kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/sched/autogroup.h>
  16 #include <linux/sched/mm.h>
  17 #include <linux/sched/coredump.h>
  18 #include <linux/sched/user.h>
  19 #include <linux/sched/numa_balancing.h>
  20 #include <linux/init.h>
  21 #include <linux/unistd.h>
  22 #include <linux/module.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/completion.h>
  25 #include <linux/personality.h>
  26 #include <linux/mempolicy.h>
  27 #include <linux/sem.h>
  28 #include <linux/file.h>
  29 #include <linux/fdtable.h>
  30 #include <linux/iocontext.h>
  31 #include <linux/key.h>
  32 #include <linux/binfmts.h>
  33 #include <linux/mman.h>
  34 #include <linux/mmu_notifier.h>
  35 #include <linux/fs.h>
  36 #include <linux/mm.h>
  37 #include <linux/vmacache.h>
  38 #include <linux/nsproxy.h>
  39 #include <linux/capability.h>
  40 #include <linux/cpu.h>
  41 #include <linux/cgroup.h>
  42 #include <linux/security.h>
  43 #include <linux/hugetlb.h>
  44 #include <linux/seccomp.h>
  45 #include <linux/swap.h>
  46 #include <linux/syscalls.h>
  47 #include <linux/jiffies.h>
  48 #include <linux/futex.h>
  49 #include <linux/compat.h>
  50 #include <linux/kthread.h>
  51 #include <linux/task_io_accounting_ops.h>
  52 #include <linux/rcupdate.h>
  53 #include <linux/ptrace.h>
  54 #include <linux/mount.h>
  55 #include <linux/audit.h>
  56 #include <linux/memcontrol.h>
  57 #include <linux/ftrace.h>
  58 #include <linux/proc_fs.h>
  59 #include <linux/profile.h>
  60 #include <linux/rmap.h>
  61 #include <linux/ksm.h>
  62 #include <linux/acct.h>
  63 #include <linux/userfaultfd_k.h>
  64 #include <linux/tsacct_kern.h>
  65 #include <linux/cn_proc.h>
  66 #include <linux/freezer.h>
  67 #include <linux/delayacct.h>
  68 #include <linux/taskstats_kern.h>
  69 #include <linux/random.h>
  70 #include <linux/tty.h>
  71 #include <linux/blkdev.h>
  72 #include <linux/fs_struct.h>
  73 #include <linux/magic.h>
  74 #include <linux/perf_event.h>
  75 #include <linux/posix-timers.h>
  76 #include <linux/user-return-notifier.h>
  77 #include <linux/oom.h>
  78 #include <linux/khugepaged.h>
  79 #include <linux/signalfd.h>
  80 #include <linux/uprobes.h>
  81 #include <linux/aio.h>
  82 #include <linux/compiler.h>
  83 #include <linux/sysctl.h>
  84 #include <linux/kcov.h>
  85
  86 #include <asm/pgtable.h>
  87 #include <asm/pgalloc.h>
  88 #include <linux/uaccess.h>
  89 #include <asm/mmu_context.h>
  90 #include <asm/cacheflush.h>
  91 #include <asm/tlbflush.h>
  92
  93 #include <trace/events/sched.h>
  94
  95 #define CREATE_TRACE_POINTS
  96 #include <trace/events/task.h>
  97
  98 /*
  99  * Minimum number of threads to boot the kernel
 100  */
 101 #define MIN_THREADS 20
 102
 103 /*
 104  * Maximum number of threads
 105  */
 106 #define MAX_THREADS FUTEX_TID_MASK
 107
 108 /*
 109  * Protected counters by write_lock_irq(&tasklist_lock)
 110  */
 111 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 112 int nr_threads;                 /* The idle threads do not count.. */
 113
 114 int max_threads;                /* tunable limit on nr_threads */
 115
 116 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 117
 118 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 119
 120 #ifdef CONFIG_PROVE_RCU
 121 int lockdep_tasklist_lock_is_held(void)
 122 {
 123         return lockdep_is_held(&tasklist_lock);
 124 }
 125 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 126 #endif /* #ifdef CONFIG_PROVE_RCU */
 127
 128 int nr_processes(void)
 129 {
 130         int cpu;
 131         int total = 0;
 132
 133         for_each_possible_cpu(cpu)
 134                 total += per_cpu(process_counts, cpu);
 135
 136         return total;
 137 }
 138
 139 void __weak arch_release_task_struct(struct task_struct *tsk)
 140 {
 141 }
 142
 143 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 144 static struct kmem_cache *task_struct_cachep;
 145
 146 static inline struct task_struct *alloc_task_struct_node(int node)
 147 {
 148         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 149 }
 150
 151 static inline void free_task_struct(struct task_struct *tsk)
 152 {
 153         kmem_cache_free(task_struct_cachep, tsk);
 154 }
 155 #endif
 156
 157 void __weak arch_release_thread_stack(unsigned long *stack)
 158 {
 159 }
 160
 161 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 162
 163 /*
 164  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 165  * kmemcache based allocator.
 166  */
 167 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 168
 169 #ifdef CONFIG_VMAP_STACK
 170 /*
 171  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 172  * flush.  Try to minimize the number of calls by caching stacks.
 173  */
 174 #define NR_CACHED_STACKS 2
 175 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 176 #endif
 177
 178 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 179 {
 180 #ifdef CONFIG_VMAP_STACK
 181         void *stack;
 182         int i;
 183
 184         local_irq_disable();
 185         for (i = 0; i < NR_CACHED_STACKS; i++) {
 186                 struct vm_struct *s = this_cpu_read(cached_stacks[i]);
 187
 188                 if (!s)
 189                         continue;
 190                 this_cpu_write(cached_stacks[i], NULL);
 191
 192                 tsk->stack_vm_area = s;
 193                 local_irq_enable();
 194                 return s->addr;
 195         }
 196         local_irq_enable();
 197
 198         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_SIZE,
 199                                      VMALLOC_START, VMALLOC_END,
 200                                      THREADINFO_GFP | __GFP_HIGHMEM,
 201                                      PAGE_KERNEL,
 202                                      0, node, __builtin_return_address(0));
 203
 204         /*
 205          * We can't call find_vm_area() in interrupt context, and
 206          * free_thread_stack() can be called in interrupt context,
 207          * so cache the vm_struct.
 208          */
 209         if (stack)
 210                 tsk->stack_vm_area = find_vm_area(stack);
 211         return stack;
 212 #else
 213         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 214                                              THREAD_SIZE_ORDER);
 215
 216         return page ? page_address(page) : NULL;
 217 #endif
 218 }
 219
 220 static inline void free_thread_stack(struct task_struct *tsk)
 221 {
 222 #ifdef CONFIG_VMAP_STACK
 223         if (task_stack_vm_area(tsk)) {
 224                 unsigned long flags;
 225                 int i;
 226
 227                 local_irq_save(flags);
 228                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 229                         if (this_cpu_read(cached_stacks[i]))
 230                                 continue;
 231
 232                         this_cpu_write(cached_stacks[i], tsk->stack_vm_area);
 233                         local_irq_restore(flags);
 234                         return;
 235                 }
 236                 local_irq_restore(flags);
 237
 238                 vfree_atomic(tsk->stack);
 239                 return;
 240         }
 241 #endif
 242
 243         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 244 }
 245 # else
 246 static struct kmem_cache *thread_stack_cache;
 247
 248 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 249                                                   int node)
 250 {
 251         return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 252 }
 253
 254 static void free_thread_stack(struct task_struct *tsk)
 255 {
 256         kmem_cache_free(thread_stack_cache, tsk->stack);
 257 }
 258
 259 void thread_stack_cache_init(void)
 260 {
 261         thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
 262                                               THREAD_SIZE, 0, NULL);
 263         BUG_ON(thread_stack_cache == NULL);
 264 }
 265 # endif
 266 #endif
 267
 268 /* SLAB cache for signal_struct structures (tsk->signal) */
 269 static struct kmem_cache *signal_cachep;
 270
 271 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 272 struct kmem_cache *sighand_cachep;
 273
 274 /* SLAB cache for files_struct structures (tsk->files) */
 275 struct kmem_cache *files_cachep;
 276
 277 /* SLAB cache for fs_struct structures (tsk->fs) */
 278 struct kmem_cache *fs_cachep;
 279
 280 /* SLAB cache for vm_area_struct structures */
 281 struct kmem_cache *vm_area_cachep;
 282
 283 /* SLAB cache for mm_struct structures (tsk->mm) */
 284 static struct kmem_cache *mm_cachep;
 285
 286 static void account_kernel_stack(struct task_struct *tsk, int account)
 287 {
 288         void *stack = task_stack_page(tsk);
 289         struct vm_struct *vm = task_stack_vm_area(tsk);
 290
 291         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 292
 293         if (vm) {
 294                 int i;
 295
 296                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 297
 298                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 299                         mod_zone_page_state(page_zone(vm->pages[i]),
 300                                             NR_KERNEL_STACK_KB,
 301                                             PAGE_SIZE / 1024 * account);
 302                 }
 303
 304                 /* All stack pages belong to the same memcg. */
 305                 memcg_kmem_update_page_stat(vm->pages[0], MEMCG_KERNEL_STACK_KB,
 306                                             account * (THREAD_SIZE / 1024));
 307         } else {
 308                 /*
 309                  * All stack pages are in the same zone and belong to the
 310                  * same memcg.
 311                  */
 312                 struct page *first_page = virt_to_page(stack);
 313
 314                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 315                                     THREAD_SIZE / 1024 * account);
 316
 317                 memcg_kmem_update_page_stat(first_page, MEMCG_KERNEL_STACK_KB,
 318                                             account * (THREAD_SIZE / 1024));
 319         }
 320 }
 321
 322 static void release_task_stack(struct task_struct *tsk)
 323 {
 324         if (WARN_ON(tsk->state != TASK_DEAD))
 325                 return;  /* Better to leak the stack than to free prematurely */
 326
 327         account_kernel_stack(tsk, -1);
 328         arch_release_thread_stack(tsk->stack);
 329         free_thread_stack(tsk);
 330         tsk->stack = NULL;
 331 #ifdef CONFIG_VMAP_STACK
 332         tsk->stack_vm_area = NULL;
 333 #endif
 334 }
 335
 336 #ifdef CONFIG_THREAD_INFO_IN_TASK
 337 void put_task_stack(struct task_struct *tsk)
 338 {
 339         if (atomic_dec_and_test(&tsk->stack_refcount))
 340                 release_task_stack(tsk);
 341 }
 342 #endif
 343
 344 void free_task(struct task_struct *tsk)
 345 {
 346 #ifndef CONFIG_THREAD_INFO_IN_TASK
 347         /*
 348          * The task is finally done with both the stack and thread_info,
 349          * so free both.
 350          */
 351         release_task_stack(tsk);
 352 #else
 353         /*
 354          * If the task had a separate stack allocation, it should be gone
 355          * by now.
 356          */
 357         WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
 358 #endif
 359         rt_mutex_debug_task_free(tsk);
 360         ftrace_graph_exit_task(tsk);
 361         put_seccomp_filter(tsk);
 362         arch_release_task_struct(tsk);
 363         if (tsk->flags & PF_KTHREAD)
 364                 free_kthread_struct(tsk);
 365         free_task_struct(tsk);
 366 }
 367 EXPORT_SYMBOL(free_task);
 368
 369 static inline void free_signal_struct(struct signal_struct *sig)
 370 {
 371         taskstats_tgid_free(sig);
 372         sched_autogroup_exit(sig);
 373         /*
 374          * __mmdrop is not safe to call from softirq context on x86 due to
 375          * pgd_dtor so postpone it to the async context
 376          */
 377         if (sig->oom_mm)
 378                 mmdrop_async(sig->oom_mm);
 379         kmem_cache_free(signal_cachep, sig);
 380 }
 381
 382 static inline void put_signal_struct(struct signal_struct *sig)
 383 {
 384         if (atomic_dec_and_test(&sig->sigcnt))
 385                 free_signal_struct(sig);
 386 }
 387
 388 void __put_task_struct(struct task_struct *tsk)
 389 {
 390         WARN_ON(!tsk->exit_state);
 391         WARN_ON(atomic_read(&tsk->usage));
 392         WARN_ON(tsk == current);
 393
 394         cgroup_free(tsk);
 395         task_numa_free(tsk);
 396         security_task_free(tsk);
 397         exit_creds(tsk);
 398         delayacct_tsk_free(tsk);
 399         put_signal_struct(tsk->signal);
 400
 401         if (!profile_handoff_task(tsk))
 402                 free_task(tsk);
 403 }
 404 EXPORT_SYMBOL_GPL(__put_task_struct);
 405
 406 void __init __weak arch_task_cache_init(void) { }
 407
 408 /*
 409  * set_max_threads
 410  */
 411 static void set_max_threads(unsigned int max_threads_suggested)
 412 {
 413         u64 threads;
 414
 415         /*
 416          * The number of threads shall be limited such that the thread
 417          * structures may only consume a small part of the available memory.
 418          */
 419         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 420                 threads = MAX_THREADS;
 421         else
 422                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 423                                     (u64) THREAD_SIZE * 8UL);
 424
 425         if (threads > max_threads_suggested)
 426                 threads = max_threads_suggested;
 427
 428         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 429 }
 430
 431 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 432 /* Initialized by the architecture: */
 433 int arch_task_struct_size __read_mostly;
 434 #endif
 435
 436 void __init fork_init(void)
 437 {
 438         int i;
 439 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 440 #ifndef ARCH_MIN_TASKALIGN
 441 #define ARCH_MIN_TASKALIGN      0
 442 #endif
 443         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 444
 445         /* create a slab on which task_structs can be allocated */
 446         task_struct_cachep = kmem_cache_create("task_struct",
 447                         arch_task_struct_size, align,
 448                         SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
 449 #endif
 450
 451         /* do the arch specific task caches init */
 452         arch_task_cache_init();
 453
 454         set_max_threads(MAX_THREADS);
 455
 456         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 457         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 458         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 459                 init_task.signal->rlim[RLIMIT_NPROC];
 460
 461         for (i = 0; i < UCOUNT_COUNTS; i++) {
 462                 init_user_ns.ucount_max[i] = max_threads/2;
 463         }
 464 }
 465
 466 int __weak arch_dup_task_struct(struct task_struct *dst,
 467                                                struct task_struct *src)
 468 {
 469         *dst = *src;
 470         return 0;
 471 }
 472
 473 void set_task_stack_end_magic(struct task_struct *tsk)
 474 {
 475         unsigned long *stackend;
 476
 477         stackend = end_of_stack(tsk);
 478         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 479 }
 480
 481 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 482 {
 483         struct task_struct *tsk;
 484         unsigned long *stack;
 485         struct vm_struct *stack_vm_area;
 486         int err;
 487
 488         if (node == NUMA_NO_NODE)
 489                 node = tsk_fork_get_node(orig);
 490         tsk = alloc_task_struct_node(node);
 491         if (!tsk)
 492                 return NULL;
 493
 494         stack = alloc_thread_stack_node(tsk, node);
 495         if (!stack)
 496                 goto free_tsk;
 497
 498         stack_vm_area = task_stack_vm_area(tsk);
 499
 500         err = arch_dup_task_struct(tsk, orig);
 501
 502         /*
 503          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 504          * sure they're properly initialized before using any stack-related
 505          * functions again.
 506          */
 507         tsk->stack = stack;
 508 #ifdef CONFIG_VMAP_STACK
 509         tsk->stack_vm_area = stack_vm_area;
 510 #endif
 511 #ifdef CONFIG_THREAD_INFO_IN_TASK
 512         atomic_set(&tsk->stack_refcount, 1);
 513 #endif
 514
 515         if (err)
 516                 goto free_stack;
 517
 518 #ifdef CONFIG_SECCOMP
 519         /*
 520          * We must handle setting up seccomp filters once we're under
 521          * the sighand lock in case orig has changed between now and
 522          * then. Until then, filter must be NULL to avoid messing up
 523          * the usage counts on the error path calling free_task.
 524          */
 525         tsk->seccomp.filter = NULL;
 526 #endif
 527
 528         setup_thread_stack(tsk, orig);
 529         clear_user_return_notifier(tsk);
 530         clear_tsk_need_resched(tsk);
 531         set_task_stack_end_magic(tsk);
 532
 533 #ifdef CONFIG_CC_STACKPROTECTOR
 534         tsk->stack_canary = get_random_int();
 535 #endif
 536
 537         /*
 538          * One for us, one for whoever does the "release_task()" (usually
 539          * parent)
 540          */
 541         atomic_set(&tsk->usage, 2);
 542 #ifdef CONFIG_BLK_DEV_IO_TRACE
 543         tsk->btrace_seq = 0;
 544 #endif
 545         tsk->splice_pipe = NULL;
 546         tsk->task_frag.page = NULL;
 547         tsk->wake_q.next = NULL;
 548
 549         account_kernel_stack(tsk, 1);
 550
 551         kcov_task_init(tsk);
 552
 553         return tsk;
 554
 555 free_stack:
 556         free_thread_stack(tsk);
 557 free_tsk:
 558         free_task_struct(tsk);
 559         return NULL;
 560 }
 561
 562 #ifdef CONFIG_MMU
 563 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 564                                         struct mm_struct *oldmm)
 565 {
 566         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 567         struct rb_node **rb_link, *rb_parent;
 568         int retval;
 569         unsigned long charge;
 570         LIST_HEAD(uf);
 571
 572         uprobe_start_dup_mmap();
 573         if (down_write_killable(&oldmm->mmap_sem)) {
 574                 retval = -EINTR;
 575                 goto fail_uprobe_end;
 576         }
 577         flush_cache_dup_mm(oldmm);
 578         uprobe_dup_mmap(oldmm, mm);
 579         /*
 580          * Not linked in yet - no deadlock potential:
 581          */
 582         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 583
 584         /* No ordering required: file already has been exposed. */
 585         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 586
 587         mm->total_vm = oldmm->total_vm;
 588         mm->data_vm = oldmm->data_vm;
 589         mm->exec_vm = oldmm->exec_vm;
 590         mm->stack_vm = oldmm->stack_vm;
 591
 592         rb_link = &mm->mm_rb.rb_node;
 593         rb_parent = NULL;
 594         pprev = &mm->mmap;
 595         retval = ksm_fork(mm, oldmm);
 596         if (retval)
 597                 goto out;
 598         retval = khugepaged_fork(mm, oldmm);
 599         if (retval)
 600                 goto out;
 601
 602         prev = NULL;
 603         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 604                 struct file *file;
 605
 606                 if (mpnt->vm_flags & VM_DONTCOPY) {
 607                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 608                         continue;
 609                 }
 610                 charge = 0;
 611                 if (mpnt->vm_flags & VM_ACCOUNT) {
 612                         unsigned long len = vma_pages(mpnt);
 613
 614                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 615                                 goto fail_nomem;
 616                         charge = len;
 617                 }
 618                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 619                 if (!tmp)
 620                         goto fail_nomem;
 621                 *tmp = *mpnt;
 622                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 623                 retval = vma_dup_policy(mpnt, tmp);
 624                 if (retval)
 625                         goto fail_nomem_policy;
 626                 tmp->vm_mm = mm;
 627                 retval = dup_userfaultfd(tmp, &uf);
 628                 if (retval)
 629                         goto fail_nomem_anon_vma_fork;
 630                 if (anon_vma_fork(tmp, mpnt))
 631                         goto fail_nomem_anon_vma_fork;
 632                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 633                 tmp->vm_next = tmp->vm_prev = NULL;
 634                 file = tmp->vm_file;
 635                 if (file) {
 636                         struct inode *inode = file_inode(file);
 637                         struct address_space *mapping = file->f_mapping;
 638
 639                         get_file(file);
 640                         if (tmp->vm_flags & VM_DENYWRITE)
 641                                 atomic_dec(&inode->i_writecount);
 642                         i_mmap_lock_write(mapping);
 643                         if (tmp->vm_flags & VM_SHARED)
 644                                 atomic_inc(&mapping->i_mmap_writable);
 645                         flush_dcache_mmap_lock(mapping);
 646                         /* insert tmp into the share list, just after mpnt */
 647                         vma_interval_tree_insert_after(tmp, mpnt,
 648                                         &mapping->i_mmap);
 649                         flush_dcache_mmap_unlock(mapping);
 650                         i_mmap_unlock_write(mapping);
 651                 }
 652
 653                 /*
 654                  * Clear hugetlb-related page reserves for children. This only
 655                  * affects MAP_PRIVATE mappings. Faults generated by the child
 656                  * are not guaranteed to succeed, even if read-only
 657                  */
 658                 if (is_vm_hugetlb_page(tmp))
 659                         reset_vma_resv_huge_pages(tmp);
 660
 661                 /*
 662                  * Link in the new vma and copy the page table entries.
 663                  */
 664                 *pprev = tmp;
 665                 pprev = &tmp->vm_next;
 666                 tmp->vm_prev = prev;
 667                 prev = tmp;
 668
 669                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 670                 rb_link = &tmp->vm_rb.rb_right;
 671                 rb_parent = &tmp->vm_rb;
 672
 673                 mm->map_count++;
 674                 retval = copy_page_range(mm, oldmm, mpnt);
 675
 676                 if (tmp->vm_ops && tmp->vm_ops->open)
 677                         tmp->vm_ops->open(tmp);
 678
 679                 if (retval)
 680                         goto out;
 681         }
 682         /* a new mm has just been created */
 683         arch_dup_mmap(oldmm, mm);
 684         retval = 0;
 685 out:
 686         up_write(&mm->mmap_sem);
 687         flush_tlb_mm(oldmm);
 688         up_write(&oldmm->mmap_sem);
 689         dup_userfaultfd_complete(&uf);
 690 fail_uprobe_end:
 691         uprobe_end_dup_mmap();
 692         return retval;
 693 fail_nomem_anon_vma_fork:
 694         mpol_put(vma_policy(tmp));
 695 fail_nomem_policy:
 696         kmem_cache_free(vm_area_cachep, tmp);
 697 fail_nomem:
 698         retval = -ENOMEM;
 699         vm_unacct_memory(charge);
 700         goto out;
 701 }
 702
 703 static inline int mm_alloc_pgd(struct mm_struct *mm)
 704 {
 705         mm->pgd = pgd_alloc(mm);
 706         if (unlikely(!mm->pgd))
 707                 return -ENOMEM;
 708         return 0;
 709 }
 710
 711 static inline void mm_free_pgd(struct mm_struct *mm)
 712 {
 713         pgd_free(mm, mm->pgd);
 714 }
 715 #else
 716 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 717 {
 718         down_write(&oldmm->mmap_sem);
 719         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 720         up_write(&oldmm->mmap_sem);
 721         return 0;
 722 }
 723 #define mm_alloc_pgd(mm)        (0)
 724 #define mm_free_pgd(mm)
 725 #endif /* CONFIG_MMU */
 726
 727 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 728
 729 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 730 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 731
 732 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 733
 734 static int __init coredump_filter_setup(char *s)
 735 {
 736         default_dump_filter =
 737                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 738                 MMF_DUMP_FILTER_MASK;
 739         return 1;
 740 }
 741
 742 __setup("coredump_filter=", coredump_filter_setup);
 743
 744 #include <linux/init_task.h>
 745
 746 static void mm_init_aio(struct mm_struct *mm)
 747 {
 748 #ifdef CONFIG_AIO
 749         spin_lock_init(&mm->ioctx_lock);
 750         mm->ioctx_table = NULL;
 751 #endif
 752 }
 753
 754 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 755 {
 756 #ifdef CONFIG_MEMCG
 757         mm->owner = p;
 758 #endif
 759 }
 760
 761 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 762         struct user_namespace *user_ns)
 763 {
 764         mm->mmap = NULL;
 765         mm->mm_rb = RB_ROOT;
 766         mm->vmacache_seqnum = 0;
 767         atomic_set(&mm->mm_users, 1);
 768         atomic_set(&mm->mm_count, 1);
 769         init_rwsem(&mm->mmap_sem);
 770         INIT_LIST_HEAD(&mm->mmlist);
 771         mm->core_state = NULL;
 772         atomic_long_set(&mm->nr_ptes, 0);
 773         mm_nr_pmds_init(mm);
 774         mm->map_count = 0;
 775         mm->locked_vm = 0;
 776         mm->pinned_vm = 0;
 777         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 778         spin_lock_init(&mm->page_table_lock);
 779         mm_init_cpumask(mm);
 780         mm_init_aio(mm);
 781         mm_init_owner(mm, p);
 782         mmu_notifier_mm_init(mm);
 783         clear_tlb_flush_pending(mm);
 784 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 785         mm->pmd_huge_pte = NULL;
 786 #endif
 787
 788         if (current->mm) {
 789                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 790                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 791         } else {
 792                 mm->flags = default_dump_filter;
 793                 mm->def_flags = 0;
 794         }
 795
 796         if (mm_alloc_pgd(mm))
 797                 goto fail_nopgd;
 798
 799         if (init_new_context(p, mm))
 800                 goto fail_nocontext;
 801
 802         mm->user_ns = get_user_ns(user_ns);
 803         return mm;
 804
 805 fail_nocontext:
 806         mm_free_pgd(mm);
 807 fail_nopgd:
 808         free_mm(mm);
 809         return NULL;
 810 }
 811
 812 static void check_mm(struct mm_struct *mm)
 813 {
 814         int i;
 815
 816         for (i = 0; i < NR_MM_COUNTERS; i++) {
 817                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 818
 819                 if (unlikely(x))
 820                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 821                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 822         }
 823
 824         if (atomic_long_read(&mm->nr_ptes))
 825                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 826                                 atomic_long_read(&mm->nr_ptes));
 827         if (mm_nr_pmds(mm))
 828                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 829                                 mm_nr_pmds(mm));
 830
 831 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 832         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 833 #endif
 834 }
 835
 836 /*
 837  * Allocate and initialize an mm_struct.
 838  */
 839 struct mm_struct *mm_alloc(void)
 840 {
 841         struct mm_struct *mm;
 842
 843         mm = allocate_mm();
 844         if (!mm)
 845                 return NULL;
 846
 847         memset(mm, 0, sizeof(*mm));
 848         return mm_init(mm, current, current_user_ns());
 849 }
 850
 851 /*
 852  * Called when the last reference to the mm
 853  * is dropped: either by a lazy thread or by
 854  * mmput. Free the page directory and the mm.
 855  */
 856 void __mmdrop(struct mm_struct *mm)
 857 {
 858         BUG_ON(mm == &init_mm);
 859         mm_free_pgd(mm);
 860         destroy_context(mm);
 861         mmu_notifier_mm_destroy(mm);
 862         check_mm(mm);
 863         put_user_ns(mm->user_ns);
 864         free_mm(mm);
 865 }
 866 EXPORT_SYMBOL_GPL(__mmdrop);
 867
 868 static inline void __mmput(struct mm_struct *mm)
 869 {
 870         VM_BUG_ON(atomic_read(&mm->mm_users));
 871
 872         uprobe_clear_state(mm);
 873         exit_aio(mm);
 874         ksm_exit(mm);
 875         khugepaged_exit(mm); /* must run before exit_mmap */
 876         exit_mmap(mm);
 877         mm_put_huge_zero_page(mm);
 878         set_mm_exe_file(mm, NULL);
 879         if (!list_empty(&mm->mmlist)) {
 880                 spin_lock(&mmlist_lock);
 881                 list_del(&mm->mmlist);
 882                 spin_unlock(&mmlist_lock);
 883         }
 884         if (mm->binfmt)
 885                 module_put(mm->binfmt->module);
 886         set_bit(MMF_OOM_SKIP, &mm->flags);
 887         mmdrop(mm);
 888 }
 889
 890 /*
 891  * Decrement the use count and release all resources for an mm.
 892  */
 893 void mmput(struct mm_struct *mm)
 894 {
 895         might_sleep();
 896
 897         if (atomic_dec_and_test(&mm->mm_users))
 898                 __mmput(mm);
 899 }
 900 EXPORT_SYMBOL_GPL(mmput);
 901
 902 #ifdef CONFIG_MMU
 903 static void mmput_async_fn(struct work_struct *work)
 904 {
 905         struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
 906         __mmput(mm);
 907 }
 908
 909 void mmput_async(struct mm_struct *mm)
 910 {
 911         if (atomic_dec_and_test(&mm->mm_users)) {
 912                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
 913                 schedule_work(&mm->async_put_work);
 914         }
 915 }
 916 #endif
 917
 918 /**
 919  * set_mm_exe_file - change a reference to the mm's executable file
 920  *
 921  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 922  *
 923  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 924  * invocations: in mmput() nobody alive left, in execve task is single
 925  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 926  * mm->exe_file, but does so without using set_mm_exe_file() in order
 927  * to do avoid the need for any locks.
 928  */
 929 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 930 {
 931         struct file *old_exe_file;
 932
 933         /*
 934          * It is safe to dereference the exe_file without RCU as
 935          * this function is only called if nobody else can access
 936          * this mm -- see comment above for justification.
 937          */
 938         old_exe_file = rcu_dereference_raw(mm->exe_file);
 939
 940         if (new_exe_file)
 941                 get_file(new_exe_file);
 942         rcu_assign_pointer(mm->exe_file, new_exe_file);
 943         if (old_exe_file)
 944                 fput(old_exe_file);
 945 }
 946
 947 /**
 948  * get_mm_exe_file - acquire a reference to the mm's executable file
 949  *
 950  * Returns %NULL if mm has no associated executable file.
 951  * User must release file via fput().
 952  */
 953 struct file *get_mm_exe_file(struct mm_struct *mm)
 954 {
 955         struct file *exe_file;
 956
 957         rcu_read_lock();
 958         exe_file = rcu_dereference(mm->exe_file);
 959         if (exe_file && !get_file_rcu(exe_file))
 960                 exe_file = NULL;
 961         rcu_read_unlock();
 962         return exe_file;
 963 }
 964 EXPORT_SYMBOL(get_mm_exe_file);
 965
 966 /**
 967  * get_task_exe_file - acquire a reference to the task's executable file
 968  *
 969  * Returns %NULL if task's mm (if any) has no associated executable file or
 970  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
 971  * User must release file via fput().
 972  */
 973 struct file *get_task_exe_file(struct task_struct *task)
 974 {
 975         struct file *exe_file = NULL;
 976         struct mm_struct *mm;
 977
 978         task_lock(task);
 979         mm = task->mm;
 980         if (mm) {
 981                 if (!(task->flags & PF_KTHREAD))
 982                         exe_file = get_mm_exe_file(mm);
 983         }
 984         task_unlock(task);
 985         return exe_file;
 986 }
 987 EXPORT_SYMBOL(get_task_exe_file);
 988
 989 /**
 990  * get_task_mm - acquire a reference to the task's mm
 991  *
 992  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 993  * this kernel workthread has transiently adopted a user mm with use_mm,
 994  * to do its AIO) is not set and if so returns a reference to it, after
 995  * bumping up the use count.  User must release the mm via mmput()
 996  * after use.  Typically used by /proc and ptrace.
 997  */
 998 struct mm_struct *get_task_mm(struct task_struct *task)
 999 {
1000         struct mm_struct *mm;
1001
1002         task_lock(task);
1003         mm = task->mm;
1004         if (mm) {
1005                 if (task->flags & PF_KTHREAD)
1006                         mm = NULL;
1007                 else
1008                         mmget(mm);
1009         }
1010         task_unlock(task);
1011         return mm;
1012 }
1013 EXPORT_SYMBOL_GPL(get_task_mm);
1014
1015 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1016 {
1017         struct mm_struct *mm;
1018         int err;
1019
1020         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1021         if (err)
1022                 return ERR_PTR(err);
1023
1024         mm = get_task_mm(task);
1025         if (mm && mm != current->mm &&
1026                         !ptrace_may_access(task, mode)) {
1027                 mmput(mm);
1028                 mm = ERR_PTR(-EACCES);
1029         }
1030         mutex_unlock(&task->signal->cred_guard_mutex);
1031
1032         return mm;
1033 }
1034
1035 static void complete_vfork_done(struct task_struct *tsk)
1036 {
1037         struct completion *vfork;
1038
1039         task_lock(tsk);
1040         vfork = tsk->vfork_done;
1041         if (likely(vfork)) {
1042                 tsk->vfork_done = NULL;
1043                 complete(vfork);
1044         }
1045         task_unlock(tsk);
1046 }
1047
1048 static int wait_for_vfork_done(struct task_struct *child,
1049                                 struct completion *vfork)
1050 {
1051         int killed;
1052
1053         freezer_do_not_count();
1054         killed = wait_for_completion_killable(vfork);
1055         freezer_count();
1056
1057         if (killed) {
1058                 task_lock(child);
1059                 child->vfork_done = NULL;
1060                 task_unlock(child);
1061         }
1062
1063         put_task_struct(child);
1064         return killed;
1065 }
1066
1067 /* Please note the differences between mmput and mm_release.
1068  * mmput is called whenever we stop holding onto a mm_struct,
1069  * error success whatever.
1070  *
1071  * mm_release is called after a mm_struct has been removed
1072  * from the current process.
1073  *
1074  * This difference is important for error handling, when we
1075  * only half set up a mm_struct for a new process and need to restore
1076  * the old one.  Because we mmput the new mm_struct before
1077  * restoring the old one. . .
1078  * Eric Biederman 10 January 1998
1079  */
1080 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1081 {
1082         /* Get rid of any futexes when releasing the mm */
1083 #ifdef CONFIG_FUTEX
1084         if (unlikely(tsk->robust_list)) {
1085                 exit_robust_list(tsk);
1086                 tsk->robust_list = NULL;
1087         }
1088 #ifdef CONFIG_COMPAT
1089         if (unlikely(tsk->compat_robust_list)) {
1090                 compat_exit_robust_list(tsk);
1091                 tsk->compat_robust_list = NULL;
1092         }
1093 #endif
1094         if (unlikely(!list_empty(&tsk->pi_state_list)))
1095                 exit_pi_state_list(tsk);
1096 #endif
1097
1098         uprobe_free_utask(tsk);
1099
1100         /* Get rid of any cached register state */
1101         deactivate_mm(tsk, mm);
1102
1103         /*
1104          * Signal userspace if we're not exiting with a core dump
1105          * because we want to leave the value intact for debugging
1106          * purposes.
1107          */
1108         if (tsk->clear_child_tid) {
1109                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1110                     atomic_read(&mm->mm_users) > 1) {
1111                         /*
1112                          * We don't check the error code - if userspace has
1113                          * not set up a proper pointer then tough luck.
1114                          */
1115                         put_user(0, tsk->clear_child_tid);
1116                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1117                                         1, NULL, NULL, 0);
1118                 }
1119                 tsk->clear_child_tid = NULL;
1120         }
1121
1122         /*
1123          * All done, finally we can wake up parent and return this mm to him.
1124          * Also kthread_stop() uses this completion for synchronization.
1125          */
1126         if (tsk->vfork_done)
1127                 complete_vfork_done(tsk);
1128 }
1129
1130 /*
1131  * Allocate a new mm structure and copy contents from the
1132  * mm structure of the passed in task structure.
1133  */
1134 static struct mm_struct *dup_mm(struct task_struct *tsk)
1135 {
1136         struct mm_struct *mm, *oldmm = current->mm;
1137         int err;
1138
1139         mm = allocate_mm();
1140         if (!mm)
1141                 goto fail_nomem;
1142
1143         memcpy(mm, oldmm, sizeof(*mm));
1144
1145         if (!mm_init(mm, tsk, mm->user_ns))
1146                 goto fail_nomem;
1147
1148         err = dup_mmap(mm, oldmm);
1149         if (err)
1150                 goto free_pt;
1151
1152         mm->hiwater_rss = get_mm_rss(mm);
1153         mm->hiwater_vm = mm->total_vm;
1154
1155         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1156                 goto free_pt;
1157
1158         return mm;
1159
1160 free_pt:
1161         /* don't put binfmt in mmput, we haven't got module yet */
1162         mm->binfmt = NULL;
1163         mmput(mm);
1164
1165 fail_nomem:
1166         return NULL;
1167 }
1168
1169 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1170 {
1171         struct mm_struct *mm, *oldmm;
1172         int retval;
1173
1174         tsk->min_flt = tsk->maj_flt = 0;
1175         tsk->nvcsw = tsk->nivcsw = 0;
1176 #ifdef CONFIG_DETECT_HUNG_TASK
1177         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1178 #endif
1179
1180         tsk->mm = NULL;
1181         tsk->active_mm = NULL;
1182
1183         /*
1184          * Are we cloning a kernel thread?
1185          *
1186          * We need to steal a active VM for that..
1187          */
1188         oldmm = current->mm;
1189         if (!oldmm)
1190                 return 0;
1191
1192         /* initialize the new vmacache entries */
1193         vmacache_flush(tsk);
1194
1195         if (clone_flags & CLONE_VM) {
1196                 mmget(oldmm);
1197                 mm = oldmm;
1198                 goto good_mm;
1199         }
1200
1201         retval = -ENOMEM;
1202         mm = dup_mm(tsk);
1203         if (!mm)
1204                 goto fail_nomem;
1205
1206 good_mm:
1207         tsk->mm = mm;
1208         tsk->active_mm = mm;
1209         return 0;
1210
1211 fail_nomem:
1212         return retval;
1213 }
1214
1215 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1216 {
1217         struct fs_struct *fs = current->fs;
1218         if (clone_flags & CLONE_FS) {
1219                 /* tsk->fs is already what we want */
1220                 spin_lock(&fs->lock);
1221                 if (fs->in_exec) {
1222                         spin_unlock(&fs->lock);
1223                         return -EAGAIN;
1224                 }
1225                 fs->users++;
1226                 spin_unlock(&fs->lock);
1227                 return 0;
1228         }
1229         tsk->fs = copy_fs_struct(fs);
1230         if (!tsk->fs)
1231                 return -ENOMEM;
1232         return 0;
1233 }
1234
1235 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1236 {
1237         struct files_struct *oldf, *newf;
1238         int error = 0;
1239
1240         /*
1241          * A background process may not have any files ...
1242          */
1243         oldf = current->files;
1244         if (!oldf)
1245                 goto out;
1246
1247         if (clone_flags & CLONE_FILES) {
1248                 atomic_inc(&oldf->count);
1249                 goto out;
1250         }
1251
1252         newf = dup_fd(oldf, &error);
1253         if (!newf)
1254                 goto out;
1255
1256         tsk->files = newf;
1257         error = 0;
1258 out:
1259         return error;
1260 }
1261
1262 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1263 {
1264 #ifdef CONFIG_BLOCK
1265         struct io_context *ioc = current->io_context;
1266         struct io_context *new_ioc;
1267
1268         if (!ioc)
1269                 return 0;
1270         /*
1271          * Share io context with parent, if CLONE_IO is set
1272          */
1273         if (clone_flags & CLONE_IO) {
1274                 ioc_task_link(ioc);
1275                 tsk->io_context = ioc;
1276         } else if (ioprio_valid(ioc->ioprio)) {
1277                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1278                 if (unlikely(!new_ioc))
1279                         return -ENOMEM;
1280
1281                 new_ioc->ioprio = ioc->ioprio;
1282                 put_io_context(new_ioc);
1283         }
1284 #endif
1285         return 0;
1286 }
1287
1288 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1289 {
1290         struct sighand_struct *sig;
1291
1292         if (clone_flags & CLONE_SIGHAND) {
1293                 atomic_inc(&current->sighand->count);
1294                 return 0;
1295         }
1296         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1297         rcu_assign_pointer(tsk->sighand, sig);
1298         if (!sig)
1299                 return -ENOMEM;
1300
1301         atomic_set(&sig->count, 1);
1302         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1303         return 0;
1304 }
1305
1306 void __cleanup_sighand(struct sighand_struct *sighand)
1307 {
1308         if (atomic_dec_and_test(&sighand->count)) {
1309                 signalfd_cleanup(sighand);
1310                 /*
1311                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1312                  * without an RCU grace period, see __lock_task_sighand().
1313                  */
1314                 kmem_cache_free(sighand_cachep, sighand);
1315         }
1316 }
1317
1318 #ifdef CONFIG_POSIX_TIMERS
1319 /*
1320  * Initialize POSIX timer handling for a thread group.
1321  */
1322 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1323 {
1324         unsigned long cpu_limit;
1325
1326         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1327         if (cpu_limit != RLIM_INFINITY) {
1328                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1329                 sig->cputimer.running = true;
1330         }
1331
1332         /* The timer lists. */
1333         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1334         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1335         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1336 }
1337 #else
1338 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1339 #endif
1340
1341 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1342 {
1343         struct signal_struct *sig;
1344
1345         if (clone_flags & CLONE_THREAD)
1346                 return 0;
1347
1348         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1349         tsk->signal = sig;
1350         if (!sig)
1351                 return -ENOMEM;
1352
1353         sig->nr_threads = 1;
1354         atomic_set(&sig->live, 1);
1355         atomic_set(&sig->sigcnt, 1);
1356
1357         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1358         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1359         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1360
1361         init_waitqueue_head(&sig->wait_chldexit);
1362         sig->curr_target = tsk;
1363         init_sigpending(&sig->shared_pending);
1364         seqlock_init(&sig->stats_lock);
1365         prev_cputime_init(&sig->prev_cputime);
1366
1367 #ifdef CONFIG_POSIX_TIMERS
1368         INIT_LIST_HEAD(&sig->posix_timers);
1369         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1370         sig->real_timer.function = it_real_fn;
1371 #endif
1372
1373         task_lock(current->group_leader);
1374         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1375         task_unlock(current->group_leader);
1376
1377         posix_cpu_timers_init_group(sig);
1378
1379         tty_audit_fork(sig);
1380         sched_autogroup_fork(sig);
1381
1382         sig->oom_score_adj = current->signal->oom_score_adj;
1383         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1384
1385         mutex_init(&sig->cred_guard_mutex);
1386
1387         return 0;
1388 }
1389
1390 static void copy_seccomp(struct task_struct *p)
1391 {
1392 #ifdef CONFIG_SECCOMP
1393         /*
1394          * Must be called with sighand->lock held, which is common to
1395          * all threads in the group. Holding cred_guard_mutex is not
1396          * needed because this new task is not yet running and cannot
1397          * be racing exec.
1398          */
1399         assert_spin_locked(&current->sighand->siglock);
1400
1401         /* Ref-count the new filter user, and assign it. */
1402         get_seccomp_filter(current);
1403         p->seccomp = current->seccomp;
1404
1405         /*
1406          * Explicitly enable no_new_privs here in case it got set
1407          * between the task_struct being duplicated and holding the
1408          * sighand lock. The seccomp state and nnp must be in sync.
1409          */
1410         if (task_no_new_privs(current))
1411                 task_set_no_new_privs(p);
1412
1413         /*
1414          * If the parent gained a seccomp mode after copying thread
1415          * flags and between before we held the sighand lock, we have
1416          * to manually enable the seccomp thread flag here.
1417          */
1418         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1419                 set_tsk_thread_flag(p, TIF_SECCOMP);
1420 #endif
1421 }
1422
1423 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1424 {
1425         current->clear_child_tid = tidptr;
1426
1427         return task_pid_vnr(current);
1428 }
1429
1430 static void rt_mutex_init_task(struct task_struct *p)
1431 {
1432         raw_spin_lock_init(&p->pi_lock);
1433 #ifdef CONFIG_RT_MUTEXES
1434         p->pi_waiters = RB_ROOT;
1435         p->pi_waiters_leftmost = NULL;
1436         p->pi_blocked_on = NULL;
1437 #endif
1438 }
1439
1440 #ifdef CONFIG_POSIX_TIMERS
1441 /*
1442  * Initialize POSIX timer handling for a single task.
1443  */
1444 static void posix_cpu_timers_init(struct task_struct *tsk)
1445 {
1446         tsk->cputime_expires.prof_exp = 0;
1447         tsk->cputime_expires.virt_exp = 0;
1448         tsk->cputime_expires.sched_exp = 0;
1449         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1450         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1451         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1452 }
1453 #else
1454 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1455 #endif
1456
1457 static inline void
1458 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1459 {
1460          task->pids[type].pid = pid;
1461 }
1462
1463 /*
1464  * This creates a new process as a copy of the old one,
1465  * but does not actually start it yet.
1466  *
1467  * It copies the registers, and all the appropriate
1468  * parts of the process environment (as per the clone
1469  * flags). The actual kick-off is left to the caller.
1470  */
1471 static __latent_entropy struct task_struct *copy_process(
1472                                         unsigned long clone_flags,
1473                                         unsigned long stack_start,
1474                                         unsigned long stack_size,
1475                                         int __user *child_tidptr,
1476                                         struct pid *pid,
1477                                         int trace,
1478                                         unsigned long tls,
1479                                         int node)
1480 {
1481         int retval;
1482         struct task_struct *p;
1483
1484         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1485                 return ERR_PTR(-EINVAL);
1486
1487         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1488                 return ERR_PTR(-EINVAL);
1489
1490         /*
1491          * Thread groups must share signals as well, and detached threads
1492          * can only be started up within the thread group.
1493          */
1494         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1495                 return ERR_PTR(-EINVAL);
1496
1497         /*
1498          * Shared signal handlers imply shared VM. By way of the above,
1499          * thread groups also imply shared VM. Blocking this case allows
1500          * for various simplifications in other code.
1501          */
1502         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1503                 return ERR_PTR(-EINVAL);
1504
1505         /*
1506          * Siblings of global init remain as zombies on exit since they are
1507          * not reaped by their parent (swapper). To solve this and to avoid
1508          * multi-rooted process trees, prevent global and container-inits
1509          * from creating siblings.
1510          */
1511         if ((clone_flags & CLONE_PARENT) &&
1512                                 current->signal->flags & SIGNAL_UNKILLABLE)
1513                 return ERR_PTR(-EINVAL);
1514
1515         /*
1516          * If the new process will be in a different pid or user namespace
1517          * do not allow it to share a thread group with the forking task.
1518          */
1519         if (clone_flags & CLONE_THREAD) {
1520                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1521                     (task_active_pid_ns(current) !=
1522                                 current->nsproxy->pid_ns_for_children))
1523                         return ERR_PTR(-EINVAL);
1524         }
1525
1526         retval = security_task_create(clone_flags);
1527         if (retval)
1528                 goto fork_out;
1529
1530         retval = -ENOMEM;
1531         p = dup_task_struct(current, node);
1532         if (!p)
1533                 goto fork_out;
1534
1535         ftrace_graph_init_task(p);
1536
1537         rt_mutex_init_task(p);
1538
1539 #ifdef CONFIG_PROVE_LOCKING
1540         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1541         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1542 #endif
1543         retval = -EAGAIN;
1544         if (atomic_read(&p->real_cred->user->processes) >=
1545                         task_rlimit(p, RLIMIT_NPROC)) {
1546                 if (p->real_cred->user != INIT_USER &&
1547                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1548                         goto bad_fork_free;
1549         }
1550         current->flags &= ~PF_NPROC_EXCEEDED;
1551
1552         retval = copy_creds(p, clone_flags);
1553         if (retval < 0)
1554                 goto bad_fork_free;
1555
1556         /*
1557          * If multiple threads are within copy_process(), then this check
1558          * triggers too late. This doesn't hurt, the check is only there
1559          * to stop root fork bombs.
1560          */
1561         retval = -EAGAIN;
1562         if (nr_threads >= max_threads)
1563                 goto bad_fork_cleanup_count;
1564
1565         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1566         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1567         p->flags |= PF_FORKNOEXEC;
1568         INIT_LIST_HEAD(&p->children);
1569         INIT_LIST_HEAD(&p->sibling);
1570         rcu_copy_process(p);
1571         p->vfork_done = NULL;
1572         spin_lock_init(&p->alloc_lock);
1573
1574         init_sigpending(&p->pending);
1575
1576         p->utime = p->stime = p->gtime = 0;
1577 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1578         p->utimescaled = p->stimescaled = 0;
1579 #endif
1580         prev_cputime_init(&p->prev_cputime);
1581
1582 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1583         seqcount_init(&p->vtime_seqcount);
1584         p->vtime_snap = 0;
1585         p->vtime_snap_whence = VTIME_INACTIVE;
1586 #endif
1587
1588 #if defined(SPLIT_RSS_COUNTING)
1589         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1590 #endif
1591
1592         p->default_timer_slack_ns = current->timer_slack_ns;
1593
1594         task_io_accounting_init(&p->ioac);
1595         acct_clear_integrals(p);
1596
1597         posix_cpu_timers_init(p);
1598
1599         p->start_time = ktime_get_ns();
1600         p->real_start_time = ktime_get_boot_ns();
1601         p->io_context = NULL;
1602         p->audit_context = NULL;
1603         cgroup_fork(p);
1604 #ifdef CONFIG_NUMA
1605         p->mempolicy = mpol_dup(p->mempolicy);
1606         if (IS_ERR(p->mempolicy)) {
1607                 retval = PTR_ERR(p->mempolicy);
1608                 p->mempolicy = NULL;
1609                 goto bad_fork_cleanup_threadgroup_lock;
1610         }
1611 #endif
1612 #ifdef CONFIG_CPUSETS
1613         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1614         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1615         seqcount_init(&p->mems_allowed_seq);
1616 #endif
1617 #ifdef CONFIG_TRACE_IRQFLAGS
1618         p->irq_events = 0;
1619         p->hardirqs_enabled = 0;
1620         p->hardirq_enable_ip = 0;
1621         p->hardirq_enable_event = 0;
1622         p->hardirq_disable_ip = _THIS_IP_;
1623         p->hardirq_disable_event = 0;
1624         p->softirqs_enabled = 1;
1625         p->softirq_enable_ip = _THIS_IP_;
1626         p->softirq_enable_event = 0;
1627         p->softirq_disable_ip = 0;
1628         p->softirq_disable_event = 0;
1629         p->hardirq_context = 0;
1630         p->softirq_context = 0;
1631 #endif
1632
1633         p->pagefault_disabled = 0;
1634
1635 #ifdef CONFIG_LOCKDEP
1636         p->lockdep_depth = 0; /* no locks held yet */
1637         p->curr_chain_key = 0;
1638         p->lockdep_recursion = 0;
1639 #endif
1640
1641 #ifdef CONFIG_DEBUG_MUTEXES
1642         p->blocked_on = NULL; /* not blocked yet */
1643 #endif
1644 #ifdef CONFIG_BCACHE
1645         p->sequential_io        = 0;
1646         p->sequential_io_avg    = 0;
1647 #endif
1648
1649         /* Perform scheduler related setup. Assign this task to a CPU. */
1650         retval = sched_fork(clone_flags, p);
1651         if (retval)
1652                 goto bad_fork_cleanup_policy;
1653
1654         retval = perf_event_init_task(p);
1655         if (retval)
1656                 goto bad_fork_cleanup_policy;
1657         retval = audit_alloc(p);
1658         if (retval)
1659                 goto bad_fork_cleanup_perf;
1660         /* copy all the process information */
1661         shm_init_task(p);
1662         retval = copy_semundo(clone_flags, p);
1663         if (retval)
1664                 goto bad_fork_cleanup_audit;
1665         retval = copy_files(clone_flags, p);
1666         if (retval)
1667                 goto bad_fork_cleanup_semundo;
1668         retval = copy_fs(clone_flags, p);
1669         if (retval)
1670                 goto bad_fork_cleanup_files;
1671         retval = copy_sighand(clone_flags, p);
1672         if (retval)
1673                 goto bad_fork_cleanup_fs;
1674         retval = copy_signal(clone_flags, p);
1675         if (retval)
1676                 goto bad_fork_cleanup_sighand;
1677         retval = copy_mm(clone_flags, p);
1678         if (retval)
1679                 goto bad_fork_cleanup_signal;
1680         retval = copy_namespaces(clone_flags, p);
1681         if (retval)
1682                 goto bad_fork_cleanup_mm;
1683         retval = copy_io(clone_flags, p);
1684         if (retval)
1685                 goto bad_fork_cleanup_namespaces;
1686         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1687         if (retval)
1688                 goto bad_fork_cleanup_io;
1689
1690         if (pid != &init_struct_pid) {
1691                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1692                 if (IS_ERR(pid)) {
1693                         retval = PTR_ERR(pid);
1694                         goto bad_fork_cleanup_thread;
1695                 }
1696         }
1697
1698         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1699         /*
1700          * Clear TID on mm_release()?
1701          */
1702         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1703 #ifdef CONFIG_BLOCK
1704         p->plug = NULL;
1705 #endif
1706 #ifdef CONFIG_FUTEX
1707         p->robust_list = NULL;
1708 #ifdef CONFIG_COMPAT
1709         p->compat_robust_list = NULL;
1710 #endif
1711         INIT_LIST_HEAD(&p->pi_state_list);
1712         p->pi_state_cache = NULL;
1713 #endif
1714         /*
1715          * sigaltstack should be cleared when sharing the same VM
1716          */
1717         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1718                 sas_ss_reset(p);
1719
1720         /*
1721          * Syscall tracing and stepping should be turned off in the
1722          * child regardless of CLONE_PTRACE.
1723          */
1724         user_disable_single_step(p);
1725         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1726 #ifdef TIF_SYSCALL_EMU
1727         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1728 #endif
1729         clear_all_latency_tracing(p);
1730
1731         /* ok, now we should be set up.. */
1732         p->pid = pid_nr(pid);
1733         if (clone_flags & CLONE_THREAD) {
1734                 p->exit_signal = -1;
1735                 p->group_leader = current->group_leader;
1736                 p->tgid = current->tgid;
1737         } else {
1738                 if (clone_flags & CLONE_PARENT)
1739                         p->exit_signal = current->group_leader->exit_signal;
1740                 else
1741                         p->exit_signal = (clone_flags & CSIGNAL);
1742                 p->group_leader = p;
1743                 p->tgid = p->pid;
1744         }
1745
1746         p->nr_dirtied = 0;
1747         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1748         p->dirty_paused_when = 0;
1749
1750         p->pdeath_signal = 0;
1751         INIT_LIST_HEAD(&p->thread_group);
1752         p->task_works = NULL;
1753
1754         cgroup_threadgroup_change_begin(current);
1755         /*
1756          * Ensure that the cgroup subsystem policies allow the new process to be
1757          * forked. It should be noted the the new process's css_set can be changed
1758          * between here and cgroup_post_fork() if an organisation operation is in
1759          * progress.
1760          */
1761         retval = cgroup_can_fork(p);
1762         if (retval)
1763                 goto bad_fork_free_pid;
1764
1765         /*
1766          * Make it visible to the rest of the system, but dont wake it up yet.
1767          * Need tasklist lock for parent etc handling!
1768          */
1769         write_lock_irq(&tasklist_lock);
1770
1771         /* CLONE_PARENT re-uses the old parent */
1772         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1773                 p->real_parent = current->real_parent;
1774                 p->parent_exec_id = current->parent_exec_id;
1775         } else {
1776                 p->real_parent = current;
1777                 p->parent_exec_id = current->self_exec_id;
1778         }
1779
1780         spin_lock(&current->sighand->siglock);
1781
1782         /*
1783          * Copy seccomp details explicitly here, in case they were changed
1784          * before holding sighand lock.
1785          */
1786         copy_seccomp(p);
1787
1788         /*
1789          * Process group and session signals need to be delivered to just the
1790          * parent before the fork or both the parent and the child after the
1791          * fork. Restart if a signal comes in before we add the new process to
1792          * it's process group.
1793          * A fatal signal pending means that current will exit, so the new
1794          * thread can't slip out of an OOM kill (or normal SIGKILL).
1795         */
1796         recalc_sigpending();
1797         if (signal_pending(current)) {
1798                 spin_unlock(&current->sighand->siglock);
1799                 write_unlock_irq(&tasklist_lock);
1800                 retval = -ERESTARTNOINTR;
1801                 goto bad_fork_cancel_cgroup;
1802         }
1803
1804         if (likely(p->pid)) {
1805                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1806
1807                 init_task_pid(p, PIDTYPE_PID, pid);
1808                 if (thread_group_leader(p)) {
1809                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1810                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1811
1812                         if (is_child_reaper(pid)) {
1813                                 ns_of_pid(pid)->child_reaper = p;
1814                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1815                         }
1816
1817                         p->signal->leader_pid = pid;
1818                         p->signal->tty = tty_kref_get(current->signal->tty);
1819                         /*
1820                          * Inherit has_child_subreaper flag under the same
1821                          * tasklist_lock with adding child to the process tree
1822                          * for propagate_has_child_subreaper optimization.
1823                          */
1824                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1825                                                          p->real_parent->signal->is_child_subreaper;
1826                         list_add_tail(&p->sibling, &p->real_parent->children);
1827                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1828                         attach_pid(p, PIDTYPE_PGID);
1829                         attach_pid(p, PIDTYPE_SID);
1830                         __this_cpu_inc(process_counts);
1831                 } else {
1832                         current->signal->nr_threads++;
1833                         atomic_inc(&current->signal->live);
1834                         atomic_inc(&current->signal->sigcnt);
1835                         list_add_tail_rcu(&p->thread_group,
1836                                           &p->group_leader->thread_group);
1837                         list_add_tail_rcu(&p->thread_node,
1838                                           &p->signal->thread_head);
1839                 }
1840                 attach_pid(p, PIDTYPE_PID);
1841                 nr_threads++;
1842         }
1843
1844         total_forks++;
1845         spin_unlock(&current->sighand->siglock);
1846         syscall_tracepoint_update(p);
1847         write_unlock_irq(&tasklist_lock);
1848
1849         proc_fork_connector(p);
1850         cgroup_post_fork(p);
1851         cgroup_threadgroup_change_end(current);
1852         perf_event_fork(p);
1853
1854         trace_task_newtask(p, clone_flags);
1855         uprobe_copy_process(p, clone_flags);
1856
1857         return p;
1858
1859 bad_fork_cancel_cgroup:
1860         cgroup_cancel_fork(p);
1861 bad_fork_free_pid:
1862         cgroup_threadgroup_change_end(current);
1863         if (pid != &init_struct_pid)
1864                 free_pid(pid);
1865 bad_fork_cleanup_thread:
1866         exit_thread(p);
1867 bad_fork_cleanup_io:
1868         if (p->io_context)
1869                 exit_io_context(p);
1870 bad_fork_cleanup_namespaces:
1871         exit_task_namespaces(p);
1872 bad_fork_cleanup_mm:
1873         if (p->mm)
1874                 mmput(p->mm);
1875 bad_fork_cleanup_signal:
1876         if (!(clone_flags & CLONE_THREAD))
1877                 free_signal_struct(p->signal);
1878 bad_fork_cleanup_sighand:
1879         __cleanup_sighand(p->sighand);
1880 bad_fork_cleanup_fs:
1881         exit_fs(p); /* blocking */
1882 bad_fork_cleanup_files:
1883         exit_files(p); /* blocking */
1884 bad_fork_cleanup_semundo:
1885         exit_sem(p);
1886 bad_fork_cleanup_audit:
1887         audit_free(p);
1888 bad_fork_cleanup_perf:
1889         perf_event_free_task(p);
1890 bad_fork_cleanup_policy:
1891 #ifdef CONFIG_NUMA
1892         mpol_put(p->mempolicy);
1893 bad_fork_cleanup_threadgroup_lock:
1894 #endif
1895         delayacct_tsk_free(p);
1896 bad_fork_cleanup_count:
1897         atomic_dec(&p->cred->user->processes);
1898         exit_creds(p);
1899 bad_fork_free:
1900         p->state = TASK_DEAD;
1901         put_task_stack(p);
1902         free_task(p);
1903 fork_out:
1904         return ERR_PTR(retval);
1905 }
1906
1907 static inline void init_idle_pids(struct pid_link *links)
1908 {
1909         enum pid_type type;
1910
1911         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1912                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1913                 links[type].pid = &init_struct_pid;
1914         }
1915 }
1916
1917 struct task_struct *fork_idle(int cpu)
1918 {
1919         struct task_struct *task;
1920         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1921                             cpu_to_node(cpu));
1922         if (!IS_ERR(task)) {
1923                 init_idle_pids(task->pids);
1924                 init_idle(task, cpu);
1925         }
1926
1927         return task;
1928 }
1929
1930 /*
1931  *  Ok, this is the main fork-routine.
1932  *
1933  * It copies the process, and if successful kick-starts
1934  * it and waits for it to finish using the VM if required.
1935  */
1936 long _do_fork(unsigned long clone_flags,
1937               unsigned long stack_start,
1938               unsigned long stack_size,
1939               int __user *parent_tidptr,
1940               int __user *child_tidptr,
1941               unsigned long tls)
1942 {
1943         struct task_struct *p;
1944         int trace = 0;
1945         long nr;
1946
1947         /*
1948          * Determine whether and which event to report to ptracer.  When
1949          * called from kernel_thread or CLONE_UNTRACED is explicitly
1950          * requested, no event is reported; otherwise, report if the event
1951          * for the type of forking is enabled.
1952          */
1953         if (!(clone_flags & CLONE_UNTRACED)) {
1954                 if (clone_flags & CLONE_VFORK)
1955                         trace = PTRACE_EVENT_VFORK;
1956                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1957                         trace = PTRACE_EVENT_CLONE;
1958                 else
1959                         trace = PTRACE_EVENT_FORK;
1960
1961                 if (likely(!ptrace_event_enabled(current, trace)))
1962                         trace = 0;
1963         }
1964
1965         p = copy_process(clone_flags, stack_start, stack_size,
1966                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
1967         add_latent_entropy();
1968         /*
1969          * Do this prior waking up the new thread - the thread pointer
1970          * might get invalid after that point, if the thread exits quickly.
1971          */
1972         if (!IS_ERR(p)) {
1973                 struct completion vfork;
1974                 struct pid *pid;
1975
1976                 trace_sched_process_fork(current, p);
1977
1978                 pid = get_task_pid(p, PIDTYPE_PID);
1979                 nr = pid_vnr(pid);
1980
1981                 if (clone_flags & CLONE_PARENT_SETTID)
1982                         put_user(nr, parent_tidptr);
1983
1984                 if (clone_flags & CLONE_VFORK) {
1985                         p->vfork_done = &vfork;
1986                         init_completion(&vfork);
1987                         get_task_struct(p);
1988                 }
1989
1990                 wake_up_new_task(p);
1991
1992                 /* forking complete and child started to run, tell ptracer */
1993                 if (unlikely(trace))
1994                         ptrace_event_pid(trace, pid);
1995
1996                 if (clone_flags & CLONE_VFORK) {
1997                         if (!wait_for_vfork_done(p, &vfork))
1998                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1999                 }
2000
2001                 put_pid(pid);
2002         } else {
2003                 nr = PTR_ERR(p);
2004         }
2005         return nr;
2006 }
2007
2008 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2009 /* For compatibility with architectures that call do_fork directly rather than
2010  * using the syscall entry points below. */
2011 long do_fork(unsigned long clone_flags,
2012               unsigned long stack_start,
2013               unsigned long stack_size,
2014               int __user *parent_tidptr,
2015               int __user *child_tidptr)
2016 {
2017         return _do_fork(clone_flags, stack_start, stack_size,
2018                         parent_tidptr, child_tidptr, 0);
2019 }
2020 #endif
2021
2022 /*
2023  * Create a kernel thread.
2024  */
2025 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2026 {
2027         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2028                 (unsigned long)arg, NULL, NULL, 0);
2029 }
2030
2031 #ifdef __ARCH_WANT_SYS_FORK
2032 SYSCALL_DEFINE0(fork)
2033 {
2034 #ifdef CONFIG_MMU
2035         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2036 #else
2037         /* can not support in nommu mode */
2038         return -EINVAL;
2039 #endif
2040 }
2041 #endif
2042
2043 #ifdef __ARCH_WANT_SYS_VFORK
2044 SYSCALL_DEFINE0(vfork)
2045 {
2046         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2047                         0, NULL, NULL, 0);
2048 }
2049 #endif
2050
2051 #ifdef __ARCH_WANT_SYS_CLONE
2052 #ifdef CONFIG_CLONE_BACKWARDS
2053 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2054                  int __user *, parent_tidptr,
2055                  unsigned long, tls,
2056                  int __user *, child_tidptr)
2057 #elif defined(CONFIG_CLONE_BACKWARDS2)
2058 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2059                  int __user *, parent_tidptr,
2060                  int __user *, child_tidptr,
2061                  unsigned long, tls)
2062 #elif defined(CONFIG_CLONE_BACKWARDS3)
2063 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2064                 int, stack_size,
2065                 int __user *, parent_tidptr,
2066                 int __user *, child_tidptr,
2067                 unsigned long, tls)
2068 #else
2069 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2070                  int __user *, parent_tidptr,
2071                  int __user *, child_tidptr,
2072                  unsigned long, tls)
2073 #endif
2074 {
2075         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2076 }
2077 #endif
2078
2079 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2080 {
2081         struct task_struct *leader, *parent, *child;
2082         int res;
2083
2084         read_lock(&tasklist_lock);
2085         leader = top = top->group_leader;
2086 down:
2087         for_each_thread(leader, parent) {
2088                 list_for_each_entry(child, &parent->children, sibling) {
2089                         res = visitor(child, data);
2090                         if (res) {
2091                                 if (res < 0)
2092                                         goto out;
2093                                 leader = child;
2094                                 goto down;
2095                         }
2096 up:
2097                         ;
2098                 }
2099         }
2100
2101         if (leader != top) {
2102                 child = leader;
2103                 parent = child->real_parent;
2104                 leader = parent->group_leader;
2105                 goto up;
2106         }
2107 out:
2108         read_unlock(&tasklist_lock);
2109 }
2110
2111 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2112 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2113 #endif
2114
2115 static void sighand_ctor(void *data)
2116 {
2117         struct sighand_struct *sighand = data;
2118
2119         spin_lock_init(&sighand->siglock);
2120         init_waitqueue_head(&sighand->signalfd_wqh);
2121 }
2122
2123 void __init proc_caches_init(void)
2124 {
2125         sighand_cachep = kmem_cache_create("sighand_cache",
2126                         sizeof(struct sighand_struct), 0,
2127                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
2128                         SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2129         signal_cachep = kmem_cache_create("signal_cache",
2130                         sizeof(struct signal_struct), 0,
2131                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2132                         NULL);
2133         files_cachep = kmem_cache_create("files_cache",
2134                         sizeof(struct files_struct), 0,
2135                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2136                         NULL);
2137         fs_cachep = kmem_cache_create("fs_cache",
2138                         sizeof(struct fs_struct), 0,
2139                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2140                         NULL);
2141         /*
2142          * FIXME! The "sizeof(struct mm_struct)" currently includes the
2143          * whole struct cpumask for the OFFSTACK case. We could change
2144          * this to *only* allocate as much of it as required by the
2145          * maximum number of CPU's we can ever have.  The cpumask_allocation
2146          * is at the end of the structure, exactly for that reason.
2147          */
2148         mm_cachep = kmem_cache_create("mm_struct",
2149                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2150                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2151                         NULL);
2152         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2153         mmap_init();
2154         nsproxy_cache_init();
2155 }
2156
2157 /*
2158  * Check constraints on flags passed to the unshare system call.
2159  */
2160 static int check_unshare_flags(unsigned long unshare_flags)
2161 {
2162         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2163                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2164                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2165                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2166                 return -EINVAL;
2167         /*
2168          * Not implemented, but pretend it works if there is nothing
2169          * to unshare.  Note that unsharing the address space or the
2170          * signal handlers also need to unshare the signal queues (aka
2171          * CLONE_THREAD).
2172          */
2173         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2174                 if (!thread_group_empty(current))
2175                         return -EINVAL;
2176         }
2177         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2178                 if (atomic_read(&current->sighand->count) > 1)
2179                         return -EINVAL;
2180         }
2181         if (unshare_flags & CLONE_VM) {
2182                 if (!current_is_single_threaded())
2183                         return -EINVAL;
2184         }
2185
2186         return 0;
2187 }
2188
2189 /*
2190  * Unshare the filesystem structure if it is being shared
2191  */
2192 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2193 {
2194         struct fs_struct *fs = current->fs;
2195
2196         if (!(unshare_flags & CLONE_FS) || !fs)
2197                 return 0;
2198
2199         /* don't need lock here; in the worst case we'll do useless copy */
2200         if (fs->users == 1)
2201                 return 0;
2202
2203         *new_fsp = copy_fs_struct(fs);
2204         if (!*new_fsp)
2205                 return -ENOMEM;
2206
2207         return 0;
2208 }
2209
2210 /*
2211  * Unshare file descriptor table if it is being shared
2212  */
2213 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2214 {
2215         struct files_struct *fd = current->files;
2216         int error = 0;
2217
2218         if ((unshare_flags & CLONE_FILES) &&
2219             (fd && atomic_read(&fd->count) > 1)) {
2220                 *new_fdp = dup_fd(fd, &error);
2221                 if (!*new_fdp)
2222                         return error;
2223         }
2224
2225         return 0;
2226 }
2227
2228 /*
2229  * unshare allows a process to 'unshare' part of the process
2230  * context which was originally shared using clone.  copy_*
2231  * functions used by do_fork() cannot be used here directly
2232  * because they modify an inactive task_struct that is being
2233  * constructed. Here we are modifying the current, active,
2234  * task_struct.
2235  */
2236 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2237 {
2238         struct fs_struct *fs, *new_fs = NULL;
2239         struct files_struct *fd, *new_fd = NULL;
2240         struct cred *new_cred = NULL;
2241         struct nsproxy *new_nsproxy = NULL;
2242         int do_sysvsem = 0;
2243         int err;
2244
2245         /*
2246          * If unsharing a user namespace must also unshare the thread group
2247          * and unshare the filesystem root and working directories.
2248          */
2249         if (unshare_flags & CLONE_NEWUSER)
2250                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2251         /*
2252          * If unsharing vm, must also unshare signal handlers.
2253          */
2254         if (unshare_flags & CLONE_VM)
2255                 unshare_flags |= CLONE_SIGHAND;
2256         /*
2257          * If unsharing a signal handlers, must also unshare the signal queues.
2258          */
2259         if (unshare_flags & CLONE_SIGHAND)
2260                 unshare_flags |= CLONE_THREAD;
2261         /*
2262          * If unsharing namespace, must also unshare filesystem information.
2263          */
2264         if (unshare_flags & CLONE_NEWNS)
2265                 unshare_flags |= CLONE_FS;
2266
2267         err = check_unshare_flags(unshare_flags);
2268         if (err)
2269                 goto bad_unshare_out;
2270         /*
2271          * CLONE_NEWIPC must also detach from the undolist: after switching
2272          * to a new ipc namespace, the semaphore arrays from the old
2273          * namespace are unreachable.
2274          */
2275         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2276                 do_sysvsem = 1;
2277         err = unshare_fs(unshare_flags, &new_fs);
2278         if (err)
2279                 goto bad_unshare_out;
2280         err = unshare_fd(unshare_flags, &new_fd);
2281         if (err)
2282                 goto bad_unshare_cleanup_fs;
2283         err = unshare_userns(unshare_flags, &new_cred);
2284         if (err)
2285                 goto bad_unshare_cleanup_fd;
2286         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2287                                          new_cred, new_fs);
2288         if (err)
2289                 goto bad_unshare_cleanup_cred;
2290
2291         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2292                 if (do_sysvsem) {
2293                         /*
2294                          * CLONE_SYSVSEM is equivalent to sys_exit().
2295                          */
2296                         exit_sem(current);
2297                 }
2298                 if (unshare_flags & CLONE_NEWIPC) {
2299                         /* Orphan segments in old ns (see sem above). */
2300                         exit_shm(current);
2301                         shm_init_task(current);
2302                 }
2303
2304                 if (new_nsproxy)
2305                         switch_task_namespaces(current, new_nsproxy);
2306
2307                 task_lock(current);
2308
2309                 if (new_fs) {
2310                         fs = current->fs;
2311                         spin_lock(&fs->lock);
2312                         current->fs = new_fs;
2313                         if (--fs->users)
2314                                 new_fs = NULL;
2315                         else
2316                                 new_fs = fs;
2317                         spin_unlock(&fs->lock);
2318                 }
2319
2320                 if (new_fd) {
2321                         fd = current->files;
2322                         current->files = new_fd;
2323                         new_fd = fd;
2324                 }
2325
2326                 task_unlock(current);
2327
2328                 if (new_cred) {
2329                         /* Install the new user namespace */
2330                         commit_creds(new_cred);
2331                         new_cred = NULL;
2332                 }
2333         }
2334
2335 bad_unshare_cleanup_cred:
2336         if (new_cred)
2337                 put_cred(new_cred);
2338 bad_unshare_cleanup_fd:
2339         if (new_fd)
2340                 put_files_struct(new_fd);
2341
2342 bad_unshare_cleanup_fs:
2343         if (new_fs)
2344                 free_fs_struct(new_fs);
2345
2346 bad_unshare_out:
2347         return err;
2348 }
2349
2350 /*
2351  *      Helper to unshare the files of the current task.
2352  *      We don't want to expose copy_files internals to
2353  *      the exec layer of the kernel.
2354  */
2355
2356 int unshare_files(struct files_struct **displaced)
2357 {
2358         struct task_struct *task = current;
2359         struct files_struct *copy = NULL;
2360         int error;
2361
2362         error = unshare_fd(CLONE_FILES, &copy);
2363         if (error || !copy) {
2364                 *displaced = NULL;
2365                 return error;
2366         }
2367         *displaced = task->files;
2368         task_lock(task);
2369         task->files = copy;
2370         task_unlock(task);
2371         return 0;
2372 }
2373
2374 int sysctl_max_threads(struct ctl_table *table, int write,
2375                        void __user *buffer, size_t *lenp, loff_t *ppos)
2376 {
2377         struct ctl_table t;
2378         int ret;
2379         int threads = max_threads;
2380         int min = MIN_THREADS;
2381         int max = MAX_THREADS;
2382
2383         t = *table;
2384         t.data = &threads;
2385         t.extra1 = &min;
2386         t.extra2 = &max;
2387
2388         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2389         if (ret || !write)
2390                 return ret;
2391
2392         set_max_threads(threads);
2393
2394         return 0;
2395 }