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/sched/stat.h>
  21 #include <linux/sched/task.h>
  22 #include <linux/sched/task_stack.h>
  23 #include <linux/sched/cputime.h>
  24 #include <linux/rtmutex.h>
  25 #include <linux/init.h>
  26 #include <linux/unistd.h>
  27 #include <linux/module.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/completion.h>
  30 #include <linux/personality.h>
  31 #include <linux/mempolicy.h>
  32 #include <linux/sem.h>
  33 #include <linux/file.h>
  34 #include <linux/fdtable.h>
  35 #include <linux/iocontext.h>
  36 #include <linux/key.h>
  37 #include <linux/binfmts.h>
  38 #include <linux/mman.h>
  39 #include <linux/mmu_notifier.h>
  40 #include <linux/fs.h>
  41 #include <linux/mm.h>
  42 #include <linux/vmacache.h>
  43 #include <linux/nsproxy.h>
  44 #include <linux/capability.h>
  45 #include <linux/cpu.h>
  46 #include <linux/cgroup.h>
  47 #include <linux/security.h>
  48 #include <linux/hugetlb.h>
  49 #include <linux/seccomp.h>
  50 #include <linux/swap.h>
  51 #include <linux/syscalls.h>
  52 #include <linux/jiffies.h>
  53 #include <linux/futex.h>
  54 #include <linux/compat.h>
  55 #include <linux/kthread.h>
  56 #include <linux/task_io_accounting_ops.h>
  57 #include <linux/rcupdate.h>
  58 #include <linux/ptrace.h>
  59 #include <linux/mount.h>
  60 #include <linux/audit.h>
  61 #include <linux/memcontrol.h>
  62 #include <linux/ftrace.h>
  63 #include <linux/proc_fs.h>
  64 #include <linux/profile.h>
  65 #include <linux/rmap.h>
  66 #include <linux/ksm.h>
  67 #include <linux/acct.h>
  68 #include <linux/userfaultfd_k.h>
  69 #include <linux/tsacct_kern.h>
  70 #include <linux/cn_proc.h>
  71 #include <linux/freezer.h>
  72 #include <linux/delayacct.h>
  73 #include <linux/taskstats_kern.h>
  74 #include <linux/random.h>
  75 #include <linux/tty.h>
  76 #include <linux/blkdev.h>
  77 #include <linux/fs_struct.h>
  78 #include <linux/magic.h>
  79 #include <linux/perf_event.h>
  80 #include <linux/posix-timers.h>
  81 #include <linux/user-return-notifier.h>
  82 #include <linux/oom.h>
  83 #include <linux/khugepaged.h>
  84 #include <linux/signalfd.h>
  85 #include <linux/uprobes.h>
  86 #include <linux/aio.h>
  87 #include <linux/compiler.h>
  88 #include <linux/sysctl.h>
  89 #include <linux/kcov.h>
  90 #include <linux/livepatch.h>
  91
  92 #include <asm/pgtable.h>
  93 #include <asm/pgalloc.h>
  94 #include <linux/uaccess.h>
  95 #include <asm/mmu_context.h>
  96 #include <asm/cacheflush.h>
  97 #include <asm/tlbflush.h>
  98
  99 #include <trace/events/sched.h>
 100
 101 #define CREATE_TRACE_POINTS
 102 #include <trace/events/task.h>
 103
 104 /*
 105  * Minimum number of threads to boot the kernel
 106  */
 107 #define MIN_THREADS 20
 108
 109 /*
 110  * Maximum number of threads
 111  */
 112 #define MAX_THREADS FUTEX_TID_MASK
 113
 114 /*
 115  * Protected counters by write_lock_irq(&tasklist_lock)
 116  */
 117 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 118 int nr_threads;                 /* The idle threads do not count.. */
 119
 120 int max_threads;                /* tunable limit on nr_threads */
 121
 122 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 123
 124 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 125
 126 #ifdef CONFIG_PROVE_RCU
 127 int lockdep_tasklist_lock_is_held(void)
 128 {
 129         return lockdep_is_held(&tasklist_lock);
 130 }
 131 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 132 #endif /* #ifdef CONFIG_PROVE_RCU */
 133
 134 int nr_processes(void)
 135 {
 136         int cpu;
 137         int total = 0;
 138
 139         for_each_possible_cpu(cpu)
 140                 total += per_cpu(process_counts, cpu);
 141
 142         return total;
 143 }
 144
 145 void __weak arch_release_task_struct(struct task_struct *tsk)
 146 {
 147 }
 148
 149 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 150 static struct kmem_cache *task_struct_cachep;
 151
 152 static inline struct task_struct *alloc_task_struct_node(int node)
 153 {
 154         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 155 }
 156
 157 static inline void free_task_struct(struct task_struct *tsk)
 158 {
 159         kmem_cache_free(task_struct_cachep, tsk);
 160 }
 161 #endif
 162
 163 void __weak arch_release_thread_stack(unsigned long *stack)
 164 {
 165 }
 166
 167 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 168
 169 /*
 170  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 171  * kmemcache based allocator.
 172  */
 173 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 174
 175 #ifdef CONFIG_VMAP_STACK
 176 /*
 177  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 178  * flush.  Try to minimize the number of calls by caching stacks.
 179  */
 180 #define NR_CACHED_STACKS 2
 181 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 182 #endif
 183
 184 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 185 {
 186 #ifdef CONFIG_VMAP_STACK
 187         void *stack;
 188         int i;
 189
 190         local_irq_disable();
 191         for (i = 0; i < NR_CACHED_STACKS; i++) {
 192                 struct vm_struct *s = this_cpu_read(cached_stacks[i]);
 193
 194                 if (!s)
 195                         continue;
 196                 this_cpu_write(cached_stacks[i], NULL);
 197
 198                 tsk->stack_vm_area = s;
 199                 local_irq_enable();
 200                 return s->addr;
 201         }
 202         local_irq_enable();
 203
 204         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_SIZE,
 205                                      VMALLOC_START, VMALLOC_END,
 206                                      THREADINFO_GFP | __GFP_HIGHMEM,
 207                                      PAGE_KERNEL,
 208                                      0, node, __builtin_return_address(0));
 209
 210         /*
 211          * We can't call find_vm_area() in interrupt context, and
 212          * free_thread_stack() can be called in interrupt context,
 213          * so cache the vm_struct.
 214          */
 215         if (stack)
 216                 tsk->stack_vm_area = find_vm_area(stack);
 217         return stack;
 218 #else
 219         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 220                                              THREAD_SIZE_ORDER);
 221
 222         return page ? page_address(page) : NULL;
 223 #endif
 224 }
 225
 226 static inline void free_thread_stack(struct task_struct *tsk)
 227 {
 228 #ifdef CONFIG_VMAP_STACK
 229         if (task_stack_vm_area(tsk)) {
 230                 unsigned long flags;
 231                 int i;
 232
 233                 local_irq_save(flags);
 234                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 235                         if (this_cpu_read(cached_stacks[i]))
 236                                 continue;
 237
 238                         this_cpu_write(cached_stacks[i], tsk->stack_vm_area);
 239                         local_irq_restore(flags);
 240                         return;
 241                 }
 242                 local_irq_restore(flags);
 243
 244                 vfree_atomic(tsk->stack);
 245                 return;
 246         }
 247 #endif
 248
 249         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 250 }
 251 # else
 252 static struct kmem_cache *thread_stack_cache;
 253
 254 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 255                                                   int node)
 256 {
 257         return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 258 }
 259
 260 static void free_thread_stack(struct task_struct *tsk)
 261 {
 262         kmem_cache_free(thread_stack_cache, tsk->stack);
 263 }
 264
 265 void thread_stack_cache_init(void)
 266 {
 267         thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
 268                                               THREAD_SIZE, 0, NULL);
 269         BUG_ON(thread_stack_cache == NULL);
 270 }
 271 # endif
 272 #endif
 273
 274 /* SLAB cache for signal_struct structures (tsk->signal) */
 275 static struct kmem_cache *signal_cachep;
 276
 277 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 278 struct kmem_cache *sighand_cachep;
 279
 280 /* SLAB cache for files_struct structures (tsk->files) */
 281 struct kmem_cache *files_cachep;
 282
 283 /* SLAB cache for fs_struct structures (tsk->fs) */
 284 struct kmem_cache *fs_cachep;
 285
 286 /* SLAB cache for vm_area_struct structures */
 287 struct kmem_cache *vm_area_cachep;
 288
 289 /* SLAB cache for mm_struct structures (tsk->mm) */
 290 static struct kmem_cache *mm_cachep;
 291
 292 static void account_kernel_stack(struct task_struct *tsk, int account)
 293 {
 294         void *stack = task_stack_page(tsk);
 295         struct vm_struct *vm = task_stack_vm_area(tsk);
 296
 297         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 298
 299         if (vm) {
 300                 int i;
 301
 302                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 303
 304                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 305                         mod_zone_page_state(page_zone(vm->pages[i]),
 306                                             NR_KERNEL_STACK_KB,
 307                                             PAGE_SIZE / 1024 * account);
 308                 }
 309
 310                 /* All stack pages belong to the same memcg. */
 311                 memcg_kmem_update_page_stat(vm->pages[0], MEMCG_KERNEL_STACK_KB,
 312                                             account * (THREAD_SIZE / 1024));
 313         } else {
 314                 /*
 315                  * All stack pages are in the same zone and belong to the
 316                  * same memcg.
 317                  */
 318                 struct page *first_page = virt_to_page(stack);
 319
 320                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 321                                     THREAD_SIZE / 1024 * account);
 322
 323                 memcg_kmem_update_page_stat(first_page, MEMCG_KERNEL_STACK_KB,
 324                                             account * (THREAD_SIZE / 1024));
 325         }
 326 }
 327
 328 static void release_task_stack(struct task_struct *tsk)
 329 {
 330         if (WARN_ON(tsk->state != TASK_DEAD))
 331                 return;  /* Better to leak the stack than to free prematurely */
 332
 333         account_kernel_stack(tsk, -1);
 334         arch_release_thread_stack(tsk->stack);
 335         free_thread_stack(tsk);
 336         tsk->stack = NULL;
 337 #ifdef CONFIG_VMAP_STACK
 338         tsk->stack_vm_area = NULL;
 339 #endif
 340 }
 341
 342 #ifdef CONFIG_THREAD_INFO_IN_TASK
 343 void put_task_stack(struct task_struct *tsk)
 344 {
 345         if (atomic_dec_and_test(&tsk->stack_refcount))
 346                 release_task_stack(tsk);
 347 }
 348 #endif
 349
 350 void free_task(struct task_struct *tsk)
 351 {
 352 #ifndef CONFIG_THREAD_INFO_IN_TASK
 353         /*
 354          * The task is finally done with both the stack and thread_info,
 355          * so free both.
 356          */
 357         release_task_stack(tsk);
 358 #else
 359         /*
 360          * If the task had a separate stack allocation, it should be gone
 361          * by now.
 362          */
 363         WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
 364 #endif
 365         rt_mutex_debug_task_free(tsk);
 366         ftrace_graph_exit_task(tsk);
 367         put_seccomp_filter(tsk);
 368         arch_release_task_struct(tsk);
 369         if (tsk->flags & PF_KTHREAD)
 370                 free_kthread_struct(tsk);
 371         free_task_struct(tsk);
 372 }
 373 EXPORT_SYMBOL(free_task);
 374
 375 static inline void free_signal_struct(struct signal_struct *sig)
 376 {
 377         taskstats_tgid_free(sig);
 378         sched_autogroup_exit(sig);
 379         /*
 380          * __mmdrop is not safe to call from softirq context on x86 due to
 381          * pgd_dtor so postpone it to the async context
 382          */
 383         if (sig->oom_mm)
 384                 mmdrop_async(sig->oom_mm);
 385         kmem_cache_free(signal_cachep, sig);
 386 }
 387
 388 static inline void put_signal_struct(struct signal_struct *sig)
 389 {
 390         if (atomic_dec_and_test(&sig->sigcnt))
 391                 free_signal_struct(sig);
 392 }
 393
 394 void __put_task_struct(struct task_struct *tsk)
 395 {
 396         WARN_ON(!tsk->exit_state);
 397         WARN_ON(atomic_read(&tsk->usage));
 398         WARN_ON(tsk == current);
 399
 400         cgroup_free(tsk);
 401         task_numa_free(tsk);
 402         security_task_free(tsk);
 403         exit_creds(tsk);
 404         delayacct_tsk_free(tsk);
 405         put_signal_struct(tsk->signal);
 406
 407         if (!profile_handoff_task(tsk))
 408                 free_task(tsk);
 409 }
 410 EXPORT_SYMBOL_GPL(__put_task_struct);
 411
 412 void __init __weak arch_task_cache_init(void) { }
 413
 414 /*
 415  * set_max_threads
 416  */
 417 static void set_max_threads(unsigned int max_threads_suggested)
 418 {
 419         u64 threads;
 420
 421         /*
 422          * The number of threads shall be limited such that the thread
 423          * structures may only consume a small part of the available memory.
 424          */
 425         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 426                 threads = MAX_THREADS;
 427         else
 428                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 429                                     (u64) THREAD_SIZE * 8UL);
 430
 431         if (threads > max_threads_suggested)
 432                 threads = max_threads_suggested;
 433
 434         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 435 }
 436
 437 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 438 /* Initialized by the architecture: */
 439 int arch_task_struct_size __read_mostly;
 440 #endif
 441
 442 void __init fork_init(void)
 443 {
 444         int i;
 445 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 446 #ifndef ARCH_MIN_TASKALIGN
 447 #define ARCH_MIN_TASKALIGN      0
 448 #endif
 449         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 450
 451         /* create a slab on which task_structs can be allocated */
 452         task_struct_cachep = kmem_cache_create("task_struct",
 453                         arch_task_struct_size, align,
 454                         SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
 455 #endif
 456
 457         /* do the arch specific task caches init */
 458         arch_task_cache_init();
 459
 460         set_max_threads(MAX_THREADS);
 461
 462         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 463         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 464         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 465                 init_task.signal->rlim[RLIMIT_NPROC];
 466
 467         for (i = 0; i < UCOUNT_COUNTS; i++) {
 468                 init_user_ns.ucount_max[i] = max_threads/2;
 469         }
 470 }
 471
 472 int __weak arch_dup_task_struct(struct task_struct *dst,
 473                                                struct task_struct *src)
 474 {
 475         *dst = *src;
 476         return 0;
 477 }
 478
 479 void set_task_stack_end_magic(struct task_struct *tsk)
 480 {
 481         unsigned long *stackend;
 482
 483         stackend = end_of_stack(tsk);
 484         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 485 }
 486
 487 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 488 {
 489         struct task_struct *tsk;
 490         unsigned long *stack;
 491         struct vm_struct *stack_vm_area;
 492         int err;
 493
 494         if (node == NUMA_NO_NODE)
 495                 node = tsk_fork_get_node(orig);
 496         tsk = alloc_task_struct_node(node);
 497         if (!tsk)
 498                 return NULL;
 499
 500         stack = alloc_thread_stack_node(tsk, node);
 501         if (!stack)
 502                 goto free_tsk;
 503
 504         stack_vm_area = task_stack_vm_area(tsk);
 505
 506         err = arch_dup_task_struct(tsk, orig);
 507
 508         /*
 509          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 510          * sure they're properly initialized before using any stack-related
 511          * functions again.
 512          */
 513         tsk->stack = stack;
 514 #ifdef CONFIG_VMAP_STACK
 515         tsk->stack_vm_area = stack_vm_area;
 516 #endif
 517 #ifdef CONFIG_THREAD_INFO_IN_TASK
 518         atomic_set(&tsk->stack_refcount, 1);
 519 #endif
 520
 521         if (err)
 522                 goto free_stack;
 523
 524 #ifdef CONFIG_SECCOMP
 525         /*
 526          * We must handle setting up seccomp filters once we're under
 527          * the sighand lock in case orig has changed between now and
 528          * then. Until then, filter must be NULL to avoid messing up
 529          * the usage counts on the error path calling free_task.
 530          */
 531         tsk->seccomp.filter = NULL;
 532 #endif
 533
 534         setup_thread_stack(tsk, orig);
 535         clear_user_return_notifier(tsk);
 536         clear_tsk_need_resched(tsk);
 537         set_task_stack_end_magic(tsk);
 538
 539 #ifdef CONFIG_CC_STACKPROTECTOR
 540         tsk->stack_canary = get_random_int();
 541 #endif
 542
 543         /*
 544          * One for us, one for whoever does the "release_task()" (usually
 545          * parent)
 546          */
 547         atomic_set(&tsk->usage, 2);
 548 #ifdef CONFIG_BLK_DEV_IO_TRACE
 549         tsk->btrace_seq = 0;
 550 #endif
 551         tsk->splice_pipe = NULL;
 552         tsk->task_frag.page = NULL;
 553         tsk->wake_q.next = NULL;
 554
 555         account_kernel_stack(tsk, 1);
 556
 557         kcov_task_init(tsk);
 558
 559         return tsk;
 560
 561 free_stack:
 562         free_thread_stack(tsk);
 563 free_tsk:
 564         free_task_struct(tsk);
 565         return NULL;
 566 }
 567
 568 #ifdef CONFIG_MMU
 569 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 570                                         struct mm_struct *oldmm)
 571 {
 572         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 573         struct rb_node **rb_link, *rb_parent;
 574         int retval;
 575         unsigned long charge;
 576         LIST_HEAD(uf);
 577
 578         uprobe_start_dup_mmap();
 579         if (down_write_killable(&oldmm->mmap_sem)) {
 580                 retval = -EINTR;
 581                 goto fail_uprobe_end;
 582         }
 583         flush_cache_dup_mm(oldmm);
 584         uprobe_dup_mmap(oldmm, mm);
 585         /*
 586          * Not linked in yet - no deadlock potential:
 587          */
 588         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 589
 590         /* No ordering required: file already has been exposed. */
 591         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 592
 593         mm->total_vm = oldmm->total_vm;
 594         mm->data_vm = oldmm->data_vm;
 595         mm->exec_vm = oldmm->exec_vm;
 596         mm->stack_vm = oldmm->stack_vm;
 597
 598         rb_link = &mm->mm_rb.rb_node;
 599         rb_parent = NULL;
 600         pprev = &mm->mmap;
 601         retval = ksm_fork(mm, oldmm);
 602         if (retval)
 603                 goto out;
 604         retval = khugepaged_fork(mm, oldmm);
 605         if (retval)
 606                 goto out;
 607
 608         prev = NULL;
 609         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 610                 struct file *file;
 611
 612                 if (mpnt->vm_flags & VM_DONTCOPY) {
 613                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 614                         continue;
 615                 }
 616                 charge = 0;
 617                 if (mpnt->vm_flags & VM_ACCOUNT) {
 618                         unsigned long len = vma_pages(mpnt);
 619
 620                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 621                                 goto fail_nomem;
 622                         charge = len;
 623                 }
 624                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 625                 if (!tmp)
 626                         goto fail_nomem;
 627                 *tmp = *mpnt;
 628                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
 629                 retval = vma_dup_policy(mpnt, tmp);
 630                 if (retval)
 631                         goto fail_nomem_policy;
 632                 tmp->vm_mm = mm;
 633                 retval = dup_userfaultfd(tmp, &uf);
 634                 if (retval)
 635                         goto fail_nomem_anon_vma_fork;
 636                 if (anon_vma_fork(tmp, mpnt))
 637                         goto fail_nomem_anon_vma_fork;
 638                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 639                 tmp->vm_next = tmp->vm_prev = NULL;
 640                 file = tmp->vm_file;
 641                 if (file) {
 642                         struct inode *inode = file_inode(file);
 643                         struct address_space *mapping = file->f_mapping;
 644
 645                         get_file(file);
 646                         if (tmp->vm_flags & VM_DENYWRITE)
 647                                 atomic_dec(&inode->i_writecount);
 648                         i_mmap_lock_write(mapping);
 649                         if (tmp->vm_flags & VM_SHARED)
 650                                 atomic_inc(&mapping->i_mmap_writable);
 651                         flush_dcache_mmap_lock(mapping);
 652                         /* insert tmp into the share list, just after mpnt */
 653                         vma_interval_tree_insert_after(tmp, mpnt,
 654                                         &mapping->i_mmap);
 655                         flush_dcache_mmap_unlock(mapping);
 656                         i_mmap_unlock_write(mapping);
 657                 }
 658
 659                 /*
 660                  * Clear hugetlb-related page reserves for children. This only
 661                  * affects MAP_PRIVATE mappings. Faults generated by the child
 662                  * are not guaranteed to succeed, even if read-only
 663                  */
 664                 if (is_vm_hugetlb_page(tmp))
 665                         reset_vma_resv_huge_pages(tmp);
 666
 667                 /*
 668                  * Link in the new vma and copy the page table entries.
 669                  */
 670                 *pprev = tmp;
 671                 pprev = &tmp->vm_next;
 672                 tmp->vm_prev = prev;
 673                 prev = tmp;
 674
 675                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 676                 rb_link = &tmp->vm_rb.rb_right;
 677                 rb_parent = &tmp->vm_rb;
 678
 679                 mm->map_count++;
 680                 retval = copy_page_range(mm, oldmm, mpnt);
 681
 682                 if (tmp->vm_ops && tmp->vm_ops->open)
 683                         tmp->vm_ops->open(tmp);
 684
 685                 if (retval)
 686                         goto out;
 687         }
 688         /* a new mm has just been created */
 689         arch_dup_mmap(oldmm, mm);
 690         retval = 0;
 691 out:
 692         up_write(&mm->mmap_sem);
 693         flush_tlb_mm(oldmm);
 694         up_write(&oldmm->mmap_sem);
 695         dup_userfaultfd_complete(&uf);
 696 fail_uprobe_end:
 697         uprobe_end_dup_mmap();
 698         return retval;
 699 fail_nomem_anon_vma_fork:
 700         mpol_put(vma_policy(tmp));
 701 fail_nomem_policy:
 702         kmem_cache_free(vm_area_cachep, tmp);
 703 fail_nomem:
 704         retval = -ENOMEM;
 705         vm_unacct_memory(charge);
 706         goto out;
 707 }
 708
 709 static inline int mm_alloc_pgd(struct mm_struct *mm)
 710 {
 711         mm->pgd = pgd_alloc(mm);
 712         if (unlikely(!mm->pgd))
 713                 return -ENOMEM;
 714         return 0;
 715 }
 716
 717 static inline void mm_free_pgd(struct mm_struct *mm)
 718 {
 719         pgd_free(mm, mm->pgd);
 720 }
 721 #else
 722 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 723 {
 724         down_write(&oldmm->mmap_sem);
 725         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 726         up_write(&oldmm->mmap_sem);
 727         return 0;
 728 }
 729 #define mm_alloc_pgd(mm)        (0)
 730 #define mm_free_pgd(mm)
 731 #endif /* CONFIG_MMU */
 732
 733 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 734
 735 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 736 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 737
 738 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 739
 740 static int __init coredump_filter_setup(char *s)
 741 {
 742         default_dump_filter =
 743                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 744                 MMF_DUMP_FILTER_MASK;
 745         return 1;
 746 }
 747
 748 __setup("coredump_filter=", coredump_filter_setup);
 749
 750 #include <linux/init_task.h>
 751
 752 static void mm_init_aio(struct mm_struct *mm)
 753 {
 754 #ifdef CONFIG_AIO
 755         spin_lock_init(&mm->ioctx_lock);
 756         mm->ioctx_table = NULL;
 757 #endif
 758 }
 759
 760 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 761 {
 762 #ifdef CONFIG_MEMCG
 763         mm->owner = p;
 764 #endif
 765 }
 766
 767 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 768         struct user_namespace *user_ns)
 769 {
 770         mm->mmap = NULL;
 771         mm->mm_rb = RB_ROOT;
 772         mm->vmacache_seqnum = 0;
 773         atomic_set(&mm->mm_users, 1);
 774         atomic_set(&mm->mm_count, 1);
 775         init_rwsem(&mm->mmap_sem);
 776         INIT_LIST_HEAD(&mm->mmlist);
 777         mm->core_state = NULL;
 778         atomic_long_set(&mm->nr_ptes, 0);
 779         mm_nr_pmds_init(mm);
 780         mm->map_count = 0;
 781         mm->locked_vm = 0;
 782         mm->pinned_vm = 0;
 783         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 784         spin_lock_init(&mm->page_table_lock);
 785         mm_init_cpumask(mm);
 786         mm_init_aio(mm);
 787         mm_init_owner(mm, p);
 788         mmu_notifier_mm_init(mm);
 789         clear_tlb_flush_pending(mm);
 790 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 791         mm->pmd_huge_pte = NULL;
 792 #endif
 793
 794         if (current->mm) {
 795                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 796                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 797         } else {
 798                 mm->flags = default_dump_filter;
 799                 mm->def_flags = 0;
 800         }
 801
 802         if (mm_alloc_pgd(mm))
 803                 goto fail_nopgd;
 804
 805         if (init_new_context(p, mm))
 806                 goto fail_nocontext;
 807
 808         mm->user_ns = get_user_ns(user_ns);
 809         return mm;
 810
 811 fail_nocontext:
 812         mm_free_pgd(mm);
 813 fail_nopgd:
 814         free_mm(mm);
 815         return NULL;
 816 }
 817
 818 static void check_mm(struct mm_struct *mm)
 819 {
 820         int i;
 821
 822         for (i = 0; i < NR_MM_COUNTERS; i++) {
 823                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 824
 825                 if (unlikely(x))
 826                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 827                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 828         }
 829
 830         if (atomic_long_read(&mm->nr_ptes))
 831                 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
 832                                 atomic_long_read(&mm->nr_ptes));
 833         if (mm_nr_pmds(mm))
 834                 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
 835                                 mm_nr_pmds(mm));
 836
 837 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 838         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 839 #endif
 840 }
 841
 842 /*
 843  * Allocate and initialize an mm_struct.
 844  */
 845 struct mm_struct *mm_alloc(void)
 846 {
 847         struct mm_struct *mm;
 848
 849         mm = allocate_mm();
 850         if (!mm)
 851                 return NULL;
 852
 853         memset(mm, 0, sizeof(*mm));
 854         return mm_init(mm, current, current_user_ns());
 855 }
 856
 857 /*
 858  * Called when the last reference to the mm
 859  * is dropped: either by a lazy thread or by
 860  * mmput. Free the page directory and the mm.
 861  */
 862 void __mmdrop(struct mm_struct *mm)
 863 {
 864         BUG_ON(mm == &init_mm);
 865         mm_free_pgd(mm);
 866         destroy_context(mm);
 867         mmu_notifier_mm_destroy(mm);
 868         check_mm(mm);
 869         put_user_ns(mm->user_ns);
 870         free_mm(mm);
 871 }
 872 EXPORT_SYMBOL_GPL(__mmdrop);
 873
 874 static inline void __mmput(struct mm_struct *mm)
 875 {
 876         VM_BUG_ON(atomic_read(&mm->mm_users));
 877
 878         uprobe_clear_state(mm);
 879         exit_aio(mm);
 880         ksm_exit(mm);
 881         khugepaged_exit(mm); /* must run before exit_mmap */
 882         exit_mmap(mm);
 883         mm_put_huge_zero_page(mm);
 884         set_mm_exe_file(mm, NULL);
 885         if (!list_empty(&mm->mmlist)) {
 886                 spin_lock(&mmlist_lock);
 887                 list_del(&mm->mmlist);
 888                 spin_unlock(&mmlist_lock);
 889         }
 890         if (mm->binfmt)
 891                 module_put(mm->binfmt->module);
 892         set_bit(MMF_OOM_SKIP, &mm->flags);
 893         mmdrop(mm);
 894 }
 895
 896 /*
 897  * Decrement the use count and release all resources for an mm.
 898  */
 899 void mmput(struct mm_struct *mm)
 900 {
 901         might_sleep();
 902
 903         if (atomic_dec_and_test(&mm->mm_users))
 904                 __mmput(mm);
 905 }
 906 EXPORT_SYMBOL_GPL(mmput);
 907
 908 #ifdef CONFIG_MMU
 909 static void mmput_async_fn(struct work_struct *work)
 910 {
 911         struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
 912         __mmput(mm);
 913 }
 914
 915 void mmput_async(struct mm_struct *mm)
 916 {
 917         if (atomic_dec_and_test(&mm->mm_users)) {
 918                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
 919                 schedule_work(&mm->async_put_work);
 920         }
 921 }
 922 #endif
 923
 924 /**
 925  * set_mm_exe_file - change a reference to the mm's executable file
 926  *
 927  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
 928  *
 929  * Main users are mmput() and sys_execve(). Callers prevent concurrent
 930  * invocations: in mmput() nobody alive left, in execve task is single
 931  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
 932  * mm->exe_file, but does so without using set_mm_exe_file() in order
 933  * to do avoid the need for any locks.
 934  */
 935 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
 936 {
 937         struct file *old_exe_file;
 938
 939         /*
 940          * It is safe to dereference the exe_file without RCU as
 941          * this function is only called if nobody else can access
 942          * this mm -- see comment above for justification.
 943          */
 944         old_exe_file = rcu_dereference_raw(mm->exe_file);
 945
 946         if (new_exe_file)
 947                 get_file(new_exe_file);
 948         rcu_assign_pointer(mm->exe_file, new_exe_file);
 949         if (old_exe_file)
 950                 fput(old_exe_file);
 951 }
 952
 953 /**
 954  * get_mm_exe_file - acquire a reference to the mm's executable file
 955  *
 956  * Returns %NULL if mm has no associated executable file.
 957  * User must release file via fput().
 958  */
 959 struct file *get_mm_exe_file(struct mm_struct *mm)
 960 {
 961         struct file *exe_file;
 962
 963         rcu_read_lock();
 964         exe_file = rcu_dereference(mm->exe_file);
 965         if (exe_file && !get_file_rcu(exe_file))
 966                 exe_file = NULL;
 967         rcu_read_unlock();
 968         return exe_file;
 969 }
 970 EXPORT_SYMBOL(get_mm_exe_file);
 971
 972 /**
 973  * get_task_exe_file - acquire a reference to the task's executable file
 974  *
 975  * Returns %NULL if task's mm (if any) has no associated executable file or
 976  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
 977  * User must release file via fput().
 978  */
 979 struct file *get_task_exe_file(struct task_struct *task)
 980 {
 981         struct file *exe_file = NULL;
 982         struct mm_struct *mm;
 983
 984         task_lock(task);
 985         mm = task->mm;
 986         if (mm) {
 987                 if (!(task->flags & PF_KTHREAD))
 988                         exe_file = get_mm_exe_file(mm);
 989         }
 990         task_unlock(task);
 991         return exe_file;
 992 }
 993 EXPORT_SYMBOL(get_task_exe_file);
 994
 995 /**
 996  * get_task_mm - acquire a reference to the task's mm
 997  *
 998  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
 999  * this kernel workthread has transiently adopted a user mm with use_mm,
1000  * to do its AIO) is not set and if so returns a reference to it, after
1001  * bumping up the use count.  User must release the mm via mmput()
1002  * after use.  Typically used by /proc and ptrace.
1003  */
1004 struct mm_struct *get_task_mm(struct task_struct *task)
1005 {
1006         struct mm_struct *mm;
1007
1008         task_lock(task);
1009         mm = task->mm;
1010         if (mm) {
1011                 if (task->flags & PF_KTHREAD)
1012                         mm = NULL;
1013                 else
1014                         mmget(mm);
1015         }
1016         task_unlock(task);
1017         return mm;
1018 }
1019 EXPORT_SYMBOL_GPL(get_task_mm);
1020
1021 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1022 {
1023         struct mm_struct *mm;
1024         int err;
1025
1026         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1027         if (err)
1028                 return ERR_PTR(err);
1029
1030         mm = get_task_mm(task);
1031         if (mm && mm != current->mm &&
1032                         !ptrace_may_access(task, mode)) {
1033                 mmput(mm);
1034                 mm = ERR_PTR(-EACCES);
1035         }
1036         mutex_unlock(&task->signal->cred_guard_mutex);
1037
1038         return mm;
1039 }
1040
1041 static void complete_vfork_done(struct task_struct *tsk)
1042 {
1043         struct completion *vfork;
1044
1045         task_lock(tsk);
1046         vfork = tsk->vfork_done;
1047         if (likely(vfork)) {
1048                 tsk->vfork_done = NULL;
1049                 complete(vfork);
1050         }
1051         task_unlock(tsk);
1052 }
1053
1054 static int wait_for_vfork_done(struct task_struct *child,
1055                                 struct completion *vfork)
1056 {
1057         int killed;
1058
1059         freezer_do_not_count();
1060         killed = wait_for_completion_killable(vfork);
1061         freezer_count();
1062
1063         if (killed) {
1064                 task_lock(child);
1065                 child->vfork_done = NULL;
1066                 task_unlock(child);
1067         }
1068
1069         put_task_struct(child);
1070         return killed;
1071 }
1072
1073 /* Please note the differences between mmput and mm_release.
1074  * mmput is called whenever we stop holding onto a mm_struct,
1075  * error success whatever.
1076  *
1077  * mm_release is called after a mm_struct has been removed
1078  * from the current process.
1079  *
1080  * This difference is important for error handling, when we
1081  * only half set up a mm_struct for a new process and need to restore
1082  * the old one.  Because we mmput the new mm_struct before
1083  * restoring the old one. . .
1084  * Eric Biederman 10 January 1998
1085  */
1086 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1087 {
1088         /* Get rid of any futexes when releasing the mm */
1089 #ifdef CONFIG_FUTEX
1090         if (unlikely(tsk->robust_list)) {
1091                 exit_robust_list(tsk);
1092                 tsk->robust_list = NULL;
1093         }
1094 #ifdef CONFIG_COMPAT
1095         if (unlikely(tsk->compat_robust_list)) {
1096                 compat_exit_robust_list(tsk);
1097                 tsk->compat_robust_list = NULL;
1098         }
1099 #endif
1100         if (unlikely(!list_empty(&tsk->pi_state_list)))
1101                 exit_pi_state_list(tsk);
1102 #endif
1103
1104         uprobe_free_utask(tsk);
1105
1106         /* Get rid of any cached register state */
1107         deactivate_mm(tsk, mm);
1108
1109         /*
1110          * Signal userspace if we're not exiting with a core dump
1111          * because we want to leave the value intact for debugging
1112          * purposes.
1113          */
1114         if (tsk->clear_child_tid) {
1115                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1116                     atomic_read(&mm->mm_users) > 1) {
1117                         /*
1118                          * We don't check the error code - if userspace has
1119                          * not set up a proper pointer then tough luck.
1120                          */
1121                         put_user(0, tsk->clear_child_tid);
1122                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1123                                         1, NULL, NULL, 0);
1124                 }
1125                 tsk->clear_child_tid = NULL;
1126         }
1127
1128         /*
1129          * All done, finally we can wake up parent and return this mm to him.
1130          * Also kthread_stop() uses this completion for synchronization.
1131          */
1132         if (tsk->vfork_done)
1133                 complete_vfork_done(tsk);
1134 }
1135
1136 /*
1137  * Allocate a new mm structure and copy contents from the
1138  * mm structure of the passed in task structure.
1139  */
1140 static struct mm_struct *dup_mm(struct task_struct *tsk)
1141 {
1142         struct mm_struct *mm, *oldmm = current->mm;
1143         int err;
1144
1145         mm = allocate_mm();
1146         if (!mm)
1147                 goto fail_nomem;
1148
1149         memcpy(mm, oldmm, sizeof(*mm));
1150
1151         if (!mm_init(mm, tsk, mm->user_ns))
1152                 goto fail_nomem;
1153
1154         err = dup_mmap(mm, oldmm);
1155         if (err)
1156                 goto free_pt;
1157
1158         mm->hiwater_rss = get_mm_rss(mm);
1159         mm->hiwater_vm = mm->total_vm;
1160
1161         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1162                 goto free_pt;
1163
1164         return mm;
1165
1166 free_pt:
1167         /* don't put binfmt in mmput, we haven't got module yet */
1168         mm->binfmt = NULL;
1169         mmput(mm);
1170
1171 fail_nomem:
1172         return NULL;
1173 }
1174
1175 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1176 {
1177         struct mm_struct *mm, *oldmm;
1178         int retval;
1179
1180         tsk->min_flt = tsk->maj_flt = 0;
1181         tsk->nvcsw = tsk->nivcsw = 0;
1182 #ifdef CONFIG_DETECT_HUNG_TASK
1183         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1184 #endif
1185
1186         tsk->mm = NULL;
1187         tsk->active_mm = NULL;
1188
1189         /*
1190          * Are we cloning a kernel thread?
1191          *
1192          * We need to steal a active VM for that..
1193          */
1194         oldmm = current->mm;
1195         if (!oldmm)
1196                 return 0;
1197
1198         /* initialize the new vmacache entries */
1199         vmacache_flush(tsk);
1200
1201         if (clone_flags & CLONE_VM) {
1202                 mmget(oldmm);
1203                 mm = oldmm;
1204                 goto good_mm;
1205         }
1206
1207         retval = -ENOMEM;
1208         mm = dup_mm(tsk);
1209         if (!mm)
1210                 goto fail_nomem;
1211
1212 good_mm:
1213         tsk->mm = mm;
1214         tsk->active_mm = mm;
1215         return 0;
1216
1217 fail_nomem:
1218         return retval;
1219 }
1220
1221 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1222 {
1223         struct fs_struct *fs = current->fs;
1224         if (clone_flags & CLONE_FS) {
1225                 /* tsk->fs is already what we want */
1226                 spin_lock(&fs->lock);
1227                 if (fs->in_exec) {
1228                         spin_unlock(&fs->lock);
1229                         return -EAGAIN;
1230                 }
1231                 fs->users++;
1232                 spin_unlock(&fs->lock);
1233                 return 0;
1234         }
1235         tsk->fs = copy_fs_struct(fs);
1236         if (!tsk->fs)
1237                 return -ENOMEM;
1238         return 0;
1239 }
1240
1241 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1242 {
1243         struct files_struct *oldf, *newf;
1244         int error = 0;
1245
1246         /*
1247          * A background process may not have any files ...
1248          */
1249         oldf = current->files;
1250         if (!oldf)
1251                 goto out;
1252
1253         if (clone_flags & CLONE_FILES) {
1254                 atomic_inc(&oldf->count);
1255                 goto out;
1256         }
1257
1258         newf = dup_fd(oldf, &error);
1259         if (!newf)
1260                 goto out;
1261
1262         tsk->files = newf;
1263         error = 0;
1264 out:
1265         return error;
1266 }
1267
1268 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1269 {
1270 #ifdef CONFIG_BLOCK
1271         struct io_context *ioc = current->io_context;
1272         struct io_context *new_ioc;
1273
1274         if (!ioc)
1275                 return 0;
1276         /*
1277          * Share io context with parent, if CLONE_IO is set
1278          */
1279         if (clone_flags & CLONE_IO) {
1280                 ioc_task_link(ioc);
1281                 tsk->io_context = ioc;
1282         } else if (ioprio_valid(ioc->ioprio)) {
1283                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1284                 if (unlikely(!new_ioc))
1285                         return -ENOMEM;
1286
1287                 new_ioc->ioprio = ioc->ioprio;
1288                 put_io_context(new_ioc);
1289         }
1290 #endif
1291         return 0;
1292 }
1293
1294 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1295 {
1296         struct sighand_struct *sig;
1297
1298         if (clone_flags & CLONE_SIGHAND) {
1299                 atomic_inc(&current->sighand->count);
1300                 return 0;
1301         }
1302         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1303         rcu_assign_pointer(tsk->sighand, sig);
1304         if (!sig)
1305                 return -ENOMEM;
1306
1307         atomic_set(&sig->count, 1);
1308         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1309         return 0;
1310 }
1311
1312 void __cleanup_sighand(struct sighand_struct *sighand)
1313 {
1314         if (atomic_dec_and_test(&sighand->count)) {
1315                 signalfd_cleanup(sighand);
1316                 /*
1317                  * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1318                  * without an RCU grace period, see __lock_task_sighand().
1319                  */
1320                 kmem_cache_free(sighand_cachep, sighand);
1321         }
1322 }
1323
1324 #ifdef CONFIG_POSIX_TIMERS
1325 /*
1326  * Initialize POSIX timer handling for a thread group.
1327  */
1328 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1329 {
1330         unsigned long cpu_limit;
1331
1332         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1333         if (cpu_limit != RLIM_INFINITY) {
1334                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1335                 sig->cputimer.running = true;
1336         }
1337
1338         /* The timer lists. */
1339         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1340         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1341         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1342 }
1343 #else
1344 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1345 #endif
1346
1347 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1348 {
1349         struct signal_struct *sig;
1350
1351         if (clone_flags & CLONE_THREAD)
1352                 return 0;
1353
1354         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1355         tsk->signal = sig;
1356         if (!sig)
1357                 return -ENOMEM;
1358
1359         sig->nr_threads = 1;
1360         atomic_set(&sig->live, 1);
1361         atomic_set(&sig->sigcnt, 1);
1362
1363         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1364         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1365         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1366
1367         init_waitqueue_head(&sig->wait_chldexit);
1368         sig->curr_target = tsk;
1369         init_sigpending(&sig->shared_pending);
1370         seqlock_init(&sig->stats_lock);
1371         prev_cputime_init(&sig->prev_cputime);
1372
1373 #ifdef CONFIG_POSIX_TIMERS
1374         INIT_LIST_HEAD(&sig->posix_timers);
1375         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1376         sig->real_timer.function = it_real_fn;
1377 #endif
1378
1379         task_lock(current->group_leader);
1380         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1381         task_unlock(current->group_leader);
1382
1383         posix_cpu_timers_init_group(sig);
1384
1385         tty_audit_fork(sig);
1386         sched_autogroup_fork(sig);
1387
1388         sig->oom_score_adj = current->signal->oom_score_adj;
1389         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1390
1391         mutex_init(&sig->cred_guard_mutex);
1392
1393         return 0;
1394 }
1395
1396 static void copy_seccomp(struct task_struct *p)
1397 {
1398 #ifdef CONFIG_SECCOMP
1399         /*
1400          * Must be called with sighand->lock held, which is common to
1401          * all threads in the group. Holding cred_guard_mutex is not
1402          * needed because this new task is not yet running and cannot
1403          * be racing exec.
1404          */
1405         assert_spin_locked(&current->sighand->siglock);
1406
1407         /* Ref-count the new filter user, and assign it. */
1408         get_seccomp_filter(current);
1409         p->seccomp = current->seccomp;
1410
1411         /*
1412          * Explicitly enable no_new_privs here in case it got set
1413          * between the task_struct being duplicated and holding the
1414          * sighand lock. The seccomp state and nnp must be in sync.
1415          */
1416         if (task_no_new_privs(current))
1417                 task_set_no_new_privs(p);
1418
1419         /*
1420          * If the parent gained a seccomp mode after copying thread
1421          * flags and between before we held the sighand lock, we have
1422          * to manually enable the seccomp thread flag here.
1423          */
1424         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1425                 set_tsk_thread_flag(p, TIF_SECCOMP);
1426 #endif
1427 }
1428
1429 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1430 {
1431         current->clear_child_tid = tidptr;
1432
1433         return task_pid_vnr(current);
1434 }
1435
1436 static void rt_mutex_init_task(struct task_struct *p)
1437 {
1438         raw_spin_lock_init(&p->pi_lock);
1439 #ifdef CONFIG_RT_MUTEXES
1440         p->pi_waiters = RB_ROOT;
1441         p->pi_waiters_leftmost = NULL;
1442         p->pi_top_task = NULL;
1443         p->pi_blocked_on = NULL;
1444 #endif
1445 }
1446
1447 #ifdef CONFIG_POSIX_TIMERS
1448 /*
1449  * Initialize POSIX timer handling for a single task.
1450  */
1451 static void posix_cpu_timers_init(struct task_struct *tsk)
1452 {
1453         tsk->cputime_expires.prof_exp = 0;
1454         tsk->cputime_expires.virt_exp = 0;
1455         tsk->cputime_expires.sched_exp = 0;
1456         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1457         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1458         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1459 }
1460 #else
1461 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1462 #endif
1463
1464 static inline void
1465 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1466 {
1467          task->pids[type].pid = pid;
1468 }
1469
1470 static inline void rcu_copy_process(struct task_struct *p)
1471 {
1472 #ifdef CONFIG_PREEMPT_RCU
1473         p->rcu_read_lock_nesting = 0;
1474         p->rcu_read_unlock_special.s = 0;
1475         p->rcu_blocked_node = NULL;
1476         INIT_LIST_HEAD(&p->rcu_node_entry);
1477 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1478 #ifdef CONFIG_TASKS_RCU
1479         p->rcu_tasks_holdout = false;
1480         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1481         p->rcu_tasks_idle_cpu = -1;
1482 #endif /* #ifdef CONFIG_TASKS_RCU */
1483 }
1484
1485 /*
1486  * This creates a new process as a copy of the old one,
1487  * but does not actually start it yet.
1488  *
1489  * It copies the registers, and all the appropriate
1490  * parts of the process environment (as per the clone
1491  * flags). The actual kick-off is left to the caller.
1492  */
1493 static __latent_entropy struct task_struct *copy_process(
1494                                         unsigned long clone_flags,
1495                                         unsigned long stack_start,
1496                                         unsigned long stack_size,
1497                                         int __user *child_tidptr,
1498                                         struct pid *pid,
1499                                         int trace,
1500                                         unsigned long tls,
1501                                         int node)
1502 {
1503         int retval;
1504         struct task_struct *p;
1505
1506         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1507                 return ERR_PTR(-EINVAL);
1508
1509         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1510                 return ERR_PTR(-EINVAL);
1511
1512         /*
1513          * Thread groups must share signals as well, and detached threads
1514          * can only be started up within the thread group.
1515          */
1516         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1517                 return ERR_PTR(-EINVAL);
1518
1519         /*
1520          * Shared signal handlers imply shared VM. By way of the above,
1521          * thread groups also imply shared VM. Blocking this case allows
1522          * for various simplifications in other code.
1523          */
1524         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1525                 return ERR_PTR(-EINVAL);
1526
1527         /*
1528          * Siblings of global init remain as zombies on exit since they are
1529          * not reaped by their parent (swapper). To solve this and to avoid
1530          * multi-rooted process trees, prevent global and container-inits
1531          * from creating siblings.
1532          */
1533         if ((clone_flags & CLONE_PARENT) &&
1534                                 current->signal->flags & SIGNAL_UNKILLABLE)
1535                 return ERR_PTR(-EINVAL);
1536
1537         /*
1538          * If the new process will be in a different pid or user namespace
1539          * do not allow it to share a thread group with the forking task.
1540          */
1541         if (clone_flags & CLONE_THREAD) {
1542                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1543                     (task_active_pid_ns(current) !=
1544                                 current->nsproxy->pid_ns_for_children))
1545                         return ERR_PTR(-EINVAL);
1546         }
1547
1548         retval = security_task_create(clone_flags);
1549         if (retval)
1550                 goto fork_out;
1551
1552         retval = -ENOMEM;
1553         p = dup_task_struct(current, node);
1554         if (!p)
1555                 goto fork_out;
1556
1557         ftrace_graph_init_task(p);
1558
1559         rt_mutex_init_task(p);
1560
1561 #ifdef CONFIG_PROVE_LOCKING
1562         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1563         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1564 #endif
1565         retval = -EAGAIN;
1566         if (atomic_read(&p->real_cred->user->processes) >=
1567                         task_rlimit(p, RLIMIT_NPROC)) {
1568                 if (p->real_cred->user != INIT_USER &&
1569                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1570                         goto bad_fork_free;
1571         }
1572         current->flags &= ~PF_NPROC_EXCEEDED;
1573
1574         retval = copy_creds(p, clone_flags);
1575         if (retval < 0)
1576                 goto bad_fork_free;
1577
1578         /*
1579          * If multiple threads are within copy_process(), then this check
1580          * triggers too late. This doesn't hurt, the check is only there
1581          * to stop root fork bombs.
1582          */
1583         retval = -EAGAIN;
1584         if (nr_threads >= max_threads)
1585                 goto bad_fork_cleanup_count;
1586
1587         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1588         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1589         p->flags |= PF_FORKNOEXEC;
1590         INIT_LIST_HEAD(&p->children);
1591         INIT_LIST_HEAD(&p->sibling);
1592         rcu_copy_process(p);
1593         p->vfork_done = NULL;
1594         spin_lock_init(&p->alloc_lock);
1595
1596         init_sigpending(&p->pending);
1597
1598         p->utime = p->stime = p->gtime = 0;
1599 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1600         p->utimescaled = p->stimescaled = 0;
1601 #endif
1602         prev_cputime_init(&p->prev_cputime);
1603
1604 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1605         seqcount_init(&p->vtime_seqcount);
1606         p->vtime_snap = 0;
1607         p->vtime_snap_whence = VTIME_INACTIVE;
1608 #endif
1609
1610 #if defined(SPLIT_RSS_COUNTING)
1611         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1612 #endif
1613
1614         p->default_timer_slack_ns = current->timer_slack_ns;
1615
1616         task_io_accounting_init(&p->ioac);
1617         acct_clear_integrals(p);
1618
1619         posix_cpu_timers_init(p);
1620
1621         p->start_time = ktime_get_ns();
1622         p->real_start_time = ktime_get_boot_ns();
1623         p->io_context = NULL;
1624         p->audit_context = NULL;
1625         cgroup_fork(p);
1626 #ifdef CONFIG_NUMA
1627         p->mempolicy = mpol_dup(p->mempolicy);
1628         if (IS_ERR(p->mempolicy)) {
1629                 retval = PTR_ERR(p->mempolicy);
1630                 p->mempolicy = NULL;
1631                 goto bad_fork_cleanup_threadgroup_lock;
1632         }
1633 #endif
1634 #ifdef CONFIG_CPUSETS
1635         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1636         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1637         seqcount_init(&p->mems_allowed_seq);
1638 #endif
1639 #ifdef CONFIG_TRACE_IRQFLAGS
1640         p->irq_events = 0;
1641         p->hardirqs_enabled = 0;
1642         p->hardirq_enable_ip = 0;
1643         p->hardirq_enable_event = 0;
1644         p->hardirq_disable_ip = _THIS_IP_;
1645         p->hardirq_disable_event = 0;
1646         p->softirqs_enabled = 1;
1647         p->softirq_enable_ip = _THIS_IP_;
1648         p->softirq_enable_event = 0;
1649         p->softirq_disable_ip = 0;
1650         p->softirq_disable_event = 0;
1651         p->hardirq_context = 0;
1652         p->softirq_context = 0;
1653 #endif
1654
1655         p->pagefault_disabled = 0;
1656
1657 #ifdef CONFIG_LOCKDEP
1658         p->lockdep_depth = 0; /* no locks held yet */
1659         p->curr_chain_key = 0;
1660         p->lockdep_recursion = 0;
1661 #endif
1662
1663 #ifdef CONFIG_DEBUG_MUTEXES
1664         p->blocked_on = NULL; /* not blocked yet */
1665 #endif
1666 #ifdef CONFIG_BCACHE
1667         p->sequential_io        = 0;
1668         p->sequential_io_avg    = 0;
1669 #endif
1670
1671         /* Perform scheduler related setup. Assign this task to a CPU. */
1672         retval = sched_fork(clone_flags, p);
1673         if (retval)
1674                 goto bad_fork_cleanup_policy;
1675
1676         retval = perf_event_init_task(p);
1677         if (retval)
1678                 goto bad_fork_cleanup_policy;
1679         retval = audit_alloc(p);
1680         if (retval)
1681                 goto bad_fork_cleanup_perf;
1682         /* copy all the process information */
1683         shm_init_task(p);
1684         retval = security_task_alloc(p, clone_flags);
1685         if (retval)
1686                 goto bad_fork_cleanup_audit;
1687         retval = copy_semundo(clone_flags, p);
1688         if (retval)
1689                 goto bad_fork_cleanup_security;
1690         retval = copy_files(clone_flags, p);
1691         if (retval)
1692                 goto bad_fork_cleanup_semundo;
1693         retval = copy_fs(clone_flags, p);
1694         if (retval)
1695                 goto bad_fork_cleanup_files;
1696         retval = copy_sighand(clone_flags, p);
1697         if (retval)
1698                 goto bad_fork_cleanup_fs;
1699         retval = copy_signal(clone_flags, p);
1700         if (retval)
1701                 goto bad_fork_cleanup_sighand;
1702         retval = copy_mm(clone_flags, p);
1703         if (retval)
1704                 goto bad_fork_cleanup_signal;
1705         retval = copy_namespaces(clone_flags, p);
1706         if (retval)
1707                 goto bad_fork_cleanup_mm;
1708         retval = copy_io(clone_flags, p);
1709         if (retval)
1710                 goto bad_fork_cleanup_namespaces;
1711         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1712         if (retval)
1713                 goto bad_fork_cleanup_io;
1714
1715         if (pid != &init_struct_pid) {
1716                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1717                 if (IS_ERR(pid)) {
1718                         retval = PTR_ERR(pid);
1719                         goto bad_fork_cleanup_thread;
1720                 }
1721         }
1722
1723         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1724         /*
1725          * Clear TID on mm_release()?
1726          */
1727         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1728 #ifdef CONFIG_BLOCK
1729         p->plug = NULL;
1730 #endif
1731 #ifdef CONFIG_FUTEX
1732         p->robust_list = NULL;
1733 #ifdef CONFIG_COMPAT
1734         p->compat_robust_list = NULL;
1735 #endif
1736         INIT_LIST_HEAD(&p->pi_state_list);
1737         p->pi_state_cache = NULL;
1738 #endif
1739         /*
1740          * sigaltstack should be cleared when sharing the same VM
1741          */
1742         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1743                 sas_ss_reset(p);
1744
1745         /*
1746          * Syscall tracing and stepping should be turned off in the
1747          * child regardless of CLONE_PTRACE.
1748          */
1749         user_disable_single_step(p);
1750         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1751 #ifdef TIF_SYSCALL_EMU
1752         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1753 #endif
1754         clear_all_latency_tracing(p);
1755
1756         /* ok, now we should be set up.. */
1757         p->pid = pid_nr(pid);
1758         if (clone_flags & CLONE_THREAD) {
1759                 p->exit_signal = -1;
1760                 p->group_leader = current->group_leader;
1761                 p->tgid = current->tgid;
1762         } else {
1763                 if (clone_flags & CLONE_PARENT)
1764                         p->exit_signal = current->group_leader->exit_signal;
1765                 else
1766                         p->exit_signal = (clone_flags & CSIGNAL);
1767                 p->group_leader = p;
1768                 p->tgid = p->pid;
1769         }
1770
1771         p->nr_dirtied = 0;
1772         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1773         p->dirty_paused_when = 0;
1774
1775         p->pdeath_signal = 0;
1776         INIT_LIST_HEAD(&p->thread_group);
1777         p->task_works = NULL;
1778
1779         cgroup_threadgroup_change_begin(current);
1780         /*
1781          * Ensure that the cgroup subsystem policies allow the new process to be
1782          * forked. It should be noted the the new process's css_set can be changed
1783          * between here and cgroup_post_fork() if an organisation operation is in
1784          * progress.
1785          */
1786         retval = cgroup_can_fork(p);
1787         if (retval)
1788                 goto bad_fork_free_pid;
1789
1790         /*
1791          * Make it visible to the rest of the system, but dont wake it up yet.
1792          * Need tasklist lock for parent etc handling!
1793          */
1794         write_lock_irq(&tasklist_lock);
1795
1796         /* CLONE_PARENT re-uses the old parent */
1797         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1798                 p->real_parent = current->real_parent;
1799                 p->parent_exec_id = current->parent_exec_id;
1800         } else {
1801                 p->real_parent = current;
1802                 p->parent_exec_id = current->self_exec_id;
1803         }
1804
1805         klp_copy_process(p);
1806
1807         spin_lock(&current->sighand->siglock);
1808
1809         /*
1810          * Copy seccomp details explicitly here, in case they were changed
1811          * before holding sighand lock.
1812          */
1813         copy_seccomp(p);
1814
1815         /*
1816          * Process group and session signals need to be delivered to just the
1817          * parent before the fork or both the parent and the child after the
1818          * fork. Restart if a signal comes in before we add the new process to
1819          * it's process group.
1820          * A fatal signal pending means that current will exit, so the new
1821          * thread can't slip out of an OOM kill (or normal SIGKILL).
1822         */
1823         recalc_sigpending();
1824         if (signal_pending(current)) {
1825                 spin_unlock(&current->sighand->siglock);
1826                 write_unlock_irq(&tasklist_lock);
1827                 retval = -ERESTARTNOINTR;
1828                 goto bad_fork_cancel_cgroup;
1829         }
1830
1831         if (likely(p->pid)) {
1832                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1833
1834                 init_task_pid(p, PIDTYPE_PID, pid);
1835                 if (thread_group_leader(p)) {
1836                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1837                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1838
1839                         if (is_child_reaper(pid)) {
1840                                 ns_of_pid(pid)->child_reaper = p;
1841                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1842                         }
1843
1844                         p->signal->leader_pid = pid;
1845                         p->signal->tty = tty_kref_get(current->signal->tty);
1846                         /*
1847                          * Inherit has_child_subreaper flag under the same
1848                          * tasklist_lock with adding child to the process tree
1849                          * for propagate_has_child_subreaper optimization.
1850                          */
1851                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1852                                                          p->real_parent->signal->is_child_subreaper;
1853                         list_add_tail(&p->sibling, &p->real_parent->children);
1854                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1855                         attach_pid(p, PIDTYPE_PGID);
1856                         attach_pid(p, PIDTYPE_SID);
1857                         __this_cpu_inc(process_counts);
1858                 } else {
1859                         current->signal->nr_threads++;
1860                         atomic_inc(&current->signal->live);
1861                         atomic_inc(&current->signal->sigcnt);
1862                         list_add_tail_rcu(&p->thread_group,
1863                                           &p->group_leader->thread_group);
1864                         list_add_tail_rcu(&p->thread_node,
1865                                           &p->signal->thread_head);
1866                 }
1867                 attach_pid(p, PIDTYPE_PID);
1868                 nr_threads++;
1869         }
1870
1871         total_forks++;
1872         spin_unlock(&current->sighand->siglock);
1873         syscall_tracepoint_update(p);
1874         write_unlock_irq(&tasklist_lock);
1875
1876         proc_fork_connector(p);
1877         cgroup_post_fork(p);
1878         cgroup_threadgroup_change_end(current);
1879         perf_event_fork(p);
1880
1881         trace_task_newtask(p, clone_flags);
1882         uprobe_copy_process(p, clone_flags);
1883
1884         return p;
1885
1886 bad_fork_cancel_cgroup:
1887         cgroup_cancel_fork(p);
1888 bad_fork_free_pid:
1889         cgroup_threadgroup_change_end(current);
1890         if (pid != &init_struct_pid)
1891                 free_pid(pid);
1892 bad_fork_cleanup_thread:
1893         exit_thread(p);
1894 bad_fork_cleanup_io:
1895         if (p->io_context)
1896                 exit_io_context(p);
1897 bad_fork_cleanup_namespaces:
1898         exit_task_namespaces(p);
1899 bad_fork_cleanup_mm:
1900         if (p->mm)
1901                 mmput(p->mm);
1902 bad_fork_cleanup_signal:
1903         if (!(clone_flags & CLONE_THREAD))
1904                 free_signal_struct(p->signal);
1905 bad_fork_cleanup_sighand:
1906         __cleanup_sighand(p->sighand);
1907 bad_fork_cleanup_fs:
1908         exit_fs(p); /* blocking */
1909 bad_fork_cleanup_files:
1910         exit_files(p); /* blocking */
1911 bad_fork_cleanup_semundo:
1912         exit_sem(p);
1913 bad_fork_cleanup_security:
1914         security_task_free(p);
1915 bad_fork_cleanup_audit:
1916         audit_free(p);
1917 bad_fork_cleanup_perf:
1918         perf_event_free_task(p);
1919 bad_fork_cleanup_policy:
1920 #ifdef CONFIG_NUMA
1921         mpol_put(p->mempolicy);
1922 bad_fork_cleanup_threadgroup_lock:
1923 #endif
1924         delayacct_tsk_free(p);
1925 bad_fork_cleanup_count:
1926         atomic_dec(&p->cred->user->processes);
1927         exit_creds(p);
1928 bad_fork_free:
1929         p->state = TASK_DEAD;
1930         put_task_stack(p);
1931         free_task(p);
1932 fork_out:
1933         return ERR_PTR(retval);
1934 }
1935
1936 static inline void init_idle_pids(struct pid_link *links)
1937 {
1938         enum pid_type type;
1939
1940         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1941                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1942                 links[type].pid = &init_struct_pid;
1943         }
1944 }
1945
1946 struct task_struct *fork_idle(int cpu)
1947 {
1948         struct task_struct *task;
1949         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1950                             cpu_to_node(cpu));
1951         if (!IS_ERR(task)) {
1952                 init_idle_pids(task->pids);
1953                 init_idle(task, cpu);
1954         }
1955
1956         return task;
1957 }
1958
1959 /*
1960  *  Ok, this is the main fork-routine.
1961  *
1962  * It copies the process, and if successful kick-starts
1963  * it and waits for it to finish using the VM if required.
1964  */
1965 long _do_fork(unsigned long clone_flags,
1966               unsigned long stack_start,
1967               unsigned long stack_size,
1968               int __user *parent_tidptr,
1969               int __user *child_tidptr,
1970               unsigned long tls)
1971 {
1972         struct task_struct *p;
1973         int trace = 0;
1974         long nr;
1975
1976         /*
1977          * Determine whether and which event to report to ptracer.  When
1978          * called from kernel_thread or CLONE_UNTRACED is explicitly
1979          * requested, no event is reported; otherwise, report if the event
1980          * for the type of forking is enabled.
1981          */
1982         if (!(clone_flags & CLONE_UNTRACED)) {
1983                 if (clone_flags & CLONE_VFORK)
1984                         trace = PTRACE_EVENT_VFORK;
1985                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1986                         trace = PTRACE_EVENT_CLONE;
1987                 else
1988                         trace = PTRACE_EVENT_FORK;
1989
1990                 if (likely(!ptrace_event_enabled(current, trace)))
1991                         trace = 0;
1992         }
1993
1994         p = copy_process(clone_flags, stack_start, stack_size,
1995                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
1996         add_latent_entropy();
1997         /*
1998          * Do this prior waking up the new thread - the thread pointer
1999          * might get invalid after that point, if the thread exits quickly.
2000          */
2001         if (!IS_ERR(p)) {
2002                 struct completion vfork;
2003                 struct pid *pid;
2004
2005                 trace_sched_process_fork(current, p);
2006
2007                 pid = get_task_pid(p, PIDTYPE_PID);
2008                 nr = pid_vnr(pid);
2009
2010                 if (clone_flags & CLONE_PARENT_SETTID)
2011                         put_user(nr, parent_tidptr);
2012
2013                 if (clone_flags & CLONE_VFORK) {
2014                         p->vfork_done = &vfork;
2015                         init_completion(&vfork);
2016                         get_task_struct(p);
2017                 }
2018
2019                 wake_up_new_task(p);
2020
2021                 /* forking complete and child started to run, tell ptracer */
2022                 if (unlikely(trace))
2023                         ptrace_event_pid(trace, pid);
2024
2025                 if (clone_flags & CLONE_VFORK) {
2026                         if (!wait_for_vfork_done(p, &vfork))
2027                                 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2028                 }
2029
2030                 put_pid(pid);
2031         } else {
2032                 nr = PTR_ERR(p);
2033         }
2034         return nr;
2035 }
2036
2037 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2038 /* For compatibility with architectures that call do_fork directly rather than
2039  * using the syscall entry points below. */
2040 long do_fork(unsigned long clone_flags,
2041               unsigned long stack_start,
2042               unsigned long stack_size,
2043               int __user *parent_tidptr,
2044               int __user *child_tidptr)
2045 {
2046         return _do_fork(clone_flags, stack_start, stack_size,
2047                         parent_tidptr, child_tidptr, 0);
2048 }
2049 #endif
2050
2051 /*
2052  * Create a kernel thread.
2053  */
2054 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2055 {
2056         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2057                 (unsigned long)arg, NULL, NULL, 0);
2058 }
2059
2060 #ifdef __ARCH_WANT_SYS_FORK
2061 SYSCALL_DEFINE0(fork)
2062 {
2063 #ifdef CONFIG_MMU
2064         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2065 #else
2066         /* can not support in nommu mode */
2067         return -EINVAL;
2068 #endif
2069 }
2070 #endif
2071
2072 #ifdef __ARCH_WANT_SYS_VFORK
2073 SYSCALL_DEFINE0(vfork)
2074 {
2075         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2076                         0, NULL, NULL, 0);
2077 }
2078 #endif
2079
2080 #ifdef __ARCH_WANT_SYS_CLONE
2081 #ifdef CONFIG_CLONE_BACKWARDS
2082 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2083                  int __user *, parent_tidptr,
2084                  unsigned long, tls,
2085                  int __user *, child_tidptr)
2086 #elif defined(CONFIG_CLONE_BACKWARDS2)
2087 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2088                  int __user *, parent_tidptr,
2089                  int __user *, child_tidptr,
2090                  unsigned long, tls)
2091 #elif defined(CONFIG_CLONE_BACKWARDS3)
2092 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2093                 int, stack_size,
2094                 int __user *, parent_tidptr,
2095                 int __user *, child_tidptr,
2096                 unsigned long, tls)
2097 #else
2098 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2099                  int __user *, parent_tidptr,
2100                  int __user *, child_tidptr,
2101                  unsigned long, tls)
2102 #endif
2103 {
2104         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2105 }
2106 #endif
2107
2108 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2109 {
2110         struct task_struct *leader, *parent, *child;
2111         int res;
2112
2113         read_lock(&tasklist_lock);
2114         leader = top = top->group_leader;
2115 down:
2116         for_each_thread(leader, parent) {
2117                 list_for_each_entry(child, &parent->children, sibling) {
2118                         res = visitor(child, data);
2119                         if (res) {
2120                                 if (res < 0)
2121                                         goto out;
2122                                 leader = child;
2123                                 goto down;
2124                         }
2125 up:
2126                         ;
2127                 }
2128         }
2129
2130         if (leader != top) {
2131                 child = leader;
2132                 parent = child->real_parent;
2133                 leader = parent->group_leader;
2134                 goto up;
2135         }
2136 out:
2137         read_unlock(&tasklist_lock);
2138 }
2139
2140 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2141 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2142 #endif
2143
2144 static void sighand_ctor(void *data)
2145 {
2146         struct sighand_struct *sighand = data;
2147
2148         spin_lock_init(&sighand->siglock);
2149         init_waitqueue_head(&sighand->signalfd_wqh);
2150 }
2151
2152 void __init proc_caches_init(void)
2153 {
2154         sighand_cachep = kmem_cache_create("sighand_cache",
2155                         sizeof(struct sighand_struct), 0,
2156                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
2157                         SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2158         signal_cachep = kmem_cache_create("signal_cache",
2159                         sizeof(struct signal_struct), 0,
2160                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2161                         NULL);
2162         files_cachep = kmem_cache_create("files_cache",
2163                         sizeof(struct files_struct), 0,
2164                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2165                         NULL);
2166         fs_cachep = kmem_cache_create("fs_cache",
2167                         sizeof(struct fs_struct), 0,
2168                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2169                         NULL);
2170         /*
2171          * FIXME! The "sizeof(struct mm_struct)" currently includes the
2172          * whole struct cpumask for the OFFSTACK case. We could change
2173          * this to *only* allocate as much of it as required by the
2174          * maximum number of CPU's we can ever have.  The cpumask_allocation
2175          * is at the end of the structure, exactly for that reason.
2176          */
2177         mm_cachep = kmem_cache_create("mm_struct",
2178                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2179                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2180                         NULL);
2181         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2182         mmap_init();
2183         nsproxy_cache_init();
2184 }
2185
2186 /*
2187  * Check constraints on flags passed to the unshare system call.
2188  */
2189 static int check_unshare_flags(unsigned long unshare_flags)
2190 {
2191         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2192                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2193                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2194                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2195                 return -EINVAL;
2196         /*
2197          * Not implemented, but pretend it works if there is nothing
2198          * to unshare.  Note that unsharing the address space or the
2199          * signal handlers also need to unshare the signal queues (aka
2200          * CLONE_THREAD).
2201          */
2202         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2203                 if (!thread_group_empty(current))
2204                         return -EINVAL;
2205         }
2206         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2207                 if (atomic_read(&current->sighand->count) > 1)
2208                         return -EINVAL;
2209         }
2210         if (unshare_flags & CLONE_VM) {
2211                 if (!current_is_single_threaded())
2212                         return -EINVAL;
2213         }
2214
2215         return 0;
2216 }
2217
2218 /*
2219  * Unshare the filesystem structure if it is being shared
2220  */
2221 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2222 {
2223         struct fs_struct *fs = current->fs;
2224
2225         if (!(unshare_flags & CLONE_FS) || !fs)
2226                 return 0;
2227
2228         /* don't need lock here; in the worst case we'll do useless copy */
2229         if (fs->users == 1)
2230                 return 0;
2231
2232         *new_fsp = copy_fs_struct(fs);
2233         if (!*new_fsp)
2234                 return -ENOMEM;
2235
2236         return 0;
2237 }
2238
2239 /*
2240  * Unshare file descriptor table if it is being shared
2241  */
2242 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2243 {
2244         struct files_struct *fd = current->files;
2245         int error = 0;
2246
2247         if ((unshare_flags & CLONE_FILES) &&
2248             (fd && atomic_read(&fd->count) > 1)) {
2249                 *new_fdp = dup_fd(fd, &error);
2250                 if (!*new_fdp)
2251                         return error;
2252         }
2253
2254         return 0;
2255 }
2256
2257 /*
2258  * unshare allows a process to 'unshare' part of the process
2259  * context which was originally shared using clone.  copy_*
2260  * functions used by do_fork() cannot be used here directly
2261  * because they modify an inactive task_struct that is being
2262  * constructed. Here we are modifying the current, active,
2263  * task_struct.
2264  */
2265 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2266 {
2267         struct fs_struct *fs, *new_fs = NULL;
2268         struct files_struct *fd, *new_fd = NULL;
2269         struct cred *new_cred = NULL;
2270         struct nsproxy *new_nsproxy = NULL;
2271         int do_sysvsem = 0;
2272         int err;
2273
2274         /*
2275          * If unsharing a user namespace must also unshare the thread group
2276          * and unshare the filesystem root and working directories.
2277          */
2278         if (unshare_flags & CLONE_NEWUSER)
2279                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2280         /*
2281          * If unsharing vm, must also unshare signal handlers.
2282          */
2283         if (unshare_flags & CLONE_VM)
2284                 unshare_flags |= CLONE_SIGHAND;
2285         /*
2286          * If unsharing a signal handlers, must also unshare the signal queues.
2287          */
2288         if (unshare_flags & CLONE_SIGHAND)
2289                 unshare_flags |= CLONE_THREAD;
2290         /*
2291          * If unsharing namespace, must also unshare filesystem information.
2292          */
2293         if (unshare_flags & CLONE_NEWNS)
2294                 unshare_flags |= CLONE_FS;
2295
2296         err = check_unshare_flags(unshare_flags);
2297         if (err)
2298                 goto bad_unshare_out;
2299         /*
2300          * CLONE_NEWIPC must also detach from the undolist: after switching
2301          * to a new ipc namespace, the semaphore arrays from the old
2302          * namespace are unreachable.
2303          */
2304         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2305                 do_sysvsem = 1;
2306         err = unshare_fs(unshare_flags, &new_fs);
2307         if (err)
2308                 goto bad_unshare_out;
2309         err = unshare_fd(unshare_flags, &new_fd);
2310         if (err)
2311                 goto bad_unshare_cleanup_fs;
2312         err = unshare_userns(unshare_flags, &new_cred);
2313         if (err)
2314                 goto bad_unshare_cleanup_fd;
2315         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2316                                          new_cred, new_fs);
2317         if (err)
2318                 goto bad_unshare_cleanup_cred;
2319
2320         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2321                 if (do_sysvsem) {
2322                         /*
2323                          * CLONE_SYSVSEM is equivalent to sys_exit().
2324                          */
2325                         exit_sem(current);
2326                 }
2327                 if (unshare_flags & CLONE_NEWIPC) {
2328                         /* Orphan segments in old ns (see sem above). */
2329                         exit_shm(current);
2330                         shm_init_task(current);
2331                 }
2332
2333                 if (new_nsproxy)
2334                         switch_task_namespaces(current, new_nsproxy);
2335
2336                 task_lock(current);
2337
2338                 if (new_fs) {
2339                         fs = current->fs;
2340                         spin_lock(&fs->lock);
2341                         current->fs = new_fs;
2342                         if (--fs->users)
2343                                 new_fs = NULL;
2344                         else
2345                                 new_fs = fs;
2346                         spin_unlock(&fs->lock);
2347                 }
2348
2349                 if (new_fd) {
2350                         fd = current->files;
2351                         current->files = new_fd;
2352                         new_fd = fd;
2353                 }
2354
2355                 task_unlock(current);
2356
2357                 if (new_cred) {
2358                         /* Install the new user namespace */
2359                         commit_creds(new_cred);
2360                         new_cred = NULL;
2361                 }
2362         }
2363
2364         perf_event_namespaces(current);
2365
2366 bad_unshare_cleanup_cred:
2367         if (new_cred)
2368                 put_cred(new_cred);
2369 bad_unshare_cleanup_fd:
2370         if (new_fd)
2371                 put_files_struct(new_fd);
2372
2373 bad_unshare_cleanup_fs:
2374         if (new_fs)
2375                 free_fs_struct(new_fs);
2376
2377 bad_unshare_out:
2378         return err;
2379 }
2380
2381 /*
2382  *      Helper to unshare the files of the current task.
2383  *      We don't want to expose copy_files internals to
2384  *      the exec layer of the kernel.
2385  */
2386
2387 int unshare_files(struct files_struct **displaced)
2388 {
2389         struct task_struct *task = current;
2390         struct files_struct *copy = NULL;
2391         int error;
2392
2393         error = unshare_fd(CLONE_FILES, &copy);
2394         if (error || !copy) {
2395                 *displaced = NULL;
2396                 return error;
2397         }
2398         *displaced = task->files;
2399         task_lock(task);
2400         task->files = copy;
2401         task_unlock(task);
2402         return 0;
2403 }
2404
2405 int sysctl_max_threads(struct ctl_table *table, int write,
2406                        void __user *buffer, size_t *lenp, loff_t *ppos)
2407 {
2408         struct ctl_table t;
2409         int ret;
2410         int threads = max_threads;
2411         int min = MIN_THREADS;
2412         int max = MAX_THREADS;
2413
2414         t = *table;
2415         t.data = &threads;
2416         t.extra1 = &min;
2417         t.extra2 = &max;
2418
2419         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2420         if (ret || !write)
2421                 return ret;
2422
2423         set_max_threads(threads);
2424
2425         return 0;
2426 }