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