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