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 #include <linux/cpufreq_times.h>
  94
  95 #include <asm/pgtable.h>
  96 #include <asm/pgalloc.h>
  97 #include <linux/uaccess.h>
  98 #include <asm/mmu_context.h>
  99 #include <asm/cacheflush.h>
 100 #include <asm/tlbflush.h>
 101
 102 #include <trace/events/sched.h>
 103
 104 #define CREATE_TRACE_POINTS
 105 #include <trace/events/task.h>
 106
 107 /*
 108  * Minimum number of threads to boot the kernel
 109  */
 110 #define MIN_THREADS 20
 111
 112 /*
 113  * Maximum number of threads
 114  */
 115 #define MAX_THREADS FUTEX_TID_MASK
 116
 117 /*
 118  * Protected counters by write_lock_irq(&tasklist_lock)
 119  */
 120 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 121 int nr_threads;                 /* The idle threads do not count.. */
 122
 123 int max_threads;                /* tunable limit on nr_threads */
 124
 125 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 126
 127 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 128
 129 #ifdef CONFIG_PROVE_RCU
 130 int lockdep_tasklist_lock_is_held(void)
 131 {
 132         return lockdep_is_held(&tasklist_lock);
 133 }
 134 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 135 #endif /* #ifdef CONFIG_PROVE_RCU */
 136
 137 int nr_processes(void)
 138 {
 139         int cpu;
 140         int total = 0;
 141
 142         for_each_possible_cpu(cpu)
 143                 total += per_cpu(process_counts, cpu);
 144
 145         return total;
 146 }
 147
 148 void __weak arch_release_task_struct(struct task_struct *tsk)
 149 {
 150 }
 151
 152 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 153 static struct kmem_cache *task_struct_cachep;
 154
 155 static inline struct task_struct *alloc_task_struct_node(int node)
 156 {
 157         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 158 }
 159
 160 static inline void free_task_struct(struct task_struct *tsk)
 161 {
 162         kmem_cache_free(task_struct_cachep, tsk);
 163 }
 164 #endif
 165
 166 void __weak arch_release_thread_stack(unsigned long *stack)
 167 {
 168 }
 169
 170 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 171
 172 /*
 173  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 174  * kmemcache based allocator.
 175  */
 176 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 177
 178 #ifdef CONFIG_VMAP_STACK
 179 /*
 180  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 181  * flush.  Try to minimize the number of calls by caching stacks.
 182  */
 183 #define NR_CACHED_STACKS 2
 184 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 185
 186 static int free_vm_stack_cache(unsigned int cpu)
 187 {
 188         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 189         int i;
 190
 191         for (i = 0; i < NR_CACHED_STACKS; i++) {
 192                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 193
 194                 if (!vm_stack)
 195                         continue;
 196
 197                 vfree(vm_stack->addr);
 198                 cached_vm_stacks[i] = NULL;
 199         }
 200
 201         return 0;
 202 }
 203 #endif
 204
 205 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 206 {
 207 #ifdef CONFIG_VMAP_STACK
 208         void *stack;
 209         int i;
 210
 211         for (i = 0; i < NR_CACHED_STACKS; i++) {
 212                 struct vm_struct *s;
 213
 214                 s = this_cpu_xchg(cached_stacks[i], NULL);
 215
 216                 if (!s)
 217                         continue;
 218
 219                 /* Clear stale pointers from reused stack. */
 220                 memset(s->addr, 0, THREAD_SIZE);
 221
 222                 tsk->stack_vm_area = s;
 223                 return s->addr;
 224         }
 225
 226         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 227                                      VMALLOC_START, VMALLOC_END,
 228                                      THREADINFO_GFP,
 229                                      PAGE_KERNEL,
 230                                      0, node, __builtin_return_address(0));
 231
 232         /*
 233          * We can't call find_vm_area() in interrupt context, and
 234          * free_thread_stack() can be called in interrupt context,
 235          * so cache the vm_struct.
 236          */
 237         if (stack)
 238                 tsk->stack_vm_area = find_vm_area(stack);
 239         return stack;
 240 #else
 241         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 242                                              THREAD_SIZE_ORDER);
 243
 244         return page ? page_address(page) : NULL;
 245 #endif
 246 }
 247
 248 static inline void free_thread_stack(struct task_struct *tsk)
 249 {
 250 #ifdef CONFIG_VMAP_STACK
 251         if (task_stack_vm_area(tsk)) {
 252                 int i;
 253
 254                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 255                         if (this_cpu_cmpxchg(cached_stacks[i],
 256                                         NULL, tsk->stack_vm_area) != NULL)
 257                                 continue;
 258
 259                         return;
 260                 }
 261
 262                 vfree_atomic(tsk->stack);
 263                 return;
 264         }
 265 #endif
 266
 267         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 268 }
 269 # else
 270 static struct kmem_cache *thread_stack_cache;
 271
 272 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 273                                                   int node)
 274 {
 275         return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 276 }
 277
 278 static void free_thread_stack(struct task_struct *tsk)
 279 {
 280         kmem_cache_free(thread_stack_cache, tsk->stack);
 281 }
 282
 283 void thread_stack_cache_init(void)
 284 {
 285         thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
 286                                               THREAD_SIZE, 0, NULL);
 287         BUG_ON(thread_stack_cache == NULL);
 288 }
 289 # endif
 290 #endif
 291
 292 /* SLAB cache for signal_struct structures (tsk->signal) */
 293 static struct kmem_cache *signal_cachep;
 294
 295 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 296 struct kmem_cache *sighand_cachep;
 297
 298 /* SLAB cache for files_struct structures (tsk->files) */
 299 struct kmem_cache *files_cachep;
 300
 301 /* SLAB cache for fs_struct structures (tsk->fs) */
 302 struct kmem_cache *fs_cachep;
 303
 304 /* SLAB cache for vm_area_struct structures */
 305 struct kmem_cache *vm_area_cachep;
 306
 307 /* SLAB cache for mm_struct structures (tsk->mm) */
 308 static struct kmem_cache *mm_cachep;
 309
 310 static void account_kernel_stack(struct task_struct *tsk, int account)
 311 {
 312         void *stack = task_stack_page(tsk);
 313         struct vm_struct *vm = task_stack_vm_area(tsk);
 314
 315         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 316
 317         if (vm) {
 318                 int i;
 319
 320                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 321
 322                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 323                         mod_zone_page_state(page_zone(vm->pages[i]),
 324                                             NR_KERNEL_STACK_KB,
 325                                             PAGE_SIZE / 1024 * account);
 326                 }
 327
 328                 /* All stack pages belong to the same memcg. */
 329                 mod_memcg_page_state(vm->pages[0], MEMCG_KERNEL_STACK_KB,
 330                                      account * (THREAD_SIZE / 1024));
 331         } else {
 332                 /*
 333                  * All stack pages are in the same zone and belong to the
 334                  * same memcg.
 335                  */
 336                 struct page *first_page = virt_to_page(stack);
 337
 338                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 339                                     THREAD_SIZE / 1024 * account);
 340
 341                 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
 342                                      account * (THREAD_SIZE / 1024));
 343         }
 344 }
 345
 346 static void release_task_stack(struct task_struct *tsk)
 347 {
 348         if (WARN_ON(tsk->state != TASK_DEAD))
 349                 return;  /* Better to leak the stack than to free prematurely */
 350
 351         account_kernel_stack(tsk, -1);
 352         arch_release_thread_stack(tsk->stack);
 353         free_thread_stack(tsk);
 354         tsk->stack = NULL;
 355 #ifdef CONFIG_VMAP_STACK
 356         tsk->stack_vm_area = NULL;
 357 #endif
 358 }
 359
 360 #ifdef CONFIG_THREAD_INFO_IN_TASK
 361 void put_task_stack(struct task_struct *tsk)
 362 {
 363         if (atomic_dec_and_test(&tsk->stack_refcount))
 364                 release_task_stack(tsk);
 365 }
 366 #endif
 367
 368 void free_task(struct task_struct *tsk)
 369 {
 370         cpufreq_task_times_exit(tsk);
 371
 372 #ifndef CONFIG_THREAD_INFO_IN_TASK
 373         /*
 374          * The task is finally done with both the stack and thread_info,
 375          * so free both.
 376          */
 377         release_task_stack(tsk);
 378 #else
 379         /*
 380          * If the task had a separate stack allocation, it should be gone
 381          * by now.
 382          */
 383         WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
 384 #endif
 385         rt_mutex_debug_task_free(tsk);
 386         ftrace_graph_exit_task(tsk);
 387         put_seccomp_filter(tsk);
 388         arch_release_task_struct(tsk);
 389         if (tsk->flags & PF_KTHREAD)
 390                 free_kthread_struct(tsk);
 391         free_task_struct(tsk);
 392 }
 393 EXPORT_SYMBOL(free_task);
 394
 395 static inline void free_signal_struct(struct signal_struct *sig)
 396 {
 397         taskstats_tgid_free(sig);
 398         sched_autogroup_exit(sig);
 399         /*
 400          * __mmdrop is not safe to call from softirq context on x86 due to
 401          * pgd_dtor so postpone it to the async context
 402          */
 403         if (sig->oom_mm)
 404                 mmdrop_async(sig->oom_mm);
 405         kmem_cache_free(signal_cachep, sig);
 406 }
 407
 408 static inline void put_signal_struct(struct signal_struct *sig)
 409 {
 410         if (atomic_dec_and_test(&sig->sigcnt))
 411                 free_signal_struct(sig);
 412 }
 413
 414 void __put_task_struct(struct task_struct *tsk)
 415 {
 416         WARN_ON(!tsk->exit_state);
 417         WARN_ON(atomic_read(&tsk->usage));
 418         WARN_ON(tsk == current);
 419
 420         cgroup_free(tsk);
 421         task_numa_free(tsk);
 422         security_task_free(tsk);
 423         exit_creds(tsk);
 424         delayacct_tsk_free(tsk);
 425         put_signal_struct(tsk->signal);
 426
 427         if (!profile_handoff_task(tsk))
 428                 free_task(tsk);
 429 }
 430 EXPORT_SYMBOL_GPL(__put_task_struct);
 431
 432 void __init __weak arch_task_cache_init(void) { }
 433
 434 /*
 435  * set_max_threads
 436  */
 437 static void set_max_threads(unsigned int max_threads_suggested)
 438 {
 439         u64 threads;
 440
 441         /*
 442          * The number of threads shall be limited such that the thread
 443          * structures may only consume a small part of the available memory.
 444          */
 445         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 446                 threads = MAX_THREADS;
 447         else
 448                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 449                                     (u64) THREAD_SIZE * 8UL);
 450
 451         if (threads > max_threads_suggested)
 452                 threads = max_threads_suggested;
 453
 454         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 455 }
 456
 457 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 458 /* Initialized by the architecture: */
 459 int arch_task_struct_size __read_mostly;
 460 #endif
 461
 462 void __init fork_init(void)
 463 {
 464         int i;
 465 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 466 #ifndef ARCH_MIN_TASKALIGN
 467 #define ARCH_MIN_TASKALIGN      0
 468 #endif
 469         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 470
 471         /* create a slab on which task_structs can be allocated */
 472         task_struct_cachep = kmem_cache_create("task_struct",
 473                         arch_task_struct_size, align,
 474                         SLAB_PANIC|SLAB_ACCOUNT, NULL);
 475 #endif
 476
 477         /* do the arch specific task caches init */
 478         arch_task_cache_init();
 479
 480         set_max_threads(MAX_THREADS);
 481
 482         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 483         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 484         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 485                 init_task.signal->rlim[RLIMIT_NPROC];
 486
 487         for (i = 0; i < UCOUNT_COUNTS; i++) {
 488                 init_user_ns.ucount_max[i] = max_threads/2;
 489         }
 490
 491 #ifdef CONFIG_VMAP_STACK
 492         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 493                           NULL, free_vm_stack_cache);
 494 #endif
 495
 496         lockdep_init_task(&init_task);
 497 }
 498
 499 int __weak arch_dup_task_struct(struct task_struct *dst,
 500                                                struct task_struct *src)
 501 {
 502         *dst = *src;
 503         return 0;
 504 }
 505
 506 void set_task_stack_end_magic(struct task_struct *tsk)
 507 {
 508         unsigned long *stackend;
 509
 510         stackend = end_of_stack(tsk);
 511         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 512 }
 513
 514 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 515 {
 516         struct task_struct *tsk;
 517         unsigned long *stack;
 518         struct vm_struct *stack_vm_area;
 519         int err;
 520
 521         if (node == NUMA_NO_NODE)
 522                 node = tsk_fork_get_node(orig);
 523         tsk = alloc_task_struct_node(node);
 524         if (!tsk)
 525                 return NULL;
 526
 527         stack = alloc_thread_stack_node(tsk, node);
 528         if (!stack)
 529                 goto free_tsk;
 530
 531         stack_vm_area = task_stack_vm_area(tsk);
 532
 533         err = arch_dup_task_struct(tsk, orig);
 534
 535         /*
 536          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 537          * sure they're properly initialized before using any stack-related
 538          * functions again.
 539          */
 540         tsk->stack = stack;
 541 #ifdef CONFIG_VMAP_STACK
 542         tsk->stack_vm_area = stack_vm_area;
 543 #endif
 544 #ifdef CONFIG_THREAD_INFO_IN_TASK
 545         atomic_set(&tsk->stack_refcount, 1);
 546 #endif
 547
 548         if (err)
 549                 goto free_stack;
 550
 551 #ifdef CONFIG_SECCOMP
 552         /*
 553          * We must handle setting up seccomp filters once we're under
 554          * the sighand lock in case orig has changed between now and
 555          * then. Until then, filter must be NULL to avoid messing up
 556          * the usage counts on the error path calling free_task.
 557          */
 558         tsk->seccomp.filter = NULL;
 559 #endif
 560
 561         setup_thread_stack(tsk, orig);
 562         clear_user_return_notifier(tsk);
 563         clear_tsk_need_resched(tsk);
 564         set_task_stack_end_magic(tsk);
 565
 566 #ifdef CONFIG_CC_STACKPROTECTOR
 567         tsk->stack_canary = get_random_canary();
 568 #endif
 569
 570         /*
 571          * One for us, one for whoever does the "release_task()" (usually
 572          * parent)
 573          */
 574         atomic_set(&tsk->usage, 2);
 575 #ifdef CONFIG_BLK_DEV_IO_TRACE
 576         tsk->btrace_seq = 0;
 577 #endif
 578         tsk->splice_pipe = NULL;
 579         tsk->task_frag.page = NULL;
 580         tsk->wake_q.next = NULL;
 581
 582         account_kernel_stack(tsk, 1);
 583
 584         kcov_task_init(tsk);
 585
 586 #ifdef CONFIG_FAULT_INJECTION
 587         tsk->fail_nth = 0;
 588 #endif
 589
 590         return tsk;
 591
 592 free_stack:
 593         free_thread_stack(tsk);
 594 free_tsk:
 595         free_task_struct(tsk);
 596         return NULL;
 597 }
 598
 599 #ifdef CONFIG_MMU
 600 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 601                                         struct mm_struct *oldmm)
 602 {
 603         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 604         struct rb_node **rb_link, *rb_parent;
 605         int retval;
 606         unsigned long charge;
 607         LIST_HEAD(uf);
 608
 609         uprobe_start_dup_mmap();
 610         if (down_write_killable(&oldmm->mmap_sem)) {
 611                 retval = -EINTR;
 612                 goto fail_uprobe_end;
 613         }
 614         flush_cache_dup_mm(oldmm);
 615         uprobe_dup_mmap(oldmm, mm);
 616         /*
 617          * Not linked in yet - no deadlock potential:
 618          */
 619         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 620
 621         /* No ordering required: file already has been exposed. */
 622         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 623
 624         mm->total_vm = oldmm->total_vm;
 625         mm->data_vm = oldmm->data_vm;
 626         mm->exec_vm = oldmm->exec_vm;
 627         mm->stack_vm = oldmm->stack_vm;
 628
 629         rb_link = &mm->mm_rb.rb_node;
 630         rb_parent = NULL;
 631         pprev = &mm->mmap;
 632         retval = ksm_fork(mm, oldmm);
 633         if (retval)
 634                 goto out;
 635         retval = khugepaged_fork(mm, oldmm);
 636         if (retval)
 637                 goto out;
 638
 639         prev = NULL;
 640         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 641                 struct file *file;
 642
 643                 if (mpnt->vm_flags & VM_DONTCOPY) {
 644                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 645                         continue;
 646                 }
 647                 charge = 0;
 648                 if (mpnt->vm_flags & VM_ACCOUNT) {
 649                         unsigned long len = vma_pages(mpnt);
 650
 651                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 652                                 goto fail_nomem;
 653                         charge = len;
 654                 }
 655                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 656                 if (!tmp)
 657                         goto fail_nomem;
 658                 *tmp = *mpnt;
 659                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 660                 retval = vma_dup_policy(mpnt, tmp);
 661                 if (retval)
 662                         goto fail_nomem_policy;
 663                 tmp->vm_mm = mm;
 664                 retval = dup_userfaultfd(tmp, &uf);
 665                 if (retval)
 666                         goto fail_nomem_anon_vma_fork;
 667                 if (tmp->vm_flags & VM_WIPEONFORK) {
 668                         /* VM_WIPEONFORK gets a clean slate in the child. */
 669                         tmp->anon_vma = NULL;
 670                         if (anon_vma_prepare(tmp))
 671                                 goto fail_nomem_anon_vma_fork;
 672                 } else if (anon_vma_fork(tmp, mpnt))
 673                         goto fail_nomem_anon_vma_fork;
 674                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 675                 tmp->vm_next = tmp->vm_prev = NULL;
 676                 file = tmp->vm_file;
 677                 if (file) {
 678                         struct inode *inode = file_inode(file);
 679                         struct address_space *mapping = file->f_mapping;
 680
 681                         get_file(file);
 682                         if (tmp->vm_flags & VM_DENYWRITE)
 683                                 atomic_dec(&inode->i_writecount);
 684                         i_mmap_lock_write(mapping);
 685                         if (tmp->vm_flags & VM_SHARED)
 686                                 atomic_inc(&mapping->i_mmap_writable);
 687                         flush_dcache_mmap_lock(mapping);
 688                         /* insert tmp into the share list, just after mpnt */
 689                         vma_interval_tree_insert_after(tmp, mpnt,
 690                                         &mapping->i_mmap);
 691                         flush_dcache_mmap_unlock(mapping);
 692                         i_mmap_unlock_write(mapping);
 693                 }
 694
 695                 /*
 696                  * Clear hugetlb-related page reserves for children. This only
 697                  * affects MAP_PRIVATE mappings. Faults generated by the child
 698                  * are not guaranteed to succeed, even if read-only
 699                  */
 700                 if (is_vm_hugetlb_page(tmp))
 701                         reset_vma_resv_huge_pages(tmp);
 702
 703                 /*
 704                  * Link in the new vma and copy the page table entries.
 705                  */
 706                 *pprev = tmp;
 707                 pprev = &tmp->vm_next;
 708                 tmp->vm_prev = prev;
 709                 prev = tmp;
 710
 711                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 712                 rb_link = &tmp->vm_rb.rb_right;
 713                 rb_parent = &tmp->vm_rb;
 714
 715                 mm->map_count++;
 716                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 717                         retval = copy_page_range(mm, oldmm, mpnt);
 718
 719                 if (tmp->vm_ops && tmp->vm_ops->open)
 720                         tmp->vm_ops->open(tmp);
 721
 722                 if (retval)
 723                         goto out;
 724         }
 725         /* a new mm has just been created */
 726         retval = arch_dup_mmap(oldmm, mm);
 727 out:
 728         up_write(&mm->mmap_sem);
 729         flush_tlb_mm(oldmm);
 730         up_write(&oldmm->mmap_sem);
 731         dup_userfaultfd_complete(&uf);
 732 fail_uprobe_end:
 733         uprobe_end_dup_mmap();
 734         return retval;
 735 fail_nomem_anon_vma_fork:
 736         mpol_put(vma_policy(tmp));
 737 fail_nomem_policy:
 738         kmem_cache_free(vm_area_cachep, tmp);
 739 fail_nomem:
 740         retval = -ENOMEM;
 741         vm_unacct_memory(charge);
 742         goto out;
 743 }
 744
 745 static inline int mm_alloc_pgd(struct mm_struct *mm)
 746 {
 747         mm->pgd = pgd_alloc(mm);
 748         if (unlikely(!mm->pgd))
 749                 return -ENOMEM;
 750         return 0;
 751 }
 752
 753 static inline void mm_free_pgd(struct mm_struct *mm)
 754 {
 755         pgd_free(mm, mm->pgd);
 756 }
 757 #else
 758 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 759 {
 760         down_write(&oldmm->mmap_sem);
 761         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 762         up_write(&oldmm->mmap_sem);
 763         return 0;
 764 }
 765 #define mm_alloc_pgd(mm)        (0)
 766 #define mm_free_pgd(mm)
 767 #endif /* CONFIG_MMU */
 768
 769 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 770
 771 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 772 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 773
 774 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 775
 776 static int __init coredump_filter_setup(char *s)
 777 {
 778         default_dump_filter =
 779                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 780                 MMF_DUMP_FILTER_MASK;
 781         return 1;
 782 }
 783
 784 __setup("coredump_filter=", coredump_filter_setup);
 785
 786 #include <linux/init_task.h>
 787
 788 static void mm_init_aio(struct mm_struct *mm)
 789 {
 790 #ifdef CONFIG_AIO
 791         spin_lock_init(&mm->ioctx_lock);
 792         mm->ioctx_table = NULL;
 793 #endif
 794 }
 795
 796 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 797 {
 798 #ifdef CONFIG_MEMCG
 799         mm->owner = p;
 800 #endif
 801 }
 802
 803 static void mm_init_uprobes_state(struct mm_struct *mm)
 804 {
 805 #ifdef CONFIG_UPROBES
 806         mm->uprobes_state.xol_area = NULL;
 807 #endif
 808 }
 809
 810 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 811         struct user_namespace *user_ns)
 812 {
 813         mm->mmap = NULL;
 814         mm->mm_rb = RB_ROOT;
 815         mm->vmacache_seqnum = 0;
 816         atomic_set(&mm->mm_users, 1);
 817         atomic_set(&mm->mm_count, 1);
 818         init_rwsem(&mm->mmap_sem);
 819         INIT_LIST_HEAD(&mm->mmlist);
 820         mm->core_state = NULL;
 821         atomic_long_set(&mm->nr_ptes, 0);
 822         mm_nr_pmds_init(mm);
 823         mm->map_count = 0;
 824         mm->locked_vm = 0;
 825         mm->pinned_vm = 0;
 826         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 827         spin_lock_init(&mm->page_table_lock);
 828         mm_init_cpumask(mm);
 829         mm_init_aio(mm);
 830         mm_init_owner(mm, p);
 831         RCU_INIT_POINTER(mm->exe_file, NULL);
 832         mmu_notifier_mm_init(mm);
 833         hmm_mm_init(mm);
 834         init_tlb_flush_pending(mm);
 835 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 836         mm->pmd_huge_pte = NULL;
 837 #endif
 838         mm_init_uprobes_state(mm);
 839
 840         if (current->mm) {
 841                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 842                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 843         } else {
 844                 mm->flags = default_dump_filter;
 845                 mm->def_flags = 0;
 846         }
 847
 848         if (mm_alloc_pgd(mm))
 849                 goto fail_nopgd;
 850
 851         if (init_new_context(p, mm))
 852                 goto fail_nocontext;
 853
 854         mm->user_ns = get_user_ns(user_ns);
 855         return mm;
 856
 857 fail_nocontext:
 858         mm_free_pgd(mm);
 859 fail_nopgd:
 860         free_mm(mm);
 861         return NULL;
 862 }
 863
 864 static void check_mm(struct mm_struct *mm)
 865 {
 866         int i;
 867
 868         for (i = 0; i < NR_MM_COUNTERS; i++) {
 869                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 870
 871                 if (unlikely(x))
 872                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 873                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 874         }
 875
 876         if (atomic_long_read(&mm->nr_ptes))
 877                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 878                                 atomic_long_read(&mm->nr_ptes));
 879         if (mm_nr_pmds(mm))
 880                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 881                                 mm_nr_pmds(mm));
 882
 883 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 884         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 885 #endif
 886 }
 887
 888 /*
 889  * Allocate and initialize an mm_struct.
 890  */
 891 struct mm_struct *mm_alloc(void)
 892 {
 893         struct mm_struct *mm;
 894
 895         mm = allocate_mm();
 896         if (!mm)
 897                 return NULL;
 898
 899         memset(mm, 0, sizeof(*mm));
 900         return mm_init(mm, current, current_user_ns());
 901 }
 902
 903 /*
 904  * Called when the last reference to the mm
 905  * is dropped: either by a lazy thread or by
 906  * mmput. Free the page directory and the mm.
 907  */
 908 void __mmdrop(struct mm_struct *mm)
 909 {
 910         BUG_ON(mm == &init_mm);
 911         mm_free_pgd(mm);
 912         destroy_context(mm);
 913         hmm_mm_destroy(mm);
 914         mmu_notifier_mm_destroy(mm);
 915         check_mm(mm);
 916         put_user_ns(mm->user_ns);
 917         free_mm(mm);
 918 }
 919 EXPORT_SYMBOL_GPL(__mmdrop);
 920
 921 static inline void __mmput(struct mm_struct *mm)
 922 {
 923         VM_BUG_ON(atomic_read(&mm->mm_users));
 924
 925         uprobe_clear_state(mm);
 926         exit_aio(mm);
 927         ksm_exit(mm);
 928         khugepaged_exit(mm); /* must run before exit_mmap */
 929         exit_mmap(mm);
 930         mm_put_huge_zero_page(mm);
 931         set_mm_exe_file(mm, NULL);
 932         if (!list_empty(&mm->mmlist)) {
 933                 spin_lock(&mmlist_lock);
 934                 list_del(&mm->mmlist);
 935                 spin_unlock(&mmlist_lock);
 936         }
 937         if (mm->binfmt)
 938                 module_put(mm->binfmt->module);
 939         mmdrop(mm);
 940 }
 941
 942 /*
 943  * Decrement the use count and release all resources for an mm.
 944  */
 945 void mmput(struct mm_struct *mm)
 946 {
 947         might_sleep();
 948
 949         if (atomic_dec_and_test(&mm->mm_users))
 950                 __mmput(mm);
 951 }
 952 EXPORT_SYMBOL_GPL(mmput);
 953
 954 #ifdef CONFIG_MMU
 955 static void mmput_async_fn(struct work_struct *work)
 956 {
 957         struct mm_struct *mm = container_of(work, struct mm_struct,
 958                                             async_put_work);
 959
 960         __mmput(mm);
 961 }
 962
 963 void mmput_async(struct mm_struct *mm)
 964 {
 965         if (atomic_dec_and_test(&mm->mm_users)) {
 966                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
 967                 schedule_work(&mm->async_put_work);
 968         }
 969 }
 970 #endif
 971
 972 /**
 973  * set_mm_exe_file - change a reference to the mm's executable file
 974  *
 975  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 976  *
 977  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 978  * invocations: in mmput() nobody alive left, in execve task is single
 979  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 980  * mm->exe_file, but does so without using set_mm_exe_file() in order
 981  * to do avoid the need for any locks.
 982  */
 983 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 984 {
 985         struct file *old_exe_file;
 986
 987         /*
 988          * It is safe to dereference the exe_file without RCU as
 989          * this function is only called if nobody else can access
 990          * this mm -- see comment above for justification.
 991          */
 992         old_exe_file = rcu_dereference_raw(mm->exe_file);
 993
 994         if (new_exe_file)
 995                 get_file(new_exe_file);
 996         rcu_assign_pointer(mm->exe_file, new_exe_file);
 997         if (old_exe_file)
 998                 fput(old_exe_file);
 999 }
1000
1001 /**
1002  * get_mm_exe_file - acquire a reference to the mm's executable file
1003  *
1004  * Returns %NULL if mm has no associated executable file.
1005  * User must release file via fput().
1006  */
1007 struct file *get_mm_exe_file(struct mm_struct *mm)
1008 {
1009         struct file *exe_file;
1010
1011         rcu_read_lock();
1012         exe_file = rcu_dereference(mm->exe_file);
1013         if (exe_file && !get_file_rcu(exe_file))
1014                 exe_file = NULL;
1015         rcu_read_unlock();
1016         return exe_file;
1017 }
1018 EXPORT_SYMBOL(get_mm_exe_file);
1019
1020 /**
1021  * get_task_exe_file - acquire a reference to the task's executable file
1022  *
1023  * Returns %NULL if task's mm (if any) has no associated executable file or
1024  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1025  * User must release file via fput().
1026  */
1027 struct file *get_task_exe_file(struct task_struct *task)
1028 {
1029         struct file *exe_file = NULL;
1030         struct mm_struct *mm;
1031
1032         task_lock(task);
1033         mm = task->mm;
1034         if (mm) {
1035                 if (!(task->flags & PF_KTHREAD))
1036                         exe_file = get_mm_exe_file(mm);
1037         }
1038         task_unlock(task);
1039         return exe_file;
1040 }
1041 EXPORT_SYMBOL(get_task_exe_file);
1042
1043 /**
1044  * get_task_mm - acquire a reference to the task's mm
1045  *
1046  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1047  * this kernel workthread has transiently adopted a user mm with use_mm,
1048  * to do its AIO) is not set and if so returns a reference to it, after
1049  * bumping up the use count.  User must release the mm via mmput()
1050  * after use.  Typically used by /proc and ptrace.
1051  */
1052 struct mm_struct *get_task_mm(struct task_struct *task)
1053 {
1054         struct mm_struct *mm;
1055
1056         task_lock(task);
1057         mm = task->mm;
1058         if (mm) {
1059                 if (task->flags & PF_KTHREAD)
1060                         mm = NULL;
1061                 else
1062                         mmget(mm);
1063         }
1064         task_unlock(task);
1065         return mm;
1066 }
1067 EXPORT_SYMBOL_GPL(get_task_mm);
1068
1069 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1070 {
1071         struct mm_struct *mm;
1072         int err;
1073
1074         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1075         if (err)
1076                 return ERR_PTR(err);
1077
1078         mm = get_task_mm(task);
1079         if (mm && mm != current->mm &&
1080                         !ptrace_may_access(task, mode)) {
1081                 mmput(mm);
1082                 mm = ERR_PTR(-EACCES);
1083         }
1084         mutex_unlock(&task->signal->cred_guard_mutex);
1085
1086         return mm;
1087 }
1088
1089 static void complete_vfork_done(struct task_struct *tsk)
1090 {
1091         struct completion *vfork;
1092
1093         task_lock(tsk);
1094         vfork = tsk->vfork_done;
1095         if (likely(vfork)) {
1096                 tsk->vfork_done = NULL;
1097                 complete(vfork);
1098         }
1099         task_unlock(tsk);
1100 }
1101
1102 static int wait_for_vfork_done(struct task_struct *child,
1103                                 struct completion *vfork)
1104 {
1105         int killed;
1106
1107         freezer_do_not_count();
1108         killed = wait_for_completion_killable(vfork);
1109         freezer_count();
1110
1111         if (killed) {
1112                 task_lock(child);
1113                 child->vfork_done = NULL;
1114                 task_unlock(child);
1115         }
1116
1117         put_task_struct(child);
1118         return killed;
1119 }
1120
1121 /* Please note the differences between mmput and mm_release.
1122  * mmput is called whenever we stop holding onto a mm_struct,
1123  * error success whatever.
1124  *
1125  * mm_release is called after a mm_struct has been removed
1126  * from the current process.
1127  *
1128  * This difference is important for error handling, when we
1129  * only half set up a mm_struct for a new process and need to restore
1130  * the old one.  Because we mmput the new mm_struct before
1131  * restoring the old one. . .
1132  * Eric Biederman 10 January 1998
1133  */
1134 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1135 {
1136         /* Get rid of any futexes when releasing the mm */
1137 #ifdef CONFIG_FUTEX
1138         if (unlikely(tsk->robust_list)) {
1139                 exit_robust_list(tsk);
1140                 tsk->robust_list = NULL;
1141         }
1142 #ifdef CONFIG_COMPAT
1143         if (unlikely(tsk->compat_robust_list)) {
1144                 compat_exit_robust_list(tsk);
1145                 tsk->compat_robust_list = NULL;
1146         }
1147 #endif
1148         if (unlikely(!list_empty(&tsk->pi_state_list)))
1149                 exit_pi_state_list(tsk);
1150 #endif
1151
1152         uprobe_free_utask(tsk);
1153
1154         /* Get rid of any cached register state */
1155         deactivate_mm(tsk, mm);
1156
1157         /*
1158          * Signal userspace if we're not exiting with a core dump
1159          * because we want to leave the value intact for debugging
1160          * purposes.
1161          */
1162         if (tsk->clear_child_tid) {
1163                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1164                     atomic_read(&mm->mm_users) > 1) {
1165                         /*
1166                          * We don't check the error code - if userspace has
1167                          * not set up a proper pointer then tough luck.
1168                          */
1169                         put_user(0, tsk->clear_child_tid);
1170                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1171                                         1, NULL, NULL, 0);
1172                 }
1173                 tsk->clear_child_tid = NULL;
1174         }
1175
1176         /*
1177          * All done, finally we can wake up parent and return this mm to him.
1178          * Also kthread_stop() uses this completion for synchronization.
1179          */
1180         if (tsk->vfork_done)
1181                 complete_vfork_done(tsk);
1182 }
1183
1184 /*
1185  * Allocate a new mm structure and copy contents from the
1186  * mm structure of the passed in task structure.
1187  */
1188 static struct mm_struct *dup_mm(struct task_struct *tsk)
1189 {
1190         struct mm_struct *mm, *oldmm = current->mm;
1191         int err;
1192
1193         mm = allocate_mm();
1194         if (!mm)
1195                 goto fail_nomem;
1196
1197         memcpy(mm, oldmm, sizeof(*mm));
1198
1199         if (!mm_init(mm, tsk, mm->user_ns))
1200                 goto fail_nomem;
1201
1202         err = dup_mmap(mm, oldmm);
1203         if (err)
1204                 goto free_pt;
1205
1206         mm->hiwater_rss = get_mm_rss(mm);
1207         mm->hiwater_vm = mm->total_vm;
1208
1209         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1210                 goto free_pt;
1211
1212         return mm;
1213
1214 free_pt:
1215         /* don't put binfmt in mmput, we haven't got module yet */
1216         mm->binfmt = NULL;
1217         mmput(mm);
1218
1219 fail_nomem:
1220         return NULL;
1221 }
1222
1223 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1224 {
1225         struct mm_struct *mm, *oldmm;
1226         int retval;
1227
1228         tsk->min_flt = tsk->maj_flt = 0;
1229         tsk->nvcsw = tsk->nivcsw = 0;
1230 #ifdef CONFIG_DETECT_HUNG_TASK
1231         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1232 #endif
1233
1234         tsk->mm = NULL;
1235         tsk->active_mm = NULL;
1236
1237         /*
1238          * Are we cloning a kernel thread?
1239          *
1240          * We need to steal a active VM for that..
1241          */
1242         oldmm = current->mm;
1243         if (!oldmm)
1244                 return 0;
1245
1246         /* initialize the new vmacache entries */
1247         vmacache_flush(tsk);
1248
1249         if (clone_flags & CLONE_VM) {
1250                 mmget(oldmm);
1251                 mm = oldmm;
1252                 goto good_mm;
1253         }
1254
1255         retval = -ENOMEM;
1256         mm = dup_mm(tsk);
1257         if (!mm)
1258                 goto fail_nomem;
1259
1260 good_mm:
1261         tsk->mm = mm;
1262         tsk->active_mm = mm;
1263         return 0;
1264
1265 fail_nomem:
1266         return retval;
1267 }
1268
1269 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1270 {
1271         struct fs_struct *fs = current->fs;
1272         if (clone_flags & CLONE_FS) {
1273                 /* tsk->fs is already what we want */
1274                 spin_lock(&fs->lock);
1275                 if (fs->in_exec) {
1276                         spin_unlock(&fs->lock);
1277                         return -EAGAIN;
1278                 }
1279                 fs->users++;
1280                 spin_unlock(&fs->lock);
1281                 return 0;
1282         }
1283         tsk->fs = copy_fs_struct(fs);
1284         if (!tsk->fs)
1285                 return -ENOMEM;
1286         return 0;
1287 }
1288
1289 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1290 {
1291         struct files_struct *oldf, *newf;
1292         int error = 0;
1293
1294         /*
1295          * A background process may not have any files ...
1296          */
1297         oldf = current->files;
1298         if (!oldf)
1299                 goto out;
1300
1301         if (clone_flags & CLONE_FILES) {
1302                 atomic_inc(&oldf->count);
1303                 goto out;
1304         }
1305
1306         newf = dup_fd(oldf, &error);
1307         if (!newf)
1308                 goto out;
1309
1310         tsk->files = newf;
1311         error = 0;
1312 out:
1313         return error;
1314 }
1315
1316 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1317 {
1318 #ifdef CONFIG_BLOCK
1319         struct io_context *ioc = current->io_context;
1320         struct io_context *new_ioc;
1321
1322         if (!ioc)
1323                 return 0;
1324         /*
1325          * Share io context with parent, if CLONE_IO is set
1326          */
1327         if (clone_flags & CLONE_IO) {
1328                 ioc_task_link(ioc);
1329                 tsk->io_context = ioc;
1330         } else if (ioprio_valid(ioc->ioprio)) {
1331                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1332                 if (unlikely(!new_ioc))
1333                         return -ENOMEM;
1334
1335                 new_ioc->ioprio = ioc->ioprio;
1336                 put_io_context(new_ioc);
1337         }
1338 #endif
1339         return 0;
1340 }
1341
1342 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1343 {
1344         struct sighand_struct *sig;
1345
1346         if (clone_flags & CLONE_SIGHAND) {
1347                 atomic_inc(&current->sighand->count);
1348                 return 0;
1349         }
1350         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1351         rcu_assign_pointer(tsk->sighand, sig);
1352         if (!sig)
1353                 return -ENOMEM;
1354
1355         atomic_set(&sig->count, 1);
1356         spin_lock_irq(&current->sighand->siglock);
1357         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1358         spin_unlock_irq(&current->sighand->siglock);
1359         return 0;
1360 }
1361
1362 void __cleanup_sighand(struct sighand_struct *sighand)
1363 {
1364         if (atomic_dec_and_test(&sighand->count)) {
1365                 signalfd_cleanup(sighand);
1366                 /*
1367                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1368                  * without an RCU grace period, see __lock_task_sighand().
1369                  */
1370                 kmem_cache_free(sighand_cachep, sighand);
1371         }
1372 }
1373
1374 #ifdef CONFIG_POSIX_TIMERS
1375 /*
1376  * Initialize POSIX timer handling for a thread group.
1377  */
1378 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1379 {
1380         unsigned long cpu_limit;
1381
1382         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1383         if (cpu_limit != RLIM_INFINITY) {
1384                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1385                 sig->cputimer.running = true;
1386         }
1387
1388         /* The timer lists. */
1389         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1390         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1391         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1392 }
1393 #else
1394 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1395 #endif
1396
1397 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1398 {
1399         struct signal_struct *sig;
1400
1401         if (clone_flags & CLONE_THREAD)
1402                 return 0;
1403
1404         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1405         tsk->signal = sig;
1406         if (!sig)
1407                 return -ENOMEM;
1408
1409         sig->nr_threads = 1;
1410         atomic_set(&sig->live, 1);
1411         atomic_set(&sig->sigcnt, 1);
1412
1413         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1414         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1415         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1416
1417         init_waitqueue_head(&sig->wait_chldexit);
1418         sig->curr_target = tsk;
1419         init_sigpending(&sig->shared_pending);
1420         seqlock_init(&sig->stats_lock);
1421         prev_cputime_init(&sig->prev_cputime);
1422
1423 #ifdef CONFIG_POSIX_TIMERS
1424         INIT_LIST_HEAD(&sig->posix_timers);
1425         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1426         sig->real_timer.function = it_real_fn;
1427 #endif
1428
1429         task_lock(current->group_leader);
1430         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1431         task_unlock(current->group_leader);
1432
1433         posix_cpu_timers_init_group(sig);
1434
1435         tty_audit_fork(sig);
1436         sched_autogroup_fork(sig);
1437
1438         sig->oom_score_adj = current->signal->oom_score_adj;
1439         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1440
1441         mutex_init(&sig->cred_guard_mutex);
1442
1443         return 0;
1444 }
1445
1446 static void copy_seccomp(struct task_struct *p)
1447 {
1448 #ifdef CONFIG_SECCOMP
1449         /*
1450          * Must be called with sighand->lock held, which is common to
1451          * all threads in the group. Holding cred_guard_mutex is not
1452          * needed because this new task is not yet running and cannot
1453          * be racing exec.
1454          */
1455         assert_spin_locked(&current->sighand->siglock);
1456
1457         /* Ref-count the new filter user, and assign it. */
1458         get_seccomp_filter(current);
1459         p->seccomp = current->seccomp;
1460
1461         /*
1462          * Explicitly enable no_new_privs here in case it got set
1463          * between the task_struct being duplicated and holding the
1464          * sighand lock. The seccomp state and nnp must be in sync.
1465          */
1466         if (task_no_new_privs(current))
1467                 task_set_no_new_privs(p);
1468
1469         /*
1470          * If the parent gained a seccomp mode after copying thread
1471          * flags and between before we held the sighand lock, we have
1472          * to manually enable the seccomp thread flag here.
1473          */
1474         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1475                 set_tsk_thread_flag(p, TIF_SECCOMP);
1476 #endif
1477 }
1478
1479 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1480 {
1481         current->clear_child_tid = tidptr;
1482
1483         return task_pid_vnr(current);
1484 }
1485
1486 static void rt_mutex_init_task(struct task_struct *p)
1487 {
1488         raw_spin_lock_init(&p->pi_lock);
1489 #ifdef CONFIG_RT_MUTEXES
1490         p->pi_waiters = RB_ROOT_CACHED;
1491         p->pi_top_task = NULL;
1492         p->pi_blocked_on = NULL;
1493 #endif
1494 }
1495
1496 #ifdef CONFIG_POSIX_TIMERS
1497 /*
1498  * Initialize POSIX timer handling for a single task.
1499  */
1500 static void posix_cpu_timers_init(struct task_struct *tsk)
1501 {
1502         tsk->cputime_expires.prof_exp = 0;
1503         tsk->cputime_expires.virt_exp = 0;
1504         tsk->cputime_expires.sched_exp = 0;
1505         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1506         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1507         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1508 }
1509 #else
1510 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1511 #endif
1512
1513 static inline void
1514 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1515 {
1516          task->pids[type].pid = pid;
1517 }
1518
1519 static inline void rcu_copy_process(struct task_struct *p)
1520 {
1521 #ifdef CONFIG_PREEMPT_RCU
1522         p->rcu_read_lock_nesting = 0;
1523         p->rcu_read_unlock_special.s = 0;
1524         p->rcu_blocked_node = NULL;
1525         INIT_LIST_HEAD(&p->rcu_node_entry);
1526 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1527 #ifdef CONFIG_TASKS_RCU
1528         p->rcu_tasks_holdout = false;
1529         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1530         p->rcu_tasks_idle_cpu = -1;
1531 #endif /* #ifdef CONFIG_TASKS_RCU */
1532 }
1533
1534 /*
1535  * This creates a new process as a copy of the old one,
1536  * but does not actually start it yet.
1537  *
1538  * It copies the registers, and all the appropriate
1539  * parts of the process environment (as per the clone
1540  * flags). The actual kick-off is left to the caller.
1541  */
1542 static __latent_entropy struct task_struct *copy_process(
1543                                         unsigned long clone_flags,
1544                                         unsigned long stack_start,
1545                                         unsigned long stack_size,
1546                                         int __user *child_tidptr,
1547                                         struct pid *pid,
1548                                         int trace,
1549                                         unsigned long tls,
1550                                         int node)
1551 {
1552         int retval;
1553         struct task_struct *p;
1554
1555         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1556                 return ERR_PTR(-EINVAL);
1557
1558         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1559                 return ERR_PTR(-EINVAL);
1560
1561         /*
1562          * Thread groups must share signals as well, and detached threads
1563          * can only be started up within the thread group.
1564          */
1565         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1566                 return ERR_PTR(-EINVAL);
1567
1568         /*
1569          * Shared signal handlers imply shared VM. By way of the above,
1570          * thread groups also imply shared VM. Blocking this case allows
1571          * for various simplifications in other code.
1572          */
1573         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1574                 return ERR_PTR(-EINVAL);
1575
1576         /*
1577          * Siblings of global init remain as zombies on exit since they are
1578          * not reaped by their parent (swapper). To solve this and to avoid
1579          * multi-rooted process trees, prevent global and container-inits
1580          * from creating siblings.
1581          */
1582         if ((clone_flags & CLONE_PARENT) &&
1583                                 current->signal->flags & SIGNAL_UNKILLABLE)
1584                 return ERR_PTR(-EINVAL);
1585
1586         /*
1587          * If the new process will be in a different pid or user namespace
1588          * do not allow it to share a thread group with the forking task.
1589          */
1590         if (clone_flags & CLONE_THREAD) {
1591                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1592                     (task_active_pid_ns(current) !=
1593                                 current->nsproxy->pid_ns_for_children))
1594                         return ERR_PTR(-EINVAL);
1595         }
1596
1597         retval = -ENOMEM;
1598         p = dup_task_struct(current, node);
1599         if (!p)
1600                 goto fork_out;
1601
1602         cpufreq_task_times_init(p);
1603
1604         /*
1605          * This _must_ happen before we call free_task(), i.e. before we jump
1606          * to any of the bad_fork_* labels. This is to avoid freeing
1607          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1608          * kernel threads (PF_KTHREAD).
1609          */
1610         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1611         /*
1612          * Clear TID on mm_release()?
1613          */
1614         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1615
1616         ftrace_graph_init_task(p);
1617
1618         rt_mutex_init_task(p);
1619
1620 #ifdef CONFIG_PROVE_LOCKING
1621         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1622         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1623 #endif
1624         retval = -EAGAIN;
1625         if (atomic_read(&p->real_cred->user->processes) >=
1626                         task_rlimit(p, RLIMIT_NPROC)) {
1627                 if (p->real_cred->user != INIT_USER &&
1628                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1629                         goto bad_fork_free;
1630         }
1631         current->flags &= ~PF_NPROC_EXCEEDED;
1632
1633         retval = copy_creds(p, clone_flags);
1634         if (retval < 0)
1635                 goto bad_fork_free;
1636
1637         /*
1638          * If multiple threads are within copy_process(), then this check
1639          * triggers too late. This doesn't hurt, the check is only there
1640          * to stop root fork bombs.
1641          */
1642         retval = -EAGAIN;
1643         if (nr_threads >= max_threads)
1644                 goto bad_fork_cleanup_count;
1645
1646         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1647         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1648         p->flags |= PF_FORKNOEXEC;
1649         INIT_LIST_HEAD(&p->children);
1650         INIT_LIST_HEAD(&p->sibling);
1651         rcu_copy_process(p);
1652         p->vfork_done = NULL;
1653         spin_lock_init(&p->alloc_lock);
1654
1655         init_sigpending(&p->pending);
1656
1657         p->utime = p->stime = p->gtime = 0;
1658 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1659         p->utimescaled = p->stimescaled = 0;
1660 #endif
1661         prev_cputime_init(&p->prev_cputime);
1662
1663 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1664         seqcount_init(&p->vtime.seqcount);
1665         p->vtime.starttime = 0;
1666         p->vtime.state = VTIME_INACTIVE;
1667 #endif
1668
1669 #if defined(SPLIT_RSS_COUNTING)
1670         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1671 #endif
1672
1673         p->default_timer_slack_ns = current->timer_slack_ns;
1674
1675         task_io_accounting_init(&p->ioac);
1676         acct_clear_integrals(p);
1677
1678         posix_cpu_timers_init(p);
1679
1680         p->io_context = NULL;
1681         p->audit_context = NULL;
1682         cgroup_fork(p);
1683 #ifdef CONFIG_NUMA
1684         p->mempolicy = mpol_dup(p->mempolicy);
1685         if (IS_ERR(p->mempolicy)) {
1686                 retval = PTR_ERR(p->mempolicy);
1687                 p->mempolicy = NULL;
1688                 goto bad_fork_cleanup_threadgroup_lock;
1689         }
1690 #endif
1691 #ifdef CONFIG_CPUSETS
1692         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1693         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1694         seqcount_init(&p->mems_allowed_seq);
1695 #endif
1696 #ifdef CONFIG_TRACE_IRQFLAGS
1697         p->irq_events = 0;
1698         p->hardirqs_enabled = 0;
1699         p->hardirq_enable_ip = 0;
1700         p->hardirq_enable_event = 0;
1701         p->hardirq_disable_ip = _THIS_IP_;
1702         p->hardirq_disable_event = 0;
1703         p->softirqs_enabled = 1;
1704         p->softirq_enable_ip = _THIS_IP_;
1705         p->softirq_enable_event = 0;
1706         p->softirq_disable_ip = 0;
1707         p->softirq_disable_event = 0;
1708         p->hardirq_context = 0;
1709         p->softirq_context = 0;
1710 #endif
1711
1712         p->pagefault_disabled = 0;
1713
1714 #ifdef CONFIG_LOCKDEP
1715         p->lockdep_depth = 0; /* no locks held yet */
1716         p->curr_chain_key = 0;
1717         p->lockdep_recursion = 0;
1718         lockdep_init_task(p);
1719 #endif
1720
1721 #ifdef CONFIG_DEBUG_MUTEXES
1722         p->blocked_on = NULL; /* not blocked yet */
1723 #endif
1724 #ifdef CONFIG_BCACHE
1725         p->sequential_io        = 0;
1726         p->sequential_io_avg    = 0;
1727 #endif
1728
1729         /* Perform scheduler related setup. Assign this task to a CPU. */
1730         retval = sched_fork(clone_flags, p);
1731         if (retval)
1732                 goto bad_fork_cleanup_policy;
1733
1734         retval = perf_event_init_task(p);
1735         if (retval)
1736                 goto bad_fork_cleanup_policy;
1737         retval = audit_alloc(p);
1738         if (retval)
1739                 goto bad_fork_cleanup_perf;
1740         /* copy all the process information */
1741         shm_init_task(p);
1742         retval = security_task_alloc(p, clone_flags);
1743         if (retval)
1744                 goto bad_fork_cleanup_audit;
1745         retval = copy_semundo(clone_flags, p);
1746         if (retval)
1747                 goto bad_fork_cleanup_security;
1748         retval = copy_files(clone_flags, p);
1749         if (retval)
1750                 goto bad_fork_cleanup_semundo;
1751         retval = copy_fs(clone_flags, p);
1752         if (retval)
1753                 goto bad_fork_cleanup_files;
1754         retval = copy_sighand(clone_flags, p);
1755         if (retval)
1756                 goto bad_fork_cleanup_fs;
1757         retval = copy_signal(clone_flags, p);
1758         if (retval)
1759                 goto bad_fork_cleanup_sighand;
1760         retval = copy_mm(clone_flags, p);
1761         if (retval)
1762                 goto bad_fork_cleanup_signal;
1763         retval = copy_namespaces(clone_flags, p);
1764         if (retval)
1765                 goto bad_fork_cleanup_mm;
1766         retval = copy_io(clone_flags, p);
1767         if (retval)
1768                 goto bad_fork_cleanup_namespaces;
1769         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1770         if (retval)
1771                 goto bad_fork_cleanup_io;
1772
1773         if (pid != &init_struct_pid) {
1774                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1775                 if (IS_ERR(pid)) {
1776                         retval = PTR_ERR(pid);
1777                         goto bad_fork_cleanup_thread;
1778                 }
1779         }
1780
1781 #ifdef CONFIG_BLOCK
1782         p->plug = NULL;
1783 #endif
1784 #ifdef CONFIG_FUTEX
1785         p->robust_list = NULL;
1786 #ifdef CONFIG_COMPAT
1787         p->compat_robust_list = NULL;
1788 #endif
1789         INIT_LIST_HEAD(&p->pi_state_list);
1790         p->pi_state_cache = NULL;
1791 #endif
1792         /*
1793          * sigaltstack should be cleared when sharing the same VM
1794          */
1795         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1796                 sas_ss_reset(p);
1797
1798         /*
1799          * Syscall tracing and stepping should be turned off in the
1800          * child regardless of CLONE_PTRACE.
1801          */
1802         user_disable_single_step(p);
1803         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1804 #ifdef TIF_SYSCALL_EMU
1805         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1806 #endif
1807         clear_all_latency_tracing(p);
1808
1809         /* ok, now we should be set up.. */
1810         p->pid = pid_nr(pid);
1811         if (clone_flags & CLONE_THREAD) {
1812                 p->exit_signal = -1;
1813                 p->group_leader = current->group_leader;
1814                 p->tgid = current->tgid;
1815         } else {
1816                 if (clone_flags & CLONE_PARENT)
1817                         p->exit_signal = current->group_leader->exit_signal;
1818                 else
1819                         p->exit_signal = (clone_flags & CSIGNAL);
1820                 p->group_leader = p;
1821                 p->tgid = p->pid;
1822         }
1823
1824         p->nr_dirtied = 0;
1825         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1826         p->dirty_paused_when = 0;
1827
1828         p->pdeath_signal = 0;
1829         INIT_LIST_HEAD(&p->thread_group);
1830         p->task_works = NULL;
1831
1832         cgroup_threadgroup_change_begin(current);
1833         /*
1834          * Ensure that the cgroup subsystem policies allow the new process to be
1835          * forked. It should be noted the the new process's css_set can be changed
1836          * between here and cgroup_post_fork() if an organisation operation is in
1837          * progress.
1838          */
1839         retval = cgroup_can_fork(p);
1840         if (retval)
1841                 goto bad_fork_free_pid;
1842
1843         /*
1844          * From this point on we must avoid any synchronous user-space
1845          * communication until we take the tasklist-lock. In particular, we do
1846          * not want user-space to be able to predict the process start-time by
1847          * stalling fork(2) after we recorded the start_time but before it is
1848          * visible to the system.
1849          */
1850
1851         p->start_time = ktime_get_ns();
1852         p->real_start_time = ktime_get_boot_ns();
1853
1854         /*
1855          * Make it visible to the rest of the system, but dont wake it up yet.
1856          * Need tasklist lock for parent etc handling!
1857          */
1858         write_lock_irq(&tasklist_lock);
1859
1860         /* CLONE_PARENT re-uses the old parent */
1861         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1862                 p->real_parent = current->real_parent;
1863                 p->parent_exec_id = current->parent_exec_id;
1864         } else {
1865                 p->real_parent = current;
1866                 p->parent_exec_id = current->self_exec_id;
1867         }
1868
1869         klp_copy_process(p);
1870
1871         spin_lock(&current->sighand->siglock);
1872
1873         /*
1874          * Copy seccomp details explicitly here, in case they were changed
1875          * before holding sighand lock.
1876          */
1877         copy_seccomp(p);
1878
1879         /*
1880          * Process group and session signals need to be delivered to just the
1881          * parent before the fork or both the parent and the child after the
1882          * fork. Restart if a signal comes in before we add the new process to
1883          * it's process group.
1884          * A fatal signal pending means that current will exit, so the new
1885          * thread can't slip out of an OOM kill (or normal SIGKILL).
1886         */
1887         recalc_sigpending();
1888         if (signal_pending(current)) {
1889                 retval = -ERESTARTNOINTR;
1890                 goto bad_fork_cancel_cgroup;
1891         }
1892         if (unlikely(!(ns_of_pid(pid)->nr_hashed & PIDNS_HASH_ADDING))) {
1893                 retval = -ENOMEM;
1894                 goto bad_fork_cancel_cgroup;
1895         }
1896
1897         if (likely(p->pid)) {
1898                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1899
1900                 init_task_pid(p, PIDTYPE_PID, pid);
1901                 if (thread_group_leader(p)) {
1902                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1903                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1904
1905                         if (is_child_reaper(pid)) {
1906                                 ns_of_pid(pid)->child_reaper = p;
1907                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1908                         }
1909
1910                         p->signal->leader_pid = pid;
1911                         p->signal->tty = tty_kref_get(current->signal->tty);
1912                         /*
1913                          * Inherit has_child_subreaper flag under the same
1914                          * tasklist_lock with adding child to the process tree
1915                          * for propagate_has_child_subreaper optimization.
1916                          */
1917                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1918                                                          p->real_parent->signal->is_child_subreaper;
1919                         list_add_tail(&p->sibling, &p->real_parent->children);
1920                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1921                         attach_pid(p, PIDTYPE_PGID);
1922                         attach_pid(p, PIDTYPE_SID);
1923                         __this_cpu_inc(process_counts);
1924                 } else {
1925                         current->signal->nr_threads++;
1926                         atomic_inc(&current->signal->live);
1927                         atomic_inc(&current->signal->sigcnt);
1928                         list_add_tail_rcu(&p->thread_group,
1929                                           &p->group_leader->thread_group);
1930                         list_add_tail_rcu(&p->thread_node,
1931                                           &p->signal->thread_head);
1932                 }
1933                 attach_pid(p, PIDTYPE_PID);
1934                 nr_threads++;
1935         }
1936
1937         total_forks++;
1938         spin_unlock(&current->sighand->siglock);
1939         syscall_tracepoint_update(p);
1940         write_unlock_irq(&tasklist_lock);
1941
1942         proc_fork_connector(p);
1943         cgroup_post_fork(p);
1944         cgroup_threadgroup_change_end(current);
1945         perf_event_fork(p);
1946
1947         trace_task_newtask(p, clone_flags);
1948         uprobe_copy_process(p, clone_flags);
1949
1950         return p;
1951
1952 bad_fork_cancel_cgroup:
1953         spin_unlock(&current->sighand->siglock);
1954         write_unlock_irq(&tasklist_lock);
1955         cgroup_cancel_fork(p);
1956 bad_fork_free_pid:
1957         cgroup_threadgroup_change_end(current);
1958         if (pid != &init_struct_pid)
1959                 free_pid(pid);
1960 bad_fork_cleanup_thread:
1961         exit_thread(p);
1962 bad_fork_cleanup_io:
1963         if (p->io_context)
1964                 exit_io_context(p);
1965 bad_fork_cleanup_namespaces:
1966         exit_task_namespaces(p);
1967 bad_fork_cleanup_mm:
1968         if (p->mm)
1969                 mmput(p->mm);
1970 bad_fork_cleanup_signal:
1971         if (!(clone_flags & CLONE_THREAD))
1972                 free_signal_struct(p->signal);
1973 bad_fork_cleanup_sighand:
1974         __cleanup_sighand(p->sighand);
1975 bad_fork_cleanup_fs:
1976         exit_fs(p); /* blocking */
1977 bad_fork_cleanup_files:
1978         exit_files(p); /* blocking */
1979 bad_fork_cleanup_semundo:
1980         exit_sem(p);
1981 bad_fork_cleanup_security:
1982         security_task_free(p);
1983 bad_fork_cleanup_audit:
1984         audit_free(p);
1985 bad_fork_cleanup_perf:
1986         perf_event_free_task(p);
1987 bad_fork_cleanup_policy:
1988         lockdep_free_task(p);
1989 #ifdef CONFIG_NUMA
1990         mpol_put(p->mempolicy);
1991 bad_fork_cleanup_threadgroup_lock:
1992 #endif
1993         delayacct_tsk_free(p);
1994 bad_fork_cleanup_count:
1995         atomic_dec(&p->cred->user->processes);
1996         exit_creds(p);
1997 bad_fork_free:
1998         p->state = TASK_DEAD;
1999         put_task_stack(p);
2000         free_task(p);
2001 fork_out:
2002         return ERR_PTR(retval);
2003 }
2004
2005 static inline void init_idle_pids(struct pid_link *links)
2006 {
2007         enum pid_type type;
2008
2009         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2010                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
2011                 links[type].pid = &init_struct_pid;
2012         }
2013 }
2014
2015 struct task_struct *fork_idle(int cpu)
2016 {
2017         struct task_struct *task;
2018         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
2019                             cpu_to_node(cpu));
2020         if (!IS_ERR(task)) {
2021                 init_idle_pids(task->pids);
2022                 init_idle(task, cpu);
2023         }
2024
2025         return task;
2026 }
2027
2028 /*
2029  *  Ok, this is the main fork-routine.
2030  *
2031  * It copies the process, and if successful kick-starts
2032  * it and waits for it to finish using the VM if required.
2033  */
2034 long _do_fork(unsigned long clone_flags,
2035               unsigned long stack_start,
2036               unsigned long stack_size,
2037               int __user *parent_tidptr,
2038               int __user *child_tidptr,
2039               unsigned long tls)
2040 {
2041         struct task_struct *p;
2042         int trace = 0;
2043         long nr;
2044
2045         /*
2046          * Determine whether and which event to report to ptracer.  When
2047          * called from kernel_thread or CLONE_UNTRACED is explicitly
2048          * requested, no event is reported; otherwise, report if the event
2049          * for the type of forking is enabled.
2050          */
2051         if (!(clone_flags & CLONE_UNTRACED)) {
2052                 if (clone_flags & CLONE_VFORK)
2053                         trace = PTRACE_EVENT_VFORK;
2054                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2055                         trace = PTRACE_EVENT_CLONE;
2056                 else
2057                         trace = PTRACE_EVENT_FORK;
2058
2059                 if (likely(!ptrace_event_enabled(current, trace)))
2060                         trace = 0;
2061         }
2062
2063         p = copy_process(clone_flags, stack_start, stack_size,
2064                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2065         add_latent_entropy();
2066         /*
2067          * Do this prior waking up the new thread - the thread pointer
2068          * might get invalid after that point, if the thread exits quickly.
2069          */
2070         if (!IS_ERR(p)) {
2071                 struct completion vfork;
2072                 struct pid *pid;
2073
2074                 cpufreq_task_times_alloc(p);
2075
2076                 trace_sched_process_fork(current, p);
2077
2078                 pid = get_task_pid(p, PIDTYPE_PID);
2079                 nr = pid_vnr(pid);
2080
2081                 if (clone_flags & CLONE_PARENT_SETTID)
2082                         put_user(nr, parent_tidptr);
2083
2084                 if (clone_flags & CLONE_VFORK) {
2085                         p->vfork_done = &vfork;
2086                         init_completion(&vfork);
2087                         get_task_struct(p);
2088                 }
2089
2090                 wake_up_new_task(p);
2091
2092                 /* forking complete and child started to run, tell ptracer */
2093                 if (unlikely(trace))
2094                         ptrace_event_pid(trace, pid);
2095
2096                 if (clone_flags & CLONE_VFORK) {
2097                         if (!wait_for_vfork_done(p, &vfork))
2098                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2099                 }
2100
2101                 put_pid(pid);
2102         } else {
2103                 nr = PTR_ERR(p);
2104         }
2105         return nr;
2106 }
2107
2108 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2109 /* For compatibility with architectures that call do_fork directly rather than
2110  * using the syscall entry points below. */
2111 long do_fork(unsigned long clone_flags,
2112               unsigned long stack_start,
2113               unsigned long stack_size,
2114               int __user *parent_tidptr,
2115               int __user *child_tidptr)
2116 {
2117         return _do_fork(clone_flags, stack_start, stack_size,
2118                         parent_tidptr, child_tidptr, 0);
2119 }
2120 #endif
2121
2122 /*
2123  * Create a kernel thread.
2124  */
2125 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2126 {
2127         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2128                 (unsigned long)arg, NULL, NULL, 0);
2129 }
2130
2131 #ifdef __ARCH_WANT_SYS_FORK
2132 SYSCALL_DEFINE0(fork)
2133 {
2134 #ifdef CONFIG_MMU
2135         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2136 #else
2137         /* can not support in nommu mode */
2138         return -EINVAL;
2139 #endif
2140 }
2141 #endif
2142
2143 #ifdef __ARCH_WANT_SYS_VFORK
2144 SYSCALL_DEFINE0(vfork)
2145 {
2146         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2147                         0, NULL, NULL, 0);
2148 }
2149 #endif
2150
2151 #ifdef __ARCH_WANT_SYS_CLONE
2152 #ifdef CONFIG_CLONE_BACKWARDS
2153 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2154                  int __user *, parent_tidptr,
2155                  unsigned long, tls,
2156                  int __user *, child_tidptr)
2157 #elif defined(CONFIG_CLONE_BACKWARDS2)
2158 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2159                  int __user *, parent_tidptr,
2160                  int __user *, child_tidptr,
2161                  unsigned long, tls)
2162 #elif defined(CONFIG_CLONE_BACKWARDS3)
2163 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2164                 int, stack_size,
2165                 int __user *, parent_tidptr,
2166                 int __user *, child_tidptr,
2167                 unsigned long, tls)
2168 #else
2169 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2170                  int __user *, parent_tidptr,
2171                  int __user *, child_tidptr,
2172                  unsigned long, tls)
2173 #endif
2174 {
2175         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2176 }
2177 #endif
2178
2179 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2180 {
2181         struct task_struct *leader, *parent, *child;
2182         int res;
2183
2184         read_lock(&tasklist_lock);
2185         leader = top = top->group_leader;
2186 down:
2187         for_each_thread(leader, parent) {
2188                 list_for_each_entry(child, &parent->children, sibling) {
2189                         res = visitor(child, data);
2190                         if (res) {
2191                                 if (res < 0)
2192                                         goto out;
2193                                 leader = child;
2194                                 goto down;
2195                         }
2196 up:
2197                         ;
2198                 }
2199         }
2200
2201         if (leader != top) {
2202                 child = leader;
2203                 parent = child->real_parent;
2204                 leader = parent->group_leader;
2205                 goto up;
2206         }
2207 out:
2208         read_unlock(&tasklist_lock);
2209 }
2210
2211 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2212 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2213 #endif
2214
2215 static void sighand_ctor(void *data)
2216 {
2217         struct sighand_struct *sighand = data;
2218
2219         spin_lock_init(&sighand->siglock);
2220         init_waitqueue_head(&sighand->signalfd_wqh);
2221 }
2222
2223 void __init proc_caches_init(void)
2224 {
2225         sighand_cachep = kmem_cache_create("sighand_cache",
2226                         sizeof(struct sighand_struct), 0,
2227                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2228                         SLAB_ACCOUNT, sighand_ctor);
2229         signal_cachep = kmem_cache_create("signal_cache",
2230                         sizeof(struct signal_struct), 0,
2231                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2232                         NULL);
2233         files_cachep = kmem_cache_create("files_cache",
2234                         sizeof(struct files_struct), 0,
2235                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2236                         NULL);
2237         fs_cachep = kmem_cache_create("fs_cache",
2238                         sizeof(struct fs_struct), 0,
2239                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2240                         NULL);
2241         /*
2242          * FIXME! The "sizeof(struct mm_struct)" currently includes the
2243          * whole struct cpumask for the OFFSTACK case. We could change
2244          * this to *only* allocate as much of it as required by the
2245          * maximum number of CPU's we can ever have.  The cpumask_allocation
2246          * is at the end of the structure, exactly for that reason.
2247          */
2248         mm_cachep = kmem_cache_create("mm_struct",
2249                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2250                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2251                         NULL);
2252         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2253         mmap_init();
2254         nsproxy_cache_init();
2255 }
2256
2257 /*
2258  * Check constraints on flags passed to the unshare system call.
2259  */
2260 static int check_unshare_flags(unsigned long unshare_flags)
2261 {
2262         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2263                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2264                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2265                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2266                 return -EINVAL;
2267         /*
2268          * Not implemented, but pretend it works if there is nothing
2269          * to unshare.  Note that unsharing the address space or the
2270          * signal handlers also need to unshare the signal queues (aka
2271          * CLONE_THREAD).
2272          */
2273         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2274                 if (!thread_group_empty(current))
2275                         return -EINVAL;
2276         }
2277         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2278                 if (atomic_read(&current->sighand->count) > 1)
2279                         return -EINVAL;
2280         }
2281         if (unshare_flags & CLONE_VM) {
2282                 if (!current_is_single_threaded())
2283                         return -EINVAL;
2284         }
2285
2286         return 0;
2287 }
2288
2289 /*
2290  * Unshare the filesystem structure if it is being shared
2291  */
2292 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2293 {
2294         struct fs_struct *fs = current->fs;
2295
2296         if (!(unshare_flags & CLONE_FS) || !fs)
2297                 return 0;
2298
2299         /* don't need lock here; in the worst case we'll do useless copy */
2300         if (fs->users == 1)
2301                 return 0;
2302
2303         *new_fsp = copy_fs_struct(fs);
2304         if (!*new_fsp)
2305                 return -ENOMEM;
2306
2307         return 0;
2308 }
2309
2310 /*
2311  * Unshare file descriptor table if it is being shared
2312  */
2313 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2314 {
2315         struct files_struct *fd = current->files;
2316         int error = 0;
2317
2318         if ((unshare_flags & CLONE_FILES) &&
2319             (fd && atomic_read(&fd->count) > 1)) {
2320                 *new_fdp = dup_fd(fd, &error);
2321                 if (!*new_fdp)
2322                         return error;
2323         }
2324
2325         return 0;
2326 }
2327
2328 /*
2329  * unshare allows a process to 'unshare' part of the process
2330  * context which was originally shared using clone.  copy_*
2331  * functions used by do_fork() cannot be used here directly
2332  * because they modify an inactive task_struct that is being
2333  * constructed. Here we are modifying the current, active,
2334  * task_struct.
2335  */
2336 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2337 {
2338         struct fs_struct *fs, *new_fs = NULL;
2339         struct files_struct *fd, *new_fd = NULL;
2340         struct cred *new_cred = NULL;
2341         struct nsproxy *new_nsproxy = NULL;
2342         int do_sysvsem = 0;
2343         int err;
2344
2345         /*
2346          * If unsharing a user namespace must also unshare the thread group
2347          * and unshare the filesystem root and working directories.
2348          */
2349         if (unshare_flags & CLONE_NEWUSER)
2350                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2351         /*
2352          * If unsharing vm, must also unshare signal handlers.
2353          */
2354         if (unshare_flags & CLONE_VM)
2355                 unshare_flags |= CLONE_SIGHAND;
2356         /*
2357          * If unsharing a signal handlers, must also unshare the signal queues.
2358          */
2359         if (unshare_flags & CLONE_SIGHAND)
2360                 unshare_flags |= CLONE_THREAD;
2361         /*
2362          * If unsharing namespace, must also unshare filesystem information.
2363          */
2364         if (unshare_flags & CLONE_NEWNS)
2365                 unshare_flags |= CLONE_FS;
2366
2367         err = check_unshare_flags(unshare_flags);
2368         if (err)
2369                 goto bad_unshare_out;
2370         /*
2371          * CLONE_NEWIPC must also detach from the undolist: after switching
2372          * to a new ipc namespace, the semaphore arrays from the old
2373          * namespace are unreachable.
2374          */
2375         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2376                 do_sysvsem = 1;
2377         err = unshare_fs(unshare_flags, &new_fs);
2378         if (err)
2379                 goto bad_unshare_out;
2380         err = unshare_fd(unshare_flags, &new_fd);
2381         if (err)
2382                 goto bad_unshare_cleanup_fs;
2383         err = unshare_userns(unshare_flags, &new_cred);
2384         if (err)
2385                 goto bad_unshare_cleanup_fd;
2386         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2387                                          new_cred, new_fs);
2388         if (err)
2389                 goto bad_unshare_cleanup_cred;
2390
2391         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2392                 if (do_sysvsem) {
2393                         /*
2394                          * CLONE_SYSVSEM is equivalent to sys_exit().
2395                          */
2396                         exit_sem(current);
2397                 }
2398                 if (unshare_flags & CLONE_NEWIPC) {
2399                         /* Orphan segments in old ns (see sem above). */
2400                         exit_shm(current);
2401                         shm_init_task(current);
2402                 }
2403
2404                 if (new_nsproxy)
2405                         switch_task_namespaces(current, new_nsproxy);
2406
2407                 task_lock(current);
2408
2409                 if (new_fs) {
2410                         fs = current->fs;
2411                         spin_lock(&fs->lock);
2412                         current->fs = new_fs;
2413                         if (--fs->users)
2414                                 new_fs = NULL;
2415                         else
2416                                 new_fs = fs;
2417                         spin_unlock(&fs->lock);
2418                 }
2419
2420                 if (new_fd) {
2421                         fd = current->files;
2422                         current->files = new_fd;
2423                         new_fd = fd;
2424                 }
2425
2426                 task_unlock(current);
2427
2428                 if (new_cred) {
2429                         /* Install the new user namespace */
2430                         commit_creds(new_cred);
2431                         new_cred = NULL;
2432                 }
2433         }
2434
2435         perf_event_namespaces(current);
2436
2437 bad_unshare_cleanup_cred:
2438         if (new_cred)
2439                 put_cred(new_cred);
2440 bad_unshare_cleanup_fd:
2441         if (new_fd)
2442                 put_files_struct(new_fd);
2443
2444 bad_unshare_cleanup_fs:
2445         if (new_fs)
2446                 free_fs_struct(new_fs);
2447
2448 bad_unshare_out:
2449         return err;
2450 }
2451
2452 /*
2453  *      Helper to unshare the files of the current task.
2454  *      We don't want to expose copy_files internals to
2455  *      the exec layer of the kernel.
2456  */
2457
2458 int unshare_files(struct files_struct **displaced)
2459 {
2460         struct task_struct *task = current;
2461         struct files_struct *copy = NULL;
2462         int error;
2463
2464         error = unshare_fd(CLONE_FILES, &copy);
2465         if (error || !copy) {
2466                 *displaced = NULL;
2467                 return error;
2468         }
2469         *displaced = task->files;
2470         task_lock(task);
2471         task->files = copy;
2472         task_unlock(task);
2473         return 0;
2474 }
2475
2476 int sysctl_max_threads(struct ctl_table *table, int write,
2477                        void __user *buffer, size_t *lenp, loff_t *ppos)
2478 {
2479         struct ctl_table t;
2480         int ret;
2481         int threads = max_threads;
2482         int min = MIN_THREADS;
2483         int max = MAX_THREADS;
2484
2485         t = *table;
2486         t.data = &threads;
2487         t.extra1 = &min;
2488         t.extra2 = &max;
2489
2490         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2491         if (ret || !write)
2492                 return ret;
2493
2494         set_max_threads(threads);
2495
2496         return 0;
2497 }