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/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/mm.h>
  32 #include <linux/vmacache.h>
  33 #include <linux/nsproxy.h>
  34 #include <linux/capability.h>
  35 #include <linux/cpu.h>
  36 #include <linux/cgroup.h>
  37 #include <linux/security.h>
  38 #include <linux/hugetlb.h>
  39 #include <linux/seccomp.h>
  40 #include <linux/swap.h>
  41 #include <linux/syscalls.h>
  42 #include <linux/jiffies.h>
  43 #include <linux/futex.h>
  44 #include <linux/compat.h>
  45 #include <linux/kthread.h>
  46 #include <linux/task_io_accounting_ops.h>
  47 #include <linux/rcupdate.h>
  48 #include <linux/ptrace.h>
  49 #include <linux/mount.h>
  50 #include <linux/audit.h>
  51 #include <linux/memcontrol.h>
  52 #include <linux/ftrace.h>
  53 #include <linux/proc_fs.h>
  54 #include <linux/profile.h>
  55 #include <linux/rmap.h>
  56 #include <linux/ksm.h>
  57 #include <linux/acct.h>
  58 #include <linux/tsacct_kern.h>
  59 #include <linux/cn_proc.h>
  60 #include <linux/freezer.h>
  61 #include <linux/delayacct.h>
  62 #include <linux/taskstats_kern.h>
  63 #include <linux/random.h>
  64 #include <linux/tty.h>
  65 #include <linux/blkdev.h>
  66 #include <linux/fs_struct.h>
  67 #include <linux/magic.h>
  68 #include <linux/perf_event.h>
  69 #include <linux/posix-timers.h>
  70 #include <linux/user-return-notifier.h>
  71 #include <linux/oom.h>
  72 #include <linux/khugepaged.h>
  73 #include <linux/signalfd.h>
  74 #include <linux/uprobes.h>
  75 #include <linux/aio.h>
  76 #include <linux/compiler.h>
  77 #include <linux/sysctl.h>
  78 #include <linux/kcov.h>
  79 #include <linux/kprobes.h>
  80
  81 #include <asm/pgtable.h>
  82 #include <asm/pgalloc.h>
  83 #include <asm/uaccess.h>
  84 #include <asm/mmu_context.h>
  85 #include <asm/cacheflush.h>
  86 #include <asm/tlbflush.h>
  87
  88 #include <trace/events/sched.h>
  89
  90 #define CREATE_TRACE_POINTS
  91 #include <trace/events/task.h>
  92
  93 /*
  94  * Minimum number of threads to boot the kernel
  95  */
  96 #define MIN_THREADS 20
  97
  98 /*
  99  * Maximum number of threads
 100  */
 101 #define MAX_THREADS FUTEX_TID_MASK
 102
 103 /*
 104  * Protected counters by write_lock_irq(&tasklist_lock)
 105  */
 106 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 107 int nr_threads;                 /* The idle threads do not count.. */
 108
 109 int max_threads;                /* tunable limit on nr_threads */
 110
 111 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 112
 113 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 114
 115 #ifdef CONFIG_PROVE_RCU
 116 int lockdep_tasklist_lock_is_held(void)
 117 {
 118         return lockdep_is_held(&tasklist_lock);
 119 }
 120 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 121 #endif /* #ifdef CONFIG_PROVE_RCU */
 122
 123 int nr_processes(void)
 124 {
 125         int cpu;
 126         int total = 0;
 127
 128         for_each_possible_cpu(cpu)
 129                 total += per_cpu(process_counts, cpu);
 130
 131         return total;
 132 }
 133
 134 void __weak arch_release_task_struct(struct task_struct *tsk)
 135 {
 136 }
 137
 138 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 139 static struct kmem_cache *task_struct_cachep;
 140
 141 static inline struct task_struct *alloc_task_struct_node(int node)
 142 {
 143         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 144 }
 145
 146 static inline void free_task_struct(struct task_struct *tsk)
 147 {
 148         kmem_cache_free(task_struct_cachep, tsk);
 149 }
 150 #endif
 151
 152 void __weak arch_release_thread_stack(unsigned long *stack)
 153 {
 154 }
 155
 156 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 157
 158 /*
 159  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 160  * kmemcache based allocator.
 161  */
 162 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 163
 164 #ifdef CONFIG_VMAP_STACK
 165 /*
 166  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 167  * flush.  Try to minimize the number of calls by caching stacks.
 168  */
 169 #define NR_CACHED_STACKS 2
 170 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 171 #endif
 172
 173 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 174 {
 175 #ifdef CONFIG_VMAP_STACK
 176         void *stack;
 177         int i;
 178
 179         local_irq_disable();
 180         for (i = 0; i < NR_CACHED_STACKS; i++) {
 181                 struct vm_struct *s = this_cpu_read(cached_stacks[i]);
 182
 183                 if (!s)
 184                         continue;
 185                 this_cpu_write(cached_stacks[i], NULL);
 186
 187                 tsk->stack_vm_area = s;
 188                 local_irq_enable();
 189                 return s->addr;
 190         }
 191         local_irq_enable();
 192
 193         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_SIZE,
 194                                      VMALLOC_START, VMALLOC_END,
 195                                      THREADINFO_GFP | __GFP_HIGHMEM,
 196                                      PAGE_KERNEL,
 197                                      0, node, __builtin_return_address(0));
 198
 199         /*
 200          * We can't call find_vm_area() in interrupt context, and
 201          * free_thread_stack() can be called in interrupt context,
 202          * so cache the vm_struct.
 203          */
 204         if (stack)
 205                 tsk->stack_vm_area = find_vm_area(stack);
 206         return stack;
 207 #else
 208         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 209                                              THREAD_SIZE_ORDER);
 210
 211         return page ? page_address(page) : NULL;
 212 #endif
 213 }
 214
 215 static inline void free_thread_stack(struct task_struct *tsk)
 216 {
 217 #ifdef CONFIG_VMAP_STACK
 218         if (task_stack_vm_area(tsk)) {
 219                 unsigned long flags;
 220                 int i;
 221
 222                 local_irq_save(flags);
 223                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 224                         if (this_cpu_read(cached_stacks[i]))
 225                                 continue;
 226
 227                         this_cpu_write(cached_stacks[i], tsk->stack_vm_area);
 228                         local_irq_restore(flags);
 229                         return;
 230                 }
 231                 local_irq_restore(flags);
 232
 233                 vfree(tsk->stack);
 234                 return;
 235         }
 236 #endif
 237
 238         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 239 }
 240 # else
 241 static struct kmem_cache *thread_stack_cache;
 242
 243 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 244                                                   int node)
 245 {
 246         return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 247 }
 248
 249 static void free_thread_stack(struct task_struct *tsk)
 250 {
 251         kmem_cache_free(thread_stack_cache, tsk->stack);
 252 }
 253
 254 void thread_stack_cache_init(void)
 255 {
 256         thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
 257                                               THREAD_SIZE, 0, NULL);
 258         BUG_ON(thread_stack_cache == NULL);
 259 }
 260 # endif
 261 #endif
 262
 263 /* SLAB cache for signal_struct structures (tsk->signal) */
 264 static struct kmem_cache *signal_cachep;
 265
 266 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 267 struct kmem_cache *sighand_cachep;
 268
 269 /* SLAB cache for files_struct structures (tsk->files) */
 270 struct kmem_cache *files_cachep;
 271
 272 /* SLAB cache for fs_struct structures (tsk->fs) */
 273 struct kmem_cache *fs_cachep;
 274
 275 /* SLAB cache for vm_area_struct structures */
 276 struct kmem_cache *vm_area_cachep;
 277
 278 /* SLAB cache for mm_struct structures (tsk->mm) */
 279 static struct kmem_cache *mm_cachep;
 280
 281 static void account_kernel_stack(struct task_struct *tsk, int account)
 282 {
 283         void *stack = task_stack_page(tsk);
 284         struct vm_struct *vm = task_stack_vm_area(tsk);
 285
 286         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 287
 288         if (vm) {
 289                 int i;
 290
 291                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 292
 293                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 294                         mod_zone_page_state(page_zone(vm->pages[i]),
 295                                             NR_KERNEL_STACK_KB,
 296                                             PAGE_SIZE / 1024 * account);
 297                 }
 298
 299                 /* All stack pages belong to the same memcg. */
 300                 memcg_kmem_update_page_stat(vm->pages[0], MEMCG_KERNEL_STACK_KB,
 301                                             account * (THREAD_SIZE / 1024));
 302         } else {
 303                 /*
 304                  * All stack pages are in the same zone and belong to the
 305                  * same memcg.
 306                  */
 307                 struct page *first_page = virt_to_page(stack);
 308
 309                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 310                                     THREAD_SIZE / 1024 * account);
 311
 312                 memcg_kmem_update_page_stat(first_page, MEMCG_KERNEL_STACK_KB,
 313                                             account * (THREAD_SIZE / 1024));
 314         }
 315 }
 316
 317 static void release_task_stack(struct task_struct *tsk)
 318 {
 319         if (WARN_ON(tsk->state != TASK_DEAD))
 320                 return;  /* Better to leak the stack than to free prematurely */
 321
 322         account_kernel_stack(tsk, -1);
 323         arch_release_thread_stack(tsk->stack);
 324         free_thread_stack(tsk);
 325         tsk->stack = NULL;
 326 #ifdef CONFIG_VMAP_STACK
 327         tsk->stack_vm_area = NULL;
 328 #endif
 329 }
 330
 331 #ifdef CONFIG_THREAD_INFO_IN_TASK
 332 void put_task_stack(struct task_struct *tsk)
 333 {
 334         if (atomic_dec_and_test(&tsk->stack_refcount))
 335                 release_task_stack(tsk);
 336 }
 337 #endif
 338
 339 void free_task(struct task_struct *tsk)
 340 {
 341 #ifndef CONFIG_THREAD_INFO_IN_TASK
 342         /*
 343          * The task is finally done with both the stack and thread_info,
 344          * so free both.
 345          */
 346         release_task_stack(tsk);
 347 #else
 348         /*
 349          * If the task had a separate stack allocation, it should be gone
 350          * by now.
 351          */
 352         WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
 353 #endif
 354         rt_mutex_debug_task_free(tsk);
 355         ftrace_graph_exit_task(tsk);
 356         put_seccomp_filter(tsk);
 357         arch_release_task_struct(tsk);
 358         free_task_struct(tsk);
 359 }
 360 EXPORT_SYMBOL(free_task);
 361
 362 static inline void free_signal_struct(struct signal_struct *sig)
 363 {
 364         taskstats_tgid_free(sig);
 365         sched_autogroup_exit(sig);
 366         /*
 367          * __mmdrop is not safe to call from softirq context on x86 due to
 368          * pgd_dtor so postpone it to the async context
 369          */
 370         if (sig->oom_mm)
 371                 mmdrop_async(sig->oom_mm);
 372         kmem_cache_free(signal_cachep, sig);
 373 }
 374
 375 static inline void put_signal_struct(struct signal_struct *sig)
 376 {
 377         if (atomic_dec_and_test(&sig->sigcnt))
 378                 free_signal_struct(sig);
 379 }
 380 #ifdef CONFIG_PREEMPT_RT_BASE
 381 static
 382 #endif
 383 void __put_task_struct(struct task_struct *tsk)
 384 {
 385         WARN_ON(!tsk->exit_state);
 386         WARN_ON(atomic_read(&tsk->usage));
 387         WARN_ON(tsk == current);
 388
 389         /*
 390          * Remove function-return probe instances associated with this
 391          * task and put them back on the free list.
 392          */
 393         kprobe_flush_task(tsk);
 394
 395         /* Task is done with its stack. */
 396         put_task_stack(tsk);
 397
 398         cgroup_free(tsk);
 399         task_numa_free(tsk);
 400         security_task_free(tsk);
 401         exit_creds(tsk);
 402         delayacct_tsk_free(tsk);
 403         put_signal_struct(tsk->signal);
 404
 405         if (!profile_handoff_task(tsk))
 406                 free_task(tsk);
 407 }
 408 #ifndef CONFIG_PREEMPT_RT_BASE
 409 EXPORT_SYMBOL_GPL(__put_task_struct);
 410 #else
 411 void __put_task_struct_cb(struct rcu_head *rhp)
 412 {
 413         struct task_struct *tsk = container_of(rhp, struct task_struct, put_rcu);
 414
 415         __put_task_struct(tsk);
 416
 417 }
 418 EXPORT_SYMBOL_GPL(__put_task_struct_cb);
 419 #endif
 420
 421 void __init __weak arch_task_cache_init(void) { }
 422
 423 /*
 424  * set_max_threads
 425  */
 426 static void set_max_threads(unsigned int max_threads_suggested)
 427 {
 428         u64 threads;
 429
 430         /*
 431          * The number of threads shall be limited such that the thread
 432          * structures may only consume a small part of the available memory.
 433          */
 434         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 435                 threads = MAX_THREADS;
 436         else
 437                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 438                                     (u64) THREAD_SIZE * 8UL);
 439
 440         if (threads > max_threads_suggested)
 441                 threads = max_threads_suggested;
 442
 443         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 444 }
 445
 446 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 447 /* Initialized by the architecture: */
 448 int arch_task_struct_size __read_mostly;
 449 #endif
 450
 451 void __init fork_init(void)
 452 {
 453         int i;
 454 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 455 #ifndef ARCH_MIN_TASKALIGN
 456 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 457 #endif
 458         /* create a slab on which task_structs can be allocated */
 459         task_struct_cachep = kmem_cache_create("task_struct",
 460                         arch_task_struct_size, ARCH_MIN_TASKALIGN,
 461                         SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
 462 #endif
 463
 464         /* do the arch specific task caches init */
 465         arch_task_cache_init();
 466
 467         set_max_threads(MAX_THREADS);
 468
 469         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 470         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 471         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 472                 init_task.signal->rlim[RLIMIT_NPROC];
 473
 474         for (i = 0; i < UCOUNT_COUNTS; i++) {
 475                 init_user_ns.ucount_max[i] = max_threads/2;
 476         }
 477 }
 478
 479 int __weak arch_dup_task_struct(struct task_struct *dst,
 480                                                struct task_struct *src)
 481 {
 482         *dst = *src;
 483         return 0;
 484 }
 485
 486 void set_task_stack_end_magic(struct task_struct *tsk)
 487 {
 488         unsigned long *stackend;
 489
 490         stackend = end_of_stack(tsk);
 491         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 492 }
 493
 494 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 495 {
 496         struct task_struct *tsk;
 497         unsigned long *stack;
 498         struct vm_struct *stack_vm_area;
 499         int err;
 500
 501         if (node == NUMA_NO_NODE)
 502                 node = tsk_fork_get_node(orig);
 503         tsk = alloc_task_struct_node(node);
 504         if (!tsk)
 505                 return NULL;
 506
 507         stack = alloc_thread_stack_node(tsk, node);
 508         if (!stack)
 509                 goto free_tsk;
 510
 511         stack_vm_area = task_stack_vm_area(tsk);
 512
 513         err = arch_dup_task_struct(tsk, orig);
 514
 515         /*
 516          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 517          * sure they're properly initialized before using any stack-related
 518          * functions again.
 519          */
 520         tsk->stack = stack;
 521 #ifdef CONFIG_VMAP_STACK
 522         tsk->stack_vm_area = stack_vm_area;
 523 #endif
 524 #ifdef CONFIG_THREAD_INFO_IN_TASK
 525         atomic_set(&tsk->stack_refcount, 1);
 526 #endif
 527
 528         if (err)
 529                 goto free_stack;
 530
 531 #ifdef CONFIG_SECCOMP
 532         /*
 533          * We must handle setting up seccomp filters once we're under
 534          * the sighand lock in case orig has changed between now and
 535          * then. Until then, filter must be NULL to avoid messing up
 536          * the usage counts on the error path calling free_task.
 537          */
 538         tsk->seccomp.filter = NULL;
 539 #endif
 540
 541         setup_thread_stack(tsk, orig);
 542         clear_user_return_notifier(tsk);
 543         clear_tsk_need_resched(tsk);
 544         set_task_stack_end_magic(tsk);
 545
 546 #ifdef CONFIG_CC_STACKPROTECTOR
 547         tsk->stack_canary = get_random_int();
 548 #endif
 549
 550         /*
 551          * One for us, one for whoever does the "release_task()" (usually
 552          * parent)
 553          */
 554         atomic_set(&tsk->usage, 2);
 555 #ifdef CONFIG_BLK_DEV_IO_TRACE
 556         tsk->btrace_seq = 0;
 557 #endif
 558         tsk->splice_pipe = NULL;
 559         tsk->task_frag.page = NULL;
 560         tsk->wake_q.next = NULL;
 561
 562         account_kernel_stack(tsk, 1);
 563
 564         kcov_task_init(tsk);
 565
 566         return tsk;
 567
 568 free_stack:
 569         free_thread_stack(tsk);
 570 free_tsk:
 571         free_task_struct(tsk);
 572         return NULL;
 573 }
 574
 575 #ifdef CONFIG_MMU
 576 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 577                                         struct mm_struct *oldmm)
 578 {
 579         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 580         struct rb_node **rb_link, *rb_parent;
 581         int retval;
 582         unsigned long charge;
 583
 584         uprobe_start_dup_mmap();
 585         if (down_write_killable(&oldmm->mmap_sem)) {
 586                 retval = -EINTR;
 587                 goto fail_uprobe_end;
 588         }
 589         flush_cache_dup_mm(oldmm);
 590         uprobe_dup_mmap(oldmm, mm);
 591         /*
 592          * Not linked in yet - no deadlock potential:
 593          */
 594         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 595
 596         /* No ordering required: file already has been exposed. */
 597         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 598
 599         mm->total_vm = oldmm->total_vm;
 600         mm->data_vm = oldmm->data_vm;
 601         mm->exec_vm = oldmm->exec_vm;
 602         mm->stack_vm = oldmm->stack_vm;
 603
 604         rb_link = &mm->mm_rb.rb_node;
 605         rb_parent = NULL;
 606         pprev = &mm->mmap;
 607         retval = ksm_fork(mm, oldmm);
 608         if (retval)
 609                 goto out;
 610         retval = khugepaged_fork(mm, oldmm);
 611         if (retval)
 612                 goto out;
 613
 614         prev = NULL;
 615         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 616                 struct file *file;
 617
 618                 if (mpnt->vm_flags & VM_DONTCOPY) {
 619                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 620                         continue;
 621                 }
 622                 charge = 0;
 623                 if (mpnt->vm_flags & VM_ACCOUNT) {
 624                         unsigned long len = vma_pages(mpnt);
 625
 626                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 627                                 goto fail_nomem;
 628                         charge = len;
 629                 }
 630                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 631                 if (!tmp)
 632                         goto fail_nomem;
 633                 *tmp = *mpnt;
 634                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 635                 retval = vma_dup_policy(mpnt, tmp);
 636                 if (retval)
 637                         goto fail_nomem_policy;
 638                 tmp->vm_mm = mm;
 639                 if (anon_vma_fork(tmp, mpnt))
 640                         goto fail_nomem_anon_vma_fork;
 641                 tmp->vm_flags &=
 642                         ~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
 643                 tmp->vm_next = tmp->vm_prev = NULL;
 644                 tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
 645                 file = tmp->vm_file;
 646                 if (file) {
 647                         struct inode *inode = file_inode(file);
 648                         struct address_space *mapping = file->f_mapping;
 649
 650                         get_file(file);
 651                         if (tmp->vm_flags & VM_DENYWRITE)
 652                                 atomic_dec(&inode->i_writecount);
 653                         i_mmap_lock_write(mapping);
 654                         if (tmp->vm_flags & VM_SHARED)
 655                                 atomic_inc(&mapping->i_mmap_writable);
 656                         flush_dcache_mmap_lock(mapping);
 657                         /* insert tmp into the share list, just after mpnt */
 658                         vma_interval_tree_insert_after(tmp, mpnt,
 659                                         &mapping->i_mmap);
 660                         flush_dcache_mmap_unlock(mapping);
 661                         i_mmap_unlock_write(mapping);
 662                 }
 663
 664                 /*
 665                  * Clear hugetlb-related page reserves for children. This only
 666                  * affects MAP_PRIVATE mappings. Faults generated by the child
 667                  * are not guaranteed to succeed, even if read-only
 668                  */
 669                 if (is_vm_hugetlb_page(tmp))
 670                         reset_vma_resv_huge_pages(tmp);
 671
 672                 /*
 673                  * Link in the new vma and copy the page table entries.
 674                  */
 675                 *pprev = tmp;
 676                 pprev = &tmp->vm_next;
 677                 tmp->vm_prev = prev;
 678                 prev = tmp;
 679
 680                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 681                 rb_link = &tmp->vm_rb.rb_right;
 682                 rb_parent = &tmp->vm_rb;
 683
 684                 mm->map_count++;
 685                 retval = copy_page_range(mm, oldmm, mpnt);
 686
 687                 if (tmp->vm_ops && tmp->vm_ops->open)
 688                         tmp->vm_ops->open(tmp);
 689
 690                 if (retval)
 691                         goto out;
 692         }
 693         /* a new mm has just been created */
 694         arch_dup_mmap(oldmm, mm);
 695         retval = 0;
 696 out:
 697         up_write(&mm->mmap_sem);
 698         flush_tlb_mm(oldmm);
 699         up_write(&oldmm->mmap_sem);
 700 fail_uprobe_end:
 701         uprobe_end_dup_mmap();
 702         return retval;
 703 fail_nomem_anon_vma_fork:
 704         mpol_put(vma_policy(tmp));
 705 fail_nomem_policy:
 706         kmem_cache_free(vm_area_cachep, tmp);
 707 fail_nomem:
 708         retval = -ENOMEM;
 709         vm_unacct_memory(charge);
 710         goto out;
 711 }
 712
 713 static inline int mm_alloc_pgd(struct mm_struct *mm)
 714 {
 715         mm->pgd = pgd_alloc(mm);
 716         if (unlikely(!mm->pgd))
 717                 return -ENOMEM;
 718         return 0;
 719 }
 720
 721 static inline void mm_free_pgd(struct mm_struct *mm)
 722 {
 723         pgd_free(mm, mm->pgd);
 724 }
 725 #else
 726 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 727 {
 728         down_write(&oldmm->mmap_sem);
 729         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 730         up_write(&oldmm->mmap_sem);
 731         return 0;
 732 }
 733 #define mm_alloc_pgd(mm)        (0)
 734 #define mm_free_pgd(mm)
 735 #endif /* CONFIG_MMU */
 736
 737 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 738
 739 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 740 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 741
 742 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 743
 744 static int __init coredump_filter_setup(char *s)
 745 {
 746         default_dump_filter =
 747                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 748                 MMF_DUMP_FILTER_MASK;
 749         return 1;
 750 }
 751
 752 __setup("coredump_filter=", coredump_filter_setup);
 753
 754 #include <linux/init_task.h>
 755
 756 static void mm_init_aio(struct mm_struct *mm)
 757 {
 758 #ifdef CONFIG_AIO
 759         spin_lock_init(&mm->ioctx_lock);
 760         mm->ioctx_table = NULL;
 761 #endif
 762 }
 763
 764 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 765 {
 766 #ifdef CONFIG_MEMCG
 767         mm->owner = p;
 768 #endif
 769 }
 770
 771 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 772 {
 773         mm->mmap = NULL;
 774         mm->mm_rb = RB_ROOT;
 775         mm->vmacache_seqnum = 0;
 776         atomic_set(&mm->mm_users, 1);
 777         atomic_set(&mm->mm_count, 1);
 778         init_rwsem(&mm->mmap_sem);
 779         INIT_LIST_HEAD(&mm->mmlist);
 780         mm->core_state = NULL;
 781         atomic_long_set(&mm->nr_ptes, 0);
 782         mm_nr_pmds_init(mm);
 783         mm->map_count = 0;
 784         mm->locked_vm = 0;
 785         mm->pinned_vm = 0;
 786         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 787         spin_lock_init(&mm->page_table_lock);
 788         mm_init_cpumask(mm);
 789         mm_init_aio(mm);
 790         mm_init_owner(mm, p);
 791         mmu_notifier_mm_init(mm);
 792         clear_tlb_flush_pending(mm);
 793 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 794         mm->pmd_huge_pte = NULL;
 795 #endif
 796
 797         if (current->mm) {
 798                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 799                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 800         } else {
 801                 mm->flags = default_dump_filter;
 802                 mm->def_flags = 0;
 803         }
 804
 805         if (mm_alloc_pgd(mm))
 806                 goto fail_nopgd;
 807
 808         if (init_new_context(p, mm))
 809                 goto fail_nocontext;
 810
 811         return mm;
 812
 813 fail_nocontext:
 814         mm_free_pgd(mm);
 815 fail_nopgd:
 816         free_mm(mm);
 817         return NULL;
 818 }
 819
 820 static void check_mm(struct mm_struct *mm)
 821 {
 822         int i;
 823
 824         for (i = 0; i < NR_MM_COUNTERS; i++) {
 825                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 826
 827                 if (unlikely(x))
 828                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 829                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 830         }
 831
 832         if (atomic_long_read(&mm->nr_ptes))
 833                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 834                                 atomic_long_read(&mm->nr_ptes));
 835         if (mm_nr_pmds(mm))
 836                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 837                                 mm_nr_pmds(mm));
 838
 839 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 840         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 841 #endif
 842 }
 843
 844 /*
 845  * Allocate and initialize an mm_struct.
 846  */
 847 struct mm_struct *mm_alloc(void)
 848 {
 849         struct mm_struct *mm;
 850
 851         mm = allocate_mm();
 852         if (!mm)
 853                 return NULL;
 854
 855         memset(mm, 0, sizeof(*mm));
 856         return mm_init(mm, current);
 857 }
 858
 859 /*
 860  * Called when the last reference to the mm
 861  * is dropped: either by a lazy thread or by
 862  * mmput. Free the page directory and the mm.
 863  */
 864 void __mmdrop(struct mm_struct *mm)
 865 {
 866         BUG_ON(mm == &init_mm);
 867         mm_free_pgd(mm);
 868         destroy_context(mm);
 869         mmu_notifier_mm_destroy(mm);
 870         check_mm(mm);
 871         free_mm(mm);
 872 }
 873 EXPORT_SYMBOL_GPL(__mmdrop);
 874
 875 #ifdef CONFIG_PREEMPT_RT_BASE
 876 /*
 877  * RCU callback for delayed mm drop. Not strictly rcu, but we don't
 878  * want another facility to make this work.
 879  */
 880 void __mmdrop_delayed(struct rcu_head *rhp)
 881 {
 882         struct mm_struct *mm = container_of(rhp, struct mm_struct, delayed_drop);
 883
 884         __mmdrop(mm);
 885 }
 886 #endif
 887
 888 static inline void __mmput(struct mm_struct *mm)
 889 {
 890         VM_BUG_ON(atomic_read(&mm->mm_users));
 891
 892         uprobe_clear_state(mm);
 893         exit_aio(mm);
 894         ksm_exit(mm);
 895         khugepaged_exit(mm); /* must run before exit_mmap */
 896         exit_mmap(mm);
 897         mm_put_huge_zero_page(mm);
 898         set_mm_exe_file(mm, NULL);
 899         if (!list_empty(&mm->mmlist)) {
 900                 spin_lock(&mmlist_lock);
 901                 list_del(&mm->mmlist);
 902                 spin_unlock(&mmlist_lock);
 903         }
 904         if (mm->binfmt)
 905                 module_put(mm->binfmt->module);
 906         set_bit(MMF_OOM_SKIP, &mm->flags);
 907         mmdrop(mm);
 908 }
 909
 910 /*
 911  * Decrement the use count and release all resources for an mm.
 912  */
 913 void mmput(struct mm_struct *mm)
 914 {
 915         might_sleep();
 916
 917         if (atomic_dec_and_test(&mm->mm_users))
 918                 __mmput(mm);
 919 }
 920 EXPORT_SYMBOL_GPL(mmput);
 921
 922 #ifdef CONFIG_MMU
 923 static void mmput_async_fn(struct work_struct *work)
 924 {
 925         struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
 926         __mmput(mm);
 927 }
 928
 929 void mmput_async(struct mm_struct *mm)
 930 {
 931         if (atomic_dec_and_test(&mm->mm_users)) {
 932                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
 933                 schedule_work(&mm->async_put_work);
 934         }
 935 }
 936 #endif
 937
 938 /**
 939  * set_mm_exe_file - change a reference to the mm's executable file
 940  *
 941  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 942  *
 943  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 944  * invocations: in mmput() nobody alive left, in execve task is single
 945  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 946  * mm->exe_file, but does so without using set_mm_exe_file() in order
 947  * to do avoid the need for any locks.
 948  */
 949 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 950 {
 951         struct file *old_exe_file;
 952
 953         /*
 954          * It is safe to dereference the exe_file without RCU as
 955          * this function is only called if nobody else can access
 956          * this mm -- see comment above for justification.
 957          */
 958         old_exe_file = rcu_dereference_raw(mm->exe_file);
 959
 960         if (new_exe_file)
 961                 get_file(new_exe_file);
 962         rcu_assign_pointer(mm->exe_file, new_exe_file);
 963         if (old_exe_file)
 964                 fput(old_exe_file);
 965 }
 966
 967 /**
 968  * get_mm_exe_file - acquire a reference to the mm's executable file
 969  *
 970  * Returns %NULL if mm has no associated executable file.
 971  * User must release file via fput().
 972  */
 973 struct file *get_mm_exe_file(struct mm_struct *mm)
 974 {
 975         struct file *exe_file;
 976
 977         rcu_read_lock();
 978         exe_file = rcu_dereference(mm->exe_file);
 979         if (exe_file && !get_file_rcu(exe_file))
 980                 exe_file = NULL;
 981         rcu_read_unlock();
 982         return exe_file;
 983 }
 984 EXPORT_SYMBOL(get_mm_exe_file);
 985
 986 /**
 987  * get_task_exe_file - acquire a reference to the task's executable file
 988  *
 989  * Returns %NULL if task's mm (if any) has no associated executable file or
 990  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
 991  * User must release file via fput().
 992  */
 993 struct file *get_task_exe_file(struct task_struct *task)
 994 {
 995         struct file *exe_file = NULL;
 996         struct mm_struct *mm;
 997
 998         task_lock(task);
 999         mm = task->mm;
1000         if (mm) {
1001                 if (!(task->flags & PF_KTHREAD))
1002                         exe_file = get_mm_exe_file(mm);
1003         }
1004         task_unlock(task);
1005         return exe_file;
1006 }
1007 EXPORT_SYMBOL(get_task_exe_file);
1008
1009 /**
1010  * get_task_mm - acquire a reference to the task's mm
1011  *
1012  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1013  * this kernel workthread has transiently adopted a user mm with use_mm,
1014  * to do its AIO) is not set and if so returns a reference to it, after
1015  * bumping up the use count.  User must release the mm via mmput()
1016  * after use.  Typically used by /proc and ptrace.
1017  */
1018 struct mm_struct *get_task_mm(struct task_struct *task)
1019 {
1020         struct mm_struct *mm;
1021
1022         task_lock(task);
1023         mm = task->mm;
1024         if (mm) {
1025                 if (task->flags & PF_KTHREAD)
1026                         mm = NULL;
1027                 else
1028                         atomic_inc(&mm->mm_users);
1029         }
1030         task_unlock(task);
1031         return mm;
1032 }
1033 EXPORT_SYMBOL_GPL(get_task_mm);
1034
1035 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1036 {
1037         struct mm_struct *mm;
1038         int err;
1039
1040         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1041         if (err)
1042                 return ERR_PTR(err);
1043
1044         mm = get_task_mm(task);
1045         if (mm && mm != current->mm &&
1046                         !ptrace_may_access(task, mode)) {
1047                 mmput(mm);
1048                 mm = ERR_PTR(-EACCES);
1049         }
1050         mutex_unlock(&task->signal->cred_guard_mutex);
1051
1052         return mm;
1053 }
1054
1055 static void complete_vfork_done(struct task_struct *tsk)
1056 {
1057         struct completion *vfork;
1058
1059         task_lock(tsk);
1060         vfork = tsk->vfork_done;
1061         if (likely(vfork)) {
1062                 tsk->vfork_done = NULL;
1063                 complete(vfork);
1064         }
1065         task_unlock(tsk);
1066 }
1067
1068 static int wait_for_vfork_done(struct task_struct *child,
1069                                 struct completion *vfork)
1070 {
1071         int killed;
1072
1073         freezer_do_not_count();
1074         killed = wait_for_completion_killable(vfork);
1075         freezer_count();
1076
1077         if (killed) {
1078                 task_lock(child);
1079                 child->vfork_done = NULL;
1080                 task_unlock(child);
1081         }
1082
1083         put_task_struct(child);
1084         return killed;
1085 }
1086
1087 /* Please note the differences between mmput and mm_release.
1088  * mmput is called whenever we stop holding onto a mm_struct,
1089  * error success whatever.
1090  *
1091  * mm_release is called after a mm_struct has been removed
1092  * from the current process.
1093  *
1094  * This difference is important for error handling, when we
1095  * only half set up a mm_struct for a new process and need to restore
1096  * the old one.  Because we mmput the new mm_struct before
1097  * restoring the old one. . .
1098  * Eric Biederman 10 January 1998
1099  */
1100 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1101 {
1102         /* Get rid of any futexes when releasing the mm */
1103 #ifdef CONFIG_FUTEX
1104         if (unlikely(tsk->robust_list)) {
1105                 exit_robust_list(tsk);
1106                 tsk->robust_list = NULL;
1107         }
1108 #ifdef CONFIG_COMPAT
1109         if (unlikely(tsk->compat_robust_list)) {
1110                 compat_exit_robust_list(tsk);
1111                 tsk->compat_robust_list = NULL;
1112         }
1113 #endif
1114         if (unlikely(!list_empty(&tsk->pi_state_list)))
1115                 exit_pi_state_list(tsk);
1116 #endif
1117
1118         uprobe_free_utask(tsk);
1119
1120         /* Get rid of any cached register state */
1121         deactivate_mm(tsk, mm);
1122
1123         /*
1124          * Signal userspace if we're not exiting with a core dump
1125          * because we want to leave the value intact for debugging
1126          * purposes.
1127          */
1128         if (tsk->clear_child_tid) {
1129                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1130                     atomic_read(&mm->mm_users) > 1) {
1131                         /*
1132                          * We don't check the error code - if userspace has
1133                          * not set up a proper pointer then tough luck.
1134                          */
1135                         put_user(0, tsk->clear_child_tid);
1136                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1137                                         1, NULL, NULL, 0);
1138                 }
1139                 tsk->clear_child_tid = NULL;
1140         }
1141
1142         /*
1143          * All done, finally we can wake up parent and return this mm to him.
1144          * Also kthread_stop() uses this completion for synchronization.
1145          */
1146         if (tsk->vfork_done)
1147                 complete_vfork_done(tsk);
1148 }
1149
1150 /*
1151  * Allocate a new mm structure and copy contents from the
1152  * mm structure of the passed in task structure.
1153  */
1154 static struct mm_struct *dup_mm(struct task_struct *tsk)
1155 {
1156         struct mm_struct *mm, *oldmm = current->mm;
1157         int err;
1158
1159         mm = allocate_mm();
1160         if (!mm)
1161                 goto fail_nomem;
1162
1163         memcpy(mm, oldmm, sizeof(*mm));
1164
1165         if (!mm_init(mm, tsk))
1166                 goto fail_nomem;
1167
1168         err = dup_mmap(mm, oldmm);
1169         if (err)
1170                 goto free_pt;
1171
1172         mm->hiwater_rss = get_mm_rss(mm);
1173         mm->hiwater_vm = mm->total_vm;
1174
1175         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1176                 goto free_pt;
1177
1178         return mm;
1179
1180 free_pt:
1181         /* don't put binfmt in mmput, we haven't got module yet */
1182         mm->binfmt = NULL;
1183         mmput(mm);
1184
1185 fail_nomem:
1186         return NULL;
1187 }
1188
1189 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1190 {
1191         struct mm_struct *mm, *oldmm;
1192         int retval;
1193
1194         tsk->min_flt = tsk->maj_flt = 0;
1195         tsk->nvcsw = tsk->nivcsw = 0;
1196 #ifdef CONFIG_DETECT_HUNG_TASK
1197         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1198 #endif
1199
1200         tsk->mm = NULL;
1201         tsk->active_mm = NULL;
1202
1203         /*
1204          * Are we cloning a kernel thread?
1205          *
1206          * We need to steal a active VM for that..
1207          */
1208         oldmm = current->mm;
1209         if (!oldmm)
1210                 return 0;
1211
1212         /* initialize the new vmacache entries */
1213         vmacache_flush(tsk);
1214
1215         if (clone_flags & CLONE_VM) {
1216                 atomic_inc(&oldmm->mm_users);
1217                 mm = oldmm;
1218                 goto good_mm;
1219         }
1220
1221         retval = -ENOMEM;
1222         mm = dup_mm(tsk);
1223         if (!mm)
1224                 goto fail_nomem;
1225
1226 good_mm:
1227         tsk->mm = mm;
1228         tsk->active_mm = mm;
1229         return 0;
1230
1231 fail_nomem:
1232         return retval;
1233 }
1234
1235 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1236 {
1237         struct fs_struct *fs = current->fs;
1238         if (clone_flags & CLONE_FS) {
1239                 /* tsk->fs is already what we want */
1240                 spin_lock(&fs->lock);
1241                 if (fs->in_exec) {
1242                         spin_unlock(&fs->lock);
1243                         return -EAGAIN;
1244                 }
1245                 fs->users++;
1246                 spin_unlock(&fs->lock);
1247                 return 0;
1248         }
1249         tsk->fs = copy_fs_struct(fs);
1250         if (!tsk->fs)
1251                 return -ENOMEM;
1252         return 0;
1253 }
1254
1255 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1256 {
1257         struct files_struct *oldf, *newf;
1258         int error = 0;
1259
1260         /*
1261          * A background process may not have any files ...
1262          */
1263         oldf = current->files;
1264         if (!oldf)
1265                 goto out;
1266
1267         if (clone_flags & CLONE_FILES) {
1268                 atomic_inc(&oldf->count);
1269                 goto out;
1270         }
1271
1272         newf = dup_fd(oldf, &error);
1273         if (!newf)
1274                 goto out;
1275
1276         tsk->files = newf;
1277         error = 0;
1278 out:
1279         return error;
1280 }
1281
1282 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1283 {
1284 #ifdef CONFIG_BLOCK
1285         struct io_context *ioc = current->io_context;
1286         struct io_context *new_ioc;
1287
1288         if (!ioc)
1289                 return 0;
1290         /*
1291          * Share io context with parent, if CLONE_IO is set
1292          */
1293         if (clone_flags & CLONE_IO) {
1294                 ioc_task_link(ioc);
1295                 tsk->io_context = ioc;
1296         } else if (ioprio_valid(ioc->ioprio)) {
1297                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1298                 if (unlikely(!new_ioc))
1299                         return -ENOMEM;
1300
1301                 new_ioc->ioprio = ioc->ioprio;
1302                 put_io_context(new_ioc);
1303         }
1304 #endif
1305         return 0;
1306 }
1307
1308 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1309 {
1310         struct sighand_struct *sig;
1311
1312         if (clone_flags & CLONE_SIGHAND) {
1313                 atomic_inc(&current->sighand->count);
1314                 return 0;
1315         }
1316         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1317         rcu_assign_pointer(tsk->sighand, sig);
1318         if (!sig)
1319                 return -ENOMEM;
1320
1321         atomic_set(&sig->count, 1);
1322         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1323         return 0;
1324 }
1325
1326 void __cleanup_sighand(struct sighand_struct *sighand)
1327 {
1328         if (atomic_dec_and_test(&sighand->count)) {
1329                 signalfd_cleanup(sighand);
1330                 /*
1331                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1332                  * without an RCU grace period, see __lock_task_sighand().
1333                  */
1334                 kmem_cache_free(sighand_cachep, sighand);
1335         }
1336 }
1337
1338 /*
1339  * Initialize POSIX timer handling for a thread group.
1340  */
1341 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1342 {
1343         unsigned long cpu_limit;
1344
1345         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1346         if (cpu_limit != RLIM_INFINITY) {
1347                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1348                 sig->cputimer.running = true;
1349         }
1350
1351         /* The timer lists. */
1352         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1353         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1354         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1355 }
1356
1357 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1358 {
1359         struct signal_struct *sig;
1360
1361         if (clone_flags & CLONE_THREAD)
1362                 return 0;
1363
1364         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1365         tsk->signal = sig;
1366         if (!sig)
1367                 return -ENOMEM;
1368
1369         sig->nr_threads = 1;
1370         atomic_set(&sig->live, 1);
1371         atomic_set(&sig->sigcnt, 1);
1372
1373         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1374         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1375         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1376
1377         init_waitqueue_head(&sig->wait_chldexit);
1378         sig->curr_target = tsk;
1379         init_sigpending(&sig->shared_pending);
1380         INIT_LIST_HEAD(&sig->posix_timers);
1381         seqlock_init(&sig->stats_lock);
1382         prev_cputime_init(&sig->prev_cputime);
1383
1384         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1385         sig->real_timer.function = it_real_fn;
1386
1387         task_lock(current->group_leader);
1388         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1389         task_unlock(current->group_leader);
1390
1391         posix_cpu_timers_init_group(sig);
1392
1393         tty_audit_fork(sig);
1394         sched_autogroup_fork(sig);
1395
1396         sig->oom_score_adj = current->signal->oom_score_adj;
1397         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1398
1399         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1400                                    current->signal->is_child_subreaper;
1401
1402         mutex_init(&sig->cred_guard_mutex);
1403
1404         return 0;
1405 }
1406
1407 static void copy_seccomp(struct task_struct *p)
1408 {
1409 #ifdef CONFIG_SECCOMP
1410         /*
1411          * Must be called with sighand->lock held, which is common to
1412          * all threads in the group. Holding cred_guard_mutex is not
1413          * needed because this new task is not yet running and cannot
1414          * be racing exec.
1415          */
1416         assert_spin_locked(&current->sighand->siglock);
1417
1418         /* Ref-count the new filter user, and assign it. */
1419         get_seccomp_filter(current);
1420         p->seccomp = current->seccomp;
1421
1422         /*
1423          * Explicitly enable no_new_privs here in case it got set
1424          * between the task_struct being duplicated and holding the
1425          * sighand lock. The seccomp state and nnp must be in sync.
1426          */
1427         if (task_no_new_privs(current))
1428                 task_set_no_new_privs(p);
1429
1430         /*
1431          * If the parent gained a seccomp mode after copying thread
1432          * flags and between before we held the sighand lock, we have
1433          * to manually enable the seccomp thread flag here.
1434          */
1435         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1436                 set_tsk_thread_flag(p, TIF_SECCOMP);
1437 #endif
1438 }
1439
1440 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1441 {
1442         current->clear_child_tid = tidptr;
1443
1444         return task_pid_vnr(current);
1445 }
1446
1447 static void rt_mutex_init_task(struct task_struct *p)
1448 {
1449         raw_spin_lock_init(&p->pi_lock);
1450 #ifdef CONFIG_RT_MUTEXES
1451         p->pi_waiters = RB_ROOT;
1452         p->pi_waiters_leftmost = NULL;
1453         p->pi_blocked_on = NULL;
1454 #endif
1455 }
1456
1457 /*
1458  * Initialize POSIX timer handling for a single task.
1459  */
1460 static void posix_cpu_timers_init(struct task_struct *tsk)
1461 {
1462 #ifdef CONFIG_PREEMPT_RT_BASE
1463         tsk->posix_timer_list = NULL;
1464 #endif
1465         tsk->cputime_expires.prof_exp = 0;
1466         tsk->cputime_expires.virt_exp = 0;
1467         tsk->cputime_expires.sched_exp = 0;
1468         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1469         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1470         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1471 }
1472
1473 static inline void
1474 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1475 {
1476          task->pids[type].pid = pid;
1477 }
1478
1479 /*
1480  * This creates a new process as a copy of the old one,
1481  * but does not actually start it yet.
1482  *
1483  * It copies the registers, and all the appropriate
1484  * parts of the process environment (as per the clone
1485  * flags). The actual kick-off is left to the caller.
1486  */
1487 static __latent_entropy struct task_struct *copy_process(
1488                                         unsigned long clone_flags,
1489                                         unsigned long stack_start,
1490                                         unsigned long stack_size,
1491                                         int __user *child_tidptr,
1492                                         struct pid *pid,
1493                                         int trace,
1494                                         unsigned long tls,
1495                                         int node)
1496 {
1497         int retval;
1498         struct task_struct *p;
1499
1500         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1501                 return ERR_PTR(-EINVAL);
1502
1503         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1504                 return ERR_PTR(-EINVAL);
1505
1506         /*
1507          * Thread groups must share signals as well, and detached threads
1508          * can only be started up within the thread group.
1509          */
1510         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1511                 return ERR_PTR(-EINVAL);
1512
1513         /*
1514          * Shared signal handlers imply shared VM. By way of the above,
1515          * thread groups also imply shared VM. Blocking this case allows
1516          * for various simplifications in other code.
1517          */
1518         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1519                 return ERR_PTR(-EINVAL);
1520
1521         /*
1522          * Siblings of global init remain as zombies on exit since they are
1523          * not reaped by their parent (swapper). To solve this and to avoid
1524          * multi-rooted process trees, prevent global and container-inits
1525          * from creating siblings.
1526          */
1527         if ((clone_flags & CLONE_PARENT) &&
1528                                 current->signal->flags & SIGNAL_UNKILLABLE)
1529                 return ERR_PTR(-EINVAL);
1530
1531         /*
1532          * If the new process will be in a different pid or user namespace
1533          * do not allow it to share a thread group with the forking task.
1534          */
1535         if (clone_flags & CLONE_THREAD) {
1536                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1537                     (task_active_pid_ns(current) !=
1538                                 current->nsproxy->pid_ns_for_children))
1539                         return ERR_PTR(-EINVAL);
1540         }
1541
1542         retval = security_task_create(clone_flags);
1543         if (retval)
1544                 goto fork_out;
1545
1546         retval = -ENOMEM;
1547         p = dup_task_struct(current, node);
1548         if (!p)
1549                 goto fork_out;
1550
1551         ftrace_graph_init_task(p);
1552
1553         rt_mutex_init_task(p);
1554
1555 #ifdef CONFIG_PROVE_LOCKING
1556         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1557         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1558 #endif
1559         retval = -EAGAIN;
1560         if (atomic_read(&p->real_cred->user->processes) >=
1561                         task_rlimit(p, RLIMIT_NPROC)) {
1562                 if (p->real_cred->user != INIT_USER &&
1563                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1564                         goto bad_fork_free;
1565         }
1566         current->flags &= ~PF_NPROC_EXCEEDED;
1567
1568         retval = copy_creds(p, clone_flags);
1569         if (retval < 0)
1570                 goto bad_fork_free;
1571
1572         /*
1573          * If multiple threads are within copy_process(), then this check
1574          * triggers too late. This doesn't hurt, the check is only there
1575          * to stop root fork bombs.
1576          */
1577         retval = -EAGAIN;
1578         if (nr_threads >= max_threads)
1579                 goto bad_fork_cleanup_count;
1580
1581         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1582         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1583         p->flags |= PF_FORKNOEXEC;
1584         INIT_LIST_HEAD(&p->children);
1585         INIT_LIST_HEAD(&p->sibling);
1586         rcu_copy_process(p);
1587         p->vfork_done = NULL;
1588         spin_lock_init(&p->alloc_lock);
1589
1590         init_sigpending(&p->pending);
1591         p->sigqueue_cache = NULL;
1592
1593         p->utime = p->stime = p->gtime = 0;
1594         p->utimescaled = p->stimescaled = 0;
1595         prev_cputime_init(&p->prev_cputime);
1596
1597 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1598         seqcount_init(&p->vtime_seqcount);
1599         p->vtime_snap = 0;
1600         p->vtime_snap_whence = VTIME_INACTIVE;
1601 #endif
1602
1603 #if defined(SPLIT_RSS_COUNTING)
1604         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1605 #endif
1606
1607         p->default_timer_slack_ns = current->timer_slack_ns;
1608
1609         task_io_accounting_init(&p->ioac);
1610         acct_clear_integrals(p);
1611
1612         posix_cpu_timers_init(p);
1613
1614         p->start_time = ktime_get_ns();
1615         p->real_start_time = ktime_get_boot_ns();
1616         p->io_context = NULL;
1617         p->audit_context = NULL;
1618         cgroup_fork(p);
1619 #ifdef CONFIG_NUMA
1620         p->mempolicy = mpol_dup(p->mempolicy);
1621         if (IS_ERR(p->mempolicy)) {
1622                 retval = PTR_ERR(p->mempolicy);
1623                 p->mempolicy = NULL;
1624                 goto bad_fork_cleanup_threadgroup_lock;
1625         }
1626 #endif
1627 #ifdef CONFIG_CPUSETS
1628         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1629         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1630         seqcount_init(&p->mems_allowed_seq);
1631 #endif
1632 #ifdef CONFIG_TRACE_IRQFLAGS
1633         p->irq_events = 0;
1634         p->hardirqs_enabled = 0;
1635         p->hardirq_enable_ip = 0;
1636         p->hardirq_enable_event = 0;
1637         p->hardirq_disable_ip = _THIS_IP_;
1638         p->hardirq_disable_event = 0;
1639         p->softirqs_enabled = 1;
1640         p->softirq_enable_ip = _THIS_IP_;
1641         p->softirq_enable_event = 0;
1642         p->softirq_disable_ip = 0;
1643         p->softirq_disable_event = 0;
1644         p->hardirq_context = 0;
1645         p->softirq_context = 0;
1646 #endif
1647
1648         p->pagefault_disabled = 0;
1649
1650 #ifdef CONFIG_LOCKDEP
1651         p->lockdep_depth = 0; /* no locks held yet */
1652         p->curr_chain_key = 0;
1653         p->lockdep_recursion = 0;
1654 #endif
1655
1656 #ifdef CONFIG_DEBUG_MUTEXES
1657         p->blocked_on = NULL; /* not blocked yet */
1658 #endif
1659 #ifdef CONFIG_BCACHE
1660         p->sequential_io        = 0;
1661         p->sequential_io_avg    = 0;
1662 #endif
1663
1664         /* Perform scheduler related setup. Assign this task to a CPU. */
1665         retval = sched_fork(clone_flags, p);
1666         if (retval)
1667                 goto bad_fork_cleanup_policy;
1668
1669         retval = perf_event_init_task(p);
1670         if (retval)
1671                 goto bad_fork_cleanup_policy;
1672         retval = audit_alloc(p);
1673         if (retval)
1674                 goto bad_fork_cleanup_perf;
1675         /* copy all the process information */
1676         shm_init_task(p);
1677         retval = copy_semundo(clone_flags, p);
1678         if (retval)
1679                 goto bad_fork_cleanup_audit;
1680         retval = copy_files(clone_flags, p);
1681         if (retval)
1682                 goto bad_fork_cleanup_semundo;
1683         retval = copy_fs(clone_flags, p);
1684         if (retval)
1685                 goto bad_fork_cleanup_files;
1686         retval = copy_sighand(clone_flags, p);
1687         if (retval)
1688                 goto bad_fork_cleanup_fs;
1689         retval = copy_signal(clone_flags, p);
1690         if (retval)
1691                 goto bad_fork_cleanup_sighand;
1692         retval = copy_mm(clone_flags, p);
1693         if (retval)
1694                 goto bad_fork_cleanup_signal;
1695         retval = copy_namespaces(clone_flags, p);
1696         if (retval)
1697                 goto bad_fork_cleanup_mm;
1698         retval = copy_io(clone_flags, p);
1699         if (retval)
1700                 goto bad_fork_cleanup_namespaces;
1701         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1702         if (retval)
1703                 goto bad_fork_cleanup_io;
1704
1705         if (pid != &init_struct_pid) {
1706                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1707                 if (IS_ERR(pid)) {
1708                         retval = PTR_ERR(pid);
1709                         goto bad_fork_cleanup_thread;
1710                 }
1711         }
1712
1713         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1714         /*
1715          * Clear TID on mm_release()?
1716          */
1717         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1718 #ifdef CONFIG_BLOCK
1719         p->plug = NULL;
1720 #endif
1721 #ifdef CONFIG_FUTEX
1722         p->robust_list = NULL;
1723 #ifdef CONFIG_COMPAT
1724         p->compat_robust_list = NULL;
1725 #endif
1726         INIT_LIST_HEAD(&p->pi_state_list);
1727         p->pi_state_cache = NULL;
1728 #endif
1729         /*
1730          * sigaltstack should be cleared when sharing the same VM
1731          */
1732         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1733                 sas_ss_reset(p);
1734
1735         /*
1736          * Syscall tracing and stepping should be turned off in the
1737          * child regardless of CLONE_PTRACE.
1738          */
1739         user_disable_single_step(p);
1740         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1741 #ifdef TIF_SYSCALL_EMU
1742         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1743 #endif
1744         clear_all_latency_tracing(p);
1745
1746         /* ok, now we should be set up.. */
1747         p->pid = pid_nr(pid);
1748         if (clone_flags & CLONE_THREAD) {
1749                 p->exit_signal = -1;
1750                 p->group_leader = current->group_leader;
1751                 p->tgid = current->tgid;
1752         } else {
1753                 if (clone_flags & CLONE_PARENT)
1754                         p->exit_signal = current->group_leader->exit_signal;
1755                 else
1756                         p->exit_signal = (clone_flags & CSIGNAL);
1757                 p->group_leader = p;
1758                 p->tgid = p->pid;
1759         }
1760
1761         p->nr_dirtied = 0;
1762         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1763         p->dirty_paused_when = 0;
1764
1765         p->pdeath_signal = 0;
1766         INIT_LIST_HEAD(&p->thread_group);
1767         p->task_works = NULL;
1768
1769         threadgroup_change_begin(current);
1770         /*
1771          * Ensure that the cgroup subsystem policies allow the new process to be
1772          * forked. It should be noted the the new process's css_set can be changed
1773          * between here and cgroup_post_fork() if an organisation operation is in
1774          * progress.
1775          */
1776         retval = cgroup_can_fork(p);
1777         if (retval)
1778                 goto bad_fork_free_pid;
1779
1780         /*
1781          * Make it visible to the rest of the system, but dont wake it up yet.
1782          * Need tasklist lock for parent etc handling!
1783          */
1784         write_lock_irq(&tasklist_lock);
1785
1786         /* CLONE_PARENT re-uses the old parent */
1787         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1788                 p->real_parent = current->real_parent;
1789                 p->parent_exec_id = current->parent_exec_id;
1790         } else {
1791                 p->real_parent = current;
1792                 p->parent_exec_id = current->self_exec_id;
1793         }
1794
1795         spin_lock(&current->sighand->siglock);
1796
1797         /*
1798          * Copy seccomp details explicitly here, in case they were changed
1799          * before holding sighand lock.
1800          */
1801         copy_seccomp(p);
1802
1803         /*
1804          * Process group and session signals need to be delivered to just the
1805          * parent before the fork or both the parent and the child after the
1806          * fork. Restart if a signal comes in before we add the new process to
1807          * it's process group.
1808          * A fatal signal pending means that current will exit, so the new
1809          * thread can't slip out of an OOM kill (or normal SIGKILL).
1810         */
1811         recalc_sigpending();
1812         if (signal_pending(current)) {
1813                 spin_unlock(&current->sighand->siglock);
1814                 write_unlock_irq(&tasklist_lock);
1815                 retval = -ERESTARTNOINTR;
1816                 goto bad_fork_cancel_cgroup;
1817         }
1818
1819         if (likely(p->pid)) {
1820                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1821
1822                 init_task_pid(p, PIDTYPE_PID, pid);
1823                 if (thread_group_leader(p)) {
1824                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1825                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1826
1827                         if (is_child_reaper(pid)) {
1828                                 ns_of_pid(pid)->child_reaper = p;
1829                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1830                         }
1831
1832                         p->signal->leader_pid = pid;
1833                         p->signal->tty = tty_kref_get(current->signal->tty);
1834                         list_add_tail(&p->sibling, &p->real_parent->children);
1835                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1836                         attach_pid(p, PIDTYPE_PGID);
1837                         attach_pid(p, PIDTYPE_SID);
1838                         __this_cpu_inc(process_counts);
1839                 } else {
1840                         current->signal->nr_threads++;
1841                         atomic_inc(&current->signal->live);
1842                         atomic_inc(&current->signal->sigcnt);
1843                         list_add_tail_rcu(&p->thread_group,
1844                                           &p->group_leader->thread_group);
1845                         list_add_tail_rcu(&p->thread_node,
1846                                           &p->signal->thread_head);
1847                 }
1848                 attach_pid(p, PIDTYPE_PID);
1849                 nr_threads++;
1850         }
1851
1852         total_forks++;
1853         spin_unlock(&current->sighand->siglock);
1854         syscall_tracepoint_update(p);
1855         write_unlock_irq(&tasklist_lock);
1856
1857         proc_fork_connector(p);
1858         cgroup_post_fork(p);
1859         threadgroup_change_end(current);
1860         perf_event_fork(p);
1861
1862         trace_task_newtask(p, clone_flags);
1863         uprobe_copy_process(p, clone_flags);
1864
1865         return p;
1866
1867 bad_fork_cancel_cgroup:
1868         cgroup_cancel_fork(p);
1869 bad_fork_free_pid:
1870         threadgroup_change_end(current);
1871         if (pid != &init_struct_pid)
1872                 free_pid(pid);
1873 bad_fork_cleanup_thread:
1874         exit_thread(p);
1875 bad_fork_cleanup_io:
1876         if (p->io_context)
1877                 exit_io_context(p);
1878 bad_fork_cleanup_namespaces:
1879         exit_task_namespaces(p);
1880 bad_fork_cleanup_mm:
1881         if (p->mm)
1882                 mmput(p->mm);
1883 bad_fork_cleanup_signal:
1884         if (!(clone_flags & CLONE_THREAD))
1885                 free_signal_struct(p->signal);
1886 bad_fork_cleanup_sighand:
1887         __cleanup_sighand(p->sighand);
1888 bad_fork_cleanup_fs:
1889         exit_fs(p); /* blocking */
1890 bad_fork_cleanup_files:
1891         exit_files(p); /* blocking */
1892 bad_fork_cleanup_semundo:
1893         exit_sem(p);
1894 bad_fork_cleanup_audit:
1895         audit_free(p);
1896 bad_fork_cleanup_perf:
1897         perf_event_free_task(p);
1898 bad_fork_cleanup_policy:
1899 #ifdef CONFIG_NUMA
1900         mpol_put(p->mempolicy);
1901 bad_fork_cleanup_threadgroup_lock:
1902 #endif
1903         delayacct_tsk_free(p);
1904 bad_fork_cleanup_count:
1905         atomic_dec(&p->cred->user->processes);
1906         exit_creds(p);
1907 bad_fork_free:
1908         p->state = TASK_DEAD;
1909         put_task_stack(p);
1910         free_task(p);
1911 fork_out:
1912         return ERR_PTR(retval);
1913 }
1914
1915 static inline void init_idle_pids(struct pid_link *links)
1916 {
1917         enum pid_type type;
1918
1919         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1920                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1921                 links[type].pid = &init_struct_pid;
1922         }
1923 }
1924
1925 struct task_struct *fork_idle(int cpu)
1926 {
1927         struct task_struct *task;
1928         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1929                             cpu_to_node(cpu));
1930         if (!IS_ERR(task)) {
1931                 init_idle_pids(task->pids);
1932                 init_idle(task, cpu);
1933         }
1934
1935         return task;
1936 }
1937
1938 /*
1939  *  Ok, this is the main fork-routine.
1940  *
1941  * It copies the process, and if successful kick-starts
1942  * it and waits for it to finish using the VM if required.
1943  */
1944 long _do_fork(unsigned long clone_flags,
1945               unsigned long stack_start,
1946               unsigned long stack_size,
1947               int __user *parent_tidptr,
1948               int __user *child_tidptr,
1949               unsigned long tls)
1950 {
1951         struct task_struct *p;
1952         int trace = 0;
1953         long nr;
1954
1955         /*
1956          * Determine whether and which event to report to ptracer.  When
1957          * called from kernel_thread or CLONE_UNTRACED is explicitly
1958          * requested, no event is reported; otherwise, report if the event
1959          * for the type of forking is enabled.
1960          */
1961         if (!(clone_flags & CLONE_UNTRACED)) {
1962                 if (clone_flags & CLONE_VFORK)
1963                         trace = PTRACE_EVENT_VFORK;
1964                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1965                         trace = PTRACE_EVENT_CLONE;
1966                 else
1967                         trace = PTRACE_EVENT_FORK;
1968
1969                 if (likely(!ptrace_event_enabled(current, trace)))
1970                         trace = 0;
1971         }
1972
1973         p = copy_process(clone_flags, stack_start, stack_size,
1974                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
1975         add_latent_entropy();
1976         /*
1977          * Do this prior waking up the new thread - the thread pointer
1978          * might get invalid after that point, if the thread exits quickly.
1979          */
1980         if (!IS_ERR(p)) {
1981                 struct completion vfork;
1982                 struct pid *pid;
1983
1984                 trace_sched_process_fork(current, p);
1985
1986                 pid = get_task_pid(p, PIDTYPE_PID);
1987                 nr = pid_vnr(pid);
1988
1989                 if (clone_flags & CLONE_PARENT_SETTID)
1990                         put_user(nr, parent_tidptr);
1991
1992                 if (clone_flags & CLONE_VFORK) {
1993                         p->vfork_done = &vfork;
1994                         init_completion(&vfork);
1995                         get_task_struct(p);
1996                 }
1997
1998                 wake_up_new_task(p);
1999
2000                 /* forking complete and child started to run, tell ptracer */
2001                 if (unlikely(trace))
2002                         ptrace_event_pid(trace, pid);
2003
2004                 if (clone_flags & CLONE_VFORK) {
2005                         if (!wait_for_vfork_done(p, &vfork))
2006                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2007                 }
2008
2009                 put_pid(pid);
2010         } else {
2011                 nr = PTR_ERR(p);
2012         }
2013         return nr;
2014 }
2015
2016 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2017 /* For compatibility with architectures that call do_fork directly rather than
2018  * using the syscall entry points below. */
2019 long do_fork(unsigned long clone_flags,
2020               unsigned long stack_start,
2021               unsigned long stack_size,
2022               int __user *parent_tidptr,
2023               int __user *child_tidptr)
2024 {
2025         return _do_fork(clone_flags, stack_start, stack_size,
2026                         parent_tidptr, child_tidptr, 0);
2027 }
2028 #endif
2029
2030 /*
2031  * Create a kernel thread.
2032  */
2033 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2034 {
2035         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2036                 (unsigned long)arg, NULL, NULL, 0);
2037 }
2038
2039 #ifdef __ARCH_WANT_SYS_FORK
2040 SYSCALL_DEFINE0(fork)
2041 {
2042 #ifdef CONFIG_MMU
2043         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2044 #else
2045         /* can not support in nommu mode */
2046         return -EINVAL;
2047 #endif
2048 }
2049 #endif
2050
2051 #ifdef __ARCH_WANT_SYS_VFORK
2052 SYSCALL_DEFINE0(vfork)
2053 {
2054         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2055                         0, NULL, NULL, 0);
2056 }
2057 #endif
2058
2059 #ifdef __ARCH_WANT_SYS_CLONE
2060 #ifdef CONFIG_CLONE_BACKWARDS
2061 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2062                  int __user *, parent_tidptr,
2063                  unsigned long, tls,
2064                  int __user *, child_tidptr)
2065 #elif defined(CONFIG_CLONE_BACKWARDS2)
2066 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2067                  int __user *, parent_tidptr,
2068                  int __user *, child_tidptr,
2069                  unsigned long, tls)
2070 #elif defined(CONFIG_CLONE_BACKWARDS3)
2071 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2072                 int, stack_size,
2073                 int __user *, parent_tidptr,
2074                 int __user *, child_tidptr,
2075                 unsigned long, tls)
2076 #else
2077 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2078                  int __user *, parent_tidptr,
2079                  int __user *, child_tidptr,
2080                  unsigned long, tls)
2081 #endif
2082 {
2083         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2084 }
2085 #endif
2086
2087 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2088 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2089 #endif
2090
2091 static void sighand_ctor(void *data)
2092 {
2093         struct sighand_struct *sighand = data;
2094
2095         spin_lock_init(&sighand->siglock);
2096         init_waitqueue_head(&sighand->signalfd_wqh);
2097 }
2098
2099 void __init proc_caches_init(void)
2100 {
2101         sighand_cachep = kmem_cache_create("sighand_cache",
2102                         sizeof(struct sighand_struct), 0,
2103                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
2104                         SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2105         signal_cachep = kmem_cache_create("signal_cache",
2106                         sizeof(struct signal_struct), 0,
2107                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2108                         NULL);
2109         files_cachep = kmem_cache_create("files_cache",
2110                         sizeof(struct files_struct), 0,
2111                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2112                         NULL);
2113         fs_cachep = kmem_cache_create("fs_cache",
2114                         sizeof(struct fs_struct), 0,
2115                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2116                         NULL);
2117         /*
2118          * FIXME! The "sizeof(struct mm_struct)" currently includes the
2119          * whole struct cpumask for the OFFSTACK case. We could change
2120          * this to *only* allocate as much of it as required by the
2121          * maximum number of CPU's we can ever have.  The cpumask_allocation
2122          * is at the end of the structure, exactly for that reason.
2123          */
2124         mm_cachep = kmem_cache_create("mm_struct",
2125                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2126                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2127                         NULL);
2128         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2129         mmap_init();
2130         nsproxy_cache_init();
2131 }
2132
2133 /*
2134  * Check constraints on flags passed to the unshare system call.
2135  */
2136 static int check_unshare_flags(unsigned long unshare_flags)
2137 {
2138         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2139                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2140                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2141                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2142                 return -EINVAL;
2143         /*
2144          * Not implemented, but pretend it works if there is nothing
2145          * to unshare.  Note that unsharing the address space or the
2146          * signal handlers also need to unshare the signal queues (aka
2147          * CLONE_THREAD).
2148          */
2149         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2150                 if (!thread_group_empty(current))
2151                         return -EINVAL;
2152         }
2153         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2154                 if (atomic_read(&current->sighand->count) > 1)
2155                         return -EINVAL;
2156         }
2157         if (unshare_flags & CLONE_VM) {
2158                 if (!current_is_single_threaded())
2159                         return -EINVAL;
2160         }
2161
2162         return 0;
2163 }
2164
2165 /*
2166  * Unshare the filesystem structure if it is being shared
2167  */
2168 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2169 {
2170         struct fs_struct *fs = current->fs;
2171
2172         if (!(unshare_flags & CLONE_FS) || !fs)
2173                 return 0;
2174
2175         /* don't need lock here; in the worst case we'll do useless copy */
2176         if (fs->users == 1)
2177                 return 0;
2178
2179         *new_fsp = copy_fs_struct(fs);
2180         if (!*new_fsp)
2181                 return -ENOMEM;
2182
2183         return 0;
2184 }
2185
2186 /*
2187  * Unshare file descriptor table if it is being shared
2188  */
2189 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2190 {
2191         struct files_struct *fd = current->files;
2192         int error = 0;
2193
2194         if ((unshare_flags & CLONE_FILES) &&
2195             (fd && atomic_read(&fd->count) > 1)) {
2196                 *new_fdp = dup_fd(fd, &error);
2197                 if (!*new_fdp)
2198                         return error;
2199         }
2200
2201         return 0;
2202 }
2203
2204 /*
2205  * unshare allows a process to 'unshare' part of the process
2206  * context which was originally shared using clone.  copy_*
2207  * functions used by do_fork() cannot be used here directly
2208  * because they modify an inactive task_struct that is being
2209  * constructed. Here we are modifying the current, active,
2210  * task_struct.
2211  */
2212 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2213 {
2214         struct fs_struct *fs, *new_fs = NULL;
2215         struct files_struct *fd, *new_fd = NULL;
2216         struct cred *new_cred = NULL;
2217         struct nsproxy *new_nsproxy = NULL;
2218         int do_sysvsem = 0;
2219         int err;
2220
2221         /*
2222          * If unsharing a user namespace must also unshare the thread group
2223          * and unshare the filesystem root and working directories.
2224          */
2225         if (unshare_flags & CLONE_NEWUSER)
2226                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2227         /*
2228          * If unsharing vm, must also unshare signal handlers.
2229          */
2230         if (unshare_flags & CLONE_VM)
2231                 unshare_flags |= CLONE_SIGHAND;
2232         /*
2233          * If unsharing a signal handlers, must also unshare the signal queues.
2234          */
2235         if (unshare_flags & CLONE_SIGHAND)
2236                 unshare_flags |= CLONE_THREAD;
2237         /*
2238          * If unsharing namespace, must also unshare filesystem information.
2239          */
2240         if (unshare_flags & CLONE_NEWNS)
2241                 unshare_flags |= CLONE_FS;
2242
2243         err = check_unshare_flags(unshare_flags);
2244         if (err)
2245                 goto bad_unshare_out;
2246         /*
2247          * CLONE_NEWIPC must also detach from the undolist: after switching
2248          * to a new ipc namespace, the semaphore arrays from the old
2249          * namespace are unreachable.
2250          */
2251         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2252                 do_sysvsem = 1;
2253         err = unshare_fs(unshare_flags, &new_fs);
2254         if (err)
2255                 goto bad_unshare_out;
2256         err = unshare_fd(unshare_flags, &new_fd);
2257         if (err)
2258                 goto bad_unshare_cleanup_fs;
2259         err = unshare_userns(unshare_flags, &new_cred);
2260         if (err)
2261                 goto bad_unshare_cleanup_fd;
2262         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2263                                          new_cred, new_fs);
2264         if (err)
2265                 goto bad_unshare_cleanup_cred;
2266
2267         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2268                 if (do_sysvsem) {
2269                         /*
2270                          * CLONE_SYSVSEM is equivalent to sys_exit().
2271                          */
2272                         exit_sem(current);
2273                 }
2274                 if (unshare_flags & CLONE_NEWIPC) {
2275                         /* Orphan segments in old ns (see sem above). */
2276                         exit_shm(current);
2277                         shm_init_task(current);
2278                 }
2279
2280                 if (new_nsproxy)
2281                         switch_task_namespaces(current, new_nsproxy);
2282
2283                 task_lock(current);
2284
2285                 if (new_fs) {
2286                         fs = current->fs;
2287                         spin_lock(&fs->lock);
2288                         current->fs = new_fs;
2289                         if (--fs->users)
2290                                 new_fs = NULL;
2291                         else
2292                                 new_fs = fs;
2293                         spin_unlock(&fs->lock);
2294                 }
2295
2296                 if (new_fd) {
2297                         fd = current->files;
2298                         current->files = new_fd;
2299                         new_fd = fd;
2300                 }
2301
2302                 task_unlock(current);
2303
2304                 if (new_cred) {
2305                         /* Install the new user namespace */
2306                         commit_creds(new_cred);
2307                         new_cred = NULL;
2308                 }
2309         }
2310
2311 bad_unshare_cleanup_cred:
2312         if (new_cred)
2313                 put_cred(new_cred);
2314 bad_unshare_cleanup_fd:
2315         if (new_fd)
2316                 put_files_struct(new_fd);
2317
2318 bad_unshare_cleanup_fs:
2319         if (new_fs)
2320                 free_fs_struct(new_fs);
2321
2322 bad_unshare_out:
2323         return err;
2324 }
2325
2326 /*
2327  *      Helper to unshare the files of the current task.
2328  *      We don't want to expose copy_files internals to
2329  *      the exec layer of the kernel.
2330  */
2331
2332 int unshare_files(struct files_struct **displaced)
2333 {
2334         struct task_struct *task = current;
2335         struct files_struct *copy = NULL;
2336         int error;
2337
2338         error = unshare_fd(CLONE_FILES, &copy);
2339         if (error || !copy) {
2340                 *displaced = NULL;
2341                 return error;
2342         }
2343         *displaced = task->files;
2344         task_lock(task);
2345         task->files = copy;
2346         task_unlock(task);
2347         return 0;
2348 }
2349
2350 int sysctl_max_threads(struct ctl_table *table, int write,
2351                        void __user *buffer, size_t *lenp, loff_t *ppos)
2352 {
2353         struct ctl_table t;
2354         int ret;
2355         int threads = max_threads;
2356         int min = MIN_THREADS;
2357         int max = MAX_THREADS;
2358
2359         t = *table;
2360         t.data = &threads;
2361         t.extra1 = &min;
2362         t.extra2 = &max;
2363
2364         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2365         if (ret || !write)
2366                 return ret;
2367
2368         set_max_threads(threads);
2369
2370         return 0;
2371 }