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