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_init(vma, mm);
 317         return vma;
 318 }
 319
 320 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 321 {
 322         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 323
 324         if (new) {
 325                 *new = *orig;
 326                 INIT_LIST_HEAD(&new->anon_vma_chain);
 327         }
 328         return new;
 329 }
 330
 331 void vm_area_free(struct vm_area_struct *vma)
 332 {
 333         kmem_cache_free(vm_area_cachep, vma);
 334 }
 335
 336 static void account_kernel_stack(struct task_struct *tsk, int account)
 337 {
 338         void *stack = task_stack_page(tsk);
 339         struct vm_struct *vm = task_stack_vm_area(tsk);
 340
 341         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 342
 343         if (vm) {
 344                 int i;
 345
 346                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 347
 348                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 349                         mod_zone_page_state(page_zone(vm->pages[i]),
 350                                             NR_KERNEL_STACK_KB,
 351                                             PAGE_SIZE / 1024 * account);
 352                 }
 353
 354                 /* All stack pages belong to the same memcg. */
 355                 mod_memcg_page_state(vm->pages[0], MEMCG_KERNEL_STACK_KB,
 356                                      account * (THREAD_SIZE / 1024));
 357         } else {
 358                 /*
 359                  * All stack pages are in the same zone and belong to the
 360                  * same memcg.
 361                  */
 362                 struct page *first_page = virt_to_page(stack);
 363
 364                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 365                                     THREAD_SIZE / 1024 * account);
 366
 367                 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
 368                                      account * (THREAD_SIZE / 1024));
 369         }
 370 }
 371
 372 static void release_task_stack(struct task_struct *tsk)
 373 {
 374         if (WARN_ON(tsk->state != TASK_DEAD))
 375                 return;  /* Better to leak the stack than to free prematurely */
 376
 377         account_kernel_stack(tsk, -1);
 378         arch_release_thread_stack(tsk->stack);
 379         free_thread_stack(tsk);
 380         tsk->stack = NULL;
 381 #ifdef CONFIG_VMAP_STACK
 382         tsk->stack_vm_area = NULL;
 383 #endif
 384 }
 385
 386 #ifdef CONFIG_THREAD_INFO_IN_TASK
 387 void put_task_stack(struct task_struct *tsk)
 388 {
 389         if (atomic_dec_and_test(&tsk->stack_refcount))
 390                 release_task_stack(tsk);
 391 }
 392 #endif
 393
 394 void free_task(struct task_struct *tsk)
 395 {
 396 #ifndef CONFIG_THREAD_INFO_IN_TASK
 397         /*
 398          * The task is finally done with both the stack and thread_info,
 399          * so free both.
 400          */
 401         release_task_stack(tsk);
 402 #else
 403         /*
 404          * If the task had a separate stack allocation, it should be gone
 405          * by now.
 406          */
 407         WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
 408 #endif
 409         rt_mutex_debug_task_free(tsk);
 410         ftrace_graph_exit_task(tsk);
 411         put_seccomp_filter(tsk);
 412         arch_release_task_struct(tsk);
 413         if (tsk->flags & PF_KTHREAD)
 414                 free_kthread_struct(tsk);
 415         free_task_struct(tsk);
 416 }
 417 EXPORT_SYMBOL(free_task);
 418
 419 #ifdef CONFIG_MMU
 420 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 421                                         struct mm_struct *oldmm)
 422 {
 423         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 424         struct rb_node **rb_link, *rb_parent;
 425         int retval;
 426         unsigned long charge;
 427         LIST_HEAD(uf);
 428
 429         uprobe_start_dup_mmap();
 430         if (down_write_killable(&oldmm->mmap_sem)) {
 431                 retval = -EINTR;
 432                 goto fail_uprobe_end;
 433         }
 434         flush_cache_dup_mm(oldmm);
 435         uprobe_dup_mmap(oldmm, mm);
 436         /*
 437          * Not linked in yet - no deadlock potential:
 438          */
 439         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 440
 441         /* No ordering required: file already has been exposed. */
 442         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 443
 444         mm->total_vm = oldmm->total_vm;
 445         mm->data_vm = oldmm->data_vm;
 446         mm->exec_vm = oldmm->exec_vm;
 447         mm->stack_vm = oldmm->stack_vm;
 448
 449         rb_link = &mm->mm_rb.rb_node;
 450         rb_parent = NULL;
 451         pprev = &mm->mmap;
 452         retval = ksm_fork(mm, oldmm);
 453         if (retval)
 454                 goto out;
 455         retval = khugepaged_fork(mm, oldmm);
 456         if (retval)
 457                 goto out;
 458
 459         prev = NULL;
 460         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 461                 struct file *file;
 462
 463                 if (mpnt->vm_flags & VM_DONTCOPY) {
 464                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 465                         continue;
 466                 }
 467                 charge = 0;
 468                 /*
 469                  * Don't duplicate many vmas if we've been oom-killed (for
 470                  * example)
 471                  */
 472                 if (fatal_signal_pending(current)) {
 473                         retval = -EINTR;
 474                         goto out;
 475                 }
 476                 if (mpnt->vm_flags & VM_ACCOUNT) {
 477                         unsigned long len = vma_pages(mpnt);
 478
 479                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 480                                 goto fail_nomem;
 481                         charge = len;
 482                 }
 483                 tmp = vm_area_dup(mpnt);
 484                 if (!tmp)
 485                         goto fail_nomem;
 486                 retval = vma_dup_policy(mpnt, tmp);
 487                 if (retval)
 488                         goto fail_nomem_policy;
 489                 tmp->vm_mm = mm;
 490                 retval = dup_userfaultfd(tmp, &uf);
 491                 if (retval)
 492                         goto fail_nomem_anon_vma_fork;
 493                 if (tmp->vm_flags & VM_WIPEONFORK) {
 494                         /* VM_WIPEONFORK gets a clean slate in the child. */
 495                         tmp->anon_vma = NULL;
 496                         if (anon_vma_prepare(tmp))
 497                                 goto fail_nomem_anon_vma_fork;
 498                 } else if (anon_vma_fork(tmp, mpnt))
 499                         goto fail_nomem_anon_vma_fork;
 500                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 501                 tmp->vm_next = tmp->vm_prev = NULL;
 502                 file = tmp->vm_file;
 503                 if (file) {
 504                         struct inode *inode = file_inode(file);
 505                         struct address_space *mapping = file->f_mapping;
 506
 507                         get_file(file);
 508                         if (tmp->vm_flags & VM_DENYWRITE)
 509                                 atomic_dec(&inode->i_writecount);
 510                         i_mmap_lock_write(mapping);
 511                         if (tmp->vm_flags & VM_SHARED)
 512                                 atomic_inc(&mapping->i_mmap_writable);
 513                         flush_dcache_mmap_lock(mapping);
 514                         /* insert tmp into the share list, just after mpnt */
 515                         vma_interval_tree_insert_after(tmp, mpnt,
 516                                         &mapping->i_mmap);
 517                         flush_dcache_mmap_unlock(mapping);
 518                         i_mmap_unlock_write(mapping);
 519                 }
 520
 521                 /*
 522                  * Clear hugetlb-related page reserves for children. This only
 523                  * affects MAP_PRIVATE mappings. Faults generated by the child
 524                  * are not guaranteed to succeed, even if read-only
 525                  */
 526                 if (is_vm_hugetlb_page(tmp))
 527                         reset_vma_resv_huge_pages(tmp);
 528
 529                 /*
 530                  * Link in the new vma and copy the page table entries.
 531                  */
 532                 *pprev = tmp;
 533                 pprev = &tmp->vm_next;
 534                 tmp->vm_prev = prev;
 535                 prev = tmp;
 536
 537                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 538                 rb_link = &tmp->vm_rb.rb_right;
 539                 rb_parent = &tmp->vm_rb;
 540
 541                 mm->map_count++;
 542                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 543                         retval = copy_page_range(mm, oldmm, mpnt);
 544
 545                 if (tmp->vm_ops && tmp->vm_ops->open)
 546                         tmp->vm_ops->open(tmp);
 547
 548                 if (retval)
 549                         goto out;
 550         }
 551         /* a new mm has just been created */
 552         arch_dup_mmap(oldmm, mm);
 553         retval = 0;
 554 out:
 555         up_write(&mm->mmap_sem);
 556         flush_tlb_mm(oldmm);
 557         up_write(&oldmm->mmap_sem);
 558         dup_userfaultfd_complete(&uf);
 559 fail_uprobe_end:
 560         uprobe_end_dup_mmap();
 561         return retval;
 562 fail_nomem_anon_vma_fork:
 563         mpol_put(vma_policy(tmp));
 564 fail_nomem_policy:
 565         vm_area_free(tmp);
 566 fail_nomem:
 567         retval = -ENOMEM;
 568         vm_unacct_memory(charge);
 569         goto out;
 570 }
 571
 572 static inline int mm_alloc_pgd(struct mm_struct *mm)
 573 {
 574         mm->pgd = pgd_alloc(mm);
 575         if (unlikely(!mm->pgd))
 576                 return -ENOMEM;
 577         return 0;
 578 }
 579
 580 static inline void mm_free_pgd(struct mm_struct *mm)
 581 {
 582         pgd_free(mm, mm->pgd);
 583 }
 584 #else
 585 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 586 {
 587         down_write(&oldmm->mmap_sem);
 588         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 589         up_write(&oldmm->mmap_sem);
 590         return 0;
 591 }
 592 #define mm_alloc_pgd(mm)        (0)
 593 #define mm_free_pgd(mm)
 594 #endif /* CONFIG_MMU */
 595
 596 static void check_mm(struct mm_struct *mm)
 597 {
 598         int i;
 599
 600         for (i = 0; i < NR_MM_COUNTERS; i++) {
 601                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 602
 603                 if (unlikely(x))
 604                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 605                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 606         }
 607
 608         if (mm_pgtables_bytes(mm))
 609                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 610                                 mm_pgtables_bytes(mm));
 611
 612 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 613         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 614 #endif
 615 }
 616
 617 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 618 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 619
 620 /*
 621  * Called when the last reference to the mm
 622  * is dropped: either by a lazy thread or by
 623  * mmput. Free the page directory and the mm.
 624  */
 625 void __mmdrop(struct mm_struct *mm)
 626 {
 627         BUG_ON(mm == &init_mm);
 628         WARN_ON_ONCE(mm == current->mm);
 629         WARN_ON_ONCE(mm == current->active_mm);
 630         mm_free_pgd(mm);
 631         destroy_context(mm);
 632         hmm_mm_destroy(mm);
 633         mmu_notifier_mm_destroy(mm);
 634         check_mm(mm);
 635         put_user_ns(mm->user_ns);
 636         free_mm(mm);
 637 }
 638 EXPORT_SYMBOL_GPL(__mmdrop);
 639
 640 static void mmdrop_async_fn(struct work_struct *work)
 641 {
 642         struct mm_struct *mm;
 643
 644         mm = container_of(work, struct mm_struct, async_put_work);
 645         __mmdrop(mm);
 646 }
 647
 648 static void mmdrop_async(struct mm_struct *mm)
 649 {
 650         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 651                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 652                 schedule_work(&mm->async_put_work);
 653         }
 654 }
 655
 656 static inline void free_signal_struct(struct signal_struct *sig)
 657 {
 658         taskstats_tgid_free(sig);
 659         sched_autogroup_exit(sig);
 660         /*
 661          * __mmdrop is not safe to call from softirq context on x86 due to
 662          * pgd_dtor so postpone it to the async context
 663          */
 664         if (sig->oom_mm)
 665                 mmdrop_async(sig->oom_mm);
 666         kmem_cache_free(signal_cachep, sig);
 667 }
 668
 669 static inline void put_signal_struct(struct signal_struct *sig)
 670 {
 671         if (atomic_dec_and_test(&sig->sigcnt))
 672                 free_signal_struct(sig);
 673 }
 674
 675 void __put_task_struct(struct task_struct *tsk)
 676 {
 677         WARN_ON(!tsk->exit_state);
 678         WARN_ON(atomic_read(&tsk->usage));
 679         WARN_ON(tsk == current);
 680
 681         cgroup_free(tsk);
 682         task_numa_free(tsk);
 683         security_task_free(tsk);
 684         exit_creds(tsk);
 685         delayacct_tsk_free(tsk);
 686         put_signal_struct(tsk->signal);
 687
 688         if (!profile_handoff_task(tsk))
 689                 free_task(tsk);
 690 }
 691 EXPORT_SYMBOL_GPL(__put_task_struct);
 692
 693 void __init __weak arch_task_cache_init(void) { }
 694
 695 /*
 696  * set_max_threads
 697  */
 698 static void set_max_threads(unsigned int max_threads_suggested)
 699 {
 700         u64 threads;
 701
 702         /*
 703          * The number of threads shall be limited such that the thread
 704          * structures may only consume a small part of the available memory.
 705          */
 706         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 707                 threads = MAX_THREADS;
 708         else
 709                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 710                                     (u64) THREAD_SIZE * 8UL);
 711
 712         if (threads > max_threads_suggested)
 713                 threads = max_threads_suggested;
 714
 715         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 716 }
 717
 718 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 719 /* Initialized by the architecture: */
 720 int arch_task_struct_size __read_mostly;
 721 #endif
 722
 723 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 724 {
 725         /* Fetch thread_struct whitelist for the architecture. */
 726         arch_thread_struct_whitelist(offset, size);
 727
 728         /*
 729          * Handle zero-sized whitelist or empty thread_struct, otherwise
 730          * adjust offset to position of thread_struct in task_struct.
 731          */
 732         if (unlikely(*size == 0))
 733                 *offset = 0;
 734         else
 735                 *offset += offsetof(struct task_struct, thread);
 736 }
 737
 738 void __init fork_init(void)
 739 {
 740         int i;
 741 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 742 #ifndef ARCH_MIN_TASKALIGN
 743 #define ARCH_MIN_TASKALIGN      0
 744 #endif
 745         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 746         unsigned long useroffset, usersize;
 747
 748         /* create a slab on which task_structs can be allocated */
 749         task_struct_whitelist(&useroffset, &usersize);
 750         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 751                         arch_task_struct_size, align,
 752                         SLAB_PANIC|SLAB_ACCOUNT,
 753                         useroffset, usersize, NULL);
 754 #endif
 755
 756         /* do the arch specific task caches init */
 757         arch_task_cache_init();
 758
 759         set_max_threads(MAX_THREADS);
 760
 761         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 762         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 763         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 764                 init_task.signal->rlim[RLIMIT_NPROC];
 765
 766         for (i = 0; i < UCOUNT_COUNTS; i++) {
 767                 init_user_ns.ucount_max[i] = max_threads/2;
 768         }
 769
 770 #ifdef CONFIG_VMAP_STACK
 771         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 772                           NULL, free_vm_stack_cache);
 773 #endif
 774
 775         lockdep_init_task(&init_task);
 776 }
 777
 778 int __weak arch_dup_task_struct(struct task_struct *dst,
 779                                                struct task_struct *src)
 780 {
 781         *dst = *src;
 782         return 0;
 783 }
 784
 785 void set_task_stack_end_magic(struct task_struct *tsk)
 786 {
 787         unsigned long *stackend;
 788
 789         stackend = end_of_stack(tsk);
 790         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 791 }
 792
 793 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 794 {
 795         struct task_struct *tsk;
 796         unsigned long *stack;
 797         struct vm_struct *stack_vm_area;
 798         int err;
 799
 800         if (node == NUMA_NO_NODE)
 801                 node = tsk_fork_get_node(orig);
 802         tsk = alloc_task_struct_node(node);
 803         if (!tsk)
 804                 return NULL;
 805
 806         stack = alloc_thread_stack_node(tsk, node);
 807         if (!stack)
 808                 goto free_tsk;
 809
 810         stack_vm_area = task_stack_vm_area(tsk);
 811
 812         err = arch_dup_task_struct(tsk, orig);
 813
 814         /*
 815          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 816          * sure they're properly initialized before using any stack-related
 817          * functions again.
 818          */
 819         tsk->stack = stack;
 820 #ifdef CONFIG_VMAP_STACK
 821         tsk->stack_vm_area = stack_vm_area;
 822 #endif
 823 #ifdef CONFIG_THREAD_INFO_IN_TASK
 824         atomic_set(&tsk->stack_refcount, 1);
 825 #endif
 826
 827         if (err)
 828                 goto free_stack;
 829
 830 #ifdef CONFIG_SECCOMP
 831         /*
 832          * We must handle setting up seccomp filters once we're under
 833          * the sighand lock in case orig has changed between now and
 834          * then. Until then, filter must be NULL to avoid messing up
 835          * the usage counts on the error path calling free_task.
 836          */
 837         tsk->seccomp.filter = NULL;
 838 #endif
 839
 840         setup_thread_stack(tsk, orig);
 841         clear_user_return_notifier(tsk);
 842         clear_tsk_need_resched(tsk);
 843         set_task_stack_end_magic(tsk);
 844
 845 #ifdef CONFIG_STACKPROTECTOR
 846         tsk->stack_canary = get_random_canary();
 847 #endif
 848
 849         /*
 850          * One for us, one for whoever does the "release_task()" (usually
 851          * parent)
 852          */
 853         atomic_set(&tsk->usage, 2);
 854 #ifdef CONFIG_BLK_DEV_IO_TRACE
 855         tsk->btrace_seq = 0;
 856 #endif
 857         tsk->splice_pipe = NULL;
 858         tsk->task_frag.page = NULL;
 859         tsk->wake_q.next = NULL;
 860
 861         account_kernel_stack(tsk, 1);
 862
 863         kcov_task_init(tsk);
 864
 865 #ifdef CONFIG_FAULT_INJECTION
 866         tsk->fail_nth = 0;
 867 #endif
 868
 869         return tsk;
 870
 871 free_stack:
 872         free_thread_stack(tsk);
 873 free_tsk:
 874         free_task_struct(tsk);
 875         return NULL;
 876 }
 877
 878 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 879
 880 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 881
 882 static int __init coredump_filter_setup(char *s)
 883 {
 884         default_dump_filter =
 885                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 886                 MMF_DUMP_FILTER_MASK;
 887         return 1;
 888 }
 889
 890 __setup("coredump_filter=", coredump_filter_setup);
 891
 892 #include <linux/init_task.h>
 893
 894 static void mm_init_aio(struct mm_struct *mm)
 895 {
 896 #ifdef CONFIG_AIO
 897         spin_lock_init(&mm->ioctx_lock);
 898         mm->ioctx_table = NULL;
 899 #endif
 900 }
 901
 902 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 903 {
 904 #ifdef CONFIG_MEMCG
 905         mm->owner = p;
 906 #endif
 907 }
 908
 909 static void mm_init_uprobes_state(struct mm_struct *mm)
 910 {
 911 #ifdef CONFIG_UPROBES
 912         mm->uprobes_state.xol_area = NULL;
 913 #endif
 914 }
 915
 916 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 917         struct user_namespace *user_ns)
 918 {
 919         mm->mmap = NULL;
 920         mm->mm_rb = RB_ROOT;
 921         mm->vmacache_seqnum = 0;
 922         atomic_set(&mm->mm_users, 1);
 923         atomic_set(&mm->mm_count, 1);
 924         init_rwsem(&mm->mmap_sem);
 925         INIT_LIST_HEAD(&mm->mmlist);
 926         mm->core_state = NULL;
 927         mm_pgtables_bytes_init(mm);
 928         mm->map_count = 0;
 929         mm->locked_vm = 0;
 930         mm->pinned_vm = 0;
 931         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 932         spin_lock_init(&mm->page_table_lock);
 933         spin_lock_init(&mm->arg_lock);
 934         mm_init_cpumask(mm);
 935         mm_init_aio(mm);
 936         mm_init_owner(mm, p);
 937         RCU_INIT_POINTER(mm->exe_file, NULL);
 938         mmu_notifier_mm_init(mm);
 939         hmm_mm_init(mm);
 940         init_tlb_flush_pending(mm);
 941 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 942         mm->pmd_huge_pte = NULL;
 943 #endif
 944         mm_init_uprobes_state(mm);
 945
 946         if (current->mm) {
 947                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 948                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 949         } else {
 950                 mm->flags = default_dump_filter;
 951                 mm->def_flags = 0;
 952         }
 953
 954         if (mm_alloc_pgd(mm))
 955                 goto fail_nopgd;
 956
 957         if (init_new_context(p, mm))
 958                 goto fail_nocontext;
 959
 960         mm->user_ns = get_user_ns(user_ns);
 961         return mm;
 962
 963 fail_nocontext:
 964         mm_free_pgd(mm);
 965 fail_nopgd:
 966         free_mm(mm);
 967         return NULL;
 968 }
 969
 970 /*
 971  * Allocate and initialize an mm_struct.
 972  */
 973 struct mm_struct *mm_alloc(void)
 974 {
 975         struct mm_struct *mm;
 976
 977         mm = allocate_mm();
 978         if (!mm)
 979                 return NULL;
 980
 981         memset(mm, 0, sizeof(*mm));
 982         return mm_init(mm, current, current_user_ns());
 983 }
 984
 985 static inline void __mmput(struct mm_struct *mm)
 986 {
 987         VM_BUG_ON(atomic_read(&mm->mm_users));
 988
 989         uprobe_clear_state(mm);
 990         exit_aio(mm);
 991         ksm_exit(mm);
 992         khugepaged_exit(mm); /* must run before exit_mmap */
 993         exit_mmap(mm);
 994         mm_put_huge_zero_page(mm);
 995         set_mm_exe_file(mm, NULL);
 996         if (!list_empty(&mm->mmlist)) {
 997                 spin_lock(&mmlist_lock);
 998                 list_del(&mm->mmlist);
 999                 spin_unlock(&mmlist_lock);
1000         }
1001         if (mm->binfmt)
1002                 module_put(mm->binfmt->module);
1003         mmdrop(mm);
1004 }
1005
1006 /*
1007  * Decrement the use count and release all resources for an mm.
1008  */
1009 void mmput(struct mm_struct *mm)
1010 {
1011         might_sleep();
1012
1013         if (atomic_dec_and_test(&mm->mm_users))
1014                 __mmput(mm);
1015 }
1016 EXPORT_SYMBOL_GPL(mmput);
1017
1018 #ifdef CONFIG_MMU
1019 static void mmput_async_fn(struct work_struct *work)
1020 {
1021         struct mm_struct *mm = container_of(work, struct mm_struct,
1022                                             async_put_work);
1023
1024         __mmput(mm);
1025 }
1026
1027 void mmput_async(struct mm_struct *mm)
1028 {
1029         if (atomic_dec_and_test(&mm->mm_users)) {
1030                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1031                 schedule_work(&mm->async_put_work);
1032         }
1033 }
1034 #endif
1035
1036 /**
1037  * set_mm_exe_file - change a reference to the mm's executable file
1038  *
1039  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1040  *
1041  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1042  * invocations: in mmput() nobody alive left, in execve task is single
1043  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1044  * mm->exe_file, but does so without using set_mm_exe_file() in order
1045  * to do avoid the need for any locks.
1046  */
1047 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1048 {
1049         struct file *old_exe_file;
1050
1051         /*
1052          * It is safe to dereference the exe_file without RCU as
1053          * this function is only called if nobody else can access
1054          * this mm -- see comment above for justification.
1055          */
1056         old_exe_file = rcu_dereference_raw(mm->exe_file);
1057
1058         if (new_exe_file)
1059                 get_file(new_exe_file);
1060         rcu_assign_pointer(mm->exe_file, new_exe_file);
1061         if (old_exe_file)
1062                 fput(old_exe_file);
1063 }
1064
1065 /**
1066  * get_mm_exe_file - acquire a reference to the mm's executable file
1067  *
1068  * Returns %NULL if mm has no associated executable file.
1069  * User must release file via fput().
1070  */
1071 struct file *get_mm_exe_file(struct mm_struct *mm)
1072 {
1073         struct file *exe_file;
1074
1075         rcu_read_lock();
1076         exe_file = rcu_dereference(mm->exe_file);
1077         if (exe_file && !get_file_rcu(exe_file))
1078                 exe_file = NULL;
1079         rcu_read_unlock();
1080         return exe_file;
1081 }
1082 EXPORT_SYMBOL(get_mm_exe_file);
1083
1084 /**
1085  * get_task_exe_file - acquire a reference to the task's executable file
1086  *
1087  * Returns %NULL if task's mm (if any) has no associated executable file or
1088  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1089  * User must release file via fput().
1090  */
1091 struct file *get_task_exe_file(struct task_struct *task)
1092 {
1093         struct file *exe_file = NULL;
1094         struct mm_struct *mm;
1095
1096         task_lock(task);
1097         mm = task->mm;
1098         if (mm) {
1099                 if (!(task->flags & PF_KTHREAD))
1100                         exe_file = get_mm_exe_file(mm);
1101         }
1102         task_unlock(task);
1103         return exe_file;
1104 }
1105 EXPORT_SYMBOL(get_task_exe_file);
1106
1107 /**
1108  * get_task_mm - acquire a reference to the task's mm
1109  *
1110  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1111  * this kernel workthread has transiently adopted a user mm with use_mm,
1112  * to do its AIO) is not set and if so returns a reference to it, after
1113  * bumping up the use count.  User must release the mm via mmput()
1114  * after use.  Typically used by /proc and ptrace.
1115  */
1116 struct mm_struct *get_task_mm(struct task_struct *task)
1117 {
1118         struct mm_struct *mm;
1119
1120         task_lock(task);
1121         mm = task->mm;
1122         if (mm) {
1123                 if (task->flags & PF_KTHREAD)
1124                         mm = NULL;
1125                 else
1126                         mmget(mm);
1127         }
1128         task_unlock(task);
1129         return mm;
1130 }
1131 EXPORT_SYMBOL_GPL(get_task_mm);
1132
1133 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1134 {
1135         struct mm_struct *mm;
1136         int err;
1137
1138         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1139         if (err)
1140                 return ERR_PTR(err);
1141
1142         mm = get_task_mm(task);
1143         if (mm && mm != current->mm &&
1144                         !ptrace_may_access(task, mode)) {
1145                 mmput(mm);
1146                 mm = ERR_PTR(-EACCES);
1147         }
1148         mutex_unlock(&task->signal->cred_guard_mutex);
1149
1150         return mm;
1151 }
1152
1153 static void complete_vfork_done(struct task_struct *tsk)
1154 {
1155         struct completion *vfork;
1156
1157         task_lock(tsk);
1158         vfork = tsk->vfork_done;
1159         if (likely(vfork)) {
1160                 tsk->vfork_done = NULL;
1161                 complete(vfork);
1162         }
1163         task_unlock(tsk);
1164 }
1165
1166 static int wait_for_vfork_done(struct task_struct *child,
1167                                 struct completion *vfork)
1168 {
1169         int killed;
1170
1171         freezer_do_not_count();
1172         killed = wait_for_completion_killable(vfork);
1173         freezer_count();
1174
1175         if (killed) {
1176                 task_lock(child);
1177                 child->vfork_done = NULL;
1178                 task_unlock(child);
1179         }
1180
1181         put_task_struct(child);
1182         return killed;
1183 }
1184
1185 /* Please note the differences between mmput and mm_release.
1186  * mmput is called whenever we stop holding onto a mm_struct,
1187  * error success whatever.
1188  *
1189  * mm_release is called after a mm_struct has been removed
1190  * from the current process.
1191  *
1192  * This difference is important for error handling, when we
1193  * only half set up a mm_struct for a new process and need to restore
1194  * the old one.  Because we mmput the new mm_struct before
1195  * restoring the old one. . .
1196  * Eric Biederman 10 January 1998
1197  */
1198 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1199 {
1200         /* Get rid of any futexes when releasing the mm */
1201 #ifdef CONFIG_FUTEX
1202         if (unlikely(tsk->robust_list)) {
1203                 exit_robust_list(tsk);
1204                 tsk->robust_list = NULL;
1205         }
1206 #ifdef CONFIG_COMPAT
1207         if (unlikely(tsk->compat_robust_list)) {
1208                 compat_exit_robust_list(tsk);
1209                 tsk->compat_robust_list = NULL;
1210         }
1211 #endif
1212         if (unlikely(!list_empty(&tsk->pi_state_list)))
1213                 exit_pi_state_list(tsk);
1214 #endif
1215
1216         uprobe_free_utask(tsk);
1217
1218         /* Get rid of any cached register state */
1219         deactivate_mm(tsk, mm);
1220
1221         /*
1222          * Signal userspace if we're not exiting with a core dump
1223          * because we want to leave the value intact for debugging
1224          * purposes.
1225          */
1226         if (tsk->clear_child_tid) {
1227                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1228                     atomic_read(&mm->mm_users) > 1) {
1229                         /*
1230                          * We don't check the error code - if userspace has
1231                          * not set up a proper pointer then tough luck.
1232                          */
1233                         put_user(0, tsk->clear_child_tid);
1234                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1235                                         1, NULL, NULL, 0, 0);
1236                 }
1237                 tsk->clear_child_tid = NULL;
1238         }
1239
1240         /*
1241          * All done, finally we can wake up parent and return this mm to him.
1242          * Also kthread_stop() uses this completion for synchronization.
1243          */
1244         if (tsk->vfork_done)
1245                 complete_vfork_done(tsk);
1246 }
1247
1248 /*
1249  * Allocate a new mm structure and copy contents from the
1250  * mm structure of the passed in task structure.
1251  */
1252 static struct mm_struct *dup_mm(struct task_struct *tsk)
1253 {
1254         struct mm_struct *mm, *oldmm = current->mm;
1255         int err;
1256
1257         mm = allocate_mm();
1258         if (!mm)
1259                 goto fail_nomem;
1260
1261         memcpy(mm, oldmm, sizeof(*mm));
1262
1263         if (!mm_init(mm, tsk, mm->user_ns))
1264                 goto fail_nomem;
1265
1266         err = dup_mmap(mm, oldmm);
1267         if (err)
1268                 goto free_pt;
1269
1270         mm->hiwater_rss = get_mm_rss(mm);
1271         mm->hiwater_vm = mm->total_vm;
1272
1273         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1274                 goto free_pt;
1275
1276         return mm;
1277
1278 free_pt:
1279         /* don't put binfmt in mmput, we haven't got module yet */
1280         mm->binfmt = NULL;
1281         mmput(mm);
1282
1283 fail_nomem:
1284         return NULL;
1285 }
1286
1287 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1288 {
1289         struct mm_struct *mm, *oldmm;
1290         int retval;
1291
1292         tsk->min_flt = tsk->maj_flt = 0;
1293         tsk->nvcsw = tsk->nivcsw = 0;
1294 #ifdef CONFIG_DETECT_HUNG_TASK
1295         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1296 #endif
1297
1298         tsk->mm = NULL;
1299         tsk->active_mm = NULL;
1300
1301         /*
1302          * Are we cloning a kernel thread?
1303          *
1304          * We need to steal a active VM for that..
1305          */
1306         oldmm = current->mm;
1307         if (!oldmm)
1308                 return 0;
1309
1310         /* initialize the new vmacache entries */
1311         vmacache_flush(tsk);
1312
1313         if (clone_flags & CLONE_VM) {
1314                 mmget(oldmm);
1315                 mm = oldmm;
1316                 goto good_mm;
1317         }
1318
1319         retval = -ENOMEM;
1320         mm = dup_mm(tsk);
1321         if (!mm)
1322                 goto fail_nomem;
1323
1324 good_mm:
1325         tsk->mm = mm;
1326         tsk->active_mm = mm;
1327         return 0;
1328
1329 fail_nomem:
1330         return retval;
1331 }
1332
1333 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1334 {
1335         struct fs_struct *fs = current->fs;
1336         if (clone_flags & CLONE_FS) {
1337                 /* tsk->fs is already what we want */
1338                 spin_lock(&fs->lock);
1339                 if (fs->in_exec) {
1340                         spin_unlock(&fs->lock);
1341                         return -EAGAIN;
1342                 }
1343                 fs->users++;
1344                 spin_unlock(&fs->lock);
1345                 return 0;
1346         }
1347         tsk->fs = copy_fs_struct(fs);
1348         if (!tsk->fs)
1349                 return -ENOMEM;
1350         return 0;
1351 }
1352
1353 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1354 {
1355         struct files_struct *oldf, *newf;
1356         int error = 0;
1357
1358         /*
1359          * A background process may not have any files ...
1360          */
1361         oldf = current->files;
1362         if (!oldf)
1363                 goto out;
1364
1365         if (clone_flags & CLONE_FILES) {
1366                 atomic_inc(&oldf->count);
1367                 goto out;
1368         }
1369
1370         newf = dup_fd(oldf, &error);
1371         if (!newf)
1372                 goto out;
1373
1374         tsk->files = newf;
1375         error = 0;
1376 out:
1377         return error;
1378 }
1379
1380 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1381 {
1382 #ifdef CONFIG_BLOCK
1383         struct io_context *ioc = current->io_context;
1384         struct io_context *new_ioc;
1385
1386         if (!ioc)
1387                 return 0;
1388         /*
1389          * Share io context with parent, if CLONE_IO is set
1390          */
1391         if (clone_flags & CLONE_IO) {
1392                 ioc_task_link(ioc);
1393                 tsk->io_context = ioc;
1394         } else if (ioprio_valid(ioc->ioprio)) {
1395                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1396                 if (unlikely(!new_ioc))
1397                         return -ENOMEM;
1398
1399                 new_ioc->ioprio = ioc->ioprio;
1400                 put_io_context(new_ioc);
1401         }
1402 #endif
1403         return 0;
1404 }
1405
1406 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1407 {
1408         struct sighand_struct *sig;
1409
1410         if (clone_flags & CLONE_SIGHAND) {
1411                 atomic_inc(&current->sighand->count);
1412                 return 0;
1413         }
1414         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1415         rcu_assign_pointer(tsk->sighand, sig);
1416         if (!sig)
1417                 return -ENOMEM;
1418
1419         atomic_set(&sig->count, 1);
1420         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1421         return 0;
1422 }
1423
1424 void __cleanup_sighand(struct sighand_struct *sighand)
1425 {
1426         if (atomic_dec_and_test(&sighand->count)) {
1427                 signalfd_cleanup(sighand);
1428                 /*
1429                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1430                  * without an RCU grace period, see __lock_task_sighand().
1431                  */
1432                 kmem_cache_free(sighand_cachep, sighand);
1433         }
1434 }
1435
1436 #ifdef CONFIG_POSIX_TIMERS
1437 /*
1438  * Initialize POSIX timer handling for a thread group.
1439  */
1440 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1441 {
1442         unsigned long cpu_limit;
1443
1444         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1445         if (cpu_limit != RLIM_INFINITY) {
1446                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1447                 sig->cputimer.running = true;
1448         }
1449
1450         /* The timer lists. */
1451         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1452         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1453         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1454 }
1455 #else
1456 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1457 #endif
1458
1459 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1460 {
1461         struct signal_struct *sig;
1462
1463         if (clone_flags & CLONE_THREAD)
1464                 return 0;
1465
1466         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1467         tsk->signal = sig;
1468         if (!sig)
1469                 return -ENOMEM;
1470
1471         sig->nr_threads = 1;
1472         atomic_set(&sig->live, 1);
1473         atomic_set(&sig->sigcnt, 1);
1474
1475         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1476         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1477         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1478
1479         init_waitqueue_head(&sig->wait_chldexit);
1480         sig->curr_target = tsk;
1481         init_sigpending(&sig->shared_pending);
1482         seqlock_init(&sig->stats_lock);
1483         prev_cputime_init(&sig->prev_cputime);
1484
1485 #ifdef CONFIG_POSIX_TIMERS
1486         INIT_LIST_HEAD(&sig->posix_timers);
1487         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1488         sig->real_timer.function = it_real_fn;
1489 #endif
1490
1491         task_lock(current->group_leader);
1492         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1493         task_unlock(current->group_leader);
1494
1495         posix_cpu_timers_init_group(sig);
1496
1497         tty_audit_fork(sig);
1498         sched_autogroup_fork(sig);
1499
1500         sig->oom_score_adj = current->signal->oom_score_adj;
1501         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1502
1503         mutex_init(&sig->cred_guard_mutex);
1504
1505         return 0;
1506 }
1507
1508 static void copy_seccomp(struct task_struct *p)
1509 {
1510 #ifdef CONFIG_SECCOMP
1511         /*
1512          * Must be called with sighand->lock held, which is common to
1513          * all threads in the group. Holding cred_guard_mutex is not
1514          * needed because this new task is not yet running and cannot
1515          * be racing exec.
1516          */
1517         assert_spin_locked(&current->sighand->siglock);
1518
1519         /* Ref-count the new filter user, and assign it. */
1520         get_seccomp_filter(current);
1521         p->seccomp = current->seccomp;
1522
1523         /*
1524          * Explicitly enable no_new_privs here in case it got set
1525          * between the task_struct being duplicated and holding the
1526          * sighand lock. The seccomp state and nnp must be in sync.
1527          */
1528         if (task_no_new_privs(current))
1529                 task_set_no_new_privs(p);
1530
1531         /*
1532          * If the parent gained a seccomp mode after copying thread
1533          * flags and between before we held the sighand lock, we have
1534          * to manually enable the seccomp thread flag here.
1535          */
1536         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1537                 set_tsk_thread_flag(p, TIF_SECCOMP);
1538 #endif
1539 }
1540
1541 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1542 {
1543         current->clear_child_tid = tidptr;
1544
1545         return task_pid_vnr(current);
1546 }
1547
1548 static void rt_mutex_init_task(struct task_struct *p)
1549 {
1550         raw_spin_lock_init(&p->pi_lock);
1551 #ifdef CONFIG_RT_MUTEXES
1552         p->pi_waiters = RB_ROOT_CACHED;
1553         p->pi_top_task = NULL;
1554         p->pi_blocked_on = NULL;
1555 #endif
1556 }
1557
1558 #ifdef CONFIG_POSIX_TIMERS
1559 /*
1560  * Initialize POSIX timer handling for a single task.
1561  */
1562 static void posix_cpu_timers_init(struct task_struct *tsk)
1563 {
1564         tsk->cputime_expires.prof_exp = 0;
1565         tsk->cputime_expires.virt_exp = 0;
1566         tsk->cputime_expires.sched_exp = 0;
1567         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1568         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1569         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1570 }
1571 #else
1572 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1573 #endif
1574
1575 static inline void
1576 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1577 {
1578          task->pids[type].pid = pid;
1579 }
1580
1581 static inline void rcu_copy_process(struct task_struct *p)
1582 {
1583 #ifdef CONFIG_PREEMPT_RCU
1584         p->rcu_read_lock_nesting = 0;
1585         p->rcu_read_unlock_special.s = 0;
1586         p->rcu_blocked_node = NULL;
1587         INIT_LIST_HEAD(&p->rcu_node_entry);
1588 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1589 #ifdef CONFIG_TASKS_RCU
1590         p->rcu_tasks_holdout = false;
1591         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1592         p->rcu_tasks_idle_cpu = -1;
1593 #endif /* #ifdef CONFIG_TASKS_RCU */
1594 }
1595
1596 /*
1597  * This creates a new process as a copy of the old one,
1598  * but does not actually start it yet.
1599  *
1600  * It copies the registers, and all the appropriate
1601  * parts of the process environment (as per the clone
1602  * flags). The actual kick-off is left to the caller.
1603  */
1604 static __latent_entropy struct task_struct *copy_process(
1605                                         unsigned long clone_flags,
1606                                         unsigned long stack_start,
1607                                         unsigned long stack_size,
1608                                         int __user *child_tidptr,
1609                                         struct pid *pid,
1610                                         int trace,
1611                                         unsigned long tls,
1612                                         int node)
1613 {
1614         int retval;
1615         struct task_struct *p;
1616
1617         /*
1618          * Don't allow sharing the root directory with processes in a different
1619          * namespace
1620          */
1621         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1622                 return ERR_PTR(-EINVAL);
1623
1624         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1625                 return ERR_PTR(-EINVAL);
1626
1627         /*
1628          * Thread groups must share signals as well, and detached threads
1629          * can only be started up within the thread group.
1630          */
1631         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1632                 return ERR_PTR(-EINVAL);
1633
1634         /*
1635          * Shared signal handlers imply shared VM. By way of the above,
1636          * thread groups also imply shared VM. Blocking this case allows
1637          * for various simplifications in other code.
1638          */
1639         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1640                 return ERR_PTR(-EINVAL);
1641
1642         /*
1643          * Siblings of global init remain as zombies on exit since they are
1644          * not reaped by their parent (swapper). To solve this and to avoid
1645          * multi-rooted process trees, prevent global and container-inits
1646          * from creating siblings.
1647          */
1648         if ((clone_flags & CLONE_PARENT) &&
1649                                 current->signal->flags & SIGNAL_UNKILLABLE)
1650                 return ERR_PTR(-EINVAL);
1651
1652         /*
1653          * If the new process will be in a different pid or user namespace
1654          * do not allow it to share a thread group with the forking task.
1655          */
1656         if (clone_flags & CLONE_THREAD) {
1657                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1658                     (task_active_pid_ns(current) !=
1659                                 current->nsproxy->pid_ns_for_children))
1660                         return ERR_PTR(-EINVAL);
1661         }
1662
1663         retval = -ENOMEM;
1664         p = dup_task_struct(current, node);
1665         if (!p)
1666                 goto fork_out;
1667
1668         /*
1669          * This _must_ happen before we call free_task(), i.e. before we jump
1670          * to any of the bad_fork_* labels. This is to avoid freeing
1671          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1672          * kernel threads (PF_KTHREAD).
1673          */
1674         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1675         /*
1676          * Clear TID on mm_release()?
1677          */
1678         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1679
1680         ftrace_graph_init_task(p);
1681
1682         rt_mutex_init_task(p);
1683
1684 #ifdef CONFIG_PROVE_LOCKING
1685         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1686         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1687 #endif
1688         retval = -EAGAIN;
1689         if (atomic_read(&p->real_cred->user->processes) >=
1690                         task_rlimit(p, RLIMIT_NPROC)) {
1691                 if (p->real_cred->user != INIT_USER &&
1692                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1693                         goto bad_fork_free;
1694         }
1695         current->flags &= ~PF_NPROC_EXCEEDED;
1696
1697         retval = copy_creds(p, clone_flags);
1698         if (retval < 0)
1699                 goto bad_fork_free;
1700
1701         /*
1702          * If multiple threads are within copy_process(), then this check
1703          * triggers too late. This doesn't hurt, the check is only there
1704          * to stop root fork bombs.
1705          */
1706         retval = -EAGAIN;
1707         if (nr_threads >= max_threads)
1708                 goto bad_fork_cleanup_count;
1709
1710         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1711         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1712         p->flags |= PF_FORKNOEXEC;
1713         INIT_LIST_HEAD(&p->children);
1714         INIT_LIST_HEAD(&p->sibling);
1715         rcu_copy_process(p);
1716         p->vfork_done = NULL;
1717         spin_lock_init(&p->alloc_lock);
1718
1719         init_sigpending(&p->pending);
1720
1721         p->utime = p->stime = p->gtime = 0;
1722 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1723         p->utimescaled = p->stimescaled = 0;
1724 #endif
1725         prev_cputime_init(&p->prev_cputime);
1726
1727 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1728         seqcount_init(&p->vtime.seqcount);
1729         p->vtime.starttime = 0;
1730         p->vtime.state = VTIME_INACTIVE;
1731 #endif
1732
1733 #if defined(SPLIT_RSS_COUNTING)
1734         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1735 #endif
1736
1737         p->default_timer_slack_ns = current->timer_slack_ns;
1738
1739         task_io_accounting_init(&p->ioac);
1740         acct_clear_integrals(p);
1741
1742         posix_cpu_timers_init(p);
1743
1744         p->start_time = ktime_get_ns();
1745         p->real_start_time = ktime_get_boot_ns();
1746         p->io_context = NULL;
1747         audit_set_context(p, NULL);
1748         cgroup_fork(p);
1749 #ifdef CONFIG_NUMA
1750         p->mempolicy = mpol_dup(p->mempolicy);
1751         if (IS_ERR(p->mempolicy)) {
1752                 retval = PTR_ERR(p->mempolicy);
1753                 p->mempolicy = NULL;
1754                 goto bad_fork_cleanup_threadgroup_lock;
1755         }
1756 #endif
1757 #ifdef CONFIG_CPUSETS
1758         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1759         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1760         seqcount_init(&p->mems_allowed_seq);
1761 #endif
1762 #ifdef CONFIG_TRACE_IRQFLAGS
1763         p->irq_events = 0;
1764         p->hardirqs_enabled = 0;
1765         p->hardirq_enable_ip = 0;
1766         p->hardirq_enable_event = 0;
1767         p->hardirq_disable_ip = _THIS_IP_;
1768         p->hardirq_disable_event = 0;
1769         p->softirqs_enabled = 1;
1770         p->softirq_enable_ip = _THIS_IP_;
1771         p->softirq_enable_event = 0;
1772         p->softirq_disable_ip = 0;
1773         p->softirq_disable_event = 0;
1774         p->hardirq_context = 0;
1775         p->softirq_context = 0;
1776 #endif
1777
1778         p->pagefault_disabled = 0;
1779
1780 #ifdef CONFIG_LOCKDEP
1781         p->lockdep_depth = 0; /* no locks held yet */
1782         p->curr_chain_key = 0;
1783         p->lockdep_recursion = 0;
1784         lockdep_init_task(p);
1785 #endif
1786
1787 #ifdef CONFIG_DEBUG_MUTEXES
1788         p->blocked_on = NULL; /* not blocked yet */
1789 #endif
1790 #ifdef CONFIG_BCACHE
1791         p->sequential_io        = 0;
1792         p->sequential_io_avg    = 0;
1793 #endif
1794
1795         /* Perform scheduler related setup. Assign this task to a CPU. */
1796         retval = sched_fork(clone_flags, p);
1797         if (retval)
1798                 goto bad_fork_cleanup_policy;
1799
1800         retval = perf_event_init_task(p);
1801         if (retval)
1802                 goto bad_fork_cleanup_policy;
1803         retval = audit_alloc(p);
1804         if (retval)
1805                 goto bad_fork_cleanup_perf;
1806         /* copy all the process information */
1807         shm_init_task(p);
1808         retval = security_task_alloc(p, clone_flags);
1809         if (retval)
1810                 goto bad_fork_cleanup_audit;
1811         retval = copy_semundo(clone_flags, p);
1812         if (retval)
1813                 goto bad_fork_cleanup_security;
1814         retval = copy_files(clone_flags, p);
1815         if (retval)
1816                 goto bad_fork_cleanup_semundo;
1817         retval = copy_fs(clone_flags, p);
1818         if (retval)
1819                 goto bad_fork_cleanup_files;
1820         retval = copy_sighand(clone_flags, p);
1821         if (retval)
1822                 goto bad_fork_cleanup_fs;
1823         retval = copy_signal(clone_flags, p);
1824         if (retval)
1825                 goto bad_fork_cleanup_sighand;
1826         retval = copy_mm(clone_flags, p);
1827         if (retval)
1828                 goto bad_fork_cleanup_signal;
1829         retval = copy_namespaces(clone_flags, p);
1830         if (retval)
1831                 goto bad_fork_cleanup_mm;
1832         retval = copy_io(clone_flags, p);
1833         if (retval)
1834                 goto bad_fork_cleanup_namespaces;
1835         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1836         if (retval)
1837                 goto bad_fork_cleanup_io;
1838
1839         if (pid != &init_struct_pid) {
1840                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1841                 if (IS_ERR(pid)) {
1842                         retval = PTR_ERR(pid);
1843                         goto bad_fork_cleanup_thread;
1844                 }
1845         }
1846
1847 #ifdef CONFIG_BLOCK
1848         p->plug = NULL;
1849 #endif
1850 #ifdef CONFIG_FUTEX
1851         p->robust_list = NULL;
1852 #ifdef CONFIG_COMPAT
1853         p->compat_robust_list = NULL;
1854 #endif
1855         INIT_LIST_HEAD(&p->pi_state_list);
1856         p->pi_state_cache = NULL;
1857 #endif
1858         /*
1859          * sigaltstack should be cleared when sharing the same VM
1860          */
1861         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1862                 sas_ss_reset(p);
1863
1864         /*
1865          * Syscall tracing and stepping should be turned off in the
1866          * child regardless of CLONE_PTRACE.
1867          */
1868         user_disable_single_step(p);
1869         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1870 #ifdef TIF_SYSCALL_EMU
1871         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1872 #endif
1873         clear_all_latency_tracing(p);
1874
1875         /* ok, now we should be set up.. */
1876         p->pid = pid_nr(pid);
1877         if (clone_flags & CLONE_THREAD) {
1878                 p->exit_signal = -1;
1879                 p->group_leader = current->group_leader;
1880                 p->tgid = current->tgid;
1881         } else {
1882                 if (clone_flags & CLONE_PARENT)
1883                         p->exit_signal = current->group_leader->exit_signal;
1884                 else
1885                         p->exit_signal = (clone_flags & CSIGNAL);
1886                 p->group_leader = p;
1887                 p->tgid = p->pid;
1888         }
1889
1890         p->nr_dirtied = 0;
1891         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1892         p->dirty_paused_when = 0;
1893
1894         p->pdeath_signal = 0;
1895         INIT_LIST_HEAD(&p->thread_group);
1896         p->task_works = NULL;
1897
1898         cgroup_threadgroup_change_begin(current);
1899         /*
1900          * Ensure that the cgroup subsystem policies allow the new process to be
1901          * forked. It should be noted the the new process's css_set can be changed
1902          * between here and cgroup_post_fork() if an organisation operation is in
1903          * progress.
1904          */
1905         retval = cgroup_can_fork(p);
1906         if (retval)
1907                 goto bad_fork_free_pid;
1908
1909         /*
1910          * Make it visible to the rest of the system, but dont wake it up yet.
1911          * Need tasklist lock for parent etc handling!
1912          */
1913         write_lock_irq(&tasklist_lock);
1914
1915         /* CLONE_PARENT re-uses the old parent */
1916         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1917                 p->real_parent = current->real_parent;
1918                 p->parent_exec_id = current->parent_exec_id;
1919         } else {
1920                 p->real_parent = current;
1921                 p->parent_exec_id = current->self_exec_id;
1922         }
1923
1924         klp_copy_process(p);
1925
1926         spin_lock(&current->sighand->siglock);
1927
1928         /*
1929          * Copy seccomp details explicitly here, in case they were changed
1930          * before holding sighand lock.
1931          */
1932         copy_seccomp(p);
1933
1934         rseq_fork(p, clone_flags);
1935
1936         /*
1937          * Process group and session signals need to be delivered to just the
1938          * parent before the fork or both the parent and the child after the
1939          * fork. Restart if a signal comes in before we add the new process to
1940          * it's process group.
1941          * A fatal signal pending means that current will exit, so the new
1942          * thread can't slip out of an OOM kill (or normal SIGKILL).
1943         */
1944         recalc_sigpending();
1945         if (signal_pending(current)) {
1946                 retval = -ERESTARTNOINTR;
1947                 goto bad_fork_cancel_cgroup;
1948         }
1949         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
1950                 retval = -ENOMEM;
1951                 goto bad_fork_cancel_cgroup;
1952         }
1953
1954         if (likely(p->pid)) {
1955                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1956
1957                 init_task_pid(p, PIDTYPE_PID, pid);
1958                 if (thread_group_leader(p)) {
1959                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1960                         init_task_pid(p, PIDTYPE_SID, task_session(current));
1961
1962                         if (is_child_reaper(pid)) {
1963                                 ns_of_pid(pid)->child_reaper = p;
1964                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1965                         }
1966
1967                         p->signal->leader_pid = pid;
1968                         p->signal->tty = tty_kref_get(current->signal->tty);
1969                         /*
1970                          * Inherit has_child_subreaper flag under the same
1971                          * tasklist_lock with adding child to the process tree
1972                          * for propagate_has_child_subreaper optimization.
1973                          */
1974                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1975                                                          p->real_parent->signal->is_child_subreaper;
1976                         list_add_tail(&p->sibling, &p->real_parent->children);
1977                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1978                         attach_pid(p, PIDTYPE_PGID);
1979                         attach_pid(p, PIDTYPE_SID);
1980                         __this_cpu_inc(process_counts);
1981                 } else {
1982                         current->signal->nr_threads++;
1983                         atomic_inc(&current->signal->live);
1984                         atomic_inc(&current->signal->sigcnt);
1985                         list_add_tail_rcu(&p->thread_group,
1986                                           &p->group_leader->thread_group);
1987                         list_add_tail_rcu(&p->thread_node,
1988                                           &p->signal->thread_head);
1989                 }
1990                 attach_pid(p, PIDTYPE_PID);
1991                 nr_threads++;
1992         }
1993
1994         total_forks++;
1995         spin_unlock(&current->sighand->siglock);
1996         syscall_tracepoint_update(p);
1997         write_unlock_irq(&tasklist_lock);
1998
1999         proc_fork_connector(p);
2000         cgroup_post_fork(p);
2001         cgroup_threadgroup_change_end(current);
2002         perf_event_fork(p);
2003
2004         trace_task_newtask(p, clone_flags);
2005         uprobe_copy_process(p, clone_flags);
2006
2007         return p;
2008
2009 bad_fork_cancel_cgroup:
2010         spin_unlock(&current->sighand->siglock);
2011         write_unlock_irq(&tasklist_lock);
2012         cgroup_cancel_fork(p);
2013 bad_fork_free_pid:
2014         cgroup_threadgroup_change_end(current);
2015         if (pid != &init_struct_pid)
2016                 free_pid(pid);
2017 bad_fork_cleanup_thread:
2018         exit_thread(p);
2019 bad_fork_cleanup_io:
2020         if (p->io_context)
2021                 exit_io_context(p);
2022 bad_fork_cleanup_namespaces:
2023         exit_task_namespaces(p);
2024 bad_fork_cleanup_mm:
2025         if (p->mm)
2026                 mmput(p->mm);
2027 bad_fork_cleanup_signal:
2028         if (!(clone_flags & CLONE_THREAD))
2029                 free_signal_struct(p->signal);
2030 bad_fork_cleanup_sighand:
2031         __cleanup_sighand(p->sighand);
2032 bad_fork_cleanup_fs:
2033         exit_fs(p); /* blocking */
2034 bad_fork_cleanup_files:
2035         exit_files(p); /* blocking */
2036 bad_fork_cleanup_semundo:
2037         exit_sem(p);
2038 bad_fork_cleanup_security:
2039         security_task_free(p);
2040 bad_fork_cleanup_audit:
2041         audit_free(p);
2042 bad_fork_cleanup_perf:
2043         perf_event_free_task(p);
2044 bad_fork_cleanup_policy:
2045         lockdep_free_task(p);
2046 #ifdef CONFIG_NUMA
2047         mpol_put(p->mempolicy);
2048 bad_fork_cleanup_threadgroup_lock:
2049 #endif
2050         delayacct_tsk_free(p);
2051 bad_fork_cleanup_count:
2052         atomic_dec(&p->cred->user->processes);
2053         exit_creds(p);
2054 bad_fork_free:
2055         p->state = TASK_DEAD;
2056         put_task_stack(p);
2057         free_task(p);
2058 fork_out:
2059         return ERR_PTR(retval);
2060 }
2061
2062 static inline void init_idle_pids(struct pid_link *links)
2063 {
2064         enum pid_type type;
2065
2066         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2067                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
2068                 links[type].pid = &init_struct_pid;
2069         }
2070 }
2071
2072 struct task_struct *fork_idle(int cpu)
2073 {
2074         struct task_struct *task;
2075         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
2076                             cpu_to_node(cpu));
2077         if (!IS_ERR(task)) {
2078                 init_idle_pids(task->pids);
2079                 init_idle(task, cpu);
2080         }
2081
2082         return task;
2083 }
2084
2085 /*
2086  *  Ok, this is the main fork-routine.
2087  *
2088  * It copies the process, and if successful kick-starts
2089  * it and waits for it to finish using the VM if required.
2090  */
2091 long _do_fork(unsigned long clone_flags,
2092               unsigned long stack_start,
2093               unsigned long stack_size,
2094               int __user *parent_tidptr,
2095               int __user *child_tidptr,
2096               unsigned long tls)
2097 {
2098         struct completion vfork;
2099         struct pid *pid;
2100         struct task_struct *p;
2101         int trace = 0;
2102         long nr;
2103
2104         /*
2105          * Determine whether and which event to report to ptracer.  When
2106          * called from kernel_thread or CLONE_UNTRACED is explicitly
2107          * requested, no event is reported; otherwise, report if the event
2108          * for the type of forking is enabled.
2109          */
2110         if (!(clone_flags & CLONE_UNTRACED)) {
2111                 if (clone_flags & CLONE_VFORK)
2112                         trace = PTRACE_EVENT_VFORK;
2113                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2114                         trace = PTRACE_EVENT_CLONE;
2115                 else
2116                         trace = PTRACE_EVENT_FORK;
2117
2118                 if (likely(!ptrace_event_enabled(current, trace)))
2119                         trace = 0;
2120         }
2121
2122         p = copy_process(clone_flags, stack_start, stack_size,
2123                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2124         add_latent_entropy();
2125
2126         if (IS_ERR(p))
2127                 return PTR_ERR(p);
2128
2129         /*
2130          * Do this prior waking up the new thread - the thread pointer
2131          * might get invalid after that point, if the thread exits quickly.
2132          */
2133         trace_sched_process_fork(current, p);
2134
2135         pid = get_task_pid(p, PIDTYPE_PID);
2136         nr = pid_vnr(pid);
2137
2138         if (clone_flags & CLONE_PARENT_SETTID)
2139                 put_user(nr, parent_tidptr);
2140
2141         if (clone_flags & CLONE_VFORK) {
2142                 p->vfork_done = &vfork;
2143                 init_completion(&vfork);
2144                 get_task_struct(p);
2145         }
2146
2147         wake_up_new_task(p);
2148
2149         /* forking complete and child started to run, tell ptracer */
2150         if (unlikely(trace))
2151                 ptrace_event_pid(trace, pid);
2152
2153         if (clone_flags & CLONE_VFORK) {
2154                 if (!wait_for_vfork_done(p, &vfork))
2155                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2156         }
2157
2158         put_pid(pid);
2159         return nr;
2160 }
2161
2162 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2163 /* For compatibility with architectures that call do_fork directly rather than
2164  * using the syscall entry points below. */
2165 long do_fork(unsigned long clone_flags,
2166               unsigned long stack_start,
2167               unsigned long stack_size,
2168               int __user *parent_tidptr,
2169               int __user *child_tidptr)
2170 {
2171         return _do_fork(clone_flags, stack_start, stack_size,
2172                         parent_tidptr, child_tidptr, 0);
2173 }
2174 #endif
2175
2176 /*
2177  * Create a kernel thread.
2178  */
2179 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2180 {
2181         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2182                 (unsigned long)arg, NULL, NULL, 0);
2183 }
2184
2185 #ifdef __ARCH_WANT_SYS_FORK
2186 SYSCALL_DEFINE0(fork)
2187 {
2188 #ifdef CONFIG_MMU
2189         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2190 #else
2191         /* can not support in nommu mode */
2192         return -EINVAL;
2193 #endif
2194 }
2195 #endif
2196
2197 #ifdef __ARCH_WANT_SYS_VFORK
2198 SYSCALL_DEFINE0(vfork)
2199 {
2200         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2201                         0, NULL, NULL, 0);
2202 }
2203 #endif
2204
2205 #ifdef __ARCH_WANT_SYS_CLONE
2206 #ifdef CONFIG_CLONE_BACKWARDS
2207 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2208                  int __user *, parent_tidptr,
2209                  unsigned long, tls,
2210                  int __user *, child_tidptr)
2211 #elif defined(CONFIG_CLONE_BACKWARDS2)
2212 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2213                  int __user *, parent_tidptr,
2214                  int __user *, child_tidptr,
2215                  unsigned long, tls)
2216 #elif defined(CONFIG_CLONE_BACKWARDS3)
2217 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2218                 int, stack_size,
2219                 int __user *, parent_tidptr,
2220                 int __user *, child_tidptr,
2221                 unsigned long, tls)
2222 #else
2223 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2224                  int __user *, parent_tidptr,
2225                  int __user *, child_tidptr,
2226                  unsigned long, tls)
2227 #endif
2228 {
2229         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2230 }
2231 #endif
2232
2233 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2234 {
2235         struct task_struct *leader, *parent, *child;
2236         int res;
2237
2238         read_lock(&tasklist_lock);
2239         leader = top = top->group_leader;
2240 down:
2241         for_each_thread(leader, parent) {
2242                 list_for_each_entry(child, &parent->children, sibling) {
2243                         res = visitor(child, data);
2244                         if (res) {
2245                                 if (res < 0)
2246                                         goto out;
2247                                 leader = child;
2248                                 goto down;
2249                         }
2250 up:
2251                         ;
2252                 }
2253         }
2254
2255         if (leader != top) {
2256                 child = leader;
2257                 parent = child->real_parent;
2258                 leader = parent->group_leader;
2259                 goto up;
2260         }
2261 out:
2262         read_unlock(&tasklist_lock);
2263 }
2264
2265 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2266 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2267 #endif
2268
2269 static void sighand_ctor(void *data)
2270 {
2271         struct sighand_struct *sighand = data;
2272
2273         spin_lock_init(&sighand->siglock);
2274         init_waitqueue_head(&sighand->signalfd_wqh);
2275 }
2276
2277 void __init proc_caches_init(void)
2278 {
2279         unsigned int mm_size;
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         /*
2299          * The mm_cpumask is located at the end of mm_struct, and is
2300          * dynamically sized based on the maximum CPU number this system
2301          * can have, taking hotplug into account (nr_cpu_ids).
2302          */
2303         mm_size = sizeof(struct mm_struct) + cpumask_size();
2304
2305         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2306                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2307                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2308                         offsetof(struct mm_struct, saved_auxv),
2309                         sizeof_field(struct mm_struct, saved_auxv),
2310                         NULL);
2311         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2312         mmap_init();
2313         nsproxy_cache_init();
2314 }
2315
2316 /*
2317  * Check constraints on flags passed to the unshare system call.
2318  */
2319 static int check_unshare_flags(unsigned long unshare_flags)
2320 {
2321         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2322                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2323                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2324                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2325                 return -EINVAL;
2326         /*
2327          * Not implemented, but pretend it works if there is nothing
2328          * to unshare.  Note that unsharing the address space or the
2329          * signal handlers also need to unshare the signal queues (aka
2330          * CLONE_THREAD).
2331          */
2332         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2333                 if (!thread_group_empty(current))
2334                         return -EINVAL;
2335         }
2336         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2337                 if (atomic_read(&current->sighand->count) > 1)
2338                         return -EINVAL;
2339         }
2340         if (unshare_flags & CLONE_VM) {
2341                 if (!current_is_single_threaded())
2342                         return -EINVAL;
2343         }
2344
2345         return 0;
2346 }
2347
2348 /*
2349  * Unshare the filesystem structure if it is being shared
2350  */
2351 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2352 {
2353         struct fs_struct *fs = current->fs;
2354
2355         if (!(unshare_flags & CLONE_FS) || !fs)
2356                 return 0;
2357
2358         /* don't need lock here; in the worst case we'll do useless copy */
2359         if (fs->users == 1)
2360                 return 0;
2361
2362         *new_fsp = copy_fs_struct(fs);
2363         if (!*new_fsp)
2364                 return -ENOMEM;
2365
2366         return 0;
2367 }
2368
2369 /*
2370  * Unshare file descriptor table if it is being shared
2371  */
2372 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2373 {
2374         struct files_struct *fd = current->files;
2375         int error = 0;
2376
2377         if ((unshare_flags & CLONE_FILES) &&
2378             (fd && atomic_read(&fd->count) > 1)) {
2379                 *new_fdp = dup_fd(fd, &error);
2380                 if (!*new_fdp)
2381                         return error;
2382         }
2383
2384         return 0;
2385 }
2386
2387 /*
2388  * unshare allows a process to 'unshare' part of the process
2389  * context which was originally shared using clone.  copy_*
2390  * functions used by do_fork() cannot be used here directly
2391  * because they modify an inactive task_struct that is being
2392  * constructed. Here we are modifying the current, active,
2393  * task_struct.
2394  */
2395 int ksys_unshare(unsigned long unshare_flags)
2396 {
2397         struct fs_struct *fs, *new_fs = NULL;
2398         struct files_struct *fd, *new_fd = NULL;
2399         struct cred *new_cred = NULL;
2400         struct nsproxy *new_nsproxy = NULL;
2401         int do_sysvsem = 0;
2402         int err;
2403
2404         /*
2405          * If unsharing a user namespace must also unshare the thread group
2406          * and unshare the filesystem root and working directories.
2407          */
2408         if (unshare_flags & CLONE_NEWUSER)
2409                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2410         /*
2411          * If unsharing vm, must also unshare signal handlers.
2412          */
2413         if (unshare_flags & CLONE_VM)
2414                 unshare_flags |= CLONE_SIGHAND;
2415         /*
2416          * If unsharing a signal handlers, must also unshare the signal queues.
2417          */
2418         if (unshare_flags & CLONE_SIGHAND)
2419                 unshare_flags |= CLONE_THREAD;
2420         /*
2421          * If unsharing namespace, must also unshare filesystem information.
2422          */
2423         if (unshare_flags & CLONE_NEWNS)
2424                 unshare_flags |= CLONE_FS;
2425
2426         err = check_unshare_flags(unshare_flags);
2427         if (err)
2428                 goto bad_unshare_out;
2429         /*
2430          * CLONE_NEWIPC must also detach from the undolist: after switching
2431          * to a new ipc namespace, the semaphore arrays from the old
2432          * namespace are unreachable.
2433          */
2434         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2435                 do_sysvsem = 1;
2436         err = unshare_fs(unshare_flags, &new_fs);
2437         if (err)
2438                 goto bad_unshare_out;
2439         err = unshare_fd(unshare_flags, &new_fd);
2440         if (err)
2441                 goto bad_unshare_cleanup_fs;
2442         err = unshare_userns(unshare_flags, &new_cred);
2443         if (err)
2444                 goto bad_unshare_cleanup_fd;
2445         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2446                                          new_cred, new_fs);
2447         if (err)
2448                 goto bad_unshare_cleanup_cred;
2449
2450         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2451                 if (do_sysvsem) {
2452                         /*
2453                          * CLONE_SYSVSEM is equivalent to sys_exit().
2454                          */
2455                         exit_sem(current);
2456                 }
2457                 if (unshare_flags & CLONE_NEWIPC) {
2458                         /* Orphan segments in old ns (see sem above). */
2459                         exit_shm(current);
2460                         shm_init_task(current);
2461                 }
2462
2463                 if (new_nsproxy)
2464                         switch_task_namespaces(current, new_nsproxy);
2465
2466                 task_lock(current);
2467
2468                 if (new_fs) {
2469                         fs = current->fs;
2470                         spin_lock(&fs->lock);
2471                         current->fs = new_fs;
2472                         if (--fs->users)
2473                                 new_fs = NULL;
2474                         else
2475                                 new_fs = fs;
2476                         spin_unlock(&fs->lock);
2477                 }
2478
2479                 if (new_fd) {
2480                         fd = current->files;
2481                         current->files = new_fd;
2482                         new_fd = fd;
2483                 }
2484
2485                 task_unlock(current);
2486
2487                 if (new_cred) {
2488                         /* Install the new user namespace */
2489                         commit_creds(new_cred);
2490                         new_cred = NULL;
2491                 }
2492         }
2493
2494         perf_event_namespaces(current);
2495
2496 bad_unshare_cleanup_cred:
2497         if (new_cred)
2498                 put_cred(new_cred);
2499 bad_unshare_cleanup_fd:
2500         if (new_fd)
2501                 put_files_struct(new_fd);
2502
2503 bad_unshare_cleanup_fs:
2504         if (new_fs)
2505                 free_fs_struct(new_fs);
2506
2507 bad_unshare_out:
2508         return err;
2509 }
2510
2511 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2512 {
2513         return ksys_unshare(unshare_flags);
2514 }
2515
2516 /*
2517  *      Helper to unshare the files of the current task.
2518  *      We don't want to expose copy_files internals to
2519  *      the exec layer of the kernel.
2520  */
2521
2522 int unshare_files(struct files_struct **displaced)
2523 {
2524         struct task_struct *task = current;
2525         struct files_struct *copy = NULL;
2526         int error;
2527
2528         error = unshare_fd(CLONE_FILES, &copy);
2529         if (error || !copy) {
2530                 *displaced = NULL;
2531                 return error;
2532         }
2533         *displaced = task->files;
2534         task_lock(task);
2535         task->files = copy;
2536         task_unlock(task);
2537         return 0;
2538 }
2539
2540 int sysctl_max_threads(struct ctl_table *table, int write,
2541                        void __user *buffer, size_t *lenp, loff_t *ppos)
2542 {
2543         struct ctl_table t;
2544         int ret;
2545         int threads = max_threads;
2546         int min = MIN_THREADS;
2547         int max = MAX_THREADS;
2548
2549         t = *table;
2550         t.data = &threads;
2551         t.extra1 = &min;
2552         t.extra2 = &max;
2553
2554         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2555         if (ret || !write)
2556                 return ret;
2557
2558         set_max_threads(threads);
2559
2560         return 0;
2561 }