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