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