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         uprobes_init();
 821 }
 822
 823 int __weak arch_dup_task_struct(struct task_struct *dst,
 824                                                struct task_struct *src)
 825 {
 826         *dst = *src;
 827         return 0;
 828 }
 829
 830 void set_task_stack_end_magic(struct task_struct *tsk)
 831 {
 832         unsigned long *stackend;
 833
 834         stackend = end_of_stack(tsk);
 835         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 836 }
 837
 838 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 839 {
 840         struct task_struct *tsk;
 841         unsigned long *stack;
 842         struct vm_struct *stack_vm_area __maybe_unused;
 843         int err;
 844
 845         if (node == NUMA_NO_NODE)
 846                 node = tsk_fork_get_node(orig);
 847         tsk = alloc_task_struct_node(node);
 848         if (!tsk)
 849                 return NULL;
 850
 851         stack = alloc_thread_stack_node(tsk, node);
 852         if (!stack)
 853                 goto free_tsk;
 854
 855         if (memcg_charge_kernel_stack(tsk))
 856                 goto free_stack;
 857
 858         stack_vm_area = task_stack_vm_area(tsk);
 859
 860         err = arch_dup_task_struct(tsk, orig);
 861
 862         /*
 863          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 864          * sure they're properly initialized before using any stack-related
 865          * functions again.
 866          */
 867         tsk->stack = stack;
 868 #ifdef CONFIG_VMAP_STACK
 869         tsk->stack_vm_area = stack_vm_area;
 870 #endif
 871 #ifdef CONFIG_THREAD_INFO_IN_TASK
 872         refcount_set(&tsk->stack_refcount, 1);
 873 #endif
 874
 875         if (err)
 876                 goto free_stack;
 877
 878 #ifdef CONFIG_SECCOMP
 879         /*
 880          * We must handle setting up seccomp filters once we're under
 881          * the sighand lock in case orig has changed between now and
 882          * then. Until then, filter must be NULL to avoid messing up
 883          * the usage counts on the error path calling free_task.
 884          */
 885         tsk->seccomp.filter = NULL;
 886 #endif
 887
 888         setup_thread_stack(tsk, orig);
 889         clear_user_return_notifier(tsk);
 890         clear_tsk_need_resched(tsk);
 891         set_task_stack_end_magic(tsk);
 892
 893 #ifdef CONFIG_STACKPROTECTOR
 894         tsk->stack_canary = get_random_canary();
 895 #endif
 896
 897         /*
 898          * One for us, one for whoever does the "release_task()" (usually
 899          * parent)
 900          */
 901         refcount_set(&tsk->usage, 2);
 902 #ifdef CONFIG_BLK_DEV_IO_TRACE
 903         tsk->btrace_seq = 0;
 904 #endif
 905         tsk->splice_pipe = NULL;
 906         tsk->task_frag.page = NULL;
 907         tsk->wake_q.next = NULL;
 908
 909         account_kernel_stack(tsk, 1);
 910
 911         kcov_task_init(tsk);
 912
 913 #ifdef CONFIG_FAULT_INJECTION
 914         tsk->fail_nth = 0;
 915 #endif
 916
 917 #ifdef CONFIG_BLK_CGROUP
 918         tsk->throttle_queue = NULL;
 919         tsk->use_memdelay = 0;
 920 #endif
 921
 922 #ifdef CONFIG_MEMCG
 923         tsk->active_memcg = NULL;
 924 #endif
 925         return tsk;
 926
 927 free_stack:
 928         free_thread_stack(tsk);
 929 free_tsk:
 930         free_task_struct(tsk);
 931         return NULL;
 932 }
 933
 934 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 935
 936 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 937
 938 static int __init coredump_filter_setup(char *s)
 939 {
 940         default_dump_filter =
 941                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 942                 MMF_DUMP_FILTER_MASK;
 943         return 1;
 944 }
 945
 946 __setup("coredump_filter=", coredump_filter_setup);
 947
 948 #include <linux/init_task.h>
 949
 950 static void mm_init_aio(struct mm_struct *mm)
 951 {
 952 #ifdef CONFIG_AIO
 953         spin_lock_init(&mm->ioctx_lock);
 954         mm->ioctx_table = NULL;
 955 #endif
 956 }
 957
 958 static __always_inline void mm_clear_owner(struct mm_struct *mm,
 959                                            struct task_struct *p)
 960 {
 961 #ifdef CONFIG_MEMCG
 962         if (mm->owner == p)
 963                 WRITE_ONCE(mm->owner, NULL);
 964 #endif
 965 }
 966
 967 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 968 {
 969 #ifdef CONFIG_MEMCG
 970         mm->owner = p;
 971 #endif
 972 }
 973
 974 static void mm_init_uprobes_state(struct mm_struct *mm)
 975 {
 976 #ifdef CONFIG_UPROBES
 977         mm->uprobes_state.xol_area = NULL;
 978 #endif
 979 }
 980
 981 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 982         struct user_namespace *user_ns)
 983 {
 984         mm->mmap = NULL;
 985         mm->mm_rb = RB_ROOT;
 986         mm->vmacache_seqnum = 0;
 987         atomic_set(&mm->mm_users, 1);
 988         atomic_set(&mm->mm_count, 1);
 989         init_rwsem(&mm->mmap_sem);
 990         INIT_LIST_HEAD(&mm->mmlist);
 991         mm->core_state = NULL;
 992         mm_pgtables_bytes_init(mm);
 993         mm->map_count = 0;
 994         mm->locked_vm = 0;
 995         atomic64_set(&mm->pinned_vm, 0);
 996         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 997         spin_lock_init(&mm->page_table_lock);
 998         spin_lock_init(&mm->arg_lock);
 999         mm_init_cpumask(mm);
1000         mm_init_aio(mm);
1001         mm_init_owner(mm, p);
1002         RCU_INIT_POINTER(mm->exe_file, NULL);
1003         mmu_notifier_mm_init(mm);
1004         hmm_mm_init(mm);
1005         init_tlb_flush_pending(mm);
1006 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1007         mm->pmd_huge_pte = NULL;
1008 #endif
1009         mm_init_uprobes_state(mm);
1010
1011         if (current->mm) {
1012                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1013                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1014         } else {
1015                 mm->flags = default_dump_filter;
1016                 mm->def_flags = 0;
1017         }
1018
1019         if (mm_alloc_pgd(mm))
1020                 goto fail_nopgd;
1021
1022         if (init_new_context(p, mm))
1023                 goto fail_nocontext;
1024
1025         mm->user_ns = get_user_ns(user_ns);
1026         return mm;
1027
1028 fail_nocontext:
1029         mm_free_pgd(mm);
1030 fail_nopgd:
1031         free_mm(mm);
1032         return NULL;
1033 }
1034
1035 /*
1036  * Allocate and initialize an mm_struct.
1037  */
1038 struct mm_struct *mm_alloc(void)
1039 {
1040         struct mm_struct *mm;
1041
1042         mm = allocate_mm();
1043         if (!mm)
1044                 return NULL;
1045
1046         memset(mm, 0, sizeof(*mm));
1047         return mm_init(mm, current, current_user_ns());
1048 }
1049
1050 static inline void __mmput(struct mm_struct *mm)
1051 {
1052         VM_BUG_ON(atomic_read(&mm->mm_users));
1053
1054         uprobe_clear_state(mm);
1055         exit_aio(mm);
1056         ksm_exit(mm);
1057         khugepaged_exit(mm); /* must run before exit_mmap */
1058         exit_mmap(mm);
1059         mm_put_huge_zero_page(mm);
1060         set_mm_exe_file(mm, NULL);
1061         if (!list_empty(&mm->mmlist)) {
1062                 spin_lock(&mmlist_lock);
1063                 list_del(&mm->mmlist);
1064                 spin_unlock(&mmlist_lock);
1065         }
1066         if (mm->binfmt)
1067                 module_put(mm->binfmt->module);
1068         mmdrop(mm);
1069 }
1070
1071 /*
1072  * Decrement the use count and release all resources for an mm.
1073  */
1074 void mmput(struct mm_struct *mm)
1075 {
1076         might_sleep();
1077
1078         if (atomic_dec_and_test(&mm->mm_users))
1079                 __mmput(mm);
1080 }
1081 EXPORT_SYMBOL_GPL(mmput);
1082
1083 #ifdef CONFIG_MMU
1084 static void mmput_async_fn(struct work_struct *work)
1085 {
1086         struct mm_struct *mm = container_of(work, struct mm_struct,
1087                                             async_put_work);
1088
1089         __mmput(mm);
1090 }
1091
1092 void mmput_async(struct mm_struct *mm)
1093 {
1094         if (atomic_dec_and_test(&mm->mm_users)) {
1095                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1096                 schedule_work(&mm->async_put_work);
1097         }
1098 }
1099 #endif
1100
1101 /**
1102  * set_mm_exe_file - change a reference to the mm's executable file
1103  *
1104  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1105  *
1106  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1107  * invocations: in mmput() nobody alive left, in execve task is single
1108  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1109  * mm->exe_file, but does so without using set_mm_exe_file() in order
1110  * to do avoid the need for any locks.
1111  */
1112 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1113 {
1114         struct file *old_exe_file;
1115
1116         /*
1117          * It is safe to dereference the exe_file without RCU as
1118          * this function is only called if nobody else can access
1119          * this mm -- see comment above for justification.
1120          */
1121         old_exe_file = rcu_dereference_raw(mm->exe_file);
1122
1123         if (new_exe_file)
1124                 get_file(new_exe_file);
1125         rcu_assign_pointer(mm->exe_file, new_exe_file);
1126         if (old_exe_file)
1127                 fput(old_exe_file);
1128 }
1129
1130 /**
1131  * get_mm_exe_file - acquire a reference to the mm's executable file
1132  *
1133  * Returns %NULL if mm has no associated executable file.
1134  * User must release file via fput().
1135  */
1136 struct file *get_mm_exe_file(struct mm_struct *mm)
1137 {
1138         struct file *exe_file;
1139
1140         rcu_read_lock();
1141         exe_file = rcu_dereference(mm->exe_file);
1142         if (exe_file && !get_file_rcu(exe_file))
1143                 exe_file = NULL;
1144         rcu_read_unlock();
1145         return exe_file;
1146 }
1147 EXPORT_SYMBOL(get_mm_exe_file);
1148
1149 /**
1150  * get_task_exe_file - acquire a reference to the task's executable file
1151  *
1152  * Returns %NULL if task's mm (if any) has no associated executable file or
1153  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1154  * User must release file via fput().
1155  */
1156 struct file *get_task_exe_file(struct task_struct *task)
1157 {
1158         struct file *exe_file = NULL;
1159         struct mm_struct *mm;
1160
1161         task_lock(task);
1162         mm = task->mm;
1163         if (mm) {
1164                 if (!(task->flags & PF_KTHREAD))
1165                         exe_file = get_mm_exe_file(mm);
1166         }
1167         task_unlock(task);
1168         return exe_file;
1169 }
1170 EXPORT_SYMBOL(get_task_exe_file);
1171
1172 /**
1173  * get_task_mm - acquire a reference to the task's mm
1174  *
1175  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1176  * this kernel workthread has transiently adopted a user mm with use_mm,
1177  * to do its AIO) is not set and if so returns a reference to it, after
1178  * bumping up the use count.  User must release the mm via mmput()
1179  * after use.  Typically used by /proc and ptrace.
1180  */
1181 struct mm_struct *get_task_mm(struct task_struct *task)
1182 {
1183         struct mm_struct *mm;
1184
1185         task_lock(task);
1186         mm = task->mm;
1187         if (mm) {
1188                 if (task->flags & PF_KTHREAD)
1189                         mm = NULL;
1190                 else
1191                         mmget(mm);
1192         }
1193         task_unlock(task);
1194         return mm;
1195 }
1196 EXPORT_SYMBOL_GPL(get_task_mm);
1197
1198 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1199 {
1200         struct mm_struct *mm;
1201         int err;
1202
1203         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1204         if (err)
1205                 return ERR_PTR(err);
1206
1207         mm = get_task_mm(task);
1208         if (mm && mm != current->mm &&
1209                         !ptrace_may_access(task, mode)) {
1210                 mmput(mm);
1211                 mm = ERR_PTR(-EACCES);
1212         }
1213         mutex_unlock(&task->signal->cred_guard_mutex);
1214
1215         return mm;
1216 }
1217
1218 static void complete_vfork_done(struct task_struct *tsk)
1219 {
1220         struct completion *vfork;
1221
1222         task_lock(tsk);
1223         vfork = tsk->vfork_done;
1224         if (likely(vfork)) {
1225                 tsk->vfork_done = NULL;
1226                 complete(vfork);
1227         }
1228         task_unlock(tsk);
1229 }
1230
1231 static int wait_for_vfork_done(struct task_struct *child,
1232                                 struct completion *vfork)
1233 {
1234         int killed;
1235
1236         freezer_do_not_count();
1237         cgroup_enter_frozen();
1238         killed = wait_for_completion_killable(vfork);
1239         cgroup_leave_frozen(false);
1240         freezer_count();
1241
1242         if (killed) {
1243                 task_lock(child);
1244                 child->vfork_done = NULL;
1245                 task_unlock(child);
1246         }
1247
1248         put_task_struct(child);
1249         return killed;
1250 }
1251
1252 /* Please note the differences between mmput and mm_release.
1253  * mmput is called whenever we stop holding onto a mm_struct,
1254  * error success whatever.
1255  *
1256  * mm_release is called after a mm_struct has been removed
1257  * from the current process.
1258  *
1259  * This difference is important for error handling, when we
1260  * only half set up a mm_struct for a new process and need to restore
1261  * the old one.  Because we mmput the new mm_struct before
1262  * restoring the old one. . .
1263  * Eric Biederman 10 January 1998
1264  */
1265 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1266 {
1267         /* Get rid of any futexes when releasing the mm */
1268 #ifdef CONFIG_FUTEX
1269         if (unlikely(tsk->robust_list)) {
1270                 exit_robust_list(tsk);
1271                 tsk->robust_list = NULL;
1272         }
1273 #ifdef CONFIG_COMPAT
1274         if (unlikely(tsk->compat_robust_list)) {
1275                 compat_exit_robust_list(tsk);
1276                 tsk->compat_robust_list = NULL;
1277         }
1278 #endif
1279         if (unlikely(!list_empty(&tsk->pi_state_list)))
1280                 exit_pi_state_list(tsk);
1281 #endif
1282
1283         uprobe_free_utask(tsk);
1284
1285         /* Get rid of any cached register state */
1286         deactivate_mm(tsk, mm);
1287
1288         /*
1289          * Signal userspace if we're not exiting with a core dump
1290          * because we want to leave the value intact for debugging
1291          * purposes.
1292          */
1293         if (tsk->clear_child_tid) {
1294                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1295                     atomic_read(&mm->mm_users) > 1) {
1296                         /*
1297                          * We don't check the error code - if userspace has
1298                          * not set up a proper pointer then tough luck.
1299                          */
1300                         put_user(0, tsk->clear_child_tid);
1301                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1302                                         1, NULL, NULL, 0, 0);
1303                 }
1304                 tsk->clear_child_tid = NULL;
1305         }
1306
1307         /*
1308          * All done, finally we can wake up parent and return this mm to him.
1309          * Also kthread_stop() uses this completion for synchronization.
1310          */
1311         if (tsk->vfork_done)
1312                 complete_vfork_done(tsk);
1313 }
1314
1315 /**
1316  * dup_mm() - duplicates an existing mm structure
1317  * @tsk: the task_struct with which the new mm will be associated.
1318  * @oldmm: the mm to duplicate.
1319  *
1320  * Allocates a new mm structure and duplicates the provided @oldmm structure
1321  * content into it.
1322  *
1323  * Return: the duplicated mm or NULL on failure.
1324  */
1325 static struct mm_struct *dup_mm(struct task_struct *tsk,
1326                                 struct mm_struct *oldmm)
1327 {
1328         struct mm_struct *mm;
1329         int err;
1330
1331         mm = allocate_mm();
1332         if (!mm)
1333                 goto fail_nomem;
1334
1335         memcpy(mm, oldmm, sizeof(*mm));
1336
1337         if (!mm_init(mm, tsk, mm->user_ns))
1338                 goto fail_nomem;
1339
1340         err = dup_mmap(mm, oldmm);
1341         if (err)
1342                 goto free_pt;
1343
1344         mm->hiwater_rss = get_mm_rss(mm);
1345         mm->hiwater_vm = mm->total_vm;
1346
1347         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1348                 goto free_pt;
1349
1350         return mm;
1351
1352 free_pt:
1353         /* don't put binfmt in mmput, we haven't got module yet */
1354         mm->binfmt = NULL;
1355         mm_init_owner(mm, NULL);
1356         mmput(mm);
1357
1358 fail_nomem:
1359         return NULL;
1360 }
1361
1362 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1363 {
1364         struct mm_struct *mm, *oldmm;
1365         int retval;
1366
1367         tsk->min_flt = tsk->maj_flt = 0;
1368         tsk->nvcsw = tsk->nivcsw = 0;
1369 #ifdef CONFIG_DETECT_HUNG_TASK
1370         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1371         tsk->last_switch_time = 0;
1372 #endif
1373
1374         tsk->mm = NULL;
1375         tsk->active_mm = NULL;
1376
1377         /*
1378          * Are we cloning a kernel thread?
1379          *
1380          * We need to steal a active VM for that..
1381          */
1382         oldmm = current->mm;
1383         if (!oldmm)
1384                 return 0;
1385
1386         /* initialize the new vmacache entries */
1387         vmacache_flush(tsk);
1388
1389         if (clone_flags & CLONE_VM) {
1390                 mmget(oldmm);
1391                 mm = oldmm;
1392                 goto good_mm;
1393         }
1394
1395         retval = -ENOMEM;
1396         mm = dup_mm(tsk, current->mm);
1397         if (!mm)
1398                 goto fail_nomem;
1399
1400 good_mm:
1401         tsk->mm = mm;
1402         tsk->active_mm = mm;
1403         return 0;
1404
1405 fail_nomem:
1406         return retval;
1407 }
1408
1409 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1410 {
1411         struct fs_struct *fs = current->fs;
1412         if (clone_flags & CLONE_FS) {
1413                 /* tsk->fs is already what we want */
1414                 spin_lock(&fs->lock);
1415                 if (fs->in_exec) {
1416                         spin_unlock(&fs->lock);
1417                         return -EAGAIN;
1418                 }
1419                 fs->users++;
1420                 spin_unlock(&fs->lock);
1421                 return 0;
1422         }
1423         tsk->fs = copy_fs_struct(fs);
1424         if (!tsk->fs)
1425                 return -ENOMEM;
1426         return 0;
1427 }
1428
1429 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1430 {
1431         struct files_struct *oldf, *newf;
1432         int error = 0;
1433
1434         /*
1435          * A background process may not have any files ...
1436          */
1437         oldf = current->files;
1438         if (!oldf)
1439                 goto out;
1440
1441         if (clone_flags & CLONE_FILES) {
1442                 atomic_inc(&oldf->count);
1443                 goto out;
1444         }
1445
1446         newf = dup_fd(oldf, &error);
1447         if (!newf)
1448                 goto out;
1449
1450         tsk->files = newf;
1451         error = 0;
1452 out:
1453         return error;
1454 }
1455
1456 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1457 {
1458 #ifdef CONFIG_BLOCK
1459         struct io_context *ioc = current->io_context;
1460         struct io_context *new_ioc;
1461
1462         if (!ioc)
1463                 return 0;
1464         /*
1465          * Share io context with parent, if CLONE_IO is set
1466          */
1467         if (clone_flags & CLONE_IO) {
1468                 ioc_task_link(ioc);
1469                 tsk->io_context = ioc;
1470         } else if (ioprio_valid(ioc->ioprio)) {
1471                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1472                 if (unlikely(!new_ioc))
1473                         return -ENOMEM;
1474
1475                 new_ioc->ioprio = ioc->ioprio;
1476                 put_io_context(new_ioc);
1477         }
1478 #endif
1479         return 0;
1480 }
1481
1482 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1483 {
1484         struct sighand_struct *sig;
1485
1486         if (clone_flags & CLONE_SIGHAND) {
1487                 refcount_inc(&current->sighand->count);
1488                 return 0;
1489         }
1490         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1491         rcu_assign_pointer(tsk->sighand, sig);
1492         if (!sig)
1493                 return -ENOMEM;
1494
1495         refcount_set(&sig->count, 1);
1496         spin_lock_irq(&current->sighand->siglock);
1497         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1498         spin_unlock_irq(&current->sighand->siglock);
1499         return 0;
1500 }
1501
1502 void __cleanup_sighand(struct sighand_struct *sighand)
1503 {
1504         if (refcount_dec_and_test(&sighand->count)) {
1505                 signalfd_cleanup(sighand);
1506                 /*
1507                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1508                  * without an RCU grace period, see __lock_task_sighand().
1509                  */
1510                 kmem_cache_free(sighand_cachep, sighand);
1511         }
1512 }
1513
1514 #ifdef CONFIG_POSIX_TIMERS
1515 /*
1516  * Initialize POSIX timer handling for a thread group.
1517  */
1518 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1519 {
1520         unsigned long cpu_limit;
1521
1522         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1523         if (cpu_limit != RLIM_INFINITY) {
1524                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1525                 sig->cputimer.running = true;
1526         }
1527
1528         /* The timer lists. */
1529         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1530         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1531         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1532 }
1533 #else
1534 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1535 #endif
1536
1537 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1538 {
1539         struct signal_struct *sig;
1540
1541         if (clone_flags & CLONE_THREAD)
1542                 return 0;
1543
1544         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1545         tsk->signal = sig;
1546         if (!sig)
1547                 return -ENOMEM;
1548
1549         sig->nr_threads = 1;
1550         atomic_set(&sig->live, 1);
1551         refcount_set(&sig->sigcnt, 1);
1552
1553         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1554         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1555         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1556
1557         init_waitqueue_head(&sig->wait_chldexit);
1558         sig->curr_target = tsk;
1559         init_sigpending(&sig->shared_pending);
1560         INIT_HLIST_HEAD(&sig->multiprocess);
1561         seqlock_init(&sig->stats_lock);
1562         prev_cputime_init(&sig->prev_cputime);
1563
1564 #ifdef CONFIG_POSIX_TIMERS
1565         INIT_LIST_HEAD(&sig->posix_timers);
1566         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1567         sig->real_timer.function = it_real_fn;
1568 #endif
1569
1570         task_lock(current->group_leader);
1571         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1572         task_unlock(current->group_leader);
1573
1574         posix_cpu_timers_init_group(sig);
1575
1576         tty_audit_fork(sig);
1577         sched_autogroup_fork(sig);
1578
1579         sig->oom_score_adj = current->signal->oom_score_adj;
1580         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1581
1582         mutex_init(&sig->cred_guard_mutex);
1583
1584         return 0;
1585 }
1586
1587 static void copy_seccomp(struct task_struct *p)
1588 {
1589 #ifdef CONFIG_SECCOMP
1590         /*
1591          * Must be called with sighand->lock held, which is common to
1592          * all threads in the group. Holding cred_guard_mutex is not
1593          * needed because this new task is not yet running and cannot
1594          * be racing exec.
1595          */
1596         assert_spin_locked(&current->sighand->siglock);
1597
1598         /* Ref-count the new filter user, and assign it. */
1599         get_seccomp_filter(current);
1600         p->seccomp = current->seccomp;
1601
1602         /*
1603          * Explicitly enable no_new_privs here in case it got set
1604          * between the task_struct being duplicated and holding the
1605          * sighand lock. The seccomp state and nnp must be in sync.
1606          */
1607         if (task_no_new_privs(current))
1608                 task_set_no_new_privs(p);
1609
1610         /*
1611          * If the parent gained a seccomp mode after copying thread
1612          * flags and between before we held the sighand lock, we have
1613          * to manually enable the seccomp thread flag here.
1614          */
1615         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1616                 set_tsk_thread_flag(p, TIF_SECCOMP);
1617 #endif
1618 }
1619
1620 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1621 {
1622         current->clear_child_tid = tidptr;
1623
1624         return task_pid_vnr(current);
1625 }
1626
1627 static void rt_mutex_init_task(struct task_struct *p)
1628 {
1629         raw_spin_lock_init(&p->pi_lock);
1630 #ifdef CONFIG_RT_MUTEXES
1631         p->pi_waiters = RB_ROOT_CACHED;
1632         p->pi_top_task = NULL;
1633         p->pi_blocked_on = NULL;
1634 #endif
1635 }
1636
1637 #ifdef CONFIG_POSIX_TIMERS
1638 /*
1639  * Initialize POSIX timer handling for a single task.
1640  */
1641 static void posix_cpu_timers_init(struct task_struct *tsk)
1642 {
1643         tsk->cputime_expires.prof_exp = 0;
1644         tsk->cputime_expires.virt_exp = 0;
1645         tsk->cputime_expires.sched_exp = 0;
1646         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1647         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1648         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1649 }
1650 #else
1651 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1652 #endif
1653
1654 static inline void init_task_pid_links(struct task_struct *task)
1655 {
1656         enum pid_type type;
1657
1658         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1659                 INIT_HLIST_NODE(&task->pid_links[type]);
1660         }
1661 }
1662
1663 static inline void
1664 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1665 {
1666         if (type == PIDTYPE_PID)
1667                 task->thread_pid = pid;
1668         else
1669                 task->signal->pids[type] = pid;
1670 }
1671
1672 static inline void rcu_copy_process(struct task_struct *p)
1673 {
1674 #ifdef CONFIG_PREEMPT_RCU
1675         p->rcu_read_lock_nesting = 0;
1676         p->rcu_read_unlock_special.s = 0;
1677         p->rcu_blocked_node = NULL;
1678         INIT_LIST_HEAD(&p->rcu_node_entry);
1679 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1680 #ifdef CONFIG_TASKS_RCU
1681         p->rcu_tasks_holdout = false;
1682         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1683         p->rcu_tasks_idle_cpu = -1;
1684 #endif /* #ifdef CONFIG_TASKS_RCU */
1685 }
1686
1687 static int pidfd_release(struct inode *inode, struct file *file)
1688 {
1689         struct pid *pid = file->private_data;
1690
1691         file->private_data = NULL;
1692         put_pid(pid);
1693         return 0;
1694 }
1695
1696 #ifdef CONFIG_PROC_FS
1697 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1698 {
1699         struct pid_namespace *ns = proc_pid_ns(file_inode(m->file));
1700         struct pid *pid = f->private_data;
1701
1702         seq_put_decimal_ull(m, "Pid:\t", pid_nr_ns(pid, ns));
1703         seq_putc(m, '\n');
1704 }
1705 #endif
1706
1707 const struct file_operations pidfd_fops = {
1708         .release = pidfd_release,
1709 #ifdef CONFIG_PROC_FS
1710         .show_fdinfo = pidfd_show_fdinfo,
1711 #endif
1712 };
1713
1714 /**
1715  * pidfd_create() - Create a new pid file descriptor.
1716  *
1717  * @pid:  struct pid that the pidfd will reference
1718  *
1719  * This creates a new pid file descriptor with the O_CLOEXEC flag set.
1720  *
1721  * Note, that this function can only be called after the fd table has
1722  * been unshared to avoid leaking the pidfd to the new process.
1723  *
1724  * Return: On success, a cloexec pidfd is returned.
1725  *         On error, a negative errno number will be returned.
1726  */
1727 static int pidfd_create(struct pid *pid)
1728 {
1729         int fd;
1730
1731         fd = anon_inode_getfd("[pidfd]", &pidfd_fops, get_pid(pid),
1732                               O_RDWR | O_CLOEXEC);
1733         if (fd < 0)
1734                 put_pid(pid);
1735
1736         return fd;
1737 }
1738
1739 static void __delayed_free_task(struct rcu_head *rhp)
1740 {
1741         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1742
1743         free_task(tsk);
1744 }
1745
1746 static __always_inline void delayed_free_task(struct task_struct *tsk)
1747 {
1748         if (IS_ENABLED(CONFIG_MEMCG))
1749                 call_rcu(&tsk->rcu, __delayed_free_task);
1750         else
1751                 free_task(tsk);
1752 }
1753
1754 /*
1755  * This creates a new process as a copy of the old one,
1756  * but does not actually start it yet.
1757  *
1758  * It copies the registers, and all the appropriate
1759  * parts of the process environment (as per the clone
1760  * flags). The actual kick-off is left to the caller.
1761  */
1762 static __latent_entropy struct task_struct *copy_process(
1763                                         struct pid *pid,
1764                                         int trace,
1765                                         int node,
1766                                         struct kernel_clone_args *args)
1767 {
1768         int pidfd = -1, retval;
1769         struct task_struct *p;
1770         struct multiprocess_signals delayed;
1771         u64 clone_flags = args->flags;
1772
1773         /*
1774          * Don't allow sharing the root directory with processes in a different
1775          * namespace
1776          */
1777         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1778                 return ERR_PTR(-EINVAL);
1779
1780         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1781                 return ERR_PTR(-EINVAL);
1782
1783         /*
1784          * Thread groups must share signals as well, and detached threads
1785          * can only be started up within the thread group.
1786          */
1787         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1788                 return ERR_PTR(-EINVAL);
1789
1790         /*
1791          * Shared signal handlers imply shared VM. By way of the above,
1792          * thread groups also imply shared VM. Blocking this case allows
1793          * for various simplifications in other code.
1794          */
1795         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1796                 return ERR_PTR(-EINVAL);
1797
1798         /*
1799          * Siblings of global init remain as zombies on exit since they are
1800          * not reaped by their parent (swapper). To solve this and to avoid
1801          * multi-rooted process trees, prevent global and container-inits
1802          * from creating siblings.
1803          */
1804         if ((clone_flags & CLONE_PARENT) &&
1805                                 current->signal->flags & SIGNAL_UNKILLABLE)
1806                 return ERR_PTR(-EINVAL);
1807
1808         /*
1809          * If the new process will be in a different pid or user namespace
1810          * do not allow it to share a thread group with the forking task.
1811          */
1812         if (clone_flags & CLONE_THREAD) {
1813                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1814                     (task_active_pid_ns(current) !=
1815                                 current->nsproxy->pid_ns_for_children))
1816                         return ERR_PTR(-EINVAL);
1817         }
1818
1819         if (clone_flags & CLONE_PIDFD) {
1820                 /*
1821                  * - CLONE_DETACHED is blocked so that we can potentially
1822                  *   reuse it later for CLONE_PIDFD.
1823                  * - CLONE_THREAD is blocked until someone really needs it.
1824                  */
1825                 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1826                         return ERR_PTR(-EINVAL);
1827         }
1828
1829         /*
1830          * Force any signals received before this point to be delivered
1831          * before the fork happens.  Collect up signals sent to multiple
1832          * processes that happen during the fork and delay them so that
1833          * they appear to happen after the fork.
1834          */
1835         sigemptyset(&delayed.signal);
1836         INIT_HLIST_NODE(&delayed.node);
1837
1838         spin_lock_irq(&current->sighand->siglock);
1839         if (!(clone_flags & CLONE_THREAD))
1840                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1841         recalc_sigpending();
1842         spin_unlock_irq(&current->sighand->siglock);
1843         retval = -ERESTARTNOINTR;
1844         if (signal_pending(current))
1845                 goto fork_out;
1846
1847         retval = -ENOMEM;
1848         p = dup_task_struct(current, node);
1849         if (!p)
1850                 goto fork_out;
1851
1852         /*
1853          * This _must_ happen before we call free_task(), i.e. before we jump
1854          * to any of the bad_fork_* labels. This is to avoid freeing
1855          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1856          * kernel threads (PF_KTHREAD).
1857          */
1858         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1859         /*
1860          * Clear TID on mm_release()?
1861          */
1862         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1863
1864         ftrace_graph_init_task(p);
1865
1866         rt_mutex_init_task(p);
1867
1868 #ifdef CONFIG_PROVE_LOCKING
1869         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1870         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1871 #endif
1872         retval = -EAGAIN;
1873         if (atomic_read(&p->real_cred->user->processes) >=
1874                         task_rlimit(p, RLIMIT_NPROC)) {
1875                 if (p->real_cred->user != INIT_USER &&
1876                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1877                         goto bad_fork_free;
1878         }
1879         current->flags &= ~PF_NPROC_EXCEEDED;
1880
1881         retval = copy_creds(p, clone_flags);
1882         if (retval < 0)
1883                 goto bad_fork_free;
1884
1885         /*
1886          * If multiple threads are within copy_process(), then this check
1887          * triggers too late. This doesn't hurt, the check is only there
1888          * to stop root fork bombs.
1889          */
1890         retval = -EAGAIN;
1891         if (nr_threads >= max_threads)
1892                 goto bad_fork_cleanup_count;
1893
1894         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1895         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1896         p->flags |= PF_FORKNOEXEC;
1897         INIT_LIST_HEAD(&p->children);
1898         INIT_LIST_HEAD(&p->sibling);
1899         rcu_copy_process(p);
1900         p->vfork_done = NULL;
1901         spin_lock_init(&p->alloc_lock);
1902
1903         init_sigpending(&p->pending);
1904
1905         p->utime = p->stime = p->gtime = 0;
1906 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1907         p->utimescaled = p->stimescaled = 0;
1908 #endif
1909         prev_cputime_init(&p->prev_cputime);
1910
1911 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1912         seqcount_init(&p->vtime.seqcount);
1913         p->vtime.starttime = 0;
1914         p->vtime.state = VTIME_INACTIVE;
1915 #endif
1916
1917 #if defined(SPLIT_RSS_COUNTING)
1918         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1919 #endif
1920
1921         p->default_timer_slack_ns = current->timer_slack_ns;
1922
1923 #ifdef CONFIG_PSI
1924         p->psi_flags = 0;
1925 #endif
1926
1927         task_io_accounting_init(&p->ioac);
1928         acct_clear_integrals(p);
1929
1930         posix_cpu_timers_init(p);
1931
1932         p->io_context = NULL;
1933         audit_set_context(p, NULL);
1934         cgroup_fork(p);
1935 #ifdef CONFIG_NUMA
1936         p->mempolicy = mpol_dup(p->mempolicy);
1937         if (IS_ERR(p->mempolicy)) {
1938                 retval = PTR_ERR(p->mempolicy);
1939                 p->mempolicy = NULL;
1940                 goto bad_fork_cleanup_threadgroup_lock;
1941         }
1942 #endif
1943 #ifdef CONFIG_CPUSETS
1944         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1945         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1946         seqcount_init(&p->mems_allowed_seq);
1947 #endif
1948 #ifdef CONFIG_TRACE_IRQFLAGS
1949         p->irq_events = 0;
1950         p->hardirqs_enabled = 0;
1951         p->hardirq_enable_ip = 0;
1952         p->hardirq_enable_event = 0;
1953         p->hardirq_disable_ip = _THIS_IP_;
1954         p->hardirq_disable_event = 0;
1955         p->softirqs_enabled = 1;
1956         p->softirq_enable_ip = _THIS_IP_;
1957         p->softirq_enable_event = 0;
1958         p->softirq_disable_ip = 0;
1959         p->softirq_disable_event = 0;
1960         p->hardirq_context = 0;
1961         p->softirq_context = 0;
1962 #endif
1963
1964         p->pagefault_disabled = 0;
1965
1966 #ifdef CONFIG_LOCKDEP
1967         p->lockdep_depth = 0; /* no locks held yet */
1968         p->curr_chain_key = 0;
1969         p->lockdep_recursion = 0;
1970         lockdep_init_task(p);
1971 #endif
1972
1973 #ifdef CONFIG_DEBUG_MUTEXES
1974         p->blocked_on = NULL; /* not blocked yet */
1975 #endif
1976 #ifdef CONFIG_BCACHE
1977         p->sequential_io        = 0;
1978         p->sequential_io_avg    = 0;
1979 #endif
1980
1981         /* Perform scheduler related setup. Assign this task to a CPU. */
1982         retval = sched_fork(clone_flags, p);
1983         if (retval)
1984                 goto bad_fork_cleanup_policy;
1985
1986         retval = perf_event_init_task(p);
1987         if (retval)
1988                 goto bad_fork_cleanup_policy;
1989         retval = audit_alloc(p);
1990         if (retval)
1991                 goto bad_fork_cleanup_perf;
1992         /* copy all the process information */
1993         shm_init_task(p);
1994         retval = security_task_alloc(p, clone_flags);
1995         if (retval)
1996                 goto bad_fork_cleanup_audit;
1997         retval = copy_semundo(clone_flags, p);
1998         if (retval)
1999                 goto bad_fork_cleanup_security;
2000         retval = copy_files(clone_flags, p);
2001         if (retval)
2002                 goto bad_fork_cleanup_semundo;
2003         retval = copy_fs(clone_flags, p);
2004         if (retval)
2005                 goto bad_fork_cleanup_files;
2006         retval = copy_sighand(clone_flags, p);
2007         if (retval)
2008                 goto bad_fork_cleanup_fs;
2009         retval = copy_signal(clone_flags, p);
2010         if (retval)
2011                 goto bad_fork_cleanup_sighand;
2012         retval = copy_mm(clone_flags, p);
2013         if (retval)
2014                 goto bad_fork_cleanup_signal;
2015         retval = copy_namespaces(clone_flags, p);
2016         if (retval)
2017                 goto bad_fork_cleanup_mm;
2018         retval = copy_io(clone_flags, p);
2019         if (retval)
2020                 goto bad_fork_cleanup_namespaces;
2021         retval = copy_thread_tls(clone_flags, args->stack, args->stack_size, p,
2022                                  args->tls);
2023         if (retval)
2024                 goto bad_fork_cleanup_io;
2025
2026         stackleak_task_init(p);
2027
2028         if (pid != &init_struct_pid) {
2029                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
2030                 if (IS_ERR(pid)) {
2031                         retval = PTR_ERR(pid);
2032                         goto bad_fork_cleanup_thread;
2033                 }
2034         }
2035
2036         /*
2037          * This has to happen after we've potentially unshared the file
2038          * descriptor table (so that the pidfd doesn't leak into the child
2039          * if the fd table isn't shared).
2040          */
2041         if (clone_flags & CLONE_PIDFD) {
2042                 retval = pidfd_create(pid);
2043                 if (retval < 0)
2044                         goto bad_fork_free_pid;
2045
2046                 pidfd = retval;
2047                 retval = put_user(pidfd, args->pidfd);
2048                 if (retval)
2049                         goto bad_fork_put_pidfd;
2050         }
2051
2052 #ifdef CONFIG_BLOCK
2053         p->plug = NULL;
2054 #endif
2055 #ifdef CONFIG_FUTEX
2056         p->robust_list = NULL;
2057 #ifdef CONFIG_COMPAT
2058         p->compat_robust_list = NULL;
2059 #endif
2060         INIT_LIST_HEAD(&p->pi_state_list);
2061         p->pi_state_cache = NULL;
2062 #endif
2063         /*
2064          * sigaltstack should be cleared when sharing the same VM
2065          */
2066         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2067                 sas_ss_reset(p);
2068
2069         /*
2070          * Syscall tracing and stepping should be turned off in the
2071          * child regardless of CLONE_PTRACE.
2072          */
2073         user_disable_single_step(p);
2074         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
2075 #ifdef TIF_SYSCALL_EMU
2076         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
2077 #endif
2078         clear_tsk_latency_tracing(p);
2079
2080         /* ok, now we should be set up.. */
2081         p->pid = pid_nr(pid);
2082         if (clone_flags & CLONE_THREAD) {
2083                 p->exit_signal = -1;
2084                 p->group_leader = current->group_leader;
2085                 p->tgid = current->tgid;
2086         } else {
2087                 if (clone_flags & CLONE_PARENT)
2088                         p->exit_signal = current->group_leader->exit_signal;
2089                 else
2090                         p->exit_signal = args->exit_signal;
2091                 p->group_leader = p;
2092                 p->tgid = p->pid;
2093         }
2094
2095         p->nr_dirtied = 0;
2096         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2097         p->dirty_paused_when = 0;
2098
2099         p->pdeath_signal = 0;
2100         INIT_LIST_HEAD(&p->thread_group);
2101         p->task_works = NULL;
2102
2103         cgroup_threadgroup_change_begin(current);
2104         /*
2105          * Ensure that the cgroup subsystem policies allow the new process to be
2106          * forked. It should be noted the the new process's css_set can be changed
2107          * between here and cgroup_post_fork() if an organisation operation is in
2108          * progress.
2109          */
2110         retval = cgroup_can_fork(p);
2111         if (retval)
2112                 goto bad_fork_cgroup_threadgroup_change_end;
2113
2114         /*
2115          * From this point on we must avoid any synchronous user-space
2116          * communication until we take the tasklist-lock. In particular, we do
2117          * not want user-space to be able to predict the process start-time by
2118          * stalling fork(2) after we recorded the start_time but before it is
2119          * visible to the system.
2120          */
2121
2122         p->start_time = ktime_get_ns();
2123         p->real_start_time = ktime_get_boot_ns();
2124
2125         /*
2126          * Make it visible to the rest of the system, but dont wake it up yet.
2127          * Need tasklist lock for parent etc handling!
2128          */
2129         write_lock_irq(&tasklist_lock);
2130
2131         /* CLONE_PARENT re-uses the old parent */
2132         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2133                 p->real_parent = current->real_parent;
2134                 p->parent_exec_id = current->parent_exec_id;
2135         } else {
2136                 p->real_parent = current;
2137                 p->parent_exec_id = current->self_exec_id;
2138         }
2139
2140         klp_copy_process(p);
2141
2142         spin_lock(&current->sighand->siglock);
2143
2144         /*
2145          * Copy seccomp details explicitly here, in case they were changed
2146          * before holding sighand lock.
2147          */
2148         copy_seccomp(p);
2149
2150         rseq_fork(p, clone_flags);
2151
2152         /* Don't start children in a dying pid namespace */
2153         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2154                 retval = -ENOMEM;
2155                 goto bad_fork_cancel_cgroup;
2156         }
2157
2158         /* Let kill terminate clone/fork in the middle */
2159         if (fatal_signal_pending(current)) {
2160                 retval = -EINTR;
2161                 goto bad_fork_cancel_cgroup;
2162         }
2163
2164
2165         init_task_pid_links(p);
2166         if (likely(p->pid)) {
2167                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2168
2169                 init_task_pid(p, PIDTYPE_PID, pid);
2170                 if (thread_group_leader(p)) {
2171                         init_task_pid(p, PIDTYPE_TGID, pid);
2172                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2173                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2174
2175                         if (is_child_reaper(pid)) {
2176                                 ns_of_pid(pid)->child_reaper = p;
2177                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2178                         }
2179                         p->signal->shared_pending.signal = delayed.signal;
2180                         p->signal->tty = tty_kref_get(current->signal->tty);
2181                         /*
2182                          * Inherit has_child_subreaper flag under the same
2183                          * tasklist_lock with adding child to the process tree
2184                          * for propagate_has_child_subreaper optimization.
2185                          */
2186                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2187                                                          p->real_parent->signal->is_child_subreaper;
2188                         list_add_tail(&p->sibling, &p->real_parent->children);
2189                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2190                         attach_pid(p, PIDTYPE_TGID);
2191                         attach_pid(p, PIDTYPE_PGID);
2192                         attach_pid(p, PIDTYPE_SID);
2193                         __this_cpu_inc(process_counts);
2194                 } else {
2195                         current->signal->nr_threads++;
2196                         atomic_inc(&current->signal->live);
2197                         refcount_inc(&current->signal->sigcnt);
2198                         task_join_group_stop(p);
2199                         list_add_tail_rcu(&p->thread_group,
2200                                           &p->group_leader->thread_group);
2201                         list_add_tail_rcu(&p->thread_node,
2202                                           &p->signal->thread_head);
2203                 }
2204                 attach_pid(p, PIDTYPE_PID);
2205                 nr_threads++;
2206         }
2207         total_forks++;
2208         hlist_del_init(&delayed.node);
2209         spin_unlock(&current->sighand->siglock);
2210         syscall_tracepoint_update(p);
2211         write_unlock_irq(&tasklist_lock);
2212
2213         proc_fork_connector(p);
2214         cgroup_post_fork(p);
2215         cgroup_threadgroup_change_end(current);
2216         perf_event_fork(p);
2217
2218         trace_task_newtask(p, clone_flags);
2219         uprobe_copy_process(p, clone_flags);
2220
2221         return p;
2222
2223 bad_fork_cancel_cgroup:
2224         spin_unlock(&current->sighand->siglock);
2225         write_unlock_irq(&tasklist_lock);
2226         cgroup_cancel_fork(p);
2227 bad_fork_cgroup_threadgroup_change_end:
2228         cgroup_threadgroup_change_end(current);
2229 bad_fork_put_pidfd:
2230         if (clone_flags & CLONE_PIDFD)
2231                 ksys_close(pidfd);
2232 bad_fork_free_pid:
2233         if (pid != &init_struct_pid)
2234                 free_pid(pid);
2235 bad_fork_cleanup_thread:
2236         exit_thread(p);
2237 bad_fork_cleanup_io:
2238         if (p->io_context)
2239                 exit_io_context(p);
2240 bad_fork_cleanup_namespaces:
2241         exit_task_namespaces(p);
2242 bad_fork_cleanup_mm:
2243         if (p->mm) {
2244                 mm_clear_owner(p->mm, p);
2245                 mmput(p->mm);
2246         }
2247 bad_fork_cleanup_signal:
2248         if (!(clone_flags & CLONE_THREAD))
2249                 free_signal_struct(p->signal);
2250 bad_fork_cleanup_sighand:
2251         __cleanup_sighand(p->sighand);
2252 bad_fork_cleanup_fs:
2253         exit_fs(p); /* blocking */
2254 bad_fork_cleanup_files:
2255         exit_files(p); /* blocking */
2256 bad_fork_cleanup_semundo:
2257         exit_sem(p);
2258 bad_fork_cleanup_security:
2259         security_task_free(p);
2260 bad_fork_cleanup_audit:
2261         audit_free(p);
2262 bad_fork_cleanup_perf:
2263         perf_event_free_task(p);
2264 bad_fork_cleanup_policy:
2265         lockdep_free_task(p);
2266 #ifdef CONFIG_NUMA
2267         mpol_put(p->mempolicy);
2268 bad_fork_cleanup_threadgroup_lock:
2269 #endif
2270         delayacct_tsk_free(p);
2271 bad_fork_cleanup_count:
2272         atomic_dec(&p->cred->user->processes);
2273         exit_creds(p);
2274 bad_fork_free:
2275         p->state = TASK_DEAD;
2276         put_task_stack(p);
2277         delayed_free_task(p);
2278 fork_out:
2279         spin_lock_irq(&current->sighand->siglock);
2280         hlist_del_init(&delayed.node);
2281         spin_unlock_irq(&current->sighand->siglock);
2282         return ERR_PTR(retval);
2283 }
2284
2285 static inline void init_idle_pids(struct task_struct *idle)
2286 {
2287         enum pid_type type;
2288
2289         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2290                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2291                 init_task_pid(idle, type, &init_struct_pid);
2292         }
2293 }
2294
2295 struct task_struct *fork_idle(int cpu)
2296 {
2297         struct task_struct *task;
2298         struct kernel_clone_args args = {
2299                 .flags = CLONE_VM,
2300         };
2301
2302         task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2303         if (!IS_ERR(task)) {
2304                 init_idle_pids(task);
2305                 init_idle(task, cpu);
2306         }
2307
2308         return task;
2309 }
2310
2311 struct mm_struct *copy_init_mm(void)
2312 {
2313         return dup_mm(NULL, &init_mm);
2314 }
2315
2316 /*
2317  *  Ok, this is the main fork-routine.
2318  *
2319  * It copies the process, and if successful kick-starts
2320  * it and waits for it to finish using the VM if required.
2321  */
2322 long _do_fork(struct kernel_clone_args *args)
2323 {
2324         u64 clone_flags = args->flags;
2325         struct completion vfork;
2326         struct pid *pid;
2327         struct task_struct *p;
2328         int trace = 0;
2329         long nr;
2330
2331         /*
2332          * Determine whether and which event to report to ptracer.  When
2333          * called from kernel_thread or CLONE_UNTRACED is explicitly
2334          * requested, no event is reported; otherwise, report if the event
2335          * for the type of forking is enabled.
2336          */
2337         if (!(clone_flags & CLONE_UNTRACED)) {
2338                 if (clone_flags & CLONE_VFORK)
2339                         trace = PTRACE_EVENT_VFORK;
2340                 else if (args->exit_signal != SIGCHLD)
2341                         trace = PTRACE_EVENT_CLONE;
2342                 else
2343                         trace = PTRACE_EVENT_FORK;
2344
2345                 if (likely(!ptrace_event_enabled(current, trace)))
2346                         trace = 0;
2347         }
2348
2349         p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2350         add_latent_entropy();
2351
2352         if (IS_ERR(p))
2353                 return PTR_ERR(p);
2354
2355         /*
2356          * Do this prior waking up the new thread - the thread pointer
2357          * might get invalid after that point, if the thread exits quickly.
2358          */
2359         trace_sched_process_fork(current, p);
2360
2361         pid = get_task_pid(p, PIDTYPE_PID);
2362         nr = pid_vnr(pid);
2363
2364         if (clone_flags & CLONE_PARENT_SETTID)
2365                 put_user(nr, args->parent_tid);
2366
2367         if (clone_flags & CLONE_VFORK) {
2368                 p->vfork_done = &vfork;
2369                 init_completion(&vfork);
2370                 get_task_struct(p);
2371         }
2372
2373         wake_up_new_task(p);
2374
2375         /* forking complete and child started to run, tell ptracer */
2376         if (unlikely(trace))
2377                 ptrace_event_pid(trace, pid);
2378
2379         if (clone_flags & CLONE_VFORK) {
2380                 if (!wait_for_vfork_done(p, &vfork))
2381                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2382         }
2383
2384         put_pid(pid);
2385         return nr;
2386 }
2387
2388 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2389 /* For compatibility with architectures that call do_fork directly rather than
2390  * using the syscall entry points below. */
2391 long do_fork(unsigned long clone_flags,
2392               unsigned long stack_start,
2393               unsigned long stack_size,
2394               int __user *parent_tidptr,
2395               int __user *child_tidptr)
2396 {
2397         struct kernel_clone_args args = {
2398                 .flags          = (clone_flags & ~CSIGNAL),
2399                 .child_tid      = child_tidptr,
2400                 .parent_tid     = parent_tidptr,
2401                 .exit_signal    = (clone_flags & CSIGNAL),
2402                 .stack          = stack_start,
2403                 .stack_size     = stack_size,
2404         };
2405
2406         return _do_fork(&args);
2407 }
2408 #endif
2409
2410 /*
2411  * Create a kernel thread.
2412  */
2413 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2414 {
2415         struct kernel_clone_args args = {
2416                 .flags          = ((flags | CLONE_VM | CLONE_UNTRACED) & ~CSIGNAL),
2417                 .exit_signal    = (flags & CSIGNAL),
2418                 .stack          = (unsigned long)fn,
2419                 .stack_size     = (unsigned long)arg,
2420         };
2421
2422         return _do_fork(&args);
2423 }
2424
2425 #ifdef __ARCH_WANT_SYS_FORK
2426 SYSCALL_DEFINE0(fork)
2427 {
2428 #ifdef CONFIG_MMU
2429         struct kernel_clone_args args = {
2430                 .exit_signal = SIGCHLD,
2431         };
2432
2433         return _do_fork(&args);
2434 #else
2435         /* can not support in nommu mode */
2436         return -EINVAL;
2437 #endif
2438 }
2439 #endif
2440
2441 #ifdef __ARCH_WANT_SYS_VFORK
2442 SYSCALL_DEFINE0(vfork)
2443 {
2444         struct kernel_clone_args args = {
2445                 .flags          = CLONE_VFORK | CLONE_VM,
2446                 .exit_signal    = SIGCHLD,
2447         };
2448
2449         return _do_fork(&args);
2450 }
2451 #endif
2452
2453 #ifdef __ARCH_WANT_SYS_CLONE
2454 #ifdef CONFIG_CLONE_BACKWARDS
2455 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2456                  int __user *, parent_tidptr,
2457                  unsigned long, tls,
2458                  int __user *, child_tidptr)
2459 #elif defined(CONFIG_CLONE_BACKWARDS2)
2460 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2461                  int __user *, parent_tidptr,
2462                  int __user *, child_tidptr,
2463                  unsigned long, tls)
2464 #elif defined(CONFIG_CLONE_BACKWARDS3)
2465 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2466                 int, stack_size,
2467                 int __user *, parent_tidptr,
2468                 int __user *, child_tidptr,
2469                 unsigned long, tls)
2470 #else
2471 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2472                  int __user *, parent_tidptr,
2473                  int __user *, child_tidptr,
2474                  unsigned long, tls)
2475 #endif
2476 {
2477         struct kernel_clone_args args = {
2478                 .flags          = (clone_flags & ~CSIGNAL),
2479                 .pidfd          = parent_tidptr,
2480                 .child_tid      = child_tidptr,
2481                 .parent_tid     = parent_tidptr,
2482                 .exit_signal    = (clone_flags & CSIGNAL),
2483                 .stack          = newsp,
2484                 .tls            = tls,
2485         };
2486
2487         /* clone(CLONE_PIDFD) uses parent_tidptr to return a pidfd */
2488         if ((clone_flags & CLONE_PIDFD) && (clone_flags & CLONE_PARENT_SETTID))
2489                 return -EINVAL;
2490
2491         return _do_fork(&args);
2492 }
2493
2494 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2495                                               struct clone_args __user *uargs,
2496                                               size_t size)
2497 {
2498         struct clone_args args;
2499
2500         if (unlikely(size > PAGE_SIZE))
2501                 return -E2BIG;
2502
2503         if (unlikely(size < sizeof(struct clone_args)))
2504                 return -EINVAL;
2505
2506         if (unlikely(!access_ok(uargs, size)))
2507                 return -EFAULT;
2508
2509         if (size > sizeof(struct clone_args)) {
2510                 unsigned char __user *addr;
2511                 unsigned char __user *end;
2512                 unsigned char val;
2513
2514                 addr = (void __user *)uargs + sizeof(struct clone_args);
2515                 end = (void __user *)uargs + size;
2516
2517                 for (; addr < end; addr++) {
2518                         if (get_user(val, addr))
2519                                 return -EFAULT;
2520                         if (val)
2521                                 return -E2BIG;
2522                 }
2523
2524                 size = sizeof(struct clone_args);
2525         }
2526
2527         if (copy_from_user(&args, uargs, size))
2528                 return -EFAULT;
2529
2530         *kargs = (struct kernel_clone_args){
2531                 .flags          = args.flags,
2532                 .pidfd          = u64_to_user_ptr(args.pidfd),
2533                 .child_tid      = u64_to_user_ptr(args.child_tid),
2534                 .parent_tid     = u64_to_user_ptr(args.parent_tid),
2535                 .exit_signal    = args.exit_signal,
2536                 .stack          = args.stack,
2537                 .stack_size     = args.stack_size,
2538                 .tls            = args.tls,
2539         };
2540
2541         return 0;
2542 }
2543
2544 static bool clone3_args_valid(const struct kernel_clone_args *kargs)
2545 {
2546         /*
2547          * All lower bits of the flag word are taken.
2548          * Verify that no other unknown flags are passed along.
2549          */
2550         if (kargs->flags & ~CLONE_LEGACY_FLAGS)
2551                 return false;
2552
2553         /*
2554          * - make the CLONE_DETACHED bit reuseable for clone3
2555          * - make the CSIGNAL bits reuseable for clone3
2556          */
2557         if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2558                 return false;
2559
2560         if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2561             kargs->exit_signal)
2562                 return false;
2563
2564         return true;
2565 }
2566
2567 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2568 {
2569         int err;
2570
2571         struct kernel_clone_args kargs;
2572
2573         err = copy_clone_args_from_user(&kargs, uargs, size);
2574         if (err)
2575                 return err;
2576
2577         if (!clone3_args_valid(&kargs))
2578                 return -EINVAL;
2579
2580         return _do_fork(&kargs);
2581 }
2582 #endif
2583
2584 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2585 {
2586         struct task_struct *leader, *parent, *child;
2587         int res;
2588
2589         read_lock(&tasklist_lock);
2590         leader = top = top->group_leader;
2591 down:
2592         for_each_thread(leader, parent) {
2593                 list_for_each_entry(child, &parent->children, sibling) {
2594                         res = visitor(child, data);
2595                         if (res) {
2596                                 if (res < 0)
2597                                         goto out;
2598                                 leader = child;
2599                                 goto down;
2600                         }
2601 up:
2602                         ;
2603                 }
2604         }
2605
2606         if (leader != top) {
2607                 child = leader;
2608                 parent = child->real_parent;
2609                 leader = parent->group_leader;
2610                 goto up;
2611         }
2612 out:
2613         read_unlock(&tasklist_lock);
2614 }
2615
2616 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2617 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2618 #endif
2619
2620 static void sighand_ctor(void *data)
2621 {
2622         struct sighand_struct *sighand = data;
2623
2624         spin_lock_init(&sighand->siglock);
2625         init_waitqueue_head(&sighand->signalfd_wqh);
2626 }
2627
2628 void __init proc_caches_init(void)
2629 {
2630         unsigned int mm_size;
2631
2632         sighand_cachep = kmem_cache_create("sighand_cache",
2633                         sizeof(struct sighand_struct), 0,
2634                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2635                         SLAB_ACCOUNT, sighand_ctor);
2636         signal_cachep = kmem_cache_create("signal_cache",
2637                         sizeof(struct signal_struct), 0,
2638                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2639                         NULL);
2640         files_cachep = kmem_cache_create("files_cache",
2641                         sizeof(struct files_struct), 0,
2642                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2643                         NULL);
2644         fs_cachep = kmem_cache_create("fs_cache",
2645                         sizeof(struct fs_struct), 0,
2646                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2647                         NULL);
2648
2649         /*
2650          * The mm_cpumask is located at the end of mm_struct, and is
2651          * dynamically sized based on the maximum CPU number this system
2652          * can have, taking hotplug into account (nr_cpu_ids).
2653          */
2654         mm_size = sizeof(struct mm_struct) + cpumask_size();
2655
2656         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2657                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2658                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2659                         offsetof(struct mm_struct, saved_auxv),
2660                         sizeof_field(struct mm_struct, saved_auxv),
2661                         NULL);
2662         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2663         mmap_init();
2664         nsproxy_cache_init();
2665 }
2666
2667 /*
2668  * Check constraints on flags passed to the unshare system call.
2669  */
2670 static int check_unshare_flags(unsigned long unshare_flags)
2671 {
2672         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2673                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2674                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2675                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2676                 return -EINVAL;
2677         /*
2678          * Not implemented, but pretend it works if there is nothing
2679          * to unshare.  Note that unsharing the address space or the
2680          * signal handlers also need to unshare the signal queues (aka
2681          * CLONE_THREAD).
2682          */
2683         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2684                 if (!thread_group_empty(current))
2685                         return -EINVAL;
2686         }
2687         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2688                 if (refcount_read(&current->sighand->count) > 1)
2689                         return -EINVAL;
2690         }
2691         if (unshare_flags & CLONE_VM) {
2692                 if (!current_is_single_threaded())
2693                         return -EINVAL;
2694         }
2695
2696         return 0;
2697 }
2698
2699 /*
2700  * Unshare the filesystem structure if it is being shared
2701  */
2702 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2703 {
2704         struct fs_struct *fs = current->fs;
2705
2706         if (!(unshare_flags & CLONE_FS) || !fs)
2707                 return 0;
2708
2709         /* don't need lock here; in the worst case we'll do useless copy */
2710         if (fs->users == 1)
2711                 return 0;
2712
2713         *new_fsp = copy_fs_struct(fs);
2714         if (!*new_fsp)
2715                 return -ENOMEM;
2716
2717         return 0;
2718 }
2719
2720 /*
2721  * Unshare file descriptor table if it is being shared
2722  */
2723 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2724 {
2725         struct files_struct *fd = current->files;
2726         int error = 0;
2727
2728         if ((unshare_flags & CLONE_FILES) &&
2729             (fd && atomic_read(&fd->count) > 1)) {
2730                 *new_fdp = dup_fd(fd, &error);
2731                 if (!*new_fdp)
2732                         return error;
2733         }
2734
2735         return 0;
2736 }
2737
2738 /*
2739  * unshare allows a process to 'unshare' part of the process
2740  * context which was originally shared using clone.  copy_*
2741  * functions used by do_fork() cannot be used here directly
2742  * because they modify an inactive task_struct that is being
2743  * constructed. Here we are modifying the current, active,
2744  * task_struct.
2745  */
2746 int ksys_unshare(unsigned long unshare_flags)
2747 {
2748         struct fs_struct *fs, *new_fs = NULL;
2749         struct files_struct *fd, *new_fd = NULL;
2750         struct cred *new_cred = NULL;
2751         struct nsproxy *new_nsproxy = NULL;
2752         int do_sysvsem = 0;
2753         int err;
2754
2755         /*
2756          * If unsharing a user namespace must also unshare the thread group
2757          * and unshare the filesystem root and working directories.
2758          */
2759         if (unshare_flags & CLONE_NEWUSER)
2760                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2761         /*
2762          * If unsharing vm, must also unshare signal handlers.
2763          */
2764         if (unshare_flags & CLONE_VM)
2765                 unshare_flags |= CLONE_SIGHAND;
2766         /*
2767          * If unsharing a signal handlers, must also unshare the signal queues.
2768          */
2769         if (unshare_flags & CLONE_SIGHAND)
2770                 unshare_flags |= CLONE_THREAD;
2771         /*
2772          * If unsharing namespace, must also unshare filesystem information.
2773          */
2774         if (unshare_flags & CLONE_NEWNS)
2775                 unshare_flags |= CLONE_FS;
2776
2777         err = check_unshare_flags(unshare_flags);
2778         if (err)
2779                 goto bad_unshare_out;
2780         /*
2781          * CLONE_NEWIPC must also detach from the undolist: after switching
2782          * to a new ipc namespace, the semaphore arrays from the old
2783          * namespace are unreachable.
2784          */
2785         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2786                 do_sysvsem = 1;
2787         err = unshare_fs(unshare_flags, &new_fs);
2788         if (err)
2789                 goto bad_unshare_out;
2790         err = unshare_fd(unshare_flags, &new_fd);
2791         if (err)
2792                 goto bad_unshare_cleanup_fs;
2793         err = unshare_userns(unshare_flags, &new_cred);
2794         if (err)
2795                 goto bad_unshare_cleanup_fd;
2796         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2797                                          new_cred, new_fs);
2798         if (err)
2799                 goto bad_unshare_cleanup_cred;
2800
2801         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2802                 if (do_sysvsem) {
2803                         /*
2804                          * CLONE_SYSVSEM is equivalent to sys_exit().
2805                          */
2806                         exit_sem(current);
2807                 }
2808                 if (unshare_flags & CLONE_NEWIPC) {
2809                         /* Orphan segments in old ns (see sem above). */
2810                         exit_shm(current);
2811                         shm_init_task(current);
2812                 }
2813
2814                 if (new_nsproxy)
2815                         switch_task_namespaces(current, new_nsproxy);
2816
2817                 task_lock(current);
2818
2819                 if (new_fs) {
2820                         fs = current->fs;
2821                         spin_lock(&fs->lock);
2822                         current->fs = new_fs;
2823                         if (--fs->users)
2824                                 new_fs = NULL;
2825                         else
2826                                 new_fs = fs;
2827                         spin_unlock(&fs->lock);
2828                 }
2829
2830                 if (new_fd) {
2831                         fd = current->files;
2832                         current->files = new_fd;
2833                         new_fd = fd;
2834                 }
2835
2836                 task_unlock(current);
2837
2838                 if (new_cred) {
2839                         /* Install the new user namespace */
2840                         commit_creds(new_cred);
2841                         new_cred = NULL;
2842                 }
2843         }
2844
2845         perf_event_namespaces(current);
2846
2847 bad_unshare_cleanup_cred:
2848         if (new_cred)
2849                 put_cred(new_cred);
2850 bad_unshare_cleanup_fd:
2851         if (new_fd)
2852                 put_files_struct(new_fd);
2853
2854 bad_unshare_cleanup_fs:
2855         if (new_fs)
2856                 free_fs_struct(new_fs);
2857
2858 bad_unshare_out:
2859         return err;
2860 }
2861
2862 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2863 {
2864         return ksys_unshare(unshare_flags);
2865 }
2866
2867 /*
2868  *      Helper to unshare the files of the current task.
2869  *      We don't want to expose copy_files internals to
2870  *      the exec layer of the kernel.
2871  */
2872
2873 int unshare_files(struct files_struct **displaced)
2874 {
2875         struct task_struct *task = current;
2876         struct files_struct *copy = NULL;
2877         int error;
2878
2879         error = unshare_fd(CLONE_FILES, &copy);
2880         if (error || !copy) {
2881                 *displaced = NULL;
2882                 return error;
2883         }
2884         *displaced = task->files;
2885         task_lock(task);
2886         task->files = copy;
2887         task_unlock(task);
2888         return 0;
2889 }
2890
2891 int sysctl_max_threads(struct ctl_table *table, int write,
2892                        void __user *buffer, size_t *lenp, loff_t *ppos)
2893 {
2894         struct ctl_table t;
2895         int ret;
2896         int threads = max_threads;
2897         int min = MIN_THREADS;
2898         int max = MAX_THREADS;
2899
2900         t = *table;
2901         t.data = &threads;
2902         t.extra1 = &min;
2903         t.extra2 = &max;
2904
2905         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2906         if (ret || !write)
2907                 return ret;
2908
2909         set_max_threads(threads);
2910
2911         return 0;
2912 }