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