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