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 #ifdef CONFIG_MMU
 950 static void mmput_async_fn(struct work_struct *work)
 951 {
 952         struct mm_struct *mm = container_of(work, struct mm_struct,
 953                                             async_put_work);
 954
 955         __mmput(mm);
 956 }
 957
 958 void mmput_async(struct mm_struct *mm)
 959 {
 960         if (atomic_dec_and_test(&mm->mm_users)) {
 961                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
 962                 schedule_work(&mm->async_put_work);
 963         }
 964 }
 965 #endif
 966
 967 /**
 968  * set_mm_exe_file - change a reference to the mm's executable file
 969  *
 970  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 971  *
 972  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 973  * invocations: in mmput() nobody alive left, in execve task is single
 974  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 975  * mm->exe_file, but does so without using set_mm_exe_file() in order
 976  * to do avoid the need for any locks.
 977  */
 978 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 979 {
 980         struct file *old_exe_file;
 981
 982         /*
 983          * It is safe to dereference the exe_file without RCU as
 984          * this function is only called if nobody else can access
 985          * this mm -- see comment above for justification.
 986          */
 987         old_exe_file = rcu_dereference_raw(mm->exe_file);
 988
 989         if (new_exe_file)
 990                 get_file(new_exe_file);
 991         rcu_assign_pointer(mm->exe_file, new_exe_file);
 992         if (old_exe_file)
 993                 fput(old_exe_file);
 994 }
 995
 996 /**
 997  * get_mm_exe_file - acquire a reference to the mm's executable file
 998  *
 999  * Returns %NULL if mm has no associated executable file.
1000  * User must release file via fput().
1001  */
1002 struct file *get_mm_exe_file(struct mm_struct *mm)
1003 {
1004         struct file *exe_file;
1005
1006         rcu_read_lock();
1007         exe_file = rcu_dereference(mm->exe_file);
1008         if (exe_file && !get_file_rcu(exe_file))
1009                 exe_file = NULL;
1010         rcu_read_unlock();
1011         return exe_file;
1012 }
1013 EXPORT_SYMBOL(get_mm_exe_file);
1014
1015 /**
1016  * get_task_exe_file - acquire a reference to the task's executable file
1017  *
1018  * Returns %NULL if task's mm (if any) has no associated executable file or
1019  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1020  * User must release file via fput().
1021  */
1022 struct file *get_task_exe_file(struct task_struct *task)
1023 {
1024         struct file *exe_file = NULL;
1025         struct mm_struct *mm;
1026
1027         task_lock(task);
1028         mm = task->mm;
1029         if (mm) {
1030                 if (!(task->flags & PF_KTHREAD))
1031                         exe_file = get_mm_exe_file(mm);
1032         }
1033         task_unlock(task);
1034         return exe_file;
1035 }
1036 EXPORT_SYMBOL(get_task_exe_file);
1037
1038 /**
1039  * get_task_mm - acquire a reference to the task's mm
1040  *
1041  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1042  * this kernel workthread has transiently adopted a user mm with use_mm,
1043  * to do its AIO) is not set and if so returns a reference to it, after
1044  * bumping up the use count.  User must release the mm via mmput()
1045  * after use.  Typically used by /proc and ptrace.
1046  */
1047 struct mm_struct *get_task_mm(struct task_struct *task)
1048 {
1049         struct mm_struct *mm;
1050
1051         task_lock(task);
1052         mm = task->mm;
1053         if (mm) {
1054                 if (task->flags & PF_KTHREAD)
1055                         mm = NULL;
1056                 else
1057                         mmget(mm);
1058         }
1059         task_unlock(task);
1060         return mm;
1061 }
1062 EXPORT_SYMBOL_GPL(get_task_mm);
1063
1064 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1065 {
1066         struct mm_struct *mm;
1067         int err;
1068
1069         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1070         if (err)
1071                 return ERR_PTR(err);
1072
1073         mm = get_task_mm(task);
1074         if (mm && mm != current->mm &&
1075                         !ptrace_may_access(task, mode)) {
1076                 mmput(mm);
1077                 mm = ERR_PTR(-EACCES);
1078         }
1079         mutex_unlock(&task->signal->cred_guard_mutex);
1080
1081         return mm;
1082 }
1083
1084 static void complete_vfork_done(struct task_struct *tsk)
1085 {
1086         struct completion *vfork;
1087
1088         task_lock(tsk);
1089         vfork = tsk->vfork_done;
1090         if (likely(vfork)) {
1091                 tsk->vfork_done = NULL;
1092                 complete(vfork);
1093         }
1094         task_unlock(tsk);
1095 }
1096
1097 static int wait_for_vfork_done(struct task_struct *child,
1098                                 struct completion *vfork)
1099 {
1100         int killed;
1101
1102         freezer_do_not_count();
1103         killed = wait_for_completion_killable(vfork);
1104         freezer_count();
1105
1106         if (killed) {
1107                 task_lock(child);
1108                 child->vfork_done = NULL;
1109                 task_unlock(child);
1110         }
1111
1112         put_task_struct(child);
1113         return killed;
1114 }
1115
1116 /* Please note the differences between mmput and mm_release.
1117  * mmput is called whenever we stop holding onto a mm_struct,
1118  * error success whatever.
1119  *
1120  * mm_release is called after a mm_struct has been removed
1121  * from the current process.
1122  *
1123  * This difference is important for error handling, when we
1124  * only half set up a mm_struct for a new process and need to restore
1125  * the old one.  Because we mmput the new mm_struct before
1126  * restoring the old one. . .
1127  * Eric Biederman 10 January 1998
1128  */
1129 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1130 {
1131         /* Get rid of any futexes when releasing the mm */
1132 #ifdef CONFIG_FUTEX
1133         if (unlikely(tsk->robust_list)) {
1134                 exit_robust_list(tsk);
1135                 tsk->robust_list = NULL;
1136         }
1137 #ifdef CONFIG_COMPAT
1138         if (unlikely(tsk->compat_robust_list)) {
1139                 compat_exit_robust_list(tsk);
1140                 tsk->compat_robust_list = NULL;
1141         }
1142 #endif
1143         if (unlikely(!list_empty(&tsk->pi_state_list)))
1144                 exit_pi_state_list(tsk);
1145 #endif
1146
1147         uprobe_free_utask(tsk);
1148
1149         /* Get rid of any cached register state */
1150         deactivate_mm(tsk, mm);
1151
1152         /*
1153          * Signal userspace if we're not exiting with a core dump
1154          * because we want to leave the value intact for debugging
1155          * purposes.
1156          */
1157         if (tsk->clear_child_tid) {
1158                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1159                     atomic_read(&mm->mm_users) > 1) {
1160                         /*
1161                          * We don't check the error code - if userspace has
1162                          * not set up a proper pointer then tough luck.
1163                          */
1164                         put_user(0, tsk->clear_child_tid);
1165                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1166                                         1, NULL, NULL, 0);
1167                 }
1168                 tsk->clear_child_tid = NULL;
1169         }
1170
1171         /*
1172          * All done, finally we can wake up parent and return this mm to him.
1173          * Also kthread_stop() uses this completion for synchronization.
1174          */
1175         if (tsk->vfork_done)
1176                 complete_vfork_done(tsk);
1177 }
1178
1179 /*
1180  * Allocate a new mm structure and copy contents from the
1181  * mm structure of the passed in task structure.
1182  */
1183 static struct mm_struct *dup_mm(struct task_struct *tsk)
1184 {
1185         struct mm_struct *mm, *oldmm = current->mm;
1186         int err;
1187
1188         mm = allocate_mm();
1189         if (!mm)
1190                 goto fail_nomem;
1191
1192         memcpy(mm, oldmm, sizeof(*mm));
1193
1194         if (!mm_init(mm, tsk, mm->user_ns))
1195                 goto fail_nomem;
1196
1197         err = dup_mmap(mm, oldmm);
1198         if (err)
1199                 goto free_pt;
1200
1201         mm->hiwater_rss = get_mm_rss(mm);
1202         mm->hiwater_vm = mm->total_vm;
1203
1204         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1205                 goto free_pt;
1206
1207         return mm;
1208
1209 free_pt:
1210         /* don't put binfmt in mmput, we haven't got module yet */
1211         mm->binfmt = NULL;
1212         mmput(mm);
1213
1214 fail_nomem:
1215         return NULL;
1216 }
1217
1218 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1219 {
1220         struct mm_struct *mm, *oldmm;
1221         int retval;
1222
1223         tsk->min_flt = tsk->maj_flt = 0;
1224         tsk->nvcsw = tsk->nivcsw = 0;
1225 #ifdef CONFIG_DETECT_HUNG_TASK
1226         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1227 #endif
1228
1229         tsk->mm = NULL;
1230         tsk->active_mm = NULL;
1231
1232         /*
1233          * Are we cloning a kernel thread?
1234          *
1235          * We need to steal a active VM for that..
1236          */
1237         oldmm = current->mm;
1238         if (!oldmm)
1239                 return 0;
1240
1241         /* initialize the new vmacache entries */
1242         vmacache_flush(tsk);
1243
1244         if (clone_flags & CLONE_VM) {
1245                 mmget(oldmm);
1246                 mm = oldmm;
1247                 goto good_mm;
1248         }
1249
1250         retval = -ENOMEM;
1251         mm = dup_mm(tsk);
1252         if (!mm)
1253                 goto fail_nomem;
1254
1255 good_mm:
1256         tsk->mm = mm;
1257         tsk->active_mm = mm;
1258         return 0;
1259
1260 fail_nomem:
1261         return retval;
1262 }
1263
1264 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1265 {
1266         struct fs_struct *fs = current->fs;
1267         if (clone_flags & CLONE_FS) {
1268                 /* tsk->fs is already what we want */
1269                 spin_lock(&fs->lock);
1270                 if (fs->in_exec) {
1271                         spin_unlock(&fs->lock);
1272                         return -EAGAIN;
1273                 }
1274                 fs->users++;
1275                 spin_unlock(&fs->lock);
1276                 return 0;
1277         }
1278         tsk->fs = copy_fs_struct(fs);
1279         if (!tsk->fs)
1280                 return -ENOMEM;
1281         return 0;
1282 }
1283
1284 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1285 {
1286         struct files_struct *oldf, *newf;
1287         int error = 0;
1288
1289         /*
1290          * A background process may not have any files ...
1291          */
1292         oldf = current->files;
1293         if (!oldf)
1294                 goto out;
1295
1296         if (clone_flags & CLONE_FILES) {
1297                 atomic_inc(&oldf->count);
1298                 goto out;
1299         }
1300
1301         newf = dup_fd(oldf, &error);
1302         if (!newf)
1303                 goto out;
1304
1305         tsk->files = newf;
1306         error = 0;
1307 out:
1308         return error;
1309 }
1310
1311 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1312 {
1313 #ifdef CONFIG_BLOCK
1314         struct io_context *ioc = current->io_context;
1315         struct io_context *new_ioc;
1316
1317         if (!ioc)
1318                 return 0;
1319         /*
1320          * Share io context with parent, if CLONE_IO is set
1321          */
1322         if (clone_flags & CLONE_IO) {
1323                 ioc_task_link(ioc);
1324                 tsk->io_context = ioc;
1325         } else if (ioprio_valid(ioc->ioprio)) {
1326                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1327                 if (unlikely(!new_ioc))
1328                         return -ENOMEM;
1329
1330                 new_ioc->ioprio = ioc->ioprio;
1331                 put_io_context(new_ioc);
1332         }
1333 #endif
1334         return 0;
1335 }
1336
1337 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1338 {
1339         struct sighand_struct *sig;
1340
1341         if (clone_flags & CLONE_SIGHAND) {
1342                 atomic_inc(&current->sighand->count);
1343                 return 0;
1344         }
1345         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1346         rcu_assign_pointer(tsk->sighand, sig);
1347         if (!sig)
1348                 return -ENOMEM;
1349
1350         atomic_set(&sig->count, 1);
1351         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1352         return 0;
1353 }
1354
1355 void __cleanup_sighand(struct sighand_struct *sighand)
1356 {
1357         if (atomic_dec_and_test(&sighand->count)) {
1358                 signalfd_cleanup(sighand);
1359                 /*
1360                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1361                  * without an RCU grace period, see __lock_task_sighand().
1362                  */
1363                 kmem_cache_free(sighand_cachep, sighand);
1364         }
1365 }
1366
1367 #ifdef CONFIG_POSIX_TIMERS
1368 /*
1369  * Initialize POSIX timer handling for a thread group.
1370  */
1371 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1372 {
1373         unsigned long cpu_limit;
1374
1375         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1376         if (cpu_limit != RLIM_INFINITY) {
1377                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1378                 sig->cputimer.running = true;
1379         }
1380
1381         /* The timer lists. */
1382         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1383         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1384         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1385 }
1386 #else
1387 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1388 #endif
1389
1390 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1391 {
1392         struct signal_struct *sig;
1393
1394         if (clone_flags & CLONE_THREAD)
1395                 return 0;
1396
1397         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1398         tsk->signal = sig;
1399         if (!sig)
1400                 return -ENOMEM;
1401
1402         sig->nr_threads = 1;
1403         atomic_set(&sig->live, 1);
1404         atomic_set(&sig->sigcnt, 1);
1405
1406         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1407         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1408         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1409
1410         init_waitqueue_head(&sig->wait_chldexit);
1411         sig->curr_target = tsk;
1412         init_sigpending(&sig->shared_pending);
1413         seqlock_init(&sig->stats_lock);
1414         prev_cputime_init(&sig->prev_cputime);
1415
1416 #ifdef CONFIG_POSIX_TIMERS
1417         INIT_LIST_HEAD(&sig->posix_timers);
1418         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1419         sig->real_timer.function = it_real_fn;
1420 #endif
1421
1422         task_lock(current->group_leader);
1423         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1424         task_unlock(current->group_leader);
1425
1426         posix_cpu_timers_init_group(sig);
1427
1428         tty_audit_fork(sig);
1429         sched_autogroup_fork(sig);
1430
1431         sig->oom_score_adj = current->signal->oom_score_adj;
1432         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1433
1434         mutex_init(&sig->cred_guard_mutex);
1435
1436         return 0;
1437 }
1438
1439 static void copy_seccomp(struct task_struct *p)
1440 {
1441 #ifdef CONFIG_SECCOMP
1442         /*
1443          * Must be called with sighand->lock held, which is common to
1444          * all threads in the group. Holding cred_guard_mutex is not
1445          * needed because this new task is not yet running and cannot
1446          * be racing exec.
1447          */
1448         assert_spin_locked(&current->sighand->siglock);
1449
1450         /* Ref-count the new filter user, and assign it. */
1451         get_seccomp_filter(current);
1452         p->seccomp = current->seccomp;
1453
1454         /*
1455          * Explicitly enable no_new_privs here in case it got set
1456          * between the task_struct being duplicated and holding the
1457          * sighand lock. The seccomp state and nnp must be in sync.
1458          */
1459         if (task_no_new_privs(current))
1460                 task_set_no_new_privs(p);
1461
1462         /*
1463          * If the parent gained a seccomp mode after copying thread
1464          * flags and between before we held the sighand lock, we have
1465          * to manually enable the seccomp thread flag here.
1466          */
1467         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1468                 set_tsk_thread_flag(p, TIF_SECCOMP);
1469 #endif
1470 }
1471
1472 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1473 {
1474         current->clear_child_tid = tidptr;
1475
1476         return task_pid_vnr(current);
1477 }
1478
1479 static void rt_mutex_init_task(struct task_struct *p)
1480 {
1481         raw_spin_lock_init(&p->pi_lock);
1482 #ifdef CONFIG_RT_MUTEXES
1483         p->pi_waiters = RB_ROOT_CACHED;
1484         p->pi_top_task = NULL;
1485         p->pi_blocked_on = NULL;
1486 #endif
1487 }
1488
1489 #ifdef CONFIG_POSIX_TIMERS
1490 /*
1491  * Initialize POSIX timer handling for a single task.
1492  */
1493 static void posix_cpu_timers_init(struct task_struct *tsk)
1494 {
1495         tsk->cputime_expires.prof_exp = 0;
1496         tsk->cputime_expires.virt_exp = 0;
1497         tsk->cputime_expires.sched_exp = 0;
1498         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1499         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1500         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1501 }
1502 #else
1503 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1504 #endif
1505
1506 static inline void
1507 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1508 {
1509          task->pids[type].pid = pid;
1510 }
1511
1512 static inline void rcu_copy_process(struct task_struct *p)
1513 {
1514 #ifdef CONFIG_PREEMPT_RCU
1515         p->rcu_read_lock_nesting = 0;
1516         p->rcu_read_unlock_special.s = 0;
1517         p->rcu_blocked_node = NULL;
1518         INIT_LIST_HEAD(&p->rcu_node_entry);
1519 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1520 #ifdef CONFIG_TASKS_RCU
1521         p->rcu_tasks_holdout = false;
1522         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1523         p->rcu_tasks_idle_cpu = -1;
1524 #endif /* #ifdef CONFIG_TASKS_RCU */
1525 }
1526
1527 /*
1528  * This creates a new process as a copy of the old one,
1529  * but does not actually start it yet.
1530  *
1531  * It copies the registers, and all the appropriate
1532  * parts of the process environment (as per the clone
1533  * flags). The actual kick-off is left to the caller.
1534  */
1535 static __latent_entropy struct task_struct *copy_process(
1536                                         unsigned long clone_flags,
1537                                         unsigned long stack_start,
1538                                         unsigned long stack_size,
1539                                         int __user *child_tidptr,
1540                                         struct pid *pid,
1541                                         int trace,
1542                                         unsigned long tls,
1543                                         int node)
1544 {
1545         int retval;
1546         struct task_struct *p;
1547
1548         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1549                 return ERR_PTR(-EINVAL);
1550
1551         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1552                 return ERR_PTR(-EINVAL);
1553
1554         /*
1555          * Thread groups must share signals as well, and detached threads
1556          * can only be started up within the thread group.
1557          */
1558         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1559                 return ERR_PTR(-EINVAL);
1560
1561         /*
1562          * Shared signal handlers imply shared VM. By way of the above,
1563          * thread groups also imply shared VM. Blocking this case allows
1564          * for various simplifications in other code.
1565          */
1566         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1567                 return ERR_PTR(-EINVAL);
1568
1569         /*
1570          * Siblings of global init remain as zombies on exit since they are
1571          * not reaped by their parent (swapper). To solve this and to avoid
1572          * multi-rooted process trees, prevent global and container-inits
1573          * from creating siblings.
1574          */
1575         if ((clone_flags & CLONE_PARENT) &&
1576                                 current->signal->flags & SIGNAL_UNKILLABLE)
1577                 return ERR_PTR(-EINVAL);
1578
1579         /*
1580          * If the new process will be in a different pid or user namespace
1581          * do not allow it to share a thread group with the forking task.
1582          */
1583         if (clone_flags & CLONE_THREAD) {
1584                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1585                     (task_active_pid_ns(current) !=
1586                                 current->nsproxy->pid_ns_for_children))
1587                         return ERR_PTR(-EINVAL);
1588         }
1589
1590         retval = -ENOMEM;
1591         p = dup_task_struct(current, node);
1592         if (!p)
1593                 goto fork_out;
1594
1595         /*
1596          * This _must_ happen before we call free_task(), i.e. before we jump
1597          * to any of the bad_fork_* labels. This is to avoid freeing
1598          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1599          * kernel threads (PF_KTHREAD).
1600          */
1601         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1602         /*
1603          * Clear TID on mm_release()?
1604          */
1605         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1606
1607         ftrace_graph_init_task(p);
1608
1609         rt_mutex_init_task(p);
1610
1611 #ifdef CONFIG_PROVE_LOCKING
1612         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1613         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1614 #endif
1615         retval = -EAGAIN;
1616         if (atomic_read(&p->real_cred->user->processes) >=
1617                         task_rlimit(p, RLIMIT_NPROC)) {
1618                 if (p->real_cred->user != INIT_USER &&
1619                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1620                         goto bad_fork_free;
1621         }
1622         current->flags &= ~PF_NPROC_EXCEEDED;
1623
1624         retval = copy_creds(p, clone_flags);
1625         if (retval < 0)
1626                 goto bad_fork_free;
1627
1628         /*
1629          * If multiple threads are within copy_process(), then this check
1630          * triggers too late. This doesn't hurt, the check is only there
1631          * to stop root fork bombs.
1632          */
1633         retval = -EAGAIN;
1634         if (nr_threads >= max_threads)
1635                 goto bad_fork_cleanup_count;
1636
1637         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1638         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1639         p->flags |= PF_FORKNOEXEC;
1640         INIT_LIST_HEAD(&p->children);
1641         INIT_LIST_HEAD(&p->sibling);
1642         rcu_copy_process(p);
1643         p->vfork_done = NULL;
1644         spin_lock_init(&p->alloc_lock);
1645
1646         init_sigpending(&p->pending);
1647
1648         p->utime = p->stime = p->gtime = 0;
1649 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1650         p->utimescaled = p->stimescaled = 0;
1651 #endif
1652         prev_cputime_init(&p->prev_cputime);
1653
1654 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1655         seqcount_init(&p->vtime.seqcount);
1656         p->vtime.starttime = 0;
1657         p->vtime.state = VTIME_INACTIVE;
1658 #endif
1659
1660 #if defined(SPLIT_RSS_COUNTING)
1661         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1662 #endif
1663
1664         p->default_timer_slack_ns = current->timer_slack_ns;
1665
1666         task_io_accounting_init(&p->ioac);
1667         acct_clear_integrals(p);
1668
1669         posix_cpu_timers_init(p);
1670
1671         p->start_time = ktime_get_ns();
1672         p->real_start_time = ktime_get_boot_ns();
1673         p->io_context = NULL;
1674         p->audit_context = NULL;
1675         cgroup_fork(p);
1676 #ifdef CONFIG_NUMA
1677         p->mempolicy = mpol_dup(p->mempolicy);
1678         if (IS_ERR(p->mempolicy)) {
1679                 retval = PTR_ERR(p->mempolicy);
1680                 p->mempolicy = NULL;
1681                 goto bad_fork_cleanup_threadgroup_lock;
1682         }
1683 #endif
1684 #ifdef CONFIG_CPUSETS
1685         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1686         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1687         seqcount_init(&p->mems_allowed_seq);
1688 #endif
1689 #ifdef CONFIG_TRACE_IRQFLAGS
1690         p->irq_events = 0;
1691         p->hardirqs_enabled = 0;
1692         p->hardirq_enable_ip = 0;
1693         p->hardirq_enable_event = 0;
1694         p->hardirq_disable_ip = _THIS_IP_;
1695         p->hardirq_disable_event = 0;
1696         p->softirqs_enabled = 1;
1697         p->softirq_enable_ip = _THIS_IP_;
1698         p->softirq_enable_event = 0;
1699         p->softirq_disable_ip = 0;
1700         p->softirq_disable_event = 0;
1701         p->hardirq_context = 0;
1702         p->softirq_context = 0;
1703 #endif
1704
1705         p->pagefault_disabled = 0;
1706
1707 #ifdef CONFIG_LOCKDEP
1708         p->lockdep_depth = 0; /* no locks held yet */
1709         p->curr_chain_key = 0;
1710         p->lockdep_recursion = 0;
1711         lockdep_init_task(p);
1712 #endif
1713
1714 #ifdef CONFIG_DEBUG_MUTEXES
1715         p->blocked_on = NULL; /* not blocked yet */
1716 #endif
1717 #ifdef CONFIG_BCACHE
1718         p->sequential_io        = 0;
1719         p->sequential_io_avg    = 0;
1720 #endif
1721
1722         /* Perform scheduler related setup. Assign this task to a CPU. */
1723         retval = sched_fork(clone_flags, p);
1724         if (retval)
1725                 goto bad_fork_cleanup_policy;
1726
1727         retval = perf_event_init_task(p);
1728         if (retval)
1729                 goto bad_fork_cleanup_policy;
1730         retval = audit_alloc(p);
1731         if (retval)
1732                 goto bad_fork_cleanup_perf;
1733         /* copy all the process information */
1734         shm_init_task(p);
1735         retval = security_task_alloc(p, clone_flags);
1736         if (retval)
1737                 goto bad_fork_cleanup_audit;
1738         retval = copy_semundo(clone_flags, p);
1739         if (retval)
1740                 goto bad_fork_cleanup_security;
1741         retval = copy_files(clone_flags, p);
1742         if (retval)
1743                 goto bad_fork_cleanup_semundo;
1744         retval = copy_fs(clone_flags, p);
1745         if (retval)
1746                 goto bad_fork_cleanup_files;
1747         retval = copy_sighand(clone_flags, p);
1748         if (retval)
1749                 goto bad_fork_cleanup_fs;
1750         retval = copy_signal(clone_flags, p);
1751         if (retval)
1752                 goto bad_fork_cleanup_sighand;
1753         retval = copy_mm(clone_flags, p);
1754         if (retval)
1755                 goto bad_fork_cleanup_signal;
1756         retval = copy_namespaces(clone_flags, p);
1757         if (retval)
1758                 goto bad_fork_cleanup_mm;
1759         retval = copy_io(clone_flags, p);
1760         if (retval)
1761                 goto bad_fork_cleanup_namespaces;
1762         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1763         if (retval)
1764                 goto bad_fork_cleanup_io;
1765
1766         if (pid != &init_struct_pid) {
1767                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1768                 if (IS_ERR(pid)) {
1769                         retval = PTR_ERR(pid);
1770                         goto bad_fork_cleanup_thread;
1771                 }
1772         }
1773
1774 #ifdef CONFIG_BLOCK
1775         p->plug = NULL;
1776 #endif
1777 #ifdef CONFIG_FUTEX
1778         p->robust_list = NULL;
1779 #ifdef CONFIG_COMPAT
1780         p->compat_robust_list = NULL;
1781 #endif
1782         INIT_LIST_HEAD(&p->pi_state_list);
1783         p->pi_state_cache = NULL;
1784 #endif
1785         /*
1786          * sigaltstack should be cleared when sharing the same VM
1787          */
1788         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1789                 sas_ss_reset(p);
1790
1791         /*
1792          * Syscall tracing and stepping should be turned off in the
1793          * child regardless of CLONE_PTRACE.
1794          */
1795         user_disable_single_step(p);
1796         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1797 #ifdef TIF_SYSCALL_EMU
1798         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1799 #endif
1800         clear_all_latency_tracing(p);
1801
1802         /* ok, now we should be set up.. */
1803         p->pid = pid_nr(pid);
1804         if (clone_flags & CLONE_THREAD) {
1805                 p->exit_signal = -1;
1806                 p->group_leader = current->group_leader;
1807                 p->tgid = current->tgid;
1808         } else {
1809                 if (clone_flags & CLONE_PARENT)
1810                         p->exit_signal = current->group_leader->exit_signal;
1811                 else
1812                         p->exit_signal = (clone_flags & CSIGNAL);
1813                 p->group_leader = p;
1814                 p->tgid = p->pid;
1815         }
1816
1817         p->nr_dirtied = 0;
1818         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1819         p->dirty_paused_when = 0;
1820
1821         p->pdeath_signal = 0;
1822         INIT_LIST_HEAD(&p->thread_group);
1823         p->task_works = NULL;
1824
1825         cgroup_threadgroup_change_begin(current);
1826         /*
1827          * Ensure that the cgroup subsystem policies allow the new process to be
1828          * forked. It should be noted the the new process's css_set can be changed
1829          * between here and cgroup_post_fork() if an organisation operation is in
1830          * progress.
1831          */
1832         retval = cgroup_can_fork(p);
1833         if (retval)
1834                 goto bad_fork_free_pid;
1835
1836         /*
1837          * Make it visible to the rest of the system, but dont wake it up yet.
1838          * Need tasklist lock for parent etc handling!
1839          */
1840         write_lock_irq(&tasklist_lock);
1841
1842         /* CLONE_PARENT re-uses the old parent */
1843         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1844                 p->real_parent = current->real_parent;
1845                 p->parent_exec_id = current->parent_exec_id;
1846         } else {
1847                 p->real_parent = current;
1848                 p->parent_exec_id = current->self_exec_id;
1849         }
1850
1851         klp_copy_process(p);
1852
1853         spin_lock(&current->sighand->siglock);
1854
1855         /*
1856          * Copy seccomp details explicitly here, in case they were changed
1857          * before holding sighand lock.
1858          */
1859         copy_seccomp(p);
1860
1861         /*
1862          * Process group and session signals need to be delivered to just the
1863          * parent before the fork or both the parent and the child after the
1864          * fork. Restart if a signal comes in before we add the new process to
1865          * it's process group.
1866          * A fatal signal pending means that current will exit, so the new
1867          * thread can't slip out of an OOM kill (or normal SIGKILL).
1868         */
1869         recalc_sigpending();
1870         if (signal_pending(current)) {
1871                 retval = -ERESTARTNOINTR;
1872                 goto bad_fork_cancel_cgroup;
1873         }
1874         if (unlikely(!(ns_of_pid(pid)->nr_hashed & PIDNS_HASH_ADDING))) {
1875                 retval = -ENOMEM;
1876                 goto bad_fork_cancel_cgroup;
1877         }
1878
1879         if (likely(p->pid)) {
1880                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1881
1882                 init_task_pid(p, PIDTYPE_PID, pid);
1883                 if (thread_group_leader(p)) {
1884                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1885                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1886
1887                         if (is_child_reaper(pid)) {
1888                                 ns_of_pid(pid)->child_reaper = p;
1889                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1890                         }
1891
1892                         p->signal->leader_pid = pid;
1893                         p->signal->tty = tty_kref_get(current->signal->tty);
1894                         /*
1895                          * Inherit has_child_subreaper flag under the same
1896                          * tasklist_lock with adding child to the process tree
1897                          * for propagate_has_child_subreaper optimization.
1898                          */
1899                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1900                                                          p->real_parent->signal->is_child_subreaper;
1901                         list_add_tail(&p->sibling, &p->real_parent->children);
1902                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1903                         attach_pid(p, PIDTYPE_PGID);
1904                         attach_pid(p, PIDTYPE_SID);
1905                         __this_cpu_inc(process_counts);
1906                 } else {
1907                         current->signal->nr_threads++;
1908                         atomic_inc(&current->signal->live);
1909                         atomic_inc(&current->signal->sigcnt);
1910                         list_add_tail_rcu(&p->thread_group,
1911                                           &p->group_leader->thread_group);
1912                         list_add_tail_rcu(&p->thread_node,
1913                                           &p->signal->thread_head);
1914                 }
1915                 attach_pid(p, PIDTYPE_PID);
1916                 nr_threads++;
1917         }
1918
1919         total_forks++;
1920         spin_unlock(&current->sighand->siglock);
1921         syscall_tracepoint_update(p);
1922         write_unlock_irq(&tasklist_lock);
1923
1924         proc_fork_connector(p);
1925         cgroup_post_fork(p);
1926         cgroup_threadgroup_change_end(current);
1927         perf_event_fork(p);
1928
1929         trace_task_newtask(p, clone_flags);
1930         uprobe_copy_process(p, clone_flags);
1931
1932         return p;
1933
1934 bad_fork_cancel_cgroup:
1935         spin_unlock(&current->sighand->siglock);
1936         write_unlock_irq(&tasklist_lock);
1937         cgroup_cancel_fork(p);
1938 bad_fork_free_pid:
1939         cgroup_threadgroup_change_end(current);
1940         if (pid != &init_struct_pid)
1941                 free_pid(pid);
1942 bad_fork_cleanup_thread:
1943         exit_thread(p);
1944 bad_fork_cleanup_io:
1945         if (p->io_context)
1946                 exit_io_context(p);
1947 bad_fork_cleanup_namespaces:
1948         exit_task_namespaces(p);
1949 bad_fork_cleanup_mm:
1950         if (p->mm)
1951                 mmput(p->mm);
1952 bad_fork_cleanup_signal:
1953         if (!(clone_flags & CLONE_THREAD))
1954                 free_signal_struct(p->signal);
1955 bad_fork_cleanup_sighand:
1956         __cleanup_sighand(p->sighand);
1957 bad_fork_cleanup_fs:
1958         exit_fs(p); /* blocking */
1959 bad_fork_cleanup_files:
1960         exit_files(p); /* blocking */
1961 bad_fork_cleanup_semundo:
1962         exit_sem(p);
1963 bad_fork_cleanup_security:
1964         security_task_free(p);
1965 bad_fork_cleanup_audit:
1966         audit_free(p);
1967 bad_fork_cleanup_perf:
1968         perf_event_free_task(p);
1969 bad_fork_cleanup_policy:
1970         lockdep_free_task(p);
1971 #ifdef CONFIG_NUMA
1972         mpol_put(p->mempolicy);
1973 bad_fork_cleanup_threadgroup_lock:
1974 #endif
1975         delayacct_tsk_free(p);
1976 bad_fork_cleanup_count:
1977         atomic_dec(&p->cred->user->processes);
1978         exit_creds(p);
1979 bad_fork_free:
1980         p->state = TASK_DEAD;
1981         put_task_stack(p);
1982         free_task(p);
1983 fork_out:
1984         return ERR_PTR(retval);
1985 }
1986
1987 static inline void init_idle_pids(struct pid_link *links)
1988 {
1989         enum pid_type type;
1990
1991         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1992                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1993                 links[type].pid = &init_struct_pid;
1994         }
1995 }
1996
1997 struct task_struct *fork_idle(int cpu)
1998 {
1999         struct task_struct *task;
2000         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
2001                             cpu_to_node(cpu));
2002         if (!IS_ERR(task)) {
2003                 init_idle_pids(task->pids);
2004                 init_idle(task, cpu);
2005         }
2006
2007         return task;
2008 }
2009
2010 /*
2011  *  Ok, this is the main fork-routine.
2012  *
2013  * It copies the process, and if successful kick-starts
2014  * it and waits for it to finish using the VM if required.
2015  */
2016 long _do_fork(unsigned long clone_flags,
2017               unsigned long stack_start,
2018               unsigned long stack_size,
2019               int __user *parent_tidptr,
2020               int __user *child_tidptr,
2021               unsigned long tls)
2022 {
2023         struct task_struct *p;
2024         int trace = 0;
2025         long nr;
2026
2027         /*
2028          * Determine whether and which event to report to ptracer.  When
2029          * called from kernel_thread or CLONE_UNTRACED is explicitly
2030          * requested, no event is reported; otherwise, report if the event
2031          * for the type of forking is enabled.
2032          */
2033         if (!(clone_flags & CLONE_UNTRACED)) {
2034                 if (clone_flags & CLONE_VFORK)
2035                         trace = PTRACE_EVENT_VFORK;
2036                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2037                         trace = PTRACE_EVENT_CLONE;
2038                 else
2039                         trace = PTRACE_EVENT_FORK;
2040
2041                 if (likely(!ptrace_event_enabled(current, trace)))
2042                         trace = 0;
2043         }
2044
2045         p = copy_process(clone_flags, stack_start, stack_size,
2046                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2047         add_latent_entropy();
2048         /*
2049          * Do this prior waking up the new thread - the thread pointer
2050          * might get invalid after that point, if the thread exits quickly.
2051          */
2052         if (!IS_ERR(p)) {
2053                 struct completion vfork;
2054                 struct pid *pid;
2055
2056                 trace_sched_process_fork(current, p);
2057
2058                 pid = get_task_pid(p, PIDTYPE_PID);
2059                 nr = pid_vnr(pid);
2060
2061                 if (clone_flags & CLONE_PARENT_SETTID)
2062                         put_user(nr, parent_tidptr);
2063
2064                 if (clone_flags & CLONE_VFORK) {
2065                         p->vfork_done = &vfork;
2066                         init_completion(&vfork);
2067                         get_task_struct(p);
2068                 }
2069
2070                 wake_up_new_task(p);
2071
2072                 /* forking complete and child started to run, tell ptracer */
2073                 if (unlikely(trace))
2074                         ptrace_event_pid(trace, pid);
2075
2076                 if (clone_flags & CLONE_VFORK) {
2077                         if (!wait_for_vfork_done(p, &vfork))
2078                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2079                 }
2080
2081                 put_pid(pid);
2082         } else {
2083                 nr = PTR_ERR(p);
2084         }
2085         return nr;
2086 }
2087
2088 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2089 /* For compatibility with architectures that call do_fork directly rather than
2090  * using the syscall entry points below. */
2091 long do_fork(unsigned long clone_flags,
2092               unsigned long stack_start,
2093               unsigned long stack_size,
2094               int __user *parent_tidptr,
2095               int __user *child_tidptr)
2096 {
2097         return _do_fork(clone_flags, stack_start, stack_size,
2098                         parent_tidptr, child_tidptr, 0);
2099 }
2100 #endif
2101
2102 /*
2103  * Create a kernel thread.
2104  */
2105 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2106 {
2107         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2108                 (unsigned long)arg, NULL, NULL, 0);
2109 }
2110
2111 #ifdef __ARCH_WANT_SYS_FORK
2112 SYSCALL_DEFINE0(fork)
2113 {
2114 #ifdef CONFIG_MMU
2115         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2116 #else
2117         /* can not support in nommu mode */
2118         return -EINVAL;
2119 #endif
2120 }
2121 #endif
2122
2123 #ifdef __ARCH_WANT_SYS_VFORK
2124 SYSCALL_DEFINE0(vfork)
2125 {
2126         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2127                         0, NULL, NULL, 0);
2128 }
2129 #endif
2130
2131 #ifdef __ARCH_WANT_SYS_CLONE
2132 #ifdef CONFIG_CLONE_BACKWARDS
2133 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2134                  int __user *, parent_tidptr,
2135                  unsigned long, tls,
2136                  int __user *, child_tidptr)
2137 #elif defined(CONFIG_CLONE_BACKWARDS2)
2138 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2139                  int __user *, parent_tidptr,
2140                  int __user *, child_tidptr,
2141                  unsigned long, tls)
2142 #elif defined(CONFIG_CLONE_BACKWARDS3)
2143 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2144                 int, stack_size,
2145                 int __user *, parent_tidptr,
2146                 int __user *, child_tidptr,
2147                 unsigned long, tls)
2148 #else
2149 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2150                  int __user *, parent_tidptr,
2151                  int __user *, child_tidptr,
2152                  unsigned long, tls)
2153 #endif
2154 {
2155         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2156 }
2157 #endif
2158
2159 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2160 {
2161         struct task_struct *leader, *parent, *child;
2162         int res;
2163
2164         read_lock(&tasklist_lock);
2165         leader = top = top->group_leader;
2166 down:
2167         for_each_thread(leader, parent) {
2168                 list_for_each_entry(child, &parent->children, sibling) {
2169                         res = visitor(child, data);
2170                         if (res) {
2171                                 if (res < 0)
2172                                         goto out;
2173                                 leader = child;
2174                                 goto down;
2175                         }
2176 up:
2177                         ;
2178                 }
2179         }
2180
2181         if (leader != top) {
2182                 child = leader;
2183                 parent = child->real_parent;
2184                 leader = parent->group_leader;
2185                 goto up;
2186         }
2187 out:
2188         read_unlock(&tasklist_lock);
2189 }
2190
2191 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2192 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2193 #endif
2194
2195 static void sighand_ctor(void *data)
2196 {
2197         struct sighand_struct *sighand = data;
2198
2199         spin_lock_init(&sighand->siglock);
2200         init_waitqueue_head(&sighand->signalfd_wqh);
2201 }
2202
2203 void __init proc_caches_init(void)
2204 {
2205         sighand_cachep = kmem_cache_create("sighand_cache",
2206                         sizeof(struct sighand_struct), 0,
2207                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2208                         SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2209         signal_cachep = kmem_cache_create("signal_cache",
2210                         sizeof(struct signal_struct), 0,
2211                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2212                         NULL);
2213         files_cachep = kmem_cache_create("files_cache",
2214                         sizeof(struct files_struct), 0,
2215                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2216                         NULL);
2217         fs_cachep = kmem_cache_create("fs_cache",
2218                         sizeof(struct fs_struct), 0,
2219                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2220                         NULL);
2221         /*
2222          * FIXME! The "sizeof(struct mm_struct)" currently includes the
2223          * whole struct cpumask for the OFFSTACK case. We could change
2224          * this to *only* allocate as much of it as required by the
2225          * maximum number of CPU's we can ever have.  The cpumask_allocation
2226          * is at the end of the structure, exactly for that reason.
2227          */
2228         mm_cachep = kmem_cache_create("mm_struct",
2229                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2230                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2231                         NULL);
2232         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2233         mmap_init();
2234         nsproxy_cache_init();
2235 }
2236
2237 /*
2238  * Check constraints on flags passed to the unshare system call.
2239  */
2240 static int check_unshare_flags(unsigned long unshare_flags)
2241 {
2242         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2243                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2244                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2245                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2246                 return -EINVAL;
2247         /*
2248          * Not implemented, but pretend it works if there is nothing
2249          * to unshare.  Note that unsharing the address space or the
2250          * signal handlers also need to unshare the signal queues (aka
2251          * CLONE_THREAD).
2252          */
2253         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2254                 if (!thread_group_empty(current))
2255                         return -EINVAL;
2256         }
2257         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2258                 if (atomic_read(&current->sighand->count) > 1)
2259                         return -EINVAL;
2260         }
2261         if (unshare_flags & CLONE_VM) {
2262                 if (!current_is_single_threaded())
2263                         return -EINVAL;
2264         }
2265
2266         return 0;
2267 }
2268
2269 /*
2270  * Unshare the filesystem structure if it is being shared
2271  */
2272 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2273 {
2274         struct fs_struct *fs = current->fs;
2275
2276         if (!(unshare_flags & CLONE_FS) || !fs)
2277                 return 0;
2278
2279         /* don't need lock here; in the worst case we'll do useless copy */
2280         if (fs->users == 1)
2281                 return 0;
2282
2283         *new_fsp = copy_fs_struct(fs);
2284         if (!*new_fsp)
2285                 return -ENOMEM;
2286
2287         return 0;
2288 }
2289
2290 /*
2291  * Unshare file descriptor table if it is being shared
2292  */
2293 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2294 {
2295         struct files_struct *fd = current->files;
2296         int error = 0;
2297
2298         if ((unshare_flags & CLONE_FILES) &&
2299             (fd && atomic_read(&fd->count) > 1)) {
2300                 *new_fdp = dup_fd(fd, &error);
2301                 if (!*new_fdp)
2302                         return error;
2303         }
2304
2305         return 0;
2306 }
2307
2308 /*
2309  * unshare allows a process to 'unshare' part of the process
2310  * context which was originally shared using clone.  copy_*
2311  * functions used by do_fork() cannot be used here directly
2312  * because they modify an inactive task_struct that is being
2313  * constructed. Here we are modifying the current, active,
2314  * task_struct.
2315  */
2316 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2317 {
2318         struct fs_struct *fs, *new_fs = NULL;
2319         struct files_struct *fd, *new_fd = NULL;
2320         struct cred *new_cred = NULL;
2321         struct nsproxy *new_nsproxy = NULL;
2322         int do_sysvsem = 0;
2323         int err;
2324
2325         /*
2326          * If unsharing a user namespace must also unshare the thread group
2327          * and unshare the filesystem root and working directories.
2328          */
2329         if (unshare_flags & CLONE_NEWUSER)
2330                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2331         /*
2332          * If unsharing vm, must also unshare signal handlers.
2333          */
2334         if (unshare_flags & CLONE_VM)
2335                 unshare_flags |= CLONE_SIGHAND;
2336         /*
2337          * If unsharing a signal handlers, must also unshare the signal queues.
2338          */
2339         if (unshare_flags & CLONE_SIGHAND)
2340                 unshare_flags |= CLONE_THREAD;
2341         /*
2342          * If unsharing namespace, must also unshare filesystem information.
2343          */
2344         if (unshare_flags & CLONE_NEWNS)
2345                 unshare_flags |= CLONE_FS;
2346
2347         err = check_unshare_flags(unshare_flags);
2348         if (err)
2349                 goto bad_unshare_out;
2350         /*
2351          * CLONE_NEWIPC must also detach from the undolist: after switching
2352          * to a new ipc namespace, the semaphore arrays from the old
2353          * namespace are unreachable.
2354          */
2355         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2356                 do_sysvsem = 1;
2357         err = unshare_fs(unshare_flags, &new_fs);
2358         if (err)
2359                 goto bad_unshare_out;
2360         err = unshare_fd(unshare_flags, &new_fd);
2361         if (err)
2362                 goto bad_unshare_cleanup_fs;
2363         err = unshare_userns(unshare_flags, &new_cred);
2364         if (err)
2365                 goto bad_unshare_cleanup_fd;
2366         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2367                                          new_cred, new_fs);
2368         if (err)
2369                 goto bad_unshare_cleanup_cred;
2370
2371         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2372                 if (do_sysvsem) {
2373                         /*
2374                          * CLONE_SYSVSEM is equivalent to sys_exit().
2375                          */
2376                         exit_sem(current);
2377                 }
2378                 if (unshare_flags & CLONE_NEWIPC) {
2379                         /* Orphan segments in old ns (see sem above). */
2380                         exit_shm(current);
2381                         shm_init_task(current);
2382                 }
2383
2384                 if (new_nsproxy)
2385                         switch_task_namespaces(current, new_nsproxy);
2386
2387                 task_lock(current);
2388
2389                 if (new_fs) {
2390                         fs = current->fs;
2391                         spin_lock(&fs->lock);
2392                         current->fs = new_fs;
2393                         if (--fs->users)
2394                                 new_fs = NULL;
2395                         else
2396                                 new_fs = fs;
2397                         spin_unlock(&fs->lock);
2398                 }
2399
2400                 if (new_fd) {
2401                         fd = current->files;
2402                         current->files = new_fd;
2403                         new_fd = fd;
2404                 }
2405
2406                 task_unlock(current);
2407
2408                 if (new_cred) {
2409                         /* Install the new user namespace */
2410                         commit_creds(new_cred);
2411                         new_cred = NULL;
2412                 }
2413         }
2414
2415         perf_event_namespaces(current);
2416
2417 bad_unshare_cleanup_cred:
2418         if (new_cred)
2419                 put_cred(new_cred);
2420 bad_unshare_cleanup_fd:
2421         if (new_fd)
2422                 put_files_struct(new_fd);
2423
2424 bad_unshare_cleanup_fs:
2425         if (new_fs)
2426                 free_fs_struct(new_fs);
2427
2428 bad_unshare_out:
2429         return err;
2430 }
2431
2432 /*
2433  *      Helper to unshare the files of the current task.
2434  *      We don't want to expose copy_files internals to
2435  *      the exec layer of the kernel.
2436  */
2437
2438 int unshare_files(struct files_struct **displaced)
2439 {
2440         struct task_struct *task = current;
2441         struct files_struct *copy = NULL;
2442         int error;
2443
2444         error = unshare_fd(CLONE_FILES, &copy);
2445         if (error || !copy) {
2446                 *displaced = NULL;
2447                 return error;
2448         }
2449         *displaced = task->files;
2450         task_lock(task);
2451         task->files = copy;
2452         task_unlock(task);
2453         return 0;
2454 }
2455
2456 int sysctl_max_threads(struct ctl_table *table, int write,
2457                        void __user *buffer, size_t *lenp, loff_t *ppos)
2458 {
2459         struct ctl_table t;
2460         int ret;
2461         int threads = max_threads;
2462         int min = MIN_THREADS;
2463         int max = MAX_THREADS;
2464
2465         t = *table;
2466         t.data = &threads;
2467         t.extra1 = &min;
2468         t.extra2 = &max;
2469
2470         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2471         if (ret || !write)
2472                 return ret;
2473
2474         set_max_threads(threads);
2475
2476         return 0;
2477 }