kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/sched/autogroup.h>
  16 #include <linux/sched/mm.h>
  17 #include <linux/sched/coredump.h>
  18 #include <linux/sched/user.h>
  19 #include <linux/sched/numa_balancing.h>
  20 #include <linux/sched/stat.h>
  21 #include <linux/sched/task.h>
  22 #include <linux/sched/task_stack.h>
  23 #include <linux/sched/cputime.h>
  24 #include <linux/rtmutex.h>
  25 #include <linux/init.h>
  26 #include <linux/unistd.h>
  27 #include <linux/module.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/completion.h>
  30 #include <linux/personality.h>
  31 #include <linux/mempolicy.h>
  32 #include <linux/sem.h>
  33 #include <linux/file.h>
  34 #include <linux/fdtable.h>
  35 #include <linux/iocontext.h>
  36 #include <linux/key.h>
  37 #include <linux/binfmts.h>
  38 #include <linux/mman.h>
  39 #include <linux/mmu_notifier.h>
  40 #include <linux/hmm.h>
  41 #include <linux/fs.h>
  42 #include <linux/mm.h>
  43 #include <linux/vmacache.h>
  44 #include <linux/nsproxy.h>
  45 #include <linux/capability.h>
  46 #include <linux/cpu.h>
  47 #include <linux/cgroup.h>
  48 #include <linux/security.h>
  49 #include <linux/hugetlb.h>
  50 #include <linux/seccomp.h>
  51 #include <linux/swap.h>
  52 #include <linux/syscalls.h>
  53 #include <linux/jiffies.h>
  54 #include <linux/futex.h>
  55 #include <linux/compat.h>
  56 #include <linux/kthread.h>
  57 #include <linux/task_io_accounting_ops.h>
  58 #include <linux/rcupdate.h>
  59 #include <linux/ptrace.h>
  60 #include <linux/mount.h>
  61 #include <linux/audit.h>
  62 #include <linux/memcontrol.h>
  63 #include <linux/ftrace.h>
  64 #include <linux/proc_fs.h>
  65 #include <linux/profile.h>
  66 #include <linux/rmap.h>
  67 #include <linux/ksm.h>
  68 #include <linux/acct.h>
  69 #include <linux/userfaultfd_k.h>
  70 #include <linux/tsacct_kern.h>
  71 #include <linux/cn_proc.h>
  72 #include <linux/freezer.h>
  73 #include <linux/delayacct.h>
  74 #include <linux/taskstats_kern.h>
  75 #include <linux/random.h>
  76 #include <linux/tty.h>
  77 #include <linux/blkdev.h>
  78 #include <linux/fs_struct.h>
  79 #include <linux/magic.h>
  80 #include <linux/sched/mm.h>
  81 #include <linux/perf_event.h>
  82 #include <linux/posix-timers.h>
  83 #include <linux/user-return-notifier.h>
  84 #include <linux/oom.h>
  85 #include <linux/khugepaged.h>
  86 #include <linux/signalfd.h>
  87 #include <linux/uprobes.h>
  88 #include <linux/aio.h>
  89 #include <linux/compiler.h>
  90 #include <linux/sysctl.h>
  91 #include <linux/kcov.h>
  92 #include <linux/livepatch.h>
  93 #include <linux/thread_info.h>
  94
  95 #include <asm/pgtable.h>
  96 #include <asm/pgalloc.h>
  97 #include <linux/uaccess.h>
  98 #include <asm/mmu_context.h>
  99 #include <asm/cacheflush.h>
 100 #include <asm/tlbflush.h>
 101
 102 #include <trace/events/sched.h>
 103
 104 #define CREATE_TRACE_POINTS
 105 #include <trace/events/task.h>
 106
 107 /*
 108  * Minimum number of threads to boot the kernel
 109  */
 110 #define MIN_THREADS 20
 111
 112 /*
 113  * Maximum number of threads
 114  */
 115 #define MAX_THREADS FUTEX_TID_MASK
 116
 117 /*
 118  * Protected counters by write_lock_irq(&tasklist_lock)
 119  */
 120 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 121 int nr_threads;                 /* The idle threads do not count.. */
 122
 123 int max_threads;                /* tunable limit on nr_threads */
 124
 125 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 126
 127 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 128
 129 #ifdef CONFIG_PROVE_RCU
 130 int lockdep_tasklist_lock_is_held(void)
 131 {
 132         return lockdep_is_held(&tasklist_lock);
 133 }
 134 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 135 #endif /* #ifdef CONFIG_PROVE_RCU */
 136
 137 int nr_processes(void)
 138 {
 139         int cpu;
 140         int total = 0;
 141
 142         for_each_possible_cpu(cpu)
 143                 total += per_cpu(process_counts, cpu);
 144
 145         return total;
 146 }
 147
 148 void __weak arch_release_task_struct(struct task_struct *tsk)
 149 {
 150 }
 151
 152 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 153 static struct kmem_cache *task_struct_cachep;
 154
 155 static inline struct task_struct *alloc_task_struct_node(int node)
 156 {
 157         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 158 }
 159
 160 static inline void free_task_struct(struct task_struct *tsk)
 161 {
 162         kmem_cache_free(task_struct_cachep, tsk);
 163 }
 164 #endif
 165
 166 void __weak arch_release_thread_stack(unsigned long *stack)
 167 {
 168 }
 169
 170 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 171
 172 /*
 173  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 174  * kmemcache based allocator.
 175  */
 176 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 177
 178 #ifdef CONFIG_VMAP_STACK
 179 /*
 180  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 181  * flush.  Try to minimize the number of calls by caching stacks.
 182  */
 183 #define NR_CACHED_STACKS 2
 184 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 185
 186 static int free_vm_stack_cache(unsigned int cpu)
 187 {
 188         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 189         int i;
 190
 191         for (i = 0; i < NR_CACHED_STACKS; i++) {
 192                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 193
 194                 if (!vm_stack)
 195                         continue;
 196
 197                 vfree(vm_stack->addr);
 198                 cached_vm_stacks[i] = NULL;
 199         }
 200
 201         return 0;
 202 }
 203 #endif
 204
 205 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 206 {
 207 #ifdef CONFIG_VMAP_STACK
 208         void *stack;
 209         int i;
 210
 211         for (i = 0; i < NR_CACHED_STACKS; i++) {
 212                 struct vm_struct *s;
 213
 214                 s = this_cpu_xchg(cached_stacks[i], NULL);
 215
 216                 if (!s)
 217                         continue;
 218
 219                 /* Clear stale pointers from reused stack. */
 220                 memset(s->addr, 0, THREAD_SIZE);
 221
 222                 tsk->stack_vm_area = s;
 223                 return s->addr;
 224         }
 225
 226         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 227                                      VMALLOC_START, VMALLOC_END,
 228                                      THREADINFO_GFP,
 229                                      PAGE_KERNEL,
 230                                      0, node, __builtin_return_address(0));
 231
 232         /*
 233          * We can't call find_vm_area() in interrupt context, and
 234          * free_thread_stack() can be called in interrupt context,
 235          * so cache the vm_struct.
 236          */
 237         if (stack)
 238                 tsk->stack_vm_area = find_vm_area(stack);
 239         return stack;
 240 #else
 241         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 242                                              THREAD_SIZE_ORDER);
 243
 244         return page ? page_address(page) : NULL;
 245 #endif
 246 }
 247
 248 static inline void free_thread_stack(struct task_struct *tsk)
 249 {
 250 #ifdef CONFIG_VMAP_STACK
 251         if (task_stack_vm_area(tsk)) {
 252                 int i;
 253
 254                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 255                         if (this_cpu_cmpxchg(cached_stacks[i],
 256                                         NULL, tsk->stack_vm_area) != NULL)
 257                                 continue;
 258
 259                         return;
 260                 }
 261
 262                 vfree_atomic(tsk->stack);
 263                 return;
 264         }
 265 #endif
 266
 267         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 268 }
 269 # else
 270 static struct kmem_cache *thread_stack_cache;
 271
 272 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 273                                                   int node)
 274 {
 275         return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 276 }
 277
 278 static void free_thread_stack(struct task_struct *tsk)
 279 {
 280         kmem_cache_free(thread_stack_cache, tsk->stack);
 281 }
 282
 283 void thread_stack_cache_init(void)
 284 {
 285         thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 286                                         THREAD_SIZE, THREAD_SIZE, 0, 0,
 287                                         THREAD_SIZE, NULL);
 288         BUG_ON(thread_stack_cache == NULL);
 289 }
 290 # endif
 291 #endif
 292
 293 /* SLAB cache for signal_struct structures (tsk->signal) */
 294 static struct kmem_cache *signal_cachep;
 295
 296 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 297 struct kmem_cache *sighand_cachep;
 298
 299 /* SLAB cache for files_struct structures (tsk->files) */
 300 struct kmem_cache *files_cachep;
 301
 302 /* SLAB cache for fs_struct structures (tsk->fs) */
 303 struct kmem_cache *fs_cachep;
 304
 305 /* SLAB cache for vm_area_struct structures */
 306 static struct kmem_cache *vm_area_cachep;
 307
 308 /* SLAB cache for mm_struct structures (tsk->mm) */
 309 static struct kmem_cache *mm_cachep;
 310
 311 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 312 {
 313         struct vm_area_struct *vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 314
 315         if (vma) {
 316                 vma->vm_mm = mm;
 317                 INIT_LIST_HEAD(&vma->anon_vma_chain);
 318         }
 319         return vma;
 320 }
 321
 322 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 323 {
 324         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 325
 326         if (new) {
 327                 *new = *orig;
 328                 INIT_LIST_HEAD(&new->anon_vma_chain);
 329         }
 330         return new;
 331 }
 332
 333 void vm_area_free(struct vm_area_struct *vma)
 334 {
 335         kmem_cache_free(vm_area_cachep, vma);
 336 }
 337
 338 static void account_kernel_stack(struct task_struct *tsk, int account)
 339 {
 340         void *stack = task_stack_page(tsk);
 341         struct vm_struct *vm = task_stack_vm_area(tsk);
 342
 343         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 344
 345         if (vm) {
 346                 int i;
 347
 348                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 349
 350                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 351                         mod_zone_page_state(page_zone(vm->pages[i]),
 352                                             NR_KERNEL_STACK_KB,
 353                                             PAGE_SIZE / 1024 * account);
 354                 }
 355
 356                 /* All stack pages belong to the same memcg. */
 357                 mod_memcg_page_state(vm->pages[0], MEMCG_KERNEL_STACK_KB,
 358                                      account * (THREAD_SIZE / 1024));
 359         } else {
 360                 /*
 361                  * All stack pages are in the same zone and belong to the
 362                  * same memcg.
 363                  */
 364                 struct page *first_page = virt_to_page(stack);
 365
 366                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 367                                     THREAD_SIZE / 1024 * account);
 368
 369                 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
 370                                      account * (THREAD_SIZE / 1024));
 371         }
 372 }
 373
 374 static void release_task_stack(struct task_struct *tsk)
 375 {
 376         if (WARN_ON(tsk->state != TASK_DEAD))
 377                 return;  /* Better to leak the stack than to free prematurely */
 378
 379         account_kernel_stack(tsk, -1);
 380         arch_release_thread_stack(tsk->stack);
 381         free_thread_stack(tsk);
 382         tsk->stack = NULL;
 383 #ifdef CONFIG_VMAP_STACK
 384         tsk->stack_vm_area = NULL;
 385 #endif
 386 }
 387
 388 #ifdef CONFIG_THREAD_INFO_IN_TASK
 389 void put_task_stack(struct task_struct *tsk)
 390 {
 391         if (atomic_dec_and_test(&tsk->stack_refcount))
 392                 release_task_stack(tsk);
 393 }
 394 #endif
 395
 396 void free_task(struct task_struct *tsk)
 397 {
 398 #ifndef CONFIG_THREAD_INFO_IN_TASK
 399         /*
 400          * The task is finally done with both the stack and thread_info,
 401          * so free both.
 402          */
 403         release_task_stack(tsk);
 404 #else
 405         /*
 406          * If the task had a separate stack allocation, it should be gone
 407          * by now.
 408          */
 409         WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
 410 #endif
 411         rt_mutex_debug_task_free(tsk);
 412         ftrace_graph_exit_task(tsk);
 413         put_seccomp_filter(tsk);
 414         arch_release_task_struct(tsk);
 415         if (tsk->flags & PF_KTHREAD)
 416                 free_kthread_struct(tsk);
 417         free_task_struct(tsk);
 418 }
 419 EXPORT_SYMBOL(free_task);
 420
 421 #ifdef CONFIG_MMU
 422 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 423                                         struct mm_struct *oldmm)
 424 {
 425         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 426         struct rb_node **rb_link, *rb_parent;
 427         int retval;
 428         unsigned long charge;
 429         LIST_HEAD(uf);
 430
 431         uprobe_start_dup_mmap();
 432         if (down_write_killable(&oldmm->mmap_sem)) {
 433                 retval = -EINTR;
 434                 goto fail_uprobe_end;
 435         }
 436         flush_cache_dup_mm(oldmm);
 437         uprobe_dup_mmap(oldmm, mm);
 438         /*
 439          * Not linked in yet - no deadlock potential:
 440          */
 441         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 442
 443         /* No ordering required: file already has been exposed. */
 444         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 445
 446         mm->total_vm = oldmm->total_vm;
 447         mm->data_vm = oldmm->data_vm;
 448         mm->exec_vm = oldmm->exec_vm;
 449         mm->stack_vm = oldmm->stack_vm;
 450
 451         rb_link = &mm->mm_rb.rb_node;
 452         rb_parent = NULL;
 453         pprev = &mm->mmap;
 454         retval = ksm_fork(mm, oldmm);
 455         if (retval)
 456                 goto out;
 457         retval = khugepaged_fork(mm, oldmm);
 458         if (retval)
 459                 goto out;
 460
 461         prev = NULL;
 462         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 463                 struct file *file;
 464
 465                 if (mpnt->vm_flags & VM_DONTCOPY) {
 466                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 467                         continue;
 468                 }
 469                 charge = 0;
 470                 /*
 471                  * Don't duplicate many vmas if we've been oom-killed (for
 472                  * example)
 473                  */
 474                 if (fatal_signal_pending(current)) {
 475                         retval = -EINTR;
 476                         goto out;
 477                 }
 478                 if (mpnt->vm_flags & VM_ACCOUNT) {
 479                         unsigned long len = vma_pages(mpnt);
 480
 481                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 482                                 goto fail_nomem;
 483                         charge = len;
 484                 }
 485                 tmp = vm_area_dup(mpnt);
 486                 if (!tmp)
 487                         goto fail_nomem;
 488                 retval = vma_dup_policy(mpnt, tmp);
 489                 if (retval)
 490                         goto fail_nomem_policy;
 491                 tmp->vm_mm = mm;
 492                 retval = dup_userfaultfd(tmp, &uf);
 493                 if (retval)
 494                         goto fail_nomem_anon_vma_fork;
 495                 if (tmp->vm_flags & VM_WIPEONFORK) {
 496                         /* VM_WIPEONFORK gets a clean slate in the child. */
 497                         tmp->anon_vma = NULL;
 498                         if (anon_vma_prepare(tmp))
 499                                 goto fail_nomem_anon_vma_fork;
 500                 } else if (anon_vma_fork(tmp, mpnt))
 501                         goto fail_nomem_anon_vma_fork;
 502                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 503                 tmp->vm_next = tmp->vm_prev = NULL;
 504                 file = tmp->vm_file;
 505                 if (file) {
 506                         struct inode *inode = file_inode(file);
 507                         struct address_space *mapping = file->f_mapping;
 508
 509                         get_file(file);
 510                         if (tmp->vm_flags & VM_DENYWRITE)
 511                                 atomic_dec(&inode->i_writecount);
 512                         i_mmap_lock_write(mapping);
 513                         if (tmp->vm_flags & VM_SHARED)
 514                                 atomic_inc(&mapping->i_mmap_writable);
 515                         flush_dcache_mmap_lock(mapping);
 516                         /* insert tmp into the share list, just after mpnt */
 517                         vma_interval_tree_insert_after(tmp, mpnt,
 518                                         &mapping->i_mmap);
 519                         flush_dcache_mmap_unlock(mapping);
 520                         i_mmap_unlock_write(mapping);
 521                 }
 522
 523                 /*
 524                  * Clear hugetlb-related page reserves for children. This only
 525                  * affects MAP_PRIVATE mappings. Faults generated by the child
 526                  * are not guaranteed to succeed, even if read-only
 527                  */
 528                 if (is_vm_hugetlb_page(tmp))
 529                         reset_vma_resv_huge_pages(tmp);
 530
 531                 /*
 532                  * Link in the new vma and copy the page table entries.
 533                  */
 534                 *pprev = tmp;
 535                 pprev = &tmp->vm_next;
 536                 tmp->vm_prev = prev;
 537                 prev = tmp;
 538
 539                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 540                 rb_link = &tmp->vm_rb.rb_right;
 541                 rb_parent = &tmp->vm_rb;
 542
 543                 mm->map_count++;
 544                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 545                         retval = copy_page_range(mm, oldmm, mpnt);
 546
 547                 if (tmp->vm_ops && tmp->vm_ops->open)
 548                         tmp->vm_ops->open(tmp);
 549
 550                 if (retval)
 551                         goto out;
 552         }
 553         /* a new mm has just been created */
 554         arch_dup_mmap(oldmm, mm);
 555         retval = 0;
 556 out:
 557         up_write(&mm->mmap_sem);
 558         flush_tlb_mm(oldmm);
 559         up_write(&oldmm->mmap_sem);
 560         dup_userfaultfd_complete(&uf);
 561 fail_uprobe_end:
 562         uprobe_end_dup_mmap();
 563         return retval;
 564 fail_nomem_anon_vma_fork:
 565         mpol_put(vma_policy(tmp));
 566 fail_nomem_policy:
 567         vm_area_free(tmp);
 568 fail_nomem:
 569         retval = -ENOMEM;
 570         vm_unacct_memory(charge);
 571         goto out;
 572 }
 573
 574 static inline int mm_alloc_pgd(struct mm_struct *mm)
 575 {
 576         mm->pgd = pgd_alloc(mm);
 577         if (unlikely(!mm->pgd))
 578                 return -ENOMEM;
 579         return 0;
 580 }
 581
 582 static inline void mm_free_pgd(struct mm_struct *mm)
 583 {
 584         pgd_free(mm, mm->pgd);
 585 }
 586 #else
 587 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 588 {
 589         down_write(&oldmm->mmap_sem);
 590         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 591         up_write(&oldmm->mmap_sem);
 592         return 0;
 593 }
 594 #define mm_alloc_pgd(mm)        (0)
 595 #define mm_free_pgd(mm)
 596 #endif /* CONFIG_MMU */
 597
 598 static void check_mm(struct mm_struct *mm)
 599 {
 600         int i;
 601
 602         for (i = 0; i < NR_MM_COUNTERS; i++) {
 603                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 604
 605                 if (unlikely(x))
 606                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 607                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 608         }
 609
 610         if (mm_pgtables_bytes(mm))
 611                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 612                                 mm_pgtables_bytes(mm));
 613
 614 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 615         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 616 #endif
 617 }
 618
 619 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 620 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 621
 622 /*
 623  * Called when the last reference to the mm
 624  * is dropped: either by a lazy thread or by
 625  * mmput. Free the page directory and the mm.
 626  */
 627 void __mmdrop(struct mm_struct *mm)
 628 {
 629         BUG_ON(mm == &init_mm);
 630         WARN_ON_ONCE(mm == current->mm);
 631         WARN_ON_ONCE(mm == current->active_mm);
 632         mm_free_pgd(mm);
 633         destroy_context(mm);
 634         hmm_mm_destroy(mm);
 635         mmu_notifier_mm_destroy(mm);
 636         check_mm(mm);
 637         put_user_ns(mm->user_ns);
 638         free_mm(mm);
 639 }
 640 EXPORT_SYMBOL_GPL(__mmdrop);
 641
 642 static void mmdrop_async_fn(struct work_struct *work)
 643 {
 644         struct mm_struct *mm;
 645
 646         mm = container_of(work, struct mm_struct, async_put_work);
 647         __mmdrop(mm);
 648 }
 649
 650 static void mmdrop_async(struct mm_struct *mm)
 651 {
 652         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 653                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 654                 schedule_work(&mm->async_put_work);
 655         }
 656 }
 657
 658 static inline void free_signal_struct(struct signal_struct *sig)
 659 {
 660         taskstats_tgid_free(sig);
 661         sched_autogroup_exit(sig);
 662         /*
 663          * __mmdrop is not safe to call from softirq context on x86 due to
 664          * pgd_dtor so postpone it to the async context
 665          */
 666         if (sig->oom_mm)
 667                 mmdrop_async(sig->oom_mm);
 668         kmem_cache_free(signal_cachep, sig);
 669 }
 670
 671 static inline void put_signal_struct(struct signal_struct *sig)
 672 {
 673         if (atomic_dec_and_test(&sig->sigcnt))
 674                 free_signal_struct(sig);
 675 }
 676
 677 void __put_task_struct(struct task_struct *tsk)
 678 {
 679         WARN_ON(!tsk->exit_state);
 680         WARN_ON(atomic_read(&tsk->usage));
 681         WARN_ON(tsk == current);
 682
 683         cgroup_free(tsk);
 684         task_numa_free(tsk);
 685         security_task_free(tsk);
 686         exit_creds(tsk);
 687         delayacct_tsk_free(tsk);
 688         put_signal_struct(tsk->signal);
 689
 690         if (!profile_handoff_task(tsk))
 691                 free_task(tsk);
 692 }
 693 EXPORT_SYMBOL_GPL(__put_task_struct);
 694
 695 void __init __weak arch_task_cache_init(void) { }
 696
 697 /*
 698  * set_max_threads
 699  */
 700 static void set_max_threads(unsigned int max_threads_suggested)
 701 {
 702         u64 threads;
 703
 704         /*
 705          * The number of threads shall be limited such that the thread
 706          * structures may only consume a small part of the available memory.
 707          */
 708         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 709                 threads = MAX_THREADS;
 710         else
 711                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 712                                     (u64) THREAD_SIZE * 8UL);
 713
 714         if (threads > max_threads_suggested)
 715                 threads = max_threads_suggested;
 716
 717         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 718 }
 719
 720 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 721 /* Initialized by the architecture: */
 722 int arch_task_struct_size __read_mostly;
 723 #endif
 724
 725 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 726 {
 727         /* Fetch thread_struct whitelist for the architecture. */
 728         arch_thread_struct_whitelist(offset, size);
 729
 730         /*
 731          * Handle zero-sized whitelist or empty thread_struct, otherwise
 732          * adjust offset to position of thread_struct in task_struct.
 733          */
 734         if (unlikely(*size == 0))
 735                 *offset = 0;
 736         else
 737                 *offset += offsetof(struct task_struct, thread);
 738 }
 739
 740 void __init fork_init(void)
 741 {
 742         int i;
 743 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 744 #ifndef ARCH_MIN_TASKALIGN
 745 #define ARCH_MIN_TASKALIGN      0
 746 #endif
 747         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 748         unsigned long useroffset, usersize;
 749
 750         /* create a slab on which task_structs can be allocated */
 751         task_struct_whitelist(&useroffset, &usersize);
 752         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 753                         arch_task_struct_size, align,
 754                         SLAB_PANIC|SLAB_ACCOUNT,
 755                         useroffset, usersize, NULL);
 756 #endif
 757
 758         /* do the arch specific task caches init */
 759         arch_task_cache_init();
 760
 761         set_max_threads(MAX_THREADS);
 762
 763         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 764         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 765         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 766                 init_task.signal->rlim[RLIMIT_NPROC];
 767
 768         for (i = 0; i < UCOUNT_COUNTS; i++) {
 769                 init_user_ns.ucount_max[i] = max_threads/2;
 770         }
 771
 772 #ifdef CONFIG_VMAP_STACK
 773         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 774                           NULL, free_vm_stack_cache);
 775 #endif
 776
 777         lockdep_init_task(&init_task);
 778 }
 779
 780 int __weak arch_dup_task_struct(struct task_struct *dst,
 781                                                struct task_struct *src)
 782 {
 783         *dst = *src;
 784         return 0;
 785 }
 786
 787 void set_task_stack_end_magic(struct task_struct *tsk)
 788 {
 789         unsigned long *stackend;
 790
 791         stackend = end_of_stack(tsk);
 792         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 793 }
 794
 795 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 796 {
 797         struct task_struct *tsk;
 798         unsigned long *stack;
 799         struct vm_struct *stack_vm_area;
 800         int err;
 801
 802         if (node == NUMA_NO_NODE)
 803                 node = tsk_fork_get_node(orig);
 804         tsk = alloc_task_struct_node(node);
 805         if (!tsk)
 806                 return NULL;
 807
 808         stack = alloc_thread_stack_node(tsk, node);
 809         if (!stack)
 810                 goto free_tsk;
 811
 812         stack_vm_area = task_stack_vm_area(tsk);
 813
 814         err = arch_dup_task_struct(tsk, orig);
 815
 816         /*
 817          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 818          * sure they're properly initialized before using any stack-related
 819          * functions again.
 820          */
 821         tsk->stack = stack;
 822 #ifdef CONFIG_VMAP_STACK
 823         tsk->stack_vm_area = stack_vm_area;
 824 #endif
 825 #ifdef CONFIG_THREAD_INFO_IN_TASK
 826         atomic_set(&tsk->stack_refcount, 1);
 827 #endif
 828
 829         if (err)
 830                 goto free_stack;
 831
 832 #ifdef CONFIG_SECCOMP
 833         /*
 834          * We must handle setting up seccomp filters once we're under
 835          * the sighand lock in case orig has changed between now and
 836          * then. Until then, filter must be NULL to avoid messing up
 837          * the usage counts on the error path calling free_task.
 838          */
 839         tsk->seccomp.filter = NULL;
 840 #endif
 841
 842         setup_thread_stack(tsk, orig);
 843         clear_user_return_notifier(tsk);
 844         clear_tsk_need_resched(tsk);
 845         set_task_stack_end_magic(tsk);
 846
 847 #ifdef CONFIG_STACKPROTECTOR
 848         tsk->stack_canary = get_random_canary();
 849 #endif
 850
 851         /*
 852          * One for us, one for whoever does the "release_task()" (usually
 853          * parent)
 854          */
 855         atomic_set(&tsk->usage, 2);
 856 #ifdef CONFIG_BLK_DEV_IO_TRACE
 857         tsk->btrace_seq = 0;
 858 #endif
 859         tsk->splice_pipe = NULL;
 860         tsk->task_frag.page = NULL;
 861         tsk->wake_q.next = NULL;
 862
 863         account_kernel_stack(tsk, 1);
 864
 865         kcov_task_init(tsk);
 866
 867 #ifdef CONFIG_FAULT_INJECTION
 868         tsk->fail_nth = 0;
 869 #endif
 870
 871         return tsk;
 872
 873 free_stack:
 874         free_thread_stack(tsk);
 875 free_tsk:
 876         free_task_struct(tsk);
 877         return NULL;
 878 }
 879
 880 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 881
 882 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 883
 884 static int __init coredump_filter_setup(char *s)
 885 {
 886         default_dump_filter =
 887                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 888                 MMF_DUMP_FILTER_MASK;
 889         return 1;
 890 }
 891
 892 __setup("coredump_filter=", coredump_filter_setup);
 893
 894 #include <linux/init_task.h>
 895
 896 static void mm_init_aio(struct mm_struct *mm)
 897 {
 898 #ifdef CONFIG_AIO
 899         spin_lock_init(&mm->ioctx_lock);
 900         mm->ioctx_table = NULL;
 901 #endif
 902 }
 903
 904 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 905 {
 906 #ifdef CONFIG_MEMCG
 907         mm->owner = p;
 908 #endif
 909 }
 910
 911 static void mm_init_uprobes_state(struct mm_struct *mm)
 912 {
 913 #ifdef CONFIG_UPROBES
 914         mm->uprobes_state.xol_area = NULL;
 915 #endif
 916 }
 917
 918 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 919         struct user_namespace *user_ns)
 920 {
 921         mm->mmap = NULL;
 922         mm->mm_rb = RB_ROOT;
 923         mm->vmacache_seqnum = 0;
 924         atomic_set(&mm->mm_users, 1);
 925         atomic_set(&mm->mm_count, 1);
 926         init_rwsem(&mm->mmap_sem);
 927         INIT_LIST_HEAD(&mm->mmlist);
 928         mm->core_state = NULL;
 929         mm_pgtables_bytes_init(mm);
 930         mm->map_count = 0;
 931         mm->locked_vm = 0;
 932         mm->pinned_vm = 0;
 933         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 934         spin_lock_init(&mm->page_table_lock);
 935         spin_lock_init(&mm->arg_lock);
 936         mm_init_cpumask(mm);
 937         mm_init_aio(mm);
 938         mm_init_owner(mm, p);
 939         RCU_INIT_POINTER(mm->exe_file, NULL);
 940         mmu_notifier_mm_init(mm);
 941         hmm_mm_init(mm);
 942         init_tlb_flush_pending(mm);
 943 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 944         mm->pmd_huge_pte = NULL;
 945 #endif
 946         mm_init_uprobes_state(mm);
 947
 948         if (current->mm) {
 949                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 950                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 951         } else {
 952                 mm->flags = default_dump_filter;
 953                 mm->def_flags = 0;
 954         }
 955
 956         if (mm_alloc_pgd(mm))
 957                 goto fail_nopgd;
 958
 959         if (init_new_context(p, mm))
 960                 goto fail_nocontext;
 961
 962         mm->user_ns = get_user_ns(user_ns);
 963         return mm;
 964
 965 fail_nocontext:
 966         mm_free_pgd(mm);
 967 fail_nopgd:
 968         free_mm(mm);
 969         return NULL;
 970 }
 971
 972 /*
 973  * Allocate and initialize an mm_struct.
 974  */
 975 struct mm_struct *mm_alloc(void)
 976 {
 977         struct mm_struct *mm;
 978
 979         mm = allocate_mm();
 980         if (!mm)
 981                 return NULL;
 982
 983         memset(mm, 0, sizeof(*mm));
 984         return mm_init(mm, current, current_user_ns());
 985 }
 986
 987 static inline void __mmput(struct mm_struct *mm)
 988 {
 989         VM_BUG_ON(atomic_read(&mm->mm_users));
 990
 991         uprobe_clear_state(mm);
 992         exit_aio(mm);
 993         ksm_exit(mm);
 994         khugepaged_exit(mm); /* must run before exit_mmap */
 995         exit_mmap(mm);
 996         mm_put_huge_zero_page(mm);
 997         set_mm_exe_file(mm, NULL);
 998         if (!list_empty(&mm->mmlist)) {
 999                 spin_lock(&mmlist_lock);
1000                 list_del(&mm->mmlist);
1001                 spin_unlock(&mmlist_lock);
1002         }
1003         if (mm->binfmt)
1004                 module_put(mm->binfmt->module);
1005         mmdrop(mm);
1006 }
1007
1008 /*
1009  * Decrement the use count and release all resources for an mm.
1010  */
1011 void mmput(struct mm_struct *mm)
1012 {
1013         might_sleep();
1014
1015         if (atomic_dec_and_test(&mm->mm_users))
1016                 __mmput(mm);
1017 }
1018 EXPORT_SYMBOL_GPL(mmput);
1019
1020 #ifdef CONFIG_MMU
1021 static void mmput_async_fn(struct work_struct *work)
1022 {
1023         struct mm_struct *mm = container_of(work, struct mm_struct,
1024                                             async_put_work);
1025
1026         __mmput(mm);
1027 }
1028
1029 void mmput_async(struct mm_struct *mm)
1030 {
1031         if (atomic_dec_and_test(&mm->mm_users)) {
1032                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1033                 schedule_work(&mm->async_put_work);
1034         }
1035 }
1036 #endif
1037
1038 /**
1039  * set_mm_exe_file - change a reference to the mm's executable file
1040  *
1041  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1042  *
1043  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1044  * invocations: in mmput() nobody alive left, in execve task is single
1045  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1046  * mm->exe_file, but does so without using set_mm_exe_file() in order
1047  * to do avoid the need for any locks.
1048  */
1049 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1050 {
1051         struct file *old_exe_file;
1052
1053         /*
1054          * It is safe to dereference the exe_file without RCU as
1055          * this function is only called if nobody else can access
1056          * this mm -- see comment above for justification.
1057          */
1058         old_exe_file = rcu_dereference_raw(mm->exe_file);
1059
1060         if (new_exe_file)
1061                 get_file(new_exe_file);
1062         rcu_assign_pointer(mm->exe_file, new_exe_file);
1063         if (old_exe_file)
1064                 fput(old_exe_file);
1065 }
1066
1067 /**
1068  * get_mm_exe_file - acquire a reference to the mm's executable file
1069  *
1070  * Returns %NULL if mm has no associated executable file.
1071  * User must release file via fput().
1072  */
1073 struct file *get_mm_exe_file(struct mm_struct *mm)
1074 {
1075         struct file *exe_file;
1076
1077         rcu_read_lock();
1078         exe_file = rcu_dereference(mm->exe_file);
1079         if (exe_file && !get_file_rcu(exe_file))
1080                 exe_file = NULL;
1081         rcu_read_unlock();
1082         return exe_file;
1083 }
1084 EXPORT_SYMBOL(get_mm_exe_file);
1085
1086 /**
1087  * get_task_exe_file - acquire a reference to the task's executable file
1088  *
1089  * Returns %NULL if task's mm (if any) has no associated executable file or
1090  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1091  * User must release file via fput().
1092  */
1093 struct file *get_task_exe_file(struct task_struct *task)
1094 {
1095         struct file *exe_file = NULL;
1096         struct mm_struct *mm;
1097
1098         task_lock(task);
1099         mm = task->mm;
1100         if (mm) {
1101                 if (!(task->flags & PF_KTHREAD))
1102                         exe_file = get_mm_exe_file(mm);
1103         }
1104         task_unlock(task);
1105         return exe_file;
1106 }
1107 EXPORT_SYMBOL(get_task_exe_file);
1108
1109 /**
1110  * get_task_mm - acquire a reference to the task's mm
1111  *
1112  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1113  * this kernel workthread has transiently adopted a user mm with use_mm,
1114  * to do its AIO) is not set and if so returns a reference to it, after
1115  * bumping up the use count.  User must release the mm via mmput()
1116  * after use.  Typically used by /proc and ptrace.
1117  */
1118 struct mm_struct *get_task_mm(struct task_struct *task)
1119 {
1120         struct mm_struct *mm;
1121
1122         task_lock(task);
1123         mm = task->mm;
1124         if (mm) {
1125                 if (task->flags & PF_KTHREAD)
1126                         mm = NULL;
1127                 else
1128                         mmget(mm);
1129         }
1130         task_unlock(task);
1131         return mm;
1132 }
1133 EXPORT_SYMBOL_GPL(get_task_mm);
1134
1135 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1136 {
1137         struct mm_struct *mm;
1138         int err;
1139
1140         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1141         if (err)
1142                 return ERR_PTR(err);
1143
1144         mm = get_task_mm(task);
1145         if (mm && mm != current->mm &&
1146                         !ptrace_may_access(task, mode)) {
1147                 mmput(mm);
1148                 mm = ERR_PTR(-EACCES);
1149         }
1150         mutex_unlock(&task->signal->cred_guard_mutex);
1151
1152         return mm;
1153 }
1154
1155 static void complete_vfork_done(struct task_struct *tsk)
1156 {
1157         struct completion *vfork;
1158
1159         task_lock(tsk);
1160         vfork = tsk->vfork_done;
1161         if (likely(vfork)) {
1162                 tsk->vfork_done = NULL;
1163                 complete(vfork);
1164         }
1165         task_unlock(tsk);
1166 }
1167
1168 static int wait_for_vfork_done(struct task_struct *child,
1169                                 struct completion *vfork)
1170 {
1171         int killed;
1172
1173         freezer_do_not_count();
1174         killed = wait_for_completion_killable(vfork);
1175         freezer_count();
1176
1177         if (killed) {
1178                 task_lock(child);
1179                 child->vfork_done = NULL;
1180                 task_unlock(child);
1181         }
1182
1183         put_task_struct(child);
1184         return killed;
1185 }
1186
1187 /* Please note the differences between mmput and mm_release.
1188  * mmput is called whenever we stop holding onto a mm_struct,
1189  * error success whatever.
1190  *
1191  * mm_release is called after a mm_struct has been removed
1192  * from the current process.
1193  *
1194  * This difference is important for error handling, when we
1195  * only half set up a mm_struct for a new process and need to restore
1196  * the old one.  Because we mmput the new mm_struct before
1197  * restoring the old one. . .
1198  * Eric Biederman 10 January 1998
1199  */
1200 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1201 {
1202         /* Get rid of any futexes when releasing the mm */
1203 #ifdef CONFIG_FUTEX
1204         if (unlikely(tsk->robust_list)) {
1205                 exit_robust_list(tsk);
1206                 tsk->robust_list = NULL;
1207         }
1208 #ifdef CONFIG_COMPAT
1209         if (unlikely(tsk->compat_robust_list)) {
1210                 compat_exit_robust_list(tsk);
1211                 tsk->compat_robust_list = NULL;
1212         }
1213 #endif
1214         if (unlikely(!list_empty(&tsk->pi_state_list)))
1215                 exit_pi_state_list(tsk);
1216 #endif
1217
1218         uprobe_free_utask(tsk);
1219
1220         /* Get rid of any cached register state */
1221         deactivate_mm(tsk, mm);
1222
1223         /*
1224          * Signal userspace if we're not exiting with a core dump
1225          * because we want to leave the value intact for debugging
1226          * purposes.
1227          */
1228         if (tsk->clear_child_tid) {
1229                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1230                     atomic_read(&mm->mm_users) > 1) {
1231                         /*
1232                          * We don't check the error code - if userspace has
1233                          * not set up a proper pointer then tough luck.
1234                          */
1235                         put_user(0, tsk->clear_child_tid);
1236                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1237                                         1, NULL, NULL, 0, 0);
1238                 }
1239                 tsk->clear_child_tid = NULL;
1240         }
1241
1242         /*
1243          * All done, finally we can wake up parent and return this mm to him.
1244          * Also kthread_stop() uses this completion for synchronization.
1245          */
1246         if (tsk->vfork_done)
1247                 complete_vfork_done(tsk);
1248 }
1249
1250 /*
1251  * Allocate a new mm structure and copy contents from the
1252  * mm structure of the passed in task structure.
1253  */
1254 static struct mm_struct *dup_mm(struct task_struct *tsk)
1255 {
1256         struct mm_struct *mm, *oldmm = current->mm;
1257         int err;
1258
1259         mm = allocate_mm();
1260         if (!mm)
1261                 goto fail_nomem;
1262
1263         memcpy(mm, oldmm, sizeof(*mm));
1264
1265         if (!mm_init(mm, tsk, mm->user_ns))
1266                 goto fail_nomem;
1267
1268         err = dup_mmap(mm, oldmm);
1269         if (err)
1270                 goto free_pt;
1271
1272         mm->hiwater_rss = get_mm_rss(mm);
1273         mm->hiwater_vm = mm->total_vm;
1274
1275         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1276                 goto free_pt;
1277
1278         return mm;
1279
1280 free_pt:
1281         /* don't put binfmt in mmput, we haven't got module yet */
1282         mm->binfmt = NULL;
1283         mmput(mm);
1284
1285 fail_nomem:
1286         return NULL;
1287 }
1288
1289 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1290 {
1291         struct mm_struct *mm, *oldmm;
1292         int retval;
1293
1294         tsk->min_flt = tsk->maj_flt = 0;
1295         tsk->nvcsw = tsk->nivcsw = 0;
1296 #ifdef CONFIG_DETECT_HUNG_TASK
1297         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1298 #endif
1299
1300         tsk->mm = NULL;
1301         tsk->active_mm = NULL;
1302
1303         /*
1304          * Are we cloning a kernel thread?
1305          *
1306          * We need to steal a active VM for that..
1307          */
1308         oldmm = current->mm;
1309         if (!oldmm)
1310                 return 0;
1311
1312         /* initialize the new vmacache entries */
1313         vmacache_flush(tsk);
1314
1315         if (clone_flags & CLONE_VM) {
1316                 mmget(oldmm);
1317                 mm = oldmm;
1318                 goto good_mm;
1319         }
1320
1321         retval = -ENOMEM;
1322         mm = dup_mm(tsk);
1323         if (!mm)
1324                 goto fail_nomem;
1325
1326 good_mm:
1327         tsk->mm = mm;
1328         tsk->active_mm = mm;
1329         return 0;
1330
1331 fail_nomem:
1332         return retval;
1333 }
1334
1335 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1336 {
1337         struct fs_struct *fs = current->fs;
1338         if (clone_flags & CLONE_FS) {
1339                 /* tsk->fs is already what we want */
1340                 spin_lock(&fs->lock);
1341                 if (fs->in_exec) {
1342                         spin_unlock(&fs->lock);
1343                         return -EAGAIN;
1344                 }
1345                 fs->users++;
1346                 spin_unlock(&fs->lock);
1347                 return 0;
1348         }
1349         tsk->fs = copy_fs_struct(fs);
1350         if (!tsk->fs)
1351                 return -ENOMEM;
1352         return 0;
1353 }
1354
1355 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1356 {
1357         struct files_struct *oldf, *newf;
1358         int error = 0;
1359
1360         /*
1361          * A background process may not have any files ...
1362          */
1363         oldf = current->files;
1364         if (!oldf)
1365                 goto out;
1366
1367         if (clone_flags & CLONE_FILES) {
1368                 atomic_inc(&oldf->count);
1369                 goto out;
1370         }
1371
1372         newf = dup_fd(oldf, &error);
1373         if (!newf)
1374                 goto out;
1375
1376         tsk->files = newf;
1377         error = 0;
1378 out:
1379         return error;
1380 }
1381
1382 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1383 {
1384 #ifdef CONFIG_BLOCK
1385         struct io_context *ioc = current->io_context;
1386         struct io_context *new_ioc;
1387
1388         if (!ioc)
1389                 return 0;
1390         /*
1391          * Share io context with parent, if CLONE_IO is set
1392          */
1393         if (clone_flags & CLONE_IO) {
1394                 ioc_task_link(ioc);
1395                 tsk->io_context = ioc;
1396         } else if (ioprio_valid(ioc->ioprio)) {
1397                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1398                 if (unlikely(!new_ioc))
1399                         return -ENOMEM;
1400
1401                 new_ioc->ioprio = ioc->ioprio;
1402                 put_io_context(new_ioc);
1403         }
1404 #endif
1405         return 0;
1406 }
1407
1408 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1409 {
1410         struct sighand_struct *sig;
1411
1412         if (clone_flags & CLONE_SIGHAND) {
1413                 atomic_inc(&current->sighand->count);
1414                 return 0;
1415         }
1416         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1417         rcu_assign_pointer(tsk->sighand, sig);
1418         if (!sig)
1419                 return -ENOMEM;
1420
1421         atomic_set(&sig->count, 1);
1422         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1423         return 0;
1424 }
1425
1426 void __cleanup_sighand(struct sighand_struct *sighand)
1427 {
1428         if (atomic_dec_and_test(&sighand->count)) {
1429                 signalfd_cleanup(sighand);
1430                 /*
1431                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1432                  * without an RCU grace period, see __lock_task_sighand().
1433                  */
1434                 kmem_cache_free(sighand_cachep, sighand);
1435         }
1436 }
1437
1438 #ifdef CONFIG_POSIX_TIMERS
1439 /*
1440  * Initialize POSIX timer handling for a thread group.
1441  */
1442 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1443 {
1444         unsigned long cpu_limit;
1445
1446         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1447         if (cpu_limit != RLIM_INFINITY) {
1448                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1449                 sig->cputimer.running = true;
1450         }
1451
1452         /* The timer lists. */
1453         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1454         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1455         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1456 }
1457 #else
1458 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1459 #endif
1460
1461 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1462 {
1463         struct signal_struct *sig;
1464
1465         if (clone_flags & CLONE_THREAD)
1466                 return 0;
1467
1468         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1469         tsk->signal = sig;
1470         if (!sig)
1471                 return -ENOMEM;
1472
1473         sig->nr_threads = 1;
1474         atomic_set(&sig->live, 1);
1475         atomic_set(&sig->sigcnt, 1);
1476
1477         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1478         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1479         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1480
1481         init_waitqueue_head(&sig->wait_chldexit);
1482         sig->curr_target = tsk;
1483         init_sigpending(&sig->shared_pending);
1484         seqlock_init(&sig->stats_lock);
1485         prev_cputime_init(&sig->prev_cputime);
1486
1487 #ifdef CONFIG_POSIX_TIMERS
1488         INIT_LIST_HEAD(&sig->posix_timers);
1489         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1490         sig->real_timer.function = it_real_fn;
1491 #endif
1492
1493         task_lock(current->group_leader);
1494         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1495         task_unlock(current->group_leader);
1496
1497         posix_cpu_timers_init_group(sig);
1498
1499         tty_audit_fork(sig);
1500         sched_autogroup_fork(sig);
1501
1502         sig->oom_score_adj = current->signal->oom_score_adj;
1503         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1504
1505         mutex_init(&sig->cred_guard_mutex);
1506
1507         return 0;
1508 }
1509
1510 static void copy_seccomp(struct task_struct *p)
1511 {
1512 #ifdef CONFIG_SECCOMP
1513         /*
1514          * Must be called with sighand->lock held, which is common to
1515          * all threads in the group. Holding cred_guard_mutex is not
1516          * needed because this new task is not yet running and cannot
1517          * be racing exec.
1518          */
1519         assert_spin_locked(&current->sighand->siglock);
1520
1521         /* Ref-count the new filter user, and assign it. */
1522         get_seccomp_filter(current);
1523         p->seccomp = current->seccomp;
1524
1525         /*
1526          * Explicitly enable no_new_privs here in case it got set
1527          * between the task_struct being duplicated and holding the
1528          * sighand lock. The seccomp state and nnp must be in sync.
1529          */
1530         if (task_no_new_privs(current))
1531                 task_set_no_new_privs(p);
1532
1533         /*
1534          * If the parent gained a seccomp mode after copying thread
1535          * flags and between before we held the sighand lock, we have
1536          * to manually enable the seccomp thread flag here.
1537          */
1538         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1539                 set_tsk_thread_flag(p, TIF_SECCOMP);
1540 #endif
1541 }
1542
1543 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1544 {
1545         current->clear_child_tid = tidptr;
1546
1547         return task_pid_vnr(current);
1548 }
1549
1550 static void rt_mutex_init_task(struct task_struct *p)
1551 {
1552         raw_spin_lock_init(&p->pi_lock);
1553 #ifdef CONFIG_RT_MUTEXES
1554         p->pi_waiters = RB_ROOT_CACHED;
1555         p->pi_top_task = NULL;
1556         p->pi_blocked_on = NULL;
1557 #endif
1558 }
1559
1560 #ifdef CONFIG_POSIX_TIMERS
1561 /*
1562  * Initialize POSIX timer handling for a single task.
1563  */
1564 static void posix_cpu_timers_init(struct task_struct *tsk)
1565 {
1566         tsk->cputime_expires.prof_exp = 0;
1567         tsk->cputime_expires.virt_exp = 0;
1568         tsk->cputime_expires.sched_exp = 0;
1569         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1570         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1571         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1572 }
1573 #else
1574 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1575 #endif
1576
1577 static inline void
1578 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1579 {
1580          task->pids[type].pid = pid;
1581 }
1582
1583 static inline void rcu_copy_process(struct task_struct *p)
1584 {
1585 #ifdef CONFIG_PREEMPT_RCU
1586         p->rcu_read_lock_nesting = 0;
1587         p->rcu_read_unlock_special.s = 0;
1588         p->rcu_blocked_node = NULL;
1589         INIT_LIST_HEAD(&p->rcu_node_entry);
1590 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1591 #ifdef CONFIG_TASKS_RCU
1592         p->rcu_tasks_holdout = false;
1593         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1594         p->rcu_tasks_idle_cpu = -1;
1595 #endif /* #ifdef CONFIG_TASKS_RCU */
1596 }
1597
1598 /*
1599  * This creates a new process as a copy of the old one,
1600  * but does not actually start it yet.
1601  *
1602  * It copies the registers, and all the appropriate
1603  * parts of the process environment (as per the clone
1604  * flags). The actual kick-off is left to the caller.
1605  */
1606 static __latent_entropy struct task_struct *copy_process(
1607                                         unsigned long clone_flags,
1608                                         unsigned long stack_start,
1609                                         unsigned long stack_size,
1610                                         int __user *child_tidptr,
1611                                         struct pid *pid,
1612                                         int trace,
1613                                         unsigned long tls,
1614                                         int node)
1615 {
1616         int retval;
1617         struct task_struct *p;
1618
1619         /*
1620          * Don't allow sharing the root directory with processes in a different
1621          * namespace
1622          */
1623         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1624                 return ERR_PTR(-EINVAL);
1625
1626         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1627                 return ERR_PTR(-EINVAL);
1628
1629         /*
1630          * Thread groups must share signals as well, and detached threads
1631          * can only be started up within the thread group.
1632          */
1633         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1634                 return ERR_PTR(-EINVAL);
1635
1636         /*
1637          * Shared signal handlers imply shared VM. By way of the above,
1638          * thread groups also imply shared VM. Blocking this case allows
1639          * for various simplifications in other code.
1640          */
1641         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1642                 return ERR_PTR(-EINVAL);
1643
1644         /*
1645          * Siblings of global init remain as zombies on exit since they are
1646          * not reaped by their parent (swapper). To solve this and to avoid
1647          * multi-rooted process trees, prevent global and container-inits
1648          * from creating siblings.
1649          */
1650         if ((clone_flags & CLONE_PARENT) &&
1651                                 current->signal->flags & SIGNAL_UNKILLABLE)
1652                 return ERR_PTR(-EINVAL);
1653
1654         /*
1655          * If the new process will be in a different pid or user namespace
1656          * do not allow it to share a thread group with the forking task.
1657          */
1658         if (clone_flags & CLONE_THREAD) {
1659                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1660                     (task_active_pid_ns(current) !=
1661                                 current->nsproxy->pid_ns_for_children))
1662                         return ERR_PTR(-EINVAL);
1663         }
1664
1665         retval = -ENOMEM;
1666         p = dup_task_struct(current, node);
1667         if (!p)
1668                 goto fork_out;
1669
1670         /*
1671          * This _must_ happen before we call free_task(), i.e. before we jump
1672          * to any of the bad_fork_* labels. This is to avoid freeing
1673          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1674          * kernel threads (PF_KTHREAD).
1675          */
1676         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1677         /*
1678          * Clear TID on mm_release()?
1679          */
1680         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1681
1682         ftrace_graph_init_task(p);
1683
1684         rt_mutex_init_task(p);
1685
1686 #ifdef CONFIG_PROVE_LOCKING
1687         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1688         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1689 #endif
1690         retval = -EAGAIN;
1691         if (atomic_read(&p->real_cred->user->processes) >=
1692                         task_rlimit(p, RLIMIT_NPROC)) {
1693                 if (p->real_cred->user != INIT_USER &&
1694                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1695                         goto bad_fork_free;
1696         }
1697         current->flags &= ~PF_NPROC_EXCEEDED;
1698
1699         retval = copy_creds(p, clone_flags);
1700         if (retval < 0)
1701                 goto bad_fork_free;
1702
1703         /*
1704          * If multiple threads are within copy_process(), then this check
1705          * triggers too late. This doesn't hurt, the check is only there
1706          * to stop root fork bombs.
1707          */
1708         retval = -EAGAIN;
1709         if (nr_threads >= max_threads)
1710                 goto bad_fork_cleanup_count;
1711
1712         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1713         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1714         p->flags |= PF_FORKNOEXEC;
1715         INIT_LIST_HEAD(&p->children);
1716         INIT_LIST_HEAD(&p->sibling);
1717         rcu_copy_process(p);
1718         p->vfork_done = NULL;
1719         spin_lock_init(&p->alloc_lock);
1720
1721         init_sigpending(&p->pending);
1722
1723         p->utime = p->stime = p->gtime = 0;
1724 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1725         p->utimescaled = p->stimescaled = 0;
1726 #endif
1727         prev_cputime_init(&p->prev_cputime);
1728
1729 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1730         seqcount_init(&p->vtime.seqcount);
1731         p->vtime.starttime = 0;
1732         p->vtime.state = VTIME_INACTIVE;
1733 #endif
1734
1735 #if defined(SPLIT_RSS_COUNTING)
1736         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1737 #endif
1738
1739         p->default_timer_slack_ns = current->timer_slack_ns;
1740
1741         task_io_accounting_init(&p->ioac);
1742         acct_clear_integrals(p);
1743
1744         posix_cpu_timers_init(p);
1745
1746         p->start_time = ktime_get_ns();
1747         p->real_start_time = ktime_get_boot_ns();
1748         p->io_context = NULL;
1749         audit_set_context(p, NULL);
1750         cgroup_fork(p);
1751 #ifdef CONFIG_NUMA
1752         p->mempolicy = mpol_dup(p->mempolicy);
1753         if (IS_ERR(p->mempolicy)) {
1754                 retval = PTR_ERR(p->mempolicy);
1755                 p->mempolicy = NULL;
1756                 goto bad_fork_cleanup_threadgroup_lock;
1757         }
1758 #endif
1759 #ifdef CONFIG_CPUSETS
1760         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1761         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1762         seqcount_init(&p->mems_allowed_seq);
1763 #endif
1764 #ifdef CONFIG_TRACE_IRQFLAGS
1765         p->irq_events = 0;
1766         p->hardirqs_enabled = 0;
1767         p->hardirq_enable_ip = 0;
1768         p->hardirq_enable_event = 0;
1769         p->hardirq_disable_ip = _THIS_IP_;
1770         p->hardirq_disable_event = 0;
1771         p->softirqs_enabled = 1;
1772         p->softirq_enable_ip = _THIS_IP_;
1773         p->softirq_enable_event = 0;
1774         p->softirq_disable_ip = 0;
1775         p->softirq_disable_event = 0;
1776         p->hardirq_context = 0;
1777         p->softirq_context = 0;
1778 #endif
1779
1780         p->pagefault_disabled = 0;
1781
1782 #ifdef CONFIG_LOCKDEP
1783         p->lockdep_depth = 0; /* no locks held yet */
1784         p->curr_chain_key = 0;
1785         p->lockdep_recursion = 0;
1786         lockdep_init_task(p);
1787 #endif
1788
1789 #ifdef CONFIG_DEBUG_MUTEXES
1790         p->blocked_on = NULL; /* not blocked yet */
1791 #endif
1792 #ifdef CONFIG_BCACHE
1793         p->sequential_io        = 0;
1794         p->sequential_io_avg    = 0;
1795 #endif
1796
1797         /* Perform scheduler related setup. Assign this task to a CPU. */
1798         retval = sched_fork(clone_flags, p);
1799         if (retval)
1800                 goto bad_fork_cleanup_policy;
1801
1802         retval = perf_event_init_task(p);
1803         if (retval)
1804                 goto bad_fork_cleanup_policy;
1805         retval = audit_alloc(p);
1806         if (retval)
1807                 goto bad_fork_cleanup_perf;
1808         /* copy all the process information */
1809         shm_init_task(p);
1810         retval = security_task_alloc(p, clone_flags);
1811         if (retval)
1812                 goto bad_fork_cleanup_audit;
1813         retval = copy_semundo(clone_flags, p);
1814         if (retval)
1815                 goto bad_fork_cleanup_security;
1816         retval = copy_files(clone_flags, p);
1817         if (retval)
1818                 goto bad_fork_cleanup_semundo;
1819         retval = copy_fs(clone_flags, p);
1820         if (retval)
1821                 goto bad_fork_cleanup_files;
1822         retval = copy_sighand(clone_flags, p);
1823         if (retval)
1824                 goto bad_fork_cleanup_fs;
1825         retval = copy_signal(clone_flags, p);
1826         if (retval)
1827                 goto bad_fork_cleanup_sighand;
1828         retval = copy_mm(clone_flags, p);
1829         if (retval)
1830                 goto bad_fork_cleanup_signal;
1831         retval = copy_namespaces(clone_flags, p);
1832         if (retval)
1833                 goto bad_fork_cleanup_mm;
1834         retval = copy_io(clone_flags, p);
1835         if (retval)
1836                 goto bad_fork_cleanup_namespaces;
1837         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1838         if (retval)
1839                 goto bad_fork_cleanup_io;
1840
1841         if (pid != &init_struct_pid) {
1842                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1843                 if (IS_ERR(pid)) {
1844                         retval = PTR_ERR(pid);
1845                         goto bad_fork_cleanup_thread;
1846                 }
1847         }
1848
1849 #ifdef CONFIG_BLOCK
1850         p->plug = NULL;
1851 #endif
1852 #ifdef CONFIG_FUTEX
1853         p->robust_list = NULL;
1854 #ifdef CONFIG_COMPAT
1855         p->compat_robust_list = NULL;
1856 #endif
1857         INIT_LIST_HEAD(&p->pi_state_list);
1858         p->pi_state_cache = NULL;
1859 #endif
1860         /*
1861          * sigaltstack should be cleared when sharing the same VM
1862          */
1863         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1864                 sas_ss_reset(p);
1865
1866         /*
1867          * Syscall tracing and stepping should be turned off in the
1868          * child regardless of CLONE_PTRACE.
1869          */
1870         user_disable_single_step(p);
1871         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1872 #ifdef TIF_SYSCALL_EMU
1873         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1874 #endif
1875         clear_all_latency_tracing(p);
1876
1877         /* ok, now we should be set up.. */
1878         p->pid = pid_nr(pid);
1879         if (clone_flags & CLONE_THREAD) {
1880                 p->exit_signal = -1;
1881                 p->group_leader = current->group_leader;
1882                 p->tgid = current->tgid;
1883         } else {
1884                 if (clone_flags & CLONE_PARENT)
1885                         p->exit_signal = current->group_leader->exit_signal;
1886                 else
1887                         p->exit_signal = (clone_flags & CSIGNAL);
1888                 p->group_leader = p;
1889                 p->tgid = p->pid;
1890         }
1891
1892         p->nr_dirtied = 0;
1893         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1894         p->dirty_paused_when = 0;
1895
1896         p->pdeath_signal = 0;
1897         INIT_LIST_HEAD(&p->thread_group);
1898         p->task_works = NULL;
1899
1900         cgroup_threadgroup_change_begin(current);
1901         /*
1902          * Ensure that the cgroup subsystem policies allow the new process to be
1903          * forked. It should be noted the the new process's css_set can be changed
1904          * between here and cgroup_post_fork() if an organisation operation is in
1905          * progress.
1906          */
1907         retval = cgroup_can_fork(p);
1908         if (retval)
1909                 goto bad_fork_free_pid;
1910
1911         /*
1912          * Make it visible to the rest of the system, but dont wake it up yet.
1913          * Need tasklist lock for parent etc handling!
1914          */
1915         write_lock_irq(&tasklist_lock);
1916
1917         /* CLONE_PARENT re-uses the old parent */
1918         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1919                 p->real_parent = current->real_parent;
1920                 p->parent_exec_id = current->parent_exec_id;
1921         } else {
1922                 p->real_parent = current;
1923                 p->parent_exec_id = current->self_exec_id;
1924         }
1925
1926         klp_copy_process(p);
1927
1928         spin_lock(&current->sighand->siglock);
1929
1930         /*
1931          * Copy seccomp details explicitly here, in case they were changed
1932          * before holding sighand lock.
1933          */
1934         copy_seccomp(p);
1935
1936         rseq_fork(p, clone_flags);
1937
1938         /*
1939          * Process group and session signals need to be delivered to just the
1940          * parent before the fork or both the parent and the child after the
1941          * fork. Restart if a signal comes in before we add the new process to
1942          * it's process group.
1943          * A fatal signal pending means that current will exit, so the new
1944          * thread can't slip out of an OOM kill (or normal SIGKILL).
1945         */
1946         recalc_sigpending();
1947         if (signal_pending(current)) {
1948                 retval = -ERESTARTNOINTR;
1949                 goto bad_fork_cancel_cgroup;
1950         }
1951         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
1952                 retval = -ENOMEM;
1953                 goto bad_fork_cancel_cgroup;
1954         }
1955
1956         if (likely(p->pid)) {
1957                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1958
1959                 init_task_pid(p, PIDTYPE_PID, pid);
1960                 if (thread_group_leader(p)) {
1961                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1962                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1963
1964                         if (is_child_reaper(pid)) {
1965                                 ns_of_pid(pid)->child_reaper = p;
1966                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1967                         }
1968
1969                         p->signal->leader_pid = pid;
1970                         p->signal->tty = tty_kref_get(current->signal->tty);
1971                         /*
1972                          * Inherit has_child_subreaper flag under the same
1973                          * tasklist_lock with adding child to the process tree
1974                          * for propagate_has_child_subreaper optimization.
1975                          */
1976                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1977                                                          p->real_parent->signal->is_child_subreaper;
1978                         list_add_tail(&p->sibling, &p->real_parent->children);
1979                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1980                         attach_pid(p, PIDTYPE_PGID);
1981                         attach_pid(p, PIDTYPE_SID);
1982                         __this_cpu_inc(process_counts);
1983                 } else {
1984                         current->signal->nr_threads++;
1985                         atomic_inc(&current->signal->live);
1986                         atomic_inc(&current->signal->sigcnt);
1987                         list_add_tail_rcu(&p->thread_group,
1988                                           &p->group_leader->thread_group);
1989                         list_add_tail_rcu(&p->thread_node,
1990                                           &p->signal->thread_head);
1991                 }
1992                 attach_pid(p, PIDTYPE_PID);
1993                 nr_threads++;
1994         }
1995
1996         total_forks++;
1997         spin_unlock(&current->sighand->siglock);
1998         syscall_tracepoint_update(p);
1999         write_unlock_irq(&tasklist_lock);
2000
2001         proc_fork_connector(p);
2002         cgroup_post_fork(p);
2003         cgroup_threadgroup_change_end(current);
2004         perf_event_fork(p);
2005
2006         trace_task_newtask(p, clone_flags);
2007         uprobe_copy_process(p, clone_flags);
2008
2009         return p;
2010
2011 bad_fork_cancel_cgroup:
2012         spin_unlock(&current->sighand->siglock);
2013         write_unlock_irq(&tasklist_lock);
2014         cgroup_cancel_fork(p);
2015 bad_fork_free_pid:
2016         cgroup_threadgroup_change_end(current);
2017         if (pid != &init_struct_pid)
2018                 free_pid(pid);
2019 bad_fork_cleanup_thread:
2020         exit_thread(p);
2021 bad_fork_cleanup_io:
2022         if (p->io_context)
2023                 exit_io_context(p);
2024 bad_fork_cleanup_namespaces:
2025         exit_task_namespaces(p);
2026 bad_fork_cleanup_mm:
2027         if (p->mm)
2028                 mmput(p->mm);
2029 bad_fork_cleanup_signal:
2030         if (!(clone_flags & CLONE_THREAD))
2031                 free_signal_struct(p->signal);
2032 bad_fork_cleanup_sighand:
2033         __cleanup_sighand(p->sighand);
2034 bad_fork_cleanup_fs:
2035         exit_fs(p); /* blocking */
2036 bad_fork_cleanup_files:
2037         exit_files(p); /* blocking */
2038 bad_fork_cleanup_semundo:
2039         exit_sem(p);
2040 bad_fork_cleanup_security:
2041         security_task_free(p);
2042 bad_fork_cleanup_audit:
2043         audit_free(p);
2044 bad_fork_cleanup_perf:
2045         perf_event_free_task(p);
2046 bad_fork_cleanup_policy:
2047         lockdep_free_task(p);
2048 #ifdef CONFIG_NUMA
2049         mpol_put(p->mempolicy);
2050 bad_fork_cleanup_threadgroup_lock:
2051 #endif
2052         delayacct_tsk_free(p);
2053 bad_fork_cleanup_count:
2054         atomic_dec(&p->cred->user->processes);
2055         exit_creds(p);
2056 bad_fork_free:
2057         p->state = TASK_DEAD;
2058         put_task_stack(p);
2059         free_task(p);
2060 fork_out:
2061         return ERR_PTR(retval);
2062 }
2063
2064 static inline void init_idle_pids(struct pid_link *links)
2065 {
2066         enum pid_type type;
2067
2068         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2069                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
2070                 links[type].pid = &init_struct_pid;
2071         }
2072 }
2073
2074 struct task_struct *fork_idle(int cpu)
2075 {
2076         struct task_struct *task;
2077         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
2078                             cpu_to_node(cpu));
2079         if (!IS_ERR(task)) {
2080                 init_idle_pids(task->pids);
2081                 init_idle(task, cpu);
2082         }
2083
2084         return task;
2085 }
2086
2087 /*
2088  *  Ok, this is the main fork-routine.
2089  *
2090  * It copies the process, and if successful kick-starts
2091  * it and waits for it to finish using the VM if required.
2092  */
2093 long _do_fork(unsigned long clone_flags,
2094               unsigned long stack_start,
2095               unsigned long stack_size,
2096               int __user *parent_tidptr,
2097               int __user *child_tidptr,
2098               unsigned long tls)
2099 {
2100         struct completion vfork;
2101         struct pid *pid;
2102         struct task_struct *p;
2103         int trace = 0;
2104         long nr;
2105
2106         /*
2107          * Determine whether and which event to report to ptracer.  When
2108          * called from kernel_thread or CLONE_UNTRACED is explicitly
2109          * requested, no event is reported; otherwise, report if the event
2110          * for the type of forking is enabled.
2111          */
2112         if (!(clone_flags & CLONE_UNTRACED)) {
2113                 if (clone_flags & CLONE_VFORK)
2114                         trace = PTRACE_EVENT_VFORK;
2115                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2116                         trace = PTRACE_EVENT_CLONE;
2117                 else
2118                         trace = PTRACE_EVENT_FORK;
2119
2120                 if (likely(!ptrace_event_enabled(current, trace)))
2121                         trace = 0;
2122         }
2123
2124         p = copy_process(clone_flags, stack_start, stack_size,
2125                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2126         add_latent_entropy();
2127
2128         if (IS_ERR(p))
2129                 return PTR_ERR(p);
2130
2131         /*
2132          * Do this prior waking up the new thread - the thread pointer
2133          * might get invalid after that point, if the thread exits quickly.
2134          */
2135         trace_sched_process_fork(current, p);
2136
2137         pid = get_task_pid(p, PIDTYPE_PID);
2138         nr = pid_vnr(pid);
2139
2140         if (clone_flags & CLONE_PARENT_SETTID)
2141                 put_user(nr, parent_tidptr);
2142
2143         if (clone_flags & CLONE_VFORK) {
2144                 p->vfork_done = &vfork;
2145                 init_completion(&vfork);
2146                 get_task_struct(p);
2147         }
2148
2149         wake_up_new_task(p);
2150
2151         /* forking complete and child started to run, tell ptracer */
2152         if (unlikely(trace))
2153                 ptrace_event_pid(trace, pid);
2154
2155         if (clone_flags & CLONE_VFORK) {
2156                 if (!wait_for_vfork_done(p, &vfork))
2157                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2158         }
2159
2160         put_pid(pid);
2161         return nr;
2162 }
2163
2164 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2165 /* For compatibility with architectures that call do_fork directly rather than
2166  * using the syscall entry points below. */
2167 long do_fork(unsigned long clone_flags,
2168               unsigned long stack_start,
2169               unsigned long stack_size,
2170               int __user *parent_tidptr,
2171               int __user *child_tidptr)
2172 {
2173         return _do_fork(clone_flags, stack_start, stack_size,
2174                         parent_tidptr, child_tidptr, 0);
2175 }
2176 #endif
2177
2178 /*
2179  * Create a kernel thread.
2180  */
2181 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2182 {
2183         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2184                 (unsigned long)arg, NULL, NULL, 0);
2185 }
2186
2187 #ifdef __ARCH_WANT_SYS_FORK
2188 SYSCALL_DEFINE0(fork)
2189 {
2190 #ifdef CONFIG_MMU
2191         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2192 #else
2193         /* can not support in nommu mode */
2194         return -EINVAL;
2195 #endif
2196 }
2197 #endif
2198
2199 #ifdef __ARCH_WANT_SYS_VFORK
2200 SYSCALL_DEFINE0(vfork)
2201 {
2202         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2203                         0, NULL, NULL, 0);
2204 }
2205 #endif
2206
2207 #ifdef __ARCH_WANT_SYS_CLONE
2208 #ifdef CONFIG_CLONE_BACKWARDS
2209 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2210                  int __user *, parent_tidptr,
2211                  unsigned long, tls,
2212                  int __user *, child_tidptr)
2213 #elif defined(CONFIG_CLONE_BACKWARDS2)
2214 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2215                  int __user *, parent_tidptr,
2216                  int __user *, child_tidptr,
2217                  unsigned long, tls)
2218 #elif defined(CONFIG_CLONE_BACKWARDS3)
2219 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2220                 int, stack_size,
2221                 int __user *, parent_tidptr,
2222                 int __user *, child_tidptr,
2223                 unsigned long, tls)
2224 #else
2225 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2226                  int __user *, parent_tidptr,
2227                  int __user *, child_tidptr,
2228                  unsigned long, tls)
2229 #endif
2230 {
2231         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2232 }
2233 #endif
2234
2235 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2236 {
2237         struct task_struct *leader, *parent, *child;
2238         int res;
2239
2240         read_lock(&tasklist_lock);
2241         leader = top = top->group_leader;
2242 down:
2243         for_each_thread(leader, parent) {
2244                 list_for_each_entry(child, &parent->children, sibling) {
2245                         res = visitor(child, data);
2246                         if (res) {
2247                                 if (res < 0)
2248                                         goto out;
2249                                 leader = child;
2250                                 goto down;
2251                         }
2252 up:
2253                         ;
2254                 }
2255         }
2256
2257         if (leader != top) {
2258                 child = leader;
2259                 parent = child->real_parent;
2260                 leader = parent->group_leader;
2261                 goto up;
2262         }
2263 out:
2264         read_unlock(&tasklist_lock);
2265 }
2266
2267 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2268 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2269 #endif
2270
2271 static void sighand_ctor(void *data)
2272 {
2273         struct sighand_struct *sighand = data;
2274
2275         spin_lock_init(&sighand->siglock);
2276         init_waitqueue_head(&sighand->signalfd_wqh);
2277 }
2278
2279 void __init proc_caches_init(void)
2280 {
2281         sighand_cachep = kmem_cache_create("sighand_cache",
2282                         sizeof(struct sighand_struct), 0,
2283                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2284                         SLAB_ACCOUNT, sighand_ctor);
2285         signal_cachep = kmem_cache_create("signal_cache",
2286                         sizeof(struct signal_struct), 0,
2287                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2288                         NULL);
2289         files_cachep = kmem_cache_create("files_cache",
2290                         sizeof(struct files_struct), 0,
2291                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2292                         NULL);
2293         fs_cachep = kmem_cache_create("fs_cache",
2294                         sizeof(struct fs_struct), 0,
2295                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2296                         NULL);
2297         /*
2298          * FIXME! The "sizeof(struct mm_struct)" currently includes the
2299          * whole struct cpumask for the OFFSTACK case. We could change
2300          * this to *only* allocate as much of it as required by the
2301          * maximum number of CPU's we can ever have.  The cpumask_allocation
2302          * is at the end of the structure, exactly for that reason.
2303          */
2304         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2305                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2306                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2307                         offsetof(struct mm_struct, saved_auxv),
2308                         sizeof_field(struct mm_struct, saved_auxv),
2309                         NULL);
2310         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2311         mmap_init();
2312         nsproxy_cache_init();
2313 }
2314
2315 /*
2316  * Check constraints on flags passed to the unshare system call.
2317  */
2318 static int check_unshare_flags(unsigned long unshare_flags)
2319 {
2320         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2321                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2322                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2323                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2324                 return -EINVAL;
2325         /*
2326          * Not implemented, but pretend it works if there is nothing
2327          * to unshare.  Note that unsharing the address space or the
2328          * signal handlers also need to unshare the signal queues (aka
2329          * CLONE_THREAD).
2330          */
2331         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2332                 if (!thread_group_empty(current))
2333                         return -EINVAL;
2334         }
2335         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2336                 if (atomic_read(&current->sighand->count) > 1)
2337                         return -EINVAL;
2338         }
2339         if (unshare_flags & CLONE_VM) {
2340                 if (!current_is_single_threaded())
2341                         return -EINVAL;
2342         }
2343
2344         return 0;
2345 }
2346
2347 /*
2348  * Unshare the filesystem structure if it is being shared
2349  */
2350 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2351 {
2352         struct fs_struct *fs = current->fs;
2353
2354         if (!(unshare_flags & CLONE_FS) || !fs)
2355                 return 0;
2356
2357         /* don't need lock here; in the worst case we'll do useless copy */
2358         if (fs->users == 1)
2359                 return 0;
2360
2361         *new_fsp = copy_fs_struct(fs);
2362         if (!*new_fsp)
2363                 return -ENOMEM;
2364
2365         return 0;
2366 }
2367
2368 /*
2369  * Unshare file descriptor table if it is being shared
2370  */
2371 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2372 {
2373         struct files_struct *fd = current->files;
2374         int error = 0;
2375
2376         if ((unshare_flags & CLONE_FILES) &&
2377             (fd && atomic_read(&fd->count) > 1)) {
2378                 *new_fdp = dup_fd(fd, &error);
2379                 if (!*new_fdp)
2380                         return error;
2381         }
2382
2383         return 0;
2384 }
2385
2386 /*
2387  * unshare allows a process to 'unshare' part of the process
2388  * context which was originally shared using clone.  copy_*
2389  * functions used by do_fork() cannot be used here directly
2390  * because they modify an inactive task_struct that is being
2391  * constructed. Here we are modifying the current, active,
2392  * task_struct.
2393  */
2394 int ksys_unshare(unsigned long unshare_flags)
2395 {
2396         struct fs_struct *fs, *new_fs = NULL;
2397         struct files_struct *fd, *new_fd = NULL;
2398         struct cred *new_cred = NULL;
2399         struct nsproxy *new_nsproxy = NULL;
2400         int do_sysvsem = 0;
2401         int err;
2402
2403         /*
2404          * If unsharing a user namespace must also unshare the thread group
2405          * and unshare the filesystem root and working directories.
2406          */
2407         if (unshare_flags & CLONE_NEWUSER)
2408                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2409         /*
2410          * If unsharing vm, must also unshare signal handlers.
2411          */
2412         if (unshare_flags & CLONE_VM)
2413                 unshare_flags |= CLONE_SIGHAND;
2414         /*
2415          * If unsharing a signal handlers, must also unshare the signal queues.
2416          */
2417         if (unshare_flags & CLONE_SIGHAND)
2418                 unshare_flags |= CLONE_THREAD;
2419         /*
2420          * If unsharing namespace, must also unshare filesystem information.
2421          */
2422         if (unshare_flags & CLONE_NEWNS)
2423                 unshare_flags |= CLONE_FS;
2424
2425         err = check_unshare_flags(unshare_flags);
2426         if (err)
2427                 goto bad_unshare_out;
2428         /*
2429          * CLONE_NEWIPC must also detach from the undolist: after switching
2430          * to a new ipc namespace, the semaphore arrays from the old
2431          * namespace are unreachable.
2432          */
2433         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2434                 do_sysvsem = 1;
2435         err = unshare_fs(unshare_flags, &new_fs);
2436         if (err)
2437                 goto bad_unshare_out;
2438         err = unshare_fd(unshare_flags, &new_fd);
2439         if (err)
2440                 goto bad_unshare_cleanup_fs;
2441         err = unshare_userns(unshare_flags, &new_cred);
2442         if (err)
2443                 goto bad_unshare_cleanup_fd;
2444         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2445                                          new_cred, new_fs);
2446         if (err)
2447                 goto bad_unshare_cleanup_cred;
2448
2449         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2450                 if (do_sysvsem) {
2451                         /*
2452                          * CLONE_SYSVSEM is equivalent to sys_exit().
2453                          */
2454                         exit_sem(current);
2455                 }
2456                 if (unshare_flags & CLONE_NEWIPC) {
2457                         /* Orphan segments in old ns (see sem above). */
2458                         exit_shm(current);
2459                         shm_init_task(current);
2460                 }
2461
2462                 if (new_nsproxy)
2463                         switch_task_namespaces(current, new_nsproxy);
2464
2465                 task_lock(current);
2466
2467                 if (new_fs) {
2468                         fs = current->fs;
2469                         spin_lock(&fs->lock);
2470                         current->fs = new_fs;
2471                         if (--fs->users)
2472                                 new_fs = NULL;
2473                         else
2474                                 new_fs = fs;
2475                         spin_unlock(&fs->lock);
2476                 }
2477
2478                 if (new_fd) {
2479                         fd = current->files;
2480                         current->files = new_fd;
2481                         new_fd = fd;
2482                 }
2483
2484                 task_unlock(current);
2485
2486                 if (new_cred) {
2487                         /* Install the new user namespace */
2488                         commit_creds(new_cred);
2489                         new_cred = NULL;
2490                 }
2491         }
2492
2493         perf_event_namespaces(current);
2494
2495 bad_unshare_cleanup_cred:
2496         if (new_cred)
2497                 put_cred(new_cred);
2498 bad_unshare_cleanup_fd:
2499         if (new_fd)
2500                 put_files_struct(new_fd);
2501
2502 bad_unshare_cleanup_fs:
2503         if (new_fs)
2504                 free_fs_struct(new_fs);
2505
2506 bad_unshare_out:
2507         return err;
2508 }
2509
2510 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2511 {
2512         return ksys_unshare(unshare_flags);
2513 }
2514
2515 /*
2516  *      Helper to unshare the files of the current task.
2517  *      We don't want to expose copy_files internals to
2518  *      the exec layer of the kernel.
2519  */
2520
2521 int unshare_files(struct files_struct **displaced)
2522 {
2523         struct task_struct *task = current;
2524         struct files_struct *copy = NULL;
2525         int error;
2526
2527         error = unshare_fd(CLONE_FILES, &copy);
2528         if (error || !copy) {
2529                 *displaced = NULL;
2530                 return error;
2531         }
2532         *displaced = task->files;
2533         task_lock(task);
2534         task->files = copy;
2535         task_unlock(task);
2536         return 0;
2537 }
2538
2539 int sysctl_max_threads(struct ctl_table *table, int write,
2540                        void __user *buffer, size_t *lenp, loff_t *ppos)
2541 {
2542         struct ctl_table t;
2543         int ret;
2544         int threads = max_threads;
2545         int min = MIN_THREADS;
2546         int max = MAX_THREADS;
2547
2548         t = *table;
2549         t.data = &threads;
2550         t.extra1 = &min;
2551         t.extra2 = &max;
2552
2553         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2554         if (ret || !write)
2555                 return ret;
2556
2557         set_max_threads(threads);
2558
2559         return 0;
2560 }