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