kernel/fork.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/kernel/fork.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  */
   7
   8 /*
   9  *  'fork.c' contains the help-routines for the 'fork' system call
  10  * (see also entry.S and others).
  11  * Fork is rather simple, once you get the hang of it, but the memory
  12  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  13  */
  14
  15 #include <linux/anon_inodes.h>
  16 #include <linux/slab.h>
  17 #include <linux/sched/autogroup.h>
  18 #include <linux/sched/mm.h>
  19 #include <linux/sched/coredump.h>
  20 #include <linux/sched/user.h>
  21 #include <linux/sched/numa_balancing.h>
  22 #include <linux/sched/stat.h>
  23 #include <linux/sched/task.h>
  24 #include <linux/sched/task_stack.h>
  25 #include <linux/sched/cputime.h>
  26 #include <linux/seq_file.h>
  27 #include <linux/rtmutex.h>
  28 #include <linux/init.h>
  29 #include <linux/unistd.h>
  30 #include <linux/module.h>
  31 #include <linux/vmalloc.h>
  32 #include <linux/completion.h>
  33 #include <linux/personality.h>
  34 #include <linux/mempolicy.h>
  35 #include <linux/sem.h>
  36 #include <linux/file.h>
  37 #include <linux/fdtable.h>
  38 #include <linux/iocontext.h>
  39 #include <linux/key.h>
  40 #include <linux/binfmts.h>
  41 #include <linux/mman.h>
  42 #include <linux/mmu_notifier.h>
  43 #include <linux/fs.h>
  44 #include <linux/mm.h>
  45 #include <linux/vmacache.h>
  46 #include <linux/nsproxy.h>
  47 #include <linux/capability.h>
  48 #include <linux/cpu.h>
  49 #include <linux/cgroup.h>
  50 #include <linux/security.h>
  51 #include <linux/hugetlb.h>
  52 #include <linux/seccomp.h>
  53 #include <linux/swap.h>
  54 #include <linux/syscalls.h>
  55 #include <linux/jiffies.h>
  56 #include <linux/futex.h>
  57 #include <linux/compat.h>
  58 #include <linux/kthread.h>
  59 #include <linux/task_io_accounting_ops.h>
  60 #include <linux/rcupdate.h>
  61 #include <linux/ptrace.h>
  62 #include <linux/mount.h>
  63 #include <linux/audit.h>
  64 #include <linux/memcontrol.h>
  65 #include <linux/ftrace.h>
  66 #include <linux/proc_fs.h>
  67 #include <linux/profile.h>
  68 #include <linux/rmap.h>
  69 #include <linux/ksm.h>
  70 #include <linux/acct.h>
  71 #include <linux/userfaultfd_k.h>
  72 #include <linux/tsacct_kern.h>
  73 #include <linux/cn_proc.h>
  74 #include <linux/freezer.h>
  75 #include <linux/delayacct.h>
  76 #include <linux/taskstats_kern.h>
  77 #include <linux/random.h>
  78 #include <linux/tty.h>
  79 #include <linux/blkdev.h>
  80 #include <linux/fs_struct.h>
  81 #include <linux/magic.h>
  82 #include <linux/perf_event.h>
  83 #include <linux/posix-timers.h>
  84 #include <linux/user-return-notifier.h>
  85 #include <linux/oom.h>
  86 #include <linux/khugepaged.h>
  87 #include <linux/signalfd.h>
  88 #include <linux/uprobes.h>
  89 #include <linux/aio.h>
  90 #include <linux/compiler.h>
  91 #include <linux/sysctl.h>
  92 #include <linux/kcov.h>
  93 #include <linux/livepatch.h>
  94 #include <linux/thread_info.h>
  95 #include <linux/stackleak.h>
  96 #include <linux/kasan.h>
  97
  98 #include <asm/pgtable.h>
  99 #include <asm/pgalloc.h>
 100 #include <linux/uaccess.h>
 101 #include <asm/mmu_context.h>
 102 #include <asm/cacheflush.h>
 103 #include <asm/tlbflush.h>
 104
 105 #include <trace/events/sched.h>
 106
 107 #define CREATE_TRACE_POINTS
 108 #include <trace/events/task.h>
 109
 110 /*
 111  * Minimum number of threads to boot the kernel
 112  */
 113 #define MIN_THREADS 20
 114
 115 /*
 116  * Maximum number of threads
 117  */
 118 #define MAX_THREADS FUTEX_TID_MASK
 119
 120 /*
 121  * Protected counters by write_lock_irq(&tasklist_lock)
 122  */
 123 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 124 int nr_threads;                 /* The idle threads do not count.. */
 125
 126 static int max_threads;         /* tunable limit on nr_threads */
 127
 128 #define NAMED_ARRAY_INDEX(x)    [x] = __stringify(x)
 129
 130 static const char * const resident_page_types[] = {
 131         NAMED_ARRAY_INDEX(MM_FILEPAGES),
 132         NAMED_ARRAY_INDEX(MM_ANONPAGES),
 133         NAMED_ARRAY_INDEX(MM_SWAPENTS),
 134         NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
 135 };
 136
 137 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 138
 139 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 140
 141 #ifdef CONFIG_PROVE_RCU
 142 int lockdep_tasklist_lock_is_held(void)
 143 {
 144         return lockdep_is_held(&tasklist_lock);
 145 }
 146 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 147 #endif /* #ifdef CONFIG_PROVE_RCU */
 148
 149 int nr_processes(void)
 150 {
 151         int cpu;
 152         int total = 0;
 153
 154         for_each_possible_cpu(cpu)
 155                 total += per_cpu(process_counts, cpu);
 156
 157         return total;
 158 }
 159
 160 void __weak arch_release_task_struct(struct task_struct *tsk)
 161 {
 162 }
 163
 164 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 165 static struct kmem_cache *task_struct_cachep;
 166
 167 static inline struct task_struct *alloc_task_struct_node(int node)
 168 {
 169         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 170 }
 171
 172 static inline void free_task_struct(struct task_struct *tsk)
 173 {
 174         kmem_cache_free(task_struct_cachep, tsk);
 175 }
 176 #endif
 177
 178 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 179
 180 /*
 181  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 182  * kmemcache based allocator.
 183  */
 184 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 185
 186 #ifdef CONFIG_VMAP_STACK
 187 /*
 188  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 189  * flush.  Try to minimize the number of calls by caching stacks.
 190  */
 191 #define NR_CACHED_STACKS 2
 192 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 193
 194 static int free_vm_stack_cache(unsigned int cpu)
 195 {
 196         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 197         int i;
 198
 199         for (i = 0; i < NR_CACHED_STACKS; i++) {
 200                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 201
 202                 if (!vm_stack)
 203                         continue;
 204
 205                 vfree(vm_stack->addr);
 206                 cached_vm_stacks[i] = NULL;
 207         }
 208
 209         return 0;
 210 }
 211 #endif
 212
 213 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 214 {
 215 #ifdef CONFIG_VMAP_STACK
 216         void *stack;
 217         int i;
 218
 219         for (i = 0; i < NR_CACHED_STACKS; i++) {
 220                 struct vm_struct *s;
 221
 222                 s = this_cpu_xchg(cached_stacks[i], NULL);
 223
 224                 if (!s)
 225                         continue;
 226
 227                 /* Clear the KASAN shadow of the stack. */
 228                 kasan_unpoison_shadow(s->addr, THREAD_SIZE);
 229
 230                 /* Clear stale pointers from reused stack. */
 231                 memset(s->addr, 0, THREAD_SIZE);
 232
 233                 tsk->stack_vm_area = s;
 234                 tsk->stack = s->addr;
 235                 return s->addr;
 236         }
 237
 238         /*
 239          * Allocated stacks are cached and later reused by new threads,
 240          * so memcg accounting is performed manually on assigning/releasing
 241          * stacks to tasks. Drop __GFP_ACCOUNT.
 242          */
 243         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 244                                      VMALLOC_START, VMALLOC_END,
 245                                      THREADINFO_GFP & ~__GFP_ACCOUNT,
 246                                      PAGE_KERNEL,
 247                                      0, node, __builtin_return_address(0));
 248
 249         /*
 250          * We can't call find_vm_area() in interrupt context, and
 251          * free_thread_stack() can be called in interrupt context,
 252          * so cache the vm_struct.
 253          */
 254         if (stack) {
 255                 tsk->stack_vm_area = find_vm_area(stack);
 256                 tsk->stack = stack;
 257         }
 258         return stack;
 259 #else
 260         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 261                                              THREAD_SIZE_ORDER);
 262
 263         if (likely(page)) {
 264                 tsk->stack = page_address(page);
 265                 return tsk->stack;
 266         }
 267         return NULL;
 268 #endif
 269 }
 270
 271 static inline void free_thread_stack(struct task_struct *tsk)
 272 {
 273 #ifdef CONFIG_VMAP_STACK
 274         struct vm_struct *vm = task_stack_vm_area(tsk);
 275
 276         if (vm) {
 277                 int i;
 278
 279                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 280                         mod_memcg_page_state(vm->pages[i],
 281                                              MEMCG_KERNEL_STACK_KB,
 282                                              -(int)(PAGE_SIZE / 1024));
 283
 284                         memcg_kmem_uncharge_page(vm->pages[i], 0);
 285                 }
 286
 287                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 288                         if (this_cpu_cmpxchg(cached_stacks[i],
 289                                         NULL, tsk->stack_vm_area) != NULL)
 290                                 continue;
 291
 292                         return;
 293                 }
 294
 295                 vfree_atomic(tsk->stack);
 296                 return;
 297         }
 298 #endif
 299
 300         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 301 }
 302 # else
 303 static struct kmem_cache *thread_stack_cache;
 304
 305 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 306                                                   int node)
 307 {
 308         unsigned long *stack;
 309         stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 310         tsk->stack = stack;
 311         return stack;
 312 }
 313
 314 static void free_thread_stack(struct task_struct *tsk)
 315 {
 316         kmem_cache_free(thread_stack_cache, tsk->stack);
 317 }
 318
 319 void thread_stack_cache_init(void)
 320 {
 321         thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 322                                         THREAD_SIZE, THREAD_SIZE, 0, 0,
 323                                         THREAD_SIZE, NULL);
 324         BUG_ON(thread_stack_cache == NULL);
 325 }
 326 # endif
 327 #endif
 328
 329 /* SLAB cache for signal_struct structures (tsk->signal) */
 330 static struct kmem_cache *signal_cachep;
 331
 332 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 333 struct kmem_cache *sighand_cachep;
 334
 335 /* SLAB cache for files_struct structures (tsk->files) */
 336 struct kmem_cache *files_cachep;
 337
 338 /* SLAB cache for fs_struct structures (tsk->fs) */
 339 struct kmem_cache *fs_cachep;
 340
 341 /* SLAB cache for vm_area_struct structures */
 342 static struct kmem_cache *vm_area_cachep;
 343
 344 /* SLAB cache for mm_struct structures (tsk->mm) */
 345 static struct kmem_cache *mm_cachep;
 346
 347 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 348 {
 349         struct vm_area_struct *vma;
 350
 351         vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 352         if (vma)
 353                 vma_init(vma, mm);
 354         return vma;
 355 }
 356
 357 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 358 {
 359         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 360
 361         if (new) {
 362                 *new = *orig;
 363                 INIT_LIST_HEAD(&new->anon_vma_chain);
 364         }
 365         return new;
 366 }
 367
 368 void vm_area_free(struct vm_area_struct *vma)
 369 {
 370         kmem_cache_free(vm_area_cachep, vma);
 371 }
 372
 373 static void account_kernel_stack(struct task_struct *tsk, int account)
 374 {
 375         void *stack = task_stack_page(tsk);
 376         struct vm_struct *vm = task_stack_vm_area(tsk);
 377
 378         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 379
 380         if (vm) {
 381                 int i;
 382
 383                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 384
 385                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 386                         mod_zone_page_state(page_zone(vm->pages[i]),
 387                                             NR_KERNEL_STACK_KB,
 388                                             PAGE_SIZE / 1024 * account);
 389                 }
 390         } else {
 391                 /*
 392                  * All stack pages are in the same zone and belong to the
 393                  * same memcg.
 394                  */
 395                 struct page *first_page = virt_to_page(stack);
 396
 397                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 398                                     THREAD_SIZE / 1024 * account);
 399
 400                 mod_memcg_obj_state(stack, MEMCG_KERNEL_STACK_KB,
 401                                     account * (THREAD_SIZE / 1024));
 402         }
 403 }
 404
 405 static int memcg_charge_kernel_stack(struct task_struct *tsk)
 406 {
 407 #ifdef CONFIG_VMAP_STACK
 408         struct vm_struct *vm = task_stack_vm_area(tsk);
 409         int ret;
 410
 411         if (vm) {
 412                 int i;
 413
 414                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 415                         /*
 416                          * If memcg_kmem_charge_page() fails, page->mem_cgroup
 417                          * pointer is NULL, and both memcg_kmem_uncharge_page()
 418                          * and mod_memcg_page_state() in free_thread_stack()
 419                          * will ignore this page. So it's safe.
 420                          */
 421                         ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL,
 422                                                      0);
 423                         if (ret)
 424                                 return ret;
 425
 426                         mod_memcg_page_state(vm->pages[i],
 427                                              MEMCG_KERNEL_STACK_KB,
 428                                              PAGE_SIZE / 1024);
 429                 }
 430         }
 431 #endif
 432         return 0;
 433 }
 434
 435 static void release_task_stack(struct task_struct *tsk)
 436 {
 437         if (WARN_ON(tsk->state != TASK_DEAD))
 438                 return;  /* Better to leak the stack than to free prematurely */
 439
 440         account_kernel_stack(tsk, -1);
 441         free_thread_stack(tsk);
 442         tsk->stack = NULL;
 443 #ifdef CONFIG_VMAP_STACK
 444         tsk->stack_vm_area = NULL;
 445 #endif
 446 }
 447
 448 #ifdef CONFIG_THREAD_INFO_IN_TASK
 449 void put_task_stack(struct task_struct *tsk)
 450 {
 451         if (refcount_dec_and_test(&tsk->stack_refcount))
 452                 release_task_stack(tsk);
 453 }
 454 #endif
 455
 456 void free_task(struct task_struct *tsk)
 457 {
 458 #ifndef CONFIG_THREAD_INFO_IN_TASK
 459         /*
 460          * The task is finally done with both the stack and thread_info,
 461          * so free both.
 462          */
 463         release_task_stack(tsk);
 464 #else
 465         /*
 466          * If the task had a separate stack allocation, it should be gone
 467          * by now.
 468          */
 469         WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
 470 #endif
 471         rt_mutex_debug_task_free(tsk);
 472         ftrace_graph_exit_task(tsk);
 473         put_seccomp_filter(tsk);
 474         arch_release_task_struct(tsk);
 475         if (tsk->flags & PF_KTHREAD)
 476                 free_kthread_struct(tsk);
 477         free_task_struct(tsk);
 478 }
 479 EXPORT_SYMBOL(free_task);
 480
 481 #ifdef CONFIG_MMU
 482 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 483                                         struct mm_struct *oldmm)
 484 {
 485         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 486         struct rb_node **rb_link, *rb_parent;
 487         int retval;
 488         unsigned long charge;
 489         LIST_HEAD(uf);
 490
 491         uprobe_start_dup_mmap();
 492         if (down_write_killable(&oldmm->mmap_sem)) {
 493                 retval = -EINTR;
 494                 goto fail_uprobe_end;
 495         }
 496         flush_cache_dup_mm(oldmm);
 497         uprobe_dup_mmap(oldmm, mm);
 498         /*
 499          * Not linked in yet - no deadlock potential:
 500          */
 501         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
 502
 503         /* No ordering required: file already has been exposed. */
 504         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 505
 506         mm->total_vm = oldmm->total_vm;
 507         mm->data_vm = oldmm->data_vm;
 508         mm->exec_vm = oldmm->exec_vm;
 509         mm->stack_vm = oldmm->stack_vm;
 510
 511         rb_link = &mm->mm_rb.rb_node;
 512         rb_parent = NULL;
 513         pprev = &mm->mmap;
 514         retval = ksm_fork(mm, oldmm);
 515         if (retval)
 516                 goto out;
 517         retval = khugepaged_fork(mm, oldmm);
 518         if (retval)
 519                 goto out;
 520
 521         prev = NULL;
 522         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 523                 struct file *file;
 524
 525                 if (mpnt->vm_flags & VM_DONTCOPY) {
 526                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 527                         continue;
 528                 }
 529                 charge = 0;
 530                 /*
 531                  * Don't duplicate many vmas if we've been oom-killed (for
 532                  * example)
 533                  */
 534                 if (fatal_signal_pending(current)) {
 535                         retval = -EINTR;
 536                         goto out;
 537                 }
 538                 if (mpnt->vm_flags & VM_ACCOUNT) {
 539                         unsigned long len = vma_pages(mpnt);
 540
 541                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 542                                 goto fail_nomem;
 543                         charge = len;
 544                 }
 545                 tmp = vm_area_dup(mpnt);
 546                 if (!tmp)
 547                         goto fail_nomem;
 548                 retval = vma_dup_policy(mpnt, tmp);
 549                 if (retval)
 550                         goto fail_nomem_policy;
 551                 tmp->vm_mm = mm;
 552                 retval = dup_userfaultfd(tmp, &uf);
 553                 if (retval)
 554                         goto fail_nomem_anon_vma_fork;
 555                 if (tmp->vm_flags & VM_WIPEONFORK) {
 556                         /* VM_WIPEONFORK gets a clean slate in the child. */
 557                         tmp->anon_vma = NULL;
 558                         if (anon_vma_prepare(tmp))
 559                                 goto fail_nomem_anon_vma_fork;
 560                 } else if (anon_vma_fork(tmp, mpnt))
 561                         goto fail_nomem_anon_vma_fork;
 562                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 563                 tmp->vm_next = tmp->vm_prev = NULL;
 564                 file = tmp->vm_file;
 565                 if (file) {
 566                         struct inode *inode = file_inode(file);
 567                         struct address_space *mapping = file->f_mapping;
 568
 569                         get_file(file);
 570                         if (tmp->vm_flags & VM_DENYWRITE)
 571                                 atomic_dec(&inode->i_writecount);
 572                         i_mmap_lock_write(mapping);
 573                         if (tmp->vm_flags & VM_SHARED)
 574                                 atomic_inc(&mapping->i_mmap_writable);
 575                         flush_dcache_mmap_lock(mapping);
 576                         /* insert tmp into the share list, just after mpnt */
 577                         vma_interval_tree_insert_after(tmp, mpnt,
 578                                         &mapping->i_mmap);
 579                         flush_dcache_mmap_unlock(mapping);
 580                         i_mmap_unlock_write(mapping);
 581                 }
 582
 583                 /*
 584                  * Clear hugetlb-related page reserves for children. This only
 585                  * affects MAP_PRIVATE mappings. Faults generated by the child
 586                  * are not guaranteed to succeed, even if read-only
 587                  */
 588                 if (is_vm_hugetlb_page(tmp))
 589                         reset_vma_resv_huge_pages(tmp);
 590
 591                 /*
 592                  * Link in the new vma and copy the page table entries.
 593                  */
 594                 *pprev = tmp;
 595                 pprev = &tmp->vm_next;
 596                 tmp->vm_prev = prev;
 597                 prev = tmp;
 598
 599                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 600                 rb_link = &tmp->vm_rb.rb_right;
 601                 rb_parent = &tmp->vm_rb;
 602
 603                 mm->map_count++;
 604                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 605                         retval = copy_page_range(mm, oldmm, mpnt);
 606
 607                 if (tmp->vm_ops && tmp->vm_ops->open)
 608                         tmp->vm_ops->open(tmp);
 609
 610                 if (retval)
 611                         goto out;
 612         }
 613         /* a new mm has just been created */
 614         retval = arch_dup_mmap(oldmm, mm);
 615 out:
 616         up_write(&mm->mmap_sem);
 617         flush_tlb_mm(oldmm);
 618         up_write(&oldmm->mmap_sem);
 619         dup_userfaultfd_complete(&uf);
 620 fail_uprobe_end:
 621         uprobe_end_dup_mmap();
 622         return retval;
 623 fail_nomem_anon_vma_fork:
 624         mpol_put(vma_policy(tmp));
 625 fail_nomem_policy:
 626         vm_area_free(tmp);
 627 fail_nomem:
 628         retval = -ENOMEM;
 629         vm_unacct_memory(charge);
 630         goto out;
 631 }
 632
 633 static inline int mm_alloc_pgd(struct mm_struct *mm)
 634 {
 635         mm->pgd = pgd_alloc(mm);
 636         if (unlikely(!mm->pgd))
 637                 return -ENOMEM;
 638         return 0;
 639 }
 640
 641 static inline void mm_free_pgd(struct mm_struct *mm)
 642 {
 643         pgd_free(mm, mm->pgd);
 644 }
 645 #else
 646 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 647 {
 648         down_write(&oldmm->mmap_sem);
 649         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 650         up_write(&oldmm->mmap_sem);
 651         return 0;
 652 }
 653 #define mm_alloc_pgd(mm)        (0)
 654 #define mm_free_pgd(mm)
 655 #endif /* CONFIG_MMU */
 656
 657 static void check_mm(struct mm_struct *mm)
 658 {
 659         int i;
 660
 661         BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
 662                          "Please make sure 'struct resident_page_types[]' is updated as well");
 663
 664         for (i = 0; i < NR_MM_COUNTERS; i++) {
 665                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 666
 667                 if (unlikely(x))
 668                         pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
 669                                  mm, resident_page_types[i], x);
 670         }
 671
 672         if (mm_pgtables_bytes(mm))
 673                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 674                                 mm_pgtables_bytes(mm));
 675
 676 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 677         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 678 #endif
 679 }
 680
 681 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 682 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 683
 684 /*
 685  * Called when the last reference to the mm
 686  * is dropped: either by a lazy thread or by
 687  * mmput. Free the page directory and the mm.
 688  */
 689 void __mmdrop(struct mm_struct *mm)
 690 {
 691         BUG_ON(mm == &init_mm);
 692         WARN_ON_ONCE(mm == current->mm);
 693         WARN_ON_ONCE(mm == current->active_mm);
 694         mm_free_pgd(mm);
 695         destroy_context(mm);
 696         mmu_notifier_subscriptions_destroy(mm);
 697         check_mm(mm);
 698         put_user_ns(mm->user_ns);
 699         free_mm(mm);
 700 }
 701 EXPORT_SYMBOL_GPL(__mmdrop);
 702
 703 static void mmdrop_async_fn(struct work_struct *work)
 704 {
 705         struct mm_struct *mm;
 706
 707         mm = container_of(work, struct mm_struct, async_put_work);
 708         __mmdrop(mm);
 709 }
 710
 711 static void mmdrop_async(struct mm_struct *mm)
 712 {
 713         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 714                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 715                 schedule_work(&mm->async_put_work);
 716         }
 717 }
 718
 719 static inline void free_signal_struct(struct signal_struct *sig)
 720 {
 721         taskstats_tgid_free(sig);
 722         sched_autogroup_exit(sig);
 723         /*
 724          * __mmdrop is not safe to call from softirq context on x86 due to
 725          * pgd_dtor so postpone it to the async context
 726          */
 727         if (sig->oom_mm)
 728                 mmdrop_async(sig->oom_mm);
 729         kmem_cache_free(signal_cachep, sig);
 730 }
 731
 732 static inline void put_signal_struct(struct signal_struct *sig)
 733 {
 734         if (refcount_dec_and_test(&sig->sigcnt))
 735                 free_signal_struct(sig);
 736 }
 737
 738 void __put_task_struct(struct task_struct *tsk)
 739 {
 740         WARN_ON(!tsk->exit_state);
 741         WARN_ON(refcount_read(&tsk->usage));
 742         WARN_ON(tsk == current);
 743
 744         cgroup_free(tsk);
 745         task_numa_free(tsk, true);
 746         security_task_free(tsk);
 747         exit_creds(tsk);
 748         delayacct_tsk_free(tsk);
 749         put_signal_struct(tsk->signal);
 750
 751         if (!profile_handoff_task(tsk))
 752                 free_task(tsk);
 753 }
 754 EXPORT_SYMBOL_GPL(__put_task_struct);
 755
 756 void __init __weak arch_task_cache_init(void) { }
 757
 758 /*
 759  * set_max_threads
 760  */
 761 static void set_max_threads(unsigned int max_threads_suggested)
 762 {
 763         u64 threads;
 764         unsigned long nr_pages = totalram_pages();
 765
 766         /*
 767          * The number of threads shall be limited such that the thread
 768          * structures may only consume a small part of the available memory.
 769          */
 770         if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
 771                 threads = MAX_THREADS;
 772         else
 773                 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
 774                                     (u64) THREAD_SIZE * 8UL);
 775
 776         if (threads > max_threads_suggested)
 777                 threads = max_threads_suggested;
 778
 779         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 780 }
 781
 782 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 783 /* Initialized by the architecture: */
 784 int arch_task_struct_size __read_mostly;
 785 #endif
 786
 787 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 788 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 789 {
 790         /* Fetch thread_struct whitelist for the architecture. */
 791         arch_thread_struct_whitelist(offset, size);
 792
 793         /*
 794          * Handle zero-sized whitelist or empty thread_struct, otherwise
 795          * adjust offset to position of thread_struct in task_struct.
 796          */
 797         if (unlikely(*size == 0))
 798                 *offset = 0;
 799         else
 800                 *offset += offsetof(struct task_struct, thread);
 801 }
 802 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
 803
 804 void __init fork_init(void)
 805 {
 806         int i;
 807 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 808 #ifndef ARCH_MIN_TASKALIGN
 809 #define ARCH_MIN_TASKALIGN      0
 810 #endif
 811         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 812         unsigned long useroffset, usersize;
 813
 814         /* create a slab on which task_structs can be allocated */
 815         task_struct_whitelist(&useroffset, &usersize);
 816         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 817                         arch_task_struct_size, align,
 818                         SLAB_PANIC|SLAB_ACCOUNT,
 819                         useroffset, usersize, NULL);
 820 #endif
 821
 822         /* do the arch specific task caches init */
 823         arch_task_cache_init();
 824
 825         set_max_threads(MAX_THREADS);
 826
 827         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 828         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 829         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 830                 init_task.signal->rlim[RLIMIT_NPROC];
 831
 832         for (i = 0; i < UCOUNT_COUNTS; i++) {
 833                 init_user_ns.ucount_max[i] = max_threads/2;
 834         }
 835
 836 #ifdef CONFIG_VMAP_STACK
 837         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 838                           NULL, free_vm_stack_cache);
 839 #endif
 840
 841         lockdep_init_task(&init_task);
 842         uprobes_init();
 843 }
 844
 845 int __weak arch_dup_task_struct(struct task_struct *dst,
 846                                                struct task_struct *src)
 847 {
 848         *dst = *src;
 849         return 0;
 850 }
 851
 852 void set_task_stack_end_magic(struct task_struct *tsk)
 853 {
 854         unsigned long *stackend;
 855
 856         stackend = end_of_stack(tsk);
 857         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 858 }
 859
 860 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 861 {
 862         struct task_struct *tsk;
 863         unsigned long *stack;
 864         struct vm_struct *stack_vm_area __maybe_unused;
 865         int err;
 866
 867         if (node == NUMA_NO_NODE)
 868                 node = tsk_fork_get_node(orig);
 869         tsk = alloc_task_struct_node(node);
 870         if (!tsk)
 871                 return NULL;
 872
 873         stack = alloc_thread_stack_node(tsk, node);
 874         if (!stack)
 875                 goto free_tsk;
 876
 877         if (memcg_charge_kernel_stack(tsk))
 878                 goto free_stack;
 879
 880         stack_vm_area = task_stack_vm_area(tsk);
 881
 882         err = arch_dup_task_struct(tsk, orig);
 883
 884         /*
 885          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 886          * sure they're properly initialized before using any stack-related
 887          * functions again.
 888          */
 889         tsk->stack = stack;
 890 #ifdef CONFIG_VMAP_STACK
 891         tsk->stack_vm_area = stack_vm_area;
 892 #endif
 893 #ifdef CONFIG_THREAD_INFO_IN_TASK
 894         refcount_set(&tsk->stack_refcount, 1);
 895 #endif
 896
 897         if (err)
 898                 goto free_stack;
 899
 900 #ifdef CONFIG_SECCOMP
 901         /*
 902          * We must handle setting up seccomp filters once we're under
 903          * the sighand lock in case orig has changed between now and
 904          * then. Until then, filter must be NULL to avoid messing up
 905          * the usage counts on the error path calling free_task.
 906          */
 907         tsk->seccomp.filter = NULL;
 908 #endif
 909
 910         setup_thread_stack(tsk, orig);
 911         clear_user_return_notifier(tsk);
 912         clear_tsk_need_resched(tsk);
 913         set_task_stack_end_magic(tsk);
 914
 915 #ifdef CONFIG_STACKPROTECTOR
 916         tsk->stack_canary = get_random_canary();
 917 #endif
 918         if (orig->cpus_ptr == &orig->cpus_mask)
 919                 tsk->cpus_ptr = &tsk->cpus_mask;
 920
 921         /*
 922          * One for the user space visible state that goes away when reaped.
 923          * One for the scheduler.
 924          */
 925         refcount_set(&tsk->rcu_users, 2);
 926         /* One for the rcu users */
 927         refcount_set(&tsk->usage, 1);
 928 #ifdef CONFIG_BLK_DEV_IO_TRACE
 929         tsk->btrace_seq = 0;
 930 #endif
 931         tsk->splice_pipe = NULL;
 932         tsk->task_frag.page = NULL;
 933         tsk->wake_q.next = NULL;
 934
 935         account_kernel_stack(tsk, 1);
 936
 937         kcov_task_init(tsk);
 938
 939 #ifdef CONFIG_FAULT_INJECTION
 940         tsk->fail_nth = 0;
 941 #endif
 942
 943 #ifdef CONFIG_BLK_CGROUP
 944         tsk->throttle_queue = NULL;
 945         tsk->use_memdelay = 0;
 946 #endif
 947
 948 #ifdef CONFIG_MEMCG
 949         tsk->active_memcg = NULL;
 950 #endif
 951         return tsk;
 952
 953 free_stack:
 954         free_thread_stack(tsk);
 955 free_tsk:
 956         free_task_struct(tsk);
 957         return NULL;
 958 }
 959
 960 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 961
 962 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 963
 964 static int __init coredump_filter_setup(char *s)
 965 {
 966         default_dump_filter =
 967                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 968                 MMF_DUMP_FILTER_MASK;
 969         return 1;
 970 }
 971
 972 __setup("coredump_filter=", coredump_filter_setup);
 973
 974 #include <linux/init_task.h>
 975
 976 static void mm_init_aio(struct mm_struct *mm)
 977 {
 978 #ifdef CONFIG_AIO
 979         spin_lock_init(&mm->ioctx_lock);
 980         mm->ioctx_table = NULL;
 981 #endif
 982 }
 983
 984 static __always_inline void mm_clear_owner(struct mm_struct *mm,
 985                                            struct task_struct *p)
 986 {
 987 #ifdef CONFIG_MEMCG
 988         if (mm->owner == p)
 989                 WRITE_ONCE(mm->owner, NULL);
 990 #endif
 991 }
 992
 993 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 994 {
 995 #ifdef CONFIG_MEMCG
 996         mm->owner = p;
 997 #endif
 998 }
 999
1000 static void mm_init_uprobes_state(struct mm_struct *mm)
1001 {
1002 #ifdef CONFIG_UPROBES
1003         mm->uprobes_state.xol_area = NULL;
1004 #endif
1005 }
1006
1007 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1008         struct user_namespace *user_ns)
1009 {
1010         mm->mmap = NULL;
1011         mm->mm_rb = RB_ROOT;
1012         mm->vmacache_seqnum = 0;
1013         atomic_set(&mm->mm_users, 1);
1014         atomic_set(&mm->mm_count, 1);
1015         init_rwsem(&mm->mmap_sem);
1016         INIT_LIST_HEAD(&mm->mmlist);
1017         mm->core_state = NULL;
1018         mm_pgtables_bytes_init(mm);
1019         mm->map_count = 0;
1020         mm->locked_vm = 0;
1021         atomic64_set(&mm->pinned_vm, 0);
1022         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1023         spin_lock_init(&mm->page_table_lock);
1024         spin_lock_init(&mm->arg_lock);
1025         mm_init_cpumask(mm);
1026         mm_init_aio(mm);
1027         mm_init_owner(mm, p);
1028         RCU_INIT_POINTER(mm->exe_file, NULL);
1029         mmu_notifier_subscriptions_init(mm);
1030         init_tlb_flush_pending(mm);
1031 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1032         mm->pmd_huge_pte = NULL;
1033 #endif
1034         mm_init_uprobes_state(mm);
1035
1036         if (current->mm) {
1037                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1038                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1039         } else {
1040                 mm->flags = default_dump_filter;
1041                 mm->def_flags = 0;
1042         }
1043
1044         if (mm_alloc_pgd(mm))
1045                 goto fail_nopgd;
1046
1047         if (init_new_context(p, mm))
1048                 goto fail_nocontext;
1049
1050         mm->user_ns = get_user_ns(user_ns);
1051         return mm;
1052
1053 fail_nocontext:
1054         mm_free_pgd(mm);
1055 fail_nopgd:
1056         free_mm(mm);
1057         return NULL;
1058 }
1059
1060 /*
1061  * Allocate and initialize an mm_struct.
1062  */
1063 struct mm_struct *mm_alloc(void)
1064 {
1065         struct mm_struct *mm;
1066
1067         mm = allocate_mm();
1068         if (!mm)
1069                 return NULL;
1070
1071         memset(mm, 0, sizeof(*mm));
1072         return mm_init(mm, current, current_user_ns());
1073 }
1074
1075 static inline void __mmput(struct mm_struct *mm)
1076 {
1077         VM_BUG_ON(atomic_read(&mm->mm_users));
1078
1079         uprobe_clear_state(mm);
1080         exit_aio(mm);
1081         ksm_exit(mm);
1082         khugepaged_exit(mm); /* must run before exit_mmap */
1083         exit_mmap(mm);
1084         mm_put_huge_zero_page(mm);
1085         set_mm_exe_file(mm, NULL);
1086         if (!list_empty(&mm->mmlist)) {
1087                 spin_lock(&mmlist_lock);
1088                 list_del(&mm->mmlist);
1089                 spin_unlock(&mmlist_lock);
1090         }
1091         if (mm->binfmt)
1092                 module_put(mm->binfmt->module);
1093         mmdrop(mm);
1094 }
1095
1096 /*
1097  * Decrement the use count and release all resources for an mm.
1098  */
1099 void mmput(struct mm_struct *mm)
1100 {
1101         might_sleep();
1102
1103         if (atomic_dec_and_test(&mm->mm_users))
1104                 __mmput(mm);
1105 }
1106 EXPORT_SYMBOL_GPL(mmput);
1107
1108 #ifdef CONFIG_MMU
1109 static void mmput_async_fn(struct work_struct *work)
1110 {
1111         struct mm_struct *mm = container_of(work, struct mm_struct,
1112                                             async_put_work);
1113
1114         __mmput(mm);
1115 }
1116
1117 void mmput_async(struct mm_struct *mm)
1118 {
1119         if (atomic_dec_and_test(&mm->mm_users)) {
1120                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1121                 schedule_work(&mm->async_put_work);
1122         }
1123 }
1124 #endif
1125
1126 /**
1127  * set_mm_exe_file - change a reference to the mm's executable file
1128  *
1129  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1130  *
1131  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1132  * invocations: in mmput() nobody alive left, in execve task is single
1133  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1134  * mm->exe_file, but does so without using set_mm_exe_file() in order
1135  * to do avoid the need for any locks.
1136  */
1137 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1138 {
1139         struct file *old_exe_file;
1140
1141         /*
1142          * It is safe to dereference the exe_file without RCU as
1143          * this function is only called if nobody else can access
1144          * this mm -- see comment above for justification.
1145          */
1146         old_exe_file = rcu_dereference_raw(mm->exe_file);
1147
1148         if (new_exe_file)
1149                 get_file(new_exe_file);
1150         rcu_assign_pointer(mm->exe_file, new_exe_file);
1151         if (old_exe_file)
1152                 fput(old_exe_file);
1153 }
1154
1155 /**
1156  * get_mm_exe_file - acquire a reference to the mm's executable file
1157  *
1158  * Returns %NULL if mm has no associated executable file.
1159  * User must release file via fput().
1160  */
1161 struct file *get_mm_exe_file(struct mm_struct *mm)
1162 {
1163         struct file *exe_file;
1164
1165         rcu_read_lock();
1166         exe_file = rcu_dereference(mm->exe_file);
1167         if (exe_file && !get_file_rcu(exe_file))
1168                 exe_file = NULL;
1169         rcu_read_unlock();
1170         return exe_file;
1171 }
1172 EXPORT_SYMBOL(get_mm_exe_file);
1173
1174 /**
1175  * get_task_exe_file - acquire a reference to the task's executable file
1176  *
1177  * Returns %NULL if task's mm (if any) has no associated executable file or
1178  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1179  * User must release file via fput().
1180  */
1181 struct file *get_task_exe_file(struct task_struct *task)
1182 {
1183         struct file *exe_file = NULL;
1184         struct mm_struct *mm;
1185
1186         task_lock(task);
1187         mm = task->mm;
1188         if (mm) {
1189                 if (!(task->flags & PF_KTHREAD))
1190                         exe_file = get_mm_exe_file(mm);
1191         }
1192         task_unlock(task);
1193         return exe_file;
1194 }
1195 EXPORT_SYMBOL(get_task_exe_file);
1196
1197 /**
1198  * get_task_mm - acquire a reference to the task's mm
1199  *
1200  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1201  * this kernel workthread has transiently adopted a user mm with use_mm,
1202  * to do its AIO) is not set and if so returns a reference to it, after
1203  * bumping up the use count.  User must release the mm via mmput()
1204  * after use.  Typically used by /proc and ptrace.
1205  */
1206 struct mm_struct *get_task_mm(struct task_struct *task)
1207 {
1208         struct mm_struct *mm;
1209
1210         task_lock(task);
1211         mm = task->mm;
1212         if (mm) {
1213                 if (task->flags & PF_KTHREAD)
1214                         mm = NULL;
1215                 else
1216                         mmget(mm);
1217         }
1218         task_unlock(task);
1219         return mm;
1220 }
1221 EXPORT_SYMBOL_GPL(get_task_mm);
1222
1223 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1224 {
1225         struct mm_struct *mm;
1226         int err;
1227
1228         err =  mutex_lock_killable(&task->signal->exec_update_mutex);
1229         if (err)
1230                 return ERR_PTR(err);
1231
1232         mm = get_task_mm(task);
1233         if (mm && mm != current->mm &&
1234                         !ptrace_may_access(task, mode)) {
1235                 mmput(mm);
1236                 mm = ERR_PTR(-EACCES);
1237         }
1238         mutex_unlock(&task->signal->exec_update_mutex);
1239
1240         return mm;
1241 }
1242
1243 static void complete_vfork_done(struct task_struct *tsk)
1244 {
1245         struct completion *vfork;
1246
1247         task_lock(tsk);
1248         vfork = tsk->vfork_done;
1249         if (likely(vfork)) {
1250                 tsk->vfork_done = NULL;
1251                 complete(vfork);
1252         }
1253         task_unlock(tsk);
1254 }
1255
1256 static int wait_for_vfork_done(struct task_struct *child,
1257                                 struct completion *vfork)
1258 {
1259         int killed;
1260
1261         freezer_do_not_count();
1262         cgroup_enter_frozen();
1263         killed = wait_for_completion_killable(vfork);
1264         cgroup_leave_frozen(false);
1265         freezer_count();
1266
1267         if (killed) {
1268                 task_lock(child);
1269                 child->vfork_done = NULL;
1270                 task_unlock(child);
1271         }
1272
1273         put_task_struct(child);
1274         return killed;
1275 }
1276
1277 /* Please note the differences between mmput and mm_release.
1278  * mmput is called whenever we stop holding onto a mm_struct,
1279  * error success whatever.
1280  *
1281  * mm_release is called after a mm_struct has been removed
1282  * from the current process.
1283  *
1284  * This difference is important for error handling, when we
1285  * only half set up a mm_struct for a new process and need to restore
1286  * the old one.  Because we mmput the new mm_struct before
1287  * restoring the old one. . .
1288  * Eric Biederman 10 January 1998
1289  */
1290 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1291 {
1292         uprobe_free_utask(tsk);
1293
1294         /* Get rid of any cached register state */
1295         deactivate_mm(tsk, mm);
1296
1297         /*
1298          * Signal userspace if we're not exiting with a core dump
1299          * because we want to leave the value intact for debugging
1300          * purposes.
1301          */
1302         if (tsk->clear_child_tid) {
1303                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1304                     atomic_read(&mm->mm_users) > 1) {
1305                         /*
1306                          * We don't check the error code - if userspace has
1307                          * not set up a proper pointer then tough luck.
1308                          */
1309                         put_user(0, tsk->clear_child_tid);
1310                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1311                                         1, NULL, NULL, 0, 0);
1312                 }
1313                 tsk->clear_child_tid = NULL;
1314         }
1315
1316         /*
1317          * All done, finally we can wake up parent and return this mm to him.
1318          * Also kthread_stop() uses this completion for synchronization.
1319          */
1320         if (tsk->vfork_done)
1321                 complete_vfork_done(tsk);
1322 }
1323
1324 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1325 {
1326         futex_exit_release(tsk);
1327         mm_release(tsk, mm);
1328 }
1329
1330 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1331 {
1332         futex_exec_release(tsk);
1333         mm_release(tsk, mm);
1334 }
1335
1336 /**
1337  * dup_mm() - duplicates an existing mm structure
1338  * @tsk: the task_struct with which the new mm will be associated.
1339  * @oldmm: the mm to duplicate.
1340  *
1341  * Allocates a new mm structure and duplicates the provided @oldmm structure
1342  * content into it.
1343  *
1344  * Return: the duplicated mm or NULL on failure.
1345  */
1346 static struct mm_struct *dup_mm(struct task_struct *tsk,
1347                                 struct mm_struct *oldmm)
1348 {
1349         struct mm_struct *mm;
1350         int err;
1351
1352         mm = allocate_mm();
1353         if (!mm)
1354                 goto fail_nomem;
1355
1356         memcpy(mm, oldmm, sizeof(*mm));
1357
1358         if (!mm_init(mm, tsk, mm->user_ns))
1359                 goto fail_nomem;
1360
1361         err = dup_mmap(mm, oldmm);
1362         if (err)
1363                 goto free_pt;
1364
1365         mm->hiwater_rss = get_mm_rss(mm);
1366         mm->hiwater_vm = mm->total_vm;
1367
1368         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1369                 goto free_pt;
1370
1371         return mm;
1372
1373 free_pt:
1374         /* don't put binfmt in mmput, we haven't got module yet */
1375         mm->binfmt = NULL;
1376         mm_init_owner(mm, NULL);
1377         mmput(mm);
1378
1379 fail_nomem:
1380         return NULL;
1381 }
1382
1383 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1384 {
1385         struct mm_struct *mm, *oldmm;
1386         int retval;
1387
1388         tsk->min_flt = tsk->maj_flt = 0;
1389         tsk->nvcsw = tsk->nivcsw = 0;
1390 #ifdef CONFIG_DETECT_HUNG_TASK
1391         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1392         tsk->last_switch_time = 0;
1393 #endif
1394
1395         tsk->mm = NULL;
1396         tsk->active_mm = NULL;
1397
1398         /*
1399          * Are we cloning a kernel thread?
1400          *
1401          * We need to steal a active VM for that..
1402          */
1403         oldmm = current->mm;
1404         if (!oldmm)
1405                 return 0;
1406
1407         /* initialize the new vmacache entries */
1408         vmacache_flush(tsk);
1409
1410         if (clone_flags & CLONE_VM) {
1411                 mmget(oldmm);
1412                 mm = oldmm;
1413                 goto good_mm;
1414         }
1415
1416         retval = -ENOMEM;
1417         mm = dup_mm(tsk, current->mm);
1418         if (!mm)
1419                 goto fail_nomem;
1420
1421 good_mm:
1422         tsk->mm = mm;
1423         tsk->active_mm = mm;
1424         return 0;
1425
1426 fail_nomem:
1427         return retval;
1428 }
1429
1430 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1431 {
1432         struct fs_struct *fs = current->fs;
1433         if (clone_flags & CLONE_FS) {
1434                 /* tsk->fs is already what we want */
1435                 spin_lock(&fs->lock);
1436                 if (fs->in_exec) {
1437                         spin_unlock(&fs->lock);
1438                         return -EAGAIN;
1439                 }
1440                 fs->users++;
1441                 spin_unlock(&fs->lock);
1442                 return 0;
1443         }
1444         tsk->fs = copy_fs_struct(fs);
1445         if (!tsk->fs)
1446                 return -ENOMEM;
1447         return 0;
1448 }
1449
1450 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1451 {
1452         struct files_struct *oldf, *newf;
1453         int error = 0;
1454
1455         /*
1456          * A background process may not have any files ...
1457          */
1458         oldf = current->files;
1459         if (!oldf)
1460                 goto out;
1461
1462         if (clone_flags & CLONE_FILES) {
1463                 atomic_inc(&oldf->count);
1464                 goto out;
1465         }
1466
1467         newf = dup_fd(oldf, &error);
1468         if (!newf)
1469                 goto out;
1470
1471         tsk->files = newf;
1472         error = 0;
1473 out:
1474         return error;
1475 }
1476
1477 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1478 {
1479 #ifdef CONFIG_BLOCK
1480         struct io_context *ioc = current->io_context;
1481         struct io_context *new_ioc;
1482
1483         if (!ioc)
1484                 return 0;
1485         /*
1486          * Share io context with parent, if CLONE_IO is set
1487          */
1488         if (clone_flags & CLONE_IO) {
1489                 ioc_task_link(ioc);
1490                 tsk->io_context = ioc;
1491         } else if (ioprio_valid(ioc->ioprio)) {
1492                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1493                 if (unlikely(!new_ioc))
1494                         return -ENOMEM;
1495
1496                 new_ioc->ioprio = ioc->ioprio;
1497                 put_io_context(new_ioc);
1498         }
1499 #endif
1500         return 0;
1501 }
1502
1503 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1504 {
1505         struct sighand_struct *sig;
1506
1507         if (clone_flags & CLONE_SIGHAND) {
1508                 refcount_inc(&current->sighand->count);
1509                 return 0;
1510         }
1511         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1512         RCU_INIT_POINTER(tsk->sighand, sig);
1513         if (!sig)
1514                 return -ENOMEM;
1515
1516         refcount_set(&sig->count, 1);
1517         spin_lock_irq(&current->sighand->siglock);
1518         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1519         spin_unlock_irq(&current->sighand->siglock);
1520
1521         /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1522         if (clone_flags & CLONE_CLEAR_SIGHAND)
1523                 flush_signal_handlers(tsk, 0);
1524
1525         return 0;
1526 }
1527
1528 void __cleanup_sighand(struct sighand_struct *sighand)
1529 {
1530         if (refcount_dec_and_test(&sighand->count)) {
1531                 signalfd_cleanup(sighand);
1532                 /*
1533                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1534                  * without an RCU grace period, see __lock_task_sighand().
1535                  */
1536                 kmem_cache_free(sighand_cachep, sighand);
1537         }
1538 }
1539
1540 /*
1541  * Initialize POSIX timer handling for a thread group.
1542  */
1543 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1544 {
1545         struct posix_cputimers *pct = &sig->posix_cputimers;
1546         unsigned long cpu_limit;
1547
1548         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1549         posix_cputimers_group_init(pct, cpu_limit);
1550 }
1551
1552 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1553 {
1554         struct signal_struct *sig;
1555
1556         if (clone_flags & CLONE_THREAD)
1557                 return 0;
1558
1559         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1560         tsk->signal = sig;
1561         if (!sig)
1562                 return -ENOMEM;
1563
1564         sig->nr_threads = 1;
1565         atomic_set(&sig->live, 1);
1566         refcount_set(&sig->sigcnt, 1);
1567
1568         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1569         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1570         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1571
1572         init_waitqueue_head(&sig->wait_chldexit);
1573         sig->curr_target = tsk;
1574         init_sigpending(&sig->shared_pending);
1575         INIT_HLIST_HEAD(&sig->multiprocess);
1576         seqlock_init(&sig->stats_lock);
1577         prev_cputime_init(&sig->prev_cputime);
1578
1579 #ifdef CONFIG_POSIX_TIMERS
1580         INIT_LIST_HEAD(&sig->posix_timers);
1581         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1582         sig->real_timer.function = it_real_fn;
1583 #endif
1584
1585         task_lock(current->group_leader);
1586         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1587         task_unlock(current->group_leader);
1588
1589         posix_cpu_timers_init_group(sig);
1590
1591         tty_audit_fork(sig);
1592         sched_autogroup_fork(sig);
1593
1594         sig->oom_score_adj = current->signal->oom_score_adj;
1595         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1596
1597         mutex_init(&sig->cred_guard_mutex);
1598         mutex_init(&sig->exec_update_mutex);
1599
1600         return 0;
1601 }
1602
1603 static void copy_seccomp(struct task_struct *p)
1604 {
1605 #ifdef CONFIG_SECCOMP
1606         /*
1607          * Must be called with sighand->lock held, which is common to
1608          * all threads in the group. Holding cred_guard_mutex is not
1609          * needed because this new task is not yet running and cannot
1610          * be racing exec.
1611          */
1612         assert_spin_locked(&current->sighand->siglock);
1613
1614         /* Ref-count the new filter user, and assign it. */
1615         get_seccomp_filter(current);
1616         p->seccomp = current->seccomp;
1617
1618         /*
1619          * Explicitly enable no_new_privs here in case it got set
1620          * between the task_struct being duplicated and holding the
1621          * sighand lock. The seccomp state and nnp must be in sync.
1622          */
1623         if (task_no_new_privs(current))
1624                 task_set_no_new_privs(p);
1625
1626         /*
1627          * If the parent gained a seccomp mode after copying thread
1628          * flags and between before we held the sighand lock, we have
1629          * to manually enable the seccomp thread flag here.
1630          */
1631         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1632                 set_tsk_thread_flag(p, TIF_SECCOMP);
1633 #endif
1634 }
1635
1636 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1637 {
1638         current->clear_child_tid = tidptr;
1639
1640         return task_pid_vnr(current);
1641 }
1642
1643 static void rt_mutex_init_task(struct task_struct *p)
1644 {
1645         raw_spin_lock_init(&p->pi_lock);
1646 #ifdef CONFIG_RT_MUTEXES
1647         p->pi_waiters = RB_ROOT_CACHED;
1648         p->pi_top_task = NULL;
1649         p->pi_blocked_on = NULL;
1650 #endif
1651 }
1652
1653 static inline void init_task_pid_links(struct task_struct *task)
1654 {
1655         enum pid_type type;
1656
1657         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1658                 INIT_HLIST_NODE(&task->pid_links[type]);
1659         }
1660 }
1661
1662 static inline void
1663 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1664 {
1665         if (type == PIDTYPE_PID)
1666                 task->thread_pid = pid;
1667         else
1668                 task->signal->pids[type] = pid;
1669 }
1670
1671 static inline void rcu_copy_process(struct task_struct *p)
1672 {
1673 #ifdef CONFIG_PREEMPT_RCU
1674         p->rcu_read_lock_nesting = 0;
1675         p->rcu_read_unlock_special.s = 0;
1676         p->rcu_blocked_node = NULL;
1677         INIT_LIST_HEAD(&p->rcu_node_entry);
1678 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1679 #ifdef CONFIG_TASKS_RCU
1680         p->rcu_tasks_holdout = false;
1681         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1682         p->rcu_tasks_idle_cpu = -1;
1683 #endif /* #ifdef CONFIG_TASKS_RCU */
1684 }
1685
1686 struct pid *pidfd_pid(const struct file *file)
1687 {
1688         if (file->f_op == &pidfd_fops)
1689                 return file->private_data;
1690
1691         return ERR_PTR(-EBADF);
1692 }
1693
1694 static int pidfd_release(struct inode *inode, struct file *file)
1695 {
1696         struct pid *pid = file->private_data;
1697
1698         file->private_data = NULL;
1699         put_pid(pid);
1700         return 0;
1701 }
1702
1703 #ifdef CONFIG_PROC_FS
1704 /**
1705  * pidfd_show_fdinfo - print information about a pidfd
1706  * @m: proc fdinfo file
1707  * @f: file referencing a pidfd
1708  *
1709  * Pid:
1710  * This function will print the pid that a given pidfd refers to in the
1711  * pid namespace of the procfs instance.
1712  * If the pid namespace of the process is not a descendant of the pid
1713  * namespace of the procfs instance 0 will be shown as its pid. This is
1714  * similar to calling getppid() on a process whose parent is outside of
1715  * its pid namespace.
1716  *
1717  * NSpid:
1718  * If pid namespaces are supported then this function will also print
1719  * the pid of a given pidfd refers to for all descendant pid namespaces
1720  * starting from the current pid namespace of the instance, i.e. the
1721  * Pid field and the first entry in the NSpid field will be identical.
1722  * If the pid namespace of the process is not a descendant of the pid
1723  * namespace of the procfs instance 0 will be shown as its first NSpid
1724  * entry and no others will be shown.
1725  * Note that this differs from the Pid and NSpid fields in
1726  * /proc/<pid>/status where Pid and NSpid are always shown relative to
1727  * the  pid namespace of the procfs instance. The difference becomes
1728  * obvious when sending around a pidfd between pid namespaces from a
1729  * different branch of the tree, i.e. where no ancestoral relation is
1730  * present between the pid namespaces:
1731  * - create two new pid namespaces ns1 and ns2 in the initial pid
1732  *   namespace (also take care to create new mount namespaces in the
1733  *   new pid namespace and mount procfs)
1734  * - create a process with a pidfd in ns1
1735  * - send pidfd from ns1 to ns2
1736  * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1737  *   have exactly one entry, which is 0
1738  */
1739 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1740 {
1741         struct pid *pid = f->private_data;
1742         struct pid_namespace *ns;
1743         pid_t nr = -1;
1744
1745         if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1746                 ns = proc_pid_ns(file_inode(m->file));
1747                 nr = pid_nr_ns(pid, ns);
1748         }
1749
1750         seq_put_decimal_ll(m, "Pid:\t", nr);
1751
1752 #ifdef CONFIG_PID_NS
1753         seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1754         if (nr > 0) {
1755                 int i;
1756
1757                 /* If nr is non-zero it means that 'pid' is valid and that
1758                  * ns, i.e. the pid namespace associated with the procfs
1759                  * instance, is in the pid namespace hierarchy of pid.
1760                  * Start at one below the already printed level.
1761                  */
1762                 for (i = ns->level + 1; i <= pid->level; i++)
1763                         seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1764         }
1765 #endif
1766         seq_putc(m, '\n');
1767 }
1768 #endif
1769
1770 /*
1771  * Poll support for process exit notification.
1772  */
1773 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1774 {
1775         struct task_struct *task;
1776         struct pid *pid = file->private_data;
1777         __poll_t poll_flags = 0;
1778
1779         poll_wait(file, &pid->wait_pidfd, pts);
1780
1781         rcu_read_lock();
1782         task = pid_task(pid, PIDTYPE_PID);
1783         /*
1784          * Inform pollers only when the whole thread group exits.
1785          * If the thread group leader exits before all other threads in the
1786          * group, then poll(2) should block, similar to the wait(2) family.
1787          */
1788         if (!task || (task->exit_state && thread_group_empty(task)))
1789                 poll_flags = EPOLLIN | EPOLLRDNORM;
1790         rcu_read_unlock();
1791
1792         return poll_flags;
1793 }
1794
1795 const struct file_operations pidfd_fops = {
1796         .release = pidfd_release,
1797         .poll = pidfd_poll,
1798 #ifdef CONFIG_PROC_FS
1799         .show_fdinfo = pidfd_show_fdinfo,
1800 #endif
1801 };
1802
1803 static void __delayed_free_task(struct rcu_head *rhp)
1804 {
1805         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1806
1807         free_task(tsk);
1808 }
1809
1810 static __always_inline void delayed_free_task(struct task_struct *tsk)
1811 {
1812         if (IS_ENABLED(CONFIG_MEMCG))
1813                 call_rcu(&tsk->rcu, __delayed_free_task);
1814         else
1815                 free_task(tsk);
1816 }
1817
1818 /*
1819  * This creates a new process as a copy of the old one,
1820  * but does not actually start it yet.
1821  *
1822  * It copies the registers, and all the appropriate
1823  * parts of the process environment (as per the clone
1824  * flags). The actual kick-off is left to the caller.
1825  */
1826 static __latent_entropy struct task_struct *copy_process(
1827                                         struct pid *pid,
1828                                         int trace,
1829                                         int node,
1830                                         struct kernel_clone_args *args)
1831 {
1832         int pidfd = -1, retval;
1833         struct task_struct *p;
1834         struct multiprocess_signals delayed;
1835         struct file *pidfile = NULL;
1836         u64 clone_flags = args->flags;
1837         struct nsproxy *nsp = current->nsproxy;
1838
1839         /*
1840          * Don't allow sharing the root directory with processes in a different
1841          * namespace
1842          */
1843         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1844                 return ERR_PTR(-EINVAL);
1845
1846         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1847                 return ERR_PTR(-EINVAL);
1848
1849         /*
1850          * Thread groups must share signals as well, and detached threads
1851          * can only be started up within the thread group.
1852          */
1853         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1854                 return ERR_PTR(-EINVAL);
1855
1856         /*
1857          * Shared signal handlers imply shared VM. By way of the above,
1858          * thread groups also imply shared VM. Blocking this case allows
1859          * for various simplifications in other code.
1860          */
1861         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1862                 return ERR_PTR(-EINVAL);
1863
1864         /*
1865          * Siblings of global init remain as zombies on exit since they are
1866          * not reaped by their parent (swapper). To solve this and to avoid
1867          * multi-rooted process trees, prevent global and container-inits
1868          * from creating siblings.
1869          */
1870         if ((clone_flags & CLONE_PARENT) &&
1871                                 current->signal->flags & SIGNAL_UNKILLABLE)
1872                 return ERR_PTR(-EINVAL);
1873
1874         /*
1875          * If the new process will be in a different pid or user namespace
1876          * do not allow it to share a thread group with the forking task.
1877          */
1878         if (clone_flags & CLONE_THREAD) {
1879                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1880                     (task_active_pid_ns(current) != nsp->pid_ns_for_children))
1881                         return ERR_PTR(-EINVAL);
1882         }
1883
1884         /*
1885          * If the new process will be in a different time namespace
1886          * do not allow it to share VM or a thread group with the forking task.
1887          */
1888         if (clone_flags & (CLONE_THREAD | CLONE_VM)) {
1889                 if (nsp->time_ns != nsp->time_ns_for_children)
1890                         return ERR_PTR(-EINVAL);
1891         }
1892
1893         if (clone_flags & CLONE_PIDFD) {
1894                 /*
1895                  * - CLONE_DETACHED is blocked so that we can potentially
1896                  *   reuse it later for CLONE_PIDFD.
1897                  * - CLONE_THREAD is blocked until someone really needs it.
1898                  */
1899                 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1900                         return ERR_PTR(-EINVAL);
1901         }
1902
1903         /*
1904          * Force any signals received before this point to be delivered
1905          * before the fork happens.  Collect up signals sent to multiple
1906          * processes that happen during the fork and delay them so that
1907          * they appear to happen after the fork.
1908          */
1909         sigemptyset(&delayed.signal);
1910         INIT_HLIST_NODE(&delayed.node);
1911
1912         spin_lock_irq(&current->sighand->siglock);
1913         if (!(clone_flags & CLONE_THREAD))
1914                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1915         recalc_sigpending();
1916         spin_unlock_irq(&current->sighand->siglock);
1917         retval = -ERESTARTNOINTR;
1918         if (signal_pending(current))
1919                 goto fork_out;
1920
1921         retval = -ENOMEM;
1922         p = dup_task_struct(current, node);
1923         if (!p)
1924                 goto fork_out;
1925
1926         /*
1927          * This _must_ happen before we call free_task(), i.e. before we jump
1928          * to any of the bad_fork_* labels. This is to avoid freeing
1929          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1930          * kernel threads (PF_KTHREAD).
1931          */
1932         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1933         /*
1934          * Clear TID on mm_release()?
1935          */
1936         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1937
1938         ftrace_graph_init_task(p);
1939
1940         rt_mutex_init_task(p);
1941
1942 #ifdef CONFIG_PROVE_LOCKING
1943         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1944         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1945 #endif
1946         retval = -EAGAIN;
1947         if (atomic_read(&p->real_cred->user->processes) >=
1948                         task_rlimit(p, RLIMIT_NPROC)) {
1949                 if (p->real_cred->user != INIT_USER &&
1950                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1951                         goto bad_fork_free;
1952         }
1953         current->flags &= ~PF_NPROC_EXCEEDED;
1954
1955         retval = copy_creds(p, clone_flags);
1956         if (retval < 0)
1957                 goto bad_fork_free;
1958
1959         /*
1960          * If multiple threads are within copy_process(), then this check
1961          * triggers too late. This doesn't hurt, the check is only there
1962          * to stop root fork bombs.
1963          */
1964         retval = -EAGAIN;
1965         if (nr_threads >= max_threads)
1966                 goto bad_fork_cleanup_count;
1967
1968         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1969         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1970         p->flags |= PF_FORKNOEXEC;
1971         INIT_LIST_HEAD(&p->children);
1972         INIT_LIST_HEAD(&p->sibling);
1973         rcu_copy_process(p);
1974         p->vfork_done = NULL;
1975         spin_lock_init(&p->alloc_lock);
1976
1977         init_sigpending(&p->pending);
1978
1979         p->utime = p->stime = p->gtime = 0;
1980 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1981         p->utimescaled = p->stimescaled = 0;
1982 #endif
1983         prev_cputime_init(&p->prev_cputime);
1984
1985 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1986         seqcount_init(&p->vtime.seqcount);
1987         p->vtime.starttime = 0;
1988         p->vtime.state = VTIME_INACTIVE;
1989 #endif
1990
1991 #if defined(SPLIT_RSS_COUNTING)
1992         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1993 #endif
1994
1995         p->default_timer_slack_ns = current->timer_slack_ns;
1996
1997 #ifdef CONFIG_PSI
1998         p->psi_flags = 0;
1999 #endif
2000
2001         task_io_accounting_init(&p->ioac);
2002         acct_clear_integrals(p);
2003
2004         posix_cputimers_init(&p->posix_cputimers);
2005
2006         p->io_context = NULL;
2007         audit_set_context(p, NULL);
2008         cgroup_fork(p);
2009 #ifdef CONFIG_NUMA
2010         p->mempolicy = mpol_dup(p->mempolicy);
2011         if (IS_ERR(p->mempolicy)) {
2012                 retval = PTR_ERR(p->mempolicy);
2013                 p->mempolicy = NULL;
2014                 goto bad_fork_cleanup_threadgroup_lock;
2015         }
2016 #endif
2017 #ifdef CONFIG_CPUSETS
2018         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2019         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2020         seqcount_init(&p->mems_allowed_seq);
2021 #endif
2022 #ifdef CONFIG_TRACE_IRQFLAGS
2023         p->irq_events = 0;
2024         p->hardirqs_enabled = 0;
2025         p->hardirq_enable_ip = 0;
2026         p->hardirq_enable_event = 0;
2027         p->hardirq_disable_ip = _THIS_IP_;
2028         p->hardirq_disable_event = 0;
2029         p->softirqs_enabled = 1;
2030         p->softirq_enable_ip = _THIS_IP_;
2031         p->softirq_enable_event = 0;
2032         p->softirq_disable_ip = 0;
2033         p->softirq_disable_event = 0;
2034         p->hardirq_context = 0;
2035         p->softirq_context = 0;
2036 #endif
2037
2038         p->pagefault_disabled = 0;
2039
2040 #ifdef CONFIG_LOCKDEP
2041         lockdep_init_task(p);
2042 #endif
2043
2044 #ifdef CONFIG_DEBUG_MUTEXES
2045         p->blocked_on = NULL; /* not blocked yet */
2046 #endif
2047 #ifdef CONFIG_BCACHE
2048         p->sequential_io        = 0;
2049         p->sequential_io_avg    = 0;
2050 #endif
2051
2052         /* Perform scheduler related setup. Assign this task to a CPU. */
2053         retval = sched_fork(clone_flags, p);
2054         if (retval)
2055                 goto bad_fork_cleanup_policy;
2056
2057         retval = perf_event_init_task(p);
2058         if (retval)
2059                 goto bad_fork_cleanup_policy;
2060         retval = audit_alloc(p);
2061         if (retval)
2062                 goto bad_fork_cleanup_perf;
2063         /* copy all the process information */
2064         shm_init_task(p);
2065         retval = security_task_alloc(p, clone_flags);
2066         if (retval)
2067                 goto bad_fork_cleanup_audit;
2068         retval = copy_semundo(clone_flags, p);
2069         if (retval)
2070                 goto bad_fork_cleanup_security;
2071         retval = copy_files(clone_flags, p);
2072         if (retval)
2073                 goto bad_fork_cleanup_semundo;
2074         retval = copy_fs(clone_flags, p);
2075         if (retval)
2076                 goto bad_fork_cleanup_files;
2077         retval = copy_sighand(clone_flags, p);
2078         if (retval)
2079                 goto bad_fork_cleanup_fs;
2080         retval = copy_signal(clone_flags, p);
2081         if (retval)
2082                 goto bad_fork_cleanup_sighand;
2083         retval = copy_mm(clone_flags, p);
2084         if (retval)
2085                 goto bad_fork_cleanup_signal;
2086         retval = copy_namespaces(clone_flags, p);
2087         if (retval)
2088                 goto bad_fork_cleanup_mm;
2089         retval = copy_io(clone_flags, p);
2090         if (retval)
2091                 goto bad_fork_cleanup_namespaces;
2092         retval = copy_thread_tls(clone_flags, args->stack, args->stack_size, p,
2093                                  args->tls);
2094         if (retval)
2095                 goto bad_fork_cleanup_io;
2096
2097         stackleak_task_init(p);
2098
2099         if (pid != &init_struct_pid) {
2100                 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2101                                 args->set_tid_size);
2102                 if (IS_ERR(pid)) {
2103                         retval = PTR_ERR(pid);
2104                         goto bad_fork_cleanup_thread;
2105                 }
2106         }
2107
2108         /*
2109          * This has to happen after we've potentially unshared the file
2110          * descriptor table (so that the pidfd doesn't leak into the child
2111          * if the fd table isn't shared).
2112          */
2113         if (clone_flags & CLONE_PIDFD) {
2114                 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2115                 if (retval < 0)
2116                         goto bad_fork_free_pid;
2117
2118                 pidfd = retval;
2119
2120                 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2121                                               O_RDWR | O_CLOEXEC);
2122                 if (IS_ERR(pidfile)) {
2123                         put_unused_fd(pidfd);
2124                         retval = PTR_ERR(pidfile);
2125                         goto bad_fork_free_pid;
2126                 }
2127                 get_pid(pid);   /* held by pidfile now */
2128
2129                 retval = put_user(pidfd, args->pidfd);
2130                 if (retval)
2131                         goto bad_fork_put_pidfd;
2132         }
2133
2134 #ifdef CONFIG_BLOCK
2135         p->plug = NULL;
2136 #endif
2137         futex_init_task(p);
2138
2139         /*
2140          * sigaltstack should be cleared when sharing the same VM
2141          */
2142         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2143                 sas_ss_reset(p);
2144
2145         /*
2146          * Syscall tracing and stepping should be turned off in the
2147          * child regardless of CLONE_PTRACE.
2148          */
2149         user_disable_single_step(p);
2150         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
2151 #ifdef TIF_SYSCALL_EMU
2152         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
2153 #endif
2154         clear_tsk_latency_tracing(p);
2155
2156         /* ok, now we should be set up.. */
2157         p->pid = pid_nr(pid);
2158         if (clone_flags & CLONE_THREAD) {
2159                 p->exit_signal = -1;
2160                 p->group_leader = current->group_leader;
2161                 p->tgid = current->tgid;
2162         } else {
2163                 if (clone_flags & CLONE_PARENT)
2164                         p->exit_signal = current->group_leader->exit_signal;
2165                 else
2166                         p->exit_signal = args->exit_signal;
2167                 p->group_leader = p;
2168                 p->tgid = p->pid;
2169         }
2170
2171         p->nr_dirtied = 0;
2172         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2173         p->dirty_paused_when = 0;
2174
2175         p->pdeath_signal = 0;
2176         INIT_LIST_HEAD(&p->thread_group);
2177         p->task_works = NULL;
2178
2179         /*
2180          * Ensure that the cgroup subsystem policies allow the new process to be
2181          * forked. It should be noted the the new process's css_set can be changed
2182          * between here and cgroup_post_fork() if an organisation operation is in
2183          * progress.
2184          */
2185         retval = cgroup_can_fork(p, args);
2186         if (retval)
2187                 goto bad_fork_put_pidfd;
2188
2189         /*
2190          * From this point on we must avoid any synchronous user-space
2191          * communication until we take the tasklist-lock. In particular, we do
2192          * not want user-space to be able to predict the process start-time by
2193          * stalling fork(2) after we recorded the start_time but before it is
2194          * visible to the system.
2195          */
2196
2197         p->start_time = ktime_get_ns();
2198         p->start_boottime = ktime_get_boottime_ns();
2199
2200         /*
2201          * Make it visible to the rest of the system, but dont wake it up yet.
2202          * Need tasklist lock for parent etc handling!
2203          */
2204         write_lock_irq(&tasklist_lock);
2205
2206         /* CLONE_PARENT re-uses the old parent */
2207         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2208                 p->real_parent = current->real_parent;
2209                 p->parent_exec_id = current->parent_exec_id;
2210         } else {
2211                 p->real_parent = current;
2212                 p->parent_exec_id = current->self_exec_id;
2213         }
2214
2215         klp_copy_process(p);
2216
2217         spin_lock(&current->sighand->siglock);
2218
2219         /*
2220          * Copy seccomp details explicitly here, in case they were changed
2221          * before holding sighand lock.
2222          */
2223         copy_seccomp(p);
2224
2225         rseq_fork(p, clone_flags);
2226
2227         /* Don't start children in a dying pid namespace */
2228         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2229                 retval = -ENOMEM;
2230                 goto bad_fork_cancel_cgroup;
2231         }
2232
2233         /* Let kill terminate clone/fork in the middle */
2234         if (fatal_signal_pending(current)) {
2235                 retval = -EINTR;
2236                 goto bad_fork_cancel_cgroup;
2237         }
2238
2239         /* past the last point of failure */
2240         if (pidfile)
2241                 fd_install(pidfd, pidfile);
2242
2243         init_task_pid_links(p);
2244         if (likely(p->pid)) {
2245                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2246
2247                 init_task_pid(p, PIDTYPE_PID, pid);
2248                 if (thread_group_leader(p)) {
2249                         init_task_pid(p, PIDTYPE_TGID, pid);
2250                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2251                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2252
2253                         if (is_child_reaper(pid)) {
2254                                 ns_of_pid(pid)->child_reaper = p;
2255                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2256                         }
2257                         p->signal->shared_pending.signal = delayed.signal;
2258                         p->signal->tty = tty_kref_get(current->signal->tty);
2259                         /*
2260                          * Inherit has_child_subreaper flag under the same
2261                          * tasklist_lock with adding child to the process tree
2262                          * for propagate_has_child_subreaper optimization.
2263                          */
2264                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2265                                                          p->real_parent->signal->is_child_subreaper;
2266                         list_add_tail(&p->sibling, &p->real_parent->children);
2267                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2268                         attach_pid(p, PIDTYPE_TGID);
2269                         attach_pid(p, PIDTYPE_PGID);
2270                         attach_pid(p, PIDTYPE_SID);
2271                         __this_cpu_inc(process_counts);
2272                 } else {
2273                         current->signal->nr_threads++;
2274                         atomic_inc(&current->signal->live);
2275                         refcount_inc(&current->signal->sigcnt);
2276                         task_join_group_stop(p);
2277                         list_add_tail_rcu(&p->thread_group,
2278                                           &p->group_leader->thread_group);
2279                         list_add_tail_rcu(&p->thread_node,
2280                                           &p->signal->thread_head);
2281                 }
2282                 attach_pid(p, PIDTYPE_PID);
2283                 nr_threads++;
2284         }
2285         total_forks++;
2286         hlist_del_init(&delayed.node);
2287         spin_unlock(&current->sighand->siglock);
2288         syscall_tracepoint_update(p);
2289         write_unlock_irq(&tasklist_lock);
2290
2291         proc_fork_connector(p);
2292         cgroup_post_fork(p, args);
2293         perf_event_fork(p);
2294
2295         trace_task_newtask(p, clone_flags);
2296         uprobe_copy_process(p, clone_flags);
2297
2298         return p;
2299
2300 bad_fork_cancel_cgroup:
2301         spin_unlock(&current->sighand->siglock);
2302         write_unlock_irq(&tasklist_lock);
2303         cgroup_cancel_fork(p, args);
2304 bad_fork_put_pidfd:
2305         if (clone_flags & CLONE_PIDFD) {
2306                 fput(pidfile);
2307                 put_unused_fd(pidfd);
2308         }
2309 bad_fork_free_pid:
2310         if (pid != &init_struct_pid)
2311                 free_pid(pid);
2312 bad_fork_cleanup_thread:
2313         exit_thread(p);
2314 bad_fork_cleanup_io:
2315         if (p->io_context)
2316                 exit_io_context(p);
2317 bad_fork_cleanup_namespaces:
2318         exit_task_namespaces(p);
2319 bad_fork_cleanup_mm:
2320         if (p->mm) {
2321                 mm_clear_owner(p->mm, p);
2322                 mmput(p->mm);
2323         }
2324 bad_fork_cleanup_signal:
2325         if (!(clone_flags & CLONE_THREAD))
2326                 free_signal_struct(p->signal);
2327 bad_fork_cleanup_sighand:
2328         __cleanup_sighand(p->sighand);
2329 bad_fork_cleanup_fs:
2330         exit_fs(p); /* blocking */
2331 bad_fork_cleanup_files:
2332         exit_files(p); /* blocking */
2333 bad_fork_cleanup_semundo:
2334         exit_sem(p);
2335 bad_fork_cleanup_security:
2336         security_task_free(p);
2337 bad_fork_cleanup_audit:
2338         audit_free(p);
2339 bad_fork_cleanup_perf:
2340         perf_event_free_task(p);
2341 bad_fork_cleanup_policy:
2342         lockdep_free_task(p);
2343 #ifdef CONFIG_NUMA
2344         mpol_put(p->mempolicy);
2345 bad_fork_cleanup_threadgroup_lock:
2346 #endif
2347         delayacct_tsk_free(p);
2348 bad_fork_cleanup_count:
2349         atomic_dec(&p->cred->user->processes);
2350         exit_creds(p);
2351 bad_fork_free:
2352         p->state = TASK_DEAD;
2353         put_task_stack(p);
2354         delayed_free_task(p);
2355 fork_out:
2356         spin_lock_irq(&current->sighand->siglock);
2357         hlist_del_init(&delayed.node);
2358         spin_unlock_irq(&current->sighand->siglock);
2359         return ERR_PTR(retval);
2360 }
2361
2362 static inline void init_idle_pids(struct task_struct *idle)
2363 {
2364         enum pid_type type;
2365
2366         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2367                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2368                 init_task_pid(idle, type, &init_struct_pid);
2369         }
2370 }
2371
2372 struct task_struct *fork_idle(int cpu)
2373 {
2374         struct task_struct *task;
2375         struct kernel_clone_args args = {
2376                 .flags = CLONE_VM,
2377         };
2378
2379         task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2380         if (!IS_ERR(task)) {
2381                 init_idle_pids(task);
2382                 init_idle(task, cpu);
2383         }
2384
2385         return task;
2386 }
2387
2388 struct mm_struct *copy_init_mm(void)
2389 {
2390         return dup_mm(NULL, &init_mm);
2391 }
2392
2393 /*
2394  *  Ok, this is the main fork-routine.
2395  *
2396  * It copies the process, and if successful kick-starts
2397  * it and waits for it to finish using the VM if required.
2398  *
2399  * args->exit_signal is expected to be checked for sanity by the caller.
2400  */
2401 long _do_fork(struct kernel_clone_args *args)
2402 {
2403         u64 clone_flags = args->flags;
2404         struct completion vfork;
2405         struct pid *pid;
2406         struct task_struct *p;
2407         int trace = 0;
2408         long nr;
2409
2410         /*
2411          * Determine whether and which event to report to ptracer.  When
2412          * called from kernel_thread or CLONE_UNTRACED is explicitly
2413          * requested, no event is reported; otherwise, report if the event
2414          * for the type of forking is enabled.
2415          */
2416         if (!(clone_flags & CLONE_UNTRACED)) {
2417                 if (clone_flags & CLONE_VFORK)
2418                         trace = PTRACE_EVENT_VFORK;
2419                 else if (args->exit_signal != SIGCHLD)
2420                         trace = PTRACE_EVENT_CLONE;
2421                 else
2422                         trace = PTRACE_EVENT_FORK;
2423
2424                 if (likely(!ptrace_event_enabled(current, trace)))
2425                         trace = 0;
2426         }
2427
2428         p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2429         add_latent_entropy();
2430
2431         if (IS_ERR(p))
2432                 return PTR_ERR(p);
2433
2434         /*
2435          * Do this prior waking up the new thread - the thread pointer
2436          * might get invalid after that point, if the thread exits quickly.
2437          */
2438         trace_sched_process_fork(current, p);
2439
2440         pid = get_task_pid(p, PIDTYPE_PID);
2441         nr = pid_vnr(pid);
2442
2443         if (clone_flags & CLONE_PARENT_SETTID)
2444                 put_user(nr, args->parent_tid);
2445
2446         if (clone_flags & CLONE_VFORK) {
2447                 p->vfork_done = &vfork;
2448                 init_completion(&vfork);
2449                 get_task_struct(p);
2450         }
2451
2452         wake_up_new_task(p);
2453
2454         /* forking complete and child started to run, tell ptracer */
2455         if (unlikely(trace))
2456                 ptrace_event_pid(trace, pid);
2457
2458         if (clone_flags & CLONE_VFORK) {
2459                 if (!wait_for_vfork_done(p, &vfork))
2460                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2461         }
2462
2463         put_pid(pid);
2464         return nr;
2465 }
2466
2467 bool legacy_clone_args_valid(const struct kernel_clone_args *kargs)
2468 {
2469         /* clone(CLONE_PIDFD) uses parent_tidptr to return a pidfd */
2470         if ((kargs->flags & CLONE_PIDFD) &&
2471             (kargs->flags & CLONE_PARENT_SETTID))
2472                 return false;
2473
2474         return true;
2475 }
2476
2477 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2478 /* For compatibility with architectures that call do_fork directly rather than
2479  * using the syscall entry points below. */
2480 long do_fork(unsigned long clone_flags,
2481               unsigned long stack_start,
2482               unsigned long stack_size,
2483               int __user *parent_tidptr,
2484               int __user *child_tidptr)
2485 {
2486         struct kernel_clone_args args = {
2487                 .flags          = (clone_flags & ~CSIGNAL),
2488                 .pidfd          = parent_tidptr,
2489                 .child_tid      = child_tidptr,
2490                 .parent_tid     = parent_tidptr,
2491                 .exit_signal    = (clone_flags & CSIGNAL),
2492                 .stack          = stack_start,
2493                 .stack_size     = stack_size,
2494         };
2495
2496         if (!legacy_clone_args_valid(&args))
2497                 return -EINVAL;
2498
2499         return _do_fork(&args);
2500 }
2501 #endif
2502
2503 /*
2504  * Create a kernel thread.
2505  */
2506 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2507 {
2508         struct kernel_clone_args args = {
2509                 .flags          = ((flags | CLONE_VM | CLONE_UNTRACED) & ~CSIGNAL),
2510                 .exit_signal    = (flags & CSIGNAL),
2511                 .stack          = (unsigned long)fn,
2512                 .stack_size     = (unsigned long)arg,
2513         };
2514
2515         return _do_fork(&args);
2516 }
2517
2518 #ifdef __ARCH_WANT_SYS_FORK
2519 SYSCALL_DEFINE0(fork)
2520 {
2521 #ifdef CONFIG_MMU
2522         struct kernel_clone_args args = {
2523                 .exit_signal = SIGCHLD,
2524         };
2525
2526         return _do_fork(&args);
2527 #else
2528         /* can not support in nommu mode */
2529         return -EINVAL;
2530 #endif
2531 }
2532 #endif
2533
2534 #ifdef __ARCH_WANT_SYS_VFORK
2535 SYSCALL_DEFINE0(vfork)
2536 {
2537         struct kernel_clone_args args = {
2538                 .flags          = CLONE_VFORK | CLONE_VM,
2539                 .exit_signal    = SIGCHLD,
2540         };
2541
2542         return _do_fork(&args);
2543 }
2544 #endif
2545
2546 #ifdef __ARCH_WANT_SYS_CLONE
2547 #ifdef CONFIG_CLONE_BACKWARDS
2548 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2549                  int __user *, parent_tidptr,
2550                  unsigned long, tls,
2551                  int __user *, child_tidptr)
2552 #elif defined(CONFIG_CLONE_BACKWARDS2)
2553 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2554                  int __user *, parent_tidptr,
2555                  int __user *, child_tidptr,
2556                  unsigned long, tls)
2557 #elif defined(CONFIG_CLONE_BACKWARDS3)
2558 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2559                 int, stack_size,
2560                 int __user *, parent_tidptr,
2561                 int __user *, child_tidptr,
2562                 unsigned long, tls)
2563 #else
2564 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2565                  int __user *, parent_tidptr,
2566                  int __user *, child_tidptr,
2567                  unsigned long, tls)
2568 #endif
2569 {
2570         struct kernel_clone_args args = {
2571                 .flags          = (clone_flags & ~CSIGNAL),
2572                 .pidfd          = parent_tidptr,
2573                 .child_tid      = child_tidptr,
2574                 .parent_tid     = parent_tidptr,
2575                 .exit_signal    = (clone_flags & CSIGNAL),
2576                 .stack          = newsp,
2577                 .tls            = tls,
2578         };
2579
2580         if (!legacy_clone_args_valid(&args))
2581                 return -EINVAL;
2582
2583         return _do_fork(&args);
2584 }
2585 #endif
2586
2587 #ifdef __ARCH_WANT_SYS_CLONE3
2588
2589 /*
2590  * copy_thread implementations handle CLONE_SETTLS by reading the TLS value from
2591  * the registers containing the syscall arguments for clone. This doesn't work
2592  * with clone3 since the TLS value is passed in clone_args instead.
2593  */
2594 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2595 #error clone3 requires copy_thread_tls support in arch
2596 #endif
2597
2598 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2599                                               struct clone_args __user *uargs,
2600                                               size_t usize)
2601 {
2602         int err;
2603         struct clone_args args;
2604         pid_t *kset_tid = kargs->set_tid;
2605
2606         if (unlikely(usize > PAGE_SIZE))
2607                 return -E2BIG;
2608         if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2609                 return -EINVAL;
2610
2611         err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2612         if (err)
2613                 return err;
2614
2615         if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2616                 return -EINVAL;
2617
2618         if (unlikely(!args.set_tid && args.set_tid_size > 0))
2619                 return -EINVAL;
2620
2621         if (unlikely(args.set_tid && args.set_tid_size == 0))
2622                 return -EINVAL;
2623
2624         /*
2625          * Verify that higher 32bits of exit_signal are unset and that
2626          * it is a valid signal
2627          */
2628         if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2629                      !valid_signal(args.exit_signal)))
2630                 return -EINVAL;
2631
2632         if ((args.flags & CLONE_INTO_CGROUP) && args.cgroup < 0)
2633                 return -EINVAL;
2634
2635         *kargs = (struct kernel_clone_args){
2636                 .flags          = args.flags,
2637                 .pidfd          = u64_to_user_ptr(args.pidfd),
2638                 .child_tid      = u64_to_user_ptr(args.child_tid),
2639                 .parent_tid     = u64_to_user_ptr(args.parent_tid),
2640                 .exit_signal    = args.exit_signal,
2641                 .stack          = args.stack,
2642                 .stack_size     = args.stack_size,
2643                 .tls            = args.tls,
2644                 .set_tid_size   = args.set_tid_size,
2645                 .cgroup         = args.cgroup,
2646         };
2647
2648         if (args.set_tid &&
2649                 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2650                         (kargs->set_tid_size * sizeof(pid_t))))
2651                 return -EFAULT;
2652
2653         kargs->set_tid = kset_tid;
2654
2655         return 0;
2656 }
2657
2658 /**
2659  * clone3_stack_valid - check and prepare stack
2660  * @kargs: kernel clone args
2661  *
2662  * Verify that the stack arguments userspace gave us are sane.
2663  * In addition, set the stack direction for userspace since it's easy for us to
2664  * determine.
2665  */
2666 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2667 {
2668         if (kargs->stack == 0) {
2669                 if (kargs->stack_size > 0)
2670                         return false;
2671         } else {
2672                 if (kargs->stack_size == 0)
2673                         return false;
2674
2675                 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2676                         return false;
2677
2678 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2679                 kargs->stack += kargs->stack_size;
2680 #endif
2681         }
2682
2683         return true;
2684 }
2685
2686 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2687 {
2688         /* Verify that no unknown flags are passed along. */
2689         if (kargs->flags &
2690             ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2691                 return false;
2692
2693         /*
2694          * - make the CLONE_DETACHED bit reuseable for clone3
2695          * - make the CSIGNAL bits reuseable for clone3
2696          */
2697         if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2698                 return false;
2699
2700         if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2701             (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2702                 return false;
2703
2704         if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2705             kargs->exit_signal)
2706                 return false;
2707
2708         if (!clone3_stack_valid(kargs))
2709                 return false;
2710
2711         return true;
2712 }
2713
2714 /**
2715  * clone3 - create a new process with specific properties
2716  * @uargs: argument structure
2717  * @size:  size of @uargs
2718  *
2719  * clone3() is the extensible successor to clone()/clone2().
2720  * It takes a struct as argument that is versioned by its size.
2721  *
2722  * Return: On success, a positive PID for the child process.
2723  *         On error, a negative errno number.
2724  */
2725 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2726 {
2727         int err;
2728
2729         struct kernel_clone_args kargs;
2730         pid_t set_tid[MAX_PID_NS_LEVEL];
2731
2732         kargs.set_tid = set_tid;
2733
2734         err = copy_clone_args_from_user(&kargs, uargs, size);
2735         if (err)
2736                 return err;
2737
2738         if (!clone3_args_valid(&kargs))
2739                 return -EINVAL;
2740
2741         return _do_fork(&kargs);
2742 }
2743 #endif
2744
2745 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2746 {
2747         struct task_struct *leader, *parent, *child;
2748         int res;
2749
2750         read_lock(&tasklist_lock);
2751         leader = top = top->group_leader;
2752 down:
2753         for_each_thread(leader, parent) {
2754                 list_for_each_entry(child, &parent->children, sibling) {
2755                         res = visitor(child, data);
2756                         if (res) {
2757                                 if (res < 0)
2758                                         goto out;
2759                                 leader = child;
2760                                 goto down;
2761                         }
2762 up:
2763                         ;
2764                 }
2765         }
2766
2767         if (leader != top) {
2768                 child = leader;
2769                 parent = child->real_parent;
2770                 leader = parent->group_leader;
2771                 goto up;
2772         }
2773 out:
2774         read_unlock(&tasklist_lock);
2775 }
2776
2777 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2778 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2779 #endif
2780
2781 static void sighand_ctor(void *data)
2782 {
2783         struct sighand_struct *sighand = data;
2784
2785         spin_lock_init(&sighand->siglock);
2786         init_waitqueue_head(&sighand->signalfd_wqh);
2787 }
2788
2789 void __init proc_caches_init(void)
2790 {
2791         unsigned int mm_size;
2792
2793         sighand_cachep = kmem_cache_create("sighand_cache",
2794                         sizeof(struct sighand_struct), 0,
2795                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2796                         SLAB_ACCOUNT, sighand_ctor);
2797         signal_cachep = kmem_cache_create("signal_cache",
2798                         sizeof(struct signal_struct), 0,
2799                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2800                         NULL);
2801         files_cachep = kmem_cache_create("files_cache",
2802                         sizeof(struct files_struct), 0,
2803                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2804                         NULL);
2805         fs_cachep = kmem_cache_create("fs_cache",
2806                         sizeof(struct fs_struct), 0,
2807                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2808                         NULL);
2809
2810         /*
2811          * The mm_cpumask is located at the end of mm_struct, and is
2812          * dynamically sized based on the maximum CPU number this system
2813          * can have, taking hotplug into account (nr_cpu_ids).
2814          */
2815         mm_size = sizeof(struct mm_struct) + cpumask_size();
2816
2817         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2818                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2819                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2820                         offsetof(struct mm_struct, saved_auxv),
2821                         sizeof_field(struct mm_struct, saved_auxv),
2822                         NULL);
2823         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2824         mmap_init();
2825         nsproxy_cache_init();
2826 }
2827
2828 /*
2829  * Check constraints on flags passed to the unshare system call.
2830  */
2831 static int check_unshare_flags(unsigned long unshare_flags)
2832 {
2833         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2834                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2835                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2836                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
2837                                 CLONE_NEWTIME))
2838                 return -EINVAL;
2839         /*
2840          * Not implemented, but pretend it works if there is nothing
2841          * to unshare.  Note that unsharing the address space or the
2842          * signal handlers also need to unshare the signal queues (aka
2843          * CLONE_THREAD).
2844          */
2845         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2846                 if (!thread_group_empty(current))
2847                         return -EINVAL;
2848         }
2849         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2850                 if (refcount_read(&current->sighand->count) > 1)
2851                         return -EINVAL;
2852         }
2853         if (unshare_flags & CLONE_VM) {
2854                 if (!current_is_single_threaded())
2855                         return -EINVAL;
2856         }
2857
2858         return 0;
2859 }
2860
2861 /*
2862  * Unshare the filesystem structure if it is being shared
2863  */
2864 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2865 {
2866         struct fs_struct *fs = current->fs;
2867
2868         if (!(unshare_flags & CLONE_FS) || !fs)
2869                 return 0;
2870
2871         /* don't need lock here; in the worst case we'll do useless copy */
2872         if (fs->users == 1)
2873                 return 0;
2874
2875         *new_fsp = copy_fs_struct(fs);
2876         if (!*new_fsp)
2877                 return -ENOMEM;
2878
2879         return 0;
2880 }
2881
2882 /*
2883  * Unshare file descriptor table if it is being shared
2884  */
2885 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2886 {
2887         struct files_struct *fd = current->files;
2888         int error = 0;
2889
2890         if ((unshare_flags & CLONE_FILES) &&
2891             (fd && atomic_read(&fd->count) > 1)) {
2892                 *new_fdp = dup_fd(fd, &error);
2893                 if (!*new_fdp)
2894                         return error;
2895         }
2896
2897         return 0;
2898 }
2899
2900 /*
2901  * unshare allows a process to 'unshare' part of the process
2902  * context which was originally shared using clone.  copy_*
2903  * functions used by do_fork() cannot be used here directly
2904  * because they modify an inactive task_struct that is being
2905  * constructed. Here we are modifying the current, active,
2906  * task_struct.
2907  */
2908 int ksys_unshare(unsigned long unshare_flags)
2909 {
2910         struct fs_struct *fs, *new_fs = NULL;
2911         struct files_struct *fd, *new_fd = NULL;
2912         struct cred *new_cred = NULL;
2913         struct nsproxy *new_nsproxy = NULL;
2914         int do_sysvsem = 0;
2915         int err;
2916
2917         /*
2918          * If unsharing a user namespace must also unshare the thread group
2919          * and unshare the filesystem root and working directories.
2920          */
2921         if (unshare_flags & CLONE_NEWUSER)
2922                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2923         /*
2924          * If unsharing vm, must also unshare signal handlers.
2925          */
2926         if (unshare_flags & CLONE_VM)
2927                 unshare_flags |= CLONE_SIGHAND;
2928         /*
2929          * If unsharing a signal handlers, must also unshare the signal queues.
2930          */
2931         if (unshare_flags & CLONE_SIGHAND)
2932                 unshare_flags |= CLONE_THREAD;
2933         /*
2934          * If unsharing namespace, must also unshare filesystem information.
2935          */
2936         if (unshare_flags & CLONE_NEWNS)
2937                 unshare_flags |= CLONE_FS;
2938
2939         err = check_unshare_flags(unshare_flags);
2940         if (err)
2941                 goto bad_unshare_out;
2942         /*
2943          * CLONE_NEWIPC must also detach from the undolist: after switching
2944          * to a new ipc namespace, the semaphore arrays from the old
2945          * namespace are unreachable.
2946          */
2947         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2948                 do_sysvsem = 1;
2949         err = unshare_fs(unshare_flags, &new_fs);
2950         if (err)
2951                 goto bad_unshare_out;
2952         err = unshare_fd(unshare_flags, &new_fd);
2953         if (err)
2954                 goto bad_unshare_cleanup_fs;
2955         err = unshare_userns(unshare_flags, &new_cred);
2956         if (err)
2957                 goto bad_unshare_cleanup_fd;
2958         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2959                                          new_cred, new_fs);
2960         if (err)
2961                 goto bad_unshare_cleanup_cred;
2962
2963         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2964                 if (do_sysvsem) {
2965                         /*
2966                          * CLONE_SYSVSEM is equivalent to sys_exit().
2967                          */
2968                         exit_sem(current);
2969                 }
2970                 if (unshare_flags & CLONE_NEWIPC) {
2971                         /* Orphan segments in old ns (see sem above). */
2972                         exit_shm(current);
2973                         shm_init_task(current);
2974                 }
2975
2976                 if (new_nsproxy)
2977                         switch_task_namespaces(current, new_nsproxy);
2978
2979                 task_lock(current);
2980
2981                 if (new_fs) {
2982                         fs = current->fs;
2983                         spin_lock(&fs->lock);
2984                         current->fs = new_fs;
2985                         if (--fs->users)
2986                                 new_fs = NULL;
2987                         else
2988                                 new_fs = fs;
2989                         spin_unlock(&fs->lock);
2990                 }
2991
2992                 if (new_fd) {
2993                         fd = current->files;
2994                         current->files = new_fd;
2995                         new_fd = fd;
2996                 }
2997
2998                 task_unlock(current);
2999
3000                 if (new_cred) {
3001                         /* Install the new user namespace */
3002                         commit_creds(new_cred);
3003                         new_cred = NULL;
3004                 }
3005         }
3006
3007         perf_event_namespaces(current);
3008
3009 bad_unshare_cleanup_cred:
3010         if (new_cred)
3011                 put_cred(new_cred);
3012 bad_unshare_cleanup_fd:
3013         if (new_fd)
3014                 put_files_struct(new_fd);
3015
3016 bad_unshare_cleanup_fs:
3017         if (new_fs)
3018                 free_fs_struct(new_fs);
3019
3020 bad_unshare_out:
3021         return err;
3022 }
3023
3024 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3025 {
3026         return ksys_unshare(unshare_flags);
3027 }
3028
3029 /*
3030  *      Helper to unshare the files of the current task.
3031  *      We don't want to expose copy_files internals to
3032  *      the exec layer of the kernel.
3033  */
3034
3035 int unshare_files(struct files_struct **displaced)
3036 {
3037         struct task_struct *task = current;
3038         struct files_struct *copy = NULL;
3039         int error;
3040
3041         error = unshare_fd(CLONE_FILES, &copy);
3042         if (error || !copy) {
3043                 *displaced = NULL;
3044                 return error;
3045         }
3046         *displaced = task->files;
3047         task_lock(task);
3048         task->files = copy;
3049         task_unlock(task);
3050         return 0;
3051 }
3052
3053 int sysctl_max_threads(struct ctl_table *table, int write,
3054                        void __user *buffer, size_t *lenp, loff_t *ppos)
3055 {
3056         struct ctl_table t;
3057         int ret;
3058         int threads = max_threads;
3059         int min = 1;
3060         int max = MAX_THREADS;
3061
3062         t = *table;
3063         t.data = &threads;
3064         t.extra1 = &min;
3065         t.extra2 = &max;
3066
3067         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3068         if (ret || !write)
3069                 return ret;
3070
3071         max_threads = threads;
3072
3073         return 0;
3074 }