kernel/fork.c

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