kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/init.h>
  16 #include <linux/unistd.h>
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/completion.h>
  20 #include <linux/personality.h>
  21 #include <linux/mempolicy.h>
  22 #include <linux/sem.h>
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/iocontext.h>
  26 #include <linux/key.h>
  27 #include <linux/binfmts.h>
  28 #include <linux/mman.h>
  29 #include <linux/mmu_notifier.h>
  30 #include <linux/fs.h>
  31 #include <linux/mm.h>
  32 #include <linux/vmacache.h>
  33 #include <linux/nsproxy.h>
  34 #include <linux/capability.h>
  35 #include <linux/cpu.h>
  36 #include <linux/cgroup.h>
  37 #include <linux/security.h>
  38 #include <linux/hugetlb.h>
  39 #include <linux/seccomp.h>
  40 #include <linux/swap.h>
  41 #include <linux/syscalls.h>
  42 #include <linux/jiffies.h>
  43 #include <linux/futex.h>
  44 #include <linux/compat.h>
  45 #include <linux/kthread.h>
  46 #include <linux/task_io_accounting_ops.h>
  47 #include <linux/rcupdate.h>
  48 #include <linux/ptrace.h>
  49 #include <linux/mount.h>
  50 #include <linux/audit.h>
  51 #include <linux/memcontrol.h>
  52 #include <linux/ftrace.h>
  53 #include <linux/proc_fs.h>
  54 #include <linux/profile.h>
  55 #include <linux/rmap.h>
  56 #include <linux/ksm.h>
  57 #include <linux/acct.h>
  58 #include <linux/tsacct_kern.h>
  59 #include <linux/cn_proc.h>
  60 #include <linux/freezer.h>
  61 #include <linux/delayacct.h>
  62 #include <linux/taskstats_kern.h>
  63 #include <linux/random.h>
  64 #include <linux/tty.h>
  65 #include <linux/blkdev.h>
  66 #include <linux/fs_struct.h>
  67 #include <linux/magic.h>
  68 #include <linux/perf_event.h>
  69 #include <linux/posix-timers.h>
  70 #include <linux/user-return-notifier.h>
  71 #include <linux/oom.h>
  72 #include <linux/khugepaged.h>
  73 #include <linux/signalfd.h>
  74 #include <linux/uprobes.h>
  75 #include <linux/aio.h>
  76 #include <linux/compiler.h>
  77 #include <linux/sysctl.h>
  78 #include <linux/kcov.h>
  79
  80 #include <asm/pgtable.h>
  81 #include <asm/pgalloc.h>
  82 #include <asm/uaccess.h>
  83 #include <asm/mmu_context.h>
  84 #include <asm/cacheflush.h>
  85 #include <asm/tlbflush.h>
  86
  87 #include <trace/events/sched.h>
  88
  89 #define CREATE_TRACE_POINTS
  90 #include <trace/events/task.h>
  91
  92 /*
  93  * Minimum number of threads to boot the kernel
  94  */
  95 #define MIN_THREADS 20
  96
  97 /*
  98  * Maximum number of threads
  99  */
 100 #define MAX_THREADS FUTEX_TID_MASK
 101
 102 /*
 103  * Protected counters by write_lock_irq(&tasklist_lock)
 104  */
 105 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 106 int nr_threads;                 /* The idle threads do not count.. */
 107
 108 int max_threads;                /* tunable limit on nr_threads */
 109
 110 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 111
 112 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 113
 114 #ifdef CONFIG_PROVE_RCU
 115 int lockdep_tasklist_lock_is_held(void)
 116 {
 117         return lockdep_is_held(&tasklist_lock);
 118 }
 119 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 120 #endif /* #ifdef CONFIG_PROVE_RCU */
 121
 122 int nr_processes(void)
 123 {
 124         int cpu;
 125         int total = 0;
 126
 127         for_each_possible_cpu(cpu)
 128                 total += per_cpu(process_counts, cpu);
 129
 130         return total;
 131 }
 132
 133 void __weak arch_release_task_struct(struct task_struct *tsk)
 134 {
 135 }
 136
 137 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 138 static struct kmem_cache *task_struct_cachep;
 139
 140 static inline struct task_struct *alloc_task_struct_node(int node)
 141 {
 142         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 143 }
 144
 145 static inline void free_task_struct(struct task_struct *tsk)
 146 {
 147         kmem_cache_free(task_struct_cachep, tsk);
 148 }
 149 #endif
 150
 151 void __weak arch_release_thread_stack(unsigned long *stack)
 152 {
 153 }
 154
 155 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 156
 157 /*
 158  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 159  * kmemcache based allocator.
 160  */
 161 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 162
 163 #ifdef CONFIG_VMAP_STACK
 164 /*
 165  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 166  * flush.  Try to minimize the number of calls by caching stacks.
 167  */
 168 #define NR_CACHED_STACKS 2
 169 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 170 #endif
 171
 172 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 173 {
 174 #ifdef CONFIG_VMAP_STACK
 175         void *stack;
 176         int i;
 177
 178         local_irq_disable();
 179         for (i = 0; i < NR_CACHED_STACKS; i++) {
 180                 struct vm_struct *s = this_cpu_read(cached_stacks[i]);
 181
 182                 if (!s)
 183                         continue;
 184                 this_cpu_write(cached_stacks[i], NULL);
 185
 186                 tsk->stack_vm_area = s;
 187                 local_irq_enable();
 188                 return s->addr;
 189         }
 190         local_irq_enable();
 191
 192         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_SIZE,
 193                                      VMALLOC_START, VMALLOC_END,
 194                                      THREADINFO_GFP | __GFP_HIGHMEM,
 195                                      PAGE_KERNEL,
 196                                      0, node, __builtin_return_address(0));
 197
 198         /*
 199          * We can't call find_vm_area() in interrupt context, and
 200          * free_thread_stack() can be called in interrupt context,
 201          * so cache the vm_struct.
 202          */
 203         if (stack)
 204                 tsk->stack_vm_area = find_vm_area(stack);
 205         return stack;
 206 #else
 207         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 208                                              THREAD_SIZE_ORDER);
 209
 210         return page ? page_address(page) : NULL;
 211 #endif
 212 }
 213
 214 static inline void free_thread_stack(struct task_struct *tsk)
 215 {
 216 #ifdef CONFIG_VMAP_STACK
 217         if (task_stack_vm_area(tsk)) {
 218                 unsigned long flags;
 219                 int i;
 220
 221                 local_irq_save(flags);
 222                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 223                         if (this_cpu_read(cached_stacks[i]))
 224                                 continue;
 225
 226                         this_cpu_write(cached_stacks[i], tsk->stack_vm_area);
 227                         local_irq_restore(flags);
 228                         return;
 229                 }
 230                 local_irq_restore(flags);
 231
 232                 vfree(tsk->stack);
 233                 return;
 234         }
 235 #endif
 236
 237         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 238 }
 239 # else
 240 static struct kmem_cache *thread_stack_cache;
 241
 242 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 243                                                   int node)
 244 {
 245         return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 246 }
 247
 248 static void free_thread_stack(struct task_struct *tsk)
 249 {
 250         kmem_cache_free(thread_stack_cache, tsk->stack);
 251 }
 252
 253 void thread_stack_cache_init(void)
 254 {
 255         thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
 256                                               THREAD_SIZE, 0, NULL);
 257         BUG_ON(thread_stack_cache == NULL);
 258 }
 259 # endif
 260 #endif
 261
 262 /* SLAB cache for signal_struct structures (tsk->signal) */
 263 static struct kmem_cache *signal_cachep;
 264
 265 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 266 struct kmem_cache *sighand_cachep;
 267
 268 /* SLAB cache for files_struct structures (tsk->files) */
 269 struct kmem_cache *files_cachep;
 270
 271 /* SLAB cache for fs_struct structures (tsk->fs) */
 272 struct kmem_cache *fs_cachep;
 273
 274 /* SLAB cache for vm_area_struct structures */
 275 struct kmem_cache *vm_area_cachep;
 276
 277 /* SLAB cache for mm_struct structures (tsk->mm) */
 278 static struct kmem_cache *mm_cachep;
 279
 280 static void account_kernel_stack(struct task_struct *tsk, int account)
 281 {
 282         void *stack = task_stack_page(tsk);
 283         struct vm_struct *vm = task_stack_vm_area(tsk);
 284
 285         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 286
 287         if (vm) {
 288                 int i;
 289
 290                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 291
 292                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 293                         mod_zone_page_state(page_zone(vm->pages[i]),
 294                                             NR_KERNEL_STACK_KB,
 295                                             PAGE_SIZE / 1024 * account);
 296                 }
 297
 298                 /* All stack pages belong to the same memcg. */
 299                 memcg_kmem_update_page_stat(vm->pages[0], MEMCG_KERNEL_STACK_KB,
 300                                             account * (THREAD_SIZE / 1024));
 301         } else {
 302                 /*
 303                  * All stack pages are in the same zone and belong to the
 304                  * same memcg.
 305                  */
 306                 struct page *first_page = virt_to_page(stack);
 307
 308                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 309                                     THREAD_SIZE / 1024 * account);
 310
 311                 memcg_kmem_update_page_stat(first_page, MEMCG_KERNEL_STACK_KB,
 312                                             account * (THREAD_SIZE / 1024));
 313         }
 314 }
 315
 316 static void release_task_stack(struct task_struct *tsk)
 317 {
 318         account_kernel_stack(tsk, -1);
 319         arch_release_thread_stack(tsk->stack);
 320         free_thread_stack(tsk);
 321         tsk->stack = NULL;
 322 #ifdef CONFIG_VMAP_STACK
 323         tsk->stack_vm_area = NULL;
 324 #endif
 325 }
 326
 327 #ifdef CONFIG_THREAD_INFO_IN_TASK
 328 void put_task_stack(struct task_struct *tsk)
 329 {
 330         if (atomic_dec_and_test(&tsk->stack_refcount))
 331                 release_task_stack(tsk);
 332 }
 333 #endif
 334
 335 void free_task(struct task_struct *tsk)
 336 {
 337 #ifndef CONFIG_THREAD_INFO_IN_TASK
 338         /*
 339          * The task is finally done with both the stack and thread_info,
 340          * so free both.
 341          */
 342         release_task_stack(tsk);
 343 #else
 344         /*
 345          * If the task had a separate stack allocation, it should be gone
 346          * by now.
 347          */
 348         WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
 349 #endif
 350         rt_mutex_debug_task_free(tsk);
 351         ftrace_graph_exit_task(tsk);
 352         put_seccomp_filter(tsk);
 353         arch_release_task_struct(tsk);
 354         free_task_struct(tsk);
 355 }
 356 EXPORT_SYMBOL(free_task);
 357
 358 static inline void free_signal_struct(struct signal_struct *sig)
 359 {
 360         taskstats_tgid_free(sig);
 361         sched_autogroup_exit(sig);
 362         /*
 363          * __mmdrop is not safe to call from softirq context on x86 due to
 364          * pgd_dtor so postpone it to the async context
 365          */
 366         if (sig->oom_mm)
 367                 mmdrop_async(sig->oom_mm);
 368         kmem_cache_free(signal_cachep, sig);
 369 }
 370
 371 static inline void put_signal_struct(struct signal_struct *sig)
 372 {
 373         if (atomic_dec_and_test(&sig->sigcnt))
 374                 free_signal_struct(sig);
 375 }
 376
 377 void __put_task_struct(struct task_struct *tsk)
 378 {
 379         WARN_ON(!tsk->exit_state);
 380         WARN_ON(atomic_read(&tsk->usage));
 381         WARN_ON(tsk == current);
 382
 383         cgroup_free(tsk);
 384         task_numa_free(tsk);
 385         security_task_free(tsk);
 386         exit_creds(tsk);
 387         delayacct_tsk_free(tsk);
 388         put_signal_struct(tsk->signal);
 389
 390         if (!profile_handoff_task(tsk))
 391                 free_task(tsk);
 392 }
 393 EXPORT_SYMBOL_GPL(__put_task_struct);
 394
 395 void __init __weak arch_task_cache_init(void) { }
 396
 397 /*
 398  * set_max_threads
 399  */
 400 static void set_max_threads(unsigned int max_threads_suggested)
 401 {
 402         u64 threads;
 403
 404         /*
 405          * The number of threads shall be limited such that the thread
 406          * structures may only consume a small part of the available memory.
 407          */
 408         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 409                 threads = MAX_THREADS;
 410         else
 411                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 412                                     (u64) THREAD_SIZE * 8UL);
 413
 414         if (threads > max_threads_suggested)
 415                 threads = max_threads_suggested;
 416
 417         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 418 }
 419
 420 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 421 /* Initialized by the architecture: */
 422 int arch_task_struct_size __read_mostly;
 423 #endif
 424
 425 void __init fork_init(void)
 426 {
 427         int i;
 428 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 429 #ifndef ARCH_MIN_TASKALIGN
 430 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
 431 #endif
 432         /* create a slab on which task_structs can be allocated */
 433         task_struct_cachep = kmem_cache_create("task_struct",
 434                         arch_task_struct_size, ARCH_MIN_TASKALIGN,
 435                         SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
 436 #endif
 437
 438         /* do the arch specific task caches init */
 439         arch_task_cache_init();
 440
 441         set_max_threads(MAX_THREADS);
 442
 443         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 444         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 445         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 446                 init_task.signal->rlim[RLIMIT_NPROC];
 447
 448         for (i = 0; i < UCOUNT_COUNTS; i++) {
 449                 init_user_ns.ucount_max[i] = max_threads/2;
 450         }
 451 }
 452
 453 int __weak arch_dup_task_struct(struct task_struct *dst,
 454                                                struct task_struct *src)
 455 {
 456         *dst = *src;
 457         return 0;
 458 }
 459
 460 void set_task_stack_end_magic(struct task_struct *tsk)
 461 {
 462         unsigned long *stackend;
 463
 464         stackend = end_of_stack(tsk);
 465         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 466 }
 467
 468 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 469 {
 470         struct task_struct *tsk;
 471         unsigned long *stack;
 472         struct vm_struct *stack_vm_area;
 473         int err;
 474
 475         if (node == NUMA_NO_NODE)
 476                 node = tsk_fork_get_node(orig);
 477         tsk = alloc_task_struct_node(node);
 478         if (!tsk)
 479                 return NULL;
 480
 481         stack = alloc_thread_stack_node(tsk, node);
 482         if (!stack)
 483                 goto free_tsk;
 484
 485         stack_vm_area = task_stack_vm_area(tsk);
 486
 487         err = arch_dup_task_struct(tsk, orig);
 488
 489         /*
 490          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 491          * sure they're properly initialized before using any stack-related
 492          * functions again.
 493          */
 494         tsk->stack = stack;
 495 #ifdef CONFIG_VMAP_STACK
 496         tsk->stack_vm_area = stack_vm_area;
 497 #endif
 498 #ifdef CONFIG_THREAD_INFO_IN_TASK
 499         atomic_set(&tsk->stack_refcount, 1);
 500 #endif
 501
 502         if (err)
 503                 goto free_stack;
 504
 505 #ifdef CONFIG_SECCOMP
 506         /*
 507          * We must handle setting up seccomp filters once we're under
 508          * the sighand lock in case orig has changed between now and
 509          * then. Until then, filter must be NULL to avoid messing up
 510          * the usage counts on the error path calling free_task.
 511          */
 512         tsk->seccomp.filter = NULL;
 513 #endif
 514
 515         setup_thread_stack(tsk, orig);
 516         clear_user_return_notifier(tsk);
 517         clear_tsk_need_resched(tsk);
 518         set_task_stack_end_magic(tsk);
 519
 520 #ifdef CONFIG_CC_STACKPROTECTOR
 521         tsk->stack_canary = get_random_int();
 522 #endif
 523
 524         /*
 525          * One for us, one for whoever does the "release_task()" (usually
 526          * parent)
 527          */
 528         atomic_set(&tsk->usage, 2);
 529 #ifdef CONFIG_BLK_DEV_IO_TRACE
 530         tsk->btrace_seq = 0;
 531 #endif
 532         tsk->splice_pipe = NULL;
 533         tsk->task_frag.page = NULL;
 534         tsk->wake_q.next = NULL;
 535
 536         account_kernel_stack(tsk, 1);
 537
 538         kcov_task_init(tsk);
 539
 540         return tsk;
 541
 542 free_stack:
 543         free_thread_stack(tsk);
 544 free_tsk:
 545         free_task_struct(tsk);
 546         return NULL;
 547 }
 548
 549 #ifdef CONFIG_MMU
 550 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 551 {
 552         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 553         struct rb_node **rb_link, *rb_parent;
 554         int retval;
 555         unsigned long charge;
 556
 557         uprobe_start_dup_mmap();
 558         if (down_write_killable(&oldmm->mmap_sem)) {
 559                 retval = -EINTR;
 560                 goto fail_uprobe_end;
 561         }
 562         flush_cache_dup_mm(oldmm);
 563         uprobe_dup_mmap(oldmm, mm);
 564         /*
 565          * Not linked in yet - no deadlock potential:
 566          */
 567         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 568
 569         /* No ordering required: file already has been exposed. */
 570         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 571
 572         mm->total_vm = oldmm->total_vm;
 573         mm->data_vm = oldmm->data_vm;
 574         mm->exec_vm = oldmm->exec_vm;
 575         mm->stack_vm = oldmm->stack_vm;
 576
 577         rb_link = &mm->mm_rb.rb_node;
 578         rb_parent = NULL;
 579         pprev = &mm->mmap;
 580         retval = ksm_fork(mm, oldmm);
 581         if (retval)
 582                 goto out;
 583         retval = khugepaged_fork(mm, oldmm);
 584         if (retval)
 585                 goto out;
 586
 587         prev = NULL;
 588         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 589                 struct file *file;
 590
 591                 if (mpnt->vm_flags & VM_DONTCOPY) {
 592                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 593                         continue;
 594                 }
 595                 charge = 0;
 596                 if (mpnt->vm_flags & VM_ACCOUNT) {
 597                         unsigned long len = vma_pages(mpnt);
 598
 599                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 600                                 goto fail_nomem;
 601                         charge = len;
 602                 }
 603                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 604                 if (!tmp)
 605                         goto fail_nomem;
 606                 *tmp = *mpnt;
 607                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 608                 retval = vma_dup_policy(mpnt, tmp);
 609                 if (retval)
 610                         goto fail_nomem_policy;
 611                 tmp->vm_mm = mm;
 612                 if (anon_vma_fork(tmp, mpnt))
 613                         goto fail_nomem_anon_vma_fork;
 614                 tmp->vm_flags &=
 615                         ~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
 616                 tmp->vm_next = tmp->vm_prev = NULL;
 617                 tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
 618                 file = tmp->vm_file;
 619                 if (file) {
 620                         struct inode *inode = file_inode(file);
 621                         struct address_space *mapping = file->f_mapping;
 622
 623                         get_file(file);
 624                         if (tmp->vm_flags & VM_DENYWRITE)
 625                                 atomic_dec(&inode->i_writecount);
 626                         i_mmap_lock_write(mapping);
 627                         if (tmp->vm_flags & VM_SHARED)
 628                                 atomic_inc(&mapping->i_mmap_writable);
 629                         flush_dcache_mmap_lock(mapping);
 630                         /* insert tmp into the share list, just after mpnt */
 631                         vma_interval_tree_insert_after(tmp, mpnt,
 632                                         &mapping->i_mmap);
 633                         flush_dcache_mmap_unlock(mapping);
 634                         i_mmap_unlock_write(mapping);
 635                 }
 636
 637                 /*
 638                  * Clear hugetlb-related page reserves for children. This only
 639                  * affects MAP_PRIVATE mappings. Faults generated by the child
 640                  * are not guaranteed to succeed, even if read-only
 641                  */
 642                 if (is_vm_hugetlb_page(tmp))
 643                         reset_vma_resv_huge_pages(tmp);
 644
 645                 /*
 646                  * Link in the new vma and copy the page table entries.
 647                  */
 648                 *pprev = tmp;
 649                 pprev = &tmp->vm_next;
 650                 tmp->vm_prev = prev;
 651                 prev = tmp;
 652
 653                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 654                 rb_link = &tmp->vm_rb.rb_right;
 655                 rb_parent = &tmp->vm_rb;
 656
 657                 mm->map_count++;
 658                 retval = copy_page_range(mm, oldmm, mpnt);
 659
 660                 if (tmp->vm_ops && tmp->vm_ops->open)
 661                         tmp->vm_ops->open(tmp);
 662
 663                 if (retval)
 664                         goto out;
 665         }
 666         /* a new mm has just been created */
 667         arch_dup_mmap(oldmm, mm);
 668         retval = 0;
 669 out:
 670         up_write(&mm->mmap_sem);
 671         flush_tlb_mm(oldmm);
 672         up_write(&oldmm->mmap_sem);
 673 fail_uprobe_end:
 674         uprobe_end_dup_mmap();
 675         return retval;
 676 fail_nomem_anon_vma_fork:
 677         mpol_put(vma_policy(tmp));
 678 fail_nomem_policy:
 679         kmem_cache_free(vm_area_cachep, tmp);
 680 fail_nomem:
 681         retval = -ENOMEM;
 682         vm_unacct_memory(charge);
 683         goto out;
 684 }
 685
 686 static inline int mm_alloc_pgd(struct mm_struct *mm)
 687 {
 688         mm->pgd = pgd_alloc(mm);
 689         if (unlikely(!mm->pgd))
 690                 return -ENOMEM;
 691         return 0;
 692 }
 693
 694 static inline void mm_free_pgd(struct mm_struct *mm)
 695 {
 696         pgd_free(mm, mm->pgd);
 697 }
 698 #else
 699 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 700 {
 701         down_write(&oldmm->mmap_sem);
 702         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 703         up_write(&oldmm->mmap_sem);
 704         return 0;
 705 }
 706 #define mm_alloc_pgd(mm)        (0)
 707 #define mm_free_pgd(mm)
 708 #endif /* CONFIG_MMU */
 709
 710 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 711
 712 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 713 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 714
 715 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 716
 717 static int __init coredump_filter_setup(char *s)
 718 {
 719         default_dump_filter =
 720                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 721                 MMF_DUMP_FILTER_MASK;
 722         return 1;
 723 }
 724
 725 __setup("coredump_filter=", coredump_filter_setup);
 726
 727 #include <linux/init_task.h>
 728
 729 static void mm_init_aio(struct mm_struct *mm)
 730 {
 731 #ifdef CONFIG_AIO
 732         spin_lock_init(&mm->ioctx_lock);
 733         mm->ioctx_table = NULL;
 734 #endif
 735 }
 736
 737 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 738 {
 739 #ifdef CONFIG_MEMCG
 740         mm->owner = p;
 741 #endif
 742 }
 743
 744 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
 745 {
 746         mm->mmap = NULL;
 747         mm->mm_rb = RB_ROOT;
 748         mm->vmacache_seqnum = 0;
 749         atomic_set(&mm->mm_users, 1);
 750         atomic_set(&mm->mm_count, 1);
 751         init_rwsem(&mm->mmap_sem);
 752         INIT_LIST_HEAD(&mm->mmlist);
 753         mm->core_state = NULL;
 754         atomic_long_set(&mm->nr_ptes, 0);
 755         mm_nr_pmds_init(mm);
 756         mm->map_count = 0;
 757         mm->locked_vm = 0;
 758         mm->pinned_vm = 0;
 759         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 760         spin_lock_init(&mm->page_table_lock);
 761         mm_init_cpumask(mm);
 762         mm_init_aio(mm);
 763         mm_init_owner(mm, p);
 764         mmu_notifier_mm_init(mm);
 765         clear_tlb_flush_pending(mm);
 766 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 767         mm->pmd_huge_pte = NULL;
 768 #endif
 769
 770         if (current->mm) {
 771                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 772                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 773         } else {
 774                 mm->flags = default_dump_filter;
 775                 mm->def_flags = 0;
 776         }
 777
 778         if (mm_alloc_pgd(mm))
 779                 goto fail_nopgd;
 780
 781         if (init_new_context(p, mm))
 782                 goto fail_nocontext;
 783
 784         return mm;
 785
 786 fail_nocontext:
 787         mm_free_pgd(mm);
 788 fail_nopgd:
 789         free_mm(mm);
 790         return NULL;
 791 }
 792
 793 static void check_mm(struct mm_struct *mm)
 794 {
 795         int i;
 796
 797         for (i = 0; i < NR_MM_COUNTERS; i++) {
 798                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 799
 800                 if (unlikely(x))
 801                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 802                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 803         }
 804
 805         if (atomic_long_read(&mm->nr_ptes))
 806                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 807                                 atomic_long_read(&mm->nr_ptes));
 808         if (mm_nr_pmds(mm))
 809                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 810                                 mm_nr_pmds(mm));
 811
 812 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 813         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 814 #endif
 815 }
 816
 817 /*
 818  * Allocate and initialize an mm_struct.
 819  */
 820 struct mm_struct *mm_alloc(void)
 821 {
 822         struct mm_struct *mm;
 823
 824         mm = allocate_mm();
 825         if (!mm)
 826                 return NULL;
 827
 828         memset(mm, 0, sizeof(*mm));
 829         return mm_init(mm, current);
 830 }
 831
 832 /*
 833  * Called when the last reference to the mm
 834  * is dropped: either by a lazy thread or by
 835  * mmput. Free the page directory and the mm.
 836  */
 837 void __mmdrop(struct mm_struct *mm)
 838 {
 839         BUG_ON(mm == &init_mm);
 840         mm_free_pgd(mm);
 841         destroy_context(mm);
 842         mmu_notifier_mm_destroy(mm);
 843         check_mm(mm);
 844         free_mm(mm);
 845 }
 846 EXPORT_SYMBOL_GPL(__mmdrop);
 847
 848 static inline void __mmput(struct mm_struct *mm)
 849 {
 850         VM_BUG_ON(atomic_read(&mm->mm_users));
 851
 852         uprobe_clear_state(mm);
 853         exit_aio(mm);
 854         ksm_exit(mm);
 855         khugepaged_exit(mm); /* must run before exit_mmap */
 856         exit_mmap(mm);
 857         set_mm_exe_file(mm, NULL);
 858         if (!list_empty(&mm->mmlist)) {
 859                 spin_lock(&mmlist_lock);
 860                 list_del(&mm->mmlist);
 861                 spin_unlock(&mmlist_lock);
 862         }
 863         if (mm->binfmt)
 864                 module_put(mm->binfmt->module);
 865         mmdrop(mm);
 866 }
 867
 868 /*
 869  * Decrement the use count and release all resources for an mm.
 870  */
 871 void mmput(struct mm_struct *mm)
 872 {
 873         might_sleep();
 874
 875         if (atomic_dec_and_test(&mm->mm_users))
 876                 __mmput(mm);
 877 }
 878 EXPORT_SYMBOL_GPL(mmput);
 879
 880 #ifdef CONFIG_MMU
 881 static void mmput_async_fn(struct work_struct *work)
 882 {
 883         struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
 884         __mmput(mm);
 885 }
 886
 887 void mmput_async(struct mm_struct *mm)
 888 {
 889         if (atomic_dec_and_test(&mm->mm_users)) {
 890                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
 891                 schedule_work(&mm->async_put_work);
 892         }
 893 }
 894 #endif
 895
 896 /**
 897  * set_mm_exe_file - change a reference to the mm's executable file
 898  *
 899  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 900  *
 901  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 902  * invocations: in mmput() nobody alive left, in execve task is single
 903  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 904  * mm->exe_file, but does so without using set_mm_exe_file() in order
 905  * to do avoid the need for any locks.
 906  */
 907 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 908 {
 909         struct file *old_exe_file;
 910
 911         /*
 912          * It is safe to dereference the exe_file without RCU as
 913          * this function is only called if nobody else can access
 914          * this mm -- see comment above for justification.
 915          */
 916         old_exe_file = rcu_dereference_raw(mm->exe_file);
 917
 918         if (new_exe_file)
 919                 get_file(new_exe_file);
 920         rcu_assign_pointer(mm->exe_file, new_exe_file);
 921         if (old_exe_file)
 922                 fput(old_exe_file);
 923 }
 924
 925 /**
 926  * get_mm_exe_file - acquire a reference to the mm's executable file
 927  *
 928  * Returns %NULL if mm has no associated executable file.
 929  * User must release file via fput().
 930  */
 931 struct file *get_mm_exe_file(struct mm_struct *mm)
 932 {
 933         struct file *exe_file;
 934
 935         rcu_read_lock();
 936         exe_file = rcu_dereference(mm->exe_file);
 937         if (exe_file && !get_file_rcu(exe_file))
 938                 exe_file = NULL;
 939         rcu_read_unlock();
 940         return exe_file;
 941 }
 942 EXPORT_SYMBOL(get_mm_exe_file);
 943
 944 /**
 945  * get_task_exe_file - acquire a reference to the task's executable file
 946  *
 947  * Returns %NULL if task's mm (if any) has no associated executable file or
 948  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
 949  * User must release file via fput().
 950  */
 951 struct file *get_task_exe_file(struct task_struct *task)
 952 {
 953         struct file *exe_file = NULL;
 954         struct mm_struct *mm;
 955
 956         task_lock(task);
 957         mm = task->mm;
 958         if (mm) {
 959                 if (!(task->flags & PF_KTHREAD))
 960                         exe_file = get_mm_exe_file(mm);
 961         }
 962         task_unlock(task);
 963         return exe_file;
 964 }
 965 EXPORT_SYMBOL(get_task_exe_file);
 966
 967 /**
 968  * get_task_mm - acquire a reference to the task's mm
 969  *
 970  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 971  * this kernel workthread has transiently adopted a user mm with use_mm,
 972  * to do its AIO) is not set and if so returns a reference to it, after
 973  * bumping up the use count.  User must release the mm via mmput()
 974  * after use.  Typically used by /proc and ptrace.
 975  */
 976 struct mm_struct *get_task_mm(struct task_struct *task)
 977 {
 978         struct mm_struct *mm;
 979
 980         task_lock(task);
 981         mm = task->mm;
 982         if (mm) {
 983                 if (task->flags & PF_KTHREAD)
 984                         mm = NULL;
 985                 else
 986                         atomic_inc(&mm->mm_users);
 987         }
 988         task_unlock(task);
 989         return mm;
 990 }
 991 EXPORT_SYMBOL_GPL(get_task_mm);
 992
 993 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 994 {
 995         struct mm_struct *mm;
 996         int err;
 997
 998         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 999         if (err)
1000                 return ERR_PTR(err);
1001
1002         mm = get_task_mm(task);
1003         if (mm && mm != current->mm &&
1004                         !ptrace_may_access(task, mode)) {
1005                 mmput(mm);
1006                 mm = ERR_PTR(-EACCES);
1007         }
1008         mutex_unlock(&task->signal->cred_guard_mutex);
1009
1010         return mm;
1011 }
1012
1013 static void complete_vfork_done(struct task_struct *tsk)
1014 {
1015         struct completion *vfork;
1016
1017         task_lock(tsk);
1018         vfork = tsk->vfork_done;
1019         if (likely(vfork)) {
1020                 tsk->vfork_done = NULL;
1021                 complete(vfork);
1022         }
1023         task_unlock(tsk);
1024 }
1025
1026 static int wait_for_vfork_done(struct task_struct *child,
1027                                 struct completion *vfork)
1028 {
1029         int killed;
1030
1031         freezer_do_not_count();
1032         killed = wait_for_completion_killable(vfork);
1033         freezer_count();
1034
1035         if (killed) {
1036                 task_lock(child);
1037                 child->vfork_done = NULL;
1038                 task_unlock(child);
1039         }
1040
1041         put_task_struct(child);
1042         return killed;
1043 }
1044
1045 /* Please note the differences between mmput and mm_release.
1046  * mmput is called whenever we stop holding onto a mm_struct,
1047  * error success whatever.
1048  *
1049  * mm_release is called after a mm_struct has been removed
1050  * from the current process.
1051  *
1052  * This difference is important for error handling, when we
1053  * only half set up a mm_struct for a new process and need to restore
1054  * the old one.  Because we mmput the new mm_struct before
1055  * restoring the old one. . .
1056  * Eric Biederman 10 January 1998
1057  */
1058 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1059 {
1060         /* Get rid of any futexes when releasing the mm */
1061 #ifdef CONFIG_FUTEX
1062         if (unlikely(tsk->robust_list)) {
1063                 exit_robust_list(tsk);
1064                 tsk->robust_list = NULL;
1065         }
1066 #ifdef CONFIG_COMPAT
1067         if (unlikely(tsk->compat_robust_list)) {
1068                 compat_exit_robust_list(tsk);
1069                 tsk->compat_robust_list = NULL;
1070         }
1071 #endif
1072         if (unlikely(!list_empty(&tsk->pi_state_list)))
1073                 exit_pi_state_list(tsk);
1074 #endif
1075
1076         uprobe_free_utask(tsk);
1077
1078         /* Get rid of any cached register state */
1079         deactivate_mm(tsk, mm);
1080
1081         /*
1082          * Signal userspace if we're not exiting with a core dump
1083          * because we want to leave the value intact for debugging
1084          * purposes.
1085          */
1086         if (tsk->clear_child_tid) {
1087                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1088                     atomic_read(&mm->mm_users) > 1) {
1089                         /*
1090                          * We don't check the error code - if userspace has
1091                          * not set up a proper pointer then tough luck.
1092                          */
1093                         put_user(0, tsk->clear_child_tid);
1094                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1095                                         1, NULL, NULL, 0);
1096                 }
1097                 tsk->clear_child_tid = NULL;
1098         }
1099
1100         /*
1101          * All done, finally we can wake up parent and return this mm to him.
1102          * Also kthread_stop() uses this completion for synchronization.
1103          */
1104         if (tsk->vfork_done)
1105                 complete_vfork_done(tsk);
1106 }
1107
1108 /*
1109  * Allocate a new mm structure and copy contents from the
1110  * mm structure of the passed in task structure.
1111  */
1112 static struct mm_struct *dup_mm(struct task_struct *tsk)
1113 {
1114         struct mm_struct *mm, *oldmm = current->mm;
1115         int err;
1116
1117         mm = allocate_mm();
1118         if (!mm)
1119                 goto fail_nomem;
1120
1121         memcpy(mm, oldmm, sizeof(*mm));
1122
1123         if (!mm_init(mm, tsk))
1124                 goto fail_nomem;
1125
1126         err = dup_mmap(mm, oldmm);
1127         if (err)
1128                 goto free_pt;
1129
1130         mm->hiwater_rss = get_mm_rss(mm);
1131         mm->hiwater_vm = mm->total_vm;
1132
1133         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1134                 goto free_pt;
1135
1136         return mm;
1137
1138 free_pt:
1139         /* don't put binfmt in mmput, we haven't got module yet */
1140         mm->binfmt = NULL;
1141         mmput(mm);
1142
1143 fail_nomem:
1144         return NULL;
1145 }
1146
1147 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1148 {
1149         struct mm_struct *mm, *oldmm;
1150         int retval;
1151
1152         tsk->min_flt = tsk->maj_flt = 0;
1153         tsk->nvcsw = tsk->nivcsw = 0;
1154 #ifdef CONFIG_DETECT_HUNG_TASK
1155         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1156 #endif
1157
1158         tsk->mm = NULL;
1159         tsk->active_mm = NULL;
1160
1161         /*
1162          * Are we cloning a kernel thread?
1163          *
1164          * We need to steal a active VM for that..
1165          */
1166         oldmm = current->mm;
1167         if (!oldmm)
1168                 return 0;
1169
1170         /* initialize the new vmacache entries */
1171         vmacache_flush(tsk);
1172
1173         if (clone_flags & CLONE_VM) {
1174                 atomic_inc(&oldmm->mm_users);
1175                 mm = oldmm;
1176                 goto good_mm;
1177         }
1178
1179         retval = -ENOMEM;
1180         mm = dup_mm(tsk);
1181         if (!mm)
1182                 goto fail_nomem;
1183
1184 good_mm:
1185         tsk->mm = mm;
1186         tsk->active_mm = mm;
1187         return 0;
1188
1189 fail_nomem:
1190         return retval;
1191 }
1192
1193 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1194 {
1195         struct fs_struct *fs = current->fs;
1196         if (clone_flags & CLONE_FS) {
1197                 /* tsk->fs is already what we want */
1198                 spin_lock(&fs->lock);
1199                 if (fs->in_exec) {
1200                         spin_unlock(&fs->lock);
1201                         return -EAGAIN;
1202                 }
1203                 fs->users++;
1204                 spin_unlock(&fs->lock);
1205                 return 0;
1206         }
1207         tsk->fs = copy_fs_struct(fs);
1208         if (!tsk->fs)
1209                 return -ENOMEM;
1210         return 0;
1211 }
1212
1213 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1214 {
1215         struct files_struct *oldf, *newf;
1216         int error = 0;
1217
1218         /*
1219          * A background process may not have any files ...
1220          */
1221         oldf = current->files;
1222         if (!oldf)
1223                 goto out;
1224
1225         if (clone_flags & CLONE_FILES) {
1226                 atomic_inc(&oldf->count);
1227                 goto out;
1228         }
1229
1230         newf = dup_fd(oldf, &error);
1231         if (!newf)
1232                 goto out;
1233
1234         tsk->files = newf;
1235         error = 0;
1236 out:
1237         return error;
1238 }
1239
1240 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1241 {
1242 #ifdef CONFIG_BLOCK
1243         struct io_context *ioc = current->io_context;
1244         struct io_context *new_ioc;
1245
1246         if (!ioc)
1247                 return 0;
1248         /*
1249          * Share io context with parent, if CLONE_IO is set
1250          */
1251         if (clone_flags & CLONE_IO) {
1252                 ioc_task_link(ioc);
1253                 tsk->io_context = ioc;
1254         } else if (ioprio_valid(ioc->ioprio)) {
1255                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1256                 if (unlikely(!new_ioc))
1257                         return -ENOMEM;
1258
1259                 new_ioc->ioprio = ioc->ioprio;
1260                 put_io_context(new_ioc);
1261         }
1262 #endif
1263         return 0;
1264 }
1265
1266 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1267 {
1268         struct sighand_struct *sig;
1269
1270         if (clone_flags & CLONE_SIGHAND) {
1271                 atomic_inc(&current->sighand->count);
1272                 return 0;
1273         }
1274         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1275         rcu_assign_pointer(tsk->sighand, sig);
1276         if (!sig)
1277                 return -ENOMEM;
1278
1279         atomic_set(&sig->count, 1);
1280         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1281         return 0;
1282 }
1283
1284 void __cleanup_sighand(struct sighand_struct *sighand)
1285 {
1286         if (atomic_dec_and_test(&sighand->count)) {
1287                 signalfd_cleanup(sighand);
1288                 /*
1289                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1290                  * without an RCU grace period, see __lock_task_sighand().
1291                  */
1292                 kmem_cache_free(sighand_cachep, sighand);
1293         }
1294 }
1295
1296 /*
1297  * Initialize POSIX timer handling for a thread group.
1298  */
1299 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1300 {
1301         unsigned long cpu_limit;
1302
1303         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1304         if (cpu_limit != RLIM_INFINITY) {
1305                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1306                 sig->cputimer.running = true;
1307         }
1308
1309         /* The timer lists. */
1310         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1311         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1312         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1313 }
1314
1315 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1316 {
1317         struct signal_struct *sig;
1318
1319         if (clone_flags & CLONE_THREAD)
1320                 return 0;
1321
1322         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1323         tsk->signal = sig;
1324         if (!sig)
1325                 return -ENOMEM;
1326
1327         sig->nr_threads = 1;
1328         atomic_set(&sig->live, 1);
1329         atomic_set(&sig->sigcnt, 1);
1330
1331         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1332         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1333         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1334
1335         init_waitqueue_head(&sig->wait_chldexit);
1336         sig->curr_target = tsk;
1337         init_sigpending(&sig->shared_pending);
1338         INIT_LIST_HEAD(&sig->posix_timers);
1339         seqlock_init(&sig->stats_lock);
1340         prev_cputime_init(&sig->prev_cputime);
1341
1342         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1343         sig->real_timer.function = it_real_fn;
1344
1345         task_lock(current->group_leader);
1346         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1347         task_unlock(current->group_leader);
1348
1349         posix_cpu_timers_init_group(sig);
1350
1351         tty_audit_fork(sig);
1352         sched_autogroup_fork(sig);
1353
1354         sig->oom_score_adj = current->signal->oom_score_adj;
1355         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1356
1357         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1358                                    current->signal->is_child_subreaper;
1359
1360         mutex_init(&sig->cred_guard_mutex);
1361
1362         return 0;
1363 }
1364
1365 static void copy_seccomp(struct task_struct *p)
1366 {
1367 #ifdef CONFIG_SECCOMP
1368         /*
1369          * Must be called with sighand->lock held, which is common to
1370          * all threads in the group. Holding cred_guard_mutex is not
1371          * needed because this new task is not yet running and cannot
1372          * be racing exec.
1373          */
1374         assert_spin_locked(&current->sighand->siglock);
1375
1376         /* Ref-count the new filter user, and assign it. */
1377         get_seccomp_filter(current);
1378         p->seccomp = current->seccomp;
1379
1380         /*
1381          * Explicitly enable no_new_privs here in case it got set
1382          * between the task_struct being duplicated and holding the
1383          * sighand lock. The seccomp state and nnp must be in sync.
1384          */
1385         if (task_no_new_privs(current))
1386                 task_set_no_new_privs(p);
1387
1388         /*
1389          * If the parent gained a seccomp mode after copying thread
1390          * flags and between before we held the sighand lock, we have
1391          * to manually enable the seccomp thread flag here.
1392          */
1393         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1394                 set_tsk_thread_flag(p, TIF_SECCOMP);
1395 #endif
1396 }
1397
1398 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1399 {
1400         current->clear_child_tid = tidptr;
1401
1402         return task_pid_vnr(current);
1403 }
1404
1405 static void rt_mutex_init_task(struct task_struct *p)
1406 {
1407         raw_spin_lock_init(&p->pi_lock);
1408 #ifdef CONFIG_RT_MUTEXES
1409         p->pi_waiters = RB_ROOT;
1410         p->pi_waiters_leftmost = NULL;
1411         p->pi_blocked_on = NULL;
1412 #endif
1413 }
1414
1415 /*
1416  * Initialize POSIX timer handling for a single task.
1417  */
1418 static void posix_cpu_timers_init(struct task_struct *tsk)
1419 {
1420         tsk->cputime_expires.prof_exp = 0;
1421         tsk->cputime_expires.virt_exp = 0;
1422         tsk->cputime_expires.sched_exp = 0;
1423         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1424         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1425         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1426 }
1427
1428 static inline void
1429 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1430 {
1431          task->pids[type].pid = pid;
1432 }
1433
1434 /*
1435  * This creates a new process as a copy of the old one,
1436  * but does not actually start it yet.
1437  *
1438  * It copies the registers, and all the appropriate
1439  * parts of the process environment (as per the clone
1440  * flags). The actual kick-off is left to the caller.
1441  */
1442 static struct task_struct *copy_process(unsigned long clone_flags,
1443                                         unsigned long stack_start,
1444                                         unsigned long stack_size,
1445                                         int __user *child_tidptr,
1446                                         struct pid *pid,
1447                                         int trace,
1448                                         unsigned long tls,
1449                                         int node)
1450 {
1451         int retval;
1452         struct task_struct *p;
1453
1454         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1455                 return ERR_PTR(-EINVAL);
1456
1457         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1458                 return ERR_PTR(-EINVAL);
1459
1460         /*
1461          * Thread groups must share signals as well, and detached threads
1462          * can only be started up within the thread group.
1463          */
1464         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1465                 return ERR_PTR(-EINVAL);
1466
1467         /*
1468          * Shared signal handlers imply shared VM. By way of the above,
1469          * thread groups also imply shared VM. Blocking this case allows
1470          * for various simplifications in other code.
1471          */
1472         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1473                 return ERR_PTR(-EINVAL);
1474
1475         /*
1476          * Siblings of global init remain as zombies on exit since they are
1477          * not reaped by their parent (swapper). To solve this and to avoid
1478          * multi-rooted process trees, prevent global and container-inits
1479          * from creating siblings.
1480          */
1481         if ((clone_flags & CLONE_PARENT) &&
1482                                 current->signal->flags & SIGNAL_UNKILLABLE)
1483                 return ERR_PTR(-EINVAL);
1484
1485         /*
1486          * If the new process will be in a different pid or user namespace
1487          * do not allow it to share a thread group with the forking task.
1488          */
1489         if (clone_flags & CLONE_THREAD) {
1490                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1491                     (task_active_pid_ns(current) !=
1492                                 current->nsproxy->pid_ns_for_children))
1493                         return ERR_PTR(-EINVAL);
1494         }
1495
1496         retval = security_task_create(clone_flags);
1497         if (retval)
1498                 goto fork_out;
1499
1500         retval = -ENOMEM;
1501         p = dup_task_struct(current, node);
1502         if (!p)
1503                 goto fork_out;
1504
1505         ftrace_graph_init_task(p);
1506
1507         rt_mutex_init_task(p);
1508
1509 #ifdef CONFIG_PROVE_LOCKING
1510         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1511         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1512 #endif
1513         retval = -EAGAIN;
1514         if (atomic_read(&p->real_cred->user->processes) >=
1515                         task_rlimit(p, RLIMIT_NPROC)) {
1516                 if (p->real_cred->user != INIT_USER &&
1517                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1518                         goto bad_fork_free;
1519         }
1520         current->flags &= ~PF_NPROC_EXCEEDED;
1521
1522         retval = copy_creds(p, clone_flags);
1523         if (retval < 0)
1524                 goto bad_fork_free;
1525
1526         /*
1527          * If multiple threads are within copy_process(), then this check
1528          * triggers too late. This doesn't hurt, the check is only there
1529          * to stop root fork bombs.
1530          */
1531         retval = -EAGAIN;
1532         if (nr_threads >= max_threads)
1533                 goto bad_fork_cleanup_count;
1534
1535         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1536         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1537         p->flags |= PF_FORKNOEXEC;
1538         INIT_LIST_HEAD(&p->children);
1539         INIT_LIST_HEAD(&p->sibling);
1540         rcu_copy_process(p);
1541         p->vfork_done = NULL;
1542         spin_lock_init(&p->alloc_lock);
1543
1544         init_sigpending(&p->pending);
1545
1546         p->utime = p->stime = p->gtime = 0;
1547         p->utimescaled = p->stimescaled = 0;
1548         prev_cputime_init(&p->prev_cputime);
1549
1550 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1551         seqcount_init(&p->vtime_seqcount);
1552         p->vtime_snap = 0;
1553         p->vtime_snap_whence = VTIME_INACTIVE;
1554 #endif
1555
1556 #if defined(SPLIT_RSS_COUNTING)
1557         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1558 #endif
1559
1560         p->default_timer_slack_ns = current->timer_slack_ns;
1561
1562         task_io_accounting_init(&p->ioac);
1563         acct_clear_integrals(p);
1564
1565         posix_cpu_timers_init(p);
1566
1567         p->start_time = ktime_get_ns();
1568         p->real_start_time = ktime_get_boot_ns();
1569         p->io_context = NULL;
1570         p->audit_context = NULL;
1571         cgroup_fork(p);
1572 #ifdef CONFIG_NUMA
1573         p->mempolicy = mpol_dup(p->mempolicy);
1574         if (IS_ERR(p->mempolicy)) {
1575                 retval = PTR_ERR(p->mempolicy);
1576                 p->mempolicy = NULL;
1577                 goto bad_fork_cleanup_threadgroup_lock;
1578         }
1579 #endif
1580 #ifdef CONFIG_CPUSETS
1581         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1582         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1583         seqcount_init(&p->mems_allowed_seq);
1584 #endif
1585 #ifdef CONFIG_TRACE_IRQFLAGS
1586         p->irq_events = 0;
1587         p->hardirqs_enabled = 0;
1588         p->hardirq_enable_ip = 0;
1589         p->hardirq_enable_event = 0;
1590         p->hardirq_disable_ip = _THIS_IP_;
1591         p->hardirq_disable_event = 0;
1592         p->softirqs_enabled = 1;
1593         p->softirq_enable_ip = _THIS_IP_;
1594         p->softirq_enable_event = 0;
1595         p->softirq_disable_ip = 0;
1596         p->softirq_disable_event = 0;
1597         p->hardirq_context = 0;
1598         p->softirq_context = 0;
1599 #endif
1600
1601         p->pagefault_disabled = 0;
1602
1603 #ifdef CONFIG_LOCKDEP
1604         p->lockdep_depth = 0; /* no locks held yet */
1605         p->curr_chain_key = 0;
1606         p->lockdep_recursion = 0;
1607 #endif
1608
1609 #ifdef CONFIG_DEBUG_MUTEXES
1610         p->blocked_on = NULL; /* not blocked yet */
1611 #endif
1612 #ifdef CONFIG_BCACHE
1613         p->sequential_io        = 0;
1614         p->sequential_io_avg    = 0;
1615 #endif
1616
1617         /* Perform scheduler related setup. Assign this task to a CPU. */
1618         retval = sched_fork(clone_flags, p);
1619         if (retval)
1620                 goto bad_fork_cleanup_policy;
1621
1622         retval = perf_event_init_task(p);
1623         if (retval)
1624                 goto bad_fork_cleanup_policy;
1625         retval = audit_alloc(p);
1626         if (retval)
1627                 goto bad_fork_cleanup_perf;
1628         /* copy all the process information */
1629         shm_init_task(p);
1630         retval = copy_semundo(clone_flags, p);
1631         if (retval)
1632                 goto bad_fork_cleanup_audit;
1633         retval = copy_files(clone_flags, p);
1634         if (retval)
1635                 goto bad_fork_cleanup_semundo;
1636         retval = copy_fs(clone_flags, p);
1637         if (retval)
1638                 goto bad_fork_cleanup_files;
1639         retval = copy_sighand(clone_flags, p);
1640         if (retval)
1641                 goto bad_fork_cleanup_fs;
1642         retval = copy_signal(clone_flags, p);
1643         if (retval)
1644                 goto bad_fork_cleanup_sighand;
1645         retval = copy_mm(clone_flags, p);
1646         if (retval)
1647                 goto bad_fork_cleanup_signal;
1648         retval = copy_namespaces(clone_flags, p);
1649         if (retval)
1650                 goto bad_fork_cleanup_mm;
1651         retval = copy_io(clone_flags, p);
1652         if (retval)
1653                 goto bad_fork_cleanup_namespaces;
1654         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1655         if (retval)
1656                 goto bad_fork_cleanup_io;
1657
1658         if (pid != &init_struct_pid) {
1659                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1660                 if (IS_ERR(pid)) {
1661                         retval = PTR_ERR(pid);
1662                         goto bad_fork_cleanup_thread;
1663                 }
1664         }
1665
1666         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1667         /*
1668          * Clear TID on mm_release()?
1669          */
1670         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1671 #ifdef CONFIG_BLOCK
1672         p->plug = NULL;
1673 #endif
1674 #ifdef CONFIG_FUTEX
1675         p->robust_list = NULL;
1676 #ifdef CONFIG_COMPAT
1677         p->compat_robust_list = NULL;
1678 #endif
1679         INIT_LIST_HEAD(&p->pi_state_list);
1680         p->pi_state_cache = NULL;
1681 #endif
1682         /*
1683          * sigaltstack should be cleared when sharing the same VM
1684          */
1685         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1686                 sas_ss_reset(p);
1687
1688         /*
1689          * Syscall tracing and stepping should be turned off in the
1690          * child regardless of CLONE_PTRACE.
1691          */
1692         user_disable_single_step(p);
1693         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1694 #ifdef TIF_SYSCALL_EMU
1695         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1696 #endif
1697         clear_all_latency_tracing(p);
1698
1699         /* ok, now we should be set up.. */
1700         p->pid = pid_nr(pid);
1701         if (clone_flags & CLONE_THREAD) {
1702                 p->exit_signal = -1;
1703                 p->group_leader = current->group_leader;
1704                 p->tgid = current->tgid;
1705         } else {
1706                 if (clone_flags & CLONE_PARENT)
1707                         p->exit_signal = current->group_leader->exit_signal;
1708                 else
1709                         p->exit_signal = (clone_flags & CSIGNAL);
1710                 p->group_leader = p;
1711                 p->tgid = p->pid;
1712         }
1713
1714         p->nr_dirtied = 0;
1715         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1716         p->dirty_paused_when = 0;
1717
1718         p->pdeath_signal = 0;
1719         INIT_LIST_HEAD(&p->thread_group);
1720         p->task_works = NULL;
1721
1722         threadgroup_change_begin(current);
1723         /*
1724          * Ensure that the cgroup subsystem policies allow the new process to be
1725          * forked. It should be noted the the new process's css_set can be changed
1726          * between here and cgroup_post_fork() if an organisation operation is in
1727          * progress.
1728          */
1729         retval = cgroup_can_fork(p);
1730         if (retval)
1731                 goto bad_fork_free_pid;
1732
1733         /*
1734          * Make it visible to the rest of the system, but dont wake it up yet.
1735          * Need tasklist lock for parent etc handling!
1736          */
1737         write_lock_irq(&tasklist_lock);
1738
1739         /* CLONE_PARENT re-uses the old parent */
1740         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1741                 p->real_parent = current->real_parent;
1742                 p->parent_exec_id = current->parent_exec_id;
1743         } else {
1744                 p->real_parent = current;
1745                 p->parent_exec_id = current->self_exec_id;
1746         }
1747
1748         spin_lock(&current->sighand->siglock);
1749
1750         /*
1751          * Copy seccomp details explicitly here, in case they were changed
1752          * before holding sighand lock.
1753          */
1754         copy_seccomp(p);
1755
1756         /*
1757          * Process group and session signals need to be delivered to just the
1758          * parent before the fork or both the parent and the child after the
1759          * fork. Restart if a signal comes in before we add the new process to
1760          * it's process group.
1761          * A fatal signal pending means that current will exit, so the new
1762          * thread can't slip out of an OOM kill (or normal SIGKILL).
1763         */
1764         recalc_sigpending();
1765         if (signal_pending(current)) {
1766                 spin_unlock(&current->sighand->siglock);
1767                 write_unlock_irq(&tasklist_lock);
1768                 retval = -ERESTARTNOINTR;
1769                 goto bad_fork_cancel_cgroup;
1770         }
1771
1772         if (likely(p->pid)) {
1773                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1774
1775                 init_task_pid(p, PIDTYPE_PID, pid);
1776                 if (thread_group_leader(p)) {
1777                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1778                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1779
1780                         if (is_child_reaper(pid)) {
1781                                 ns_of_pid(pid)->child_reaper = p;
1782                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1783                         }
1784
1785                         p->signal->leader_pid = pid;
1786                         p->signal->tty = tty_kref_get(current->signal->tty);
1787                         list_add_tail(&p->sibling, &p->real_parent->children);
1788                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1789                         attach_pid(p, PIDTYPE_PGID);
1790                         attach_pid(p, PIDTYPE_SID);
1791                         __this_cpu_inc(process_counts);
1792                 } else {
1793                         current->signal->nr_threads++;
1794                         atomic_inc(&current->signal->live);
1795                         atomic_inc(&current->signal->sigcnt);
1796                         list_add_tail_rcu(&p->thread_group,
1797                                           &p->group_leader->thread_group);
1798                         list_add_tail_rcu(&p->thread_node,
1799                                           &p->signal->thread_head);
1800                 }
1801                 attach_pid(p, PIDTYPE_PID);
1802                 nr_threads++;
1803         }
1804
1805         total_forks++;
1806         spin_unlock(&current->sighand->siglock);
1807         syscall_tracepoint_update(p);
1808         write_unlock_irq(&tasklist_lock);
1809
1810         proc_fork_connector(p);
1811         cgroup_post_fork(p);
1812         threadgroup_change_end(current);
1813         perf_event_fork(p);
1814
1815         trace_task_newtask(p, clone_flags);
1816         uprobe_copy_process(p, clone_flags);
1817
1818         return p;
1819
1820 bad_fork_cancel_cgroup:
1821         cgroup_cancel_fork(p);
1822 bad_fork_free_pid:
1823         threadgroup_change_end(current);
1824         if (pid != &init_struct_pid)
1825                 free_pid(pid);
1826 bad_fork_cleanup_thread:
1827         exit_thread(p);
1828 bad_fork_cleanup_io:
1829         if (p->io_context)
1830                 exit_io_context(p);
1831 bad_fork_cleanup_namespaces:
1832         exit_task_namespaces(p);
1833 bad_fork_cleanup_mm:
1834         if (p->mm)
1835                 mmput(p->mm);
1836 bad_fork_cleanup_signal:
1837         if (!(clone_flags & CLONE_THREAD))
1838                 free_signal_struct(p->signal);
1839 bad_fork_cleanup_sighand:
1840         __cleanup_sighand(p->sighand);
1841 bad_fork_cleanup_fs:
1842         exit_fs(p); /* blocking */
1843 bad_fork_cleanup_files:
1844         exit_files(p); /* blocking */
1845 bad_fork_cleanup_semundo:
1846         exit_sem(p);
1847 bad_fork_cleanup_audit:
1848         audit_free(p);
1849 bad_fork_cleanup_perf:
1850         perf_event_free_task(p);
1851 bad_fork_cleanup_policy:
1852 #ifdef CONFIG_NUMA
1853         mpol_put(p->mempolicy);
1854 bad_fork_cleanup_threadgroup_lock:
1855 #endif
1856         delayacct_tsk_free(p);
1857 bad_fork_cleanup_count:
1858         atomic_dec(&p->cred->user->processes);
1859         exit_creds(p);
1860 bad_fork_free:
1861         put_task_stack(p);
1862         free_task(p);
1863 fork_out:
1864         return ERR_PTR(retval);
1865 }
1866
1867 static inline void init_idle_pids(struct pid_link *links)
1868 {
1869         enum pid_type type;
1870
1871         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1872                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1873                 links[type].pid = &init_struct_pid;
1874         }
1875 }
1876
1877 struct task_struct *fork_idle(int cpu)
1878 {
1879         struct task_struct *task;
1880         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1881                             cpu_to_node(cpu));
1882         if (!IS_ERR(task)) {
1883                 init_idle_pids(task->pids);
1884                 init_idle(task, cpu);
1885         }
1886
1887         return task;
1888 }
1889
1890 /*
1891  *  Ok, this is the main fork-routine.
1892  *
1893  * It copies the process, and if successful kick-starts
1894  * it and waits for it to finish using the VM if required.
1895  */
1896 long _do_fork(unsigned long clone_flags,
1897               unsigned long stack_start,
1898               unsigned long stack_size,
1899               int __user *parent_tidptr,
1900               int __user *child_tidptr,
1901               unsigned long tls)
1902 {
1903         struct task_struct *p;
1904         int trace = 0;
1905         long nr;
1906
1907         /*
1908          * Determine whether and which event to report to ptracer.  When
1909          * called from kernel_thread or CLONE_UNTRACED is explicitly
1910          * requested, no event is reported; otherwise, report if the event
1911          * for the type of forking is enabled.
1912          */
1913         if (!(clone_flags & CLONE_UNTRACED)) {
1914                 if (clone_flags & CLONE_VFORK)
1915                         trace = PTRACE_EVENT_VFORK;
1916                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1917                         trace = PTRACE_EVENT_CLONE;
1918                 else
1919                         trace = PTRACE_EVENT_FORK;
1920
1921                 if (likely(!ptrace_event_enabled(current, trace)))
1922                         trace = 0;
1923         }
1924
1925         p = copy_process(clone_flags, stack_start, stack_size,
1926                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
1927         /*
1928          * Do this prior waking up the new thread - the thread pointer
1929          * might get invalid after that point, if the thread exits quickly.
1930          */
1931         if (!IS_ERR(p)) {
1932                 struct completion vfork;
1933                 struct pid *pid;
1934
1935                 trace_sched_process_fork(current, p);
1936
1937                 pid = get_task_pid(p, PIDTYPE_PID);
1938                 nr = pid_vnr(pid);
1939
1940                 if (clone_flags & CLONE_PARENT_SETTID)
1941                         put_user(nr, parent_tidptr);
1942
1943                 if (clone_flags & CLONE_VFORK) {
1944                         p->vfork_done = &vfork;
1945                         init_completion(&vfork);
1946                         get_task_struct(p);
1947                 }
1948
1949                 wake_up_new_task(p);
1950
1951                 /* forking complete and child started to run, tell ptracer */
1952                 if (unlikely(trace))
1953                         ptrace_event_pid(trace, pid);
1954
1955                 if (clone_flags & CLONE_VFORK) {
1956                         if (!wait_for_vfork_done(p, &vfork))
1957                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1958                 }
1959
1960                 put_pid(pid);
1961         } else {
1962                 nr = PTR_ERR(p);
1963         }
1964         return nr;
1965 }
1966
1967 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1968 /* For compatibility with architectures that call do_fork directly rather than
1969  * using the syscall entry points below. */
1970 long do_fork(unsigned long clone_flags,
1971               unsigned long stack_start,
1972               unsigned long stack_size,
1973               int __user *parent_tidptr,
1974               int __user *child_tidptr)
1975 {
1976         return _do_fork(clone_flags, stack_start, stack_size,
1977                         parent_tidptr, child_tidptr, 0);
1978 }
1979 #endif
1980
1981 /*
1982  * Create a kernel thread.
1983  */
1984 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1985 {
1986         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1987                 (unsigned long)arg, NULL, NULL, 0);
1988 }
1989
1990 #ifdef __ARCH_WANT_SYS_FORK
1991 SYSCALL_DEFINE0(fork)
1992 {
1993 #ifdef CONFIG_MMU
1994         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1995 #else
1996         /* can not support in nommu mode */
1997         return -EINVAL;
1998 #endif
1999 }
2000 #endif
2001
2002 #ifdef __ARCH_WANT_SYS_VFORK
2003 SYSCALL_DEFINE0(vfork)
2004 {
2005         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2006                         0, NULL, NULL, 0);
2007 }
2008 #endif
2009
2010 #ifdef __ARCH_WANT_SYS_CLONE
2011 #ifdef CONFIG_CLONE_BACKWARDS
2012 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2013                  int __user *, parent_tidptr,
2014                  unsigned long, tls,
2015                  int __user *, child_tidptr)
2016 #elif defined(CONFIG_CLONE_BACKWARDS2)
2017 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2018                  int __user *, parent_tidptr,
2019                  int __user *, child_tidptr,
2020                  unsigned long, tls)
2021 #elif defined(CONFIG_CLONE_BACKWARDS3)
2022 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2023                 int, stack_size,
2024                 int __user *, parent_tidptr,
2025                 int __user *, child_tidptr,
2026                 unsigned long, tls)
2027 #else
2028 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2029                  int __user *, parent_tidptr,
2030                  int __user *, child_tidptr,
2031                  unsigned long, tls)
2032 #endif
2033 {
2034         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2035 }
2036 #endif
2037
2038 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2039 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2040 #endif
2041
2042 static void sighand_ctor(void *data)
2043 {
2044         struct sighand_struct *sighand = data;
2045
2046         spin_lock_init(&sighand->siglock);
2047         init_waitqueue_head(&sighand->signalfd_wqh);
2048 }
2049
2050 void __init proc_caches_init(void)
2051 {
2052         sighand_cachep = kmem_cache_create("sighand_cache",
2053                         sizeof(struct sighand_struct), 0,
2054                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
2055                         SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2056         signal_cachep = kmem_cache_create("signal_cache",
2057                         sizeof(struct signal_struct), 0,
2058                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2059                         NULL);
2060         files_cachep = kmem_cache_create("files_cache",
2061                         sizeof(struct files_struct), 0,
2062                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2063                         NULL);
2064         fs_cachep = kmem_cache_create("fs_cache",
2065                         sizeof(struct fs_struct), 0,
2066                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2067                         NULL);
2068         /*
2069          * FIXME! The "sizeof(struct mm_struct)" currently includes the
2070          * whole struct cpumask for the OFFSTACK case. We could change
2071          * this to *only* allocate as much of it as required by the
2072          * maximum number of CPU's we can ever have.  The cpumask_allocation
2073          * is at the end of the structure, exactly for that reason.
2074          */
2075         mm_cachep = kmem_cache_create("mm_struct",
2076                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2077                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2078                         NULL);
2079         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2080         mmap_init();
2081         nsproxy_cache_init();
2082 }
2083
2084 /*
2085  * Check constraints on flags passed to the unshare system call.
2086  */
2087 static int check_unshare_flags(unsigned long unshare_flags)
2088 {
2089         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2090                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2091                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2092                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2093                 return -EINVAL;
2094         /*
2095          * Not implemented, but pretend it works if there is nothing
2096          * to unshare.  Note that unsharing the address space or the
2097          * signal handlers also need to unshare the signal queues (aka
2098          * CLONE_THREAD).
2099          */
2100         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2101                 if (!thread_group_empty(current))
2102                         return -EINVAL;
2103         }
2104         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2105                 if (atomic_read(&current->sighand->count) > 1)
2106                         return -EINVAL;
2107         }
2108         if (unshare_flags & CLONE_VM) {
2109                 if (!current_is_single_threaded())
2110                         return -EINVAL;
2111         }
2112
2113         return 0;
2114 }
2115
2116 /*
2117  * Unshare the filesystem structure if it is being shared
2118  */
2119 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2120 {
2121         struct fs_struct *fs = current->fs;
2122
2123         if (!(unshare_flags & CLONE_FS) || !fs)
2124                 return 0;
2125
2126         /* don't need lock here; in the worst case we'll do useless copy */
2127         if (fs->users == 1)
2128                 return 0;
2129
2130         *new_fsp = copy_fs_struct(fs);
2131         if (!*new_fsp)
2132                 return -ENOMEM;
2133
2134         return 0;
2135 }
2136
2137 /*
2138  * Unshare file descriptor table if it is being shared
2139  */
2140 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2141 {
2142         struct files_struct *fd = current->files;
2143         int error = 0;
2144
2145         if ((unshare_flags & CLONE_FILES) &&
2146             (fd && atomic_read(&fd->count) > 1)) {
2147                 *new_fdp = dup_fd(fd, &error);
2148                 if (!*new_fdp)
2149                         return error;
2150         }
2151
2152         return 0;
2153 }
2154
2155 /*
2156  * unshare allows a process to 'unshare' part of the process
2157  * context which was originally shared using clone.  copy_*
2158  * functions used by do_fork() cannot be used here directly
2159  * because they modify an inactive task_struct that is being
2160  * constructed. Here we are modifying the current, active,
2161  * task_struct.
2162  */
2163 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2164 {
2165         struct fs_struct *fs, *new_fs = NULL;
2166         struct files_struct *fd, *new_fd = NULL;
2167         struct cred *new_cred = NULL;
2168         struct nsproxy *new_nsproxy = NULL;
2169         int do_sysvsem = 0;
2170         int err;
2171
2172         /*
2173          * If unsharing a user namespace must also unshare the thread group
2174          * and unshare the filesystem root and working directories.
2175          */
2176         if (unshare_flags & CLONE_NEWUSER)
2177                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2178         /*
2179          * If unsharing vm, must also unshare signal handlers.
2180          */
2181         if (unshare_flags & CLONE_VM)
2182                 unshare_flags |= CLONE_SIGHAND;
2183         /*
2184          * If unsharing a signal handlers, must also unshare the signal queues.
2185          */
2186         if (unshare_flags & CLONE_SIGHAND)
2187                 unshare_flags |= CLONE_THREAD;
2188         /*
2189          * If unsharing namespace, must also unshare filesystem information.
2190          */
2191         if (unshare_flags & CLONE_NEWNS)
2192                 unshare_flags |= CLONE_FS;
2193
2194         err = check_unshare_flags(unshare_flags);
2195         if (err)
2196                 goto bad_unshare_out;
2197         /*
2198          * CLONE_NEWIPC must also detach from the undolist: after switching
2199          * to a new ipc namespace, the semaphore arrays from the old
2200          * namespace are unreachable.
2201          */
2202         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2203                 do_sysvsem = 1;
2204         err = unshare_fs(unshare_flags, &new_fs);
2205         if (err)
2206                 goto bad_unshare_out;
2207         err = unshare_fd(unshare_flags, &new_fd);
2208         if (err)
2209                 goto bad_unshare_cleanup_fs;
2210         err = unshare_userns(unshare_flags, &new_cred);
2211         if (err)
2212                 goto bad_unshare_cleanup_fd;
2213         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2214                                          new_cred, new_fs);
2215         if (err)
2216                 goto bad_unshare_cleanup_cred;
2217
2218         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2219                 if (do_sysvsem) {
2220                         /*
2221                          * CLONE_SYSVSEM is equivalent to sys_exit().
2222                          */
2223                         exit_sem(current);
2224                 }
2225                 if (unshare_flags & CLONE_NEWIPC) {
2226                         /* Orphan segments in old ns (see sem above). */
2227                         exit_shm(current);
2228                         shm_init_task(current);
2229                 }
2230
2231                 if (new_nsproxy)
2232                         switch_task_namespaces(current, new_nsproxy);
2233
2234                 task_lock(current);
2235
2236                 if (new_fs) {
2237                         fs = current->fs;
2238                         spin_lock(&fs->lock);
2239                         current->fs = new_fs;
2240                         if (--fs->users)
2241                                 new_fs = NULL;
2242                         else
2243                                 new_fs = fs;
2244                         spin_unlock(&fs->lock);
2245                 }
2246
2247                 if (new_fd) {
2248                         fd = current->files;
2249                         current->files = new_fd;
2250                         new_fd = fd;
2251                 }
2252
2253                 task_unlock(current);
2254
2255                 if (new_cred) {
2256                         /* Install the new user namespace */
2257                         commit_creds(new_cred);
2258                         new_cred = NULL;
2259                 }
2260         }
2261
2262 bad_unshare_cleanup_cred:
2263         if (new_cred)
2264                 put_cred(new_cred);
2265 bad_unshare_cleanup_fd:
2266         if (new_fd)
2267                 put_files_struct(new_fd);
2268
2269 bad_unshare_cleanup_fs:
2270         if (new_fs)
2271                 free_fs_struct(new_fs);
2272
2273 bad_unshare_out:
2274         return err;
2275 }
2276
2277 /*
2278  *      Helper to unshare the files of the current task.
2279  *      We don't want to expose copy_files internals to
2280  *      the exec layer of the kernel.
2281  */
2282
2283 int unshare_files(struct files_struct **displaced)
2284 {
2285         struct task_struct *task = current;
2286         struct files_struct *copy = NULL;
2287         int error;
2288
2289         error = unshare_fd(CLONE_FILES, &copy);
2290         if (error || !copy) {
2291                 *displaced = NULL;
2292                 return error;
2293         }
2294         *displaced = task->files;
2295         task_lock(task);
2296         task->files = copy;
2297         task_unlock(task);
2298         return 0;
2299 }
2300
2301 int sysctl_max_threads(struct ctl_table *table, int write,
2302                        void __user *buffer, size_t *lenp, loff_t *ppos)
2303 {
2304         struct ctl_table t;
2305         int ret;
2306         int threads = max_threads;
2307         int min = MIN_THREADS;
2308         int max = MAX_THREADS;
2309
2310         t = *table;
2311         t.data = &threads;
2312         t.extra1 = &min;
2313         t.extra2 = &max;
2314
2315         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2316         if (ret || !write)
2317                 return ret;
2318
2319         set_max_threads(threads);
2320
2321         return 0;
2322 }