kernel/fork.c

   1 /*
   2  *  linux/kernel/fork.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  *  'fork.c' contains the help-routines for the 'fork' system call
   9  * (see also entry.S and others).
  10  * Fork is rather simple, once you get the hang of it, but the memory
  11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  12  */
  13
  14 #include <linux/slab.h>
  15 #include <linux/sched/autogroup.h>
  16 #include <linux/sched/mm.h>
  17 #include <linux/sched/coredump.h>
  18 #include <linux/sched/user.h>
  19 #include <linux/sched/numa_balancing.h>
  20 #include <linux/sched/stat.h>
  21 #include <linux/sched/task.h>
  22 #include <linux/sched/task_stack.h>
  23 #include <linux/sched/cputime.h>
  24 #include <linux/rtmutex.h>
  25 #include <linux/init.h>
  26 #include <linux/unistd.h>
  27 #include <linux/module.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/completion.h>
  30 #include <linux/personality.h>
  31 #include <linux/mempolicy.h>
  32 #include <linux/sem.h>
  33 #include <linux/file.h>
  34 #include <linux/fdtable.h>
  35 #include <linux/iocontext.h>
  36 #include <linux/key.h>
  37 #include <linux/binfmts.h>
  38 #include <linux/mman.h>
  39 #include <linux/mmu_notifier.h>
  40 #include <linux/hmm.h>
  41 #include <linux/fs.h>
  42 #include <linux/mm.h>
  43 #include <linux/vmacache.h>
  44 #include <linux/nsproxy.h>
  45 #include <linux/capability.h>
  46 #include <linux/cpu.h>
  47 #include <linux/cgroup.h>
  48 #include <linux/security.h>
  49 #include <linux/hugetlb.h>
  50 #include <linux/seccomp.h>
  51 #include <linux/swap.h>
  52 #include <linux/syscalls.h>
  53 #include <linux/jiffies.h>
  54 #include <linux/futex.h>
  55 #include <linux/compat.h>
  56 #include <linux/kthread.h>
  57 #include <linux/task_io_accounting_ops.h>
  58 #include <linux/rcupdate.h>
  59 #include <linux/ptrace.h>
  60 #include <linux/mount.h>
  61 #include <linux/audit.h>
  62 #include <linux/memcontrol.h>
  63 #include <linux/ftrace.h>
  64 #include <linux/proc_fs.h>
  65 #include <linux/profile.h>
  66 #include <linux/rmap.h>
  67 #include <linux/ksm.h>
  68 #include <linux/acct.h>
  69 #include <linux/userfaultfd_k.h>
  70 #include <linux/tsacct_kern.h>
  71 #include <linux/cn_proc.h>
  72 #include <linux/freezer.h>
  73 #include <linux/delayacct.h>
  74 #include <linux/taskstats_kern.h>
  75 #include <linux/random.h>
  76 #include <linux/tty.h>
  77 #include <linux/blkdev.h>
  78 #include <linux/fs_struct.h>
  79 #include <linux/magic.h>
  80 #include <linux/sched/mm.h>
  81 #include <linux/perf_event.h>
  82 #include <linux/posix-timers.h>
  83 #include <linux/user-return-notifier.h>
  84 #include <linux/oom.h>
  85 #include <linux/khugepaged.h>
  86 #include <linux/signalfd.h>
  87 #include <linux/uprobes.h>
  88 #include <linux/aio.h>
  89 #include <linux/compiler.h>
  90 #include <linux/sysctl.h>
  91 #include <linux/kcov.h>
  92 #include <linux/livepatch.h>
  93 #include <linux/thread_info.h>
  94
  95 #include <asm/pgtable.h>
  96 #include <asm/pgalloc.h>
  97 #include <linux/uaccess.h>
  98 #include <asm/mmu_context.h>
  99 #include <asm/cacheflush.h>
 100 #include <asm/tlbflush.h>
 101
 102 #include <trace/events/sched.h>
 103
 104 #define CREATE_TRACE_POINTS
 105 #include <trace/events/task.h>
 106
 107 /*
 108  * Minimum number of threads to boot the kernel
 109  */
 110 #define MIN_THREADS 20
 111
 112 /*
 113  * Maximum number of threads
 114  */
 115 #define MAX_THREADS FUTEX_TID_MASK
 116
 117 /*
 118  * Protected counters by write_lock_irq(&tasklist_lock)
 119  */
 120 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 121 int nr_threads;                 /* The idle threads do not count.. */
 122
 123 int max_threads;                /* tunable limit on nr_threads */
 124
 125 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 126
 127 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 128
 129 #ifdef CONFIG_PROVE_RCU
 130 int lockdep_tasklist_lock_is_held(void)
 131 {
 132         return lockdep_is_held(&tasklist_lock);
 133 }
 134 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 135 #endif /* #ifdef CONFIG_PROVE_RCU */
 136
 137 int nr_processes(void)
 138 {
 139         int cpu;
 140         int total = 0;
 141
 142         for_each_possible_cpu(cpu)
 143                 total += per_cpu(process_counts, cpu);
 144
 145         return total;
 146 }
 147
 148 void __weak arch_release_task_struct(struct task_struct *tsk)
 149 {
 150 }
 151
 152 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 153 static struct kmem_cache *task_struct_cachep;
 154
 155 static inline struct task_struct *alloc_task_struct_node(int node)
 156 {
 157         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 158 }
 159
 160 static inline void free_task_struct(struct task_struct *tsk)
 161 {
 162         kmem_cache_free(task_struct_cachep, tsk);
 163 }
 164 #endif
 165
 166 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 167
 168 /*
 169  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 170  * kmemcache based allocator.
 171  */
 172 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 173
 174 #ifdef CONFIG_VMAP_STACK
 175 /*
 176  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 177  * flush.  Try to minimize the number of calls by caching stacks.
 178  */
 179 #define NR_CACHED_STACKS 2
 180 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 181
 182 static int free_vm_stack_cache(unsigned int cpu)
 183 {
 184         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 185         int i;
 186
 187         for (i = 0; i < NR_CACHED_STACKS; i++) {
 188                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 189
 190                 if (!vm_stack)
 191                         continue;
 192
 193                 vfree(vm_stack->addr);
 194                 cached_vm_stacks[i] = NULL;
 195         }
 196
 197         return 0;
 198 }
 199 #endif
 200
 201 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 202 {
 203 #ifdef CONFIG_VMAP_STACK
 204         void *stack;
 205         int i;
 206
 207         for (i = 0; i < NR_CACHED_STACKS; i++) {
 208                 struct vm_struct *s;
 209
 210                 s = this_cpu_xchg(cached_stacks[i], NULL);
 211
 212                 if (!s)
 213                         continue;
 214
 215                 /* Clear stale pointers from reused stack. */
 216                 memset(s->addr, 0, THREAD_SIZE);
 217
 218                 tsk->stack_vm_area = s;
 219                 return s->addr;
 220         }
 221
 222         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 223                                      VMALLOC_START, VMALLOC_END,
 224                                      THREADINFO_GFP,
 225                                      PAGE_KERNEL,
 226                                      0, node, __builtin_return_address(0));
 227
 228         /*
 229          * We can't call find_vm_area() in interrupt context, and
 230          * free_thread_stack() can be called in interrupt context,
 231          * so cache the vm_struct.
 232          */
 233         if (stack)
 234                 tsk->stack_vm_area = find_vm_area(stack);
 235         return stack;
 236 #else
 237         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 238                                              THREAD_SIZE_ORDER);
 239
 240         return page ? page_address(page) : NULL;
 241 #endif
 242 }
 243
 244 static inline void free_thread_stack(struct task_struct *tsk)
 245 {
 246 #ifdef CONFIG_VMAP_STACK
 247         if (task_stack_vm_area(tsk)) {
 248                 int i;
 249
 250                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 251                         if (this_cpu_cmpxchg(cached_stacks[i],
 252                                         NULL, tsk->stack_vm_area) != NULL)
 253                                 continue;
 254
 255                         return;
 256                 }
 257
 258                 vfree_atomic(tsk->stack);
 259                 return;
 260         }
 261 #endif
 262
 263         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 264 }
 265 # else
 266 static struct kmem_cache *thread_stack_cache;
 267
 268 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 269                                                   int node)
 270 {
 271         return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 272 }
 273
 274 static void free_thread_stack(struct task_struct *tsk)
 275 {
 276         kmem_cache_free(thread_stack_cache, tsk->stack);
 277 }
 278
 279 void thread_stack_cache_init(void)
 280 {
 281         thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 282                                         THREAD_SIZE, THREAD_SIZE, 0, 0,
 283                                         THREAD_SIZE, NULL);
 284         BUG_ON(thread_stack_cache == NULL);
 285 }
 286 # endif
 287 #endif
 288
 289 /* SLAB cache for signal_struct structures (tsk->signal) */
 290 static struct kmem_cache *signal_cachep;
 291
 292 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 293 struct kmem_cache *sighand_cachep;
 294
 295 /* SLAB cache for files_struct structures (tsk->files) */
 296 struct kmem_cache *files_cachep;
 297
 298 /* SLAB cache for fs_struct structures (tsk->fs) */
 299 struct kmem_cache *fs_cachep;
 300
 301 /* SLAB cache for vm_area_struct structures */
 302 static struct kmem_cache *vm_area_cachep;
 303
 304 /* SLAB cache for mm_struct structures (tsk->mm) */
 305 static struct kmem_cache *mm_cachep;
 306
 307 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 308 {
 309         struct vm_area_struct *vma;
 310
 311         vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 312         if (vma)
 313                 vma_init(vma, mm);
 314         return vma;
 315 }
 316
 317 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 318 {
 319         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 320
 321         if (new) {
 322                 *new = *orig;
 323                 INIT_LIST_HEAD(&new->anon_vma_chain);
 324         }
 325         return new;
 326 }
 327
 328 void vm_area_free(struct vm_area_struct *vma)
 329 {
 330         kmem_cache_free(vm_area_cachep, vma);
 331 }
 332
 333 static void account_kernel_stack(struct task_struct *tsk, int account)
 334 {
 335         void *stack = task_stack_page(tsk);
 336         struct vm_struct *vm = task_stack_vm_area(tsk);
 337
 338         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 339
 340         if (vm) {
 341                 int i;
 342
 343                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 344
 345                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 346                         mod_zone_page_state(page_zone(vm->pages[i]),
 347                                             NR_KERNEL_STACK_KB,
 348                                             PAGE_SIZE / 1024 * account);
 349                 }
 350
 351                 /* All stack pages belong to the same memcg. */
 352                 mod_memcg_page_state(vm->pages[0], MEMCG_KERNEL_STACK_KB,
 353                                      account * (THREAD_SIZE / 1024));
 354         } else {
 355                 /*
 356                  * All stack pages are in the same zone and belong to the
 357                  * same memcg.
 358                  */
 359                 struct page *first_page = virt_to_page(stack);
 360
 361                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 362                                     THREAD_SIZE / 1024 * account);
 363
 364                 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
 365                                      account * (THREAD_SIZE / 1024));
 366         }
 367 }
 368
 369 static void release_task_stack(struct task_struct *tsk)
 370 {
 371         if (WARN_ON(tsk->state != TASK_DEAD))
 372                 return;  /* Better to leak the stack than to free prematurely */
 373
 374         account_kernel_stack(tsk, -1);
 375         free_thread_stack(tsk);
 376         tsk->stack = NULL;
 377 #ifdef CONFIG_VMAP_STACK
 378         tsk->stack_vm_area = NULL;
 379 #endif
 380 }
 381
 382 #ifdef CONFIG_THREAD_INFO_IN_TASK
 383 void put_task_stack(struct task_struct *tsk)
 384 {
 385         if (atomic_dec_and_test(&tsk->stack_refcount))
 386                 release_task_stack(tsk);
 387 }
 388 #endif
 389
 390 void free_task(struct task_struct *tsk)
 391 {
 392 #ifndef CONFIG_THREAD_INFO_IN_TASK
 393         /*
 394          * The task is finally done with both the stack and thread_info,
 395          * so free both.
 396          */
 397         release_task_stack(tsk);
 398 #else
 399         /*
 400          * If the task had a separate stack allocation, it should be gone
 401          * by now.
 402          */
 403         WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
 404 #endif
 405         rt_mutex_debug_task_free(tsk);
 406         ftrace_graph_exit_task(tsk);
 407         put_seccomp_filter(tsk);
 408         arch_release_task_struct(tsk);
 409         if (tsk->flags & PF_KTHREAD)
 410                 free_kthread_struct(tsk);
 411         free_task_struct(tsk);
 412 }
 413 EXPORT_SYMBOL(free_task);
 414
 415 #ifdef CONFIG_MMU
 416 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 417                                         struct mm_struct *oldmm)
 418 {
 419         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 420         struct rb_node **rb_link, *rb_parent;
 421         int retval;
 422         unsigned long charge;
 423         LIST_HEAD(uf);
 424
 425         uprobe_start_dup_mmap();
 426         if (down_write_killable(&oldmm->mmap_sem)) {
 427                 retval = -EINTR;
 428                 goto fail_uprobe_end;
 429         }
 430         flush_cache_dup_mm(oldmm);
 431         uprobe_dup_mmap(oldmm, mm);
 432         /*
 433          * Not linked in yet - no deadlock potential:
 434          */
 435         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 436
 437         /* No ordering required: file already has been exposed. */
 438         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 439
 440         mm->total_vm = oldmm->total_vm;
 441         mm->data_vm = oldmm->data_vm;
 442         mm->exec_vm = oldmm->exec_vm;
 443         mm->stack_vm = oldmm->stack_vm;
 444
 445         rb_link = &mm->mm_rb.rb_node;
 446         rb_parent = NULL;
 447         pprev = &mm->mmap;
 448         retval = ksm_fork(mm, oldmm);
 449         if (retval)
 450                 goto out;
 451         retval = khugepaged_fork(mm, oldmm);
 452         if (retval)
 453                 goto out;
 454
 455         prev = NULL;
 456         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 457                 struct file *file;
 458
 459                 if (mpnt->vm_flags & VM_DONTCOPY) {
 460                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 461                         continue;
 462                 }
 463                 charge = 0;
 464                 /*
 465                  * Don't duplicate many vmas if we've been oom-killed (for
 466                  * example)
 467                  */
 468                 if (fatal_signal_pending(current)) {
 469                         retval = -EINTR;
 470                         goto out;
 471                 }
 472                 if (mpnt->vm_flags & VM_ACCOUNT) {
 473                         unsigned long len = vma_pages(mpnt);
 474
 475                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 476                                 goto fail_nomem;
 477                         charge = len;
 478                 }
 479                 tmp = vm_area_dup(mpnt);
 480                 if (!tmp)
 481                         goto fail_nomem;
 482                 retval = vma_dup_policy(mpnt, tmp);
 483                 if (retval)
 484                         goto fail_nomem_policy;
 485                 tmp->vm_mm = mm;
 486                 retval = dup_userfaultfd(tmp, &uf);
 487                 if (retval)
 488                         goto fail_nomem_anon_vma_fork;
 489                 if (tmp->vm_flags & VM_WIPEONFORK) {
 490                         /* VM_WIPEONFORK gets a clean slate in the child. */
 491                         tmp->anon_vma = NULL;
 492                         if (anon_vma_prepare(tmp))
 493                                 goto fail_nomem_anon_vma_fork;
 494                 } else if (anon_vma_fork(tmp, mpnt))
 495                         goto fail_nomem_anon_vma_fork;
 496                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 497                 tmp->vm_next = tmp->vm_prev = NULL;
 498                 file = tmp->vm_file;
 499                 if (file) {
 500                         struct inode *inode = file_inode(file);
 501                         struct address_space *mapping = file->f_mapping;
 502
 503                         get_file(file);
 504                         if (tmp->vm_flags & VM_DENYWRITE)
 505                                 atomic_dec(&inode->i_writecount);
 506                         i_mmap_lock_write(mapping);
 507                         if (tmp->vm_flags & VM_SHARED)
 508                                 atomic_inc(&mapping->i_mmap_writable);
 509                         flush_dcache_mmap_lock(mapping);
 510                         /* insert tmp into the share list, just after mpnt */
 511                         vma_interval_tree_insert_after(tmp, mpnt,
 512                                         &mapping->i_mmap);
 513                         flush_dcache_mmap_unlock(mapping);
 514                         i_mmap_unlock_write(mapping);
 515                 }
 516
 517                 /*
 518                  * Clear hugetlb-related page reserves for children. This only
 519                  * affects MAP_PRIVATE mappings. Faults generated by the child
 520                  * are not guaranteed to succeed, even if read-only
 521                  */
 522                 if (is_vm_hugetlb_page(tmp))
 523                         reset_vma_resv_huge_pages(tmp);
 524
 525                 /*
 526                  * Link in the new vma and copy the page table entries.
 527                  */
 528                 *pprev = tmp;
 529                 pprev = &tmp->vm_next;
 530                 tmp->vm_prev = prev;
 531                 prev = tmp;
 532
 533                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 534                 rb_link = &tmp->vm_rb.rb_right;
 535                 rb_parent = &tmp->vm_rb;
 536
 537                 mm->map_count++;
 538                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 539                         retval = copy_page_range(mm, oldmm, mpnt);
 540
 541                 if (tmp->vm_ops && tmp->vm_ops->open)
 542                         tmp->vm_ops->open(tmp);
 543
 544                 if (retval)
 545                         goto out;
 546         }
 547         /* a new mm has just been created */
 548         retval = arch_dup_mmap(oldmm, mm);
 549 out:
 550         up_write(&mm->mmap_sem);
 551         flush_tlb_mm(oldmm);
 552         up_write(&oldmm->mmap_sem);
 553         dup_userfaultfd_complete(&uf);
 554 fail_uprobe_end:
 555         uprobe_end_dup_mmap();
 556         return retval;
 557 fail_nomem_anon_vma_fork:
 558         mpol_put(vma_policy(tmp));
 559 fail_nomem_policy:
 560         vm_area_free(tmp);
 561 fail_nomem:
 562         retval = -ENOMEM;
 563         vm_unacct_memory(charge);
 564         goto out;
 565 }
 566
 567 static inline int mm_alloc_pgd(struct mm_struct *mm)
 568 {
 569         mm->pgd = pgd_alloc(mm);
 570         if (unlikely(!mm->pgd))
 571                 return -ENOMEM;
 572         return 0;
 573 }
 574
 575 static inline void mm_free_pgd(struct mm_struct *mm)
 576 {
 577         pgd_free(mm, mm->pgd);
 578 }
 579 #else
 580 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 581 {
 582         down_write(&oldmm->mmap_sem);
 583         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 584         up_write(&oldmm->mmap_sem);
 585         return 0;
 586 }
 587 #define mm_alloc_pgd(mm)        (0)
 588 #define mm_free_pgd(mm)
 589 #endif /* CONFIG_MMU */
 590
 591 static void check_mm(struct mm_struct *mm)
 592 {
 593         int i;
 594
 595         for (i = 0; i < NR_MM_COUNTERS; i++) {
 596                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 597
 598                 if (unlikely(x))
 599                         printk(KERN_ALERT "BUG: Bad rss-counter state "
 600                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
 601         }
 602
 603         if (mm_pgtables_bytes(mm))
 604                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 605                                 mm_pgtables_bytes(mm));
 606
 607 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 608         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 609 #endif
 610 }
 611
 612 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 613 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 614
 615 /*
 616  * Called when the last reference to the mm
 617  * is dropped: either by a lazy thread or by
 618  * mmput. Free the page directory and the mm.
 619  */
 620 void __mmdrop(struct mm_struct *mm)
 621 {
 622         BUG_ON(mm == &init_mm);
 623         WARN_ON_ONCE(mm == current->mm);
 624         WARN_ON_ONCE(mm == current->active_mm);
 625         mm_free_pgd(mm);
 626         destroy_context(mm);
 627         hmm_mm_destroy(mm);
 628         mmu_notifier_mm_destroy(mm);
 629         check_mm(mm);
 630         put_user_ns(mm->user_ns);
 631         free_mm(mm);
 632 }
 633 EXPORT_SYMBOL_GPL(__mmdrop);
 634
 635 static void mmdrop_async_fn(struct work_struct *work)
 636 {
 637         struct mm_struct *mm;
 638
 639         mm = container_of(work, struct mm_struct, async_put_work);
 640         __mmdrop(mm);
 641 }
 642
 643 static void mmdrop_async(struct mm_struct *mm)
 644 {
 645         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 646                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 647                 schedule_work(&mm->async_put_work);
 648         }
 649 }
 650
 651 static inline void free_signal_struct(struct signal_struct *sig)
 652 {
 653         taskstats_tgid_free(sig);
 654         sched_autogroup_exit(sig);
 655         /*
 656          * __mmdrop is not safe to call from softirq context on x86 due to
 657          * pgd_dtor so postpone it to the async context
 658          */
 659         if (sig->oom_mm)
 660                 mmdrop_async(sig->oom_mm);
 661         kmem_cache_free(signal_cachep, sig);
 662 }
 663
 664 static inline void put_signal_struct(struct signal_struct *sig)
 665 {
 666         if (atomic_dec_and_test(&sig->sigcnt))
 667                 free_signal_struct(sig);
 668 }
 669
 670 void __put_task_struct(struct task_struct *tsk)
 671 {
 672         WARN_ON(!tsk->exit_state);
 673         WARN_ON(atomic_read(&tsk->usage));
 674         WARN_ON(tsk == current);
 675
 676         cgroup_free(tsk);
 677         task_numa_free(tsk, true);
 678         security_task_free(tsk);
 679         exit_creds(tsk);
 680         delayacct_tsk_free(tsk);
 681         put_signal_struct(tsk->signal);
 682
 683         if (!profile_handoff_task(tsk))
 684                 free_task(tsk);
 685 }
 686 EXPORT_SYMBOL_GPL(__put_task_struct);
 687
 688 void __init __weak arch_task_cache_init(void) { }
 689
 690 /*
 691  * set_max_threads
 692  */
 693 static void set_max_threads(unsigned int max_threads_suggested)
 694 {
 695         u64 threads;
 696
 697         /*
 698          * The number of threads shall be limited such that the thread
 699          * structures may only consume a small part of the available memory.
 700          */
 701         if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
 702                 threads = MAX_THREADS;
 703         else
 704                 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
 705                                     (u64) THREAD_SIZE * 8UL);
 706
 707         if (threads > max_threads_suggested)
 708                 threads = max_threads_suggested;
 709
 710         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 711 }
 712
 713 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 714 /* Initialized by the architecture: */
 715 int arch_task_struct_size __read_mostly;
 716 #endif
 717
 718 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 719 {
 720         /* Fetch thread_struct whitelist for the architecture. */
 721         arch_thread_struct_whitelist(offset, size);
 722
 723         /*
 724          * Handle zero-sized whitelist or empty thread_struct, otherwise
 725          * adjust offset to position of thread_struct in task_struct.
 726          */
 727         if (unlikely(*size == 0))
 728                 *offset = 0;
 729         else
 730                 *offset += offsetof(struct task_struct, thread);
 731 }
 732
 733 void __init fork_init(void)
 734 {
 735         int i;
 736 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 737 #ifndef ARCH_MIN_TASKALIGN
 738 #define ARCH_MIN_TASKALIGN      0
 739 #endif
 740         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 741         unsigned long useroffset, usersize;
 742
 743         /* create a slab on which task_structs can be allocated */
 744         task_struct_whitelist(&useroffset, &usersize);
 745         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 746                         arch_task_struct_size, align,
 747                         SLAB_PANIC|SLAB_ACCOUNT,
 748                         useroffset, usersize, NULL);
 749 #endif
 750
 751         /* do the arch specific task caches init */
 752         arch_task_cache_init();
 753
 754         set_max_threads(MAX_THREADS);
 755
 756         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 757         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 758         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 759                 init_task.signal->rlim[RLIMIT_NPROC];
 760
 761         for (i = 0; i < UCOUNT_COUNTS; i++) {
 762                 init_user_ns.ucount_max[i] = max_threads/2;
 763         }
 764
 765 #ifdef CONFIG_VMAP_STACK
 766         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 767                           NULL, free_vm_stack_cache);
 768 #endif
 769
 770         lockdep_init_task(&init_task);
 771 }
 772
 773 int __weak arch_dup_task_struct(struct task_struct *dst,
 774                                                struct task_struct *src)
 775 {
 776         *dst = *src;
 777         return 0;
 778 }
 779
 780 void set_task_stack_end_magic(struct task_struct *tsk)
 781 {
 782         unsigned long *stackend;
 783
 784         stackend = end_of_stack(tsk);
 785         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 786 }
 787
 788 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 789 {
 790         struct task_struct *tsk;
 791         unsigned long *stack;
 792         struct vm_struct *stack_vm_area;
 793         int err;
 794
 795         if (node == NUMA_NO_NODE)
 796                 node = tsk_fork_get_node(orig);
 797         tsk = alloc_task_struct_node(node);
 798         if (!tsk)
 799                 return NULL;
 800
 801         stack = alloc_thread_stack_node(tsk, node);
 802         if (!stack)
 803                 goto free_tsk;
 804
 805         stack_vm_area = task_stack_vm_area(tsk);
 806
 807         err = arch_dup_task_struct(tsk, orig);
 808
 809         /*
 810          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 811          * sure they're properly initialized before using any stack-related
 812          * functions again.
 813          */
 814         tsk->stack = stack;
 815 #ifdef CONFIG_VMAP_STACK
 816         tsk->stack_vm_area = stack_vm_area;
 817 #endif
 818 #ifdef CONFIG_THREAD_INFO_IN_TASK
 819         atomic_set(&tsk->stack_refcount, 1);
 820 #endif
 821
 822         if (err)
 823                 goto free_stack;
 824
 825 #ifdef CONFIG_SECCOMP
 826         /*
 827          * We must handle setting up seccomp filters once we're under
 828          * the sighand lock in case orig has changed between now and
 829          * then. Until then, filter must be NULL to avoid messing up
 830          * the usage counts on the error path calling free_task.
 831          */
 832         tsk->seccomp.filter = NULL;
 833 #endif
 834
 835         setup_thread_stack(tsk, orig);
 836         clear_user_return_notifier(tsk);
 837         clear_tsk_need_resched(tsk);
 838         set_task_stack_end_magic(tsk);
 839
 840 #ifdef CONFIG_STACKPROTECTOR
 841         tsk->stack_canary = get_random_canary();
 842 #endif
 843
 844         /*
 845          * One for us, one for whoever does the "release_task()" (usually
 846          * parent)
 847          */
 848         atomic_set(&tsk->usage, 2);
 849 #ifdef CONFIG_BLK_DEV_IO_TRACE
 850         tsk->btrace_seq = 0;
 851 #endif
 852         tsk->splice_pipe = NULL;
 853         tsk->task_frag.page = NULL;
 854         tsk->wake_q.next = NULL;
 855
 856         account_kernel_stack(tsk, 1);
 857
 858         kcov_task_init(tsk);
 859
 860 #ifdef CONFIG_FAULT_INJECTION
 861         tsk->fail_nth = 0;
 862 #endif
 863
 864 #ifdef CONFIG_BLK_CGROUP
 865         tsk->throttle_queue = NULL;
 866         tsk->use_memdelay = 0;
 867 #endif
 868
 869 #ifdef CONFIG_MEMCG
 870         tsk->active_memcg = NULL;
 871 #endif
 872         return tsk;
 873
 874 free_stack:
 875         free_thread_stack(tsk);
 876 free_tsk:
 877         free_task_struct(tsk);
 878         return NULL;
 879 }
 880
 881 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 882
 883 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 884
 885 static int __init coredump_filter_setup(char *s)
 886 {
 887         default_dump_filter =
 888                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 889                 MMF_DUMP_FILTER_MASK;
 890         return 1;
 891 }
 892
 893 __setup("coredump_filter=", coredump_filter_setup);
 894
 895 #include <linux/init_task.h>
 896
 897 static void mm_init_aio(struct mm_struct *mm)
 898 {
 899 #ifdef CONFIG_AIO
 900         spin_lock_init(&mm->ioctx_lock);
 901         mm->ioctx_table = NULL;
 902 #endif
 903 }
 904
 905 static __always_inline void mm_clear_owner(struct mm_struct *mm,
 906                                            struct task_struct *p)
 907 {
 908 #ifdef CONFIG_MEMCG
 909         if (mm->owner == p)
 910                 WRITE_ONCE(mm->owner, NULL);
 911 #endif
 912 }
 913
 914 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 915 {
 916 #ifdef CONFIG_MEMCG
 917         mm->owner = p;
 918 #endif
 919 }
 920
 921 static void mm_init_uprobes_state(struct mm_struct *mm)
 922 {
 923 #ifdef CONFIG_UPROBES
 924         mm->uprobes_state.xol_area = NULL;
 925 #endif
 926 }
 927
 928 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 929         struct user_namespace *user_ns)
 930 {
 931         mm->mmap = NULL;
 932         mm->mm_rb = RB_ROOT;
 933         mm->vmacache_seqnum = 0;
 934         atomic_set(&mm->mm_users, 1);
 935         atomic_set(&mm->mm_count, 1);
 936         init_rwsem(&mm->mmap_sem);
 937         INIT_LIST_HEAD(&mm->mmlist);
 938         mm->core_state = NULL;
 939         mm_pgtables_bytes_init(mm);
 940         mm->map_count = 0;
 941         mm->locked_vm = 0;
 942         mm->pinned_vm = 0;
 943         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
 944         spin_lock_init(&mm->page_table_lock);
 945         spin_lock_init(&mm->arg_lock);
 946         mm_init_cpumask(mm);
 947         mm_init_aio(mm);
 948         mm_init_owner(mm, p);
 949         RCU_INIT_POINTER(mm->exe_file, NULL);
 950         mmu_notifier_mm_init(mm);
 951         hmm_mm_init(mm);
 952         init_tlb_flush_pending(mm);
 953 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 954         mm->pmd_huge_pte = NULL;
 955 #endif
 956         mm_init_uprobes_state(mm);
 957
 958         if (current->mm) {
 959                 mm->flags = current->mm->flags & MMF_INIT_MASK;
 960                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
 961         } else {
 962                 mm->flags = default_dump_filter;
 963                 mm->def_flags = 0;
 964         }
 965
 966         if (mm_alloc_pgd(mm))
 967                 goto fail_nopgd;
 968
 969         if (init_new_context(p, mm))
 970                 goto fail_nocontext;
 971
 972         mm->user_ns = get_user_ns(user_ns);
 973         return mm;
 974
 975 fail_nocontext:
 976         mm_free_pgd(mm);
 977 fail_nopgd:
 978         free_mm(mm);
 979         return NULL;
 980 }
 981
 982 /*
 983  * Allocate and initialize an mm_struct.
 984  */
 985 struct mm_struct *mm_alloc(void)
 986 {
 987         struct mm_struct *mm;
 988
 989         mm = allocate_mm();
 990         if (!mm)
 991                 return NULL;
 992
 993         memset(mm, 0, sizeof(*mm));
 994         return mm_init(mm, current, current_user_ns());
 995 }
 996
 997 static inline void __mmput(struct mm_struct *mm)
 998 {
 999         VM_BUG_ON(atomic_read(&mm->mm_users));
1000
1001         uprobe_clear_state(mm);
1002         exit_aio(mm);
1003         ksm_exit(mm);
1004         khugepaged_exit(mm); /* must run before exit_mmap */
1005         exit_mmap(mm);
1006         mm_put_huge_zero_page(mm);
1007         set_mm_exe_file(mm, NULL);
1008         if (!list_empty(&mm->mmlist)) {
1009                 spin_lock(&mmlist_lock);
1010                 list_del(&mm->mmlist);
1011                 spin_unlock(&mmlist_lock);
1012         }
1013         if (mm->binfmt)
1014                 module_put(mm->binfmt->module);
1015         mmdrop(mm);
1016 }
1017
1018 /*
1019  * Decrement the use count and release all resources for an mm.
1020  */
1021 void mmput(struct mm_struct *mm)
1022 {
1023         might_sleep();
1024
1025         if (atomic_dec_and_test(&mm->mm_users))
1026                 __mmput(mm);
1027 }
1028 EXPORT_SYMBOL_GPL(mmput);
1029
1030 #ifdef CONFIG_MMU
1031 static void mmput_async_fn(struct work_struct *work)
1032 {
1033         struct mm_struct *mm = container_of(work, struct mm_struct,
1034                                             async_put_work);
1035
1036         __mmput(mm);
1037 }
1038
1039 void mmput_async(struct mm_struct *mm)
1040 {
1041         if (atomic_dec_and_test(&mm->mm_users)) {
1042                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1043                 schedule_work(&mm->async_put_work);
1044         }
1045 }
1046 #endif
1047
1048 /**
1049  * set_mm_exe_file - change a reference to the mm's executable file
1050  *
1051  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1052  *
1053  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1054  * invocations: in mmput() nobody alive left, in execve task is single
1055  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1056  * mm->exe_file, but does so without using set_mm_exe_file() in order
1057  * to do avoid the need for any locks.
1058  */
1059 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1060 {
1061         struct file *old_exe_file;
1062
1063         /*
1064          * It is safe to dereference the exe_file without RCU as
1065          * this function is only called if nobody else can access
1066          * this mm -- see comment above for justification.
1067          */
1068         old_exe_file = rcu_dereference_raw(mm->exe_file);
1069
1070         if (new_exe_file)
1071                 get_file(new_exe_file);
1072         rcu_assign_pointer(mm->exe_file, new_exe_file);
1073         if (old_exe_file)
1074                 fput(old_exe_file);
1075 }
1076
1077 /**
1078  * get_mm_exe_file - acquire a reference to the mm's executable file
1079  *
1080  * Returns %NULL if mm has no associated executable file.
1081  * User must release file via fput().
1082  */
1083 struct file *get_mm_exe_file(struct mm_struct *mm)
1084 {
1085         struct file *exe_file;
1086
1087         rcu_read_lock();
1088         exe_file = rcu_dereference(mm->exe_file);
1089         if (exe_file && !get_file_rcu(exe_file))
1090                 exe_file = NULL;
1091         rcu_read_unlock();
1092         return exe_file;
1093 }
1094 EXPORT_SYMBOL(get_mm_exe_file);
1095
1096 /**
1097  * get_task_exe_file - acquire a reference to the task's executable file
1098  *
1099  * Returns %NULL if task's mm (if any) has no associated executable file or
1100  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1101  * User must release file via fput().
1102  */
1103 struct file *get_task_exe_file(struct task_struct *task)
1104 {
1105         struct file *exe_file = NULL;
1106         struct mm_struct *mm;
1107
1108         task_lock(task);
1109         mm = task->mm;
1110         if (mm) {
1111                 if (!(task->flags & PF_KTHREAD))
1112                         exe_file = get_mm_exe_file(mm);
1113         }
1114         task_unlock(task);
1115         return exe_file;
1116 }
1117 EXPORT_SYMBOL(get_task_exe_file);
1118
1119 /**
1120  * get_task_mm - acquire a reference to the task's mm
1121  *
1122  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1123  * this kernel workthread has transiently adopted a user mm with use_mm,
1124  * to do its AIO) is not set and if so returns a reference to it, after
1125  * bumping up the use count.  User must release the mm via mmput()
1126  * after use.  Typically used by /proc and ptrace.
1127  */
1128 struct mm_struct *get_task_mm(struct task_struct *task)
1129 {
1130         struct mm_struct *mm;
1131
1132         task_lock(task);
1133         mm = task->mm;
1134         if (mm) {
1135                 if (task->flags & PF_KTHREAD)
1136                         mm = NULL;
1137                 else
1138                         mmget(mm);
1139         }
1140         task_unlock(task);
1141         return mm;
1142 }
1143 EXPORT_SYMBOL_GPL(get_task_mm);
1144
1145 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1146 {
1147         struct mm_struct *mm;
1148         int err;
1149
1150         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1151         if (err)
1152                 return ERR_PTR(err);
1153
1154         mm = get_task_mm(task);
1155         if (mm && mm != current->mm &&
1156                         !ptrace_may_access(task, mode)) {
1157                 mmput(mm);
1158                 mm = ERR_PTR(-EACCES);
1159         }
1160         mutex_unlock(&task->signal->cred_guard_mutex);
1161
1162         return mm;
1163 }
1164
1165 static void complete_vfork_done(struct task_struct *tsk)
1166 {
1167         struct completion *vfork;
1168
1169         task_lock(tsk);
1170         vfork = tsk->vfork_done;
1171         if (likely(vfork)) {
1172                 tsk->vfork_done = NULL;
1173                 complete(vfork);
1174         }
1175         task_unlock(tsk);
1176 }
1177
1178 static int wait_for_vfork_done(struct task_struct *child,
1179                                 struct completion *vfork)
1180 {
1181         int killed;
1182
1183         freezer_do_not_count();
1184         killed = wait_for_completion_killable(vfork);
1185         freezer_count();
1186
1187         if (killed) {
1188                 task_lock(child);
1189                 child->vfork_done = NULL;
1190                 task_unlock(child);
1191         }
1192
1193         put_task_struct(child);
1194         return killed;
1195 }
1196
1197 /* Please note the differences between mmput and mm_release.
1198  * mmput is called whenever we stop holding onto a mm_struct,
1199  * error success whatever.
1200  *
1201  * mm_release is called after a mm_struct has been removed
1202  * from the current process.
1203  *
1204  * This difference is important for error handling, when we
1205  * only half set up a mm_struct for a new process and need to restore
1206  * the old one.  Because we mmput the new mm_struct before
1207  * restoring the old one. . .
1208  * Eric Biederman 10 January 1998
1209  */
1210 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1211 {
1212         /* Get rid of any futexes when releasing the mm */
1213 #ifdef CONFIG_FUTEX
1214         if (unlikely(tsk->robust_list)) {
1215                 exit_robust_list(tsk);
1216                 tsk->robust_list = NULL;
1217         }
1218 #ifdef CONFIG_COMPAT
1219         if (unlikely(tsk->compat_robust_list)) {
1220                 compat_exit_robust_list(tsk);
1221                 tsk->compat_robust_list = NULL;
1222         }
1223 #endif
1224         if (unlikely(!list_empty(&tsk->pi_state_list)))
1225                 exit_pi_state_list(tsk);
1226 #endif
1227
1228         uprobe_free_utask(tsk);
1229
1230         /* Get rid of any cached register state */
1231         deactivate_mm(tsk, mm);
1232
1233         /*
1234          * Signal userspace if we're not exiting with a core dump
1235          * because we want to leave the value intact for debugging
1236          * purposes.
1237          */
1238         if (tsk->clear_child_tid) {
1239                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1240                     atomic_read(&mm->mm_users) > 1) {
1241                         /*
1242                          * We don't check the error code - if userspace has
1243                          * not set up a proper pointer then tough luck.
1244                          */
1245                         put_user(0, tsk->clear_child_tid);
1246                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1247                                         1, NULL, NULL, 0, 0);
1248                 }
1249                 tsk->clear_child_tid = NULL;
1250         }
1251
1252         /*
1253          * All done, finally we can wake up parent and return this mm to him.
1254          * Also kthread_stop() uses this completion for synchronization.
1255          */
1256         if (tsk->vfork_done)
1257                 complete_vfork_done(tsk);
1258 }
1259
1260 /*
1261  * Allocate a new mm structure and copy contents from the
1262  * mm structure of the passed in task structure.
1263  */
1264 static struct mm_struct *dup_mm(struct task_struct *tsk)
1265 {
1266         struct mm_struct *mm, *oldmm = current->mm;
1267         int err;
1268
1269         mm = allocate_mm();
1270         if (!mm)
1271                 goto fail_nomem;
1272
1273         memcpy(mm, oldmm, sizeof(*mm));
1274
1275         if (!mm_init(mm, tsk, mm->user_ns))
1276                 goto fail_nomem;
1277
1278         err = dup_mmap(mm, oldmm);
1279         if (err)
1280                 goto free_pt;
1281
1282         mm->hiwater_rss = get_mm_rss(mm);
1283         mm->hiwater_vm = mm->total_vm;
1284
1285         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1286                 goto free_pt;
1287
1288         return mm;
1289
1290 free_pt:
1291         /* don't put binfmt in mmput, we haven't got module yet */
1292         mm->binfmt = NULL;
1293         mm_init_owner(mm, NULL);
1294         mmput(mm);
1295
1296 fail_nomem:
1297         return NULL;
1298 }
1299
1300 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1301 {
1302         struct mm_struct *mm, *oldmm;
1303         int retval;
1304
1305         tsk->min_flt = tsk->maj_flt = 0;
1306         tsk->nvcsw = tsk->nivcsw = 0;
1307 #ifdef CONFIG_DETECT_HUNG_TASK
1308         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1309         tsk->last_switch_time = 0;
1310 #endif
1311
1312         tsk->mm = NULL;
1313         tsk->active_mm = NULL;
1314
1315         /*
1316          * Are we cloning a kernel thread?
1317          *
1318          * We need to steal a active VM for that..
1319          */
1320         oldmm = current->mm;
1321         if (!oldmm)
1322                 return 0;
1323
1324         /* initialize the new vmacache entries */
1325         vmacache_flush(tsk);
1326
1327         if (clone_flags & CLONE_VM) {
1328                 mmget(oldmm);
1329                 mm = oldmm;
1330                 goto good_mm;
1331         }
1332
1333         retval = -ENOMEM;
1334         mm = dup_mm(tsk);
1335         if (!mm)
1336                 goto fail_nomem;
1337
1338 good_mm:
1339         tsk->mm = mm;
1340         tsk->active_mm = mm;
1341         return 0;
1342
1343 fail_nomem:
1344         return retval;
1345 }
1346
1347 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1348 {
1349         struct fs_struct *fs = current->fs;
1350         if (clone_flags & CLONE_FS) {
1351                 /* tsk->fs is already what we want */
1352                 spin_lock(&fs->lock);
1353                 if (fs->in_exec) {
1354                         spin_unlock(&fs->lock);
1355                         return -EAGAIN;
1356                 }
1357                 fs->users++;
1358                 spin_unlock(&fs->lock);
1359                 return 0;
1360         }
1361         tsk->fs = copy_fs_struct(fs);
1362         if (!tsk->fs)
1363                 return -ENOMEM;
1364         return 0;
1365 }
1366
1367 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1368 {
1369         struct files_struct *oldf, *newf;
1370         int error = 0;
1371
1372         /*
1373          * A background process may not have any files ...
1374          */
1375         oldf = current->files;
1376         if (!oldf)
1377                 goto out;
1378
1379         if (clone_flags & CLONE_FILES) {
1380                 atomic_inc(&oldf->count);
1381                 goto out;
1382         }
1383
1384         newf = dup_fd(oldf, &error);
1385         if (!newf)
1386                 goto out;
1387
1388         tsk->files = newf;
1389         error = 0;
1390 out:
1391         return error;
1392 }
1393
1394 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1395 {
1396 #ifdef CONFIG_BLOCK
1397         struct io_context *ioc = current->io_context;
1398         struct io_context *new_ioc;
1399
1400         if (!ioc)
1401                 return 0;
1402         /*
1403          * Share io context with parent, if CLONE_IO is set
1404          */
1405         if (clone_flags & CLONE_IO) {
1406                 ioc_task_link(ioc);
1407                 tsk->io_context = ioc;
1408         } else if (ioprio_valid(ioc->ioprio)) {
1409                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1410                 if (unlikely(!new_ioc))
1411                         return -ENOMEM;
1412
1413                 new_ioc->ioprio = ioc->ioprio;
1414                 put_io_context(new_ioc);
1415         }
1416 #endif
1417         return 0;
1418 }
1419
1420 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1421 {
1422         struct sighand_struct *sig;
1423
1424         if (clone_flags & CLONE_SIGHAND) {
1425                 atomic_inc(&current->sighand->count);
1426                 return 0;
1427         }
1428         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1429         rcu_assign_pointer(tsk->sighand, sig);
1430         if (!sig)
1431                 return -ENOMEM;
1432
1433         atomic_set(&sig->count, 1);
1434         spin_lock_irq(&current->sighand->siglock);
1435         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1436         spin_unlock_irq(&current->sighand->siglock);
1437         return 0;
1438 }
1439
1440 void __cleanup_sighand(struct sighand_struct *sighand)
1441 {
1442         if (atomic_dec_and_test(&sighand->count)) {
1443                 signalfd_cleanup(sighand);
1444                 /*
1445                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1446                  * without an RCU grace period, see __lock_task_sighand().
1447                  */
1448                 kmem_cache_free(sighand_cachep, sighand);
1449         }
1450 }
1451
1452 #ifdef CONFIG_POSIX_TIMERS
1453 /*
1454  * Initialize POSIX timer handling for a thread group.
1455  */
1456 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1457 {
1458         unsigned long cpu_limit;
1459
1460         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1461         if (cpu_limit != RLIM_INFINITY) {
1462                 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1463                 sig->cputimer.running = true;
1464         }
1465
1466         /* The timer lists. */
1467         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1468         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1469         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1470 }
1471 #else
1472 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1473 #endif
1474
1475 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1476 {
1477         struct signal_struct *sig;
1478
1479         if (clone_flags & CLONE_THREAD)
1480                 return 0;
1481
1482         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1483         tsk->signal = sig;
1484         if (!sig)
1485                 return -ENOMEM;
1486
1487         sig->nr_threads = 1;
1488         atomic_set(&sig->live, 1);
1489         atomic_set(&sig->sigcnt, 1);
1490
1491         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1492         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1493         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1494
1495         init_waitqueue_head(&sig->wait_chldexit);
1496         sig->curr_target = tsk;
1497         init_sigpending(&sig->shared_pending);
1498         INIT_HLIST_HEAD(&sig->multiprocess);
1499         seqlock_init(&sig->stats_lock);
1500         prev_cputime_init(&sig->prev_cputime);
1501
1502 #ifdef CONFIG_POSIX_TIMERS
1503         INIT_LIST_HEAD(&sig->posix_timers);
1504         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1505         sig->real_timer.function = it_real_fn;
1506 #endif
1507
1508         task_lock(current->group_leader);
1509         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1510         task_unlock(current->group_leader);
1511
1512         posix_cpu_timers_init_group(sig);
1513
1514         tty_audit_fork(sig);
1515         sched_autogroup_fork(sig);
1516
1517         sig->oom_score_adj = current->signal->oom_score_adj;
1518         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1519
1520         mutex_init(&sig->cred_guard_mutex);
1521
1522         return 0;
1523 }
1524
1525 static void copy_seccomp(struct task_struct *p)
1526 {
1527 #ifdef CONFIG_SECCOMP
1528         /*
1529          * Must be called with sighand->lock held, which is common to
1530          * all threads in the group. Holding cred_guard_mutex is not
1531          * needed because this new task is not yet running and cannot
1532          * be racing exec.
1533          */
1534         assert_spin_locked(&current->sighand->siglock);
1535
1536         /* Ref-count the new filter user, and assign it. */
1537         get_seccomp_filter(current);
1538         p->seccomp = current->seccomp;
1539
1540         /*
1541          * Explicitly enable no_new_privs here in case it got set
1542          * between the task_struct being duplicated and holding the
1543          * sighand lock. The seccomp state and nnp must be in sync.
1544          */
1545         if (task_no_new_privs(current))
1546                 task_set_no_new_privs(p);
1547
1548         /*
1549          * If the parent gained a seccomp mode after copying thread
1550          * flags and between before we held the sighand lock, we have
1551          * to manually enable the seccomp thread flag here.
1552          */
1553         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1554                 set_tsk_thread_flag(p, TIF_SECCOMP);
1555 #endif
1556 }
1557
1558 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1559 {
1560         current->clear_child_tid = tidptr;
1561
1562         return task_pid_vnr(current);
1563 }
1564
1565 static void rt_mutex_init_task(struct task_struct *p)
1566 {
1567         raw_spin_lock_init(&p->pi_lock);
1568 #ifdef CONFIG_RT_MUTEXES
1569         p->pi_waiters = RB_ROOT_CACHED;
1570         p->pi_top_task = NULL;
1571         p->pi_blocked_on = NULL;
1572 #endif
1573 }
1574
1575 #ifdef CONFIG_POSIX_TIMERS
1576 /*
1577  * Initialize POSIX timer handling for a single task.
1578  */
1579 static void posix_cpu_timers_init(struct task_struct *tsk)
1580 {
1581         tsk->cputime_expires.prof_exp = 0;
1582         tsk->cputime_expires.virt_exp = 0;
1583         tsk->cputime_expires.sched_exp = 0;
1584         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1585         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1586         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1587 }
1588 #else
1589 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1590 #endif
1591
1592 static inline void init_task_pid_links(struct task_struct *task)
1593 {
1594         enum pid_type type;
1595
1596         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1597                 INIT_HLIST_NODE(&task->pid_links[type]);
1598         }
1599 }
1600
1601 static inline void
1602 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1603 {
1604         if (type == PIDTYPE_PID)
1605                 task->thread_pid = pid;
1606         else
1607                 task->signal->pids[type] = pid;
1608 }
1609
1610 static inline void rcu_copy_process(struct task_struct *p)
1611 {
1612 #ifdef CONFIG_PREEMPT_RCU
1613         p->rcu_read_lock_nesting = 0;
1614         p->rcu_read_unlock_special.s = 0;
1615         p->rcu_blocked_node = NULL;
1616         INIT_LIST_HEAD(&p->rcu_node_entry);
1617 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1618 #ifdef CONFIG_TASKS_RCU
1619         p->rcu_tasks_holdout = false;
1620         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1621         p->rcu_tasks_idle_cpu = -1;
1622 #endif /* #ifdef CONFIG_TASKS_RCU */
1623 }
1624
1625 static void __delayed_free_task(struct rcu_head *rhp)
1626 {
1627         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1628
1629         free_task(tsk);
1630 }
1631
1632 static __always_inline void delayed_free_task(struct task_struct *tsk)
1633 {
1634         if (IS_ENABLED(CONFIG_MEMCG))
1635                 call_rcu(&tsk->rcu, __delayed_free_task);
1636         else
1637                 free_task(tsk);
1638 }
1639
1640 /*
1641  * This creates a new process as a copy of the old one,
1642  * but does not actually start it yet.
1643  *
1644  * It copies the registers, and all the appropriate
1645  * parts of the process environment (as per the clone
1646  * flags). The actual kick-off is left to the caller.
1647  */
1648 static __latent_entropy struct task_struct *copy_process(
1649                                         unsigned long clone_flags,
1650                                         unsigned long stack_start,
1651                                         unsigned long stack_size,
1652                                         int __user *child_tidptr,
1653                                         struct pid *pid,
1654                                         int trace,
1655                                         unsigned long tls,
1656                                         int node)
1657 {
1658         int retval;
1659         struct task_struct *p;
1660         struct multiprocess_signals delayed;
1661
1662         /*
1663          * Don't allow sharing the root directory with processes in a different
1664          * namespace
1665          */
1666         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1667                 return ERR_PTR(-EINVAL);
1668
1669         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1670                 return ERR_PTR(-EINVAL);
1671
1672         /*
1673          * Thread groups must share signals as well, and detached threads
1674          * can only be started up within the thread group.
1675          */
1676         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1677                 return ERR_PTR(-EINVAL);
1678
1679         /*
1680          * Shared signal handlers imply shared VM. By way of the above,
1681          * thread groups also imply shared VM. Blocking this case allows
1682          * for various simplifications in other code.
1683          */
1684         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1685                 return ERR_PTR(-EINVAL);
1686
1687         /*
1688          * Siblings of global init remain as zombies on exit since they are
1689          * not reaped by their parent (swapper). To solve this and to avoid
1690          * multi-rooted process trees, prevent global and container-inits
1691          * from creating siblings.
1692          */
1693         if ((clone_flags & CLONE_PARENT) &&
1694                                 current->signal->flags & SIGNAL_UNKILLABLE)
1695                 return ERR_PTR(-EINVAL);
1696
1697         /*
1698          * If the new process will be in a different pid or user namespace
1699          * do not allow it to share a thread group with the forking task.
1700          */
1701         if (clone_flags & CLONE_THREAD) {
1702                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1703                     (task_active_pid_ns(current) !=
1704                                 current->nsproxy->pid_ns_for_children))
1705                         return ERR_PTR(-EINVAL);
1706         }
1707
1708         /*
1709          * Force any signals received before this point to be delivered
1710          * before the fork happens.  Collect up signals sent to multiple
1711          * processes that happen during the fork and delay them so that
1712          * they appear to happen after the fork.
1713          */
1714         sigemptyset(&delayed.signal);
1715         INIT_HLIST_NODE(&delayed.node);
1716
1717         spin_lock_irq(&current->sighand->siglock);
1718         if (!(clone_flags & CLONE_THREAD))
1719                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1720         recalc_sigpending();
1721         spin_unlock_irq(&current->sighand->siglock);
1722         retval = -ERESTARTNOINTR;
1723         if (signal_pending(current))
1724                 goto fork_out;
1725
1726         retval = -ENOMEM;
1727         p = dup_task_struct(current, node);
1728         if (!p)
1729                 goto fork_out;
1730
1731         /*
1732          * This _must_ happen before we call free_task(), i.e. before we jump
1733          * to any of the bad_fork_* labels. This is to avoid freeing
1734          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1735          * kernel threads (PF_KTHREAD).
1736          */
1737         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1738         /*
1739          * Clear TID on mm_release()?
1740          */
1741         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1742
1743         ftrace_graph_init_task(p);
1744
1745         rt_mutex_init_task(p);
1746
1747 #ifdef CONFIG_PROVE_LOCKING
1748         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1749         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1750 #endif
1751         retval = -EAGAIN;
1752         if (atomic_read(&p->real_cred->user->processes) >=
1753                         task_rlimit(p, RLIMIT_NPROC)) {
1754                 if (p->real_cred->user != INIT_USER &&
1755                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1756                         goto bad_fork_free;
1757         }
1758         current->flags &= ~PF_NPROC_EXCEEDED;
1759
1760         retval = copy_creds(p, clone_flags);
1761         if (retval < 0)
1762                 goto bad_fork_free;
1763
1764         /*
1765          * If multiple threads are within copy_process(), then this check
1766          * triggers too late. This doesn't hurt, the check is only there
1767          * to stop root fork bombs.
1768          */
1769         retval = -EAGAIN;
1770         if (nr_threads >= max_threads)
1771                 goto bad_fork_cleanup_count;
1772
1773         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1774         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1775         p->flags |= PF_FORKNOEXEC;
1776         INIT_LIST_HEAD(&p->children);
1777         INIT_LIST_HEAD(&p->sibling);
1778         rcu_copy_process(p);
1779         p->vfork_done = NULL;
1780         spin_lock_init(&p->alloc_lock);
1781
1782         init_sigpending(&p->pending);
1783
1784         p->utime = p->stime = p->gtime = 0;
1785 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1786         p->utimescaled = p->stimescaled = 0;
1787 #endif
1788         prev_cputime_init(&p->prev_cputime);
1789
1790 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1791         seqcount_init(&p->vtime.seqcount);
1792         p->vtime.starttime = 0;
1793         p->vtime.state = VTIME_INACTIVE;
1794 #endif
1795
1796 #if defined(SPLIT_RSS_COUNTING)
1797         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1798 #endif
1799
1800         p->default_timer_slack_ns = current->timer_slack_ns;
1801
1802         task_io_accounting_init(&p->ioac);
1803         acct_clear_integrals(p);
1804
1805         posix_cpu_timers_init(p);
1806
1807         p->io_context = NULL;
1808         audit_set_context(p, NULL);
1809         cgroup_fork(p);
1810 #ifdef CONFIG_NUMA
1811         p->mempolicy = mpol_dup(p->mempolicy);
1812         if (IS_ERR(p->mempolicy)) {
1813                 retval = PTR_ERR(p->mempolicy);
1814                 p->mempolicy = NULL;
1815                 goto bad_fork_cleanup_threadgroup_lock;
1816         }
1817 #endif
1818 #ifdef CONFIG_CPUSETS
1819         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1820         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1821         seqcount_init(&p->mems_allowed_seq);
1822 #endif
1823 #ifdef CONFIG_TRACE_IRQFLAGS
1824         p->irq_events = 0;
1825         p->hardirqs_enabled = 0;
1826         p->hardirq_enable_ip = 0;
1827         p->hardirq_enable_event = 0;
1828         p->hardirq_disable_ip = _THIS_IP_;
1829         p->hardirq_disable_event = 0;
1830         p->softirqs_enabled = 1;
1831         p->softirq_enable_ip = _THIS_IP_;
1832         p->softirq_enable_event = 0;
1833         p->softirq_disable_ip = 0;
1834         p->softirq_disable_event = 0;
1835         p->hardirq_context = 0;
1836         p->softirq_context = 0;
1837 #endif
1838
1839         p->pagefault_disabled = 0;
1840
1841 #ifdef CONFIG_LOCKDEP
1842         p->lockdep_depth = 0; /* no locks held yet */
1843         p->curr_chain_key = 0;
1844         p->lockdep_recursion = 0;
1845         lockdep_init_task(p);
1846 #endif
1847
1848 #ifdef CONFIG_DEBUG_MUTEXES
1849         p->blocked_on = NULL; /* not blocked yet */
1850 #endif
1851 #ifdef CONFIG_BCACHE
1852         p->sequential_io        = 0;
1853         p->sequential_io_avg    = 0;
1854 #endif
1855
1856         /* Perform scheduler related setup. Assign this task to a CPU. */
1857         retval = sched_fork(clone_flags, p);
1858         if (retval)
1859                 goto bad_fork_cleanup_policy;
1860
1861         retval = perf_event_init_task(p);
1862         if (retval)
1863                 goto bad_fork_cleanup_policy;
1864         retval = audit_alloc(p);
1865         if (retval)
1866                 goto bad_fork_cleanup_perf;
1867         /* copy all the process information */
1868         shm_init_task(p);
1869         retval = security_task_alloc(p, clone_flags);
1870         if (retval)
1871                 goto bad_fork_cleanup_audit;
1872         retval = copy_semundo(clone_flags, p);
1873         if (retval)
1874                 goto bad_fork_cleanup_security;
1875         retval = copy_files(clone_flags, p);
1876         if (retval)
1877                 goto bad_fork_cleanup_semundo;
1878         retval = copy_fs(clone_flags, p);
1879         if (retval)
1880                 goto bad_fork_cleanup_files;
1881         retval = copy_sighand(clone_flags, p);
1882         if (retval)
1883                 goto bad_fork_cleanup_fs;
1884         retval = copy_signal(clone_flags, p);
1885         if (retval)
1886                 goto bad_fork_cleanup_sighand;
1887         retval = copy_mm(clone_flags, p);
1888         if (retval)
1889                 goto bad_fork_cleanup_signal;
1890         retval = copy_namespaces(clone_flags, p);
1891         if (retval)
1892                 goto bad_fork_cleanup_mm;
1893         retval = copy_io(clone_flags, p);
1894         if (retval)
1895                 goto bad_fork_cleanup_namespaces;
1896         retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1897         if (retval)
1898                 goto bad_fork_cleanup_io;
1899
1900         if (pid != &init_struct_pid) {
1901                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1902                 if (IS_ERR(pid)) {
1903                         retval = PTR_ERR(pid);
1904                         goto bad_fork_cleanup_thread;
1905                 }
1906         }
1907
1908 #ifdef CONFIG_BLOCK
1909         p->plug = NULL;
1910 #endif
1911 #ifdef CONFIG_FUTEX
1912         p->robust_list = NULL;
1913 #ifdef CONFIG_COMPAT
1914         p->compat_robust_list = NULL;
1915 #endif
1916         INIT_LIST_HEAD(&p->pi_state_list);
1917         p->pi_state_cache = NULL;
1918 #endif
1919         /*
1920          * sigaltstack should be cleared when sharing the same VM
1921          */
1922         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1923                 sas_ss_reset(p);
1924
1925         /*
1926          * Syscall tracing and stepping should be turned off in the
1927          * child regardless of CLONE_PTRACE.
1928          */
1929         user_disable_single_step(p);
1930         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1931 #ifdef TIF_SYSCALL_EMU
1932         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1933 #endif
1934         clear_all_latency_tracing(p);
1935
1936         /* ok, now we should be set up.. */
1937         p->pid = pid_nr(pid);
1938         if (clone_flags & CLONE_THREAD) {
1939                 p->exit_signal = -1;
1940                 p->group_leader = current->group_leader;
1941                 p->tgid = current->tgid;
1942         } else {
1943                 if (clone_flags & CLONE_PARENT)
1944                         p->exit_signal = current->group_leader->exit_signal;
1945                 else
1946                         p->exit_signal = (clone_flags & CSIGNAL);
1947                 p->group_leader = p;
1948                 p->tgid = p->pid;
1949         }
1950
1951         p->nr_dirtied = 0;
1952         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1953         p->dirty_paused_when = 0;
1954
1955         p->pdeath_signal = 0;
1956         INIT_LIST_HEAD(&p->thread_group);
1957         p->task_works = NULL;
1958
1959         cgroup_threadgroup_change_begin(current);
1960         /*
1961          * Ensure that the cgroup subsystem policies allow the new process to be
1962          * forked. It should be noted the the new process's css_set can be changed
1963          * between here and cgroup_post_fork() if an organisation operation is in
1964          * progress.
1965          */
1966         retval = cgroup_can_fork(p);
1967         if (retval)
1968                 goto bad_fork_free_pid;
1969
1970         /*
1971          * From this point on we must avoid any synchronous user-space
1972          * communication until we take the tasklist-lock. In particular, we do
1973          * not want user-space to be able to predict the process start-time by
1974          * stalling fork(2) after we recorded the start_time but before it is
1975          * visible to the system.
1976          */
1977
1978         p->start_time = ktime_get_ns();
1979         p->real_start_time = ktime_get_boot_ns();
1980
1981         /*
1982          * Make it visible to the rest of the system, but dont wake it up yet.
1983          * Need tasklist lock for parent etc handling!
1984          */
1985         write_lock_irq(&tasklist_lock);
1986
1987         /* CLONE_PARENT re-uses the old parent */
1988         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1989                 p->real_parent = current->real_parent;
1990                 p->parent_exec_id = current->parent_exec_id;
1991         } else {
1992                 p->real_parent = current;
1993                 p->parent_exec_id = current->self_exec_id;
1994         }
1995
1996         klp_copy_process(p);
1997
1998         spin_lock(&current->sighand->siglock);
1999
2000         /*
2001          * Copy seccomp details explicitly here, in case they were changed
2002          * before holding sighand lock.
2003          */
2004         copy_seccomp(p);
2005
2006         rseq_fork(p, clone_flags);
2007
2008         /* Don't start children in a dying pid namespace */
2009         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2010                 retval = -ENOMEM;
2011                 goto bad_fork_cancel_cgroup;
2012         }
2013
2014         /* Let kill terminate clone/fork in the middle */
2015         if (fatal_signal_pending(current)) {
2016                 retval = -EINTR;
2017                 goto bad_fork_cancel_cgroup;
2018         }
2019
2020
2021         init_task_pid_links(p);
2022         if (likely(p->pid)) {
2023                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2024
2025                 init_task_pid(p, PIDTYPE_PID, pid);
2026                 if (thread_group_leader(p)) {
2027                         init_task_pid(p, PIDTYPE_TGID, pid);
2028                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2029                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2030
2031                         if (is_child_reaper(pid)) {
2032                                 ns_of_pid(pid)->child_reaper = p;
2033                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2034                         }
2035                         p->signal->shared_pending.signal = delayed.signal;
2036                         p->signal->tty = tty_kref_get(current->signal->tty);
2037                         /*
2038                          * Inherit has_child_subreaper flag under the same
2039                          * tasklist_lock with adding child to the process tree
2040                          * for propagate_has_child_subreaper optimization.
2041                          */
2042                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2043                                                          p->real_parent->signal->is_child_subreaper;
2044                         list_add_tail(&p->sibling, &p->real_parent->children);
2045                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2046                         attach_pid(p, PIDTYPE_TGID);
2047                         attach_pid(p, PIDTYPE_PGID);
2048                         attach_pid(p, PIDTYPE_SID);
2049                         __this_cpu_inc(process_counts);
2050                 } else {
2051                         current->signal->nr_threads++;
2052                         atomic_inc(&current->signal->live);
2053                         atomic_inc(&current->signal->sigcnt);
2054                         task_join_group_stop(p);
2055                         list_add_tail_rcu(&p->thread_group,
2056                                           &p->group_leader->thread_group);
2057                         list_add_tail_rcu(&p->thread_node,
2058                                           &p->signal->thread_head);
2059                 }
2060                 attach_pid(p, PIDTYPE_PID);
2061                 nr_threads++;
2062         }
2063         total_forks++;
2064         hlist_del_init(&delayed.node);
2065         spin_unlock(&current->sighand->siglock);
2066         syscall_tracepoint_update(p);
2067         write_unlock_irq(&tasklist_lock);
2068
2069         proc_fork_connector(p);
2070         cgroup_post_fork(p);
2071         cgroup_threadgroup_change_end(current);
2072         perf_event_fork(p);
2073
2074         trace_task_newtask(p, clone_flags);
2075         uprobe_copy_process(p, clone_flags);
2076
2077         return p;
2078
2079 bad_fork_cancel_cgroup:
2080         spin_unlock(&current->sighand->siglock);
2081         write_unlock_irq(&tasklist_lock);
2082         cgroup_cancel_fork(p);
2083 bad_fork_free_pid:
2084         cgroup_threadgroup_change_end(current);
2085         if (pid != &init_struct_pid)
2086                 free_pid(pid);
2087 bad_fork_cleanup_thread:
2088         exit_thread(p);
2089 bad_fork_cleanup_io:
2090         if (p->io_context)
2091                 exit_io_context(p);
2092 bad_fork_cleanup_namespaces:
2093         exit_task_namespaces(p);
2094 bad_fork_cleanup_mm:
2095         if (p->mm) {
2096                 mm_clear_owner(p->mm, p);
2097                 mmput(p->mm);
2098         }
2099 bad_fork_cleanup_signal:
2100         if (!(clone_flags & CLONE_THREAD))
2101                 free_signal_struct(p->signal);
2102 bad_fork_cleanup_sighand:
2103         __cleanup_sighand(p->sighand);
2104 bad_fork_cleanup_fs:
2105         exit_fs(p); /* blocking */
2106 bad_fork_cleanup_files:
2107         exit_files(p); /* blocking */
2108 bad_fork_cleanup_semundo:
2109         exit_sem(p);
2110 bad_fork_cleanup_security:
2111         security_task_free(p);
2112 bad_fork_cleanup_audit:
2113         audit_free(p);
2114 bad_fork_cleanup_perf:
2115         perf_event_free_task(p);
2116 bad_fork_cleanup_policy:
2117         lockdep_free_task(p);
2118 #ifdef CONFIG_NUMA
2119         mpol_put(p->mempolicy);
2120 bad_fork_cleanup_threadgroup_lock:
2121 #endif
2122         delayacct_tsk_free(p);
2123 bad_fork_cleanup_count:
2124         atomic_dec(&p->cred->user->processes);
2125         exit_creds(p);
2126 bad_fork_free:
2127         p->state = TASK_DEAD;
2128         put_task_stack(p);
2129         delayed_free_task(p);
2130 fork_out:
2131         spin_lock_irq(&current->sighand->siglock);
2132         hlist_del_init(&delayed.node);
2133         spin_unlock_irq(&current->sighand->siglock);
2134         return ERR_PTR(retval);
2135 }
2136
2137 static inline void init_idle_pids(struct task_struct *idle)
2138 {
2139         enum pid_type type;
2140
2141         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2142                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2143                 init_task_pid(idle, type, &init_struct_pid);
2144         }
2145 }
2146
2147 struct task_struct *fork_idle(int cpu)
2148 {
2149         struct task_struct *task;
2150         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
2151                             cpu_to_node(cpu));
2152         if (!IS_ERR(task)) {
2153                 init_idle_pids(task);
2154                 init_idle(task, cpu);
2155         }
2156
2157         return task;
2158 }
2159
2160 /*
2161  *  Ok, this is the main fork-routine.
2162  *
2163  * It copies the process, and if successful kick-starts
2164  * it and waits for it to finish using the VM if required.
2165  */
2166 long _do_fork(unsigned long clone_flags,
2167               unsigned long stack_start,
2168               unsigned long stack_size,
2169               int __user *parent_tidptr,
2170               int __user *child_tidptr,
2171               unsigned long tls)
2172 {
2173         struct completion vfork;
2174         struct pid *pid;
2175         struct task_struct *p;
2176         int trace = 0;
2177         long nr;
2178
2179         /*
2180          * Determine whether and which event to report to ptracer.  When
2181          * called from kernel_thread or CLONE_UNTRACED is explicitly
2182          * requested, no event is reported; otherwise, report if the event
2183          * for the type of forking is enabled.
2184          */
2185         if (!(clone_flags & CLONE_UNTRACED)) {
2186                 if (clone_flags & CLONE_VFORK)
2187                         trace = PTRACE_EVENT_VFORK;
2188                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2189                         trace = PTRACE_EVENT_CLONE;
2190                 else
2191                         trace = PTRACE_EVENT_FORK;
2192
2193                 if (likely(!ptrace_event_enabled(current, trace)))
2194                         trace = 0;
2195         }
2196
2197         p = copy_process(clone_flags, stack_start, stack_size,
2198                          child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2199         add_latent_entropy();
2200
2201         if (IS_ERR(p))
2202                 return PTR_ERR(p);
2203
2204         /*
2205          * Do this prior waking up the new thread - the thread pointer
2206          * might get invalid after that point, if the thread exits quickly.
2207          */
2208         trace_sched_process_fork(current, p);
2209
2210         pid = get_task_pid(p, PIDTYPE_PID);
2211         nr = pid_vnr(pid);
2212
2213         if (clone_flags & CLONE_PARENT_SETTID)
2214                 put_user(nr, parent_tidptr);
2215
2216         if (clone_flags & CLONE_VFORK) {
2217                 p->vfork_done = &vfork;
2218                 init_completion(&vfork);
2219                 get_task_struct(p);
2220         }
2221
2222         wake_up_new_task(p);
2223
2224         /* forking complete and child started to run, tell ptracer */
2225         if (unlikely(trace))
2226                 ptrace_event_pid(trace, pid);
2227
2228         if (clone_flags & CLONE_VFORK) {
2229                 if (!wait_for_vfork_done(p, &vfork))
2230                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2231         }
2232
2233         put_pid(pid);
2234         return nr;
2235 }
2236
2237 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2238 /* For compatibility with architectures that call do_fork directly rather than
2239  * using the syscall entry points below. */
2240 long do_fork(unsigned long clone_flags,
2241               unsigned long stack_start,
2242               unsigned long stack_size,
2243               int __user *parent_tidptr,
2244               int __user *child_tidptr)
2245 {
2246         return _do_fork(clone_flags, stack_start, stack_size,
2247                         parent_tidptr, child_tidptr, 0);
2248 }
2249 #endif
2250
2251 /*
2252  * Create a kernel thread.
2253  */
2254 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2255 {
2256         return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2257                 (unsigned long)arg, NULL, NULL, 0);
2258 }
2259
2260 #ifdef __ARCH_WANT_SYS_FORK
2261 SYSCALL_DEFINE0(fork)
2262 {
2263 #ifdef CONFIG_MMU
2264         return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2265 #else
2266         /* can not support in nommu mode */
2267         return -EINVAL;
2268 #endif
2269 }
2270 #endif
2271
2272 #ifdef __ARCH_WANT_SYS_VFORK
2273 SYSCALL_DEFINE0(vfork)
2274 {
2275         return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2276                         0, NULL, NULL, 0);
2277 }
2278 #endif
2279
2280 #ifdef __ARCH_WANT_SYS_CLONE
2281 #ifdef CONFIG_CLONE_BACKWARDS
2282 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2283                  int __user *, parent_tidptr,
2284                  unsigned long, tls,
2285                  int __user *, child_tidptr)
2286 #elif defined(CONFIG_CLONE_BACKWARDS2)
2287 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2288                  int __user *, parent_tidptr,
2289                  int __user *, child_tidptr,
2290                  unsigned long, tls)
2291 #elif defined(CONFIG_CLONE_BACKWARDS3)
2292 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2293                 int, stack_size,
2294                 int __user *, parent_tidptr,
2295                 int __user *, child_tidptr,
2296                 unsigned long, tls)
2297 #else
2298 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2299                  int __user *, parent_tidptr,
2300                  int __user *, child_tidptr,
2301                  unsigned long, tls)
2302 #endif
2303 {
2304         return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2305 }
2306 #endif
2307
2308 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2309 {
2310         struct task_struct *leader, *parent, *child;
2311         int res;
2312
2313         read_lock(&tasklist_lock);
2314         leader = top = top->group_leader;
2315 down:
2316         for_each_thread(leader, parent) {
2317                 list_for_each_entry(child, &parent->children, sibling) {
2318                         res = visitor(child, data);
2319                         if (res) {
2320                                 if (res < 0)
2321                                         goto out;
2322                                 leader = child;
2323                                 goto down;
2324                         }
2325 up:
2326                         ;
2327                 }
2328         }
2329
2330         if (leader != top) {
2331                 child = leader;
2332                 parent = child->real_parent;
2333                 leader = parent->group_leader;
2334                 goto up;
2335         }
2336 out:
2337         read_unlock(&tasklist_lock);
2338 }
2339
2340 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2341 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2342 #endif
2343
2344 static void sighand_ctor(void *data)
2345 {
2346         struct sighand_struct *sighand = data;
2347
2348         spin_lock_init(&sighand->siglock);
2349         init_waitqueue_head(&sighand->signalfd_wqh);
2350 }
2351
2352 void __init proc_caches_init(void)
2353 {
2354         unsigned int mm_size;
2355
2356         sighand_cachep = kmem_cache_create("sighand_cache",
2357                         sizeof(struct sighand_struct), 0,
2358                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2359                         SLAB_ACCOUNT, sighand_ctor);
2360         signal_cachep = kmem_cache_create("signal_cache",
2361                         sizeof(struct signal_struct), 0,
2362                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2363                         NULL);
2364         files_cachep = kmem_cache_create("files_cache",
2365                         sizeof(struct files_struct), 0,
2366                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2367                         NULL);
2368         fs_cachep = kmem_cache_create("fs_cache",
2369                         sizeof(struct fs_struct), 0,
2370                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2371                         NULL);
2372
2373         /*
2374          * The mm_cpumask is located at the end of mm_struct, and is
2375          * dynamically sized based on the maximum CPU number this system
2376          * can have, taking hotplug into account (nr_cpu_ids).
2377          */
2378         mm_size = sizeof(struct mm_struct) + cpumask_size();
2379
2380         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2381                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2382                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2383                         offsetof(struct mm_struct, saved_auxv),
2384                         sizeof_field(struct mm_struct, saved_auxv),
2385                         NULL);
2386         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2387         mmap_init();
2388         nsproxy_cache_init();
2389 }
2390
2391 /*
2392  * Check constraints on flags passed to the unshare system call.
2393  */
2394 static int check_unshare_flags(unsigned long unshare_flags)
2395 {
2396         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2397                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2398                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2399                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2400                 return -EINVAL;
2401         /*
2402          * Not implemented, but pretend it works if there is nothing
2403          * to unshare.  Note that unsharing the address space or the
2404          * signal handlers also need to unshare the signal queues (aka
2405          * CLONE_THREAD).
2406          */
2407         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2408                 if (!thread_group_empty(current))
2409                         return -EINVAL;
2410         }
2411         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2412                 if (atomic_read(&current->sighand->count) > 1)
2413                         return -EINVAL;
2414         }
2415         if (unshare_flags & CLONE_VM) {
2416                 if (!current_is_single_threaded())
2417                         return -EINVAL;
2418         }
2419
2420         return 0;
2421 }
2422
2423 /*
2424  * Unshare the filesystem structure if it is being shared
2425  */
2426 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2427 {
2428         struct fs_struct *fs = current->fs;
2429
2430         if (!(unshare_flags & CLONE_FS) || !fs)
2431                 return 0;
2432
2433         /* don't need lock here; in the worst case we'll do useless copy */
2434         if (fs->users == 1)
2435                 return 0;
2436
2437         *new_fsp = copy_fs_struct(fs);
2438         if (!*new_fsp)
2439                 return -ENOMEM;
2440
2441         return 0;
2442 }
2443
2444 /*
2445  * Unshare file descriptor table if it is being shared
2446  */
2447 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2448 {
2449         struct files_struct *fd = current->files;
2450         int error = 0;
2451
2452         if ((unshare_flags & CLONE_FILES) &&
2453             (fd && atomic_read(&fd->count) > 1)) {
2454                 *new_fdp = dup_fd(fd, &error);
2455                 if (!*new_fdp)
2456                         return error;
2457         }
2458
2459         return 0;
2460 }
2461
2462 /*
2463  * unshare allows a process to 'unshare' part of the process
2464  * context which was originally shared using clone.  copy_*
2465  * functions used by do_fork() cannot be used here directly
2466  * because they modify an inactive task_struct that is being
2467  * constructed. Here we are modifying the current, active,
2468  * task_struct.
2469  */
2470 int ksys_unshare(unsigned long unshare_flags)
2471 {
2472         struct fs_struct *fs, *new_fs = NULL;
2473         struct files_struct *fd, *new_fd = NULL;
2474         struct cred *new_cred = NULL;
2475         struct nsproxy *new_nsproxy = NULL;
2476         int do_sysvsem = 0;
2477         int err;
2478
2479         /*
2480          * If unsharing a user namespace must also unshare the thread group
2481          * and unshare the filesystem root and working directories.
2482          */
2483         if (unshare_flags & CLONE_NEWUSER)
2484                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2485         /*
2486          * If unsharing vm, must also unshare signal handlers.
2487          */
2488         if (unshare_flags & CLONE_VM)
2489                 unshare_flags |= CLONE_SIGHAND;
2490         /*
2491          * If unsharing a signal handlers, must also unshare the signal queues.
2492          */
2493         if (unshare_flags & CLONE_SIGHAND)
2494                 unshare_flags |= CLONE_THREAD;
2495         /*
2496          * If unsharing namespace, must also unshare filesystem information.
2497          */
2498         if (unshare_flags & CLONE_NEWNS)
2499                 unshare_flags |= CLONE_FS;
2500
2501         err = check_unshare_flags(unshare_flags);
2502         if (err)
2503                 goto bad_unshare_out;
2504         /*
2505          * CLONE_NEWIPC must also detach from the undolist: after switching
2506          * to a new ipc namespace, the semaphore arrays from the old
2507          * namespace are unreachable.
2508          */
2509         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2510                 do_sysvsem = 1;
2511         err = unshare_fs(unshare_flags, &new_fs);
2512         if (err)
2513                 goto bad_unshare_out;
2514         err = unshare_fd(unshare_flags, &new_fd);
2515         if (err)
2516                 goto bad_unshare_cleanup_fs;
2517         err = unshare_userns(unshare_flags, &new_cred);
2518         if (err)
2519                 goto bad_unshare_cleanup_fd;
2520         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2521                                          new_cred, new_fs);
2522         if (err)
2523                 goto bad_unshare_cleanup_cred;
2524
2525         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2526                 if (do_sysvsem) {
2527                         /*
2528                          * CLONE_SYSVSEM is equivalent to sys_exit().
2529                          */
2530                         exit_sem(current);
2531                 }
2532                 if (unshare_flags & CLONE_NEWIPC) {
2533                         /* Orphan segments in old ns (see sem above). */
2534                         exit_shm(current);
2535                         shm_init_task(current);
2536                 }
2537
2538                 if (new_nsproxy)
2539                         switch_task_namespaces(current, new_nsproxy);
2540
2541                 task_lock(current);
2542
2543                 if (new_fs) {
2544                         fs = current->fs;
2545                         spin_lock(&fs->lock);
2546                         current->fs = new_fs;
2547                         if (--fs->users)
2548                                 new_fs = NULL;
2549                         else
2550                                 new_fs = fs;
2551                         spin_unlock(&fs->lock);
2552                 }
2553
2554                 if (new_fd) {
2555                         fd = current->files;
2556                         current->files = new_fd;
2557                         new_fd = fd;
2558                 }
2559
2560                 task_unlock(current);
2561
2562                 if (new_cred) {
2563                         /* Install the new user namespace */
2564                         commit_creds(new_cred);
2565                         new_cred = NULL;
2566                 }
2567         }
2568
2569         perf_event_namespaces(current);
2570
2571 bad_unshare_cleanup_cred:
2572         if (new_cred)
2573                 put_cred(new_cred);
2574 bad_unshare_cleanup_fd:
2575         if (new_fd)
2576                 put_files_struct(new_fd);
2577
2578 bad_unshare_cleanup_fs:
2579         if (new_fs)
2580                 free_fs_struct(new_fs);
2581
2582 bad_unshare_out:
2583         return err;
2584 }
2585
2586 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2587 {
2588         return ksys_unshare(unshare_flags);
2589 }
2590
2591 /*
2592  *      Helper to unshare the files of the current task.
2593  *      We don't want to expose copy_files internals to
2594  *      the exec layer of the kernel.
2595  */
2596
2597 int unshare_files(struct files_struct **displaced)
2598 {
2599         struct task_struct *task = current;
2600         struct files_struct *copy = NULL;
2601         int error;
2602
2603         error = unshare_fd(CLONE_FILES, &copy);
2604         if (error || !copy) {
2605                 *displaced = NULL;
2606                 return error;
2607         }
2608         *displaced = task->files;
2609         task_lock(task);
2610         task->files = copy;
2611         task_unlock(task);
2612         return 0;
2613 }
2614
2615 int sysctl_max_threads(struct ctl_table *table, int write,
2616                        void __user *buffer, size_t *lenp, loff_t *ppos)
2617 {
2618         struct ctl_table t;
2619         int ret;
2620         int threads = max_threads;
2621         int min = 1;
2622         int max = MAX_THREADS;
2623
2624         t = *table;
2625         t.data = &threads;
2626         t.extra1 = &min;
2627         t.extra2 = &max;
2628
2629         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2630         if (ret || !write)
2631                 return ret;
2632
2633         max_threads = threads;
2634
2635         return 0;
2636 }