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