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/hmm.h>
  41 #include <linux/fs.h>
  42 #include <linux/mm.h>
  43 #include <linux/vmacache.h>
  44 #include <linux/nsproxy.h>
  45 #include <linux/capability.h>
  46 #include <linux/cpu.h>
  47 #include <linux/cgroup.h>
  48 #include <linux/security.h>
  49 #include <linux/hugetlb.h>
  50 #include <linux/seccomp.h>
  51 #include <linux/swap.h>
  52 #include <linux/syscalls.h>
  53 #include <linux/jiffies.h>
  54 #include <linux/futex.h>
  55 #include <linux/compat.h>
  56 #include <linux/kthread.h>
  57 #include <linux/task_io_accounting_ops.h>
  58 #include <linux/rcupdate.h>
  59 #include <linux/ptrace.h>
  60 #include <linux/mount.h>
  61 #include <linux/audit.h>
  62 #include <linux/memcontrol.h>
  63 #include <linux/ftrace.h>
  64 #include <linux/proc_fs.h>
  65 #include <linux/profile.h>
  66 #include <linux/rmap.h>
  67 #include <linux/ksm.h>
  68 #include <linux/acct.h>
  69 #include <linux/userfaultfd_k.h>
  70 #include <linux/tsacct_kern.h>
  71 #include <linux/cn_proc.h>
  72 #include <linux/freezer.h>
  73 #include <linux/delayacct.h>
  74 #include <linux/taskstats_kern.h>
  75 #include <linux/random.h>
  76 #include <linux/tty.h>
  77 #include <linux/blkdev.h>
  78 #include <linux/fs_struct.h>
  79 #include <linux/magic.h>
  80 #include <linux/perf_event.h>
  81 #include <linux/posix-timers.h>
  82 #include <linux/user-return-notifier.h>
  83 #include <linux/oom.h>
  84 #include <linux/khugepaged.h>
  85 #include <linux/signalfd.h>
  86 #include <linux/uprobes.h>
  87 #include <linux/aio.h>
  88 #include <linux/compiler.h>
  89 #include <linux/sysctl.h>
  90 #include <linux/kcov.h>
  91 #include <linux/livepatch.h>
  92 #include <linux/thread_info.h>
  93
  94 #include <asm/pgtable.h>
  95 #include <asm/pgalloc.h>
  96 #include <linux/uaccess.h>
  97 #include <asm/mmu_context.h>
  98 #include <asm/cacheflush.h>
  99 #include <asm/tlbflush.h>
 100
 101 #include <trace/events/sched.h>
 102
 103 #define CREATE_TRACE_POINTS
 104 #include <trace/events/task.h>
 105
 106 /*
 107  * Minimum number of threads to boot the kernel
 108  */
 109 #define MIN_THREADS 20
 110
 111 /*
 112  * Maximum number of threads
 113  */
 114 #define MAX_THREADS FUTEX_TID_MASK
 115
 116 /*
 117  * Protected counters by write_lock_irq(&tasklist_lock)
 118  */
 119 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 120 int nr_threads;                 /* The idle threads do not count.. */
 121
 122 int max_threads;                /* tunable limit on nr_threads */
 123
 124 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 125
 126 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 127
 128 #ifdef CONFIG_PROVE_RCU
 129 int lockdep_tasklist_lock_is_held(void)
 130 {
 131         return lockdep_is_held(&tasklist_lock);
 132 }
 133 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 134 #endif /* #ifdef CONFIG_PROVE_RCU */
 135
 136 int nr_processes(void)
 137 {
 138         int cpu;
 139         int total = 0;
 140
 141         for_each_possible_cpu(cpu)
 142                 total += per_cpu(process_counts, cpu);
 143
 144         return total;
 145 }
 146
 147 void __weak arch_release_task_struct(struct task_struct *tsk)
 148 {
 149 }
 150
 151 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 152 static struct kmem_cache *task_struct_cachep;
 153
 154 static inline struct task_struct *alloc_task_struct_node(int node)
 155 {
 156         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 157 }
 158
 159 static inline void free_task_struct(struct task_struct *tsk)
 160 {
 161         kmem_cache_free(task_struct_cachep, tsk);
 162 }
 163 #endif
 164
 165 void __weak arch_release_thread_stack(unsigned long *stack)
 166 {
 167 }
 168
 169 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 170
 171 /*
 172  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 173  * kmemcache based allocator.
 174  */
 175 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 176
 177 #ifdef CONFIG_VMAP_STACK
 178 /*
 179  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 180  * flush.  Try to minimize the number of calls by caching stacks.
 181  */
 182 #define NR_CACHED_STACKS 2
 183 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 184
 185 static int free_vm_stack_cache(unsigned int cpu)
 186 {
 187         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 188         int i;
 189
 190         for (i = 0; i < NR_CACHED_STACKS; i++) {
 191                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 192
 193                 if (!vm_stack)
 194                         continue;
 195
 196                 vfree(vm_stack->addr);
 197                 cached_vm_stacks[i] = NULL;
 198         }
 199
 200         return 0;
 201 }
 202 #endif
 203
 204 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 205 {
 206 #ifdef CONFIG_VMAP_STACK
 207         void *stack;
 208         int i;
 209
 210         for (i = 0; i < NR_CACHED_STACKS; i++) {
 211                 struct vm_struct *s;
 212
 213                 s = this_cpu_xchg(cached_stacks[i], NULL);
 214
 215                 if (!s)
 216                         continue;
 217
 218                 tsk->stack_vm_area = s;
 219                 return s->addr;
 220         }
 221
 222         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 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                 int i;
 249
 250                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 251                         if (this_cpu_cmpxchg(cached_stacks[i],
 252                                         NULL, tsk->stack_vm_area) != NULL)
 253                                 continue;
 254
 255                         return;
 256                 }
 257
 258                 vfree_atomic(tsk->stack);
 259                 return;
 260         }
 261 #endif
 262
 263         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 264 }
 265 # else
 266 static struct kmem_cache *thread_stack_cache;
 267
 268 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 269                                                   int node)
 270 {
 271         return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 272 }
 273
 274 static void free_thread_stack(struct task_struct *tsk)
 275 {
 276         kmem_cache_free(thread_stack_cache, tsk->stack);
 277 }
 278
 279 void thread_stack_cache_init(void)
 280 {
 281         thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
 282                                               THREAD_SIZE, 0, NULL);
 283         BUG_ON(thread_stack_cache == NULL);
 284 }
 285 # endif
 286 #endif
 287
 288 /* SLAB cache for signal_struct structures (tsk->signal) */
 289 static struct kmem_cache *signal_cachep;
 290
 291 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 292 struct kmem_cache *sighand_cachep;
 293
 294 /* SLAB cache for files_struct structures (tsk->files) */
 295 struct kmem_cache *files_cachep;
 296
 297 /* SLAB cache for fs_struct structures (tsk->fs) */
 298 struct kmem_cache *fs_cachep;
 299
 300 /* SLAB cache for vm_area_struct structures */
 301 struct kmem_cache *vm_area_cachep;
 302
 303 /* SLAB cache for mm_struct structures (tsk->mm) */
 304 static struct kmem_cache *mm_cachep;
 305
 306 static void account_kernel_stack(struct task_struct *tsk, int account)
 307 {
 308         void *stack = task_stack_page(tsk);
 309         struct vm_struct *vm = task_stack_vm_area(tsk);
 310
 311         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 312
 313         if (vm) {
 314                 int i;
 315
 316                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 317
 318                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 319                         mod_zone_page_state(page_zone(vm->pages[i]),
 320                                             NR_KERNEL_STACK_KB,
 321                                             PAGE_SIZE / 1024 * account);
 322                 }
 323
 324                 /* All stack pages belong to the same memcg. */
 325                 mod_memcg_page_state(vm->pages[0], MEMCG_KERNEL_STACK_KB,
 326                                      account * (THREAD_SIZE / 1024));
 327         } else {
 328                 /*
 329                  * All stack pages are in the same zone and belong to the
 330                  * same memcg.
 331                  */
 332                 struct page *first_page = virt_to_page(stack);
 333
 334                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 335                                     THREAD_SIZE / 1024 * account);
 336
 337                 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
 338                                      account * (THREAD_SIZE / 1024));
 339         }
 340 }
 341
 342 static void release_task_stack(struct task_struct *tsk)
 343 {
 344         if (WARN_ON(tsk->state != TASK_DEAD))
 345                 return;  /* Better to leak the stack than to free prematurely */
 346
 347         account_kernel_stack(tsk, -1);
 348         arch_release_thread_stack(tsk->stack);
 349         free_thread_stack(tsk);
 350         tsk->stack = NULL;
 351 #ifdef CONFIG_VMAP_STACK
 352         tsk->stack_vm_area = NULL;
 353 #endif
 354 }
 355
 356 #ifdef CONFIG_THREAD_INFO_IN_TASK
 357 void put_task_stack(struct task_struct *tsk)
 358 {
 359         if (atomic_dec_and_test(&tsk->stack_refcount))
 360                 release_task_stack(tsk);
 361 }
 362 #endif
 363
 364 void free_task(struct task_struct *tsk)
 365 {
 366 #ifndef CONFIG_THREAD_INFO_IN_TASK
 367         /*
 368          * The task is finally done with both the stack and thread_info,
 369          * so free both.
 370          */
 371         release_task_stack(tsk);
 372 #else
 373         /*
 374          * If the task had a separate stack allocation, it should be gone
 375          * by now.
 376          */
 377         WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
 378 #endif
 379         rt_mutex_debug_task_free(tsk);
 380         ftrace_graph_exit_task(tsk);
 381         put_seccomp_filter(tsk);
 382         arch_release_task_struct(tsk);
 383         if (tsk->flags & PF_KTHREAD)
 384                 free_kthread_struct(tsk);
 385         free_task_struct(tsk);
 386 }
 387 EXPORT_SYMBOL(free_task);
 388
 389 static inline void free_signal_struct(struct signal_struct *sig)
 390 {
 391         taskstats_tgid_free(sig);
 392         sched_autogroup_exit(sig);
 393         /*
 394          * __mmdrop is not safe to call from softirq context on x86 due to
 395          * pgd_dtor so postpone it to the async context
 396          */
 397         if (sig->oom_mm)
 398                 mmdrop_async(sig->oom_mm);
 399         kmem_cache_free(signal_cachep, sig);
 400 }
 401
 402 static inline void put_signal_struct(struct signal_struct *sig)
 403 {
 404         if (atomic_dec_and_test(&sig->sigcnt))
 405                 free_signal_struct(sig);
 406 }
 407
 408 void __put_task_struct(struct task_struct *tsk)
 409 {
 410         WARN_ON(!tsk->exit_state);
 411         WARN_ON(atomic_read(&tsk->usage));
 412         WARN_ON(tsk == current);
 413
 414         cgroup_free(tsk);
 415         task_numa_free(tsk);
 416         security_task_free(tsk);
 417         exit_creds(tsk);
 418         delayacct_tsk_free(tsk);
 419         put_signal_struct(tsk->signal);
 420
 421         if (!profile_handoff_task(tsk))
 422                 free_task(tsk);
 423 }
 424 EXPORT_SYMBOL_GPL(__put_task_struct);
 425
 426 void __init __weak arch_task_cache_init(void) { }
 427
 428 /*
 429  * set_max_threads
 430  */
 431 static void set_max_threads(unsigned int max_threads_suggested)
 432 {
 433         u64 threads;
 434
 435         /*
 436          * The number of threads shall be limited such that the thread
 437          * structures may only consume a small part of the available memory.
 438          */
 439         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 440                 threads = MAX_THREADS;
 441         else
 442                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 443                                     (u64) THREAD_SIZE * 8UL);
 444
 445         if (threads > max_threads_suggested)
 446                 threads = max_threads_suggested;
 447
 448         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 449 }
 450
 451 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 452 /* Initialized by the architecture: */
 453 int arch_task_struct_size __read_mostly;
 454 #endif
 455
 456 void __init fork_init(void)
 457 {
 458         int i;
 459 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 460 #ifndef ARCH_MIN_TASKALIGN
 461 #define ARCH_MIN_TASKALIGN      0
 462 #endif
 463         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 464
 465         /* create a slab on which task_structs can be allocated */
 466         task_struct_cachep = kmem_cache_create("task_struct",
 467                         arch_task_struct_size, align,
 468                         SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
 469 #endif
 470
 471         /* do the arch specific task caches init */
 472         arch_task_cache_init();
 473
 474         set_max_threads(MAX_THREADS);
 475
 476         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 477         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 478         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 479                 init_task.signal->rlim[RLIMIT_NPROC];
 480
 481         for (i = 0; i < UCOUNT_COUNTS; i++) {
 482                 init_user_ns.ucount_max[i] = max_threads/2;
 483         }
 484
 485 #ifdef CONFIG_VMAP_STACK
 486         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 487                           NULL, free_vm_stack_cache);
 488 #endif
 489
 490         lockdep_init_task(&init_task);
 491 }
 492
 493 int __weak arch_dup_task_struct(struct task_struct *dst,
 494                                                struct task_struct *src)
 495 {
 496         *dst = *src;
 497         return 0;
 498 }
 499
 500 void set_task_stack_end_magic(struct task_struct *tsk)
 501 {
 502         unsigned long *stackend;
 503
 504         stackend = end_of_stack(tsk);
 505         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 506 }
 507
 508 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 509 {
 510         struct task_struct *tsk;
 511         unsigned long *stack;
 512         struct vm_struct *stack_vm_area;
 513         int err;
 514
 515         if (node == NUMA_NO_NODE)
 516                 node = tsk_fork_get_node(orig);
 517         tsk = alloc_task_struct_node(node);
 518         if (!tsk)
 519                 return NULL;
 520
 521         stack = alloc_thread_stack_node(tsk, node);
 522         if (!stack)
 523                 goto free_tsk;
 524
 525         stack_vm_area = task_stack_vm_area(tsk);
 526
 527         err = arch_dup_task_struct(tsk, orig);
 528
 529         /*
 530          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 531          * sure they're properly initialized before using any stack-related
 532          * functions again.
 533          */
 534         tsk->stack = stack;
 535 #ifdef CONFIG_VMAP_STACK
 536         tsk->stack_vm_area = stack_vm_area;
 537 #endif
 538 #ifdef CONFIG_THREAD_INFO_IN_TASK
 539         atomic_set(&tsk->stack_refcount, 1);
 540 #endif
 541
 542         if (err)
 543                 goto free_stack;
 544
 545 #ifdef CONFIG_SECCOMP
 546         /*
 547          * We must handle setting up seccomp filters once we're under
 548          * the sighand lock in case orig has changed between now and
 549          * then. Until then, filter must be NULL to avoid messing up
 550          * the usage counts on the error path calling free_task.
 551          */
 552         tsk->seccomp.filter = NULL;
 553 #endif
 554
 555         setup_thread_stack(tsk, orig);
 556         clear_user_return_notifier(tsk);
 557         clear_tsk_need_resched(tsk);
 558         set_task_stack_end_magic(tsk);
 559
 560 #ifdef CONFIG_CC_STACKPROTECTOR
 561         tsk->stack_canary = get_random_canary();
 562 #endif
 563
 564         /*
 565          * One for us, one for whoever does the "release_task()" (usually
 566          * parent)
 567          */
 568         atomic_set(&tsk->usage, 2);
 569 #ifdef CONFIG_BLK_DEV_IO_TRACE
 570         tsk->btrace_seq = 0;
 571 #endif
 572         tsk->splice_pipe = NULL;
 573         tsk->task_frag.page = NULL;
 574         tsk->wake_q.next = NULL;
 575
 576         account_kernel_stack(tsk, 1);
 577
 578         kcov_task_init(tsk);
 579
 580 #ifdef CONFIG_FAULT_INJECTION
 581         tsk->fail_nth = 0;
 582 #endif
 583
 584         return tsk;
 585
 586 free_stack:
 587         free_thread_stack(tsk);
 588 free_tsk:
 589         free_task_struct(tsk);
 590         return NULL;
 591 }
 592
 593 #ifdef CONFIG_MMU
 594 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 595                                         struct mm_struct *oldmm)
 596 {
 597         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 598         struct rb_node **rb_link, *rb_parent;
 599         int retval;
 600         unsigned long charge;
 601         LIST_HEAD(uf);
 602
 603         uprobe_start_dup_mmap();
 604         if (down_write_killable(&oldmm->mmap_sem)) {
 605                 retval = -EINTR;
 606                 goto fail_uprobe_end;
 607         }
 608         flush_cache_dup_mm(oldmm);
 609         uprobe_dup_mmap(oldmm, mm);
 610         /*
 611          * Not linked in yet - no deadlock potential:
 612          */
 613         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 614
 615         /* No ordering required: file already has been exposed. */
 616         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 617
 618         mm->total_vm = oldmm->total_vm;
 619         mm->data_vm = oldmm->data_vm;
 620         mm->exec_vm = oldmm->exec_vm;
 621         mm->stack_vm = oldmm->stack_vm;
 622
 623         rb_link = &mm->mm_rb.rb_node;
 624         rb_parent = NULL;
 625         pprev = &mm->mmap;
 626         retval = ksm_fork(mm, oldmm);
 627         if (retval)
 628                 goto out;
 629         retval = khugepaged_fork(mm, oldmm);
 630         if (retval)
 631                 goto out;
 632
 633         prev = NULL;
 634         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 635                 struct file *file;
 636
 637                 if (mpnt->vm_flags & VM_DONTCOPY) {
 638                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 639                         continue;
 640                 }
 641                 charge = 0;
 642                 if (mpnt->vm_flags & VM_ACCOUNT) {
 643                         unsigned long len = vma_pages(mpnt);
 644
 645                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 646                                 goto fail_nomem;
 647                         charge = len;
 648                 }
 649                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 650                 if (!tmp)
 651                         goto fail_nomem;
 652                 *tmp = *mpnt;
 653                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 654                 retval = vma_dup_policy(mpnt, tmp);
 655                 if (retval)
 656                         goto fail_nomem_policy;
 657                 tmp->vm_mm = mm;
 658                 retval = dup_userfaultfd(tmp, &uf);
 659                 if (retval)
 660                         goto fail_nomem_anon_vma_fork;
 661                 if (tmp->vm_flags & VM_WIPEONFORK) {
 662                         /* VM_WIPEONFORK gets a clean slate in the child. */
 663                         tmp->anon_vma = NULL;
 664                         if (anon_vma_prepare(tmp))
 665                                 goto fail_nomem_anon_vma_fork;
 666                 } else if (anon_vma_fork(tmp, mpnt))
 667                         goto fail_nomem_anon_vma_fork;
 668                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 669                 tmp->vm_next = tmp->vm_prev = NULL;
 670                 file = tmp->vm_file;
 671                 if (file) {
 672                         struct inode *inode = file_inode(file);
 673                         struct address_space *mapping = file->f_mapping;
 674
 675                         get_file(file);
 676                         if (tmp->vm_flags & VM_DENYWRITE)
 677                                 atomic_dec(&inode->i_writecount);
 678                         i_mmap_lock_write(mapping);
 679                         if (tmp->vm_flags & VM_SHARED)
 680                                 atomic_inc(&mapping->i_mmap_writable);
 681                         flush_dcache_mmap_lock(mapping);
 682                         /* insert tmp into the share list, just after mpnt */
 683                         vma_interval_tree_insert_after(tmp, mpnt,
 684                                         &mapping->i_mmap);
 685                         flush_dcache_mmap_unlock(mapping);
 686                         i_mmap_unlock_write(mapping);
 687                 }
 688
 689                 /*
 690                  * Clear hugetlb-related page reserves for children. This only
 691                  * affects MAP_PRIVATE mappings. Faults generated by the child
 692                  * are not guaranteed to succeed, even if read-only
 693                  */
 694                 if (is_vm_hugetlb_page(tmp))
 695                         reset_vma_resv_huge_pages(tmp);
 696
 697                 /*
 698                  * Link in the new vma and copy the page table entries.
 699                  */
 700                 *pprev = tmp;
 701                 pprev = &tmp->vm_next;
 702                 tmp->vm_prev = prev;
 703                 prev = tmp;
 704
 705                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 706                 rb_link = &tmp->vm_rb.rb_right;
 707                 rb_parent = &tmp->vm_rb;
 708
 709                 mm->map_count++;
 710                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 711                         retval = copy_page_range(mm, oldmm, mpnt);
 712
 713                 if (tmp->vm_ops && tmp->vm_ops->open)
 714                         tmp->vm_ops->open(tmp);
 715
 716                 if (retval)
 717                         goto out;
 718         }
 719         /* a new mm has just been created */
 720         arch_dup_mmap(oldmm, mm);
 721         retval = 0;
 722 out:
 723         up_write(&mm->mmap_sem);
 724         flush_tlb_mm(oldmm);
 725         up_write(&oldmm->mmap_sem);
 726         dup_userfaultfd_complete(&uf);
 727 fail_uprobe_end:
 728         uprobe_end_dup_mmap();
 729         return retval;
 730 fail_nomem_anon_vma_fork:
 731         mpol_put(vma_policy(tmp));
 732 fail_nomem_policy:
 733         kmem_cache_free(vm_area_cachep, tmp);
 734 fail_nomem:
 735         retval = -ENOMEM;
 736         vm_unacct_memory(charge);
 737         goto out;
 738 }
 739
 740 static inline int mm_alloc_pgd(struct mm_struct *mm)
 741 {
 742         mm->pgd = pgd_alloc(mm);
 743         if (unlikely(!mm->pgd))
 744                 return -ENOMEM;
 745         return 0;
 746 }
 747
 748 static inline void mm_free_pgd(struct mm_struct *mm)
 749 {
 750         pgd_free(mm, mm->pgd);
 751 }
 752 #else
 753 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 754 {
 755         down_write(&oldmm->mmap_sem);
 756         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 757         up_write(&oldmm->mmap_sem);
 758         return 0;
 759 }
 760 #define mm_alloc_pgd(mm)        (0)
 761 #define mm_free_pgd(mm)
 762 #endif /* CONFIG_MMU */
 763
 764 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 765
 766 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 767 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 768
 769 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 770
 771 static int __init coredump_filter_setup(char *s)
 772 {
 773         default_dump_filter =
 774                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 775                 MMF_DUMP_FILTER_MASK;
 776         return 1;
 777 }
 778
 779 __setup("coredump_filter=", coredump_filter_setup);
 780
 781 #include <linux/init_task.h>
 782
 783 static void mm_init_aio(struct mm_struct *mm)
 784 {
 785 #ifdef CONFIG_AIO
 786         spin_lock_init(&mm->ioctx_lock);
 787         mm->ioctx_table = NULL;
 788 #endif
 789 }
 790
 791 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 792 {
 793 #ifdef CONFIG_MEMCG
 794         mm->owner = p;
 795 #endif
 796 }
 797
 798 static void mm_init_uprobes_state(struct mm_struct *mm)
 799 {
 800 #ifdef CONFIG_UPROBES
 801         mm->uprobes_state.xol_area = NULL;
 802 #endif
 803 }
 804
 805 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 806         struct user_namespace *user_ns)
 807 {
 808         mm->mmap = NULL;
 809         mm->mm_rb = RB_ROOT;
 810         mm->vmacache_seqnum = 0;
 811         atomic_set(&mm->mm_users, 1);
 812         atomic_set(&mm->mm_count, 1);
 813         init_rwsem(&mm->mmap_sem);
 814         INIT_LIST_HEAD(&mm->mmlist);
 815         mm->core_state = NULL;
 816         atomic_long_set(&mm->nr_ptes, 0);
 817         mm_nr_pmds_init(mm);
 818         mm->map_count = 0;
 819         mm->locked_vm = 0;
 820         mm->pinned_vm = 0;
 821         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 822         spin_lock_init(&mm->page_table_lock);
 823         mm_init_cpumask(mm);
 824         mm_init_aio(mm);
 825         mm_init_owner(mm, p);
 826         RCU_INIT_POINTER(mm->exe_file, NULL);
 827         mmu_notifier_mm_init(mm);
 828         hmm_mm_init(mm);
 829         init_tlb_flush_pending(mm);
 830 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 831         mm->pmd_huge_pte = NULL;
 832 #endif
 833         mm_init_uprobes_state(mm);
 834
 835         if (current->mm) {
 836                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 837                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 838         } else {
 839                 mm->flags = default_dump_filter;
 840                 mm->def_flags = 0;
 841         }
 842
 843         if (mm_alloc_pgd(mm))
 844                 goto fail_nopgd;
 845
 846         if (init_new_context(p, mm))
 847                 goto fail_nocontext;
 848
 849         mm->user_ns = get_user_ns(user_ns);
 850         return mm;
 851
 852 fail_nocontext:
 853         mm_free_pgd(mm);
 854 fail_nopgd:
 855         free_mm(mm);
 856         return NULL;
 857 }
 858
 859 static void check_mm(struct mm_struct *mm)
 860 {
 861         int i;
 862
 863         for (i = 0; i < NR_MM_COUNTERS; i++) {
 864                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 865
 866                 if (unlikely(x))
 867                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 868                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 869         }
 870
 871         if (atomic_long_read(&mm->nr_ptes))
 872                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 873                                 atomic_long_read(&mm->nr_ptes));
 874         if (mm_nr_pmds(mm))
 875                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 876                                 mm_nr_pmds(mm));
 877
 878 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 879         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 880 #endif
 881 }
 882
 883 /*
 884  * Allocate and initialize an mm_struct.
 885  */
 886 struct mm_struct *mm_alloc(void)
 887 {
 888         struct mm_struct *mm;
 889
 890         mm = allocate_mm();
 891         if (!mm)
 892                 return NULL;
 893
 894         memset(mm, 0, sizeof(*mm));
 895         return mm_init(mm, current, current_user_ns());
 896 }
 897
 898 /*
 899  * Called when the last reference to the mm
 900  * is dropped: either by a lazy thread or by
 901  * mmput. Free the page directory and the mm.
 902  */
 903 void __mmdrop(struct mm_struct *mm)
 904 {
 905         BUG_ON(mm == &init_mm);
 906         mm_free_pgd(mm);
 907         destroy_context(mm);
 908         hmm_mm_destroy(mm);
 909         mmu_notifier_mm_destroy(mm);
 910         check_mm(mm);
 911         put_user_ns(mm->user_ns);
 912         free_mm(mm);
 913 }
 914 EXPORT_SYMBOL_GPL(__mmdrop);
 915
 916 static inline void __mmput(struct mm_struct *mm)
 917 {
 918         VM_BUG_ON(atomic_read(&mm->mm_users));
 919
 920         uprobe_clear_state(mm);
 921         exit_aio(mm);
 922         ksm_exit(mm);
 923         khugepaged_exit(mm); /* must run before exit_mmap */
 924         exit_mmap(mm);
 925         mm_put_huge_zero_page(mm);
 926         set_mm_exe_file(mm, NULL);
 927         if (!list_empty(&mm->mmlist)) {
 928                 spin_lock(&mmlist_lock);
 929                 list_del(&mm->mmlist);
 930                 spin_unlock(&mmlist_lock);
 931         }
 932         if (mm->binfmt)
 933                 module_put(mm->binfmt->module);
 934         mmdrop(mm);
 935 }
 936
 937 /*
 938  * Decrement the use count and release all resources for an mm.
 939  */
 940 void mmput(struct mm_struct *mm)
 941 {
 942         might_sleep();
 943
 944         if (atomic_dec_and_test(&mm->mm_users))
 945                 __mmput(mm);
 946 }
 947 EXPORT_SYMBOL_GPL(mmput);
 948
 949 /**
 950  * set_mm_exe_file - change a reference to the mm's executable file
 951  *
 952  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 953  *
 954  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 955  * invocations: in mmput() nobody alive left, in execve task is single
 956  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 957  * mm->exe_file, but does so without using set_mm_exe_file() in order
 958  * to do avoid the need for any locks.
 959  */
 960 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 961 {
 962         struct file *old_exe_file;
 963
 964         /*
 965          * It is safe to dereference the exe_file without RCU as
 966          * this function is only called if nobody else can access
 967          * this mm -- see comment above for justification.
 968          */
 969         old_exe_file = rcu_dereference_raw(mm->exe_file);
 970
 971         if (new_exe_file)
 972                 get_file(new_exe_file);
 973         rcu_assign_pointer(mm->exe_file, new_exe_file);
 974         if (old_exe_file)
 975                 fput(old_exe_file);
 976 }
 977
 978 /**
 979  * get_mm_exe_file - acquire a reference to the mm's executable file
 980  *
 981  * Returns %NULL if mm has no associated executable file.
 982  * User must release file via fput().
 983  */
 984 struct file *get_mm_exe_file(struct mm_struct *mm)
 985 {
 986         struct file *exe_file;
 987
 988         rcu_read_lock();
 989         exe_file = rcu_dereference(mm->exe_file);
 990         if (exe_file && !get_file_rcu(exe_file))
 991                 exe_file = NULL;
 992         rcu_read_unlock();
 993         return exe_file;
 994 }
 995 EXPORT_SYMBOL(get_mm_exe_file);
 996
 997 /**
 998  * get_task_exe_file - acquire a reference to the task's executable file
 999  *
1000  * Returns %NULL if task's mm (if any) has no associated executable file or
1001  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1002  * User must release file via fput().
1003  */
1004 struct file *get_task_exe_file(struct task_struct *task)
1005 {
1006         struct file *exe_file = NULL;
1007         struct mm_struct *mm;
1008
1009         task_lock(task);
1010         mm = task->mm;
1011         if (mm) {
1012                 if (!(task->flags & PF_KTHREAD))
1013                         exe_file = get_mm_exe_file(mm);
1014         }
1015         task_unlock(task);
1016         return exe_file;
1017 }
1018 EXPORT_SYMBOL(get_task_exe_file);
1019
1020 /**
1021  * get_task_mm - acquire a reference to the task's mm
1022  *
1023  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1024  * this kernel workthread has transiently adopted a user mm with use_mm,
1025  * to do its AIO) is not set and if so returns a reference to it, after
1026  * bumping up the use count.  User must release the mm via mmput()
1027  * after use.  Typically used by /proc and ptrace.
1028  */
1029 struct mm_struct *get_task_mm(struct task_struct *task)
1030 {
1031         struct mm_struct *mm;
1032
1033         task_lock(task);
1034         mm = task->mm;
1035         if (mm) {
1036                 if (task->flags & PF_KTHREAD)
1037                         mm = NULL;
1038                 else
1039                         mmget(mm);
1040         }
1041         task_unlock(task);
1042         return mm;
1043 }
1044 EXPORT_SYMBOL_GPL(get_task_mm);
1045
1046 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1047 {
1048         struct mm_struct *mm;
1049         int err;
1050
1051         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1052         if (err)
1053                 return ERR_PTR(err);
1054
1055         mm = get_task_mm(task);
1056         if (mm && mm != current->mm &&
1057                         !ptrace_may_access(task, mode)) {
1058                 mmput(mm);
1059                 mm = ERR_PTR(-EACCES);
1060         }
1061         mutex_unlock(&task->signal->cred_guard_mutex);
1062
1063         return mm;
1064 }
1065
1066 static void complete_vfork_done(struct task_struct *tsk)
1067 {
1068         struct completion *vfork;
1069
1070         task_lock(tsk);
1071         vfork = tsk->vfork_done;
1072         if (likely(vfork)) {
1073                 tsk->vfork_done = NULL;
1074                 complete(vfork);
1075         }
1076         task_unlock(tsk);
1077 }
1078
1079 static int wait_for_vfork_done(struct task_struct *child,
1080                                 struct completion *vfork)
1081 {
1082         int killed;
1083
1084         freezer_do_not_count();
1085         killed = wait_for_completion_killable(vfork);
1086         freezer_count();
1087
1088         if (killed) {
1089                 task_lock(child);
1090                 child->vfork_done = NULL;
1091                 task_unlock(child);
1092         }
1093
1094         put_task_struct(child);
1095         return killed;
1096 }
1097
1098 /* Please note the differences between mmput and mm_release.
1099  * mmput is called whenever we stop holding onto a mm_struct,
1100  * error success whatever.
1101  *
1102  * mm_release is called after a mm_struct has been removed
1103  * from the current process.
1104  *
1105  * This difference is important for error handling, when we
1106  * only half set up a mm_struct for a new process and need to restore
1107  * the old one.  Because we mmput the new mm_struct before
1108  * restoring the old one. . .
1109  * Eric Biederman 10 January 1998
1110  */
1111 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1112 {
1113         /* Get rid of any futexes when releasing the mm */
1114 #ifdef CONFIG_FUTEX
1115         if (unlikely(tsk->robust_list)) {
1116                 exit_robust_list(tsk);
1117                 tsk->robust_list = NULL;
1118         }
1119 #ifdef CONFIG_COMPAT
1120         if (unlikely(tsk->compat_robust_list)) {
1121                 compat_exit_robust_list(tsk);
1122                 tsk->compat_robust_list = NULL;
1123         }
1124 #endif
1125         if (unlikely(!list_empty(&tsk->pi_state_list)))
1126                 exit_pi_state_list(tsk);
1127 #endif
1128
1129         uprobe_free_utask(tsk);
1130
1131         /* Get rid of any cached register state */
1132         deactivate_mm(tsk, mm);
1133
1134         /*
1135          * Signal userspace if we're not exiting with a core dump
1136          * because we want to leave the value intact for debugging
1137          * purposes.
1138          */
1139         if (tsk->clear_child_tid) {
1140                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1141                     atomic_read(&mm->mm_users) > 1) {
1142                         /*
1143                          * We don't check the error code - if userspace has
1144                          * not set up a proper pointer then tough luck.
1145                          */
1146                         put_user(0, tsk->clear_child_tid);
1147                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1148                                         1, NULL, NULL, 0);
1149                 }
1150                 tsk->clear_child_tid = NULL;
1151         }
1152
1153         /*
1154          * All done, finally we can wake up parent and return this mm to him.
1155          * Also kthread_stop() uses this completion for synchronization.
1156          */
1157         if (tsk->vfork_done)
1158                 complete_vfork_done(tsk);
1159 }
1160
1161 /*
1162  * Allocate a new mm structure and copy contents from the
1163  * mm structure of the passed in task structure.
1164  */
1165 static struct mm_struct *dup_mm(struct task_struct *tsk)
1166 {
1167         struct mm_struct *mm, *oldmm = current->mm;
1168         int err;
1169
1170         mm = allocate_mm();
1171         if (!mm)
1172                 goto fail_nomem;
1173
1174         memcpy(mm, oldmm, sizeof(*mm));
1175
1176         if (!mm_init(mm, tsk, mm->user_ns))
1177                 goto fail_nomem;
1178
1179         err = dup_mmap(mm, oldmm);
1180         if (err)
1181                 goto free_pt;
1182
1183         mm->hiwater_rss = get_mm_rss(mm);
1184         mm->hiwater_vm = mm->total_vm;
1185
1186         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1187                 goto free_pt;
1188
1189         return mm;
1190
1191 free_pt:
1192         /* don't put binfmt in mmput, we haven't got module yet */
1193         mm->binfmt = NULL;
1194         mmput(mm);
1195
1196 fail_nomem:
1197         return NULL;
1198 }
1199
1200 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1201 {
1202         struct mm_struct *mm, *oldmm;
1203         int retval;
1204
1205         tsk->min_flt = tsk->maj_flt = 0;
1206         tsk->nvcsw = tsk->nivcsw = 0;
1207 #ifdef CONFIG_DETECT_HUNG_TASK
1208         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1209 #endif
1210
1211         tsk->mm = NULL;
1212         tsk->active_mm = NULL;
1213
1214         /*
1215          * Are we cloning a kernel thread?
1216          *
1217          * We need to steal a active VM for that..
1218          */
1219         oldmm = current->mm;
1220         if (!oldmm)
1221                 return 0;
1222
1223         /* initialize the new vmacache entries */
1224         vmacache_flush(tsk);
1225
1226         if (clone_flags & CLONE_VM) {
1227                 mmget(oldmm);
1228                 mm = oldmm;
1229                 goto good_mm;
1230         }
1231
1232         retval = -ENOMEM;
1233         mm = dup_mm(tsk);
1234         if (!mm)
1235                 goto fail_nomem;
1236
1237 good_mm:
1238         tsk->mm = mm;
1239         tsk->active_mm = mm;
1240         return 0;
1241
1242 fail_nomem:
1243         return retval;
1244 }
1245
1246 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1247 {
1248         struct fs_struct *fs = current->fs;
1249         if (clone_flags & CLONE_FS) {
1250                 /* tsk->fs is already what we want */
1251                 spin_lock(&fs->lock);
1252                 if (fs->in_exec) {
1253                         spin_unlock(&fs->lock);
1254                         return -EAGAIN;
1255                 }
1256                 fs->users++;
1257                 spin_unlock(&fs->lock);
1258                 return 0;
1259         }
1260         tsk->fs = copy_fs_struct(fs);
1261         if (!tsk->fs)
1262                 return -ENOMEM;
1263         return 0;
1264 }
1265
1266 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1267 {
1268         struct files_struct *oldf, *newf;
1269         int error = 0;
1270
1271         /*
1272          * A background process may not have any files ...
1273          */
1274         oldf = current->files;
1275         if (!oldf)
1276                 goto out;
1277
1278         if (clone_flags & CLONE_FILES) {
1279                 atomic_inc(&oldf->count);
1280                 goto out;
1281         }
1282
1283         newf = dup_fd(oldf, &error);
1284         if (!newf)
1285                 goto out;
1286
1287         tsk->files = newf;
1288         error = 0;
1289 out:
1290         return error;
1291 }
1292
1293 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1294 {
1295 #ifdef CONFIG_BLOCK
1296         struct io_context *ioc = current->io_context;
1297         struct io_context *new_ioc;
1298
1299         if (!ioc)
1300                 return 0;
1301         /*
1302          * Share io context with parent, if CLONE_IO is set
1303          */
1304         if (clone_flags & CLONE_IO) {
1305                 ioc_task_link(ioc);
1306                 tsk->io_context = ioc;
1307         } else if (ioprio_valid(ioc->ioprio)) {
1308                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1309                 if (unlikely(!new_ioc))
1310                         return -ENOMEM;
1311
1312                 new_ioc->ioprio = ioc->ioprio;
1313                 put_io_context(new_ioc);
1314         }
1315 #endif
1316         return 0;
1317 }
1318
1319 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1320 {
1321         struct sighand_struct *sig;
1322
1323         if (clone_flags & CLONE_SIGHAND) {
1324                 atomic_inc(&current->sighand->count);
1325                 return 0;
1326         }
1327         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1328         rcu_assign_pointer(tsk->sighand, sig);
1329         if (!sig)
1330                 return -ENOMEM;
1331
1332         atomic_set(&sig->count, 1);
1333         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1334         return 0;
1335 }
1336
1337 void __cleanup_sighand(struct sighand_struct *sighand)
1338 {
1339         if (atomic_dec_and_test(&sighand->count)) {
1340                 signalfd_cleanup(sighand);
1341                 /*
1342                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1343                  * without an RCU grace period, see __lock_task_sighand().
1344                  */
1345                 kmem_cache_free(sighand_cachep, sighand);
1346         }
1347 }
1348
1349 #ifdef CONFIG_POSIX_TIMERS
1350 /*
1351  * Initialize POSIX timer handling for a thread group.
1352  */
1353 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1354 {
1355         unsigned long cpu_limit;
1356
1357         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1358         if (cpu_limit != RLIM_INFINITY) {
1359                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1360                 sig->cputimer.running = true;
1361         }
1362
1363         /* The timer lists. */
1364         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1365         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1366         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1367 }
1368 #else
1369 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1370 #endif
1371
1372 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1373 {
1374         struct signal_struct *sig;
1375
1376         if (clone_flags & CLONE_THREAD)
1377                 return 0;
1378
1379         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1380         tsk->signal = sig;
1381         if (!sig)
1382                 return -ENOMEM;
1383
1384         sig->nr_threads = 1;
1385         atomic_set(&sig->live, 1);
1386         atomic_set(&sig->sigcnt, 1);
1387
1388         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1389         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1390         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1391
1392         init_waitqueue_head(&sig->wait_chldexit);
1393         sig->curr_target = tsk;
1394         init_sigpending(&sig->shared_pending);
1395         seqlock_init(&sig->stats_lock);
1396         prev_cputime_init(&sig->prev_cputime);
1397
1398 #ifdef CONFIG_POSIX_TIMERS
1399         INIT_LIST_HEAD(&sig->posix_timers);
1400         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1401         sig->real_timer.function = it_real_fn;
1402 #endif
1403
1404         task_lock(current->group_leader);
1405         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1406         task_unlock(current->group_leader);
1407
1408         posix_cpu_timers_init_group(sig);
1409
1410         tty_audit_fork(sig);
1411         sched_autogroup_fork(sig);
1412
1413         sig->oom_score_adj = current->signal->oom_score_adj;
1414         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1415
1416         mutex_init(&sig->cred_guard_mutex);
1417
1418         return 0;
1419 }
1420
1421 static void copy_seccomp(struct task_struct *p)
1422 {
1423 #ifdef CONFIG_SECCOMP
1424         /*
1425          * Must be called with sighand->lock held, which is common to
1426          * all threads in the group. Holding cred_guard_mutex is not
1427          * needed because this new task is not yet running and cannot
1428          * be racing exec.
1429          */
1430         assert_spin_locked(&current->sighand->siglock);
1431
1432         /* Ref-count the new filter user, and assign it. */
1433         get_seccomp_filter(current);
1434         p->seccomp = current->seccomp;
1435
1436         /*
1437          * Explicitly enable no_new_privs here in case it got set
1438          * between the task_struct being duplicated and holding the
1439          * sighand lock. The seccomp state and nnp must be in sync.
1440          */
1441         if (task_no_new_privs(current))
1442                 task_set_no_new_privs(p);
1443
1444         /*
1445          * If the parent gained a seccomp mode after copying thread
1446          * flags and between before we held the sighand lock, we have
1447          * to manually enable the seccomp thread flag here.
1448          */
1449         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1450                 set_tsk_thread_flag(p, TIF_SECCOMP);
1451 #endif
1452 }
1453
1454 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1455 {
1456         current->clear_child_tid = tidptr;
1457
1458         return task_pid_vnr(current);
1459 }
1460
1461 static void rt_mutex_init_task(struct task_struct *p)
1462 {
1463         raw_spin_lock_init(&p->pi_lock);
1464 #ifdef CONFIG_RT_MUTEXES
1465         p->pi_waiters = RB_ROOT_CACHED;
1466         p->pi_top_task = NULL;
1467         p->pi_blocked_on = NULL;
1468 #endif
1469 }
1470
1471 #ifdef CONFIG_POSIX_TIMERS
1472 /*
1473  * Initialize POSIX timer handling for a single task.
1474  */
1475 static void posix_cpu_timers_init(struct task_struct *tsk)
1476 {
1477         tsk->cputime_expires.prof_exp = 0;
1478         tsk->cputime_expires.virt_exp = 0;
1479         tsk->cputime_expires.sched_exp = 0;
1480         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1481         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1482         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1483 }
1484 #else
1485 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1486 #endif
1487
1488 static inline void
1489 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1490 {
1491          task->pids[type].pid = pid;
1492 }
1493
1494 static inline void rcu_copy_process(struct task_struct *p)
1495 {
1496 #ifdef CONFIG_PREEMPT_RCU
1497         p->rcu_read_lock_nesting = 0;
1498         p->rcu_read_unlock_special.s = 0;
1499         p->rcu_blocked_node = NULL;
1500         INIT_LIST_HEAD(&p->rcu_node_entry);
1501 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1502 #ifdef CONFIG_TASKS_RCU
1503         p->rcu_tasks_holdout = false;
1504         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1505         p->rcu_tasks_idle_cpu = -1;
1506 #endif /* #ifdef CONFIG_TASKS_RCU */
1507 }
1508
1509 /*
1510  * This creates a new process as a copy of the old one,
1511  * but does not actually start it yet.
1512  *
1513  * It copies the registers, and all the appropriate
1514  * parts of the process environment (as per the clone
1515  * flags). The actual kick-off is left to the caller.
1516  */
1517 static __latent_entropy struct task_struct *copy_process(
1518                                         unsigned long clone_flags,
1519                                         unsigned long stack_start,
1520                                         unsigned long stack_size,
1521                                         int __user *child_tidptr,
1522                                         struct pid *pid,
1523                                         int trace,
1524                                         unsigned long tls,
1525                                         int node)
1526 {
1527         int retval;
1528         struct task_struct *p;
1529
1530         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1531                 return ERR_PTR(-EINVAL);
1532
1533         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1534                 return ERR_PTR(-EINVAL);
1535
1536         /*
1537          * Thread groups must share signals as well, and detached threads
1538          * can only be started up within the thread group.
1539          */
1540         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1541                 return ERR_PTR(-EINVAL);
1542
1543         /*
1544          * Shared signal handlers imply shared VM. By way of the above,
1545          * thread groups also imply shared VM. Blocking this case allows
1546          * for various simplifications in other code.
1547          */
1548         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1549                 return ERR_PTR(-EINVAL);
1550
1551         /*
1552          * Siblings of global init remain as zombies on exit since they are
1553          * not reaped by their parent (swapper). To solve this and to avoid
1554          * multi-rooted process trees, prevent global and container-inits
1555          * from creating siblings.
1556          */
1557         if ((clone_flags & CLONE_PARENT) &&
1558                                 current->signal->flags & SIGNAL_UNKILLABLE)
1559                 return ERR_PTR(-EINVAL);
1560
1561         /*
1562          * If the new process will be in a different pid or user namespace
1563          * do not allow it to share a thread group with the forking task.
1564          */
1565         if (clone_flags & CLONE_THREAD) {
1566                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1567                     (task_active_pid_ns(current) !=
1568                                 current->nsproxy->pid_ns_for_children))
1569                         return ERR_PTR(-EINVAL);
1570         }
1571
1572         retval = security_task_create(clone_flags);
1573         if (retval)
1574                 goto fork_out;
1575
1576         retval = -ENOMEM;
1577         p = dup_task_struct(current, node);
1578         if (!p)
1579                 goto fork_out;
1580
1581         /*
1582          * This _must_ happen before we call free_task(), i.e. before we jump
1583          * to any of the bad_fork_* labels. This is to avoid freeing
1584          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1585          * kernel threads (PF_KTHREAD).
1586          */
1587         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1588         /*
1589          * Clear TID on mm_release()?
1590          */
1591         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1592
1593         ftrace_graph_init_task(p);
1594
1595         rt_mutex_init_task(p);
1596
1597 #ifdef CONFIG_PROVE_LOCKING
1598         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1599         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1600 #endif
1601         retval = -EAGAIN;
1602         if (atomic_read(&p->real_cred->user->processes) >=
1603                         task_rlimit(p, RLIMIT_NPROC)) {
1604                 if (p->real_cred->user != INIT_USER &&
1605                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1606                         goto bad_fork_free;
1607         }
1608         current->flags &= ~PF_NPROC_EXCEEDED;
1609
1610         retval = copy_creds(p, clone_flags);
1611         if (retval < 0)
1612                 goto bad_fork_free;
1613
1614         /*
1615          * If multiple threads are within copy_process(), then this check
1616          * triggers too late. This doesn't hurt, the check is only there
1617          * to stop root fork bombs.
1618          */
1619         retval = -EAGAIN;
1620         if (nr_threads >= max_threads)
1621                 goto bad_fork_cleanup_count;
1622
1623         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1624         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1625         p->flags |= PF_FORKNOEXEC;
1626         INIT_LIST_HEAD(&p->children);
1627         INIT_LIST_HEAD(&p->sibling);
1628         rcu_copy_process(p);
1629         p->vfork_done = NULL;
1630         spin_lock_init(&p->alloc_lock);
1631
1632         init_sigpending(&p->pending);
1633
1634         p->utime = p->stime = p->gtime = 0;
1635 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1636         p->utimescaled = p->stimescaled = 0;
1637 #endif
1638         prev_cputime_init(&p->prev_cputime);
1639
1640 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1641         seqcount_init(&p->vtime.seqcount);
1642         p->vtime.starttime = 0;
1643         p->vtime.state = VTIME_INACTIVE;
1644 #endif
1645
1646 #if defined(SPLIT_RSS_COUNTING)
1647         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1648 #endif
1649
1650         p->default_timer_slack_ns = current->timer_slack_ns;
1651
1652         task_io_accounting_init(&p->ioac);
1653         acct_clear_integrals(p);
1654
1655         posix_cpu_timers_init(p);
1656
1657         p->start_time = ktime_get_ns();
1658         p->real_start_time = ktime_get_boot_ns();
1659         p->io_context = NULL;
1660         p->audit_context = NULL;
1661         cgroup_fork(p);
1662 #ifdef CONFIG_NUMA
1663         p->mempolicy = mpol_dup(p->mempolicy);
1664         if (IS_ERR(p->mempolicy)) {
1665                 retval = PTR_ERR(p->mempolicy);
1666                 p->mempolicy = NULL;
1667                 goto bad_fork_cleanup_threadgroup_lock;
1668         }
1669 #endif
1670 #ifdef CONFIG_CPUSETS
1671         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1672         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1673         seqcount_init(&p->mems_allowed_seq);
1674 #endif
1675 #ifdef CONFIG_TRACE_IRQFLAGS
1676         p->irq_events = 0;
1677         p->hardirqs_enabled = 0;
1678         p->hardirq_enable_ip = 0;
1679         p->hardirq_enable_event = 0;
1680         p->hardirq_disable_ip = _THIS_IP_;
1681         p->hardirq_disable_event = 0;
1682         p->softirqs_enabled = 1;
1683         p->softirq_enable_ip = _THIS_IP_;
1684         p->softirq_enable_event = 0;
1685         p->softirq_disable_ip = 0;
1686         p->softirq_disable_event = 0;
1687         p->hardirq_context = 0;
1688         p->softirq_context = 0;
1689 #endif
1690
1691         p->pagefault_disabled = 0;
1692
1693 #ifdef CONFIG_LOCKDEP
1694         p->lockdep_depth = 0; /* no locks held yet */
1695         p->curr_chain_key = 0;
1696         p->lockdep_recursion = 0;
1697         lockdep_init_task(p);
1698 #endif
1699
1700 #ifdef CONFIG_DEBUG_MUTEXES
1701         p->blocked_on = NULL; /* not blocked yet */
1702 #endif
1703 #ifdef CONFIG_BCACHE
1704         p->sequential_io        = 0;
1705         p->sequential_io_avg    = 0;
1706 #endif
1707
1708         /* Perform scheduler related setup. Assign this task to a CPU. */
1709         retval = sched_fork(clone_flags, p);
1710         if (retval)
1711                 goto bad_fork_cleanup_policy;
1712
1713         retval = perf_event_init_task(p);
1714         if (retval)
1715                 goto bad_fork_cleanup_policy;
1716         retval = audit_alloc(p);
1717         if (retval)
1718                 goto bad_fork_cleanup_perf;
1719         /* copy all the process information */
1720         shm_init_task(p);
1721         retval = security_task_alloc(p, clone_flags);
1722         if (retval)
1723                 goto bad_fork_cleanup_audit;
1724         retval = copy_semundo(clone_flags, p);
1725         if (retval)
1726                 goto bad_fork_cleanup_security;
1727         retval = copy_files(clone_flags, p);
1728         if (retval)
1729                 goto bad_fork_cleanup_semundo;
1730         retval = copy_fs(clone_flags, p);
1731         if (retval)
1732                 goto bad_fork_cleanup_files;
1733         retval = copy_sighand(clone_flags, p);
1734         if (retval)
1735                 goto bad_fork_cleanup_fs;
1736         retval = copy_signal(clone_flags, p);
1737         if (retval)
1738                 goto bad_fork_cleanup_sighand;
1739         retval = copy_mm(clone_flags, p);
1740         if (retval)
1741                 goto bad_fork_cleanup_signal;
1742         retval = copy_namespaces(clone_flags, p);
1743         if (retval)
1744                 goto bad_fork_cleanup_mm;
1745         retval = copy_io(clone_flags, p);
1746         if (retval)
1747                 goto bad_fork_cleanup_namespaces;
1748         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1749         if (retval)
1750                 goto bad_fork_cleanup_io;
1751
1752         if (pid != &init_struct_pid) {
1753                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1754                 if (IS_ERR(pid)) {
1755                         retval = PTR_ERR(pid);
1756                         goto bad_fork_cleanup_thread;
1757                 }
1758         }
1759
1760 #ifdef CONFIG_BLOCK
1761         p->plug = NULL;
1762 #endif
1763 #ifdef CONFIG_FUTEX
1764         p->robust_list = NULL;
1765 #ifdef CONFIG_COMPAT
1766         p->compat_robust_list = NULL;
1767 #endif
1768         INIT_LIST_HEAD(&p->pi_state_list);
1769         p->pi_state_cache = NULL;
1770 #endif
1771         /*
1772          * sigaltstack should be cleared when sharing the same VM
1773          */
1774         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1775                 sas_ss_reset(p);
1776
1777         /*
1778          * Syscall tracing and stepping should be turned off in the
1779          * child regardless of CLONE_PTRACE.
1780          */
1781         user_disable_single_step(p);
1782         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1783 #ifdef TIF_SYSCALL_EMU
1784         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1785 #endif
1786         clear_all_latency_tracing(p);
1787
1788         /* ok, now we should be set up.. */
1789         p->pid = pid_nr(pid);
1790         if (clone_flags & CLONE_THREAD) {
1791                 p->exit_signal = -1;
1792                 p->group_leader = current->group_leader;
1793                 p->tgid = current->tgid;
1794         } else {
1795                 if (clone_flags & CLONE_PARENT)
1796                         p->exit_signal = current->group_leader->exit_signal;
1797                 else
1798                         p->exit_signal = (clone_flags & CSIGNAL);
1799                 p->group_leader = p;
1800                 p->tgid = p->pid;
1801         }
1802
1803         p->nr_dirtied = 0;
1804         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1805         p->dirty_paused_when = 0;
1806
1807         p->pdeath_signal = 0;
1808         INIT_LIST_HEAD(&p->thread_group);
1809         p->task_works = NULL;
1810
1811         cgroup_threadgroup_change_begin(current);
1812         /*
1813          * Ensure that the cgroup subsystem policies allow the new process to be
1814          * forked. It should be noted the the new process's css_set can be changed
1815          * between here and cgroup_post_fork() if an organisation operation is in
1816          * progress.
1817          */
1818         retval = cgroup_can_fork(p);
1819         if (retval)
1820                 goto bad_fork_free_pid;
1821
1822         /*
1823          * Make it visible to the rest of the system, but dont wake it up yet.
1824          * Need tasklist lock for parent etc handling!
1825          */
1826         write_lock_irq(&tasklist_lock);
1827
1828         /* CLONE_PARENT re-uses the old parent */
1829         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1830                 p->real_parent = current->real_parent;
1831                 p->parent_exec_id = current->parent_exec_id;
1832         } else {
1833                 p->real_parent = current;
1834                 p->parent_exec_id = current->self_exec_id;
1835         }
1836
1837         klp_copy_process(p);
1838
1839         spin_lock(&current->sighand->siglock);
1840
1841         /*
1842          * Copy seccomp details explicitly here, in case they were changed
1843          * before holding sighand lock.
1844          */
1845         copy_seccomp(p);
1846
1847         /*
1848          * Process group and session signals need to be delivered to just the
1849          * parent before the fork or both the parent and the child after the
1850          * fork. Restart if a signal comes in before we add the new process to
1851          * it's process group.
1852          * A fatal signal pending means that current will exit, so the new
1853          * thread can't slip out of an OOM kill (or normal SIGKILL).
1854         */
1855         recalc_sigpending();
1856         if (signal_pending(current)) {
1857                 retval = -ERESTARTNOINTR;
1858                 goto bad_fork_cancel_cgroup;
1859         }
1860         if (unlikely(!(ns_of_pid(pid)->nr_hashed & PIDNS_HASH_ADDING))) {
1861                 retval = -ENOMEM;
1862                 goto bad_fork_cancel_cgroup;
1863         }
1864
1865         if (likely(p->pid)) {
1866                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1867
1868                 init_task_pid(p, PIDTYPE_PID, pid);
1869                 if (thread_group_leader(p)) {
1870                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1871                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1872
1873                         if (is_child_reaper(pid)) {
1874                                 ns_of_pid(pid)->child_reaper = p;
1875                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1876                         }
1877
1878                         p->signal->leader_pid = pid;
1879                         p->signal->tty = tty_kref_get(current->signal->tty);
1880                         /*
1881                          * Inherit has_child_subreaper flag under the same
1882                          * tasklist_lock with adding child to the process tree
1883                          * for propagate_has_child_subreaper optimization.
1884                          */
1885                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1886                                                          p->real_parent->signal->is_child_subreaper;
1887                         list_add_tail(&p->sibling, &p->real_parent->children);
1888                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1889                         attach_pid(p, PIDTYPE_PGID);
1890                         attach_pid(p, PIDTYPE_SID);
1891                         __this_cpu_inc(process_counts);
1892                 } else {
1893                         current->signal->nr_threads++;
1894                         atomic_inc(&current->signal->live);
1895                         atomic_inc(&current->signal->sigcnt);
1896                         list_add_tail_rcu(&p->thread_group,
1897                                           &p->group_leader->thread_group);
1898                         list_add_tail_rcu(&p->thread_node,
1899                                           &p->signal->thread_head);
1900                 }
1901                 attach_pid(p, PIDTYPE_PID);
1902                 nr_threads++;
1903         }
1904
1905         total_forks++;
1906         spin_unlock(&current->sighand->siglock);
1907         syscall_tracepoint_update(p);
1908         write_unlock_irq(&tasklist_lock);
1909
1910         proc_fork_connector(p);
1911         cgroup_post_fork(p);
1912         cgroup_threadgroup_change_end(current);
1913         perf_event_fork(p);
1914
1915         trace_task_newtask(p, clone_flags);
1916         uprobe_copy_process(p, clone_flags);
1917
1918         return p;
1919
1920 bad_fork_cancel_cgroup:
1921         spin_unlock(&current->sighand->siglock);
1922         write_unlock_irq(&tasklist_lock);
1923         cgroup_cancel_fork(p);
1924 bad_fork_free_pid:
1925         cgroup_threadgroup_change_end(current);
1926         if (pid != &init_struct_pid)
1927                 free_pid(pid);
1928 bad_fork_cleanup_thread:
1929         exit_thread(p);
1930 bad_fork_cleanup_io:
1931         if (p->io_context)
1932                 exit_io_context(p);
1933 bad_fork_cleanup_namespaces:
1934         exit_task_namespaces(p);
1935 bad_fork_cleanup_mm:
1936         if (p->mm)
1937                 mmput(p->mm);
1938 bad_fork_cleanup_signal:
1939         if (!(clone_flags & CLONE_THREAD))
1940                 free_signal_struct(p->signal);
1941 bad_fork_cleanup_sighand:
1942         __cleanup_sighand(p->sighand);
1943 bad_fork_cleanup_fs:
1944         exit_fs(p); /* blocking */
1945 bad_fork_cleanup_files:
1946         exit_files(p); /* blocking */
1947 bad_fork_cleanup_semundo:
1948         exit_sem(p);
1949 bad_fork_cleanup_security:
1950         security_task_free(p);
1951 bad_fork_cleanup_audit:
1952         audit_free(p);
1953 bad_fork_cleanup_perf:
1954         perf_event_free_task(p);
1955 bad_fork_cleanup_policy:
1956         lockdep_free_task(p);
1957 #ifdef CONFIG_NUMA
1958         mpol_put(p->mempolicy);
1959 bad_fork_cleanup_threadgroup_lock:
1960 #endif
1961         delayacct_tsk_free(p);
1962 bad_fork_cleanup_count:
1963         atomic_dec(&p->cred->user->processes);
1964         exit_creds(p);
1965 bad_fork_free:
1966         p->state = TASK_DEAD;
1967         put_task_stack(p);
1968         free_task(p);
1969 fork_out:
1970         return ERR_PTR(retval);
1971 }
1972
1973 static inline void init_idle_pids(struct pid_link *links)
1974 {
1975         enum pid_type type;
1976
1977         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1978                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1979                 links[type].pid = &init_struct_pid;
1980         }
1981 }
1982
1983 struct task_struct *fork_idle(int cpu)
1984 {
1985         struct task_struct *task;
1986         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1987                             cpu_to_node(cpu));
1988         if (!IS_ERR(task)) {
1989                 init_idle_pids(task->pids);
1990                 init_idle(task, cpu);
1991         }
1992
1993         return task;
1994 }
1995
1996 /*
1997  *  Ok, this is the main fork-routine.
1998  *
1999  * It copies the process, and if successful kick-starts
2000  * it and waits for it to finish using the VM if required.
2001  */
2002 long _do_fork(unsigned long clone_flags,
2003               unsigned long stack_start,
2004               unsigned long stack_size,
2005               int __user *parent_tidptr,
2006               int __user *child_tidptr,
2007               unsigned long tls)
2008 {
2009         struct task_struct *p;
2010         int trace = 0;
2011         long nr;
2012
2013         /*
2014          * Determine whether and which event to report to ptracer.  When
2015          * called from kernel_thread or CLONE_UNTRACED is explicitly
2016          * requested, no event is reported; otherwise, report if the event
2017          * for the type of forking is enabled.
2018          */
2019         if (!(clone_flags & CLONE_UNTRACED)) {
2020                 if (clone_flags & CLONE_VFORK)
2021                         trace = PTRACE_EVENT_VFORK;
2022                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2023                         trace = PTRACE_EVENT_CLONE;
2024                 else
2025                         trace = PTRACE_EVENT_FORK;
2026
2027                 if (likely(!ptrace_event_enabled(current, trace)))
2028                         trace = 0;
2029         }
2030
2031         p = copy_process(clone_flags, stack_start, stack_size,
2032                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2033         add_latent_entropy();
2034         /*
2035          * Do this prior waking up the new thread - the thread pointer
2036          * might get invalid after that point, if the thread exits quickly.
2037          */
2038         if (!IS_ERR(p)) {
2039                 struct completion vfork;
2040                 struct pid *pid;
2041
2042                 trace_sched_process_fork(current, p);
2043
2044                 pid = get_task_pid(p, PIDTYPE_PID);
2045                 nr = pid_vnr(pid);
2046
2047                 if (clone_flags & CLONE_PARENT_SETTID)
2048                         put_user(nr, parent_tidptr);
2049
2050                 if (clone_flags & CLONE_VFORK) {
2051                         p->vfork_done = &vfork;
2052                         init_completion(&vfork);
2053                         get_task_struct(p);
2054                 }
2055
2056                 wake_up_new_task(p);
2057
2058                 /* forking complete and child started to run, tell ptracer */
2059                 if (unlikely(trace))
2060                         ptrace_event_pid(trace, pid);
2061
2062                 if (clone_flags & CLONE_VFORK) {
2063                         if (!wait_for_vfork_done(p, &vfork))
2064                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2065                 }
2066
2067                 put_pid(pid);
2068         } else {
2069                 nr = PTR_ERR(p);
2070         }
2071         return nr;
2072 }
2073
2074 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2075 /* For compatibility with architectures that call do_fork directly rather than
2076  * using the syscall entry points below. */
2077 long do_fork(unsigned long clone_flags,
2078               unsigned long stack_start,
2079               unsigned long stack_size,
2080               int __user *parent_tidptr,
2081               int __user *child_tidptr)
2082 {
2083         return _do_fork(clone_flags, stack_start, stack_size,
2084                         parent_tidptr, child_tidptr, 0);
2085 }
2086 #endif
2087
2088 /*
2089  * Create a kernel thread.
2090  */
2091 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2092 {
2093         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2094                 (unsigned long)arg, NULL, NULL, 0);
2095 }
2096
2097 #ifdef __ARCH_WANT_SYS_FORK
2098 SYSCALL_DEFINE0(fork)
2099 {
2100 #ifdef CONFIG_MMU
2101         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2102 #else
2103         /* can not support in nommu mode */
2104         return -EINVAL;
2105 #endif
2106 }
2107 #endif
2108
2109 #ifdef __ARCH_WANT_SYS_VFORK
2110 SYSCALL_DEFINE0(vfork)
2111 {
2112         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2113                         0, NULL, NULL, 0);
2114 }
2115 #endif
2116
2117 #ifdef __ARCH_WANT_SYS_CLONE
2118 #ifdef CONFIG_CLONE_BACKWARDS
2119 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2120                  int __user *, parent_tidptr,
2121                  unsigned long, tls,
2122                  int __user *, child_tidptr)
2123 #elif defined(CONFIG_CLONE_BACKWARDS2)
2124 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2125                  int __user *, parent_tidptr,
2126                  int __user *, child_tidptr,
2127                  unsigned long, tls)
2128 #elif defined(CONFIG_CLONE_BACKWARDS3)
2129 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2130                 int, stack_size,
2131                 int __user *, parent_tidptr,
2132                 int __user *, child_tidptr,
2133                 unsigned long, tls)
2134 #else
2135 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2136                  int __user *, parent_tidptr,
2137                  int __user *, child_tidptr,
2138                  unsigned long, tls)
2139 #endif
2140 {
2141         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2142 }
2143 #endif
2144
2145 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2146 {
2147         struct task_struct *leader, *parent, *child;
2148         int res;
2149
2150         read_lock(&tasklist_lock);
2151         leader = top = top->group_leader;
2152 down:
2153         for_each_thread(leader, parent) {
2154                 list_for_each_entry(child, &parent->children, sibling) {
2155                         res = visitor(child, data);
2156                         if (res) {
2157                                 if (res < 0)
2158                                         goto out;
2159                                 leader = child;
2160                                 goto down;
2161                         }
2162 up:
2163                         ;
2164                 }
2165         }
2166
2167         if (leader != top) {
2168                 child = leader;
2169                 parent = child->real_parent;
2170                 leader = parent->group_leader;
2171                 goto up;
2172         }
2173 out:
2174         read_unlock(&tasklist_lock);
2175 }
2176
2177 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2178 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2179 #endif
2180
2181 static void sighand_ctor(void *data)
2182 {
2183         struct sighand_struct *sighand = data;
2184
2185         spin_lock_init(&sighand->siglock);
2186         init_waitqueue_head(&sighand->signalfd_wqh);
2187 }
2188
2189 void __init proc_caches_init(void)
2190 {
2191         sighand_cachep = kmem_cache_create("sighand_cache",
2192                         sizeof(struct sighand_struct), 0,
2193                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2194                         SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2195         signal_cachep = kmem_cache_create("signal_cache",
2196                         sizeof(struct signal_struct), 0,
2197                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2198                         NULL);
2199         files_cachep = kmem_cache_create("files_cache",
2200                         sizeof(struct files_struct), 0,
2201                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2202                         NULL);
2203         fs_cachep = kmem_cache_create("fs_cache",
2204                         sizeof(struct fs_struct), 0,
2205                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2206                         NULL);
2207         /*
2208          * FIXME! The "sizeof(struct mm_struct)" currently includes the
2209          * whole struct cpumask for the OFFSTACK case. We could change
2210          * this to *only* allocate as much of it as required by the
2211          * maximum number of CPU's we can ever have.  The cpumask_allocation
2212          * is at the end of the structure, exactly for that reason.
2213          */
2214         mm_cachep = kmem_cache_create("mm_struct",
2215                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2216                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2217                         NULL);
2218         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2219         mmap_init();
2220         nsproxy_cache_init();
2221 }
2222
2223 /*
2224  * Check constraints on flags passed to the unshare system call.
2225  */
2226 static int check_unshare_flags(unsigned long unshare_flags)
2227 {
2228         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2229                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2230                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2231                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2232                 return -EINVAL;
2233         /*
2234          * Not implemented, but pretend it works if there is nothing
2235          * to unshare.  Note that unsharing the address space or the
2236          * signal handlers also need to unshare the signal queues (aka
2237          * CLONE_THREAD).
2238          */
2239         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2240                 if (!thread_group_empty(current))
2241                         return -EINVAL;
2242         }
2243         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2244                 if (atomic_read(&current->sighand->count) > 1)
2245                         return -EINVAL;
2246         }
2247         if (unshare_flags & CLONE_VM) {
2248                 if (!current_is_single_threaded())
2249                         return -EINVAL;
2250         }
2251
2252         return 0;
2253 }
2254
2255 /*
2256  * Unshare the filesystem structure if it is being shared
2257  */
2258 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2259 {
2260         struct fs_struct *fs = current->fs;
2261
2262         if (!(unshare_flags & CLONE_FS) || !fs)
2263                 return 0;
2264
2265         /* don't need lock here; in the worst case we'll do useless copy */
2266         if (fs->users == 1)
2267                 return 0;
2268
2269         *new_fsp = copy_fs_struct(fs);
2270         if (!*new_fsp)
2271                 return -ENOMEM;
2272
2273         return 0;
2274 }
2275
2276 /*
2277  * Unshare file descriptor table if it is being shared
2278  */
2279 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2280 {
2281         struct files_struct *fd = current->files;
2282         int error = 0;
2283
2284         if ((unshare_flags & CLONE_FILES) &&
2285             (fd && atomic_read(&fd->count) > 1)) {
2286                 *new_fdp = dup_fd(fd, &error);
2287                 if (!*new_fdp)
2288                         return error;
2289         }
2290
2291         return 0;
2292 }
2293
2294 /*
2295  * unshare allows a process to 'unshare' part of the process
2296  * context which was originally shared using clone.  copy_*
2297  * functions used by do_fork() cannot be used here directly
2298  * because they modify an inactive task_struct that is being
2299  * constructed. Here we are modifying the current, active,
2300  * task_struct.
2301  */
2302 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2303 {
2304         struct fs_struct *fs, *new_fs = NULL;
2305         struct files_struct *fd, *new_fd = NULL;
2306         struct cred *new_cred = NULL;
2307         struct nsproxy *new_nsproxy = NULL;
2308         int do_sysvsem = 0;
2309         int err;
2310
2311         /*
2312          * If unsharing a user namespace must also unshare the thread group
2313          * and unshare the filesystem root and working directories.
2314          */
2315         if (unshare_flags & CLONE_NEWUSER)
2316                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2317         /*
2318          * If unsharing vm, must also unshare signal handlers.
2319          */
2320         if (unshare_flags & CLONE_VM)
2321                 unshare_flags |= CLONE_SIGHAND;
2322         /*
2323          * If unsharing a signal handlers, must also unshare the signal queues.
2324          */
2325         if (unshare_flags & CLONE_SIGHAND)
2326                 unshare_flags |= CLONE_THREAD;
2327         /*
2328          * If unsharing namespace, must also unshare filesystem information.
2329          */
2330         if (unshare_flags & CLONE_NEWNS)
2331                 unshare_flags |= CLONE_FS;
2332
2333         err = check_unshare_flags(unshare_flags);
2334         if (err)
2335                 goto bad_unshare_out;
2336         /*
2337          * CLONE_NEWIPC must also detach from the undolist: after switching
2338          * to a new ipc namespace, the semaphore arrays from the old
2339          * namespace are unreachable.
2340          */
2341         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2342                 do_sysvsem = 1;
2343         err = unshare_fs(unshare_flags, &new_fs);
2344         if (err)
2345                 goto bad_unshare_out;
2346         err = unshare_fd(unshare_flags, &new_fd);
2347         if (err)
2348                 goto bad_unshare_cleanup_fs;
2349         err = unshare_userns(unshare_flags, &new_cred);
2350         if (err)
2351                 goto bad_unshare_cleanup_fd;
2352         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2353                                          new_cred, new_fs);
2354         if (err)
2355                 goto bad_unshare_cleanup_cred;
2356
2357         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2358                 if (do_sysvsem) {
2359                         /*
2360                          * CLONE_SYSVSEM is equivalent to sys_exit().
2361                          */
2362                         exit_sem(current);
2363                 }
2364                 if (unshare_flags & CLONE_NEWIPC) {
2365                         /* Orphan segments in old ns (see sem above). */
2366                         exit_shm(current);
2367                         shm_init_task(current);
2368                 }
2369
2370                 if (new_nsproxy)
2371                         switch_task_namespaces(current, new_nsproxy);
2372
2373                 task_lock(current);
2374
2375                 if (new_fs) {
2376                         fs = current->fs;
2377                         spin_lock(&fs->lock);
2378                         current->fs = new_fs;
2379                         if (--fs->users)
2380                                 new_fs = NULL;
2381                         else
2382                                 new_fs = fs;
2383                         spin_unlock(&fs->lock);
2384                 }
2385
2386                 if (new_fd) {
2387                         fd = current->files;
2388                         current->files = new_fd;
2389                         new_fd = fd;
2390                 }
2391
2392                 task_unlock(current);
2393
2394                 if (new_cred) {
2395                         /* Install the new user namespace */
2396                         commit_creds(new_cred);
2397                         new_cred = NULL;
2398                 }
2399         }
2400
2401         perf_event_namespaces(current);
2402
2403 bad_unshare_cleanup_cred:
2404         if (new_cred)
2405                 put_cred(new_cred);
2406 bad_unshare_cleanup_fd:
2407         if (new_fd)
2408                 put_files_struct(new_fd);
2409
2410 bad_unshare_cleanup_fs:
2411         if (new_fs)
2412                 free_fs_struct(new_fs);
2413
2414 bad_unshare_out:
2415         return err;
2416 }
2417
2418 /*
2419  *      Helper to unshare the files of the current task.
2420  *      We don't want to expose copy_files internals to
2421  *      the exec layer of the kernel.
2422  */
2423
2424 int unshare_files(struct files_struct **displaced)
2425 {
2426         struct task_struct *task = current;
2427         struct files_struct *copy = NULL;
2428         int error;
2429
2430         error = unshare_fd(CLONE_FILES, &copy);
2431         if (error || !copy) {
2432                 *displaced = NULL;
2433                 return error;
2434         }
2435         *displaced = task->files;
2436         task_lock(task);
2437         task->files = copy;
2438         task_unlock(task);
2439         return 0;
2440 }
2441
2442 int sysctl_max_threads(struct ctl_table *table, int write,
2443                        void __user *buffer, size_t *lenp, loff_t *ppos)
2444 {
2445         struct ctl_table t;
2446         int ret;
2447         int threads = max_threads;
2448         int min = MIN_THREADS;
2449         int max = MAX_THREADS;
2450
2451         t = *table;
2452         t.data = &threads;
2453         t.extra1 = &min;
2454         t.extra2 = &max;
2455
2456         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2457         if (ret || !write)
2458                 return ret;
2459
2460         set_max_threads(threads);
2461
2462         return 0;
2463 }