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/init.h>
  20 #include <linux/unistd.h>
  21 #include <linux/module.h>
  22 #include <linux/vmalloc.h>
  23 #include <linux/completion.h>
  24 #include <linux/personality.h>
  25 #include <linux/mempolicy.h>
  26 #include <linux/sem.h>
  27 #include <linux/file.h>
  28 #include <linux/fdtable.h>
  29 #include <linux/iocontext.h>
  30 #include <linux/key.h>
  31 #include <linux/binfmts.h>
  32 #include <linux/mman.h>
  33 #include <linux/mmu_notifier.h>
  34 #include <linux/fs.h>
  35 #include <linux/mm.h>
  36 #include <linux/vmacache.h>
  37 #include <linux/nsproxy.h>
  38 #include <linux/capability.h>
  39 #include <linux/cpu.h>
  40 #include <linux/cgroup.h>
  41 #include <linux/security.h>
  42 #include <linux/hugetlb.h>
  43 #include <linux/seccomp.h>
  44 #include <linux/swap.h>
  45 #include <linux/syscalls.h>
  46 #include <linux/jiffies.h>
  47 #include <linux/futex.h>
  48 #include <linux/compat.h>
  49 #include <linux/kthread.h>
  50 #include <linux/task_io_accounting_ops.h>
  51 #include <linux/rcupdate.h>
  52 #include <linux/ptrace.h>
  53 #include <linux/mount.h>
  54 #include <linux/audit.h>
  55 #include <linux/memcontrol.h>
  56 #include <linux/ftrace.h>
  57 #include <linux/proc_fs.h>
  58 #include <linux/profile.h>
  59 #include <linux/rmap.h>
  60 #include <linux/ksm.h>
  61 #include <linux/acct.h>
  62 #include <linux/userfaultfd_k.h>
  63 #include <linux/tsacct_kern.h>
  64 #include <linux/cn_proc.h>
  65 #include <linux/freezer.h>
  66 #include <linux/delayacct.h>
  67 #include <linux/taskstats_kern.h>
  68 #include <linux/random.h>
  69 #include <linux/tty.h>
  70 #include <linux/blkdev.h>
  71 #include <linux/fs_struct.h>
  72 #include <linux/magic.h>
  73 #include <linux/perf_event.h>
  74 #include <linux/posix-timers.h>
  75 #include <linux/user-return-notifier.h>
  76 #include <linux/oom.h>
  77 #include <linux/khugepaged.h>
  78 #include <linux/signalfd.h>
  79 #include <linux/uprobes.h>
  80 #include <linux/aio.h>
  81 #include <linux/compiler.h>
  82 #include <linux/sysctl.h>
  83 #include <linux/kcov.h>
  84
  85 #include <asm/pgtable.h>
  86 #include <asm/pgalloc.h>
  87 #include <linux/uaccess.h>
  88 #include <asm/mmu_context.h>
  89 #include <asm/cacheflush.h>
  90 #include <asm/tlbflush.h>
  91
  92 #include <trace/events/sched.h>
  93
  94 #define CREATE_TRACE_POINTS
  95 #include <trace/events/task.h>
  96
  97 /*
  98  * Minimum number of threads to boot the kernel
  99  */
 100 #define MIN_THREADS 20
 101
 102 /*
 103  * Maximum number of threads
 104  */
 105 #define MAX_THREADS FUTEX_TID_MASK
 106
 107 /*
 108  * Protected counters by write_lock_irq(&tasklist_lock)
 109  */
 110 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 111 int nr_threads;                 /* The idle threads do not count.. */
 112
 113 int max_threads;                /* tunable limit on nr_threads */
 114
 115 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 116
 117 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 118
 119 #ifdef CONFIG_PROVE_RCU
 120 int lockdep_tasklist_lock_is_held(void)
 121 {
 122         return lockdep_is_held(&tasklist_lock);
 123 }
 124 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 125 #endif /* #ifdef CONFIG_PROVE_RCU */
 126
 127 int nr_processes(void)
 128 {
 129         int cpu;
 130         int total = 0;
 131
 132         for_each_possible_cpu(cpu)
 133                 total += per_cpu(process_counts, cpu);
 134
 135         return total;
 136 }
 137
 138 void __weak arch_release_task_struct(struct task_struct *tsk)
 139 {
 140 }
 141
 142 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 143 static struct kmem_cache *task_struct_cachep;
 144
 145 static inline struct task_struct *alloc_task_struct_node(int node)
 146 {
 147         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 148 }
 149
 150 static inline void free_task_struct(struct task_struct *tsk)
 151 {
 152         kmem_cache_free(task_struct_cachep, tsk);
 153 }
 154 #endif
 155
 156 void __weak arch_release_thread_stack(unsigned long *stack)
 157 {
 158 }
 159
 160 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 161
 162 /*
 163  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 164  * kmemcache based allocator.
 165  */
 166 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 167
 168 #ifdef CONFIG_VMAP_STACK
 169 /*
 170  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 171  * flush.  Try to minimize the number of calls by caching stacks.
 172  */
 173 #define NR_CACHED_STACKS 2
 174 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 175 #endif
 176
 177 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 178 {
 179 #ifdef CONFIG_VMAP_STACK
 180         void *stack;
 181         int i;
 182
 183         local_irq_disable();
 184         for (i = 0; i < NR_CACHED_STACKS; i++) {
 185                 struct vm_struct *s = this_cpu_read(cached_stacks[i]);
 186
 187                 if (!s)
 188                         continue;
 189                 this_cpu_write(cached_stacks[i], NULL);
 190
 191                 tsk->stack_vm_area = s;
 192                 local_irq_enable();
 193                 return s->addr;
 194         }
 195         local_irq_enable();
 196
 197         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_SIZE,
 198                                      VMALLOC_START, VMALLOC_END,
 199                                      THREADINFO_GFP | __GFP_HIGHMEM,
 200                                      PAGE_KERNEL,
 201                                      0, node, __builtin_return_address(0));
 202
 203         /*
 204          * We can't call find_vm_area() in interrupt context, and
 205          * free_thread_stack() can be called in interrupt context,
 206          * so cache the vm_struct.
 207          */
 208         if (stack)
 209                 tsk->stack_vm_area = find_vm_area(stack);
 210         return stack;
 211 #else
 212         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 213                                              THREAD_SIZE_ORDER);
 214
 215         return page ? page_address(page) : NULL;
 216 #endif
 217 }
 218
 219 static inline void free_thread_stack(struct task_struct *tsk)
 220 {
 221 #ifdef CONFIG_VMAP_STACK
 222         if (task_stack_vm_area(tsk)) {
 223                 unsigned long flags;
 224                 int i;
 225
 226                 local_irq_save(flags);
 227                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 228                         if (this_cpu_read(cached_stacks[i]))
 229                                 continue;
 230
 231                         this_cpu_write(cached_stacks[i], tsk->stack_vm_area);
 232                         local_irq_restore(flags);
 233                         return;
 234                 }
 235                 local_irq_restore(flags);
 236
 237                 vfree_atomic(tsk->stack);
 238                 return;
 239         }
 240 #endif
 241
 242         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 243 }
 244 # else
 245 static struct kmem_cache *thread_stack_cache;
 246
 247 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 248                                                   int node)
 249 {
 250         return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 251 }
 252
 253 static void free_thread_stack(struct task_struct *tsk)
 254 {
 255         kmem_cache_free(thread_stack_cache, tsk->stack);
 256 }
 257
 258 void thread_stack_cache_init(void)
 259 {
 260         thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
 261                                               THREAD_SIZE, 0, NULL);
 262         BUG_ON(thread_stack_cache == NULL);
 263 }
 264 # endif
 265 #endif
 266
 267 /* SLAB cache for signal_struct structures (tsk->signal) */
 268 static struct kmem_cache *signal_cachep;
 269
 270 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 271 struct kmem_cache *sighand_cachep;
 272
 273 /* SLAB cache for files_struct structures (tsk->files) */
 274 struct kmem_cache *files_cachep;
 275
 276 /* SLAB cache for fs_struct structures (tsk->fs) */
 277 struct kmem_cache *fs_cachep;
 278
 279 /* SLAB cache for vm_area_struct structures */
 280 struct kmem_cache *vm_area_cachep;
 281
 282 /* SLAB cache for mm_struct structures (tsk->mm) */
 283 static struct kmem_cache *mm_cachep;
 284
 285 static void account_kernel_stack(struct task_struct *tsk, int account)
 286 {
 287         void *stack = task_stack_page(tsk);
 288         struct vm_struct *vm = task_stack_vm_area(tsk);
 289
 290         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 291
 292         if (vm) {
 293                 int i;
 294
 295                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 296
 297                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 298                         mod_zone_page_state(page_zone(vm->pages[i]),
 299                                             NR_KERNEL_STACK_KB,
 300                                             PAGE_SIZE / 1024 * account);
 301                 }
 302
 303                 /* All stack pages belong to the same memcg. */
 304                 memcg_kmem_update_page_stat(vm->pages[0], MEMCG_KERNEL_STACK_KB,
 305                                             account * (THREAD_SIZE / 1024));
 306         } else {
 307                 /*
 308                  * All stack pages are in the same zone and belong to the
 309                  * same memcg.
 310                  */
 311                 struct page *first_page = virt_to_page(stack);
 312
 313                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 314                                     THREAD_SIZE / 1024 * account);
 315
 316                 memcg_kmem_update_page_stat(first_page, MEMCG_KERNEL_STACK_KB,
 317                                             account * (THREAD_SIZE / 1024));
 318         }
 319 }
 320
 321 static void release_task_stack(struct task_struct *tsk)
 322 {
 323         if (WARN_ON(tsk->state != TASK_DEAD))
 324                 return;  /* Better to leak the stack than to free prematurely */
 325
 326         account_kernel_stack(tsk, -1);
 327         arch_release_thread_stack(tsk->stack);
 328         free_thread_stack(tsk);
 329         tsk->stack = NULL;
 330 #ifdef CONFIG_VMAP_STACK
 331         tsk->stack_vm_area = NULL;
 332 #endif
 333 }
 334
 335 #ifdef CONFIG_THREAD_INFO_IN_TASK
 336 void put_task_stack(struct task_struct *tsk)
 337 {
 338         if (atomic_dec_and_test(&tsk->stack_refcount))
 339                 release_task_stack(tsk);
 340 }
 341 #endif
 342
 343 void free_task(struct task_struct *tsk)
 344 {
 345 #ifndef CONFIG_THREAD_INFO_IN_TASK
 346         /*
 347          * The task is finally done with both the stack and thread_info,
 348          * so free both.
 349          */
 350         release_task_stack(tsk);
 351 #else
 352         /*
 353          * If the task had a separate stack allocation, it should be gone
 354          * by now.
 355          */
 356         WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
 357 #endif
 358         rt_mutex_debug_task_free(tsk);
 359         ftrace_graph_exit_task(tsk);
 360         put_seccomp_filter(tsk);
 361         arch_release_task_struct(tsk);
 362         if (tsk->flags & PF_KTHREAD)
 363                 free_kthread_struct(tsk);
 364         free_task_struct(tsk);
 365 }
 366 EXPORT_SYMBOL(free_task);
 367
 368 static inline void free_signal_struct(struct signal_struct *sig)
 369 {
 370         taskstats_tgid_free(sig);
 371         sched_autogroup_exit(sig);
 372         /*
 373          * __mmdrop is not safe to call from softirq context on x86 due to
 374          * pgd_dtor so postpone it to the async context
 375          */
 376         if (sig->oom_mm)
 377                 mmdrop_async(sig->oom_mm);
 378         kmem_cache_free(signal_cachep, sig);
 379 }
 380
 381 static inline void put_signal_struct(struct signal_struct *sig)
 382 {
 383         if (atomic_dec_and_test(&sig->sigcnt))
 384                 free_signal_struct(sig);
 385 }
 386
 387 void __put_task_struct(struct task_struct *tsk)
 388 {
 389         WARN_ON(!tsk->exit_state);
 390         WARN_ON(atomic_read(&tsk->usage));
 391         WARN_ON(tsk == current);
 392
 393         cgroup_free(tsk);
 394         task_numa_free(tsk);
 395         security_task_free(tsk);
 396         exit_creds(tsk);
 397         delayacct_tsk_free(tsk);
 398         put_signal_struct(tsk->signal);
 399
 400         if (!profile_handoff_task(tsk))
 401                 free_task(tsk);
 402 }
 403 EXPORT_SYMBOL_GPL(__put_task_struct);
 404
 405 void __init __weak arch_task_cache_init(void) { }
 406
 407 /*
 408  * set_max_threads
 409  */
 410 static void set_max_threads(unsigned int max_threads_suggested)
 411 {
 412         u64 threads;
 413
 414         /*
 415          * The number of threads shall be limited such that the thread
 416          * structures may only consume a small part of the available memory.
 417          */
 418         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 419                 threads = MAX_THREADS;
 420         else
 421                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 422                                     (u64) THREAD_SIZE * 8UL);
 423
 424         if (threads > max_threads_suggested)
 425                 threads = max_threads_suggested;
 426
 427         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 428 }
 429
 430 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 431 /* Initialized by the architecture: */
 432 int arch_task_struct_size __read_mostly;
 433 #endif
 434
 435 void __init fork_init(void)
 436 {
 437         int i;
 438 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 439 #ifndef ARCH_MIN_TASKALIGN
 440 #define ARCH_MIN_TASKALIGN      0
 441 #endif
 442         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 443
 444         /* create a slab on which task_structs can be allocated */
 445         task_struct_cachep = kmem_cache_create("task_struct",
 446                         arch_task_struct_size, align,
 447                         SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
 448 #endif
 449
 450         /* do the arch specific task caches init */
 451         arch_task_cache_init();
 452
 453         set_max_threads(MAX_THREADS);
 454
 455         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 456         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 457         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 458                 init_task.signal->rlim[RLIMIT_NPROC];
 459
 460         for (i = 0; i < UCOUNT_COUNTS; i++) {
 461                 init_user_ns.ucount_max[i] = max_threads/2;
 462         }
 463 }
 464
 465 int __weak arch_dup_task_struct(struct task_struct *dst,
 466                                                struct task_struct *src)
 467 {
 468         *dst = *src;
 469         return 0;
 470 }
 471
 472 void set_task_stack_end_magic(struct task_struct *tsk)
 473 {
 474         unsigned long *stackend;
 475
 476         stackend = end_of_stack(tsk);
 477         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 478 }
 479
 480 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 481 {
 482         struct task_struct *tsk;
 483         unsigned long *stack;
 484         struct vm_struct *stack_vm_area;
 485         int err;
 486
 487         if (node == NUMA_NO_NODE)
 488                 node = tsk_fork_get_node(orig);
 489         tsk = alloc_task_struct_node(node);
 490         if (!tsk)
 491                 return NULL;
 492
 493         stack = alloc_thread_stack_node(tsk, node);
 494         if (!stack)
 495                 goto free_tsk;
 496
 497         stack_vm_area = task_stack_vm_area(tsk);
 498
 499         err = arch_dup_task_struct(tsk, orig);
 500
 501         /*
 502          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 503          * sure they're properly initialized before using any stack-related
 504          * functions again.
 505          */
 506         tsk->stack = stack;
 507 #ifdef CONFIG_VMAP_STACK
 508         tsk->stack_vm_area = stack_vm_area;
 509 #endif
 510 #ifdef CONFIG_THREAD_INFO_IN_TASK
 511         atomic_set(&tsk->stack_refcount, 1);
 512 #endif
 513
 514         if (err)
 515                 goto free_stack;
 516
 517 #ifdef CONFIG_SECCOMP
 518         /*
 519          * We must handle setting up seccomp filters once we're under
 520          * the sighand lock in case orig has changed between now and
 521          * then. Until then, filter must be NULL to avoid messing up
 522          * the usage counts on the error path calling free_task.
 523          */
 524         tsk->seccomp.filter = NULL;
 525 #endif
 526
 527         setup_thread_stack(tsk, orig);
 528         clear_user_return_notifier(tsk);
 529         clear_tsk_need_resched(tsk);
 530         set_task_stack_end_magic(tsk);
 531
 532 #ifdef CONFIG_CC_STACKPROTECTOR
 533         tsk->stack_canary = get_random_int();
 534 #endif
 535
 536         /*
 537          * One for us, one for whoever does the "release_task()" (usually
 538          * parent)
 539          */
 540         atomic_set(&tsk->usage, 2);
 541 #ifdef CONFIG_BLK_DEV_IO_TRACE
 542         tsk->btrace_seq = 0;
 543 #endif
 544         tsk->splice_pipe = NULL;
 545         tsk->task_frag.page = NULL;
 546         tsk->wake_q.next = NULL;
 547
 548         account_kernel_stack(tsk, 1);
 549
 550         kcov_task_init(tsk);
 551
 552         return tsk;
 553
 554 free_stack:
 555         free_thread_stack(tsk);
 556 free_tsk:
 557         free_task_struct(tsk);
 558         return NULL;
 559 }
 560
 561 #ifdef CONFIG_MMU
 562 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 563                                         struct mm_struct *oldmm)
 564 {
 565         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 566         struct rb_node **rb_link, *rb_parent;
 567         int retval;
 568         unsigned long charge;
 569         LIST_HEAD(uf);
 570
 571         uprobe_start_dup_mmap();
 572         if (down_write_killable(&oldmm->mmap_sem)) {
 573                 retval = -EINTR;
 574                 goto fail_uprobe_end;
 575         }
 576         flush_cache_dup_mm(oldmm);
 577         uprobe_dup_mmap(oldmm, mm);
 578         /*
 579          * Not linked in yet - no deadlock potential:
 580          */
 581         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 582
 583         /* No ordering required: file already has been exposed. */
 584         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 585
 586         mm->total_vm = oldmm->total_vm;
 587         mm->data_vm = oldmm->data_vm;
 588         mm->exec_vm = oldmm->exec_vm;
 589         mm->stack_vm = oldmm->stack_vm;
 590
 591         rb_link = &mm->mm_rb.rb_node;
 592         rb_parent = NULL;
 593         pprev = &mm->mmap;
 594         retval = ksm_fork(mm, oldmm);
 595         if (retval)
 596                 goto out;
 597         retval = khugepaged_fork(mm, oldmm);
 598         if (retval)
 599                 goto out;
 600
 601         prev = NULL;
 602         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 603                 struct file *file;
 604
 605                 if (mpnt->vm_flags & VM_DONTCOPY) {
 606                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 607                         continue;
 608                 }
 609                 charge = 0;
 610                 if (mpnt->vm_flags & VM_ACCOUNT) {
 611                         unsigned long len = vma_pages(mpnt);
 612
 613                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 614                                 goto fail_nomem;
 615                         charge = len;
 616                 }
 617                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 618                 if (!tmp)
 619                         goto fail_nomem;
 620                 *tmp = *mpnt;
 621                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 622                 retval = vma_dup_policy(mpnt, tmp);
 623                 if (retval)
 624                         goto fail_nomem_policy;
 625                 tmp->vm_mm = mm;
 626                 retval = dup_userfaultfd(tmp, &uf);
 627                 if (retval)
 628                         goto fail_nomem_anon_vma_fork;
 629                 if (anon_vma_fork(tmp, mpnt))
 630                         goto fail_nomem_anon_vma_fork;
 631                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 632                 tmp->vm_next = tmp->vm_prev = NULL;
 633                 file = tmp->vm_file;
 634                 if (file) {
 635                         struct inode *inode = file_inode(file);
 636                         struct address_space *mapping = file->f_mapping;
 637
 638                         get_file(file);
 639                         if (tmp->vm_flags & VM_DENYWRITE)
 640                                 atomic_dec(&inode->i_writecount);
 641                         i_mmap_lock_write(mapping);
 642                         if (tmp->vm_flags & VM_SHARED)
 643                                 atomic_inc(&mapping->i_mmap_writable);
 644                         flush_dcache_mmap_lock(mapping);
 645                         /* insert tmp into the share list, just after mpnt */
 646                         vma_interval_tree_insert_after(tmp, mpnt,
 647                                         &mapping->i_mmap);
 648                         flush_dcache_mmap_unlock(mapping);
 649                         i_mmap_unlock_write(mapping);
 650                 }
 651
 652                 /*
 653                  * Clear hugetlb-related page reserves for children. This only
 654                  * affects MAP_PRIVATE mappings. Faults generated by the child
 655                  * are not guaranteed to succeed, even if read-only
 656                  */
 657                 if (is_vm_hugetlb_page(tmp))
 658                         reset_vma_resv_huge_pages(tmp);
 659
 660                 /*
 661                  * Link in the new vma and copy the page table entries.
 662                  */
 663                 *pprev = tmp;
 664                 pprev = &tmp->vm_next;
 665                 tmp->vm_prev = prev;
 666                 prev = tmp;
 667
 668                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 669                 rb_link = &tmp->vm_rb.rb_right;
 670                 rb_parent = &tmp->vm_rb;
 671
 672                 mm->map_count++;
 673                 retval = copy_page_range(mm, oldmm, mpnt);
 674
 675                 if (tmp->vm_ops && tmp->vm_ops->open)
 676                         tmp->vm_ops->open(tmp);
 677
 678                 if (retval)
 679                         goto out;
 680         }
 681         /* a new mm has just been created */
 682         arch_dup_mmap(oldmm, mm);
 683         retval = 0;
 684 out:
 685         up_write(&mm->mmap_sem);
 686         flush_tlb_mm(oldmm);
 687         up_write(&oldmm->mmap_sem);
 688         dup_userfaultfd_complete(&uf);
 689 fail_uprobe_end:
 690         uprobe_end_dup_mmap();
 691         return retval;
 692 fail_nomem_anon_vma_fork:
 693         mpol_put(vma_policy(tmp));
 694 fail_nomem_policy:
 695         kmem_cache_free(vm_area_cachep, tmp);
 696 fail_nomem:
 697         retval = -ENOMEM;
 698         vm_unacct_memory(charge);
 699         goto out;
 700 }
 701
 702 static inline int mm_alloc_pgd(struct mm_struct *mm)
 703 {
 704         mm->pgd = pgd_alloc(mm);
 705         if (unlikely(!mm->pgd))
 706                 return -ENOMEM;
 707         return 0;
 708 }
 709
 710 static inline void mm_free_pgd(struct mm_struct *mm)
 711 {
 712         pgd_free(mm, mm->pgd);
 713 }
 714 #else
 715 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 716 {
 717         down_write(&oldmm->mmap_sem);
 718         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 719         up_write(&oldmm->mmap_sem);
 720         return 0;
 721 }
 722 #define mm_alloc_pgd(mm)        (0)
 723 #define mm_free_pgd(mm)
 724 #endif /* CONFIG_MMU */
 725
 726 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 727
 728 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 729 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 730
 731 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 732
 733 static int __init coredump_filter_setup(char *s)
 734 {
 735         default_dump_filter =
 736                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 737                 MMF_DUMP_FILTER_MASK;
 738         return 1;
 739 }
 740
 741 __setup("coredump_filter=", coredump_filter_setup);
 742
 743 #include <linux/init_task.h>
 744
 745 static void mm_init_aio(struct mm_struct *mm)
 746 {
 747 #ifdef CONFIG_AIO
 748         spin_lock_init(&mm->ioctx_lock);
 749         mm->ioctx_table = NULL;
 750 #endif
 751 }
 752
 753 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 754 {
 755 #ifdef CONFIG_MEMCG
 756         mm->owner = p;
 757 #endif
 758 }
 759
 760 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 761         struct user_namespace *user_ns)
 762 {
 763         mm->mmap = NULL;
 764         mm->mm_rb = RB_ROOT;
 765         mm->vmacache_seqnum = 0;
 766         atomic_set(&mm->mm_users, 1);
 767         atomic_set(&mm->mm_count, 1);
 768         init_rwsem(&mm->mmap_sem);
 769         INIT_LIST_HEAD(&mm->mmlist);
 770         mm->core_state = NULL;
 771         atomic_long_set(&mm->nr_ptes, 0);
 772         mm_nr_pmds_init(mm);
 773         mm->map_count = 0;
 774         mm->locked_vm = 0;
 775         mm->pinned_vm = 0;
 776         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 777         spin_lock_init(&mm->page_table_lock);
 778         mm_init_cpumask(mm);
 779         mm_init_aio(mm);
 780         mm_init_owner(mm, p);
 781         mmu_notifier_mm_init(mm);
 782         clear_tlb_flush_pending(mm);
 783 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 784         mm->pmd_huge_pte = NULL;
 785 #endif
 786
 787         if (current->mm) {
 788                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 789                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 790         } else {
 791                 mm->flags = default_dump_filter;
 792                 mm->def_flags = 0;
 793         }
 794
 795         if (mm_alloc_pgd(mm))
 796                 goto fail_nopgd;
 797
 798         if (init_new_context(p, mm))
 799                 goto fail_nocontext;
 800
 801         mm->user_ns = get_user_ns(user_ns);
 802         return mm;
 803
 804 fail_nocontext:
 805         mm_free_pgd(mm);
 806 fail_nopgd:
 807         free_mm(mm);
 808         return NULL;
 809 }
 810
 811 static void check_mm(struct mm_struct *mm)
 812 {
 813         int i;
 814
 815         for (i = 0; i < NR_MM_COUNTERS; i++) {
 816                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 817
 818                 if (unlikely(x))
 819                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 820                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 821         }
 822
 823         if (atomic_long_read(&mm->nr_ptes))
 824                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 825                                 atomic_long_read(&mm->nr_ptes));
 826         if (mm_nr_pmds(mm))
 827                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 828                                 mm_nr_pmds(mm));
 829
 830 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 831         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 832 #endif
 833 }
 834
 835 /*
 836  * Allocate and initialize an mm_struct.
 837  */
 838 struct mm_struct *mm_alloc(void)
 839 {
 840         struct mm_struct *mm;
 841
 842         mm = allocate_mm();
 843         if (!mm)
 844                 return NULL;
 845
 846         memset(mm, 0, sizeof(*mm));
 847         return mm_init(mm, current, current_user_ns());
 848 }
 849
 850 /*
 851  * Called when the last reference to the mm
 852  * is dropped: either by a lazy thread or by
 853  * mmput. Free the page directory and the mm.
 854  */
 855 void __mmdrop(struct mm_struct *mm)
 856 {
 857         BUG_ON(mm == &init_mm);
 858         mm_free_pgd(mm);
 859         destroy_context(mm);
 860         mmu_notifier_mm_destroy(mm);
 861         check_mm(mm);
 862         put_user_ns(mm->user_ns);
 863         free_mm(mm);
 864 }
 865 EXPORT_SYMBOL_GPL(__mmdrop);
 866
 867 static inline void __mmput(struct mm_struct *mm)
 868 {
 869         VM_BUG_ON(atomic_read(&mm->mm_users));
 870
 871         uprobe_clear_state(mm);
 872         exit_aio(mm);
 873         ksm_exit(mm);
 874         khugepaged_exit(mm); /* must run before exit_mmap */
 875         exit_mmap(mm);
 876         mm_put_huge_zero_page(mm);
 877         set_mm_exe_file(mm, NULL);
 878         if (!list_empty(&mm->mmlist)) {
 879                 spin_lock(&mmlist_lock);
 880                 list_del(&mm->mmlist);
 881                 spin_unlock(&mmlist_lock);
 882         }
 883         if (mm->binfmt)
 884                 module_put(mm->binfmt->module);
 885         set_bit(MMF_OOM_SKIP, &mm->flags);
 886         mmdrop(mm);
 887 }
 888
 889 /*
 890  * Decrement the use count and release all resources for an mm.
 891  */
 892 void mmput(struct mm_struct *mm)
 893 {
 894         might_sleep();
 895
 896         if (atomic_dec_and_test(&mm->mm_users))
 897                 __mmput(mm);
 898 }
 899 EXPORT_SYMBOL_GPL(mmput);
 900
 901 #ifdef CONFIG_MMU
 902 static void mmput_async_fn(struct work_struct *work)
 903 {
 904         struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
 905         __mmput(mm);
 906 }
 907
 908 void mmput_async(struct mm_struct *mm)
 909 {
 910         if (atomic_dec_and_test(&mm->mm_users)) {
 911                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
 912                 schedule_work(&mm->async_put_work);
 913         }
 914 }
 915 #endif
 916
 917 /**
 918  * set_mm_exe_file - change a reference to the mm's executable file
 919  *
 920  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 921  *
 922  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 923  * invocations: in mmput() nobody alive left, in execve task is single
 924  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 925  * mm->exe_file, but does so without using set_mm_exe_file() in order
 926  * to do avoid the need for any locks.
 927  */
 928 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 929 {
 930         struct file *old_exe_file;
 931
 932         /*
 933          * It is safe to dereference the exe_file without RCU as
 934          * this function is only called if nobody else can access
 935          * this mm -- see comment above for justification.
 936          */
 937         old_exe_file = rcu_dereference_raw(mm->exe_file);
 938
 939         if (new_exe_file)
 940                 get_file(new_exe_file);
 941         rcu_assign_pointer(mm->exe_file, new_exe_file);
 942         if (old_exe_file)
 943                 fput(old_exe_file);
 944 }
 945
 946 /**
 947  * get_mm_exe_file - acquire a reference to the mm's executable file
 948  *
 949  * Returns %NULL if mm has no associated executable file.
 950  * User must release file via fput().
 951  */
 952 struct file *get_mm_exe_file(struct mm_struct *mm)
 953 {
 954         struct file *exe_file;
 955
 956         rcu_read_lock();
 957         exe_file = rcu_dereference(mm->exe_file);
 958         if (exe_file && !get_file_rcu(exe_file))
 959                 exe_file = NULL;
 960         rcu_read_unlock();
 961         return exe_file;
 962 }
 963 EXPORT_SYMBOL(get_mm_exe_file);
 964
 965 /**
 966  * get_task_exe_file - acquire a reference to the task's executable file
 967  *
 968  * Returns %NULL if task's mm (if any) has no associated executable file or
 969  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
 970  * User must release file via fput().
 971  */
 972 struct file *get_task_exe_file(struct task_struct *task)
 973 {
 974         struct file *exe_file = NULL;
 975         struct mm_struct *mm;
 976
 977         task_lock(task);
 978         mm = task->mm;
 979         if (mm) {
 980                 if (!(task->flags & PF_KTHREAD))
 981                         exe_file = get_mm_exe_file(mm);
 982         }
 983         task_unlock(task);
 984         return exe_file;
 985 }
 986 EXPORT_SYMBOL(get_task_exe_file);
 987
 988 /**
 989  * get_task_mm - acquire a reference to the task's mm
 990  *
 991  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 992  * this kernel workthread has transiently adopted a user mm with use_mm,
 993  * to do its AIO) is not set and if so returns a reference to it, after
 994  * bumping up the use count.  User must release the mm via mmput()
 995  * after use.  Typically used by /proc and ptrace.
 996  */
 997 struct mm_struct *get_task_mm(struct task_struct *task)
 998 {
 999         struct mm_struct *mm;
1000
1001         task_lock(task);
1002         mm = task->mm;
1003         if (mm) {
1004                 if (task->flags & PF_KTHREAD)
1005                         mm = NULL;
1006                 else
1007                         mmget(mm);
1008         }
1009         task_unlock(task);
1010         return mm;
1011 }
1012 EXPORT_SYMBOL_GPL(get_task_mm);
1013
1014 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1015 {
1016         struct mm_struct *mm;
1017         int err;
1018
1019         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1020         if (err)
1021                 return ERR_PTR(err);
1022
1023         mm = get_task_mm(task);
1024         if (mm && mm != current->mm &&
1025                         !ptrace_may_access(task, mode)) {
1026                 mmput(mm);
1027                 mm = ERR_PTR(-EACCES);
1028         }
1029         mutex_unlock(&task->signal->cred_guard_mutex);
1030
1031         return mm;
1032 }
1033
1034 static void complete_vfork_done(struct task_struct *tsk)
1035 {
1036         struct completion *vfork;
1037
1038         task_lock(tsk);
1039         vfork = tsk->vfork_done;
1040         if (likely(vfork)) {
1041                 tsk->vfork_done = NULL;
1042                 complete(vfork);
1043         }
1044         task_unlock(tsk);
1045 }
1046
1047 static int wait_for_vfork_done(struct task_struct *child,
1048                                 struct completion *vfork)
1049 {
1050         int killed;
1051
1052         freezer_do_not_count();
1053         killed = wait_for_completion_killable(vfork);
1054         freezer_count();
1055
1056         if (killed) {
1057                 task_lock(child);
1058                 child->vfork_done = NULL;
1059                 task_unlock(child);
1060         }
1061
1062         put_task_struct(child);
1063         return killed;
1064 }
1065
1066 /* Please note the differences between mmput and mm_release.
1067  * mmput is called whenever we stop holding onto a mm_struct,
1068  * error success whatever.
1069  *
1070  * mm_release is called after a mm_struct has been removed
1071  * from the current process.
1072  *
1073  * This difference is important for error handling, when we
1074  * only half set up a mm_struct for a new process and need to restore
1075  * the old one.  Because we mmput the new mm_struct before
1076  * restoring the old one. . .
1077  * Eric Biederman 10 January 1998
1078  */
1079 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1080 {
1081         /* Get rid of any futexes when releasing the mm */
1082 #ifdef CONFIG_FUTEX
1083         if (unlikely(tsk->robust_list)) {
1084                 exit_robust_list(tsk);
1085                 tsk->robust_list = NULL;
1086         }
1087 #ifdef CONFIG_COMPAT
1088         if (unlikely(tsk->compat_robust_list)) {
1089                 compat_exit_robust_list(tsk);
1090                 tsk->compat_robust_list = NULL;
1091         }
1092 #endif
1093         if (unlikely(!list_empty(&tsk->pi_state_list)))
1094                 exit_pi_state_list(tsk);
1095 #endif
1096
1097         uprobe_free_utask(tsk);
1098
1099         /* Get rid of any cached register state */
1100         deactivate_mm(tsk, mm);
1101
1102         /*
1103          * Signal userspace if we're not exiting with a core dump
1104          * because we want to leave the value intact for debugging
1105          * purposes.
1106          */
1107         if (tsk->clear_child_tid) {
1108                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1109                     atomic_read(&mm->mm_users) > 1) {
1110                         /*
1111                          * We don't check the error code - if userspace has
1112                          * not set up a proper pointer then tough luck.
1113                          */
1114                         put_user(0, tsk->clear_child_tid);
1115                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1116                                         1, NULL, NULL, 0);
1117                 }
1118                 tsk->clear_child_tid = NULL;
1119         }
1120
1121         /*
1122          * All done, finally we can wake up parent and return this mm to him.
1123          * Also kthread_stop() uses this completion for synchronization.
1124          */
1125         if (tsk->vfork_done)
1126                 complete_vfork_done(tsk);
1127 }
1128
1129 /*
1130  * Allocate a new mm structure and copy contents from the
1131  * mm structure of the passed in task structure.
1132  */
1133 static struct mm_struct *dup_mm(struct task_struct *tsk)
1134 {
1135         struct mm_struct *mm, *oldmm = current->mm;
1136         int err;
1137
1138         mm = allocate_mm();
1139         if (!mm)
1140                 goto fail_nomem;
1141
1142         memcpy(mm, oldmm, sizeof(*mm));
1143
1144         if (!mm_init(mm, tsk, mm->user_ns))
1145                 goto fail_nomem;
1146
1147         err = dup_mmap(mm, oldmm);
1148         if (err)
1149                 goto free_pt;
1150
1151         mm->hiwater_rss = get_mm_rss(mm);
1152         mm->hiwater_vm = mm->total_vm;
1153
1154         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1155                 goto free_pt;
1156
1157         return mm;
1158
1159 free_pt:
1160         /* don't put binfmt in mmput, we haven't got module yet */
1161         mm->binfmt = NULL;
1162         mmput(mm);
1163
1164 fail_nomem:
1165         return NULL;
1166 }
1167
1168 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1169 {
1170         struct mm_struct *mm, *oldmm;
1171         int retval;
1172
1173         tsk->min_flt = tsk->maj_flt = 0;
1174         tsk->nvcsw = tsk->nivcsw = 0;
1175 #ifdef CONFIG_DETECT_HUNG_TASK
1176         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1177 #endif
1178
1179         tsk->mm = NULL;
1180         tsk->active_mm = NULL;
1181
1182         /*
1183          * Are we cloning a kernel thread?
1184          *
1185          * We need to steal a active VM for that..
1186          */
1187         oldmm = current->mm;
1188         if (!oldmm)
1189                 return 0;
1190
1191         /* initialize the new vmacache entries */
1192         vmacache_flush(tsk);
1193
1194         if (clone_flags & CLONE_VM) {
1195                 mmget(oldmm);
1196                 mm = oldmm;
1197                 goto good_mm;
1198         }
1199
1200         retval = -ENOMEM;
1201         mm = dup_mm(tsk);
1202         if (!mm)
1203                 goto fail_nomem;
1204
1205 good_mm:
1206         tsk->mm = mm;
1207         tsk->active_mm = mm;
1208         return 0;
1209
1210 fail_nomem:
1211         return retval;
1212 }
1213
1214 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1215 {
1216         struct fs_struct *fs = current->fs;
1217         if (clone_flags & CLONE_FS) {
1218                 /* tsk->fs is already what we want */
1219                 spin_lock(&fs->lock);
1220                 if (fs->in_exec) {
1221                         spin_unlock(&fs->lock);
1222                         return -EAGAIN;
1223                 }
1224                 fs->users++;
1225                 spin_unlock(&fs->lock);
1226                 return 0;
1227         }
1228         tsk->fs = copy_fs_struct(fs);
1229         if (!tsk->fs)
1230                 return -ENOMEM;
1231         return 0;
1232 }
1233
1234 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1235 {
1236         struct files_struct *oldf, *newf;
1237         int error = 0;
1238
1239         /*
1240          * A background process may not have any files ...
1241          */
1242         oldf = current->files;
1243         if (!oldf)
1244                 goto out;
1245
1246         if (clone_flags & CLONE_FILES) {
1247                 atomic_inc(&oldf->count);
1248                 goto out;
1249         }
1250
1251         newf = dup_fd(oldf, &error);
1252         if (!newf)
1253                 goto out;
1254
1255         tsk->files = newf;
1256         error = 0;
1257 out:
1258         return error;
1259 }
1260
1261 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1262 {
1263 #ifdef CONFIG_BLOCK
1264         struct io_context *ioc = current->io_context;
1265         struct io_context *new_ioc;
1266
1267         if (!ioc)
1268                 return 0;
1269         /*
1270          * Share io context with parent, if CLONE_IO is set
1271          */
1272         if (clone_flags & CLONE_IO) {
1273                 ioc_task_link(ioc);
1274                 tsk->io_context = ioc;
1275         } else if (ioprio_valid(ioc->ioprio)) {
1276                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1277                 if (unlikely(!new_ioc))
1278                         return -ENOMEM;
1279
1280                 new_ioc->ioprio = ioc->ioprio;
1281                 put_io_context(new_ioc);
1282         }
1283 #endif
1284         return 0;
1285 }
1286
1287 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1288 {
1289         struct sighand_struct *sig;
1290
1291         if (clone_flags & CLONE_SIGHAND) {
1292                 atomic_inc(&current->sighand->count);
1293                 return 0;
1294         }
1295         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1296         rcu_assign_pointer(tsk->sighand, sig);
1297         if (!sig)
1298                 return -ENOMEM;
1299
1300         atomic_set(&sig->count, 1);
1301         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1302         return 0;
1303 }
1304
1305 void __cleanup_sighand(struct sighand_struct *sighand)
1306 {
1307         if (atomic_dec_and_test(&sighand->count)) {
1308                 signalfd_cleanup(sighand);
1309                 /*
1310                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1311                  * without an RCU grace period, see __lock_task_sighand().
1312                  */
1313                 kmem_cache_free(sighand_cachep, sighand);
1314         }
1315 }
1316
1317 #ifdef CONFIG_POSIX_TIMERS
1318 /*
1319  * Initialize POSIX timer handling for a thread group.
1320  */
1321 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1322 {
1323         unsigned long cpu_limit;
1324
1325         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1326         if (cpu_limit != RLIM_INFINITY) {
1327                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1328                 sig->cputimer.running = true;
1329         }
1330
1331         /* The timer lists. */
1332         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1333         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1334         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1335 }
1336 #else
1337 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1338 #endif
1339
1340 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1341 {
1342         struct signal_struct *sig;
1343
1344         if (clone_flags & CLONE_THREAD)
1345                 return 0;
1346
1347         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1348         tsk->signal = sig;
1349         if (!sig)
1350                 return -ENOMEM;
1351
1352         sig->nr_threads = 1;
1353         atomic_set(&sig->live, 1);
1354         atomic_set(&sig->sigcnt, 1);
1355
1356         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1357         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1358         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1359
1360         init_waitqueue_head(&sig->wait_chldexit);
1361         sig->curr_target = tsk;
1362         init_sigpending(&sig->shared_pending);
1363         seqlock_init(&sig->stats_lock);
1364         prev_cputime_init(&sig->prev_cputime);
1365
1366 #ifdef CONFIG_POSIX_TIMERS
1367         INIT_LIST_HEAD(&sig->posix_timers);
1368         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1369         sig->real_timer.function = it_real_fn;
1370 #endif
1371
1372         task_lock(current->group_leader);
1373         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1374         task_unlock(current->group_leader);
1375
1376         posix_cpu_timers_init_group(sig);
1377
1378         tty_audit_fork(sig);
1379         sched_autogroup_fork(sig);
1380
1381         sig->oom_score_adj = current->signal->oom_score_adj;
1382         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1383
1384         mutex_init(&sig->cred_guard_mutex);
1385
1386         return 0;
1387 }
1388
1389 static void copy_seccomp(struct task_struct *p)
1390 {
1391 #ifdef CONFIG_SECCOMP
1392         /*
1393          * Must be called with sighand->lock held, which is common to
1394          * all threads in the group. Holding cred_guard_mutex is not
1395          * needed because this new task is not yet running and cannot
1396          * be racing exec.
1397          */
1398         assert_spin_locked(&current->sighand->siglock);
1399
1400         /* Ref-count the new filter user, and assign it. */
1401         get_seccomp_filter(current);
1402         p->seccomp = current->seccomp;
1403
1404         /*
1405          * Explicitly enable no_new_privs here in case it got set
1406          * between the task_struct being duplicated and holding the
1407          * sighand lock. The seccomp state and nnp must be in sync.
1408          */
1409         if (task_no_new_privs(current))
1410                 task_set_no_new_privs(p);
1411
1412         /*
1413          * If the parent gained a seccomp mode after copying thread
1414          * flags and between before we held the sighand lock, we have
1415          * to manually enable the seccomp thread flag here.
1416          */
1417         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1418                 set_tsk_thread_flag(p, TIF_SECCOMP);
1419 #endif
1420 }
1421
1422 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1423 {
1424         current->clear_child_tid = tidptr;
1425
1426         return task_pid_vnr(current);
1427 }
1428
1429 static void rt_mutex_init_task(struct task_struct *p)
1430 {
1431         raw_spin_lock_init(&p->pi_lock);
1432 #ifdef CONFIG_RT_MUTEXES
1433         p->pi_waiters = RB_ROOT;
1434         p->pi_waiters_leftmost = NULL;
1435         p->pi_blocked_on = NULL;
1436 #endif
1437 }
1438
1439 #ifdef CONFIG_POSIX_TIMERS
1440 /*
1441  * Initialize POSIX timer handling for a single task.
1442  */
1443 static void posix_cpu_timers_init(struct task_struct *tsk)
1444 {
1445         tsk->cputime_expires.prof_exp = 0;
1446         tsk->cputime_expires.virt_exp = 0;
1447         tsk->cputime_expires.sched_exp = 0;
1448         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1449         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1450         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1451 }
1452 #else
1453 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1454 #endif
1455
1456 static inline void
1457 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1458 {
1459          task->pids[type].pid = pid;
1460 }
1461
1462 /*
1463  * This creates a new process as a copy of the old one,
1464  * but does not actually start it yet.
1465  *
1466  * It copies the registers, and all the appropriate
1467  * parts of the process environment (as per the clone
1468  * flags). The actual kick-off is left to the caller.
1469  */
1470 static __latent_entropy struct task_struct *copy_process(
1471                                         unsigned long clone_flags,
1472                                         unsigned long stack_start,
1473                                         unsigned long stack_size,
1474                                         int __user *child_tidptr,
1475                                         struct pid *pid,
1476                                         int trace,
1477                                         unsigned long tls,
1478                                         int node)
1479 {
1480         int retval;
1481         struct task_struct *p;
1482
1483         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1484                 return ERR_PTR(-EINVAL);
1485
1486         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1487                 return ERR_PTR(-EINVAL);
1488
1489         /*
1490          * Thread groups must share signals as well, and detached threads
1491          * can only be started up within the thread group.
1492          */
1493         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1494                 return ERR_PTR(-EINVAL);
1495
1496         /*
1497          * Shared signal handlers imply shared VM. By way of the above,
1498          * thread groups also imply shared VM. Blocking this case allows
1499          * for various simplifications in other code.
1500          */
1501         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1502                 return ERR_PTR(-EINVAL);
1503
1504         /*
1505          * Siblings of global init remain as zombies on exit since they are
1506          * not reaped by their parent (swapper). To solve this and to avoid
1507          * multi-rooted process trees, prevent global and container-inits
1508          * from creating siblings.
1509          */
1510         if ((clone_flags & CLONE_PARENT) &&
1511                                 current->signal->flags & SIGNAL_UNKILLABLE)
1512                 return ERR_PTR(-EINVAL);
1513
1514         /*
1515          * If the new process will be in a different pid or user namespace
1516          * do not allow it to share a thread group with the forking task.
1517          */
1518         if (clone_flags & CLONE_THREAD) {
1519                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1520                     (task_active_pid_ns(current) !=
1521                                 current->nsproxy->pid_ns_for_children))
1522                         return ERR_PTR(-EINVAL);
1523         }
1524
1525         retval = security_task_create(clone_flags);
1526         if (retval)
1527                 goto fork_out;
1528
1529         retval = -ENOMEM;
1530         p = dup_task_struct(current, node);
1531         if (!p)
1532                 goto fork_out;
1533
1534         ftrace_graph_init_task(p);
1535
1536         rt_mutex_init_task(p);
1537
1538 #ifdef CONFIG_PROVE_LOCKING
1539         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1540         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1541 #endif
1542         retval = -EAGAIN;
1543         if (atomic_read(&p->real_cred->user->processes) >=
1544                         task_rlimit(p, RLIMIT_NPROC)) {
1545                 if (p->real_cred->user != INIT_USER &&
1546                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1547                         goto bad_fork_free;
1548         }
1549         current->flags &= ~PF_NPROC_EXCEEDED;
1550
1551         retval = copy_creds(p, clone_flags);
1552         if (retval < 0)
1553                 goto bad_fork_free;
1554
1555         /*
1556          * If multiple threads are within copy_process(), then this check
1557          * triggers too late. This doesn't hurt, the check is only there
1558          * to stop root fork bombs.
1559          */
1560         retval = -EAGAIN;
1561         if (nr_threads >= max_threads)
1562                 goto bad_fork_cleanup_count;
1563
1564         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1565         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1566         p->flags |= PF_FORKNOEXEC;
1567         INIT_LIST_HEAD(&p->children);
1568         INIT_LIST_HEAD(&p->sibling);
1569         rcu_copy_process(p);
1570         p->vfork_done = NULL;
1571         spin_lock_init(&p->alloc_lock);
1572
1573         init_sigpending(&p->pending);
1574
1575         p->utime = p->stime = p->gtime = 0;
1576 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1577         p->utimescaled = p->stimescaled = 0;
1578 #endif
1579         prev_cputime_init(&p->prev_cputime);
1580
1581 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1582         seqcount_init(&p->vtime_seqcount);
1583         p->vtime_snap = 0;
1584         p->vtime_snap_whence = VTIME_INACTIVE;
1585 #endif
1586
1587 #if defined(SPLIT_RSS_COUNTING)
1588         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1589 #endif
1590
1591         p->default_timer_slack_ns = current->timer_slack_ns;
1592
1593         task_io_accounting_init(&p->ioac);
1594         acct_clear_integrals(p);
1595
1596         posix_cpu_timers_init(p);
1597
1598         p->start_time = ktime_get_ns();
1599         p->real_start_time = ktime_get_boot_ns();
1600         p->io_context = NULL;
1601         p->audit_context = NULL;
1602         cgroup_fork(p);
1603 #ifdef CONFIG_NUMA
1604         p->mempolicy = mpol_dup(p->mempolicy);
1605         if (IS_ERR(p->mempolicy)) {
1606                 retval = PTR_ERR(p->mempolicy);
1607                 p->mempolicy = NULL;
1608                 goto bad_fork_cleanup_threadgroup_lock;
1609         }
1610 #endif
1611 #ifdef CONFIG_CPUSETS
1612         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1613         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1614         seqcount_init(&p->mems_allowed_seq);
1615 #endif
1616 #ifdef CONFIG_TRACE_IRQFLAGS
1617         p->irq_events = 0;
1618         p->hardirqs_enabled = 0;
1619         p->hardirq_enable_ip = 0;
1620         p->hardirq_enable_event = 0;
1621         p->hardirq_disable_ip = _THIS_IP_;
1622         p->hardirq_disable_event = 0;
1623         p->softirqs_enabled = 1;
1624         p->softirq_enable_ip = _THIS_IP_;
1625         p->softirq_enable_event = 0;
1626         p->softirq_disable_ip = 0;
1627         p->softirq_disable_event = 0;
1628         p->hardirq_context = 0;
1629         p->softirq_context = 0;
1630 #endif
1631
1632         p->pagefault_disabled = 0;
1633
1634 #ifdef CONFIG_LOCKDEP
1635         p->lockdep_depth = 0; /* no locks held yet */
1636         p->curr_chain_key = 0;
1637         p->lockdep_recursion = 0;
1638 #endif
1639
1640 #ifdef CONFIG_DEBUG_MUTEXES
1641         p->blocked_on = NULL; /* not blocked yet */
1642 #endif
1643 #ifdef CONFIG_BCACHE
1644         p->sequential_io        = 0;
1645         p->sequential_io_avg    = 0;
1646 #endif
1647
1648         /* Perform scheduler related setup. Assign this task to a CPU. */
1649         retval = sched_fork(clone_flags, p);
1650         if (retval)
1651                 goto bad_fork_cleanup_policy;
1652
1653         retval = perf_event_init_task(p);
1654         if (retval)
1655                 goto bad_fork_cleanup_policy;
1656         retval = audit_alloc(p);
1657         if (retval)
1658                 goto bad_fork_cleanup_perf;
1659         /* copy all the process information */
1660         shm_init_task(p);
1661         retval = copy_semundo(clone_flags, p);
1662         if (retval)
1663                 goto bad_fork_cleanup_audit;
1664         retval = copy_files(clone_flags, p);
1665         if (retval)
1666                 goto bad_fork_cleanup_semundo;
1667         retval = copy_fs(clone_flags, p);
1668         if (retval)
1669                 goto bad_fork_cleanup_files;
1670         retval = copy_sighand(clone_flags, p);
1671         if (retval)
1672                 goto bad_fork_cleanup_fs;
1673         retval = copy_signal(clone_flags, p);
1674         if (retval)
1675                 goto bad_fork_cleanup_sighand;
1676         retval = copy_mm(clone_flags, p);
1677         if (retval)
1678                 goto bad_fork_cleanup_signal;
1679         retval = copy_namespaces(clone_flags, p);
1680         if (retval)
1681                 goto bad_fork_cleanup_mm;
1682         retval = copy_io(clone_flags, p);
1683         if (retval)
1684                 goto bad_fork_cleanup_namespaces;
1685         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1686         if (retval)
1687                 goto bad_fork_cleanup_io;
1688
1689         if (pid != &init_struct_pid) {
1690                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1691                 if (IS_ERR(pid)) {
1692                         retval = PTR_ERR(pid);
1693                         goto bad_fork_cleanup_thread;
1694                 }
1695         }
1696
1697         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1698         /*
1699          * Clear TID on mm_release()?
1700          */
1701         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1702 #ifdef CONFIG_BLOCK
1703         p->plug = NULL;
1704 #endif
1705 #ifdef CONFIG_FUTEX
1706         p->robust_list = NULL;
1707 #ifdef CONFIG_COMPAT
1708         p->compat_robust_list = NULL;
1709 #endif
1710         INIT_LIST_HEAD(&p->pi_state_list);
1711         p->pi_state_cache = NULL;
1712 #endif
1713         /*
1714          * sigaltstack should be cleared when sharing the same VM
1715          */
1716         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1717                 sas_ss_reset(p);
1718
1719         /*
1720          * Syscall tracing and stepping should be turned off in the
1721          * child regardless of CLONE_PTRACE.
1722          */
1723         user_disable_single_step(p);
1724         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1725 #ifdef TIF_SYSCALL_EMU
1726         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1727 #endif
1728         clear_all_latency_tracing(p);
1729
1730         /* ok, now we should be set up.. */
1731         p->pid = pid_nr(pid);
1732         if (clone_flags & CLONE_THREAD) {
1733                 p->exit_signal = -1;
1734                 p->group_leader = current->group_leader;
1735                 p->tgid = current->tgid;
1736         } else {
1737                 if (clone_flags & CLONE_PARENT)
1738                         p->exit_signal = current->group_leader->exit_signal;
1739                 else
1740                         p->exit_signal = (clone_flags & CSIGNAL);
1741                 p->group_leader = p;
1742                 p->tgid = p->pid;
1743         }
1744
1745         p->nr_dirtied = 0;
1746         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1747         p->dirty_paused_when = 0;
1748
1749         p->pdeath_signal = 0;
1750         INIT_LIST_HEAD(&p->thread_group);
1751         p->task_works = NULL;
1752
1753         cgroup_threadgroup_change_begin(current);
1754         /*
1755          * Ensure that the cgroup subsystem policies allow the new process to be
1756          * forked. It should be noted the the new process's css_set can be changed
1757          * between here and cgroup_post_fork() if an organisation operation is in
1758          * progress.
1759          */
1760         retval = cgroup_can_fork(p);
1761         if (retval)
1762                 goto bad_fork_free_pid;
1763
1764         /*
1765          * Make it visible to the rest of the system, but dont wake it up yet.
1766          * Need tasklist lock for parent etc handling!
1767          */
1768         write_lock_irq(&tasklist_lock);
1769
1770         /* CLONE_PARENT re-uses the old parent */
1771         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1772                 p->real_parent = current->real_parent;
1773                 p->parent_exec_id = current->parent_exec_id;
1774         } else {
1775                 p->real_parent = current;
1776                 p->parent_exec_id = current->self_exec_id;
1777         }
1778
1779         spin_lock(&current->sighand->siglock);
1780
1781         /*
1782          * Copy seccomp details explicitly here, in case they were changed
1783          * before holding sighand lock.
1784          */
1785         copy_seccomp(p);
1786
1787         /*
1788          * Process group and session signals need to be delivered to just the
1789          * parent before the fork or both the parent and the child after the
1790          * fork. Restart if a signal comes in before we add the new process to
1791          * it's process group.
1792          * A fatal signal pending means that current will exit, so the new
1793          * thread can't slip out of an OOM kill (or normal SIGKILL).
1794         */
1795         recalc_sigpending();
1796         if (signal_pending(current)) {
1797                 spin_unlock(&current->sighand->siglock);
1798                 write_unlock_irq(&tasklist_lock);
1799                 retval = -ERESTARTNOINTR;
1800                 goto bad_fork_cancel_cgroup;
1801         }
1802
1803         if (likely(p->pid)) {
1804                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1805
1806                 init_task_pid(p, PIDTYPE_PID, pid);
1807                 if (thread_group_leader(p)) {
1808                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1809                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1810
1811                         if (is_child_reaper(pid)) {
1812                                 ns_of_pid(pid)->child_reaper = p;
1813                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1814                         }
1815
1816                         p->signal->leader_pid = pid;
1817                         p->signal->tty = tty_kref_get(current->signal->tty);
1818                         /*
1819                          * Inherit has_child_subreaper flag under the same
1820                          * tasklist_lock with adding child to the process tree
1821                          * for propagate_has_child_subreaper optimization.
1822                          */
1823                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1824                                                          p->real_parent->signal->is_child_subreaper;
1825                         list_add_tail(&p->sibling, &p->real_parent->children);
1826                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1827                         attach_pid(p, PIDTYPE_PGID);
1828                         attach_pid(p, PIDTYPE_SID);
1829                         __this_cpu_inc(process_counts);
1830                 } else {
1831                         current->signal->nr_threads++;
1832                         atomic_inc(&current->signal->live);
1833                         atomic_inc(&current->signal->sigcnt);
1834                         list_add_tail_rcu(&p->thread_group,
1835                                           &p->group_leader->thread_group);
1836                         list_add_tail_rcu(&p->thread_node,
1837                                           &p->signal->thread_head);
1838                 }
1839                 attach_pid(p, PIDTYPE_PID);
1840                 nr_threads++;
1841         }
1842
1843         total_forks++;
1844         spin_unlock(&current->sighand->siglock);
1845         syscall_tracepoint_update(p);
1846         write_unlock_irq(&tasklist_lock);
1847
1848         proc_fork_connector(p);
1849         cgroup_post_fork(p);
1850         cgroup_threadgroup_change_end(current);
1851         perf_event_fork(p);
1852
1853         trace_task_newtask(p, clone_flags);
1854         uprobe_copy_process(p, clone_flags);
1855
1856         return p;
1857
1858 bad_fork_cancel_cgroup:
1859         cgroup_cancel_fork(p);
1860 bad_fork_free_pid:
1861         cgroup_threadgroup_change_end(current);
1862         if (pid != &init_struct_pid)
1863                 free_pid(pid);
1864 bad_fork_cleanup_thread:
1865         exit_thread(p);
1866 bad_fork_cleanup_io:
1867         if (p->io_context)
1868                 exit_io_context(p);
1869 bad_fork_cleanup_namespaces:
1870         exit_task_namespaces(p);
1871 bad_fork_cleanup_mm:
1872         if (p->mm)
1873                 mmput(p->mm);
1874 bad_fork_cleanup_signal:
1875         if (!(clone_flags & CLONE_THREAD))
1876                 free_signal_struct(p->signal);
1877 bad_fork_cleanup_sighand:
1878         __cleanup_sighand(p->sighand);
1879 bad_fork_cleanup_fs:
1880         exit_fs(p); /* blocking */
1881 bad_fork_cleanup_files:
1882         exit_files(p); /* blocking */
1883 bad_fork_cleanup_semundo:
1884         exit_sem(p);
1885 bad_fork_cleanup_audit:
1886         audit_free(p);
1887 bad_fork_cleanup_perf:
1888         perf_event_free_task(p);
1889 bad_fork_cleanup_policy:
1890 #ifdef CONFIG_NUMA
1891         mpol_put(p->mempolicy);
1892 bad_fork_cleanup_threadgroup_lock:
1893 #endif
1894         delayacct_tsk_free(p);
1895 bad_fork_cleanup_count:
1896         atomic_dec(&p->cred->user->processes);
1897         exit_creds(p);
1898 bad_fork_free:
1899         p->state = TASK_DEAD;
1900         put_task_stack(p);
1901         free_task(p);
1902 fork_out:
1903         return ERR_PTR(retval);
1904 }
1905
1906 static inline void init_idle_pids(struct pid_link *links)
1907 {
1908         enum pid_type type;
1909
1910         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1911                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1912                 links[type].pid = &init_struct_pid;
1913         }
1914 }
1915
1916 struct task_struct *fork_idle(int cpu)
1917 {
1918         struct task_struct *task;
1919         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1920                             cpu_to_node(cpu));
1921         if (!IS_ERR(task)) {
1922                 init_idle_pids(task->pids);
1923                 init_idle(task, cpu);
1924         }
1925
1926         return task;
1927 }
1928
1929 /*
1930  *  Ok, this is the main fork-routine.
1931  *
1932  * It copies the process, and if successful kick-starts
1933  * it and waits for it to finish using the VM if required.
1934  */
1935 long _do_fork(unsigned long clone_flags,
1936               unsigned long stack_start,
1937               unsigned long stack_size,
1938               int __user *parent_tidptr,
1939               int __user *child_tidptr,
1940               unsigned long tls)
1941 {
1942         struct task_struct *p;
1943         int trace = 0;
1944         long nr;
1945
1946         /*
1947          * Determine whether and which event to report to ptracer.  When
1948          * called from kernel_thread or CLONE_UNTRACED is explicitly
1949          * requested, no event is reported; otherwise, report if the event
1950          * for the type of forking is enabled.
1951          */
1952         if (!(clone_flags & CLONE_UNTRACED)) {
1953                 if (clone_flags & CLONE_VFORK)
1954                         trace = PTRACE_EVENT_VFORK;
1955                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1956                         trace = PTRACE_EVENT_CLONE;
1957                 else
1958                         trace = PTRACE_EVENT_FORK;
1959
1960                 if (likely(!ptrace_event_enabled(current, trace)))
1961                         trace = 0;
1962         }
1963
1964         p = copy_process(clone_flags, stack_start, stack_size,
1965                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
1966         add_latent_entropy();
1967         /*
1968          * Do this prior waking up the new thread - the thread pointer
1969          * might get invalid after that point, if the thread exits quickly.
1970          */
1971         if (!IS_ERR(p)) {
1972                 struct completion vfork;
1973                 struct pid *pid;
1974
1975                 trace_sched_process_fork(current, p);
1976
1977                 pid = get_task_pid(p, PIDTYPE_PID);
1978                 nr = pid_vnr(pid);
1979
1980                 if (clone_flags & CLONE_PARENT_SETTID)
1981                         put_user(nr, parent_tidptr);
1982
1983                 if (clone_flags & CLONE_VFORK) {
1984                         p->vfork_done = &vfork;
1985                         init_completion(&vfork);
1986                         get_task_struct(p);
1987                 }
1988
1989                 wake_up_new_task(p);
1990
1991                 /* forking complete and child started to run, tell ptracer */
1992                 if (unlikely(trace))
1993                         ptrace_event_pid(trace, pid);
1994
1995                 if (clone_flags & CLONE_VFORK) {
1996                         if (!wait_for_vfork_done(p, &vfork))
1997                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1998                 }
1999
2000                 put_pid(pid);
2001         } else {
2002                 nr = PTR_ERR(p);
2003         }
2004         return nr;
2005 }
2006
2007 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2008 /* For compatibility with architectures that call do_fork directly rather than
2009  * using the syscall entry points below. */
2010 long do_fork(unsigned long clone_flags,
2011               unsigned long stack_start,
2012               unsigned long stack_size,
2013               int __user *parent_tidptr,
2014               int __user *child_tidptr)
2015 {
2016         return _do_fork(clone_flags, stack_start, stack_size,
2017                         parent_tidptr, child_tidptr, 0);
2018 }
2019 #endif
2020
2021 /*
2022  * Create a kernel thread.
2023  */
2024 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2025 {
2026         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2027                 (unsigned long)arg, NULL, NULL, 0);
2028 }
2029
2030 #ifdef __ARCH_WANT_SYS_FORK
2031 SYSCALL_DEFINE0(fork)
2032 {
2033 #ifdef CONFIG_MMU
2034         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2035 #else
2036         /* can not support in nommu mode */
2037         return -EINVAL;
2038 #endif
2039 }
2040 #endif
2041
2042 #ifdef __ARCH_WANT_SYS_VFORK
2043 SYSCALL_DEFINE0(vfork)
2044 {
2045         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2046                         0, NULL, NULL, 0);
2047 }
2048 #endif
2049
2050 #ifdef __ARCH_WANT_SYS_CLONE
2051 #ifdef CONFIG_CLONE_BACKWARDS
2052 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2053                  int __user *, parent_tidptr,
2054                  unsigned long, tls,
2055                  int __user *, child_tidptr)
2056 #elif defined(CONFIG_CLONE_BACKWARDS2)
2057 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2058                  int __user *, parent_tidptr,
2059                  int __user *, child_tidptr,
2060                  unsigned long, tls)
2061 #elif defined(CONFIG_CLONE_BACKWARDS3)
2062 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2063                 int, stack_size,
2064                 int __user *, parent_tidptr,
2065                 int __user *, child_tidptr,
2066                 unsigned long, tls)
2067 #else
2068 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2069                  int __user *, parent_tidptr,
2070                  int __user *, child_tidptr,
2071                  unsigned long, tls)
2072 #endif
2073 {
2074         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2075 }
2076 #endif
2077
2078 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2079 {
2080         struct task_struct *leader, *parent, *child;
2081         int res;
2082
2083         read_lock(&tasklist_lock);
2084         leader = top = top->group_leader;
2085 down:
2086         for_each_thread(leader, parent) {
2087                 list_for_each_entry(child, &parent->children, sibling) {
2088                         res = visitor(child, data);
2089                         if (res) {
2090                                 if (res < 0)
2091                                         goto out;
2092                                 leader = child;
2093                                 goto down;
2094                         }
2095 up:
2096                         ;
2097                 }
2098         }
2099
2100         if (leader != top) {
2101                 child = leader;
2102                 parent = child->real_parent;
2103                 leader = parent->group_leader;
2104                 goto up;
2105         }
2106 out:
2107         read_unlock(&tasklist_lock);
2108 }
2109
2110 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2111 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2112 #endif
2113
2114 static void sighand_ctor(void *data)
2115 {
2116         struct sighand_struct *sighand = data;
2117
2118         spin_lock_init(&sighand->siglock);
2119         init_waitqueue_head(&sighand->signalfd_wqh);
2120 }
2121
2122 void __init proc_caches_init(void)
2123 {
2124         sighand_cachep = kmem_cache_create("sighand_cache",
2125                         sizeof(struct sighand_struct), 0,
2126                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
2127                         SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2128         signal_cachep = kmem_cache_create("signal_cache",
2129                         sizeof(struct signal_struct), 0,
2130                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2131                         NULL);
2132         files_cachep = kmem_cache_create("files_cache",
2133                         sizeof(struct files_struct), 0,
2134                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2135                         NULL);
2136         fs_cachep = kmem_cache_create("fs_cache",
2137                         sizeof(struct fs_struct), 0,
2138                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2139                         NULL);
2140         /*
2141          * FIXME! The "sizeof(struct mm_struct)" currently includes the
2142          * whole struct cpumask for the OFFSTACK case. We could change
2143          * this to *only* allocate as much of it as required by the
2144          * maximum number of CPU's we can ever have.  The cpumask_allocation
2145          * is at the end of the structure, exactly for that reason.
2146          */
2147         mm_cachep = kmem_cache_create("mm_struct",
2148                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2149                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2150                         NULL);
2151         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2152         mmap_init();
2153         nsproxy_cache_init();
2154 }
2155
2156 /*
2157  * Check constraints on flags passed to the unshare system call.
2158  */
2159 static int check_unshare_flags(unsigned long unshare_flags)
2160 {
2161         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2162                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2163                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2164                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2165                 return -EINVAL;
2166         /*
2167          * Not implemented, but pretend it works if there is nothing
2168          * to unshare.  Note that unsharing the address space or the
2169          * signal handlers also need to unshare the signal queues (aka
2170          * CLONE_THREAD).
2171          */
2172         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2173                 if (!thread_group_empty(current))
2174                         return -EINVAL;
2175         }
2176         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2177                 if (atomic_read(&current->sighand->count) > 1)
2178                         return -EINVAL;
2179         }
2180         if (unshare_flags & CLONE_VM) {
2181                 if (!current_is_single_threaded())
2182                         return -EINVAL;
2183         }
2184
2185         return 0;
2186 }
2187
2188 /*
2189  * Unshare the filesystem structure if it is being shared
2190  */
2191 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2192 {
2193         struct fs_struct *fs = current->fs;
2194
2195         if (!(unshare_flags & CLONE_FS) || !fs)
2196                 return 0;
2197
2198         /* don't need lock here; in the worst case we'll do useless copy */
2199         if (fs->users == 1)
2200                 return 0;
2201
2202         *new_fsp = copy_fs_struct(fs);
2203         if (!*new_fsp)
2204                 return -ENOMEM;
2205
2206         return 0;
2207 }
2208
2209 /*
2210  * Unshare file descriptor table if it is being shared
2211  */
2212 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2213 {
2214         struct files_struct *fd = current->files;
2215         int error = 0;
2216
2217         if ((unshare_flags & CLONE_FILES) &&
2218             (fd && atomic_read(&fd->count) > 1)) {
2219                 *new_fdp = dup_fd(fd, &error);
2220                 if (!*new_fdp)
2221                         return error;
2222         }
2223
2224         return 0;
2225 }
2226
2227 /*
2228  * unshare allows a process to 'unshare' part of the process
2229  * context which was originally shared using clone.  copy_*
2230  * functions used by do_fork() cannot be used here directly
2231  * because they modify an inactive task_struct that is being
2232  * constructed. Here we are modifying the current, active,
2233  * task_struct.
2234  */
2235 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2236 {
2237         struct fs_struct *fs, *new_fs = NULL;
2238         struct files_struct *fd, *new_fd = NULL;
2239         struct cred *new_cred = NULL;
2240         struct nsproxy *new_nsproxy = NULL;
2241         int do_sysvsem = 0;
2242         int err;
2243
2244         /*
2245          * If unsharing a user namespace must also unshare the thread group
2246          * and unshare the filesystem root and working directories.
2247          */
2248         if (unshare_flags & CLONE_NEWUSER)
2249                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2250         /*
2251          * If unsharing vm, must also unshare signal handlers.
2252          */
2253         if (unshare_flags & CLONE_VM)
2254                 unshare_flags |= CLONE_SIGHAND;
2255         /*
2256          * If unsharing a signal handlers, must also unshare the signal queues.
2257          */
2258         if (unshare_flags & CLONE_SIGHAND)
2259                 unshare_flags |= CLONE_THREAD;
2260         /*
2261          * If unsharing namespace, must also unshare filesystem information.
2262          */
2263         if (unshare_flags & CLONE_NEWNS)
2264                 unshare_flags |= CLONE_FS;
2265
2266         err = check_unshare_flags(unshare_flags);
2267         if (err)
2268                 goto bad_unshare_out;
2269         /*
2270          * CLONE_NEWIPC must also detach from the undolist: after switching
2271          * to a new ipc namespace, the semaphore arrays from the old
2272          * namespace are unreachable.
2273          */
2274         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2275                 do_sysvsem = 1;
2276         err = unshare_fs(unshare_flags, &new_fs);
2277         if (err)
2278                 goto bad_unshare_out;
2279         err = unshare_fd(unshare_flags, &new_fd);
2280         if (err)
2281                 goto bad_unshare_cleanup_fs;
2282         err = unshare_userns(unshare_flags, &new_cred);
2283         if (err)
2284                 goto bad_unshare_cleanup_fd;
2285         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2286                                          new_cred, new_fs);
2287         if (err)
2288                 goto bad_unshare_cleanup_cred;
2289
2290         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2291                 if (do_sysvsem) {
2292                         /*
2293                          * CLONE_SYSVSEM is equivalent to sys_exit().
2294                          */
2295                         exit_sem(current);
2296                 }
2297                 if (unshare_flags & CLONE_NEWIPC) {
2298                         /* Orphan segments in old ns (see sem above). */
2299                         exit_shm(current);
2300                         shm_init_task(current);
2301                 }
2302
2303                 if (new_nsproxy)
2304                         switch_task_namespaces(current, new_nsproxy);
2305
2306                 task_lock(current);
2307
2308                 if (new_fs) {
2309                         fs = current->fs;
2310                         spin_lock(&fs->lock);
2311                         current->fs = new_fs;
2312                         if (--fs->users)
2313                                 new_fs = NULL;
2314                         else
2315                                 new_fs = fs;
2316                         spin_unlock(&fs->lock);
2317                 }
2318
2319                 if (new_fd) {
2320                         fd = current->files;
2321                         current->files = new_fd;
2322                         new_fd = fd;
2323                 }
2324
2325                 task_unlock(current);
2326
2327                 if (new_cred) {
2328                         /* Install the new user namespace */
2329                         commit_creds(new_cred);
2330                         new_cred = NULL;
2331                 }
2332         }
2333
2334 bad_unshare_cleanup_cred:
2335         if (new_cred)
2336                 put_cred(new_cred);
2337 bad_unshare_cleanup_fd:
2338         if (new_fd)
2339                 put_files_struct(new_fd);
2340
2341 bad_unshare_cleanup_fs:
2342         if (new_fs)
2343                 free_fs_struct(new_fs);
2344
2345 bad_unshare_out:
2346         return err;
2347 }
2348
2349 /*
2350  *      Helper to unshare the files of the current task.
2351  *      We don't want to expose copy_files internals to
2352  *      the exec layer of the kernel.
2353  */
2354
2355 int unshare_files(struct files_struct **displaced)
2356 {
2357         struct task_struct *task = current;
2358         struct files_struct *copy = NULL;
2359         int error;
2360
2361         error = unshare_fd(CLONE_FILES, &copy);
2362         if (error || !copy) {
2363                 *displaced = NULL;
2364                 return error;
2365         }
2366         *displaced = task->files;
2367         task_lock(task);
2368         task->files = copy;
2369         task_unlock(task);
2370         return 0;
2371 }
2372
2373 int sysctl_max_threads(struct ctl_table *table, int write,
2374                        void __user *buffer, size_t *lenp, loff_t *ppos)
2375 {
2376         struct ctl_table t;
2377         int ret;
2378         int threads = max_threads;
2379         int min = MIN_THREADS;
2380         int max = MAX_THREADS;
2381
2382         t = *table;
2383         t.data = &threads;
2384         t.extra1 = &min;
2385         t.extra2 = &max;
2386
2387         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2388         if (ret || !write)
2389                 return ret;
2390
2391         set_max_threads(threads);
2392
2393         return 0;
2394 }