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