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         set_bit(MMF_OOM_SKIP, &mm->flags);
 866         mmdrop(mm);
 867 }
 868
 869 /*
 870  * Decrement the use count and release all resources for an mm.
 871  */
 872 void mmput(struct mm_struct *mm)
 873 {
 874         might_sleep();
 875
 876         if (atomic_dec_and_test(&mm->mm_users))
 877                 __mmput(mm);
 878 }
 879 EXPORT_SYMBOL_GPL(mmput);
 880
 881 #ifdef CONFIG_MMU
 882 static void mmput_async_fn(struct work_struct *work)
 883 {
 884         struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
 885         __mmput(mm);
 886 }
 887
 888 void mmput_async(struct mm_struct *mm)
 889 {
 890         if (atomic_dec_and_test(&mm->mm_users)) {
 891                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
 892                 schedule_work(&mm->async_put_work);
 893         }
 894 }
 895 #endif
 896
 897 /**
 898  * set_mm_exe_file - change a reference to the mm's executable file
 899  *
 900  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 901  *
 902  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 903  * invocations: in mmput() nobody alive left, in execve task is single
 904  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 905  * mm->exe_file, but does so without using set_mm_exe_file() in order
 906  * to do avoid the need for any locks.
 907  */
 908 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 909 {
 910         struct file *old_exe_file;
 911
 912         /*
 913          * It is safe to dereference the exe_file without RCU as
 914          * this function is only called if nobody else can access
 915          * this mm -- see comment above for justification.
 916          */
 917         old_exe_file = rcu_dereference_raw(mm->exe_file);
 918
 919         if (new_exe_file)
 920                 get_file(new_exe_file);
 921         rcu_assign_pointer(mm->exe_file, new_exe_file);
 922         if (old_exe_file)
 923                 fput(old_exe_file);
 924 }
 925
 926 /**
 927  * get_mm_exe_file - acquire a reference to the mm's executable file
 928  *
 929  * Returns %NULL if mm has no associated executable file.
 930  * User must release file via fput().
 931  */
 932 struct file *get_mm_exe_file(struct mm_struct *mm)
 933 {
 934         struct file *exe_file;
 935
 936         rcu_read_lock();
 937         exe_file = rcu_dereference(mm->exe_file);
 938         if (exe_file && !get_file_rcu(exe_file))
 939                 exe_file = NULL;
 940         rcu_read_unlock();
 941         return exe_file;
 942 }
 943 EXPORT_SYMBOL(get_mm_exe_file);
 944
 945 /**
 946  * get_task_exe_file - acquire a reference to the task's executable file
 947  *
 948  * Returns %NULL if task's mm (if any) has no associated executable file or
 949  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
 950  * User must release file via fput().
 951  */
 952 struct file *get_task_exe_file(struct task_struct *task)
 953 {
 954         struct file *exe_file = NULL;
 955         struct mm_struct *mm;
 956
 957         task_lock(task);
 958         mm = task->mm;
 959         if (mm) {
 960                 if (!(task->flags & PF_KTHREAD))
 961                         exe_file = get_mm_exe_file(mm);
 962         }
 963         task_unlock(task);
 964         return exe_file;
 965 }
 966 EXPORT_SYMBOL(get_task_exe_file);
 967
 968 /**
 969  * get_task_mm - acquire a reference to the task's mm
 970  *
 971  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 972  * this kernel workthread has transiently adopted a user mm with use_mm,
 973  * to do its AIO) is not set and if so returns a reference to it, after
 974  * bumping up the use count.  User must release the mm via mmput()
 975  * after use.  Typically used by /proc and ptrace.
 976  */
 977 struct mm_struct *get_task_mm(struct task_struct *task)
 978 {
 979         struct mm_struct *mm;
 980
 981         task_lock(task);
 982         mm = task->mm;
 983         if (mm) {
 984                 if (task->flags & PF_KTHREAD)
 985                         mm = NULL;
 986                 else
 987                         atomic_inc(&mm->mm_users);
 988         }
 989         task_unlock(task);
 990         return mm;
 991 }
 992 EXPORT_SYMBOL_GPL(get_task_mm);
 993
 994 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 995 {
 996         struct mm_struct *mm;
 997         int err;
 998
 999         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1000         if (err)
1001                 return ERR_PTR(err);
1002
1003         mm = get_task_mm(task);
1004         if (mm && mm != current->mm &&
1005                         !ptrace_may_access(task, mode)) {
1006                 mmput(mm);
1007                 mm = ERR_PTR(-EACCES);
1008         }
1009         mutex_unlock(&task->signal->cred_guard_mutex);
1010
1011         return mm;
1012 }
1013
1014 static void complete_vfork_done(struct task_struct *tsk)
1015 {
1016         struct completion *vfork;
1017
1018         task_lock(tsk);
1019         vfork = tsk->vfork_done;
1020         if (likely(vfork)) {
1021                 tsk->vfork_done = NULL;
1022                 complete(vfork);
1023         }
1024         task_unlock(tsk);
1025 }
1026
1027 static int wait_for_vfork_done(struct task_struct *child,
1028                                 struct completion *vfork)
1029 {
1030         int killed;
1031
1032         freezer_do_not_count();
1033         killed = wait_for_completion_killable(vfork);
1034         freezer_count();
1035
1036         if (killed) {
1037                 task_lock(child);
1038                 child->vfork_done = NULL;
1039                 task_unlock(child);
1040         }
1041
1042         put_task_struct(child);
1043         return killed;
1044 }
1045
1046 /* Please note the differences between mmput and mm_release.
1047  * mmput is called whenever we stop holding onto a mm_struct,
1048  * error success whatever.
1049  *
1050  * mm_release is called after a mm_struct has been removed
1051  * from the current process.
1052  *
1053  * This difference is important for error handling, when we
1054  * only half set up a mm_struct for a new process and need to restore
1055  * the old one.  Because we mmput the new mm_struct before
1056  * restoring the old one. . .
1057  * Eric Biederman 10 January 1998
1058  */
1059 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1060 {
1061         /* Get rid of any futexes when releasing the mm */
1062 #ifdef CONFIG_FUTEX
1063         if (unlikely(tsk->robust_list)) {
1064                 exit_robust_list(tsk);
1065                 tsk->robust_list = NULL;
1066         }
1067 #ifdef CONFIG_COMPAT
1068         if (unlikely(tsk->compat_robust_list)) {
1069                 compat_exit_robust_list(tsk);
1070                 tsk->compat_robust_list = NULL;
1071         }
1072 #endif
1073         if (unlikely(!list_empty(&tsk->pi_state_list)))
1074                 exit_pi_state_list(tsk);
1075 #endif
1076
1077         uprobe_free_utask(tsk);
1078
1079         /* Get rid of any cached register state */
1080         deactivate_mm(tsk, mm);
1081
1082         /*
1083          * Signal userspace if we're not exiting with a core dump
1084          * because we want to leave the value intact for debugging
1085          * purposes.
1086          */
1087         if (tsk->clear_child_tid) {
1088                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1089                     atomic_read(&mm->mm_users) > 1) {
1090                         /*
1091                          * We don't check the error code - if userspace has
1092                          * not set up a proper pointer then tough luck.
1093                          */
1094                         put_user(0, tsk->clear_child_tid);
1095                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1096                                         1, NULL, NULL, 0);
1097                 }
1098                 tsk->clear_child_tid = NULL;
1099         }
1100
1101         /*
1102          * All done, finally we can wake up parent and return this mm to him.
1103          * Also kthread_stop() uses this completion for synchronization.
1104          */
1105         if (tsk->vfork_done)
1106                 complete_vfork_done(tsk);
1107 }
1108
1109 /*
1110  * Allocate a new mm structure and copy contents from the
1111  * mm structure of the passed in task structure.
1112  */
1113 static struct mm_struct *dup_mm(struct task_struct *tsk)
1114 {
1115         struct mm_struct *mm, *oldmm = current->mm;
1116         int err;
1117
1118         mm = allocate_mm();
1119         if (!mm)
1120                 goto fail_nomem;
1121
1122         memcpy(mm, oldmm, sizeof(*mm));
1123
1124         if (!mm_init(mm, tsk))
1125                 goto fail_nomem;
1126
1127         err = dup_mmap(mm, oldmm);
1128         if (err)
1129                 goto free_pt;
1130
1131         mm->hiwater_rss = get_mm_rss(mm);
1132         mm->hiwater_vm = mm->total_vm;
1133
1134         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1135                 goto free_pt;
1136
1137         return mm;
1138
1139 free_pt:
1140         /* don't put binfmt in mmput, we haven't got module yet */
1141         mm->binfmt = NULL;
1142         mmput(mm);
1143
1144 fail_nomem:
1145         return NULL;
1146 }
1147
1148 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1149 {
1150         struct mm_struct *mm, *oldmm;
1151         int retval;
1152
1153         tsk->min_flt = tsk->maj_flt = 0;
1154         tsk->nvcsw = tsk->nivcsw = 0;
1155 #ifdef CONFIG_DETECT_HUNG_TASK
1156         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1157 #endif
1158
1159         tsk->mm = NULL;
1160         tsk->active_mm = NULL;
1161
1162         /*
1163          * Are we cloning a kernel thread?
1164          *
1165          * We need to steal a active VM for that..
1166          */
1167         oldmm = current->mm;
1168         if (!oldmm)
1169                 return 0;
1170
1171         /* initialize the new vmacache entries */
1172         vmacache_flush(tsk);
1173
1174         if (clone_flags & CLONE_VM) {
1175                 atomic_inc(&oldmm->mm_users);
1176                 mm = oldmm;
1177                 goto good_mm;
1178         }
1179
1180         retval = -ENOMEM;
1181         mm = dup_mm(tsk);
1182         if (!mm)
1183                 goto fail_nomem;
1184
1185 good_mm:
1186         tsk->mm = mm;
1187         tsk->active_mm = mm;
1188         return 0;
1189
1190 fail_nomem:
1191         return retval;
1192 }
1193
1194 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1195 {
1196         struct fs_struct *fs = current->fs;
1197         if (clone_flags & CLONE_FS) {
1198                 /* tsk->fs is already what we want */
1199                 spin_lock(&fs->lock);
1200                 if (fs->in_exec) {
1201                         spin_unlock(&fs->lock);
1202                         return -EAGAIN;
1203                 }
1204                 fs->users++;
1205                 spin_unlock(&fs->lock);
1206                 return 0;
1207         }
1208         tsk->fs = copy_fs_struct(fs);
1209         if (!tsk->fs)
1210                 return -ENOMEM;
1211         return 0;
1212 }
1213
1214 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1215 {
1216         struct files_struct *oldf, *newf;
1217         int error = 0;
1218
1219         /*
1220          * A background process may not have any files ...
1221          */
1222         oldf = current->files;
1223         if (!oldf)
1224                 goto out;
1225
1226         if (clone_flags & CLONE_FILES) {
1227                 atomic_inc(&oldf->count);
1228                 goto out;
1229         }
1230
1231         newf = dup_fd(oldf, &error);
1232         if (!newf)
1233                 goto out;
1234
1235         tsk->files = newf;
1236         error = 0;
1237 out:
1238         return error;
1239 }
1240
1241 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1242 {
1243 #ifdef CONFIG_BLOCK
1244         struct io_context *ioc = current->io_context;
1245         struct io_context *new_ioc;
1246
1247         if (!ioc)
1248                 return 0;
1249         /*
1250          * Share io context with parent, if CLONE_IO is set
1251          */
1252         if (clone_flags & CLONE_IO) {
1253                 ioc_task_link(ioc);
1254                 tsk->io_context = ioc;
1255         } else if (ioprio_valid(ioc->ioprio)) {
1256                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1257                 if (unlikely(!new_ioc))
1258                         return -ENOMEM;
1259
1260                 new_ioc->ioprio = ioc->ioprio;
1261                 put_io_context(new_ioc);
1262         }
1263 #endif
1264         return 0;
1265 }
1266
1267 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1268 {
1269         struct sighand_struct *sig;
1270
1271         if (clone_flags & CLONE_SIGHAND) {
1272                 atomic_inc(&current->sighand->count);
1273                 return 0;
1274         }
1275         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1276         rcu_assign_pointer(tsk->sighand, sig);
1277         if (!sig)
1278                 return -ENOMEM;
1279
1280         atomic_set(&sig->count, 1);
1281         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1282         return 0;
1283 }
1284
1285 void __cleanup_sighand(struct sighand_struct *sighand)
1286 {
1287         if (atomic_dec_and_test(&sighand->count)) {
1288                 signalfd_cleanup(sighand);
1289                 /*
1290                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1291                  * without an RCU grace period, see __lock_task_sighand().
1292                  */
1293                 kmem_cache_free(sighand_cachep, sighand);
1294         }
1295 }
1296
1297 /*
1298  * Initialize POSIX timer handling for a thread group.
1299  */
1300 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1301 {
1302         unsigned long cpu_limit;
1303
1304         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1305         if (cpu_limit != RLIM_INFINITY) {
1306                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1307                 sig->cputimer.running = true;
1308         }
1309
1310         /* The timer lists. */
1311         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1312         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1313         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1314 }
1315
1316 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1317 {
1318         struct signal_struct *sig;
1319
1320         if (clone_flags & CLONE_THREAD)
1321                 return 0;
1322
1323         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1324         tsk->signal = sig;
1325         if (!sig)
1326                 return -ENOMEM;
1327
1328         sig->nr_threads = 1;
1329         atomic_set(&sig->live, 1);
1330         atomic_set(&sig->sigcnt, 1);
1331
1332         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1333         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1334         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1335
1336         init_waitqueue_head(&sig->wait_chldexit);
1337         sig->curr_target = tsk;
1338         init_sigpending(&sig->shared_pending);
1339         INIT_LIST_HEAD(&sig->posix_timers);
1340         seqlock_init(&sig->stats_lock);
1341         prev_cputime_init(&sig->prev_cputime);
1342
1343         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1344         sig->real_timer.function = it_real_fn;
1345
1346         task_lock(current->group_leader);
1347         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1348         task_unlock(current->group_leader);
1349
1350         posix_cpu_timers_init_group(sig);
1351
1352         tty_audit_fork(sig);
1353         sched_autogroup_fork(sig);
1354
1355         sig->oom_score_adj = current->signal->oom_score_adj;
1356         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1357
1358         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1359                                    current->signal->is_child_subreaper;
1360
1361         mutex_init(&sig->cred_guard_mutex);
1362
1363         return 0;
1364 }
1365
1366 static void copy_seccomp(struct task_struct *p)
1367 {
1368 #ifdef CONFIG_SECCOMP
1369         /*
1370          * Must be called with sighand->lock held, which is common to
1371          * all threads in the group. Holding cred_guard_mutex is not
1372          * needed because this new task is not yet running and cannot
1373          * be racing exec.
1374          */
1375         assert_spin_locked(&current->sighand->siglock);
1376
1377         /* Ref-count the new filter user, and assign it. */
1378         get_seccomp_filter(current);
1379         p->seccomp = current->seccomp;
1380
1381         /*
1382          * Explicitly enable no_new_privs here in case it got set
1383          * between the task_struct being duplicated and holding the
1384          * sighand lock. The seccomp state and nnp must be in sync.
1385          */
1386         if (task_no_new_privs(current))
1387                 task_set_no_new_privs(p);
1388
1389         /*
1390          * If the parent gained a seccomp mode after copying thread
1391          * flags and between before we held the sighand lock, we have
1392          * to manually enable the seccomp thread flag here.
1393          */
1394         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1395                 set_tsk_thread_flag(p, TIF_SECCOMP);
1396 #endif
1397 }
1398
1399 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1400 {
1401         current->clear_child_tid = tidptr;
1402
1403         return task_pid_vnr(current);
1404 }
1405
1406 static void rt_mutex_init_task(struct task_struct *p)
1407 {
1408         raw_spin_lock_init(&p->pi_lock);
1409 #ifdef CONFIG_RT_MUTEXES
1410         p->pi_waiters = RB_ROOT;
1411         p->pi_waiters_leftmost = NULL;
1412         p->pi_blocked_on = NULL;
1413 #endif
1414 }
1415
1416 /*
1417  * Initialize POSIX timer handling for a single task.
1418  */
1419 static void posix_cpu_timers_init(struct task_struct *tsk)
1420 {
1421         tsk->cputime_expires.prof_exp = 0;
1422         tsk->cputime_expires.virt_exp = 0;
1423         tsk->cputime_expires.sched_exp = 0;
1424         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1425         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1426         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1427 }
1428
1429 static inline void
1430 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1431 {
1432          task->pids[type].pid = pid;
1433 }
1434
1435 /*
1436  * This creates a new process as a copy of the old one,
1437  * but does not actually start it yet.
1438  *
1439  * It copies the registers, and all the appropriate
1440  * parts of the process environment (as per the clone
1441  * flags). The actual kick-off is left to the caller.
1442  */
1443 static struct task_struct *copy_process(unsigned long clone_flags,
1444                                         unsigned long stack_start,
1445                                         unsigned long stack_size,
1446                                         int __user *child_tidptr,
1447                                         struct pid *pid,
1448                                         int trace,
1449                                         unsigned long tls,
1450                                         int node)
1451 {
1452         int retval;
1453         struct task_struct *p;
1454
1455         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1456                 return ERR_PTR(-EINVAL);
1457
1458         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1459                 return ERR_PTR(-EINVAL);
1460
1461         /*
1462          * Thread groups must share signals as well, and detached threads
1463          * can only be started up within the thread group.
1464          */
1465         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1466                 return ERR_PTR(-EINVAL);
1467
1468         /*
1469          * Shared signal handlers imply shared VM. By way of the above,
1470          * thread groups also imply shared VM. Blocking this case allows
1471          * for various simplifications in other code.
1472          */
1473         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1474                 return ERR_PTR(-EINVAL);
1475
1476         /*
1477          * Siblings of global init remain as zombies on exit since they are
1478          * not reaped by their parent (swapper). To solve this and to avoid
1479          * multi-rooted process trees, prevent global and container-inits
1480          * from creating siblings.
1481          */
1482         if ((clone_flags & CLONE_PARENT) &&
1483                                 current->signal->flags & SIGNAL_UNKILLABLE)
1484                 return ERR_PTR(-EINVAL);
1485
1486         /*
1487          * If the new process will be in a different pid or user namespace
1488          * do not allow it to share a thread group with the forking task.
1489          */
1490         if (clone_flags & CLONE_THREAD) {
1491                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1492                     (task_active_pid_ns(current) !=
1493                                 current->nsproxy->pid_ns_for_children))
1494                         return ERR_PTR(-EINVAL);
1495         }
1496
1497         retval = security_task_create(clone_flags);
1498         if (retval)
1499                 goto fork_out;
1500
1501         retval = -ENOMEM;
1502         p = dup_task_struct(current, node);
1503         if (!p)
1504                 goto fork_out;
1505
1506         ftrace_graph_init_task(p);
1507
1508         rt_mutex_init_task(p);
1509
1510 #ifdef CONFIG_PROVE_LOCKING
1511         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1512         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1513 #endif
1514         retval = -EAGAIN;
1515         if (atomic_read(&p->real_cred->user->processes) >=
1516                         task_rlimit(p, RLIMIT_NPROC)) {
1517                 if (p->real_cred->user != INIT_USER &&
1518                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1519                         goto bad_fork_free;
1520         }
1521         current->flags &= ~PF_NPROC_EXCEEDED;
1522
1523         retval = copy_creds(p, clone_flags);
1524         if (retval < 0)
1525                 goto bad_fork_free;
1526
1527         /*
1528          * If multiple threads are within copy_process(), then this check
1529          * triggers too late. This doesn't hurt, the check is only there
1530          * to stop root fork bombs.
1531          */
1532         retval = -EAGAIN;
1533         if (nr_threads >= max_threads)
1534                 goto bad_fork_cleanup_count;
1535
1536         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1537         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1538         p->flags |= PF_FORKNOEXEC;
1539         INIT_LIST_HEAD(&p->children);
1540         INIT_LIST_HEAD(&p->sibling);
1541         rcu_copy_process(p);
1542         p->vfork_done = NULL;
1543         spin_lock_init(&p->alloc_lock);
1544
1545         init_sigpending(&p->pending);
1546
1547         p->utime = p->stime = p->gtime = 0;
1548         p->utimescaled = p->stimescaled = 0;
1549         prev_cputime_init(&p->prev_cputime);
1550
1551 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1552         seqcount_init(&p->vtime_seqcount);
1553         p->vtime_snap = 0;
1554         p->vtime_snap_whence = VTIME_INACTIVE;
1555 #endif
1556
1557 #if defined(SPLIT_RSS_COUNTING)
1558         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1559 #endif
1560
1561         p->default_timer_slack_ns = current->timer_slack_ns;
1562
1563         task_io_accounting_init(&p->ioac);
1564         acct_clear_integrals(p);
1565
1566         posix_cpu_timers_init(p);
1567
1568         p->start_time = ktime_get_ns();
1569         p->real_start_time = ktime_get_boot_ns();
1570         p->io_context = NULL;
1571         p->audit_context = NULL;
1572         cgroup_fork(p);
1573 #ifdef CONFIG_NUMA
1574         p->mempolicy = mpol_dup(p->mempolicy);
1575         if (IS_ERR(p->mempolicy)) {
1576                 retval = PTR_ERR(p->mempolicy);
1577                 p->mempolicy = NULL;
1578                 goto bad_fork_cleanup_threadgroup_lock;
1579         }
1580 #endif
1581 #ifdef CONFIG_CPUSETS
1582         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1583         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1584         seqcount_init(&p->mems_allowed_seq);
1585 #endif
1586 #ifdef CONFIG_TRACE_IRQFLAGS
1587         p->irq_events = 0;
1588         p->hardirqs_enabled = 0;
1589         p->hardirq_enable_ip = 0;
1590         p->hardirq_enable_event = 0;
1591         p->hardirq_disable_ip = _THIS_IP_;
1592         p->hardirq_disable_event = 0;
1593         p->softirqs_enabled = 1;
1594         p->softirq_enable_ip = _THIS_IP_;
1595         p->softirq_enable_event = 0;
1596         p->softirq_disable_ip = 0;
1597         p->softirq_disable_event = 0;
1598         p->hardirq_context = 0;
1599         p->softirq_context = 0;
1600 #endif
1601
1602         p->pagefault_disabled = 0;
1603
1604 #ifdef CONFIG_LOCKDEP
1605         p->lockdep_depth = 0; /* no locks held yet */
1606         p->curr_chain_key = 0;
1607         p->lockdep_recursion = 0;
1608 #endif
1609
1610 #ifdef CONFIG_DEBUG_MUTEXES
1611         p->blocked_on = NULL; /* not blocked yet */
1612 #endif
1613 #ifdef CONFIG_BCACHE
1614         p->sequential_io        = 0;
1615         p->sequential_io_avg    = 0;
1616 #endif
1617
1618         /* Perform scheduler related setup. Assign this task to a CPU. */
1619         retval = sched_fork(clone_flags, p);
1620         if (retval)
1621                 goto bad_fork_cleanup_policy;
1622
1623         retval = perf_event_init_task(p);
1624         if (retval)
1625                 goto bad_fork_cleanup_policy;
1626         retval = audit_alloc(p);
1627         if (retval)
1628                 goto bad_fork_cleanup_perf;
1629         /* copy all the process information */
1630         shm_init_task(p);
1631         retval = copy_semundo(clone_flags, p);
1632         if (retval)
1633                 goto bad_fork_cleanup_audit;
1634         retval = copy_files(clone_flags, p);
1635         if (retval)
1636                 goto bad_fork_cleanup_semundo;
1637         retval = copy_fs(clone_flags, p);
1638         if (retval)
1639                 goto bad_fork_cleanup_files;
1640         retval = copy_sighand(clone_flags, p);
1641         if (retval)
1642                 goto bad_fork_cleanup_fs;
1643         retval = copy_signal(clone_flags, p);
1644         if (retval)
1645                 goto bad_fork_cleanup_sighand;
1646         retval = copy_mm(clone_flags, p);
1647         if (retval)
1648                 goto bad_fork_cleanup_signal;
1649         retval = copy_namespaces(clone_flags, p);
1650         if (retval)
1651                 goto bad_fork_cleanup_mm;
1652         retval = copy_io(clone_flags, p);
1653         if (retval)
1654                 goto bad_fork_cleanup_namespaces;
1655         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1656         if (retval)
1657                 goto bad_fork_cleanup_io;
1658
1659         if (pid != &init_struct_pid) {
1660                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1661                 if (IS_ERR(pid)) {
1662                         retval = PTR_ERR(pid);
1663                         goto bad_fork_cleanup_thread;
1664                 }
1665         }
1666
1667         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1668         /*
1669          * Clear TID on mm_release()?
1670          */
1671         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1672 #ifdef CONFIG_BLOCK
1673         p->plug = NULL;
1674 #endif
1675 #ifdef CONFIG_FUTEX
1676         p->robust_list = NULL;
1677 #ifdef CONFIG_COMPAT
1678         p->compat_robust_list = NULL;
1679 #endif
1680         INIT_LIST_HEAD(&p->pi_state_list);
1681         p->pi_state_cache = NULL;
1682 #endif
1683         /*
1684          * sigaltstack should be cleared when sharing the same VM
1685          */
1686         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1687                 sas_ss_reset(p);
1688
1689         /*
1690          * Syscall tracing and stepping should be turned off in the
1691          * child regardless of CLONE_PTRACE.
1692          */
1693         user_disable_single_step(p);
1694         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1695 #ifdef TIF_SYSCALL_EMU
1696         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1697 #endif
1698         clear_all_latency_tracing(p);
1699
1700         /* ok, now we should be set up.. */
1701         p->pid = pid_nr(pid);
1702         if (clone_flags & CLONE_THREAD) {
1703                 p->exit_signal = -1;
1704                 p->group_leader = current->group_leader;
1705                 p->tgid = current->tgid;
1706         } else {
1707                 if (clone_flags & CLONE_PARENT)
1708                         p->exit_signal = current->group_leader->exit_signal;
1709                 else
1710                         p->exit_signal = (clone_flags & CSIGNAL);
1711                 p->group_leader = p;
1712                 p->tgid = p->pid;
1713         }
1714
1715         p->nr_dirtied = 0;
1716         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1717         p->dirty_paused_when = 0;
1718
1719         p->pdeath_signal = 0;
1720         INIT_LIST_HEAD(&p->thread_group);
1721         p->task_works = NULL;
1722
1723         threadgroup_change_begin(current);
1724         /*
1725          * Ensure that the cgroup subsystem policies allow the new process to be
1726          * forked. It should be noted the the new process's css_set can be changed
1727          * between here and cgroup_post_fork() if an organisation operation is in
1728          * progress.
1729          */
1730         retval = cgroup_can_fork(p);
1731         if (retval)
1732                 goto bad_fork_free_pid;
1733
1734         /*
1735          * Make it visible to the rest of the system, but dont wake it up yet.
1736          * Need tasklist lock for parent etc handling!
1737          */
1738         write_lock_irq(&tasklist_lock);
1739
1740         /* CLONE_PARENT re-uses the old parent */
1741         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1742                 p->real_parent = current->real_parent;
1743                 p->parent_exec_id = current->parent_exec_id;
1744         } else {
1745                 p->real_parent = current;
1746                 p->parent_exec_id = current->self_exec_id;
1747         }
1748
1749         spin_lock(&current->sighand->siglock);
1750
1751         /*
1752          * Copy seccomp details explicitly here, in case they were changed
1753          * before holding sighand lock.
1754          */
1755         copy_seccomp(p);
1756
1757         /*
1758          * Process group and session signals need to be delivered to just the
1759          * parent before the fork or both the parent and the child after the
1760          * fork. Restart if a signal comes in before we add the new process to
1761          * it's process group.
1762          * A fatal signal pending means that current will exit, so the new
1763          * thread can't slip out of an OOM kill (or normal SIGKILL).
1764         */
1765         recalc_sigpending();
1766         if (signal_pending(current)) {
1767                 spin_unlock(&current->sighand->siglock);
1768                 write_unlock_irq(&tasklist_lock);
1769                 retval = -ERESTARTNOINTR;
1770                 goto bad_fork_cancel_cgroup;
1771         }
1772
1773         if (likely(p->pid)) {
1774                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1775
1776                 init_task_pid(p, PIDTYPE_PID, pid);
1777                 if (thread_group_leader(p)) {
1778                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1779                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1780
1781                         if (is_child_reaper(pid)) {
1782                                 ns_of_pid(pid)->child_reaper = p;
1783                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1784                         }
1785
1786                         p->signal->leader_pid = pid;
1787                         p->signal->tty = tty_kref_get(current->signal->tty);
1788                         list_add_tail(&p->sibling, &p->real_parent->children);
1789                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1790                         attach_pid(p, PIDTYPE_PGID);
1791                         attach_pid(p, PIDTYPE_SID);
1792                         __this_cpu_inc(process_counts);
1793                 } else {
1794                         current->signal->nr_threads++;
1795                         atomic_inc(&current->signal->live);
1796                         atomic_inc(&current->signal->sigcnt);
1797                         list_add_tail_rcu(&p->thread_group,
1798                                           &p->group_leader->thread_group);
1799                         list_add_tail_rcu(&p->thread_node,
1800                                           &p->signal->thread_head);
1801                 }
1802                 attach_pid(p, PIDTYPE_PID);
1803                 nr_threads++;
1804         }
1805
1806         total_forks++;
1807         spin_unlock(&current->sighand->siglock);
1808         syscall_tracepoint_update(p);
1809         write_unlock_irq(&tasklist_lock);
1810
1811         proc_fork_connector(p);
1812         cgroup_post_fork(p);
1813         threadgroup_change_end(current);
1814         perf_event_fork(p);
1815
1816         trace_task_newtask(p, clone_flags);
1817         uprobe_copy_process(p, clone_flags);
1818
1819         return p;
1820
1821 bad_fork_cancel_cgroup:
1822         cgroup_cancel_fork(p);
1823 bad_fork_free_pid:
1824         threadgroup_change_end(current);
1825         if (pid != &init_struct_pid)
1826                 free_pid(pid);
1827 bad_fork_cleanup_thread:
1828         exit_thread(p);
1829 bad_fork_cleanup_io:
1830         if (p->io_context)
1831                 exit_io_context(p);
1832 bad_fork_cleanup_namespaces:
1833         exit_task_namespaces(p);
1834 bad_fork_cleanup_mm:
1835         if (p->mm)
1836                 mmput(p->mm);
1837 bad_fork_cleanup_signal:
1838         if (!(clone_flags & CLONE_THREAD))
1839                 free_signal_struct(p->signal);
1840 bad_fork_cleanup_sighand:
1841         __cleanup_sighand(p->sighand);
1842 bad_fork_cleanup_fs:
1843         exit_fs(p); /* blocking */
1844 bad_fork_cleanup_files:
1845         exit_files(p); /* blocking */
1846 bad_fork_cleanup_semundo:
1847         exit_sem(p);
1848 bad_fork_cleanup_audit:
1849         audit_free(p);
1850 bad_fork_cleanup_perf:
1851         perf_event_free_task(p);
1852 bad_fork_cleanup_policy:
1853 #ifdef CONFIG_NUMA
1854         mpol_put(p->mempolicy);
1855 bad_fork_cleanup_threadgroup_lock:
1856 #endif
1857         delayacct_tsk_free(p);
1858 bad_fork_cleanup_count:
1859         atomic_dec(&p->cred->user->processes);
1860         exit_creds(p);
1861 bad_fork_free:
1862         put_task_stack(p);
1863         free_task(p);
1864 fork_out:
1865         return ERR_PTR(retval);
1866 }
1867
1868 static inline void init_idle_pids(struct pid_link *links)
1869 {
1870         enum pid_type type;
1871
1872         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1873                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1874                 links[type].pid = &init_struct_pid;
1875         }
1876 }
1877
1878 struct task_struct *fork_idle(int cpu)
1879 {
1880         struct task_struct *task;
1881         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1882                             cpu_to_node(cpu));
1883         if (!IS_ERR(task)) {
1884                 init_idle_pids(task->pids);
1885                 init_idle(task, cpu);
1886         }
1887
1888         return task;
1889 }
1890
1891 /*
1892  *  Ok, this is the main fork-routine.
1893  *
1894  * It copies the process, and if successful kick-starts
1895  * it and waits for it to finish using the VM if required.
1896  */
1897 long _do_fork(unsigned long clone_flags,
1898               unsigned long stack_start,
1899               unsigned long stack_size,
1900               int __user *parent_tidptr,
1901               int __user *child_tidptr,
1902               unsigned long tls)
1903 {
1904         struct task_struct *p;
1905         int trace = 0;
1906         long nr;
1907
1908         /*
1909          * Determine whether and which event to report to ptracer.  When
1910          * called from kernel_thread or CLONE_UNTRACED is explicitly
1911          * requested, no event is reported; otherwise, report if the event
1912          * for the type of forking is enabled.
1913          */
1914         if (!(clone_flags & CLONE_UNTRACED)) {
1915                 if (clone_flags & CLONE_VFORK)
1916                         trace = PTRACE_EVENT_VFORK;
1917                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1918                         trace = PTRACE_EVENT_CLONE;
1919                 else
1920                         trace = PTRACE_EVENT_FORK;
1921
1922                 if (likely(!ptrace_event_enabled(current, trace)))
1923                         trace = 0;
1924         }
1925
1926         p = copy_process(clone_flags, stack_start, stack_size,
1927                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
1928         /*
1929          * Do this prior waking up the new thread - the thread pointer
1930          * might get invalid after that point, if the thread exits quickly.
1931          */
1932         if (!IS_ERR(p)) {
1933                 struct completion vfork;
1934                 struct pid *pid;
1935
1936                 trace_sched_process_fork(current, p);
1937
1938                 pid = get_task_pid(p, PIDTYPE_PID);
1939                 nr = pid_vnr(pid);
1940
1941                 if (clone_flags & CLONE_PARENT_SETTID)
1942                         put_user(nr, parent_tidptr);
1943
1944                 if (clone_flags & CLONE_VFORK) {
1945                         p->vfork_done = &vfork;
1946                         init_completion(&vfork);
1947                         get_task_struct(p);
1948                 }
1949
1950                 wake_up_new_task(p);
1951
1952                 /* forking complete and child started to run, tell ptracer */
1953                 if (unlikely(trace))
1954                         ptrace_event_pid(trace, pid);
1955
1956                 if (clone_flags & CLONE_VFORK) {
1957                         if (!wait_for_vfork_done(p, &vfork))
1958                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1959                 }
1960
1961                 put_pid(pid);
1962         } else {
1963                 nr = PTR_ERR(p);
1964         }
1965         return nr;
1966 }
1967
1968 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1969 /* For compatibility with architectures that call do_fork directly rather than
1970  * using the syscall entry points below. */
1971 long do_fork(unsigned long clone_flags,
1972               unsigned long stack_start,
1973               unsigned long stack_size,
1974               int __user *parent_tidptr,
1975               int __user *child_tidptr)
1976 {
1977         return _do_fork(clone_flags, stack_start, stack_size,
1978                         parent_tidptr, child_tidptr, 0);
1979 }
1980 #endif
1981
1982 /*
1983  * Create a kernel thread.
1984  */
1985 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1986 {
1987         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1988                 (unsigned long)arg, NULL, NULL, 0);
1989 }
1990
1991 #ifdef __ARCH_WANT_SYS_FORK
1992 SYSCALL_DEFINE0(fork)
1993 {
1994 #ifdef CONFIG_MMU
1995         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1996 #else
1997         /* can not support in nommu mode */
1998         return -EINVAL;
1999 #endif
2000 }
2001 #endif
2002
2003 #ifdef __ARCH_WANT_SYS_VFORK
2004 SYSCALL_DEFINE0(vfork)
2005 {
2006         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2007                         0, NULL, NULL, 0);
2008 }
2009 #endif
2010
2011 #ifdef __ARCH_WANT_SYS_CLONE
2012 #ifdef CONFIG_CLONE_BACKWARDS
2013 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2014                  int __user *, parent_tidptr,
2015                  unsigned long, tls,
2016                  int __user *, child_tidptr)
2017 #elif defined(CONFIG_CLONE_BACKWARDS2)
2018 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2019                  int __user *, parent_tidptr,
2020                  int __user *, child_tidptr,
2021                  unsigned long, tls)
2022 #elif defined(CONFIG_CLONE_BACKWARDS3)
2023 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2024                 int, stack_size,
2025                 int __user *, parent_tidptr,
2026                 int __user *, child_tidptr,
2027                 unsigned long, tls)
2028 #else
2029 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2030                  int __user *, parent_tidptr,
2031                  int __user *, child_tidptr,
2032                  unsigned long, tls)
2033 #endif
2034 {
2035         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2036 }
2037 #endif
2038
2039 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2040 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2041 #endif
2042
2043 static void sighand_ctor(void *data)
2044 {
2045         struct sighand_struct *sighand = data;
2046
2047         spin_lock_init(&sighand->siglock);
2048         init_waitqueue_head(&sighand->signalfd_wqh);
2049 }
2050
2051 void __init proc_caches_init(void)
2052 {
2053         sighand_cachep = kmem_cache_create("sighand_cache",
2054                         sizeof(struct sighand_struct), 0,
2055                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
2056                         SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2057         signal_cachep = kmem_cache_create("signal_cache",
2058                         sizeof(struct signal_struct), 0,
2059                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2060                         NULL);
2061         files_cachep = kmem_cache_create("files_cache",
2062                         sizeof(struct files_struct), 0,
2063                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2064                         NULL);
2065         fs_cachep = kmem_cache_create("fs_cache",
2066                         sizeof(struct fs_struct), 0,
2067                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2068                         NULL);
2069         /*
2070          * FIXME! The "sizeof(struct mm_struct)" currently includes the
2071          * whole struct cpumask for the OFFSTACK case. We could change
2072          * this to *only* allocate as much of it as required by the
2073          * maximum number of CPU's we can ever have.  The cpumask_allocation
2074          * is at the end of the structure, exactly for that reason.
2075          */
2076         mm_cachep = kmem_cache_create("mm_struct",
2077                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2078                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2079                         NULL);
2080         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2081         mmap_init();
2082         nsproxy_cache_init();
2083 }
2084
2085 /*
2086  * Check constraints on flags passed to the unshare system call.
2087  */
2088 static int check_unshare_flags(unsigned long unshare_flags)
2089 {
2090         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2091                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2092                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2093                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2094                 return -EINVAL;
2095         /*
2096          * Not implemented, but pretend it works if there is nothing
2097          * to unshare.  Note that unsharing the address space or the
2098          * signal handlers also need to unshare the signal queues (aka
2099          * CLONE_THREAD).
2100          */
2101         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2102                 if (!thread_group_empty(current))
2103                         return -EINVAL;
2104         }
2105         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2106                 if (atomic_read(&current->sighand->count) > 1)
2107                         return -EINVAL;
2108         }
2109         if (unshare_flags & CLONE_VM) {
2110                 if (!current_is_single_threaded())
2111                         return -EINVAL;
2112         }
2113
2114         return 0;
2115 }
2116
2117 /*
2118  * Unshare the filesystem structure if it is being shared
2119  */
2120 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2121 {
2122         struct fs_struct *fs = current->fs;
2123
2124         if (!(unshare_flags & CLONE_FS) || !fs)
2125                 return 0;
2126
2127         /* don't need lock here; in the worst case we'll do useless copy */
2128         if (fs->users == 1)
2129                 return 0;
2130
2131         *new_fsp = copy_fs_struct(fs);
2132         if (!*new_fsp)
2133                 return -ENOMEM;
2134
2135         return 0;
2136 }
2137
2138 /*
2139  * Unshare file descriptor table if it is being shared
2140  */
2141 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2142 {
2143         struct files_struct *fd = current->files;
2144         int error = 0;
2145
2146         if ((unshare_flags & CLONE_FILES) &&
2147             (fd && atomic_read(&fd->count) > 1)) {
2148                 *new_fdp = dup_fd(fd, &error);
2149                 if (!*new_fdp)
2150                         return error;
2151         }
2152
2153         return 0;
2154 }
2155
2156 /*
2157  * unshare allows a process to 'unshare' part of the process
2158  * context which was originally shared using clone.  copy_*
2159  * functions used by do_fork() cannot be used here directly
2160  * because they modify an inactive task_struct that is being
2161  * constructed. Here we are modifying the current, active,
2162  * task_struct.
2163  */
2164 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2165 {
2166         struct fs_struct *fs, *new_fs = NULL;
2167         struct files_struct *fd, *new_fd = NULL;
2168         struct cred *new_cred = NULL;
2169         struct nsproxy *new_nsproxy = NULL;
2170         int do_sysvsem = 0;
2171         int err;
2172
2173         /*
2174          * If unsharing a user namespace must also unshare the thread group
2175          * and unshare the filesystem root and working directories.
2176          */
2177         if (unshare_flags & CLONE_NEWUSER)
2178                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2179         /*
2180          * If unsharing vm, must also unshare signal handlers.
2181          */
2182         if (unshare_flags & CLONE_VM)
2183                 unshare_flags |= CLONE_SIGHAND;
2184         /*
2185          * If unsharing a signal handlers, must also unshare the signal queues.
2186          */
2187         if (unshare_flags & CLONE_SIGHAND)
2188                 unshare_flags |= CLONE_THREAD;
2189         /*
2190          * If unsharing namespace, must also unshare filesystem information.
2191          */
2192         if (unshare_flags & CLONE_NEWNS)
2193                 unshare_flags |= CLONE_FS;
2194
2195         err = check_unshare_flags(unshare_flags);
2196         if (err)
2197                 goto bad_unshare_out;
2198         /*
2199          * CLONE_NEWIPC must also detach from the undolist: after switching
2200          * to a new ipc namespace, the semaphore arrays from the old
2201          * namespace are unreachable.
2202          */
2203         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2204                 do_sysvsem = 1;
2205         err = unshare_fs(unshare_flags, &new_fs);
2206         if (err)
2207                 goto bad_unshare_out;
2208         err = unshare_fd(unshare_flags, &new_fd);
2209         if (err)
2210                 goto bad_unshare_cleanup_fs;
2211         err = unshare_userns(unshare_flags, &new_cred);
2212         if (err)
2213                 goto bad_unshare_cleanup_fd;
2214         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2215                                          new_cred, new_fs);
2216         if (err)
2217                 goto bad_unshare_cleanup_cred;
2218
2219         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2220                 if (do_sysvsem) {
2221                         /*
2222                          * CLONE_SYSVSEM is equivalent to sys_exit().
2223                          */
2224                         exit_sem(current);
2225                 }
2226                 if (unshare_flags & CLONE_NEWIPC) {
2227                         /* Orphan segments in old ns (see sem above). */
2228                         exit_shm(current);
2229                         shm_init_task(current);
2230                 }
2231
2232                 if (new_nsproxy)
2233                         switch_task_namespaces(current, new_nsproxy);
2234
2235                 task_lock(current);
2236
2237                 if (new_fs) {
2238                         fs = current->fs;
2239                         spin_lock(&fs->lock);
2240                         current->fs = new_fs;
2241                         if (--fs->users)
2242                                 new_fs = NULL;
2243                         else
2244                                 new_fs = fs;
2245                         spin_unlock(&fs->lock);
2246                 }
2247
2248                 if (new_fd) {
2249                         fd = current->files;
2250                         current->files = new_fd;
2251                         new_fd = fd;
2252                 }
2253
2254                 task_unlock(current);
2255
2256                 if (new_cred) {
2257                         /* Install the new user namespace */
2258                         commit_creds(new_cred);
2259                         new_cred = NULL;
2260                 }
2261         }
2262
2263 bad_unshare_cleanup_cred:
2264         if (new_cred)
2265                 put_cred(new_cred);
2266 bad_unshare_cleanup_fd:
2267         if (new_fd)
2268                 put_files_struct(new_fd);
2269
2270 bad_unshare_cleanup_fs:
2271         if (new_fs)
2272                 free_fs_struct(new_fs);
2273
2274 bad_unshare_out:
2275         return err;
2276 }
2277
2278 /*
2279  *      Helper to unshare the files of the current task.
2280  *      We don't want to expose copy_files internals to
2281  *      the exec layer of the kernel.
2282  */
2283
2284 int unshare_files(struct files_struct **displaced)
2285 {
2286         struct task_struct *task = current;
2287         struct files_struct *copy = NULL;
2288         int error;
2289
2290         error = unshare_fd(CLONE_FILES, &copy);
2291         if (error || !copy) {
2292                 *displaced = NULL;
2293                 return error;
2294         }
2295         *displaced = task->files;
2296         task_lock(task);
2297         task->files = copy;
2298         task_unlock(task);
2299         return 0;
2300 }
2301
2302 int sysctl_max_threads(struct ctl_table *table, int write,
2303                        void __user *buffer, size_t *lenp, loff_t *ppos)
2304 {
2305         struct ctl_table t;
2306         int ret;
2307         int threads = max_threads;
2308         int min = MIN_THREADS;
2309         int max = MAX_THREADS;
2310
2311         t = *table;
2312         t.data = &threads;
2313         t.extra1 = &min;
2314         t.extra2 = &max;
2315
2316         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2317         if (ret || !write)
2318                 return ret;
2319
2320         set_max_threads(threads);
2321
2322         return 0;
2323 }