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/anon_inodes.h>
  15 #include <linux/slab.h>
  16 #include <linux/sched/autogroup.h>
  17 #include <linux/sched/mm.h>
  18 #include <linux/sched/coredump.h>
  19 #include <linux/sched/user.h>
  20 #include <linux/sched/numa_balancing.h>
  21 #include <linux/sched/stat.h>
  22 #include <linux/sched/task.h>
  23 #include <linux/sched/task_stack.h>
  24 #include <linux/sched/cputime.h>
  25 #include <linux/seq_file.h>
  26 #include <linux/rtmutex.h>
  27 #include <linux/init.h>
  28 #include <linux/unistd.h>
  29 #include <linux/module.h>
  30 #include <linux/vmalloc.h>
  31 #include <linux/completion.h>
  32 #include <linux/personality.h>
  33 #include <linux/mempolicy.h>
  34 #include <linux/sem.h>
  35 #include <linux/file.h>
  36 #include <linux/fdtable.h>
  37 #include <linux/iocontext.h>
  38 #include <linux/key.h>
  39 #include <linux/binfmts.h>
  40 #include <linux/mman.h>
  41 #include <linux/mmu_notifier.h>
  42 #include <linux/hmm.h>
  43 #include <linux/fs.h>
  44 #include <linux/mm.h>
  45 #include <linux/vmacache.h>
  46 #include <linux/nsproxy.h>
  47 #include <linux/capability.h>
  48 #include <linux/cpu.h>
  49 #include <linux/cgroup.h>
  50 #include <linux/security.h>
  51 #include <linux/hugetlb.h>
  52 #include <linux/seccomp.h>
  53 #include <linux/swap.h>
  54 #include <linux/syscalls.h>
  55 #include <linux/jiffies.h>
  56 #include <linux/futex.h>
  57 #include <linux/compat.h>
  58 #include <linux/kthread.h>
  59 #include <linux/task_io_accounting_ops.h>
  60 #include <linux/rcupdate.h>
  61 #include <linux/ptrace.h>
  62 #include <linux/mount.h>
  63 #include <linux/audit.h>
  64 #include <linux/memcontrol.h>
  65 #include <linux/ftrace.h>
  66 #include <linux/proc_fs.h>
  67 #include <linux/profile.h>
  68 #include <linux/rmap.h>
  69 #include <linux/ksm.h>
  70 #include <linux/acct.h>
  71 #include <linux/userfaultfd_k.h>
  72 #include <linux/tsacct_kern.h>
  73 #include <linux/cn_proc.h>
  74 #include <linux/freezer.h>
  75 #include <linux/delayacct.h>
  76 #include <linux/taskstats_kern.h>
  77 #include <linux/random.h>
  78 #include <linux/tty.h>
  79 #include <linux/blkdev.h>
  80 #include <linux/fs_struct.h>
  81 #include <linux/magic.h>
  82 #include <linux/perf_event.h>
  83 #include <linux/posix-timers.h>
  84 #include <linux/user-return-notifier.h>
  85 #include <linux/oom.h>
  86 #include <linux/khugepaged.h>
  87 #include <linux/signalfd.h>
  88 #include <linux/uprobes.h>
  89 #include <linux/aio.h>
  90 #include <linux/compiler.h>
  91 #include <linux/sysctl.h>
  92 #include <linux/kcov.h>
  93 #include <linux/livepatch.h>
  94 #include <linux/thread_info.h>
  95 #include <linux/stackleak.h>
  96
  97 #include <asm/pgtable.h>
  98 #include <asm/pgalloc.h>
  99 #include <linux/uaccess.h>
 100 #include <asm/mmu_context.h>
 101 #include <asm/cacheflush.h>
 102 #include <asm/tlbflush.h>
 103
 104 #include <trace/events/sched.h>
 105
 106 #define CREATE_TRACE_POINTS
 107 #include <trace/events/task.h>
 108
 109 /*
 110  * Minimum number of threads to boot the kernel
 111  */
 112 #define MIN_THREADS 20
 113
 114 /*
 115  * Maximum number of threads
 116  */
 117 #define MAX_THREADS FUTEX_TID_MASK
 118
 119 /*
 120  * Protected counters by write_lock_irq(&tasklist_lock)
 121  */
 122 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 123 int nr_threads;                 /* The idle threads do not count.. */
 124
 125 int max_threads;                /* tunable limit on nr_threads */
 126
 127 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 128
 129 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 130
 131 #ifdef CONFIG_PROVE_RCU
 132 int lockdep_tasklist_lock_is_held(void)
 133 {
 134         return lockdep_is_held(&tasklist_lock);
 135 }
 136 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 137 #endif /* #ifdef CONFIG_PROVE_RCU */
 138
 139 int nr_processes(void)
 140 {
 141         int cpu;
 142         int total = 0;
 143
 144         for_each_possible_cpu(cpu)
 145                 total += per_cpu(process_counts, cpu);
 146
 147         return total;
 148 }
 149
 150 void __weak arch_release_task_struct(struct task_struct *tsk)
 151 {
 152 }
 153
 154 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 155 static struct kmem_cache *task_struct_cachep;
 156
 157 static inline struct task_struct *alloc_task_struct_node(int node)
 158 {
 159         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 160 }
 161
 162 static inline void free_task_struct(struct task_struct *tsk)
 163 {
 164         kmem_cache_free(task_struct_cachep, tsk);
 165 }
 166 #endif
 167
 168 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 169
 170 /*
 171  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 172  * kmemcache based allocator.
 173  */
 174 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 175
 176 #ifdef CONFIG_VMAP_STACK
 177 /*
 178  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 179  * flush.  Try to minimize the number of calls by caching stacks.
 180  */
 181 #define NR_CACHED_STACKS 2
 182 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 183
 184 static int free_vm_stack_cache(unsigned int cpu)
 185 {
 186         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 187         int i;
 188
 189         for (i = 0; i < NR_CACHED_STACKS; i++) {
 190                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 191
 192                 if (!vm_stack)
 193                         continue;
 194
 195                 vfree(vm_stack->addr);
 196                 cached_vm_stacks[i] = NULL;
 197         }
 198
 199         return 0;
 200 }
 201 #endif
 202
 203 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 204 {
 205 #ifdef CONFIG_VMAP_STACK
 206         void *stack;
 207         int i;
 208
 209         for (i = 0; i < NR_CACHED_STACKS; i++) {
 210                 struct vm_struct *s;
 211
 212                 s = this_cpu_xchg(cached_stacks[i], NULL);
 213
 214                 if (!s)
 215                         continue;
 216
 217                 /* Clear stale pointers from reused stack. */
 218                 memset(s->addr, 0, THREAD_SIZE);
 219
 220                 tsk->stack_vm_area = s;
 221                 tsk->stack = s->addr;
 222                 return s->addr;
 223         }
 224
 225         /*
 226          * Allocated stacks are cached and later reused by new threads,
 227          * so memcg accounting is performed manually on assigning/releasing
 228          * stacks to tasks. Drop __GFP_ACCOUNT.
 229          */
 230         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 231                                      VMALLOC_START, VMALLOC_END,
 232                                      THREADINFO_GFP & ~__GFP_ACCOUNT,
 233                                      PAGE_KERNEL,
 234                                      0, node, __builtin_return_address(0));
 235
 236         /*
 237          * We can't call find_vm_area() in interrupt context, and
 238          * free_thread_stack() can be called in interrupt context,
 239          * so cache the vm_struct.
 240          */
 241         if (stack) {
 242                 tsk->stack_vm_area = find_vm_area(stack);
 243                 tsk->stack = stack;
 244         }
 245         return stack;
 246 #else
 247         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 248                                              THREAD_SIZE_ORDER);
 249
 250         return page ? page_address(page) : NULL;
 251 #endif
 252 }
 253
 254 static inline void free_thread_stack(struct task_struct *tsk)
 255 {
 256 #ifdef CONFIG_VMAP_STACK
 257         struct vm_struct *vm = task_stack_vm_area(tsk);
 258
 259         if (vm) {
 260                 int i;
 261
 262                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 263                         mod_memcg_page_state(vm->pages[i],
 264                                              MEMCG_KERNEL_STACK_KB,
 265                                              -(int)(PAGE_SIZE / 1024));
 266
 267                         memcg_kmem_uncharge(vm->pages[i], 0);
 268                 }
 269
 270                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 271                         if (this_cpu_cmpxchg(cached_stacks[i],
 272                                         NULL, tsk->stack_vm_area) != NULL)
 273                                 continue;
 274
 275                         return;
 276                 }
 277
 278                 vfree_atomic(tsk->stack);
 279                 return;
 280         }
 281 #endif
 282
 283         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 284 }
 285 # else
 286 static struct kmem_cache *thread_stack_cache;
 287
 288 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 289                                                   int node)
 290 {
 291         unsigned long *stack;
 292         stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 293         tsk->stack = stack;
 294         return stack;
 295 }
 296
 297 static void free_thread_stack(struct task_struct *tsk)
 298 {
 299         kmem_cache_free(thread_stack_cache, tsk->stack);
 300 }
 301
 302 void thread_stack_cache_init(void)
 303 {
 304         thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 305                                         THREAD_SIZE, THREAD_SIZE, 0, 0,
 306                                         THREAD_SIZE, NULL);
 307         BUG_ON(thread_stack_cache == NULL);
 308 }
 309 # endif
 310 #endif
 311
 312 /* SLAB cache for signal_struct structures (tsk->signal) */
 313 static struct kmem_cache *signal_cachep;
 314
 315 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 316 struct kmem_cache *sighand_cachep;
 317
 318 /* SLAB cache for files_struct structures (tsk->files) */
 319 struct kmem_cache *files_cachep;
 320
 321 /* SLAB cache for fs_struct structures (tsk->fs) */
 322 struct kmem_cache *fs_cachep;
 323
 324 /* SLAB cache for vm_area_struct structures */
 325 static struct kmem_cache *vm_area_cachep;
 326
 327 /* SLAB cache for mm_struct structures (tsk->mm) */
 328 static struct kmem_cache *mm_cachep;
 329
 330 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 331 {
 332         struct vm_area_struct *vma;
 333
 334         vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 335         if (vma)
 336                 vma_init(vma, mm);
 337         return vma;
 338 }
 339
 340 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 341 {
 342         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 343
 344         if (new) {
 345                 *new = *orig;
 346                 INIT_LIST_HEAD(&new->anon_vma_chain);
 347         }
 348         return new;
 349 }
 350
 351 void vm_area_free(struct vm_area_struct *vma)
 352 {
 353         kmem_cache_free(vm_area_cachep, vma);
 354 }
 355
 356 static void account_kernel_stack(struct task_struct *tsk, int account)
 357 {
 358         void *stack = task_stack_page(tsk);
 359         struct vm_struct *vm = task_stack_vm_area(tsk);
 360
 361         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 362
 363         if (vm) {
 364                 int i;
 365
 366                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 367
 368                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 369                         mod_zone_page_state(page_zone(vm->pages[i]),
 370                                             NR_KERNEL_STACK_KB,
 371                                             PAGE_SIZE / 1024 * account);
 372                 }
 373         } else {
 374                 /*
 375                  * All stack pages are in the same zone and belong to the
 376                  * same memcg.
 377                  */
 378                 struct page *first_page = virt_to_page(stack);
 379
 380                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 381                                     THREAD_SIZE / 1024 * account);
 382
 383                 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
 384                                      account * (THREAD_SIZE / 1024));
 385         }
 386 }
 387
 388 static int memcg_charge_kernel_stack(struct task_struct *tsk)
 389 {
 390 #ifdef CONFIG_VMAP_STACK
 391         struct vm_struct *vm = task_stack_vm_area(tsk);
 392         int ret;
 393
 394         if (vm) {
 395                 int i;
 396
 397                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 398                         /*
 399                          * If memcg_kmem_charge() fails, page->mem_cgroup
 400                          * pointer is NULL, and both memcg_kmem_uncharge()
 401                          * and mod_memcg_page_state() in free_thread_stack()
 402                          * will ignore this page. So it's safe.
 403                          */
 404                         ret = memcg_kmem_charge(vm->pages[i], GFP_KERNEL, 0);
 405                         if (ret)
 406                                 return ret;
 407
 408                         mod_memcg_page_state(vm->pages[i],
 409                                              MEMCG_KERNEL_STACK_KB,
 410                                              PAGE_SIZE / 1024);
 411                 }
 412         }
 413 #endif
 414         return 0;
 415 }
 416
 417 static void release_task_stack(struct task_struct *tsk)
 418 {
 419         if (WARN_ON(tsk->state != TASK_DEAD))
 420                 return;  /* Better to leak the stack than to free prematurely */
 421
 422         account_kernel_stack(tsk, -1);
 423         free_thread_stack(tsk);
 424         tsk->stack = NULL;
 425 #ifdef CONFIG_VMAP_STACK
 426         tsk->stack_vm_area = NULL;
 427 #endif
 428 }
 429
 430 #ifdef CONFIG_THREAD_INFO_IN_TASK
 431 void put_task_stack(struct task_struct *tsk)
 432 {
 433         if (refcount_dec_and_test(&tsk->stack_refcount))
 434                 release_task_stack(tsk);
 435 }
 436 #endif
 437
 438 void free_task(struct task_struct *tsk)
 439 {
 440 #ifndef CONFIG_THREAD_INFO_IN_TASK
 441         /*
 442          * The task is finally done with both the stack and thread_info,
 443          * so free both.
 444          */
 445         release_task_stack(tsk);
 446 #else
 447         /*
 448          * If the task had a separate stack allocation, it should be gone
 449          * by now.
 450          */
 451         WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
 452 #endif
 453         rt_mutex_debug_task_free(tsk);
 454         ftrace_graph_exit_task(tsk);
 455         put_seccomp_filter(tsk);
 456         arch_release_task_struct(tsk);
 457         if (tsk->flags & PF_KTHREAD)
 458                 free_kthread_struct(tsk);
 459         free_task_struct(tsk);
 460 }
 461 EXPORT_SYMBOL(free_task);
 462
 463 #ifdef CONFIG_MMU
 464 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 465                                         struct mm_struct *oldmm)
 466 {
 467         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 468         struct rb_node **rb_link, *rb_parent;
 469         int retval;
 470         unsigned long charge;
 471         LIST_HEAD(uf);
 472
 473         uprobe_start_dup_mmap();
 474         if (down_write_killable(&oldmm->mmap_sem)) {
 475                 retval = -EINTR;
 476                 goto fail_uprobe_end;
 477         }
 478         flush_cache_dup_mm(oldmm);
 479         uprobe_dup_mmap(oldmm, mm);
 480         /*
 481          * Not linked in yet - no deadlock potential:
 482          */
 483         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 484
 485         /* No ordering required: file already has been exposed. */
 486         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 487
 488         mm->total_vm = oldmm->total_vm;
 489         mm->data_vm = oldmm->data_vm;
 490         mm->exec_vm = oldmm->exec_vm;
 491         mm->stack_vm = oldmm->stack_vm;
 492
 493         rb_link = &mm->mm_rb.rb_node;
 494         rb_parent = NULL;
 495         pprev = &mm->mmap;
 496         retval = ksm_fork(mm, oldmm);
 497         if (retval)
 498                 goto out;
 499         retval = khugepaged_fork(mm, oldmm);
 500         if (retval)
 501                 goto out;
 502
 503         prev = NULL;
 504         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 505                 struct file *file;
 506
 507                 if (mpnt->vm_flags & VM_DONTCOPY) {
 508                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 509                         continue;
 510                 }
 511                 charge = 0;
 512                 /*
 513                  * Don't duplicate many vmas if we've been oom-killed (for
 514                  * example)
 515                  */
 516                 if (fatal_signal_pending(current)) {
 517                         retval = -EINTR;
 518                         goto out;
 519                 }
 520                 if (mpnt->vm_flags & VM_ACCOUNT) {
 521                         unsigned long len = vma_pages(mpnt);
 522
 523                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 524                                 goto fail_nomem;
 525                         charge = len;
 526                 }
 527                 tmp = vm_area_dup(mpnt);
 528                 if (!tmp)
 529                         goto fail_nomem;
 530                 retval = vma_dup_policy(mpnt, tmp);
 531                 if (retval)
 532                         goto fail_nomem_policy;
 533                 tmp->vm_mm = mm;
 534                 retval = dup_userfaultfd(tmp, &uf);
 535                 if (retval)
 536                         goto fail_nomem_anon_vma_fork;
 537                 if (tmp->vm_flags & VM_WIPEONFORK) {
 538                         /* VM_WIPEONFORK gets a clean slate in the child. */
 539                         tmp->anon_vma = NULL;
 540                         if (anon_vma_prepare(tmp))
 541                                 goto fail_nomem_anon_vma_fork;
 542                 } else if (anon_vma_fork(tmp, mpnt))
 543                         goto fail_nomem_anon_vma_fork;
 544                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 545                 tmp->vm_next = tmp->vm_prev = NULL;
 546                 file = tmp->vm_file;
 547                 if (file) {
 548                         struct inode *inode = file_inode(file);
 549                         struct address_space *mapping = file->f_mapping;
 550
 551                         get_file(file);
 552                         if (tmp->vm_flags & VM_DENYWRITE)
 553                                 atomic_dec(&inode->i_writecount);
 554                         i_mmap_lock_write(mapping);
 555                         if (tmp->vm_flags & VM_SHARED)
 556                                 atomic_inc(&mapping->i_mmap_writable);
 557                         flush_dcache_mmap_lock(mapping);
 558                         /* insert tmp into the share list, just after mpnt */
 559                         vma_interval_tree_insert_after(tmp, mpnt,
 560                                         &mapping->i_mmap);
 561                         flush_dcache_mmap_unlock(mapping);
 562                         i_mmap_unlock_write(mapping);
 563                 }
 564
 565                 /*
 566                  * Clear hugetlb-related page reserves for children. This only
 567                  * affects MAP_PRIVATE mappings. Faults generated by the child
 568                  * are not guaranteed to succeed, even if read-only
 569                  */
 570                 if (is_vm_hugetlb_page(tmp))
 571                         reset_vma_resv_huge_pages(tmp);
 572
 573                 /*
 574                  * Link in the new vma and copy the page table entries.
 575                  */
 576                 *pprev = tmp;
 577                 pprev = &tmp->vm_next;
 578                 tmp->vm_prev = prev;
 579                 prev = tmp;
 580
 581                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 582                 rb_link = &tmp->vm_rb.rb_right;
 583                 rb_parent = &tmp->vm_rb;
 584
 585                 mm->map_count++;
 586                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 587                         retval = copy_page_range(mm, oldmm, mpnt);
 588
 589                 if (tmp->vm_ops && tmp->vm_ops->open)
 590                         tmp->vm_ops->open(tmp);
 591
 592                 if (retval)
 593                         goto out;
 594         }
 595         /* a new mm has just been created */
 596         retval = arch_dup_mmap(oldmm, mm);
 597 out:
 598         up_write(&mm->mmap_sem);
 599         flush_tlb_mm(oldmm);
 600         up_write(&oldmm->mmap_sem);
 601         dup_userfaultfd_complete(&uf);
 602 fail_uprobe_end:
 603         uprobe_end_dup_mmap();
 604         return retval;
 605 fail_nomem_anon_vma_fork:
 606         mpol_put(vma_policy(tmp));
 607 fail_nomem_policy:
 608         vm_area_free(tmp);
 609 fail_nomem:
 610         retval = -ENOMEM;
 611         vm_unacct_memory(charge);
 612         goto out;
 613 }
 614
 615 static inline int mm_alloc_pgd(struct mm_struct *mm)
 616 {
 617         mm->pgd = pgd_alloc(mm);
 618         if (unlikely(!mm->pgd))
 619                 return -ENOMEM;
 620         return 0;
 621 }
 622
 623 static inline void mm_free_pgd(struct mm_struct *mm)
 624 {
 625         pgd_free(mm, mm->pgd);
 626 }
 627 #else
 628 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 629 {
 630         down_write(&oldmm->mmap_sem);
 631         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 632         up_write(&oldmm->mmap_sem);
 633         return 0;
 634 }
 635 #define mm_alloc_pgd(mm)        (0)
 636 #define mm_free_pgd(mm)
 637 #endif /* CONFIG_MMU */
 638
 639 static void check_mm(struct mm_struct *mm)
 640 {
 641         int i;
 642
 643         for (i = 0; i < NR_MM_COUNTERS; i++) {
 644                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 645
 646                 if (unlikely(x))
 647                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 648                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 649         }
 650
 651         if (mm_pgtables_bytes(mm))
 652                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 653                                 mm_pgtables_bytes(mm));
 654
 655 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 656         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 657 #endif
 658 }
 659
 660 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 661 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 662
 663 /*
 664  * Called when the last reference to the mm
 665  * is dropped: either by a lazy thread or by
 666  * mmput. Free the page directory and the mm.
 667  */
 668 void __mmdrop(struct mm_struct *mm)
 669 {
 670         BUG_ON(mm == &init_mm);
 671         WARN_ON_ONCE(mm == current->mm);
 672         WARN_ON_ONCE(mm == current->active_mm);
 673         mm_free_pgd(mm);
 674         destroy_context(mm);
 675         hmm_mm_destroy(mm);
 676         mmu_notifier_mm_destroy(mm);
 677         check_mm(mm);
 678         put_user_ns(mm->user_ns);
 679         free_mm(mm);
 680 }
 681 EXPORT_SYMBOL_GPL(__mmdrop);
 682
 683 static void mmdrop_async_fn(struct work_struct *work)
 684 {
 685         struct mm_struct *mm;
 686
 687         mm = container_of(work, struct mm_struct, async_put_work);
 688         __mmdrop(mm);
 689 }
 690
 691 static void mmdrop_async(struct mm_struct *mm)
 692 {
 693         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 694                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 695                 schedule_work(&mm->async_put_work);
 696         }
 697 }
 698
 699 static inline void free_signal_struct(struct signal_struct *sig)
 700 {
 701         taskstats_tgid_free(sig);
 702         sched_autogroup_exit(sig);
 703         /*
 704          * __mmdrop is not safe to call from softirq context on x86 due to
 705          * pgd_dtor so postpone it to the async context
 706          */
 707         if (sig->oom_mm)
 708                 mmdrop_async(sig->oom_mm);
 709         kmem_cache_free(signal_cachep, sig);
 710 }
 711
 712 static inline void put_signal_struct(struct signal_struct *sig)
 713 {
 714         if (refcount_dec_and_test(&sig->sigcnt))
 715                 free_signal_struct(sig);
 716 }
 717
 718 void __put_task_struct(struct task_struct *tsk)
 719 {
 720         WARN_ON(!tsk->exit_state);
 721         WARN_ON(refcount_read(&tsk->usage));
 722         WARN_ON(tsk == current);
 723
 724         cgroup_free(tsk);
 725         task_numa_free(tsk);
 726         security_task_free(tsk);
 727         exit_creds(tsk);
 728         delayacct_tsk_free(tsk);
 729         put_signal_struct(tsk->signal);
 730
 731         if (!profile_handoff_task(tsk))
 732                 free_task(tsk);
 733 }
 734 EXPORT_SYMBOL_GPL(__put_task_struct);
 735
 736 void __init __weak arch_task_cache_init(void) { }
 737
 738 /*
 739  * set_max_threads
 740  */
 741 static void set_max_threads(unsigned int max_threads_suggested)
 742 {
 743         u64 threads;
 744         unsigned long nr_pages = totalram_pages();
 745
 746         /*
 747          * The number of threads shall be limited such that the thread
 748          * structures may only consume a small part of the available memory.
 749          */
 750         if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
 751                 threads = MAX_THREADS;
 752         else
 753                 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
 754                                     (u64) THREAD_SIZE * 8UL);
 755
 756         if (threads > max_threads_suggested)
 757                 threads = max_threads_suggested;
 758
 759         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 760 }
 761
 762 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 763 /* Initialized by the architecture: */
 764 int arch_task_struct_size __read_mostly;
 765 #endif
 766
 767 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 768 {
 769         /* Fetch thread_struct whitelist for the architecture. */
 770         arch_thread_struct_whitelist(offset, size);
 771
 772         /*
 773          * Handle zero-sized whitelist or empty thread_struct, otherwise
 774          * adjust offset to position of thread_struct in task_struct.
 775          */
 776         if (unlikely(*size == 0))
 777                 *offset = 0;
 778         else
 779                 *offset += offsetof(struct task_struct, thread);
 780 }
 781
 782 void __init fork_init(void)
 783 {
 784         int i;
 785 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 786 #ifndef ARCH_MIN_TASKALIGN
 787 #define ARCH_MIN_TASKALIGN      0
 788 #endif
 789         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 790         unsigned long useroffset, usersize;
 791
 792         /* create a slab on which task_structs can be allocated */
 793         task_struct_whitelist(&useroffset, &usersize);
 794         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 795                         arch_task_struct_size, align,
 796                         SLAB_PANIC|SLAB_ACCOUNT,
 797                         useroffset, usersize, NULL);
 798 #endif
 799
 800         /* do the arch specific task caches init */
 801         arch_task_cache_init();
 802
 803         set_max_threads(MAX_THREADS);
 804
 805         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 806         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 807         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 808                 init_task.signal->rlim[RLIMIT_NPROC];
 809
 810         for (i = 0; i < UCOUNT_COUNTS; i++) {
 811                 init_user_ns.ucount_max[i] = max_threads/2;
 812         }
 813
 814 #ifdef CONFIG_VMAP_STACK
 815         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 816                           NULL, free_vm_stack_cache);
 817 #endif
 818
 819         lockdep_init_task(&init_task);
 820 }
 821
 822 int __weak arch_dup_task_struct(struct task_struct *dst,
 823                                                struct task_struct *src)
 824 {
 825         *dst = *src;
 826         return 0;
 827 }
 828
 829 void set_task_stack_end_magic(struct task_struct *tsk)
 830 {
 831         unsigned long *stackend;
 832
 833         stackend = end_of_stack(tsk);
 834         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 835 }
 836
 837 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 838 {
 839         struct task_struct *tsk;
 840         unsigned long *stack;
 841         struct vm_struct *stack_vm_area __maybe_unused;
 842         int err;
 843
 844         if (node == NUMA_NO_NODE)
 845                 node = tsk_fork_get_node(orig);
 846         tsk = alloc_task_struct_node(node);
 847         if (!tsk)
 848                 return NULL;
 849
 850         stack = alloc_thread_stack_node(tsk, node);
 851         if (!stack)
 852                 goto free_tsk;
 853
 854         if (memcg_charge_kernel_stack(tsk))
 855                 goto free_stack;
 856
 857         stack_vm_area = task_stack_vm_area(tsk);
 858
 859         err = arch_dup_task_struct(tsk, orig);
 860
 861         /*
 862          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 863          * sure they're properly initialized before using any stack-related
 864          * functions again.
 865          */
 866         tsk->stack = stack;
 867 #ifdef CONFIG_VMAP_STACK
 868         tsk->stack_vm_area = stack_vm_area;
 869 #endif
 870 #ifdef CONFIG_THREAD_INFO_IN_TASK
 871         refcount_set(&tsk->stack_refcount, 1);
 872 #endif
 873
 874         if (err)
 875                 goto free_stack;
 876
 877 #ifdef CONFIG_SECCOMP
 878         /*
 879          * We must handle setting up seccomp filters once we're under
 880          * the sighand lock in case orig has changed between now and
 881          * then. Until then, filter must be NULL to avoid messing up
 882          * the usage counts on the error path calling free_task.
 883          */
 884         tsk->seccomp.filter = NULL;
 885 #endif
 886
 887         setup_thread_stack(tsk, orig);
 888         clear_user_return_notifier(tsk);
 889         clear_tsk_need_resched(tsk);
 890         set_task_stack_end_magic(tsk);
 891
 892 #ifdef CONFIG_STACKPROTECTOR
 893         tsk->stack_canary = get_random_canary();
 894 #endif
 895
 896         /*
 897          * One for us, one for whoever does the "release_task()" (usually
 898          * parent)
 899          */
 900         refcount_set(&tsk->usage, 2);
 901 #ifdef CONFIG_BLK_DEV_IO_TRACE
 902         tsk->btrace_seq = 0;
 903 #endif
 904         tsk->splice_pipe = NULL;
 905         tsk->task_frag.page = NULL;
 906         tsk->wake_q.next = NULL;
 907
 908         account_kernel_stack(tsk, 1);
 909
 910         kcov_task_init(tsk);
 911
 912 #ifdef CONFIG_FAULT_INJECTION
 913         tsk->fail_nth = 0;
 914 #endif
 915
 916 #ifdef CONFIG_BLK_CGROUP
 917         tsk->throttle_queue = NULL;
 918         tsk->use_memdelay = 0;
 919 #endif
 920
 921 #ifdef CONFIG_MEMCG
 922         tsk->active_memcg = NULL;
 923 #endif
 924         return tsk;
 925
 926 free_stack:
 927         free_thread_stack(tsk);
 928 free_tsk:
 929         free_task_struct(tsk);
 930         return NULL;
 931 }
 932
 933 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 934
 935 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 936
 937 static int __init coredump_filter_setup(char *s)
 938 {
 939         default_dump_filter =
 940                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 941                 MMF_DUMP_FILTER_MASK;
 942         return 1;
 943 }
 944
 945 __setup("coredump_filter=", coredump_filter_setup);
 946
 947 #include <linux/init_task.h>
 948
 949 static void mm_init_aio(struct mm_struct *mm)
 950 {
 951 #ifdef CONFIG_AIO
 952         spin_lock_init(&mm->ioctx_lock);
 953         mm->ioctx_table = NULL;
 954 #endif
 955 }
 956
 957 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 958 {
 959 #ifdef CONFIG_MEMCG
 960         mm->owner = p;
 961 #endif
 962 }
 963
 964 static void mm_init_uprobes_state(struct mm_struct *mm)
 965 {
 966 #ifdef CONFIG_UPROBES
 967         mm->uprobes_state.xol_area = NULL;
 968 #endif
 969 }
 970
 971 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 972         struct user_namespace *user_ns)
 973 {
 974         mm->mmap = NULL;
 975         mm->mm_rb = RB_ROOT;
 976         mm->vmacache_seqnum = 0;
 977         atomic_set(&mm->mm_users, 1);
 978         atomic_set(&mm->mm_count, 1);
 979         init_rwsem(&mm->mmap_sem);
 980         INIT_LIST_HEAD(&mm->mmlist);
 981         mm->core_state = NULL;
 982         mm_pgtables_bytes_init(mm);
 983         mm->map_count = 0;
 984         mm->locked_vm = 0;
 985         atomic64_set(&mm->pinned_vm, 0);
 986         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 987         spin_lock_init(&mm->page_table_lock);
 988         spin_lock_init(&mm->arg_lock);
 989         mm_init_cpumask(mm);
 990         mm_init_aio(mm);
 991         mm_init_owner(mm, p);
 992         RCU_INIT_POINTER(mm->exe_file, NULL);
 993         mmu_notifier_mm_init(mm);
 994         hmm_mm_init(mm);
 995         init_tlb_flush_pending(mm);
 996 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 997         mm->pmd_huge_pte = NULL;
 998 #endif
 999         mm_init_uprobes_state(mm);
1000
1001         if (current->mm) {
1002                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1003                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1004         } else {
1005                 mm->flags = default_dump_filter;
1006                 mm->def_flags = 0;
1007         }
1008
1009         if (mm_alloc_pgd(mm))
1010                 goto fail_nopgd;
1011
1012         if (init_new_context(p, mm))
1013                 goto fail_nocontext;
1014
1015         mm->user_ns = get_user_ns(user_ns);
1016         return mm;
1017
1018 fail_nocontext:
1019         mm_free_pgd(mm);
1020 fail_nopgd:
1021         free_mm(mm);
1022         return NULL;
1023 }
1024
1025 /*
1026  * Allocate and initialize an mm_struct.
1027  */
1028 struct mm_struct *mm_alloc(void)
1029 {
1030         struct mm_struct *mm;
1031
1032         mm = allocate_mm();
1033         if (!mm)
1034                 return NULL;
1035
1036         memset(mm, 0, sizeof(*mm));
1037         return mm_init(mm, current, current_user_ns());
1038 }
1039
1040 static inline void __mmput(struct mm_struct *mm)
1041 {
1042         VM_BUG_ON(atomic_read(&mm->mm_users));
1043
1044         uprobe_clear_state(mm);
1045         exit_aio(mm);
1046         ksm_exit(mm);
1047         khugepaged_exit(mm); /* must run before exit_mmap */
1048         exit_mmap(mm);
1049         mm_put_huge_zero_page(mm);
1050         set_mm_exe_file(mm, NULL);
1051         if (!list_empty(&mm->mmlist)) {
1052                 spin_lock(&mmlist_lock);
1053                 list_del(&mm->mmlist);
1054                 spin_unlock(&mmlist_lock);
1055         }
1056         if (mm->binfmt)
1057                 module_put(mm->binfmt->module);
1058         mmdrop(mm);
1059 }
1060
1061 /*
1062  * Decrement the use count and release all resources for an mm.
1063  */
1064 void mmput(struct mm_struct *mm)
1065 {
1066         might_sleep();
1067
1068         if (atomic_dec_and_test(&mm->mm_users))
1069                 __mmput(mm);
1070 }
1071 EXPORT_SYMBOL_GPL(mmput);
1072
1073 #ifdef CONFIG_MMU
1074 static void mmput_async_fn(struct work_struct *work)
1075 {
1076         struct mm_struct *mm = container_of(work, struct mm_struct,
1077                                             async_put_work);
1078
1079         __mmput(mm);
1080 }
1081
1082 void mmput_async(struct mm_struct *mm)
1083 {
1084         if (atomic_dec_and_test(&mm->mm_users)) {
1085                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1086                 schedule_work(&mm->async_put_work);
1087         }
1088 }
1089 #endif
1090
1091 /**
1092  * set_mm_exe_file - change a reference to the mm's executable file
1093  *
1094  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1095  *
1096  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1097  * invocations: in mmput() nobody alive left, in execve task is single
1098  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1099  * mm->exe_file, but does so without using set_mm_exe_file() in order
1100  * to do avoid the need for any locks.
1101  */
1102 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1103 {
1104         struct file *old_exe_file;
1105
1106         /*
1107          * It is safe to dereference the exe_file without RCU as
1108          * this function is only called if nobody else can access
1109          * this mm -- see comment above for justification.
1110          */
1111         old_exe_file = rcu_dereference_raw(mm->exe_file);
1112
1113         if (new_exe_file)
1114                 get_file(new_exe_file);
1115         rcu_assign_pointer(mm->exe_file, new_exe_file);
1116         if (old_exe_file)
1117                 fput(old_exe_file);
1118 }
1119
1120 /**
1121  * get_mm_exe_file - acquire a reference to the mm's executable file
1122  *
1123  * Returns %NULL if mm has no associated executable file.
1124  * User must release file via fput().
1125  */
1126 struct file *get_mm_exe_file(struct mm_struct *mm)
1127 {
1128         struct file *exe_file;
1129
1130         rcu_read_lock();
1131         exe_file = rcu_dereference(mm->exe_file);
1132         if (exe_file && !get_file_rcu(exe_file))
1133                 exe_file = NULL;
1134         rcu_read_unlock();
1135         return exe_file;
1136 }
1137 EXPORT_SYMBOL(get_mm_exe_file);
1138
1139 /**
1140  * get_task_exe_file - acquire a reference to the task's executable file
1141  *
1142  * Returns %NULL if task's mm (if any) has no associated executable file or
1143  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1144  * User must release file via fput().
1145  */
1146 struct file *get_task_exe_file(struct task_struct *task)
1147 {
1148         struct file *exe_file = NULL;
1149         struct mm_struct *mm;
1150
1151         task_lock(task);
1152         mm = task->mm;
1153         if (mm) {
1154                 if (!(task->flags & PF_KTHREAD))
1155                         exe_file = get_mm_exe_file(mm);
1156         }
1157         task_unlock(task);
1158         return exe_file;
1159 }
1160 EXPORT_SYMBOL(get_task_exe_file);
1161
1162 /**
1163  * get_task_mm - acquire a reference to the task's mm
1164  *
1165  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1166  * this kernel workthread has transiently adopted a user mm with use_mm,
1167  * to do its AIO) is not set and if so returns a reference to it, after
1168  * bumping up the use count.  User must release the mm via mmput()
1169  * after use.  Typically used by /proc and ptrace.
1170  */
1171 struct mm_struct *get_task_mm(struct task_struct *task)
1172 {
1173         struct mm_struct *mm;
1174
1175         task_lock(task);
1176         mm = task->mm;
1177         if (mm) {
1178                 if (task->flags & PF_KTHREAD)
1179                         mm = NULL;
1180                 else
1181                         mmget(mm);
1182         }
1183         task_unlock(task);
1184         return mm;
1185 }
1186 EXPORT_SYMBOL_GPL(get_task_mm);
1187
1188 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1189 {
1190         struct mm_struct *mm;
1191         int err;
1192
1193         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1194         if (err)
1195                 return ERR_PTR(err);
1196
1197         mm = get_task_mm(task);
1198         if (mm && mm != current->mm &&
1199                         !ptrace_may_access(task, mode)) {
1200                 mmput(mm);
1201                 mm = ERR_PTR(-EACCES);
1202         }
1203         mutex_unlock(&task->signal->cred_guard_mutex);
1204
1205         return mm;
1206 }
1207
1208 static void complete_vfork_done(struct task_struct *tsk)
1209 {
1210         struct completion *vfork;
1211
1212         task_lock(tsk);
1213         vfork = tsk->vfork_done;
1214         if (likely(vfork)) {
1215                 tsk->vfork_done = NULL;
1216                 complete(vfork);
1217         }
1218         task_unlock(tsk);
1219 }
1220
1221 static int wait_for_vfork_done(struct task_struct *child,
1222                                 struct completion *vfork)
1223 {
1224         int killed;
1225
1226         freezer_do_not_count();
1227         killed = wait_for_completion_killable(vfork);
1228         freezer_count();
1229
1230         if (killed) {
1231                 task_lock(child);
1232                 child->vfork_done = NULL;
1233                 task_unlock(child);
1234         }
1235
1236         put_task_struct(child);
1237         return killed;
1238 }
1239
1240 /* Please note the differences between mmput and mm_release.
1241  * mmput is called whenever we stop holding onto a mm_struct,
1242  * error success whatever.
1243  *
1244  * mm_release is called after a mm_struct has been removed
1245  * from the current process.
1246  *
1247  * This difference is important for error handling, when we
1248  * only half set up a mm_struct for a new process and need to restore
1249  * the old one.  Because we mmput the new mm_struct before
1250  * restoring the old one. . .
1251  * Eric Biederman 10 January 1998
1252  */
1253 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1254 {
1255         /* Get rid of any futexes when releasing the mm */
1256 #ifdef CONFIG_FUTEX
1257         if (unlikely(tsk->robust_list)) {
1258                 exit_robust_list(tsk);
1259                 tsk->robust_list = NULL;
1260         }
1261 #ifdef CONFIG_COMPAT
1262         if (unlikely(tsk->compat_robust_list)) {
1263                 compat_exit_robust_list(tsk);
1264                 tsk->compat_robust_list = NULL;
1265         }
1266 #endif
1267         if (unlikely(!list_empty(&tsk->pi_state_list)))
1268                 exit_pi_state_list(tsk);
1269 #endif
1270
1271         uprobe_free_utask(tsk);
1272
1273         /* Get rid of any cached register state */
1274         deactivate_mm(tsk, mm);
1275
1276         /*
1277          * Signal userspace if we're not exiting with a core dump
1278          * because we want to leave the value intact for debugging
1279          * purposes.
1280          */
1281         if (tsk->clear_child_tid) {
1282                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1283                     atomic_read(&mm->mm_users) > 1) {
1284                         /*
1285                          * We don't check the error code - if userspace has
1286                          * not set up a proper pointer then tough luck.
1287                          */
1288                         put_user(0, tsk->clear_child_tid);
1289                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1290                                         1, NULL, NULL, 0, 0);
1291                 }
1292                 tsk->clear_child_tid = NULL;
1293         }
1294
1295         /*
1296          * All done, finally we can wake up parent and return this mm to him.
1297          * Also kthread_stop() uses this completion for synchronization.
1298          */
1299         if (tsk->vfork_done)
1300                 complete_vfork_done(tsk);
1301 }
1302
1303 /*
1304  * Allocate a new mm structure and copy contents from the
1305  * mm structure of the passed in task structure.
1306  */
1307 static struct mm_struct *dup_mm(struct task_struct *tsk)
1308 {
1309         struct mm_struct *mm, *oldmm = current->mm;
1310         int err;
1311
1312         mm = allocate_mm();
1313         if (!mm)
1314                 goto fail_nomem;
1315
1316         memcpy(mm, oldmm, sizeof(*mm));
1317
1318         if (!mm_init(mm, tsk, mm->user_ns))
1319                 goto fail_nomem;
1320
1321         err = dup_mmap(mm, oldmm);
1322         if (err)
1323                 goto free_pt;
1324
1325         mm->hiwater_rss = get_mm_rss(mm);
1326         mm->hiwater_vm = mm->total_vm;
1327
1328         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1329                 goto free_pt;
1330
1331         return mm;
1332
1333 free_pt:
1334         /* don't put binfmt in mmput, we haven't got module yet */
1335         mm->binfmt = NULL;
1336         mmput(mm);
1337
1338 fail_nomem:
1339         return NULL;
1340 }
1341
1342 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1343 {
1344         struct mm_struct *mm, *oldmm;
1345         int retval;
1346
1347         tsk->min_flt = tsk->maj_flt = 0;
1348         tsk->nvcsw = tsk->nivcsw = 0;
1349 #ifdef CONFIG_DETECT_HUNG_TASK
1350         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1351         tsk->last_switch_time = 0;
1352 #endif
1353
1354         tsk->mm = NULL;
1355         tsk->active_mm = NULL;
1356
1357         /*
1358          * Are we cloning a kernel thread?
1359          *
1360          * We need to steal a active VM for that..
1361          */
1362         oldmm = current->mm;
1363         if (!oldmm)
1364                 return 0;
1365
1366         /* initialize the new vmacache entries */
1367         vmacache_flush(tsk);
1368
1369         if (clone_flags & CLONE_VM) {
1370                 mmget(oldmm);
1371                 mm = oldmm;
1372                 goto good_mm;
1373         }
1374
1375         retval = -ENOMEM;
1376         mm = dup_mm(tsk);
1377         if (!mm)
1378                 goto fail_nomem;
1379
1380 good_mm:
1381         tsk->mm = mm;
1382         tsk->active_mm = mm;
1383         return 0;
1384
1385 fail_nomem:
1386         return retval;
1387 }
1388
1389 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1390 {
1391         struct fs_struct *fs = current->fs;
1392         if (clone_flags & CLONE_FS) {
1393                 /* tsk->fs is already what we want */
1394                 spin_lock(&fs->lock);
1395                 if (fs->in_exec) {
1396                         spin_unlock(&fs->lock);
1397                         return -EAGAIN;
1398                 }
1399                 fs->users++;
1400                 spin_unlock(&fs->lock);
1401                 return 0;
1402         }
1403         tsk->fs = copy_fs_struct(fs);
1404         if (!tsk->fs)
1405                 return -ENOMEM;
1406         return 0;
1407 }
1408
1409 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1410 {
1411         struct files_struct *oldf, *newf;
1412         int error = 0;
1413
1414         /*
1415          * A background process may not have any files ...
1416          */
1417         oldf = current->files;
1418         if (!oldf)
1419                 goto out;
1420
1421         if (clone_flags & CLONE_FILES) {
1422                 atomic_inc(&oldf->count);
1423                 goto out;
1424         }
1425
1426         newf = dup_fd(oldf, &error);
1427         if (!newf)
1428                 goto out;
1429
1430         tsk->files = newf;
1431         error = 0;
1432 out:
1433         return error;
1434 }
1435
1436 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1437 {
1438 #ifdef CONFIG_BLOCK
1439         struct io_context *ioc = current->io_context;
1440         struct io_context *new_ioc;
1441
1442         if (!ioc)
1443                 return 0;
1444         /*
1445          * Share io context with parent, if CLONE_IO is set
1446          */
1447         if (clone_flags & CLONE_IO) {
1448                 ioc_task_link(ioc);
1449                 tsk->io_context = ioc;
1450         } else if (ioprio_valid(ioc->ioprio)) {
1451                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1452                 if (unlikely(!new_ioc))
1453                         return -ENOMEM;
1454
1455                 new_ioc->ioprio = ioc->ioprio;
1456                 put_io_context(new_ioc);
1457         }
1458 #endif
1459         return 0;
1460 }
1461
1462 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1463 {
1464         struct sighand_struct *sig;
1465
1466         if (clone_flags & CLONE_SIGHAND) {
1467                 refcount_inc(&current->sighand->count);
1468                 return 0;
1469         }
1470         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1471         rcu_assign_pointer(tsk->sighand, sig);
1472         if (!sig)
1473                 return -ENOMEM;
1474
1475         refcount_set(&sig->count, 1);
1476         spin_lock_irq(&current->sighand->siglock);
1477         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1478         spin_unlock_irq(&current->sighand->siglock);
1479         return 0;
1480 }
1481
1482 void __cleanup_sighand(struct sighand_struct *sighand)
1483 {
1484         if (refcount_dec_and_test(&sighand->count)) {
1485                 signalfd_cleanup(sighand);
1486                 /*
1487                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1488                  * without an RCU grace period, see __lock_task_sighand().
1489                  */
1490                 kmem_cache_free(sighand_cachep, sighand);
1491         }
1492 }
1493
1494 #ifdef CONFIG_POSIX_TIMERS
1495 /*
1496  * Initialize POSIX timer handling for a thread group.
1497  */
1498 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1499 {
1500         unsigned long cpu_limit;
1501
1502         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1503         if (cpu_limit != RLIM_INFINITY) {
1504                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1505                 sig->cputimer.running = true;
1506         }
1507
1508         /* The timer lists. */
1509         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1510         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1511         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1512 }
1513 #else
1514 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1515 #endif
1516
1517 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1518 {
1519         struct signal_struct *sig;
1520
1521         if (clone_flags & CLONE_THREAD)
1522                 return 0;
1523
1524         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1525         tsk->signal = sig;
1526         if (!sig)
1527                 return -ENOMEM;
1528
1529         sig->nr_threads = 1;
1530         atomic_set(&sig->live, 1);
1531         refcount_set(&sig->sigcnt, 1);
1532
1533         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1534         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1535         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1536
1537         init_waitqueue_head(&sig->wait_chldexit);
1538         sig->curr_target = tsk;
1539         init_sigpending(&sig->shared_pending);
1540         INIT_HLIST_HEAD(&sig->multiprocess);
1541         seqlock_init(&sig->stats_lock);
1542         prev_cputime_init(&sig->prev_cputime);
1543
1544 #ifdef CONFIG_POSIX_TIMERS
1545         INIT_LIST_HEAD(&sig->posix_timers);
1546         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1547         sig->real_timer.function = it_real_fn;
1548 #endif
1549
1550         task_lock(current->group_leader);
1551         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1552         task_unlock(current->group_leader);
1553
1554         posix_cpu_timers_init_group(sig);
1555
1556         tty_audit_fork(sig);
1557         sched_autogroup_fork(sig);
1558
1559         sig->oom_score_adj = current->signal->oom_score_adj;
1560         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1561
1562         mutex_init(&sig->cred_guard_mutex);
1563
1564         return 0;
1565 }
1566
1567 static void copy_seccomp(struct task_struct *p)
1568 {
1569 #ifdef CONFIG_SECCOMP
1570         /*
1571          * Must be called with sighand->lock held, which is common to
1572          * all threads in the group. Holding cred_guard_mutex is not
1573          * needed because this new task is not yet running and cannot
1574          * be racing exec.
1575          */
1576         assert_spin_locked(&current->sighand->siglock);
1577
1578         /* Ref-count the new filter user, and assign it. */
1579         get_seccomp_filter(current);
1580         p->seccomp = current->seccomp;
1581
1582         /*
1583          * Explicitly enable no_new_privs here in case it got set
1584          * between the task_struct being duplicated and holding the
1585          * sighand lock. The seccomp state and nnp must be in sync.
1586          */
1587         if (task_no_new_privs(current))
1588                 task_set_no_new_privs(p);
1589
1590         /*
1591          * If the parent gained a seccomp mode after copying thread
1592          * flags and between before we held the sighand lock, we have
1593          * to manually enable the seccomp thread flag here.
1594          */
1595         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1596                 set_tsk_thread_flag(p, TIF_SECCOMP);
1597 #endif
1598 }
1599
1600 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1601 {
1602         current->clear_child_tid = tidptr;
1603
1604         return task_pid_vnr(current);
1605 }
1606
1607 static void rt_mutex_init_task(struct task_struct *p)
1608 {
1609         raw_spin_lock_init(&p->pi_lock);
1610 #ifdef CONFIG_RT_MUTEXES
1611         p->pi_waiters = RB_ROOT_CACHED;
1612         p->pi_top_task = NULL;
1613         p->pi_blocked_on = NULL;
1614 #endif
1615 }
1616
1617 #ifdef CONFIG_POSIX_TIMERS
1618 /*
1619  * Initialize POSIX timer handling for a single task.
1620  */
1621 static void posix_cpu_timers_init(struct task_struct *tsk)
1622 {
1623         tsk->cputime_expires.prof_exp = 0;
1624         tsk->cputime_expires.virt_exp = 0;
1625         tsk->cputime_expires.sched_exp = 0;
1626         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1627         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1628         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1629 }
1630 #else
1631 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1632 #endif
1633
1634 static inline void init_task_pid_links(struct task_struct *task)
1635 {
1636         enum pid_type type;
1637
1638         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1639                 INIT_HLIST_NODE(&task->pid_links[type]);
1640         }
1641 }
1642
1643 static inline void
1644 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1645 {
1646         if (type == PIDTYPE_PID)
1647                 task->thread_pid = pid;
1648         else
1649                 task->signal->pids[type] = pid;
1650 }
1651
1652 static inline void rcu_copy_process(struct task_struct *p)
1653 {
1654 #ifdef CONFIG_PREEMPT_RCU
1655         p->rcu_read_lock_nesting = 0;
1656         p->rcu_read_unlock_special.s = 0;
1657         p->rcu_blocked_node = NULL;
1658         INIT_LIST_HEAD(&p->rcu_node_entry);
1659 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1660 #ifdef CONFIG_TASKS_RCU
1661         p->rcu_tasks_holdout = false;
1662         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1663         p->rcu_tasks_idle_cpu = -1;
1664 #endif /* #ifdef CONFIG_TASKS_RCU */
1665 }
1666
1667 static int pidfd_release(struct inode *inode, struct file *file)
1668 {
1669         struct pid *pid = file->private_data;
1670
1671         file->private_data = NULL;
1672         put_pid(pid);
1673         return 0;
1674 }
1675
1676 #ifdef CONFIG_PROC_FS
1677 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1678 {
1679         struct pid_namespace *ns = proc_pid_ns(file_inode(m->file));
1680         struct pid *pid = f->private_data;
1681
1682         seq_put_decimal_ull(m, "Pid:\t", pid_nr_ns(pid, ns));
1683         seq_putc(m, '\n');
1684 }
1685 #endif
1686
1687 const struct file_operations pidfd_fops = {
1688         .release = pidfd_release,
1689 #ifdef CONFIG_PROC_FS
1690         .show_fdinfo = pidfd_show_fdinfo,
1691 #endif
1692 };
1693
1694 /**
1695  * pidfd_create() - Create a new pid file descriptor.
1696  *
1697  * @pid:  struct pid that the pidfd will reference
1698  *
1699  * This creates a new pid file descriptor with the O_CLOEXEC flag set.
1700  *
1701  * Note, that this function can only be called after the fd table has
1702  * been unshared to avoid leaking the pidfd to the new process.
1703  *
1704  * Return: On success, a cloexec pidfd is returned.
1705  *         On error, a negative errno number will be returned.
1706  */
1707 static int pidfd_create(struct pid *pid)
1708 {
1709         int fd;
1710
1711         fd = anon_inode_getfd("[pidfd]", &pidfd_fops, get_pid(pid),
1712                               O_RDWR | O_CLOEXEC);
1713         if (fd < 0)
1714                 put_pid(pid);
1715
1716         return fd;
1717 }
1718
1719 /*
1720  * This creates a new process as a copy of the old one,
1721  * but does not actually start it yet.
1722  *
1723  * It copies the registers, and all the appropriate
1724  * parts of the process environment (as per the clone
1725  * flags). The actual kick-off is left to the caller.
1726  */
1727 static __latent_entropy struct task_struct *copy_process(
1728                                         unsigned long clone_flags,
1729                                         unsigned long stack_start,
1730                                         unsigned long stack_size,
1731                                         int __user *parent_tidptr,
1732                                         int __user *child_tidptr,
1733                                         struct pid *pid,
1734                                         int trace,
1735                                         unsigned long tls,
1736                                         int node)
1737 {
1738         int pidfd = -1, retval;
1739         struct task_struct *p;
1740         struct multiprocess_signals delayed;
1741
1742         /*
1743          * Don't allow sharing the root directory with processes in a different
1744          * namespace
1745          */
1746         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1747                 return ERR_PTR(-EINVAL);
1748
1749         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1750                 return ERR_PTR(-EINVAL);
1751
1752         /*
1753          * Thread groups must share signals as well, and detached threads
1754          * can only be started up within the thread group.
1755          */
1756         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1757                 return ERR_PTR(-EINVAL);
1758
1759         /*
1760          * Shared signal handlers imply shared VM. By way of the above,
1761          * thread groups also imply shared VM. Blocking this case allows
1762          * for various simplifications in other code.
1763          */
1764         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1765                 return ERR_PTR(-EINVAL);
1766
1767         /*
1768          * Siblings of global init remain as zombies on exit since they are
1769          * not reaped by their parent (swapper). To solve this and to avoid
1770          * multi-rooted process trees, prevent global and container-inits
1771          * from creating siblings.
1772          */
1773         if ((clone_flags & CLONE_PARENT) &&
1774                                 current->signal->flags & SIGNAL_UNKILLABLE)
1775                 return ERR_PTR(-EINVAL);
1776
1777         /*
1778          * If the new process will be in a different pid or user namespace
1779          * do not allow it to share a thread group with the forking task.
1780          */
1781         if (clone_flags & CLONE_THREAD) {
1782                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1783                     (task_active_pid_ns(current) !=
1784                                 current->nsproxy->pid_ns_for_children))
1785                         return ERR_PTR(-EINVAL);
1786         }
1787
1788         if (clone_flags & CLONE_PIDFD) {
1789                 int reserved;
1790
1791                 /*
1792                  * - CLONE_PARENT_SETTID is useless for pidfds and also
1793                  *   parent_tidptr is used to return pidfds.
1794                  * - CLONE_DETACHED is blocked so that we can potentially
1795                  *   reuse it later for CLONE_PIDFD.
1796                  * - CLONE_THREAD is blocked until someone really needs it.
1797                  */
1798                 if (clone_flags &
1799                     (CLONE_DETACHED | CLONE_PARENT_SETTID | CLONE_THREAD))
1800                         return ERR_PTR(-EINVAL);
1801
1802                 /*
1803                  * Verify that parent_tidptr is sane so we can potentially
1804                  * reuse it later.
1805                  */
1806                 if (get_user(reserved, parent_tidptr))
1807                         return ERR_PTR(-EFAULT);
1808
1809                 if (reserved != 0)
1810                         return ERR_PTR(-EINVAL);
1811         }
1812
1813         /*
1814          * Force any signals received before this point to be delivered
1815          * before the fork happens.  Collect up signals sent to multiple
1816          * processes that happen during the fork and delay them so that
1817          * they appear to happen after the fork.
1818          */
1819         sigemptyset(&delayed.signal);
1820         INIT_HLIST_NODE(&delayed.node);
1821
1822         spin_lock_irq(&current->sighand->siglock);
1823         if (!(clone_flags & CLONE_THREAD))
1824                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1825         recalc_sigpending();
1826         spin_unlock_irq(&current->sighand->siglock);
1827         retval = -ERESTARTNOINTR;
1828         if (signal_pending(current))
1829                 goto fork_out;
1830
1831         retval = -ENOMEM;
1832         p = dup_task_struct(current, node);
1833         if (!p)
1834                 goto fork_out;
1835
1836         /*
1837          * This _must_ happen before we call free_task(), i.e. before we jump
1838          * to any of the bad_fork_* labels. This is to avoid freeing
1839          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1840          * kernel threads (PF_KTHREAD).
1841          */
1842         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1843         /*
1844          * Clear TID on mm_release()?
1845          */
1846         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1847
1848         ftrace_graph_init_task(p);
1849
1850         rt_mutex_init_task(p);
1851
1852 #ifdef CONFIG_PROVE_LOCKING
1853         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1854         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1855 #endif
1856         retval = -EAGAIN;
1857         if (atomic_read(&p->real_cred->user->processes) >=
1858                         task_rlimit(p, RLIMIT_NPROC)) {
1859                 if (p->real_cred->user != INIT_USER &&
1860                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1861                         goto bad_fork_free;
1862         }
1863         current->flags &= ~PF_NPROC_EXCEEDED;
1864
1865         retval = copy_creds(p, clone_flags);
1866         if (retval < 0)
1867                 goto bad_fork_free;
1868
1869         /*
1870          * If multiple threads are within copy_process(), then this check
1871          * triggers too late. This doesn't hurt, the check is only there
1872          * to stop root fork bombs.
1873          */
1874         retval = -EAGAIN;
1875         if (nr_threads >= max_threads)
1876                 goto bad_fork_cleanup_count;
1877
1878         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1879         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1880         p->flags |= PF_FORKNOEXEC;
1881         INIT_LIST_HEAD(&p->children);
1882         INIT_LIST_HEAD(&p->sibling);
1883         rcu_copy_process(p);
1884         p->vfork_done = NULL;
1885         spin_lock_init(&p->alloc_lock);
1886
1887         init_sigpending(&p->pending);
1888
1889         p->utime = p->stime = p->gtime = 0;
1890 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1891         p->utimescaled = p->stimescaled = 0;
1892 #endif
1893         prev_cputime_init(&p->prev_cputime);
1894
1895 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1896         seqcount_init(&p->vtime.seqcount);
1897         p->vtime.starttime = 0;
1898         p->vtime.state = VTIME_INACTIVE;
1899 #endif
1900
1901 #if defined(SPLIT_RSS_COUNTING)
1902         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1903 #endif
1904
1905         p->default_timer_slack_ns = current->timer_slack_ns;
1906
1907 #ifdef CONFIG_PSI
1908         p->psi_flags = 0;
1909 #endif
1910
1911         task_io_accounting_init(&p->ioac);
1912         acct_clear_integrals(p);
1913
1914         posix_cpu_timers_init(p);
1915
1916         p->io_context = NULL;
1917         audit_set_context(p, NULL);
1918         cgroup_fork(p);
1919 #ifdef CONFIG_NUMA
1920         p->mempolicy = mpol_dup(p->mempolicy);
1921         if (IS_ERR(p->mempolicy)) {
1922                 retval = PTR_ERR(p->mempolicy);
1923                 p->mempolicy = NULL;
1924                 goto bad_fork_cleanup_threadgroup_lock;
1925         }
1926 #endif
1927 #ifdef CONFIG_CPUSETS
1928         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1929         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1930         seqcount_init(&p->mems_allowed_seq);
1931 #endif
1932 #ifdef CONFIG_TRACE_IRQFLAGS
1933         p->irq_events = 0;
1934         p->hardirqs_enabled = 0;
1935         p->hardirq_enable_ip = 0;
1936         p->hardirq_enable_event = 0;
1937         p->hardirq_disable_ip = _THIS_IP_;
1938         p->hardirq_disable_event = 0;
1939         p->softirqs_enabled = 1;
1940         p->softirq_enable_ip = _THIS_IP_;
1941         p->softirq_enable_event = 0;
1942         p->softirq_disable_ip = 0;
1943         p->softirq_disable_event = 0;
1944         p->hardirq_context = 0;
1945         p->softirq_context = 0;
1946 #endif
1947
1948         p->pagefault_disabled = 0;
1949
1950 #ifdef CONFIG_LOCKDEP
1951         p->lockdep_depth = 0; /* no locks held yet */
1952         p->curr_chain_key = 0;
1953         p->lockdep_recursion = 0;
1954         lockdep_init_task(p);
1955 #endif
1956
1957 #ifdef CONFIG_DEBUG_MUTEXES
1958         p->blocked_on = NULL; /* not blocked yet */
1959 #endif
1960 #ifdef CONFIG_BCACHE
1961         p->sequential_io        = 0;
1962         p->sequential_io_avg    = 0;
1963 #endif
1964
1965         /* Perform scheduler related setup. Assign this task to a CPU. */
1966         retval = sched_fork(clone_flags, p);
1967         if (retval)
1968                 goto bad_fork_cleanup_policy;
1969
1970         retval = perf_event_init_task(p);
1971         if (retval)
1972                 goto bad_fork_cleanup_policy;
1973         retval = audit_alloc(p);
1974         if (retval)
1975                 goto bad_fork_cleanup_perf;
1976         /* copy all the process information */
1977         shm_init_task(p);
1978         retval = security_task_alloc(p, clone_flags);
1979         if (retval)
1980                 goto bad_fork_cleanup_audit;
1981         retval = copy_semundo(clone_flags, p);
1982         if (retval)
1983                 goto bad_fork_cleanup_security;
1984         retval = copy_files(clone_flags, p);
1985         if (retval)
1986                 goto bad_fork_cleanup_semundo;
1987         retval = copy_fs(clone_flags, p);
1988         if (retval)
1989                 goto bad_fork_cleanup_files;
1990         retval = copy_sighand(clone_flags, p);
1991         if (retval)
1992                 goto bad_fork_cleanup_fs;
1993         retval = copy_signal(clone_flags, p);
1994         if (retval)
1995                 goto bad_fork_cleanup_sighand;
1996         retval = copy_mm(clone_flags, p);
1997         if (retval)
1998                 goto bad_fork_cleanup_signal;
1999         retval = copy_namespaces(clone_flags, p);
2000         if (retval)
2001                 goto bad_fork_cleanup_mm;
2002         retval = copy_io(clone_flags, p);
2003         if (retval)
2004                 goto bad_fork_cleanup_namespaces;
2005         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
2006         if (retval)
2007                 goto bad_fork_cleanup_io;
2008
2009         stackleak_task_init(p);
2010
2011         if (pid != &init_struct_pid) {
2012                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
2013                 if (IS_ERR(pid)) {
2014                         retval = PTR_ERR(pid);
2015                         goto bad_fork_cleanup_thread;
2016                 }
2017         }
2018
2019         /*
2020          * This has to happen after we've potentially unshared the file
2021          * descriptor table (so that the pidfd doesn't leak into the child
2022          * if the fd table isn't shared).
2023          */
2024         if (clone_flags & CLONE_PIDFD) {
2025                 retval = pidfd_create(pid);
2026                 if (retval < 0)
2027                         goto bad_fork_free_pid;
2028
2029                 pidfd = retval;
2030                 retval = put_user(pidfd, parent_tidptr);
2031                 if (retval)
2032                         goto bad_fork_put_pidfd;
2033         }
2034
2035 #ifdef CONFIG_BLOCK
2036         p->plug = NULL;
2037 #endif
2038 #ifdef CONFIG_FUTEX
2039         p->robust_list = NULL;
2040 #ifdef CONFIG_COMPAT
2041         p->compat_robust_list = NULL;
2042 #endif
2043         INIT_LIST_HEAD(&p->pi_state_list);
2044         p->pi_state_cache = NULL;
2045 #endif
2046         /*
2047          * sigaltstack should be cleared when sharing the same VM
2048          */
2049         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2050                 sas_ss_reset(p);
2051
2052         /*
2053          * Syscall tracing and stepping should be turned off in the
2054          * child regardless of CLONE_PTRACE.
2055          */
2056         user_disable_single_step(p);
2057         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
2058 #ifdef TIF_SYSCALL_EMU
2059         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
2060 #endif
2061         clear_all_latency_tracing(p);
2062
2063         /* ok, now we should be set up.. */
2064         p->pid = pid_nr(pid);
2065         if (clone_flags & CLONE_THREAD) {
2066                 p->exit_signal = -1;
2067                 p->group_leader = current->group_leader;
2068                 p->tgid = current->tgid;
2069         } else {
2070                 if (clone_flags & CLONE_PARENT)
2071                         p->exit_signal = current->group_leader->exit_signal;
2072                 else
2073                         p->exit_signal = (clone_flags & CSIGNAL);
2074                 p->group_leader = p;
2075                 p->tgid = p->pid;
2076         }
2077
2078         p->nr_dirtied = 0;
2079         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2080         p->dirty_paused_when = 0;
2081
2082         p->pdeath_signal = 0;
2083         INIT_LIST_HEAD(&p->thread_group);
2084         p->task_works = NULL;
2085
2086         cgroup_threadgroup_change_begin(current);
2087         /*
2088          * Ensure that the cgroup subsystem policies allow the new process to be
2089          * forked. It should be noted the the new process's css_set can be changed
2090          * between here and cgroup_post_fork() if an organisation operation is in
2091          * progress.
2092          */
2093         retval = cgroup_can_fork(p);
2094         if (retval)
2095                 goto bad_fork_put_pidfd;
2096
2097         /*
2098          * From this point on we must avoid any synchronous user-space
2099          * communication until we take the tasklist-lock. In particular, we do
2100          * not want user-space to be able to predict the process start-time by
2101          * stalling fork(2) after we recorded the start_time but before it is
2102          * visible to the system.
2103          */
2104
2105         p->start_time = ktime_get_ns();
2106         p->real_start_time = ktime_get_boot_ns();
2107
2108         /*
2109          * Make it visible to the rest of the system, but dont wake it up yet.
2110          * Need tasklist lock for parent etc handling!
2111          */
2112         write_lock_irq(&tasklist_lock);
2113
2114         /* CLONE_PARENT re-uses the old parent */
2115         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2116                 p->real_parent = current->real_parent;
2117                 p->parent_exec_id = current->parent_exec_id;
2118         } else {
2119                 p->real_parent = current;
2120                 p->parent_exec_id = current->self_exec_id;
2121         }
2122
2123         klp_copy_process(p);
2124
2125         spin_lock(&current->sighand->siglock);
2126
2127         /*
2128          * Copy seccomp details explicitly here, in case they were changed
2129          * before holding sighand lock.
2130          */
2131         copy_seccomp(p);
2132
2133         rseq_fork(p, clone_flags);
2134
2135         /* Don't start children in a dying pid namespace */
2136         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2137                 retval = -ENOMEM;
2138                 goto bad_fork_cancel_cgroup;
2139         }
2140
2141         /* Let kill terminate clone/fork in the middle */
2142         if (fatal_signal_pending(current)) {
2143                 retval = -EINTR;
2144                 goto bad_fork_cancel_cgroup;
2145         }
2146
2147
2148         init_task_pid_links(p);
2149         if (likely(p->pid)) {
2150                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2151
2152                 init_task_pid(p, PIDTYPE_PID, pid);
2153                 if (thread_group_leader(p)) {
2154                         init_task_pid(p, PIDTYPE_TGID, pid);
2155                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2156                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2157
2158                         if (is_child_reaper(pid)) {
2159                                 ns_of_pid(pid)->child_reaper = p;
2160                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2161                         }
2162                         p->signal->shared_pending.signal = delayed.signal;
2163                         p->signal->tty = tty_kref_get(current->signal->tty);
2164                         /*
2165                          * Inherit has_child_subreaper flag under the same
2166                          * tasklist_lock with adding child to the process tree
2167                          * for propagate_has_child_subreaper optimization.
2168                          */
2169                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2170                                                          p->real_parent->signal->is_child_subreaper;
2171                         list_add_tail(&p->sibling, &p->real_parent->children);
2172                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2173                         attach_pid(p, PIDTYPE_TGID);
2174                         attach_pid(p, PIDTYPE_PGID);
2175                         attach_pid(p, PIDTYPE_SID);
2176                         __this_cpu_inc(process_counts);
2177                 } else {
2178                         current->signal->nr_threads++;
2179                         atomic_inc(&current->signal->live);
2180                         refcount_inc(&current->signal->sigcnt);
2181                         task_join_group_stop(p);
2182                         list_add_tail_rcu(&p->thread_group,
2183                                           &p->group_leader->thread_group);
2184                         list_add_tail_rcu(&p->thread_node,
2185                                           &p->signal->thread_head);
2186                 }
2187                 attach_pid(p, PIDTYPE_PID);
2188                 nr_threads++;
2189         }
2190         total_forks++;
2191         hlist_del_init(&delayed.node);
2192         spin_unlock(&current->sighand->siglock);
2193         syscall_tracepoint_update(p);
2194         write_unlock_irq(&tasklist_lock);
2195
2196         proc_fork_connector(p);
2197         cgroup_post_fork(p);
2198         cgroup_threadgroup_change_end(current);
2199         perf_event_fork(p);
2200
2201         trace_task_newtask(p, clone_flags);
2202         uprobe_copy_process(p, clone_flags);
2203
2204         return p;
2205
2206 bad_fork_cancel_cgroup:
2207         spin_unlock(&current->sighand->siglock);
2208         write_unlock_irq(&tasklist_lock);
2209         cgroup_cancel_fork(p);
2210 bad_fork_put_pidfd:
2211         if (clone_flags & CLONE_PIDFD)
2212                 ksys_close(pidfd);
2213 bad_fork_free_pid:
2214         cgroup_threadgroup_change_end(current);
2215         if (pid != &init_struct_pid)
2216                 free_pid(pid);
2217 bad_fork_cleanup_thread:
2218         exit_thread(p);
2219 bad_fork_cleanup_io:
2220         if (p->io_context)
2221                 exit_io_context(p);
2222 bad_fork_cleanup_namespaces:
2223         exit_task_namespaces(p);
2224 bad_fork_cleanup_mm:
2225         if (p->mm)
2226                 mmput(p->mm);
2227 bad_fork_cleanup_signal:
2228         if (!(clone_flags & CLONE_THREAD))
2229                 free_signal_struct(p->signal);
2230 bad_fork_cleanup_sighand:
2231         __cleanup_sighand(p->sighand);
2232 bad_fork_cleanup_fs:
2233         exit_fs(p); /* blocking */
2234 bad_fork_cleanup_files:
2235         exit_files(p); /* blocking */
2236 bad_fork_cleanup_semundo:
2237         exit_sem(p);
2238 bad_fork_cleanup_security:
2239         security_task_free(p);
2240 bad_fork_cleanup_audit:
2241         audit_free(p);
2242 bad_fork_cleanup_perf:
2243         perf_event_free_task(p);
2244 bad_fork_cleanup_policy:
2245         lockdep_free_task(p);
2246 #ifdef CONFIG_NUMA
2247         mpol_put(p->mempolicy);
2248 bad_fork_cleanup_threadgroup_lock:
2249 #endif
2250         delayacct_tsk_free(p);
2251 bad_fork_cleanup_count:
2252         atomic_dec(&p->cred->user->processes);
2253         exit_creds(p);
2254 bad_fork_free:
2255         p->state = TASK_DEAD;
2256         put_task_stack(p);
2257         free_task(p);
2258 fork_out:
2259         spin_lock_irq(&current->sighand->siglock);
2260         hlist_del_init(&delayed.node);
2261         spin_unlock_irq(&current->sighand->siglock);
2262         return ERR_PTR(retval);
2263 }
2264
2265 static inline void init_idle_pids(struct task_struct *idle)
2266 {
2267         enum pid_type type;
2268
2269         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2270                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2271                 init_task_pid(idle, type, &init_struct_pid);
2272         }
2273 }
2274
2275 struct task_struct *fork_idle(int cpu)
2276 {
2277         struct task_struct *task;
2278         task = copy_process(CLONE_VM, 0, 0, NULL, NULL, &init_struct_pid, 0, 0,
2279                             cpu_to_node(cpu));
2280         if (!IS_ERR(task)) {
2281                 init_idle_pids(task);
2282                 init_idle(task, cpu);
2283         }
2284
2285         return task;
2286 }
2287
2288 /*
2289  *  Ok, this is the main fork-routine.
2290  *
2291  * It copies the process, and if successful kick-starts
2292  * it and waits for it to finish using the VM if required.
2293  */
2294 long _do_fork(unsigned long clone_flags,
2295               unsigned long stack_start,
2296               unsigned long stack_size,
2297               int __user *parent_tidptr,
2298               int __user *child_tidptr,
2299               unsigned long tls)
2300 {
2301         struct completion vfork;
2302         struct pid *pid;
2303         struct task_struct *p;
2304         int trace = 0;
2305         long nr;
2306
2307         /*
2308          * Determine whether and which event to report to ptracer.  When
2309          * called from kernel_thread or CLONE_UNTRACED is explicitly
2310          * requested, no event is reported; otherwise, report if the event
2311          * for the type of forking is enabled.
2312          */
2313         if (!(clone_flags & CLONE_UNTRACED)) {
2314                 if (clone_flags & CLONE_VFORK)
2315                         trace = PTRACE_EVENT_VFORK;
2316                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2317                         trace = PTRACE_EVENT_CLONE;
2318                 else
2319                         trace = PTRACE_EVENT_FORK;
2320
2321                 if (likely(!ptrace_event_enabled(current, trace)))
2322                         trace = 0;
2323         }
2324
2325         p = copy_process(clone_flags, stack_start, stack_size, parent_tidptr,
2326                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2327         add_latent_entropy();
2328
2329         if (IS_ERR(p))
2330                 return PTR_ERR(p);
2331
2332         /*
2333          * Do this prior waking up the new thread - the thread pointer
2334          * might get invalid after that point, if the thread exits quickly.
2335          */
2336         trace_sched_process_fork(current, p);
2337
2338         pid = get_task_pid(p, PIDTYPE_PID);
2339         nr = pid_vnr(pid);
2340
2341         if (clone_flags & CLONE_PARENT_SETTID)
2342                 put_user(nr, parent_tidptr);
2343
2344         if (clone_flags & CLONE_VFORK) {
2345                 p->vfork_done = &vfork;
2346                 init_completion(&vfork);
2347                 get_task_struct(p);
2348         }
2349
2350         wake_up_new_task(p);
2351
2352         /* forking complete and child started to run, tell ptracer */
2353         if (unlikely(trace))
2354                 ptrace_event_pid(trace, pid);
2355
2356         if (clone_flags & CLONE_VFORK) {
2357                 if (!wait_for_vfork_done(p, &vfork))
2358                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2359         }
2360
2361         put_pid(pid);
2362         return nr;
2363 }
2364
2365 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2366 /* For compatibility with architectures that call do_fork directly rather than
2367  * using the syscall entry points below. */
2368 long do_fork(unsigned long clone_flags,
2369               unsigned long stack_start,
2370               unsigned long stack_size,
2371               int __user *parent_tidptr,
2372               int __user *child_tidptr)
2373 {
2374         return _do_fork(clone_flags, stack_start, stack_size,
2375                         parent_tidptr, child_tidptr, 0);
2376 }
2377 #endif
2378
2379 /*
2380  * Create a kernel thread.
2381  */
2382 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2383 {
2384         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2385                 (unsigned long)arg, NULL, NULL, 0);
2386 }
2387
2388 #ifdef __ARCH_WANT_SYS_FORK
2389 SYSCALL_DEFINE0(fork)
2390 {
2391 #ifdef CONFIG_MMU
2392         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2393 #else
2394         /* can not support in nommu mode */
2395         return -EINVAL;
2396 #endif
2397 }
2398 #endif
2399
2400 #ifdef __ARCH_WANT_SYS_VFORK
2401 SYSCALL_DEFINE0(vfork)
2402 {
2403         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2404                         0, NULL, NULL, 0);
2405 }
2406 #endif
2407
2408 #ifdef __ARCH_WANT_SYS_CLONE
2409 #ifdef CONFIG_CLONE_BACKWARDS
2410 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2411                  int __user *, parent_tidptr,
2412                  unsigned long, tls,
2413                  int __user *, child_tidptr)
2414 #elif defined(CONFIG_CLONE_BACKWARDS2)
2415 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2416                  int __user *, parent_tidptr,
2417                  int __user *, child_tidptr,
2418                  unsigned long, tls)
2419 #elif defined(CONFIG_CLONE_BACKWARDS3)
2420 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2421                 int, stack_size,
2422                 int __user *, parent_tidptr,
2423                 int __user *, child_tidptr,
2424                 unsigned long, tls)
2425 #else
2426 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2427                  int __user *, parent_tidptr,
2428                  int __user *, child_tidptr,
2429                  unsigned long, tls)
2430 #endif
2431 {
2432         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2433 }
2434 #endif
2435
2436 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2437 {
2438         struct task_struct *leader, *parent, *child;
2439         int res;
2440
2441         read_lock(&tasklist_lock);
2442         leader = top = top->group_leader;
2443 down:
2444         for_each_thread(leader, parent) {
2445                 list_for_each_entry(child, &parent->children, sibling) {
2446                         res = visitor(child, data);
2447                         if (res) {
2448                                 if (res < 0)
2449                                         goto out;
2450                                 leader = child;
2451                                 goto down;
2452                         }
2453 up:
2454                         ;
2455                 }
2456         }
2457
2458         if (leader != top) {
2459                 child = leader;
2460                 parent = child->real_parent;
2461                 leader = parent->group_leader;
2462                 goto up;
2463         }
2464 out:
2465         read_unlock(&tasklist_lock);
2466 }
2467
2468 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2469 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2470 #endif
2471
2472 static void sighand_ctor(void *data)
2473 {
2474         struct sighand_struct *sighand = data;
2475
2476         spin_lock_init(&sighand->siglock);
2477         init_waitqueue_head(&sighand->signalfd_wqh);
2478 }
2479
2480 void __init proc_caches_init(void)
2481 {
2482         unsigned int mm_size;
2483
2484         sighand_cachep = kmem_cache_create("sighand_cache",
2485                         sizeof(struct sighand_struct), 0,
2486                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2487                         SLAB_ACCOUNT, sighand_ctor);
2488         signal_cachep = kmem_cache_create("signal_cache",
2489                         sizeof(struct signal_struct), 0,
2490                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2491                         NULL);
2492         files_cachep = kmem_cache_create("files_cache",
2493                         sizeof(struct files_struct), 0,
2494                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2495                         NULL);
2496         fs_cachep = kmem_cache_create("fs_cache",
2497                         sizeof(struct fs_struct), 0,
2498                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2499                         NULL);
2500
2501         /*
2502          * The mm_cpumask is located at the end of mm_struct, and is
2503          * dynamically sized based on the maximum CPU number this system
2504          * can have, taking hotplug into account (nr_cpu_ids).
2505          */
2506         mm_size = sizeof(struct mm_struct) + cpumask_size();
2507
2508         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2509                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2510                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2511                         offsetof(struct mm_struct, saved_auxv),
2512                         sizeof_field(struct mm_struct, saved_auxv),
2513                         NULL);
2514         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2515         mmap_init();
2516         nsproxy_cache_init();
2517 }
2518
2519 /*
2520  * Check constraints on flags passed to the unshare system call.
2521  */
2522 static int check_unshare_flags(unsigned long unshare_flags)
2523 {
2524         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2525                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2526                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2527                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2528                 return -EINVAL;
2529         /*
2530          * Not implemented, but pretend it works if there is nothing
2531          * to unshare.  Note that unsharing the address space or the
2532          * signal handlers also need to unshare the signal queues (aka
2533          * CLONE_THREAD).
2534          */
2535         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2536                 if (!thread_group_empty(current))
2537                         return -EINVAL;
2538         }
2539         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2540                 if (refcount_read(&current->sighand->count) > 1)
2541                         return -EINVAL;
2542         }
2543         if (unshare_flags & CLONE_VM) {
2544                 if (!current_is_single_threaded())
2545                         return -EINVAL;
2546         }
2547
2548         return 0;
2549 }
2550
2551 /*
2552  * Unshare the filesystem structure if it is being shared
2553  */
2554 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2555 {
2556         struct fs_struct *fs = current->fs;
2557
2558         if (!(unshare_flags & CLONE_FS) || !fs)
2559                 return 0;
2560
2561         /* don't need lock here; in the worst case we'll do useless copy */
2562         if (fs->users == 1)
2563                 return 0;
2564
2565         *new_fsp = copy_fs_struct(fs);
2566         if (!*new_fsp)
2567                 return -ENOMEM;
2568
2569         return 0;
2570 }
2571
2572 /*
2573  * Unshare file descriptor table if it is being shared
2574  */
2575 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2576 {
2577         struct files_struct *fd = current->files;
2578         int error = 0;
2579
2580         if ((unshare_flags & CLONE_FILES) &&
2581             (fd && atomic_read(&fd->count) > 1)) {
2582                 *new_fdp = dup_fd(fd, &error);
2583                 if (!*new_fdp)
2584                         return error;
2585         }
2586
2587         return 0;
2588 }
2589
2590 /*
2591  * unshare allows a process to 'unshare' part of the process
2592  * context which was originally shared using clone.  copy_*
2593  * functions used by do_fork() cannot be used here directly
2594  * because they modify an inactive task_struct that is being
2595  * constructed. Here we are modifying the current, active,
2596  * task_struct.
2597  */
2598 int ksys_unshare(unsigned long unshare_flags)
2599 {
2600         struct fs_struct *fs, *new_fs = NULL;
2601         struct files_struct *fd, *new_fd = NULL;
2602         struct cred *new_cred = NULL;
2603         struct nsproxy *new_nsproxy = NULL;
2604         int do_sysvsem = 0;
2605         int err;
2606
2607         /*
2608          * If unsharing a user namespace must also unshare the thread group
2609          * and unshare the filesystem root and working directories.
2610          */
2611         if (unshare_flags & CLONE_NEWUSER)
2612                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2613         /*
2614          * If unsharing vm, must also unshare signal handlers.
2615          */
2616         if (unshare_flags & CLONE_VM)
2617                 unshare_flags |= CLONE_SIGHAND;
2618         /*
2619          * If unsharing a signal handlers, must also unshare the signal queues.
2620          */
2621         if (unshare_flags & CLONE_SIGHAND)
2622                 unshare_flags |= CLONE_THREAD;
2623         /*
2624          * If unsharing namespace, must also unshare filesystem information.
2625          */
2626         if (unshare_flags & CLONE_NEWNS)
2627                 unshare_flags |= CLONE_FS;
2628
2629         err = check_unshare_flags(unshare_flags);
2630         if (err)
2631                 goto bad_unshare_out;
2632         /*
2633          * CLONE_NEWIPC must also detach from the undolist: after switching
2634          * to a new ipc namespace, the semaphore arrays from the old
2635          * namespace are unreachable.
2636          */
2637         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2638                 do_sysvsem = 1;
2639         err = unshare_fs(unshare_flags, &new_fs);
2640         if (err)
2641                 goto bad_unshare_out;
2642         err = unshare_fd(unshare_flags, &new_fd);
2643         if (err)
2644                 goto bad_unshare_cleanup_fs;
2645         err = unshare_userns(unshare_flags, &new_cred);
2646         if (err)
2647                 goto bad_unshare_cleanup_fd;
2648         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2649                                          new_cred, new_fs);
2650         if (err)
2651                 goto bad_unshare_cleanup_cred;
2652
2653         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2654                 if (do_sysvsem) {
2655                         /*
2656                          * CLONE_SYSVSEM is equivalent to sys_exit().
2657                          */
2658                         exit_sem(current);
2659                 }
2660                 if (unshare_flags & CLONE_NEWIPC) {
2661                         /* Orphan segments in old ns (see sem above). */
2662                         exit_shm(current);
2663                         shm_init_task(current);
2664                 }
2665
2666                 if (new_nsproxy)
2667                         switch_task_namespaces(current, new_nsproxy);
2668
2669                 task_lock(current);
2670
2671                 if (new_fs) {
2672                         fs = current->fs;
2673                         spin_lock(&fs->lock);
2674                         current->fs = new_fs;
2675                         if (--fs->users)
2676                                 new_fs = NULL;
2677                         else
2678                                 new_fs = fs;
2679                         spin_unlock(&fs->lock);
2680                 }
2681
2682                 if (new_fd) {
2683                         fd = current->files;
2684                         current->files = new_fd;
2685                         new_fd = fd;
2686                 }
2687
2688                 task_unlock(current);
2689
2690                 if (new_cred) {
2691                         /* Install the new user namespace */
2692                         commit_creds(new_cred);
2693                         new_cred = NULL;
2694                 }
2695         }
2696
2697         perf_event_namespaces(current);
2698
2699 bad_unshare_cleanup_cred:
2700         if (new_cred)
2701                 put_cred(new_cred);
2702 bad_unshare_cleanup_fd:
2703         if (new_fd)
2704                 put_files_struct(new_fd);
2705
2706 bad_unshare_cleanup_fs:
2707         if (new_fs)
2708                 free_fs_struct(new_fs);
2709
2710 bad_unshare_out:
2711         return err;
2712 }
2713
2714 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2715 {
2716         return ksys_unshare(unshare_flags);
2717 }
2718
2719 /*
2720  *      Helper to unshare the files of the current task.
2721  *      We don't want to expose copy_files internals to
2722  *      the exec layer of the kernel.
2723  */
2724
2725 int unshare_files(struct files_struct **displaced)
2726 {
2727         struct task_struct *task = current;
2728         struct files_struct *copy = NULL;
2729         int error;
2730
2731         error = unshare_fd(CLONE_FILES, &copy);
2732         if (error || !copy) {
2733                 *displaced = NULL;
2734                 return error;
2735         }
2736         *displaced = task->files;
2737         task_lock(task);
2738         task->files = copy;
2739         task_unlock(task);
2740         return 0;
2741 }
2742
2743 int sysctl_max_threads(struct ctl_table *table, int write,
2744                        void __user *buffer, size_t *lenp, loff_t *ppos)
2745 {
2746         struct ctl_table t;
2747         int ret;
2748         int threads = max_threads;
2749         int min = MIN_THREADS;
2750         int max = MAX_THREADS;
2751
2752         t = *table;
2753         t.data = &threads;
2754         t.extra1 = &min;
2755         t.extra2 = &max;
2756
2757         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2758         if (ret || !write)
2759                 return ret;
2760
2761         set_max_threads(threads);
2762
2763         return 0;
2764 }