task_struct *copy_process(
1526                                         unsigned long clone_flags,
1527                                         unsigned long stack_start,
1528                                         unsigned long stack_size,
1529                                         int __user *child_tidptr,
1530                                         struct pid *pid,
1531                                         int trace,
1532                                         unsigned long tls,
1533                                         int node)
1534 {
1535         int retval;
1536         struct task_struct *p;
1537
1538         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1539                 return ERR_PTR(-EINVAL);
1540
1541         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1542                 return ERR_PTR(-EINVAL);
1543
1544         /*
1545          * Thread groups must share signals as well, and detached threads
1546          * can only be started up within the thread group.
1547          */
1548         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1549                 return ERR_PTR(-EINVAL);
1550
1551         /*
1552          * Shared signal handlers imply shared VM. By way of the above,
1553          * thread groups also imply shared VM. Blocking this case allows
1554          * for various simplifications in other code.
1555          */
1556         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1557                 return ERR_PTR(-EINVAL);
1558
1559         /*
1560          * Siblings of global init remain as zombies on exit since they are
1561          * not reaped by their parent (swapper). To solve this and to avoid
1562          * multi-rooted process trees, prevent global and container-inits
1563          * from creating siblings.
1564          */
1565         if ((clone_flags & CLONE_PARENT) &&
1566                                 current->signal->flags & SIGNAL_UNKILLABLE)
1567                 return ERR_PTR(-EINVAL);
1568
1569         /*
1570          * If the new process will be in a different pid or user namespace
1571          * do not allow it to share a thread group with the forking task.
1572          */
1573         if (clone_flags & CLONE_THREAD) {
1574                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1575                     (task_active_pid_ns(current) !=
1576                                 current->nsproxy->pid_ns_for_children))
1577                         return ERR_PTR(-EINVAL);
1578         }
1579
1580         retval = security_task_create(clone_flags);
1581         if (retval)
1582                 goto fork_out;
1583
1584         retval = -ENOMEM;
1585         p = dup_task_struct(current, node);
1586         if (!p)
1587                 goto fork_out;
1588
1589         /*
1590          * This _must_ happen before we call free_task(), i.e. before we jump
1591          * to any of the bad_fork_* labels. This is to avoid freeing
1592          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1593          * kernel threads (PF_KTHREAD).
1594          */
1595         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1596         /*
1597          * Clear TID on mm_release()?
1598          */
1599         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1600
1601         ftrace_graph_init_task(p);
1602
1603         rt_mutex_init_task(p);
1604
1605 #ifdef CONFIG_PROVE_LOCKING
1606         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1607         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1608 #endif
1609         retval = -EAGAIN;
1610         if (atomic_read(&p->real_cred->user->processes) >=
1611                         task_rlimit(p, RLIMIT_NPROC)) {
1612                 if (p->real_cred->user != INIT_USER &&
1613                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1614                         goto bad_fork_free;
1615         }
1616         current->flags &= ~PF_NPROC_EXCEEDED;
1617
1618         retval = copy_creds(p, clone_flags);
1619         if (retval < 0)
1620                 goto bad_fork_free;
1621
1622         /*
1623          * If multiple threads are within copy_process(), then this check
1624          * triggers too late. This doesn't hurt, the check is only there
1625          * to stop root fork bombs.
1626          */
1627         retval = -EAGAIN;
1628         if (nr_threads >= max_threads)
1629                 goto bad_fork_cleanup_count;
1630
1631         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1632         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1633         p->flags |= PF_FORKNOEXEC;
1634         INIT_LIST_HEAD(&p->children);
1635         INIT_LIST_HEAD(&p->sibling);
1636         rcu_copy_process(p);
1637         p->vfork_done = NULL;
1638         spin_lock_init(&p->alloc_lock);
1639
1640         init_sigpending(&p->pending);
1641
1642         p->utime = p->stime = p->gtime = 0;
1643 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1644         p->utimescaled = p->stimescaled = 0;
1645 #endif
1646         prev_cputime_init(&p->prev_cputime);
1647
1648 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1649         seqcount_init(&p->vtime.seqcount);
1650         p->vtime.starttime = 0;
1651         p->vtime.state = VTIME_INACTIVE;
1652 #endif
1653
1654 #if defined(SPLIT_RSS_COUNTING)
1655         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1656 #endif
1657
1658         p->default_timer_slack_ns = current->timer_slack_ns;
1659
1660         task_io_accounting_init(&p->ioac);
1661         acct_clear_integrals(p);
1662
1663         posix_cpu_timers_init(p);
1664
1665         p->start_time = ktime_get_ns();
1666         p->real_start_time = ktime_get_boot_ns();
1667         p->io_context = NULL;
1668         p->audit_context = NULL;
1669         cgroup_fork(p);
1670 #ifdef CONFIG_NUMA
1671         p->mempolicy = mpol_dup(p->mempolicy);
1672         if (IS_ERR(p->mempolicy)) {
1673                 retval = PTR_ERR(p->mempolicy);
1674                 p->mempolicy = NULL;
1675                 goto bad_fork_cleanup_threadgroup_lock;
1676         }
1677 #endif
1678 #ifdef CONFIG_CPUSETS
1679         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1680         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1681         seqcount_init(&p->mems_allowed_seq);
1682 #endif
1683 #ifdef CONFIG_TRACE_IRQFLAGS
1684         p->irq_events = 0;
1685         p->hardirqs_enabled = 0;
1686         p->hardirq_enable_ip = 0;
1687         p->hardirq_enable_event = 0;
1688         p->hardirq_disable_ip = _THIS_IP_;
1689         p->hardirq_disable_event = 0;
1690         p->softirqs_enabled = 1;
1691         p->softirq_enable_ip = _THIS_IP_;
1692         p->softirq_enable_event = 0;
1693         p->softirq_disable_ip = 0;
1694         p->softirq_disable_event = 0;
1695         p->hardirq_context = 0;
1696         p->softirq_context = 0;
1697 #endif
1698
1699         p->pagefault_disabled = 0;
1700
1701 #ifdef CONFIG_LOCKDEP
1702         p->lockdep_depth = 0; /* no locks held yet */
1703         p->curr_chain_key = 0;
1704         p->lockdep_recursion = 0;
1705         lockdep_init_task(p);
1706 #endif
1707
1708 #ifdef CONFIG_DEBUG_MUTEXES
1709         p->blocked_on = NULL; /* not blocked yet */
1710 #endif
1711 #ifdef CONFIG_BCACHE
1712         p->sequential_io        = 0;
1713         p->sequential_io_avg    = 0;
1714 #endif
1715
1716         /* Perform scheduler related setup. Assign this task to a CPU. */
1717         retval = sched_fork(clone_flags, p);
1718         if (retval)
1719                 goto bad_fork_cleanup_policy;
1720
1721         retval = perf_event_init_task(p);
1722         if (retval)
1723                 goto bad_fork_cleanup_policy;
1724         retval = audit_alloc(p);
1725         if (retval)
1726                 goto bad_fork_cleanup_perf;
1727         /* copy all the process information */
1728         shm_init_task(p);
1729         retval = security_task_alloc(p, clone_flags);
1730         if (retval)
1731                 goto bad_fork_cleanup_audit;
1732         retval = copy_semundo(clone_flags, p);
1733         if (retval)
1734                 goto bad_fork_cleanup_security;
1735         retval = copy_files(clone_flags, p);
1736         if (retval)
1737                 goto bad_fork_cleanup_semundo;
1738         retval = copy_fs(clone_flags, p);
1739         if (retval)
1740                 goto bad_fork_cleanup_files;
1741         retval = copy_sighand(clone_flags, p);
1742         if (retval)
1743                 goto bad_fork_cleanup_fs;
1744         retval = copy_signal(clone_flags, p);
1745         if (retval)
1746                 goto bad_fork_cleanup_sighand;
1747         retval = copy_mm(clone_flags, p);
1748         if (retval)
1749                 goto bad_fork_cleanup_signal;
1750         retval = copy_namespaces(clone_flags, p);
1751         if (retval)
1752                 goto bad_fork_cleanup_mm;
1753         retval = copy_io(clone_flags, p);
1754         if (retval)
1755                 goto bad_fork_cleanup_namespaces;
1756         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1757         if (retval)
1758                 goto bad_fork_cleanup_io;
1759
1760         if (pid != &init_struct_pid) {
1761                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1762                 if (IS_ERR(pid)) {
1763                         retval = PTR_ERR(pid);
1764                         goto bad_fork_cleanup_thread;
1765                 }
1766         }
1767
1768 #ifdef CONFIG_BLOCK
1769         p->plug = NULL;
1770 #endif
1771 #ifdef CONFIG_FUTEX
1772         p->robust_list = NULL;
1773 #ifdef CONFIG_COMPAT
1774         p->compat_robust_list = NULL;
1775 #endif
1776         INIT_LIST_HEAD(&p->pi_state_list);
1777         p->pi_state_cache = NULL;
1778 #endif
1779         /*
1780          * sigaltstack should be cleared when sharing the same VM
1781          */
1782         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1783                 sas_ss_reset(p);
1784
1785         /*
1786          * Syscall tracing and stepping should be turned off in the
1787          * child regardless of CLONE_PTRACE.
1788          */
1789         user_disable_single_step(p);
1790         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1791 #ifdef TIF_SYSCALL_EMU
1792         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1793 #endif
1794         clear_all_latency_tracing(p);
1795
1796         /* ok, now we should be set up.. */
1797         p->pid = pid_nr(pid);
1798         if (clone_flags & CLONE_THREAD) {
1799                 p->exit_signal = -1;
1800                 p->group_leader = current->group_leader;
1801                 p->tgid = current->tgid;
1802         } else {
1803                 if (clone_flags & CLONE_PARENT)
1804                         p->exit_signal = current->group_leader->exit_signal;
1805                 else
1806                         p->exit_signal = (clone_flags & CSIGNAL);
1807                 p->group_leader = p;
1808                 p->tgid = p->pid;
1809         }
1810
1811         p->nr_dirtied = 0;
1812         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1813         p->dirty_paused_when = 0;
1814
1815         p->pdeath_signal = 0;
1816         INIT_LIST_HEAD(&p->thread_group);
1817         p->task_works = NULL;
1818
1819         cgroup_threadgroup_change_begin(current);
1820         /*
1821          * Ensure that the cgroup subsystem policies allow the new process to be
1822          * forked. It should be noted the the new process's css_set can be changed
1823          * between here and cgroup_post_fork() if an organisation operation is in
1824          * progress.
1825          */
1826         retval = cgroup_can_fork(p);
1827         if (retval)
1828                 goto bad_fork_free_pid;
1829
1830         /*
1831          * Make it visible to the rest of the system, but dont wake it up yet.
1832          * Need tasklist lock for parent etc handling!
1833          */
1834         write_lock_irq(&tasklist_lock);
1835
1836         /* CLONE_PARENT re-uses the old parent */
1837         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1838                 p->real_parent = current->real_parent;
1839                 p->parent_exec_id = current->parent_exec_id;
1840         } else {
1841                 p->real_parent = current;
1842                 p->parent_exec_id = current->self_exec_id;
1843         }
1844
1845         klp_copy_process(p);
1846
1847         spin_lock(&current->sighand->siglock);
1848
1849         /*
1850          * Copy seccomp details explicitly here, in case they were changed
1851          * before holding sighand lock.
1852          */
1853         copy_seccomp(p);
1854
1855         /*
1856          * Process group and session signals need to be delivered to just the
1857          * parent before the fork or both the parent and the child after the
1858          * fork. Restart if a signal comes in before we add the new process to
1859          * it's process group.
1860          * A fatal signal pending means that current will exit, so the new
1861          * thread can't slip out of an OOM kill (or normal SIGKILL).
1862         */
1863         recalc_sigpending();
1864         if (signal_pending(current)) {
1865                 retval = -ERESTARTNOINTR;
1866                 goto bad_fork_cancel_cgroup;
1867         }
1868         if (unlikely(!(ns_of_pid(pid)->nr_hashed & PIDNS_HASH_ADDING))) {
1869                 retval = -ENOMEM;
1870                 goto bad_fork_cancel_cgroup;
1871         }
1872
1873         if (likely(p->pid)) {
1874                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1875
1876                 init_task_pid(p, PIDTYPE_PID, pid);
1877                 if (thread_group_leader(p)) {
1878                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1879                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1880
1881                         if (is_child_reaper(pid)) {
1882                                 ns_of_pid(pid)->child_reaper = p;
1883                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1884                         }
1885
1886                         p->signal->leader_pid = pid;
1887                         p->signal->tty = tty_kref_get(current->signal->tty);
1888                         /*
1889                          * Inherit has_child_subreaper flag under the same
1890                          * tasklist_lock with adding child to the process tree
1891                          * for propagate_has_child_subreaper optimization.
1892                          */
1893                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1894                                                          p->real_parent->signal->is_child_subreaper;
1895                         list_add_tail(&p->sibling, &p->real_parent->children);
1896                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1897                         attach_pid(p, PIDTYPE_PGID);
1898                         attach_pid(p, PIDTYPE_SID);
1899                         __this_cpu_inc(process_counts);
1900                 } else {
1901                         current->signal->nr_threads++;
1902                         atomic_inc(&current->signal->live);
1903                         atomic_inc(&current->signal->sigcnt);
1904                         list_add_tail_rcu(&p->thread_group,
1905                                           &p->group_leader->thread_group);
1906                         list_add_tail_rcu(&p->thread_node,
1907                                           &p->signal->thread_head);
1908                 }
1909                 attach_pid(p, PIDTYPE_PID);
1910                 nr_threads++;
1911         }
1912
1913         total_forks++;
1914         spin_unlock(&current->sighand->siglock);
1915         syscall_tracepoint_update(p);
1916         write_unlock_irq(&tasklist_lock);
1917
1918         proc_fork_connector(p);
1919         cgroup_post_fork(p);
1920         cgroup_threadgroup_change_end(current);
1921         perf_event_fork(p);
1922
1923         trace_task_newtask(p, clone_flags);
1924         uprobe_copy_process(p, clone_flags);
1925
1926         return p;
1927
1928 bad_fork_cancel_cgroup:
1929         spin_unlock(&current->sighand->siglock);
1930         write_unlock_irq(&tasklist_lock);
1931         cgroup_cancel_fork(p);
1932 bad_fork_free_pid:
1933         cgroup_threadgroup_change_end(current);
1934         if (pid != &init_struct_pid)
1935                 free_pid(pid);
1936 bad_fork_cleanup_thread:
1937         exit_thread(p);
1938 bad_fork_cleanup_io:
1939         if (p->io_context)
1940                 exit_io_context(p);
1941 bad_fork_cleanup_namespaces:
1942         exit_task_namespaces(p);
1943 bad_fork_cleanup_mm:
1944         if (p->mm)
1945                 mmput(p->mm);
1946 bad_fork_cleanup_signal:
1947         if (!(clone_flags & CLONE_THREAD))
1948                 free_signal_struct(p->signal);
1949 bad_fork_cleanup_sighand:
1950         __cleanup_sighand(p->sighand);
1951 bad_fork_cleanup_fs:
1952         exit_fs(p); /* blocking */
1953 bad_fork_cleanup_files:
1954         exit_files(p); /* blocking */
1955 bad_fork_cleanup_semundo:
1956         exit_sem(p);
1957 bad_fork_cleanup_security:
1958         security_task_free(p);
1959 bad_fork_cleanup_audit:
1960         audit_free(p);
1961 bad_fork_cleanup_perf:
1962         perf_event_free_task(p);
1963 bad_fork_cleanup_policy:
1964         lockdep_free_task(p);
1965 #ifdef CONFIG_NUMA
1966         mpol_put(p->mempolicy);
1967 bad_fork_cleanup_threadgroup_lock:
1968 #endif
1969         delayacct_tsk_free(p);
1970 bad_fork_cleanup_count:
1971         atomic_dec(&p->cred->user->processes);
1972         exit_creds(p);
1973 bad_fork_free:
1974         p->state = TASK_DEAD;
1975         put_task_stack(p);
1976         free_task(p);
1977 fork_out:
1978         return ERR_PTR(retval);
1979 }
1980
1981 static inline void init_idle_pids(struct pid_link *links)
1982 {
1983         enum pid_type type;
1984
1985         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1986                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1987                 links[type].pid = &init_struct_pid;
1988         }
1989 }
1990
1991 struct task_struct *fork_idle(int cpu)
1992 {
1993         struct task_struct *task;
1994         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1995                             cpu_to_node(cpu));
1996         if (!IS_ERR(task)) {
1997                 init_idle_pids(task->pids);
1998                 init_idle(task, cpu);
1999         }
2000
2001         return task;
2002 }
2003
2004 /*
2005  *  Ok, this is the main fork-routine.
2006  *
2007  * It copies the process, and if successful kick-starts
2008  * it and waits for it to finish using the VM if required.
2009  */
2010 long _do_fork(unsigned long clone_flags,
2011               unsigned long stack_start,
2012               unsigned long stack_size,
2013               int __user *parent_tidptr,
2014               int __user *child_tidptr,
2015               unsigned long tls)
2016 {
2017         struct task_struct *p;
2018         int trace = 0;
2019         long nr;
2020
2021         /*
2022          * Determine whether and which event to report to ptracer.  When
2023          * called from kernel_thread or CLONE_UNTRACED is explicitly
2024          * requested, no event is reported; otherwise, report if the event
2025          * for the type of forking is enabled.
2026          */
2027         if (!(clone_flags & CLONE_UNTRACED)) {
2028                 if (clone_flags & CLONE_VFORK)
2029                         trace = PTRACE_EVENT_VFORK;
2030                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2031                         trace = PTRACE_EVENT_CLONE;
2032                 else
2033                         trace = PTRACE_EVENT_FORK;
2034
2035                 if (likely(!ptrace_event_enabled(current, trace)))
2036                         trace = 0;
2037         }
2038
2039         p = copy_process(clone_flags, stack_start, stack_size,
2040                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2041         add_latent_entropy();
2042         /*
2043          * Do this prior waking up the new thread - the thread pointer
2044          * might get invalid after that point, if the thread exits quickly.
2045          */
2046         if (!IS_ERR(p)) {
2047                 struct completion vfork;
2048                 struct pid *pid;
2049
2050                 trace_sched_process_fork(current, p);
2051
2052                 pid = get_task_pid(p, PIDTYPE_PID);
2053                 nr = pid_vnr(pid);
2054
2055                 if (clone_flags & CLONE_PARENT_SETTID)
2056                         put_user(nr, parent_tidptr);
2057
2058                 if (clone_flags & CLONE_VFORK) {
2059                         p->vfork_done = &vfork;
2060                         init_completion(&vfork);
2061                         get_task_struct(p);
2062                 }
2063
2064                 wake_up_new_task(p);
2065
2066                 /* forking complete and child started to run, tell ptracer */
2067                 if (unlikely(trace))
2068                         ptrace_event_pid(trace, pid);
2069
2070                 if (clone_flags & CLONE_VFORK) {
2071                         if (!wait_for_vfork_done(p, &vfork))
2072                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2073                 }
2074
2075                 put_pid(pid);
2076         } else {
2077                 nr = PTR_ERR(p);
2078         }
2079         return nr;
2080 }
2081
2082 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2083 /* For compatibility with architectures that call do_fork directly rather than
2084  * using the syscall entry points below. */
2085 long do_fork(unsigned long clone_flags,
2086               unsigned long stack_start,
2087               unsigned long stack_size,
2088               int __user *parent_tidptr,
2089               int __user *child_tidptr)
2090 {
2091         return _do_fork(clone_flags, stack_start, stack_size,
2092                         parent_tidptr, child_tidptr, 0);
2093 }
2094 #endif
2095
2096 /*
2097  * Create a kernel thread.
2098  */
2099 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2100 {
2101         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2102                 (unsigned long)arg, NULL, NULL, 0);
2103 }
2104
2105 #ifdef __ARCH_WANT_SYS_FORK
2106 SYSCALL_DEFINE0(fork)
2107 {
2108 #ifdef CONFIG_MMU
2109         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2110 #else
2111         /* can not support in nommu mode */
2112         return -EINVAL;
2113 #endif
2114 }
2115 #endif
2116
2117 #ifdef __ARCH_WANT_SYS_VFORK
2118 SYSCALL_DEFINE0(vfork)
2119 {
2120         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2121                         0, NULL, NULL, 0);
2122 }
2123 #endif
2124
2125 #ifdef __ARCH_WANT_SYS_CLONE
2126 #ifdef CONFIG_CLONE_BACKWARDS
2127 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2128                  int __user *, parent_tidptr,
2129                  unsigned long, tls,
2130                  int __user *, child_tidptr)
2131 #elif defined(CONFIG_CLONE_BACKWARDS2)
2132 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2133                  int __user *, parent_tidptr,
2134                  int __user *, child_tidptr,
2135                  unsigned long, tls)
2136 #elif defined(CONFIG_CLONE_BACKWARDS3)
2137 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2138                 int, stack_size,
2139                 int __user *, parent_tidptr,
2140                 int __user *, child_tidptr,
2141                 unsigned long, tls)
2142 #else
2143 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2144                  int __user *, parent_tidptr,
2145                  int __user *, child_tidptr,
2146                  unsigned long, tls)
2147 #endif
2148 {
2149         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2150 }
2151 #endif
2152
2153 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2154 {
2155         struct task_struct *leader, *parent, *child;
2156         int res;
2157
2158         read_lock(&tasklist_lock);
2159         leader = top = top->group_leader;
2160 down:
2161         for_each_thread(leader, parent) {
2162                 list_for_each_entry(child, &parent->children, sibling) {
2163                         res = visitor(child, data);
2164                         if (res) {
2165                                 if (res < 0)
2166                                         goto out;
2167                                 leader = child;
2168                                 goto down;
2169                         }
2170 up:
2171                         ;
2172                 }
2173         }
2174
2175         if (leader != top) {
2176                 child = leader;
2177                 parent = child->real_parent;
2178                 leader = parent->group_leader;
2179                 goto up;
2180         }
2181 out:
2182         read_unlock(&tasklist_lock);
2183 }
2184
2185 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2186 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2187 #endif
2188
2189 static void sighand_ctor(void *data)
2190 {
2191         struct sighand_struct *sighand = data;
2192
2193         spin_lock_init(&sighand->siglock);
2194         init_waitqueue_head(&sighand->signalfd_wqh);
2195 }
2196
2197 void __init proc_caches_init(void)
2198 {
2199         sighand_cachep = kmem_cache_create("sighand_cache",
2200                         sizeof(struct sighand_struct), 0,
2201                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2202                         SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2203         signal_cachep = kmem_cache_create("signal_cache",
2204                         sizeof(struct signal_struct), 0,
2205                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2206                         NULL);
2207         files_cachep = kmem_cache_create("files_cache",
2208                         sizeof(struct files_struct), 0,
2209                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2210                         NULL);
2211         fs_cachep = kmem_cache_create("fs_cache",
2212                         sizeof(struct fs_struct), 0,
2213                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2214                         NULL);
2215         /*
2216          * FIXME! The "sizeof(struct mm_struct)" currently includes the
2217          * whole struct cpumask for the OFFSTACK case. We could change
2218          * this to *only* allocate as much of it as required by the
2219          * maximum number of CPU's we can ever have.  The cpumask_allocation
2220          * is at the end of the structure, exactly for that reason.
2221          */
2222         mm_cachep = kmem_cache_create("mm_struct",
2223                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2224                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2225                         NULL);
2226         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2227         mmap_init();
2228         nsproxy_cache_init();
2229 }
2230
2231 /*
2232  * Check constraints on flags passed to the unshare system call.
2233  */
2234 static int check_unshare_flags(unsigned long unshare_flags)
2235 {
2236         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2237                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2238                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2239                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2240                 return -EINVAL;
2241         /*
2242          * Not implemented, but pretend it works if there is nothing
2243          * to unshare.  Note that unsharing the address space or the
2244          * signal handlers also need to unshare the signal queues (aka
2245          * CLONE_THREAD).
2246          */
2247         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2248                 if (!thread_group_empty(current))
2249                         return -EINVAL;
2250         }
2251         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2252                 if (atomic_read(&current->sighand->count) > 1)
2253                         return -EINVAL;
2254         }
2255         if (unshare_flags & CLONE_VM) {
2256                 if (!current_is_single_threaded())
2257                         return -EINVAL;
2258         }
2259
2260         return 0;
2261 }
2262
2263 /*
2264  * Unshare the filesystem structure if it is being shared
2265  */
2266 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2267 {
2268         struct fs_struct *fs = current->fs;
2269
2270         if (!(unshare_flags & CLONE_FS) || !fs)
2271                 return 0;
2272
2273         /* don't need lock here; in the worst case we'll do useless copy */
2274         if (fs->users == 1)
2275                 return 0;
2276
2277         *new_fsp = copy_fs_struct(fs);
2278         if (!*new_fsp)
2279                 return -ENOMEM;
2280
2281         return 0;
2282 }
2283
2284 /*
2285  * Unshare file descriptor table if it is being shared
2286  */
2287 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2288 {
2289         struct files_struct *fd = current->files;
2290         int error = 0;
2291
2292         if ((unshare_flags & CLONE_FILES) &&
2293             (fd && atomic_read(&fd->count) > 1)) {
2294                 *new_fdp = dup_fd(fd, &error);
2295                 if (!*new_fdp)
2296                         return error;
2297         }
2298
2299         return 0;
2300 }
2301
2302 /*
2303  * unshare allows a process to 'unshare' part of the process
2304  * context which was originally shared using clone.  copy_*
2305  * functions used by do_fork() cannot be used here directly
2306  * because they modify an inactive task_struct that is being
2307  * constructed. Here we are modifying the current, active,
2308  * task_struct.
2309  */
2310 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2311 {
2312         struct fs_struct *fs, *new_fs = NULL;
2313         struct files_struct *fd, *new_fd = NULL;
2314         struct cred *new_cred = NULL;
2315         struct nsproxy *new_nsproxy = NULL;
2316         int do_sysvsem = 0;
2317         int err;
2318
2319         /*
2320          * If unsharing a user namespace must also unshare the thread group
2321          * and unshare the filesystem root and working directories.
2322          */
2323         if (unshare_flags & CLONE_NEWUSER)
2324                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2325         /*
2326          * If unsharing vm, must also unshare signal handlers.
2327          */
2328         if (unshare_flags & CLONE_VM)
2329                 unshare_flags |= CLONE_SIGHAND;
2330         /*
2331          * If unsharing a signal handlers, must also unshare the signal queues.
2332          */
2333         if (unshare_flags & CLONE_SIGHAND)
2334                 unshare_flags |= CLONE_THREAD;
2335         /*
2336          * If unsharing namespace, must also unshare filesystem information.
2337          */
2338         if (unshare_flags & CLONE_NEWNS)
2339                 unshare_flags |= CLONE_FS;
2340
2341         err = check_unshare_flags(unshare_flags);
2342         if (err)
2343                 goto bad_unshare_out;
2344         /*
2345          * CLONE_NEWIPC must also detach from the undolist: after switching
2346          * to a new ipc namespace, the semaphore arrays from the old
2347          * namespace are unreachable.
2348          */
2349         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2350                 do_sysvsem = 1;
2351         err = unshare_fs(unshare_flags, &new_fs);
2352         if (err)
2353                 goto bad_unshare_out;
2354         err = unshare_fd(unshare_flags, &new_fd);
2355         if (err)
2356                 goto bad_unshare_cleanup_fs;
2357         err = unshare_userns(unshare_flags, &new_cred);
2358         if (err)
2359                 goto bad_unshare_cleanup_fd;
2360         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2361                                          new_cred, new_fs);
2362         if (err)
2363                 goto bad_unshare_cleanup_cred;
2364
2365         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2366                 if (do_sysvsem) {
2367                         /*
2368                          * CLONE_SYSVSEM is equivalent to sys_exit().
2369                          */
2370                         exit_sem(current);
2371                 }
2372                 if (unshare_flags & CLONE_NEWIPC) {
2373                         /* Orphan segments in old ns (see sem above). */
2374                         exit_shm(current);
2375                         shm_init_task(current);
2376                 }
2377
2378                 if (new_nsproxy)
2379                         switch_task_namespaces(current, new_nsproxy);
2380
2381                 task_lock(current);
2382
2383                 if (new_fs) {
2384                         fs = current->fs;
2385                         spin_lock(&fs->lock);
2386                         current->fs = new_fs;
2387                         if (--fs->users)
2388                                 new_fs = NULL;
2389                         else
2390                                 new_fs = fs;
2391                         spin_unlock(&fs->lock);
2392                 }
2393
2394                 if (new_fd) {
2395                         fd = current->files;
2396                         current->files = new_fd;
2397                         new_fd = fd;
2398                 }
2399
2400                 task_unlock(current);
2401
2402                 if (new_cred) {
2403                         /* Install the new user namespace */
2404                         commit_creds(new_cred);
2405                         new_cred = NULL;
2406                 }
2407         }
2408
2409         perf_event_namespaces(current);
2410
2411 bad_unshare_cleanup_cred:
2412         if (new_cred)
2413                 put_cred(new_cred);
2414 bad_unshare_cleanup_fd:
2415         if (new_fd)
2416                 put_files_struct(new_fd);
2417
2418 bad_unshare_cleanup_fs:
2419         if (new_fs)
2420                 free_fs_struct(new_fs);
2421
2422 bad_unshare_out:
2423         return err;
2424 }
2425
2426 /*
2427  *      Helper to unshare the files of the current task.
2428  *      We don't want to expose copy_files internals to
2429  *      the exec layer of the kernel.
2430  */
2431
2432 int unshare_files(struct files_struct **displaced)
2433 {
2434         struct task_struct *task = current;
2435         struct files_struct *copy = NULL;
2436         int error;
2437
2438         error = unshare_fd(CLONE_FILES, &copy);
2439         if (error || !copy) {
2440                 *displaced = NULL;
2441                 return error;
2442         }
2443         *displaced = task->files;
2444         task_lock(task);
2445         task->files = copy;
2446         task_unlock(task);
2447         return 0;
2448 }
2449
2450 int sysctl_max_threads(struct ctl_table *table, int write,
2451                        void __user *buffer, size_t *lenp, loff_t *ppos)
2452 {
2453         struct ctl_table t;
2454         int ret;
2455         int threads = max_threads;
2456         int min = MIN_THREADS;
2457         int max = MAX_THREADS;
2458
2459         t = *table;
2460         t.data = &threads;
2461         t.extra1 = &min;
2462         t.extra2 = &max;
2463
2464         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2465         if (ret || !write)
2466                 return ret;
2467
2468         set_max_threads(threads);
2469
2470         return 0;
2471 }