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/kmsan.h>
  41 #include <linux/binfmts.h>
  42 #include <linux/mman.h>
  43 #include <linux/mmu_notifier.h>
  44 #include <linux/fs.h>
  45 #include <linux/mm.h>
  46 #include <linux/mm_inline.h>
  47 #include <linux/nsproxy.h>
  48 #include <linux/capability.h>
  49 #include <linux/cpu.h>
  50 #include <linux/cgroup.h>
  51 #include <linux/security.h>
  52 #include <linux/hugetlb.h>
  53 #include <linux/seccomp.h>
  54 #include <linux/swap.h>
  55 #include <linux/syscalls.h>
  56 #include <linux/jiffies.h>
  57 #include <linux/futex.h>
  58 #include <linux/compat.h>
  59 #include <linux/kthread.h>
  60 #include <linux/task_io_accounting_ops.h>
  61 #include <linux/rcupdate.h>
  62 #include <linux/ptrace.h>
  63 #include <linux/mount.h>
  64 #include <linux/audit.h>
  65 #include <linux/memcontrol.h>
  66 #include <linux/ftrace.h>
  67 #include <linux/proc_fs.h>
  68 #include <linux/profile.h>
  69 #include <linux/rmap.h>
  70 #include <linux/ksm.h>
  71 #include <linux/acct.h>
  72 #include <linux/userfaultfd_k.h>
  73 #include <linux/tsacct_kern.h>
  74 #include <linux/cn_proc.h>
  75 #include <linux/freezer.h>
  76 #include <linux/delayacct.h>
  77 #include <linux/taskstats_kern.h>
  78 #include <linux/tty.h>
  79 #include <linux/fs_struct.h>
  80 #include <linux/magic.h>
  81 #include <linux/perf_event.h>
  82 #include <linux/posix-timers.h>
  83 #include <linux/user-return-notifier.h>
  84 #include <linux/oom.h>
  85 #include <linux/khugepaged.h>
  86 #include <linux/signalfd.h>
  87 #include <linux/uprobes.h>
  88 #include <linux/aio.h>
  89 #include <linux/compiler.h>
  90 #include <linux/sysctl.h>
  91 #include <linux/kcov.h>
  92 #include <linux/livepatch.h>
  93 #include <linux/thread_info.h>
  94 #include <linux/stackleak.h>
  95 #include <linux/kasan.h>
  96 #include <linux/scs.h>
  97 #include <linux/io_uring.h>
  98 #include <linux/bpf.h>
  99 #include <linux/stackprotector.h>
 100
 101 #include <asm/pgalloc.h>
 102 #include <linux/uaccess.h>
 103 #include <asm/mmu_context.h>
 104 #include <asm/cacheflush.h>
 105 #include <asm/tlbflush.h>
 106
 107 #include <trace/events/sched.h>
 108
 109 #define CREATE_TRACE_POINTS
 110 #include <trace/events/task.h>
 111
 112 /*
 113  * Minimum number of threads to boot the kernel
 114  */
 115 #define MIN_THREADS 20
 116
 117 /*
 118  * Maximum number of threads
 119  */
 120 #define MAX_THREADS FUTEX_TID_MASK
 121
 122 /*
 123  * Protected counters by write_lock_irq(&tasklist_lock)
 124  */
 125 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 126 int nr_threads;                 /* The idle threads do not count.. */
 127
 128 static int max_threads;         /* tunable limit on nr_threads */
 129
 130 #define NAMED_ARRAY_INDEX(x)    [x] = __stringify(x)
 131
 132 static const char * const resident_page_types[] = {
 133         NAMED_ARRAY_INDEX(MM_FILEPAGES),
 134         NAMED_ARRAY_INDEX(MM_ANONPAGES),
 135         NAMED_ARRAY_INDEX(MM_SWAPENTS),
 136         NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
 137 };
 138
 139 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 140
 141 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 142
 143 #ifdef CONFIG_PROVE_RCU
 144 int lockdep_tasklist_lock_is_held(void)
 145 {
 146         return lockdep_is_held(&tasklist_lock);
 147 }
 148 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 149 #endif /* #ifdef CONFIG_PROVE_RCU */
 150
 151 int nr_processes(void)
 152 {
 153         int cpu;
 154         int total = 0;
 155
 156         for_each_possible_cpu(cpu)
 157                 total += per_cpu(process_counts, cpu);
 158
 159         return total;
 160 }
 161
 162 void __weak arch_release_task_struct(struct task_struct *tsk)
 163 {
 164 }
 165
 166 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 167 static struct kmem_cache *task_struct_cachep;
 168
 169 static inline struct task_struct *alloc_task_struct_node(int node)
 170 {
 171         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 172 }
 173
 174 static inline void free_task_struct(struct task_struct *tsk)
 175 {
 176         kmem_cache_free(task_struct_cachep, tsk);
 177 }
 178 #endif
 179
 180 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 181
 182 /*
 183  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 184  * kmemcache based allocator.
 185  */
 186 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 187
 188 #  ifdef CONFIG_VMAP_STACK
 189 /*
 190  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 191  * flush.  Try to minimize the number of calls by caching stacks.
 192  */
 193 #define NR_CACHED_STACKS 2
 194 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 195
 196 struct vm_stack {
 197         struct rcu_head rcu;
 198         struct vm_struct *stack_vm_area;
 199 };
 200
 201 static bool try_release_thread_stack_to_cache(struct vm_struct *vm)
 202 {
 203         unsigned int i;
 204
 205         for (i = 0; i < NR_CACHED_STACKS; i++) {
 206                 if (this_cpu_cmpxchg(cached_stacks[i], NULL, vm) != NULL)
 207                         continue;
 208                 return true;
 209         }
 210         return false;
 211 }
 212
 213 static void thread_stack_free_rcu(struct rcu_head *rh)
 214 {
 215         struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
 216
 217         if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
 218                 return;
 219
 220         vfree(vm_stack);
 221 }
 222
 223 static void thread_stack_delayed_free(struct task_struct *tsk)
 224 {
 225         struct vm_stack *vm_stack = tsk->stack;
 226
 227         vm_stack->stack_vm_area = tsk->stack_vm_area;
 228         call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
 229 }
 230
 231 static int free_vm_stack_cache(unsigned int cpu)
 232 {
 233         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 234         int i;
 235
 236         for (i = 0; i < NR_CACHED_STACKS; i++) {
 237                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 238
 239                 if (!vm_stack)
 240                         continue;
 241
 242                 vfree(vm_stack->addr);
 243                 cached_vm_stacks[i] = NULL;
 244         }
 245
 246         return 0;
 247 }
 248
 249 static int memcg_charge_kernel_stack(struct vm_struct *vm)
 250 {
 251         int i;
 252         int ret;
 253
 254         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 255         BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 256
 257         for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 258                 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0);
 259                 if (ret)
 260                         goto err;
 261         }
 262         return 0;
 263 err:
 264         /*
 265          * If memcg_kmem_charge_page() fails, page's memory cgroup pointer is
 266          * NULL, and memcg_kmem_uncharge_page() in free_thread_stack() will
 267          * ignore this page.
 268          */
 269         for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 270                 memcg_kmem_uncharge_page(vm->pages[i], 0);
 271         return ret;
 272 }
 273
 274 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 275 {
 276         struct vm_struct *vm;
 277         void *stack;
 278         int i;
 279
 280         for (i = 0; i < NR_CACHED_STACKS; i++) {
 281                 struct vm_struct *s;
 282
 283                 s = this_cpu_xchg(cached_stacks[i], NULL);
 284
 285                 if (!s)
 286                         continue;
 287
 288                 /* Reset stack metadata. */
 289                 kasan_unpoison_range(s->addr, THREAD_SIZE);
 290
 291                 stack = kasan_reset_tag(s->addr);
 292
 293                 /* Clear stale pointers from reused stack. */
 294                 memset(stack, 0, THREAD_SIZE);
 295
 296                 if (memcg_charge_kernel_stack(s)) {
 297                         vfree(s->addr);
 298                         return -ENOMEM;
 299                 }
 300
 301                 tsk->stack_vm_area = s;
 302                 tsk->stack = stack;
 303                 return 0;
 304         }
 305
 306         /*
 307          * Allocated stacks are cached and later reused by new threads,
 308          * so memcg accounting is performed manually on assigning/releasing
 309          * stacks to tasks. Drop __GFP_ACCOUNT.
 310          */
 311         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 312                                      VMALLOC_START, VMALLOC_END,
 313                                      THREADINFO_GFP & ~__GFP_ACCOUNT,
 314                                      PAGE_KERNEL,
 315                                      0, node, __builtin_return_address(0));
 316         if (!stack)
 317                 return -ENOMEM;
 318
 319         vm = find_vm_area(stack);
 320         if (memcg_charge_kernel_stack(vm)) {
 321                 vfree(stack);
 322                 return -ENOMEM;
 323         }
 324         /*
 325          * We can't call find_vm_area() in interrupt context, and
 326          * free_thread_stack() can be called in interrupt context,
 327          * so cache the vm_struct.
 328          */
 329         tsk->stack_vm_area = vm;
 330         stack = kasan_reset_tag(stack);
 331         tsk->stack = stack;
 332         return 0;
 333 }
 334
 335 static void free_thread_stack(struct task_struct *tsk)
 336 {
 337         if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
 338                 thread_stack_delayed_free(tsk);
 339
 340         tsk->stack = NULL;
 341         tsk->stack_vm_area = NULL;
 342 }
 343
 344 #  else /* !CONFIG_VMAP_STACK */
 345
 346 static void thread_stack_free_rcu(struct rcu_head *rh)
 347 {
 348         __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
 349 }
 350
 351 static void thread_stack_delayed_free(struct task_struct *tsk)
 352 {
 353         struct rcu_head *rh = tsk->stack;
 354
 355         call_rcu(rh, thread_stack_free_rcu);
 356 }
 357
 358 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 359 {
 360         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 361                                              THREAD_SIZE_ORDER);
 362
 363         if (likely(page)) {
 364                 tsk->stack = kasan_reset_tag(page_address(page));
 365                 return 0;
 366         }
 367         return -ENOMEM;
 368 }
 369
 370 static void free_thread_stack(struct task_struct *tsk)
 371 {
 372         thread_stack_delayed_free(tsk);
 373         tsk->stack = NULL;
 374 }
 375
 376 #  endif /* CONFIG_VMAP_STACK */
 377 # else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
 378
 379 static struct kmem_cache *thread_stack_cache;
 380
 381 static void thread_stack_free_rcu(struct rcu_head *rh)
 382 {
 383         kmem_cache_free(thread_stack_cache, rh);
 384 }
 385
 386 static void thread_stack_delayed_free(struct task_struct *tsk)
 387 {
 388         struct rcu_head *rh = tsk->stack;
 389
 390         call_rcu(rh, thread_stack_free_rcu);
 391 }
 392
 393 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 394 {
 395         unsigned long *stack;
 396         stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 397         stack = kasan_reset_tag(stack);
 398         tsk->stack = stack;
 399         return stack ? 0 : -ENOMEM;
 400 }
 401
 402 static void free_thread_stack(struct task_struct *tsk)
 403 {
 404         thread_stack_delayed_free(tsk);
 405         tsk->stack = NULL;
 406 }
 407
 408 void thread_stack_cache_init(void)
 409 {
 410         thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 411                                         THREAD_SIZE, THREAD_SIZE, 0, 0,
 412                                         THREAD_SIZE, NULL);
 413         BUG_ON(thread_stack_cache == NULL);
 414 }
 415
 416 # endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
 417 #else /* CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
 418
 419 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 420 {
 421         unsigned long *stack;
 422
 423         stack = arch_alloc_thread_stack_node(tsk, node);
 424         tsk->stack = stack;
 425         return stack ? 0 : -ENOMEM;
 426 }
 427
 428 static void free_thread_stack(struct task_struct *tsk)
 429 {
 430         arch_free_thread_stack(tsk);
 431         tsk->stack = NULL;
 432 }
 433
 434 #endif /* !CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
 435
 436 /* SLAB cache for signal_struct structures (tsk->signal) */
 437 static struct kmem_cache *signal_cachep;
 438
 439 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 440 struct kmem_cache *sighand_cachep;
 441
 442 /* SLAB cache for files_struct structures (tsk->files) */
 443 struct kmem_cache *files_cachep;
 444
 445 /* SLAB cache for fs_struct structures (tsk->fs) */
 446 struct kmem_cache *fs_cachep;
 447
 448 /* SLAB cache for vm_area_struct structures */
 449 static struct kmem_cache *vm_area_cachep;
 450
 451 /* SLAB cache for mm_struct structures (tsk->mm) */
 452 static struct kmem_cache *mm_cachep;
 453
 454 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 455 {
 456         struct vm_area_struct *vma;
 457
 458         vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 459         if (vma)
 460                 vma_init(vma, mm);
 461         return vma;
 462 }
 463
 464 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 465 {
 466         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 467
 468         if (new) {
 469                 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
 470                 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
 471                 /*
 472                  * orig->shared.rb may be modified concurrently, but the clone
 473                  * will be reinitialized.
 474                  */
 475                 data_race(memcpy(new, orig, sizeof(*new)));
 476                 INIT_LIST_HEAD(&new->anon_vma_chain);
 477                 dup_anon_vma_name(orig, new);
 478         }
 479         return new;
 480 }
 481
 482 static void __vm_area_free(struct vm_area_struct *vma)
 483 {
 484         free_anon_vma_name(vma);
 485         kmem_cache_free(vm_area_cachep, vma);
 486 }
 487
 488 #ifdef CONFIG_PER_VMA_LOCK
 489 static void vm_area_free_rcu_cb(struct rcu_head *head)
 490 {
 491         struct vm_area_struct *vma = container_of(head, struct vm_area_struct,
 492                                                   vm_rcu);
 493         __vm_area_free(vma);
 494 }
 495 #endif
 496
 497 void vm_area_free(struct vm_area_struct *vma)
 498 {
 499 #ifdef CONFIG_PER_VMA_LOCK
 500         call_rcu(&vma->vm_rcu, vm_area_free_rcu_cb);
 501 #else
 502         __vm_area_free(vma);
 503 #endif
 504 }
 505
 506 static void account_kernel_stack(struct task_struct *tsk, int account)
 507 {
 508         if (IS_ENABLED(CONFIG_VMAP_STACK)) {
 509                 struct vm_struct *vm = task_stack_vm_area(tsk);
 510                 int i;
 511
 512                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 513                         mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
 514                                               account * (PAGE_SIZE / 1024));
 515         } else {
 516                 void *stack = task_stack_page(tsk);
 517
 518                 /* All stack pages are in the same node. */
 519                 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
 520                                       account * (THREAD_SIZE / 1024));
 521         }
 522 }
 523
 524 void exit_task_stack_account(struct task_struct *tsk)
 525 {
 526         account_kernel_stack(tsk, -1);
 527
 528         if (IS_ENABLED(CONFIG_VMAP_STACK)) {
 529                 struct vm_struct *vm;
 530                 int i;
 531
 532                 vm = task_stack_vm_area(tsk);
 533                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 534                         memcg_kmem_uncharge_page(vm->pages[i], 0);
 535         }
 536 }
 537
 538 static void release_task_stack(struct task_struct *tsk)
 539 {
 540         if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
 541                 return;  /* Better to leak the stack than to free prematurely */
 542
 543         free_thread_stack(tsk);
 544 }
 545
 546 #ifdef CONFIG_THREAD_INFO_IN_TASK
 547 void put_task_stack(struct task_struct *tsk)
 548 {
 549         if (refcount_dec_and_test(&tsk->stack_refcount))
 550                 release_task_stack(tsk);
 551 }
 552 #endif
 553
 554 void free_task(struct task_struct *tsk)
 555 {
 556 #ifdef CONFIG_SECCOMP
 557         WARN_ON_ONCE(tsk->seccomp.filter);
 558 #endif
 559         release_user_cpus_ptr(tsk);
 560         scs_release(tsk);
 561
 562 #ifndef CONFIG_THREAD_INFO_IN_TASK
 563         /*
 564          * The task is finally done with both the stack and thread_info,
 565          * so free both.
 566          */
 567         release_task_stack(tsk);
 568 #else
 569         /*
 570          * If the task had a separate stack allocation, it should be gone
 571          * by now.
 572          */
 573         WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
 574 #endif
 575         rt_mutex_debug_task_free(tsk);
 576         ftrace_graph_exit_task(tsk);
 577         arch_release_task_struct(tsk);
 578         if (tsk->flags & PF_KTHREAD)
 579                 free_kthread_struct(tsk);
 580         free_task_struct(tsk);
 581 }
 582 EXPORT_SYMBOL(free_task);
 583
 584 static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
 585 {
 586         struct file *exe_file;
 587
 588         exe_file = get_mm_exe_file(oldmm);
 589         RCU_INIT_POINTER(mm->exe_file, exe_file);
 590         /*
 591          * We depend on the oldmm having properly denied write access to the
 592          * exe_file already.
 593          */
 594         if (exe_file && deny_write_access(exe_file))
 595                 pr_warn_once("deny_write_access() failed in %s\n", __func__);
 596 }
 597
 598 #ifdef CONFIG_MMU
 599 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 600                                         struct mm_struct *oldmm)
 601 {
 602         struct vm_area_struct *mpnt, *tmp;
 603         int retval;
 604         unsigned long charge = 0;
 605         LIST_HEAD(uf);
 606         VMA_ITERATOR(old_vmi, oldmm, 0);
 607         VMA_ITERATOR(vmi, mm, 0);
 608
 609         uprobe_start_dup_mmap();
 610         if (mmap_write_lock_killable(oldmm)) {
 611                 retval = -EINTR;
 612                 goto fail_uprobe_end;
 613         }
 614         flush_cache_dup_mm(oldmm);
 615         uprobe_dup_mmap(oldmm, mm);
 616         /*
 617          * Not linked in yet - no deadlock potential:
 618          */
 619         mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
 620
 621         /* No ordering required: file already has been exposed. */
 622         dup_mm_exe_file(mm, oldmm);
 623
 624         mm->total_vm = oldmm->total_vm;
 625         mm->data_vm = oldmm->data_vm;
 626         mm->exec_vm = oldmm->exec_vm;
 627         mm->stack_vm = oldmm->stack_vm;
 628
 629         retval = ksm_fork(mm, oldmm);
 630         if (retval)
 631                 goto out;
 632         khugepaged_fork(mm, oldmm);
 633
 634         retval = vma_iter_bulk_alloc(&vmi, oldmm->map_count);
 635         if (retval)
 636                 goto out;
 637
 638         for_each_vma(old_vmi, mpnt) {
 639                 struct file *file;
 640
 641                 if (mpnt->vm_flags & VM_DONTCOPY) {
 642                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 643                         continue;
 644                 }
 645                 charge = 0;
 646                 /*
 647                  * Don't duplicate many vmas if we've been oom-killed (for
 648                  * example)
 649                  */
 650                 if (fatal_signal_pending(current)) {
 651                         retval = -EINTR;
 652                         goto loop_out;
 653                 }
 654                 if (mpnt->vm_flags & VM_ACCOUNT) {
 655                         unsigned long len = vma_pages(mpnt);
 656
 657                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 658                                 goto fail_nomem;
 659                         charge = len;
 660                 }
 661                 tmp = vm_area_dup(mpnt);
 662                 if (!tmp)
 663                         goto fail_nomem;
 664                 retval = vma_dup_policy(mpnt, tmp);
 665                 if (retval)
 666                         goto fail_nomem_policy;
 667                 tmp->vm_mm = mm;
 668                 retval = dup_userfaultfd(tmp, &uf);
 669                 if (retval)
 670                         goto fail_nomem_anon_vma_fork;
 671                 if (tmp->vm_flags & VM_WIPEONFORK) {
 672                         /*
 673                          * VM_WIPEONFORK gets a clean slate in the child.
 674                          * Don't prepare anon_vma until fault since we don't
 675                          * copy page for current vma.
 676                          */
 677                         tmp->anon_vma = NULL;
 678                 } else if (anon_vma_fork(tmp, mpnt))
 679                         goto fail_nomem_anon_vma_fork;
 680                 vm_flags_clear(tmp, VM_LOCKED_MASK);
 681                 file = tmp->vm_file;
 682                 if (file) {
 683                         struct address_space *mapping = file->f_mapping;
 684
 685                         get_file(file);
 686                         i_mmap_lock_write(mapping);
 687                         if (tmp->vm_flags & VM_SHARED)
 688                                 mapping_allow_writable(mapping);
 689                         flush_dcache_mmap_lock(mapping);
 690                         /* insert tmp into the share list, just after mpnt */
 691                         vma_interval_tree_insert_after(tmp, mpnt,
 692                                         &mapping->i_mmap);
 693                         flush_dcache_mmap_unlock(mapping);
 694                         i_mmap_unlock_write(mapping);
 695                 }
 696
 697                 /*
 698                  * Copy/update hugetlb private vma information.
 699                  */
 700                 if (is_vm_hugetlb_page(tmp))
 701                         hugetlb_dup_vma_private(tmp);
 702
 703                 /* Link the vma into the MT */
 704                 if (vma_iter_bulk_store(&vmi, tmp))
 705                         goto fail_nomem_vmi_store;
 706
 707                 mm->map_count++;
 708                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 709                         retval = copy_page_range(tmp, mpnt);
 710
 711                 if (tmp->vm_ops && tmp->vm_ops->open)
 712                         tmp->vm_ops->open(tmp);
 713
 714                 if (retval)
 715                         goto loop_out;
 716         }
 717         /* a new mm has just been created */
 718         retval = arch_dup_mmap(oldmm, mm);
 719 loop_out:
 720         vma_iter_free(&vmi);
 721 out:
 722         mmap_write_unlock(mm);
 723         flush_tlb_mm(oldmm);
 724         mmap_write_unlock(oldmm);
 725         dup_userfaultfd_complete(&uf);
 726 fail_uprobe_end:
 727         uprobe_end_dup_mmap();
 728         return retval;
 729
 730 fail_nomem_vmi_store:
 731         unlink_anon_vmas(tmp);
 732 fail_nomem_anon_vma_fork:
 733         mpol_put(vma_policy(tmp));
 734 fail_nomem_policy:
 735         vm_area_free(tmp);
 736 fail_nomem:
 737         retval = -ENOMEM;
 738         vm_unacct_memory(charge);
 739         goto loop_out;
 740 }
 741
 742 static inline int mm_alloc_pgd(struct mm_struct *mm)
 743 {
 744         mm->pgd = pgd_alloc(mm);
 745         if (unlikely(!mm->pgd))
 746                 return -ENOMEM;
 747         return 0;
 748 }
 749
 750 static inline void mm_free_pgd(struct mm_struct *mm)
 751 {
 752         pgd_free(mm, mm->pgd);
 753 }
 754 #else
 755 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 756 {
 757         mmap_write_lock(oldmm);
 758         dup_mm_exe_file(mm, oldmm);
 759         mmap_write_unlock(oldmm);
 760         return 0;
 761 }
 762 #define mm_alloc_pgd(mm)        (0)
 763 #define mm_free_pgd(mm)
 764 #endif /* CONFIG_MMU */
 765
 766 static void check_mm(struct mm_struct *mm)
 767 {
 768         int i;
 769
 770         BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
 771                          "Please make sure 'struct resident_page_types[]' is updated as well");
 772
 773         for (i = 0; i < NR_MM_COUNTERS; i++) {
 774                 long x = percpu_counter_sum(&mm->rss_stat[i]);
 775
 776                 if (unlikely(x))
 777                         pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
 778                                  mm, resident_page_types[i], x);
 779         }
 780
 781         if (mm_pgtables_bytes(mm))
 782                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 783                                 mm_pgtables_bytes(mm));
 784
 785 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 786         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 787 #endif
 788 }
 789
 790 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 791 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 792
 793 static void do_check_lazy_tlb(void *arg)
 794 {
 795         struct mm_struct *mm = arg;
 796
 797         WARN_ON_ONCE(current->active_mm == mm);
 798 }
 799
 800 static void do_shoot_lazy_tlb(void *arg)
 801 {
 802         struct mm_struct *mm = arg;
 803
 804         if (current->active_mm == mm) {
 805                 WARN_ON_ONCE(current->mm);
 806                 current->active_mm = &init_mm;
 807                 switch_mm(mm, &init_mm, current);
 808         }
 809 }
 810
 811 static void cleanup_lazy_tlbs(struct mm_struct *mm)
 812 {
 813         if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) {
 814                 /*
 815                  * In this case, lazy tlb mms are refounted and would not reach
 816                  * __mmdrop until all CPUs have switched away and mmdrop()ed.
 817                  */
 818                 return;
 819         }
 820
 821         /*
 822          * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
 823          * requires lazy mm users to switch to another mm when the refcount
 824          * drops to zero, before the mm is freed. This requires IPIs here to
 825          * switch kernel threads to init_mm.
 826          *
 827          * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
 828          * switch with the final userspace teardown TLB flush which leaves the
 829          * mm lazy on this CPU but no others, reducing the need for additional
 830          * IPIs here. There are cases where a final IPI is still required here,
 831          * such as the final mmdrop being performed on a different CPU than the
 832          * one exiting, or kernel threads using the mm when userspace exits.
 833          *
 834          * IPI overheads have not found to be expensive, but they could be
 835          * reduced in a number of possible ways, for example (roughly
 836          * increasing order of complexity):
 837          * - The last lazy reference created by exit_mm() could instead switch
 838          *   to init_mm, however it's probable this will run on the same CPU
 839          *   immediately afterwards, so this may not reduce IPIs much.
 840          * - A batch of mms requiring IPIs could be gathered and freed at once.
 841          * - CPUs store active_mm where it can be remotely checked without a
 842          *   lock, to filter out false-positives in the cpumask.
 843          * - After mm_users or mm_count reaches zero, switching away from the
 844          *   mm could clear mm_cpumask to reduce some IPIs, perhaps together
 845          *   with some batching or delaying of the final IPIs.
 846          * - A delayed freeing and RCU-like quiescing sequence based on mm
 847          *   switching to avoid IPIs completely.
 848          */
 849         on_each_cpu_mask(mm_cpumask(mm), do_shoot_lazy_tlb, (void *)mm, 1);
 850         if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES))
 851                 on_each_cpu(do_check_lazy_tlb, (void *)mm, 1);
 852 }
 853
 854 /*
 855  * Called when the last reference to the mm
 856  * is dropped: either by a lazy thread or by
 857  * mmput. Free the page directory and the mm.
 858  */
 859 void __mmdrop(struct mm_struct *mm)
 860 {
 861         int i;
 862
 863         BUG_ON(mm == &init_mm);
 864         WARN_ON_ONCE(mm == current->mm);
 865
 866         /* Ensure no CPUs are using this as their lazy tlb mm */
 867         cleanup_lazy_tlbs(mm);
 868
 869         WARN_ON_ONCE(mm == current->active_mm);
 870         mm_free_pgd(mm);
 871         destroy_context(mm);
 872         mmu_notifier_subscriptions_destroy(mm);
 873         check_mm(mm);
 874         put_user_ns(mm->user_ns);
 875         mm_pasid_drop(mm);
 876
 877         for (i = 0; i < NR_MM_COUNTERS; i++)
 878                 percpu_counter_destroy(&mm->rss_stat[i]);
 879         free_mm(mm);
 880 }
 881 EXPORT_SYMBOL_GPL(__mmdrop);
 882
 883 static void mmdrop_async_fn(struct work_struct *work)
 884 {
 885         struct mm_struct *mm;
 886
 887         mm = container_of(work, struct mm_struct, async_put_work);
 888         __mmdrop(mm);
 889 }
 890
 891 static void mmdrop_async(struct mm_struct *mm)
 892 {
 893         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 894                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 895                 schedule_work(&mm->async_put_work);
 896         }
 897 }
 898
 899 static inline void free_signal_struct(struct signal_struct *sig)
 900 {
 901         taskstats_tgid_free(sig);
 902         sched_autogroup_exit(sig);
 903         /*
 904          * __mmdrop is not safe to call from softirq context on x86 due to
 905          * pgd_dtor so postpone it to the async context
 906          */
 907         if (sig->oom_mm)
 908                 mmdrop_async(sig->oom_mm);
 909         kmem_cache_free(signal_cachep, sig);
 910 }
 911
 912 static inline void put_signal_struct(struct signal_struct *sig)
 913 {
 914         if (refcount_dec_and_test(&sig->sigcnt))
 915                 free_signal_struct(sig);
 916 }
 917
 918 void __put_task_struct(struct task_struct *tsk)
 919 {
 920         WARN_ON(!tsk->exit_state);
 921         WARN_ON(refcount_read(&tsk->usage));
 922         WARN_ON(tsk == current);
 923
 924         io_uring_free(tsk);
 925         cgroup_free(tsk);
 926         task_numa_free(tsk, true);
 927         security_task_free(tsk);
 928         bpf_task_storage_free(tsk);
 929         exit_creds(tsk);
 930         delayacct_tsk_free(tsk);
 931         put_signal_struct(tsk->signal);
 932         sched_core_free(tsk);
 933         free_task(tsk);
 934 }
 935 EXPORT_SYMBOL_GPL(__put_task_struct);
 936
 937 void __init __weak arch_task_cache_init(void) { }
 938
 939 /*
 940  * set_max_threads
 941  */
 942 static void set_max_threads(unsigned int max_threads_suggested)
 943 {
 944         u64 threads;
 945         unsigned long nr_pages = totalram_pages();
 946
 947         /*
 948          * The number of threads shall be limited such that the thread
 949          * structures may only consume a small part of the available memory.
 950          */
 951         if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
 952                 threads = MAX_THREADS;
 953         else
 954                 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
 955                                     (u64) THREAD_SIZE * 8UL);
 956
 957         if (threads > max_threads_suggested)
 958                 threads = max_threads_suggested;
 959
 960         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 961 }
 962
 963 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 964 /* Initialized by the architecture: */
 965 int arch_task_struct_size __read_mostly;
 966 #endif
 967
 968 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 969 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 970 {
 971         /* Fetch thread_struct whitelist for the architecture. */
 972         arch_thread_struct_whitelist(offset, size);
 973
 974         /*
 975          * Handle zero-sized whitelist or empty thread_struct, otherwise
 976          * adjust offset to position of thread_struct in task_struct.
 977          */
 978         if (unlikely(*size == 0))
 979                 *offset = 0;
 980         else
 981                 *offset += offsetof(struct task_struct, thread);
 982 }
 983 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
 984
 985 void __init fork_init(void)
 986 {
 987         int i;
 988 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 989 #ifndef ARCH_MIN_TASKALIGN
 990 #define ARCH_MIN_TASKALIGN      0
 991 #endif
 992         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 993         unsigned long useroffset, usersize;
 994
 995         /* create a slab on which task_structs can be allocated */
 996         task_struct_whitelist(&useroffset, &usersize);
 997         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 998                         arch_task_struct_size, align,
 999                         SLAB_PANIC|SLAB_ACCOUNT,
1000                         useroffset, usersize, NULL);
1001 #endif
1002
1003         /* do the arch specific task caches init */
1004         arch_task_cache_init();
1005
1006         set_max_threads(MAX_THREADS);
1007
1008         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
1009         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
1010         init_task.signal->rlim[RLIMIT_SIGPENDING] =
1011                 init_task.signal->rlim[RLIMIT_NPROC];
1012
1013         for (i = 0; i < UCOUNT_COUNTS; i++)
1014                 init_user_ns.ucount_max[i] = max_threads/2;
1015
1016         set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC,      RLIM_INFINITY);
1017         set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE,   RLIM_INFINITY);
1018         set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
1019         set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK,    RLIM_INFINITY);
1020
1021 #ifdef CONFIG_VMAP_STACK
1022         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
1023                           NULL, free_vm_stack_cache);
1024 #endif
1025
1026         scs_init();
1027
1028         lockdep_init_task(&init_task);
1029         uprobes_init();
1030 }
1031
1032 int __weak arch_dup_task_struct(struct task_struct *dst,
1033                                                struct task_struct *src)
1034 {
1035         *dst = *src;
1036         return 0;
1037 }
1038
1039 void set_task_stack_end_magic(struct task_struct *tsk)
1040 {
1041         unsigned long *stackend;
1042
1043         stackend = end_of_stack(tsk);
1044         *stackend = STACK_END_MAGIC;    /* for overflow detection */
1045 }
1046
1047 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
1048 {
1049         struct task_struct *tsk;
1050         int err;
1051
1052         if (node == NUMA_NO_NODE)
1053                 node = tsk_fork_get_node(orig);
1054         tsk = alloc_task_struct_node(node);
1055         if (!tsk)
1056                 return NULL;
1057
1058         err = arch_dup_task_struct(tsk, orig);
1059         if (err)
1060                 goto free_tsk;
1061
1062         err = alloc_thread_stack_node(tsk, node);
1063         if (err)
1064                 goto free_tsk;
1065
1066 #ifdef CONFIG_THREAD_INFO_IN_TASK
1067         refcount_set(&tsk->stack_refcount, 1);
1068 #endif
1069         account_kernel_stack(tsk, 1);
1070
1071         err = scs_prepare(tsk, node);
1072         if (err)
1073                 goto free_stack;
1074
1075 #ifdef CONFIG_SECCOMP
1076         /*
1077          * We must handle setting up seccomp filters once we're under
1078          * the sighand lock in case orig has changed between now and
1079          * then. Until then, filter must be NULL to avoid messing up
1080          * the usage counts on the error path calling free_task.
1081          */
1082         tsk->seccomp.filter = NULL;
1083 #endif
1084
1085         setup_thread_stack(tsk, orig);
1086         clear_user_return_notifier(tsk);
1087         clear_tsk_need_resched(tsk);
1088         set_task_stack_end_magic(tsk);
1089         clear_syscall_work_syscall_user_dispatch(tsk);
1090
1091 #ifdef CONFIG_STACKPROTECTOR
1092         tsk->stack_canary = get_random_canary();
1093 #endif
1094         if (orig->cpus_ptr == &orig->cpus_mask)
1095                 tsk->cpus_ptr = &tsk->cpus_mask;
1096         dup_user_cpus_ptr(tsk, orig, node);
1097
1098         /*
1099          * One for the user space visible state that goes away when reaped.
1100          * One for the scheduler.
1101          */
1102         refcount_set(&tsk->rcu_users, 2);
1103         /* One for the rcu users */
1104         refcount_set(&tsk->usage, 1);
1105 #ifdef CONFIG_BLK_DEV_IO_TRACE
1106         tsk->btrace_seq = 0;
1107 #endif
1108         tsk->splice_pipe = NULL;
1109         tsk->task_frag.page = NULL;
1110         tsk->wake_q.next = NULL;
1111         tsk->worker_private = NULL;
1112
1113         kcov_task_init(tsk);
1114         kmsan_task_create(tsk);
1115         kmap_local_fork(tsk);
1116
1117 #ifdef CONFIG_FAULT_INJECTION
1118         tsk->fail_nth = 0;
1119 #endif
1120
1121 #ifdef CONFIG_BLK_CGROUP
1122         tsk->throttle_disk = NULL;
1123         tsk->use_memdelay = 0;
1124 #endif
1125
1126 #ifdef CONFIG_IOMMU_SVA
1127         tsk->pasid_activated = 0;
1128 #endif
1129
1130 #ifdef CONFIG_MEMCG
1131         tsk->active_memcg = NULL;
1132 #endif
1133
1134 #ifdef CONFIG_CPU_SUP_INTEL
1135         tsk->reported_split_lock = 0;
1136 #endif
1137
1138 #ifdef CONFIG_SCHED_MM_CID
1139         tsk->mm_cid = -1;
1140         tsk->mm_cid_active = 0;
1141 #endif
1142         return tsk;
1143
1144 free_stack:
1145         exit_task_stack_account(tsk);
1146         free_thread_stack(tsk);
1147 free_tsk:
1148         free_task_struct(tsk);
1149         return NULL;
1150 }
1151
1152 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1153
1154 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1155
1156 static int __init coredump_filter_setup(char *s)
1157 {
1158         default_dump_filter =
1159                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1160                 MMF_DUMP_FILTER_MASK;
1161         return 1;
1162 }
1163
1164 __setup("coredump_filter=", coredump_filter_setup);
1165
1166 #include <linux/init_task.h>
1167
1168 static void mm_init_aio(struct mm_struct *mm)
1169 {
1170 #ifdef CONFIG_AIO
1171         spin_lock_init(&mm->ioctx_lock);
1172         mm->ioctx_table = NULL;
1173 #endif
1174 }
1175
1176 static __always_inline void mm_clear_owner(struct mm_struct *mm,
1177                                            struct task_struct *p)
1178 {
1179 #ifdef CONFIG_MEMCG
1180         if (mm->owner == p)
1181                 WRITE_ONCE(mm->owner, NULL);
1182 #endif
1183 }
1184
1185 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1186 {
1187 #ifdef CONFIG_MEMCG
1188         mm->owner = p;
1189 #endif
1190 }
1191
1192 static void mm_init_uprobes_state(struct mm_struct *mm)
1193 {
1194 #ifdef CONFIG_UPROBES
1195         mm->uprobes_state.xol_area = NULL;
1196 #endif
1197 }
1198
1199 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1200         struct user_namespace *user_ns)
1201 {
1202         int i;
1203
1204         mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1205         mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1206         atomic_set(&mm->mm_users, 1);
1207         atomic_set(&mm->mm_count, 1);
1208         seqcount_init(&mm->write_protect_seq);
1209         mmap_init_lock(mm);
1210         INIT_LIST_HEAD(&mm->mmlist);
1211         mm_pgtables_bytes_init(mm);
1212         mm->map_count = 0;
1213         mm->locked_vm = 0;
1214         atomic64_set(&mm->pinned_vm, 0);
1215         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1216         spin_lock_init(&mm->page_table_lock);
1217         spin_lock_init(&mm->arg_lock);
1218         mm_init_cpumask(mm);
1219         mm_init_aio(mm);
1220         mm_init_owner(mm, p);
1221         mm_pasid_init(mm);
1222         RCU_INIT_POINTER(mm->exe_file, NULL);
1223         mmu_notifier_subscriptions_init(mm);
1224         init_tlb_flush_pending(mm);
1225 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1226         mm->pmd_huge_pte = NULL;
1227 #endif
1228         mm_init_uprobes_state(mm);
1229         hugetlb_count_init(mm);
1230
1231         if (current->mm) {
1232                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1233                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1234         } else {
1235                 mm->flags = default_dump_filter;
1236                 mm->def_flags = 0;
1237         }
1238
1239         if (mm_alloc_pgd(mm))
1240                 goto fail_nopgd;
1241
1242         if (init_new_context(p, mm))
1243                 goto fail_nocontext;
1244
1245         for (i = 0; i < NR_MM_COUNTERS; i++)
1246                 if (percpu_counter_init(&mm->rss_stat[i], 0, GFP_KERNEL_ACCOUNT))
1247                         goto fail_pcpu;
1248
1249         mm->user_ns = get_user_ns(user_ns);
1250         lru_gen_init_mm(mm);
1251         mm_init_cid(mm);
1252         return mm;
1253
1254 fail_pcpu:
1255         while (i > 0)
1256                 percpu_counter_destroy(&mm->rss_stat[--i]);
1257 fail_nocontext:
1258         mm_free_pgd(mm);
1259 fail_nopgd:
1260         free_mm(mm);
1261         return NULL;
1262 }
1263
1264 /*
1265  * Allocate and initialize an mm_struct.
1266  */
1267 struct mm_struct *mm_alloc(void)
1268 {
1269         struct mm_struct *mm;
1270
1271         mm = allocate_mm();
1272         if (!mm)
1273                 return NULL;
1274
1275         memset(mm, 0, sizeof(*mm));
1276         return mm_init(mm, current, current_user_ns());
1277 }
1278
1279 static inline void __mmput(struct mm_struct *mm)
1280 {
1281         VM_BUG_ON(atomic_read(&mm->mm_users));
1282
1283         uprobe_clear_state(mm);
1284         exit_aio(mm);
1285         ksm_exit(mm);
1286         khugepaged_exit(mm); /* must run before exit_mmap */
1287         exit_mmap(mm);
1288         mm_put_huge_zero_page(mm);
1289         set_mm_exe_file(mm, NULL);
1290         if (!list_empty(&mm->mmlist)) {
1291                 spin_lock(&mmlist_lock);
1292                 list_del(&mm->mmlist);
1293                 spin_unlock(&mmlist_lock);
1294         }
1295         if (mm->binfmt)
1296                 module_put(mm->binfmt->module);
1297         lru_gen_del_mm(mm);
1298         mmdrop(mm);
1299 }
1300
1301 /*
1302  * Decrement the use count and release all resources for an mm.
1303  */
1304 void mmput(struct mm_struct *mm)
1305 {
1306         might_sleep();
1307
1308         if (atomic_dec_and_test(&mm->mm_users))
1309                 __mmput(mm);
1310 }
1311 EXPORT_SYMBOL_GPL(mmput);
1312
1313 #ifdef CONFIG_MMU
1314 static void mmput_async_fn(struct work_struct *work)
1315 {
1316         struct mm_struct *mm = container_of(work, struct mm_struct,
1317                                             async_put_work);
1318
1319         __mmput(mm);
1320 }
1321
1322 void mmput_async(struct mm_struct *mm)
1323 {
1324         if (atomic_dec_and_test(&mm->mm_users)) {
1325                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1326                 schedule_work(&mm->async_put_work);
1327         }
1328 }
1329 EXPORT_SYMBOL_GPL(mmput_async);
1330 #endif
1331
1332 /**
1333  * set_mm_exe_file - change a reference to the mm's executable file
1334  *
1335  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1336  *
1337  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1338  * invocations: in mmput() nobody alive left, in execve task is single
1339  * threaded.
1340  *
1341  * Can only fail if new_exe_file != NULL.
1342  */
1343 int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1344 {
1345         struct file *old_exe_file;
1346
1347         /*
1348          * It is safe to dereference the exe_file without RCU as
1349          * this function is only called if nobody else can access
1350          * this mm -- see comment above for justification.
1351          */
1352         old_exe_file = rcu_dereference_raw(mm->exe_file);
1353
1354         if (new_exe_file) {
1355                 /*
1356                  * We expect the caller (i.e., sys_execve) to already denied
1357                  * write access, so this is unlikely to fail.
1358                  */
1359                 if (unlikely(deny_write_access(new_exe_file)))
1360                         return -EACCES;
1361                 get_file(new_exe_file);
1362         }
1363         rcu_assign_pointer(mm->exe_file, new_exe_file);
1364         if (old_exe_file) {
1365                 allow_write_access(old_exe_file);
1366                 fput(old_exe_file);
1367         }
1368         return 0;
1369 }
1370
1371 /**
1372  * replace_mm_exe_file - replace a reference to the mm's executable file
1373  *
1374  * This changes mm's executable file (shown as symlink /proc/[pid]/exe),
1375  * dealing with concurrent invocation and without grabbing the mmap lock in
1376  * write mode.
1377  *
1378  * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1379  */
1380 int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1381 {
1382         struct vm_area_struct *vma;
1383         struct file *old_exe_file;
1384         int ret = 0;
1385
1386         /* Forbid mm->exe_file change if old file still mapped. */
1387         old_exe_file = get_mm_exe_file(mm);
1388         if (old_exe_file) {
1389                 VMA_ITERATOR(vmi, mm, 0);
1390                 mmap_read_lock(mm);
1391                 for_each_vma(vmi, vma) {
1392                         if (!vma->vm_file)
1393                                 continue;
1394                         if (path_equal(&vma->vm_file->f_path,
1395                                        &old_exe_file->f_path)) {
1396                                 ret = -EBUSY;
1397                                 break;
1398                         }
1399                 }
1400                 mmap_read_unlock(mm);
1401                 fput(old_exe_file);
1402                 if (ret)
1403                         return ret;
1404         }
1405
1406         /* set the new file, lockless */
1407         ret = deny_write_access(new_exe_file);
1408         if (ret)
1409                 return -EACCES;
1410         get_file(new_exe_file);
1411
1412         old_exe_file = xchg(&mm->exe_file, new_exe_file);
1413         if (old_exe_file) {
1414                 /*
1415                  * Don't race with dup_mmap() getting the file and disallowing
1416                  * write access while someone might open the file writable.
1417                  */
1418                 mmap_read_lock(mm);
1419                 allow_write_access(old_exe_file);
1420                 fput(old_exe_file);
1421                 mmap_read_unlock(mm);
1422         }
1423         return 0;
1424 }
1425
1426 /**
1427  * get_mm_exe_file - acquire a reference to the mm's executable file
1428  *
1429  * Returns %NULL if mm has no associated executable file.
1430  * User must release file via fput().
1431  */
1432 struct file *get_mm_exe_file(struct mm_struct *mm)
1433 {
1434         struct file *exe_file;
1435
1436         rcu_read_lock();
1437         exe_file = rcu_dereference(mm->exe_file);
1438         if (exe_file && !get_file_rcu(exe_file))
1439                 exe_file = NULL;
1440         rcu_read_unlock();
1441         return exe_file;
1442 }
1443
1444 /**
1445  * get_task_exe_file - acquire a reference to the task's executable file
1446  *
1447  * Returns %NULL if task's mm (if any) has no associated executable file or
1448  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1449  * User must release file via fput().
1450  */
1451 struct file *get_task_exe_file(struct task_struct *task)
1452 {
1453         struct file *exe_file = NULL;
1454         struct mm_struct *mm;
1455
1456         task_lock(task);
1457         mm = task->mm;
1458         if (mm) {
1459                 if (!(task->flags & PF_KTHREAD))
1460                         exe_file = get_mm_exe_file(mm);
1461         }
1462         task_unlock(task);
1463         return exe_file;
1464 }
1465
1466 /**
1467  * get_task_mm - acquire a reference to the task's mm
1468  *
1469  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1470  * this kernel workthread has transiently adopted a user mm with use_mm,
1471  * to do its AIO) is not set and if so returns a reference to it, after
1472  * bumping up the use count.  User must release the mm via mmput()
1473  * after use.  Typically used by /proc and ptrace.
1474  */
1475 struct mm_struct *get_task_mm(struct task_struct *task)
1476 {
1477         struct mm_struct *mm;
1478
1479         task_lock(task);
1480         mm = task->mm;
1481         if (mm) {
1482                 if (task->flags & PF_KTHREAD)
1483                         mm = NULL;
1484                 else
1485                         mmget(mm);
1486         }
1487         task_unlock(task);
1488         return mm;
1489 }
1490 EXPORT_SYMBOL_GPL(get_task_mm);
1491
1492 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1493 {
1494         struct mm_struct *mm;
1495         int err;
1496
1497         err =  down_read_killable(&task->signal->exec_update_lock);
1498         if (err)
1499                 return ERR_PTR(err);
1500
1501         mm = get_task_mm(task);
1502         if (mm && mm != current->mm &&
1503                         !ptrace_may_access(task, mode)) {
1504                 mmput(mm);
1505                 mm = ERR_PTR(-EACCES);
1506         }
1507         up_read(&task->signal->exec_update_lock);
1508
1509         return mm;
1510 }
1511
1512 static void complete_vfork_done(struct task_struct *tsk)
1513 {
1514         struct completion *vfork;
1515
1516         task_lock(tsk);
1517         vfork = tsk->vfork_done;
1518         if (likely(vfork)) {
1519                 tsk->vfork_done = NULL;
1520                 complete(vfork);
1521         }
1522         task_unlock(tsk);
1523 }
1524
1525 static int wait_for_vfork_done(struct task_struct *child,
1526                                 struct completion *vfork)
1527 {
1528         unsigned int state = TASK_UNINTERRUPTIBLE|TASK_KILLABLE|TASK_FREEZABLE;
1529         int killed;
1530
1531         cgroup_enter_frozen();
1532         killed = wait_for_completion_state(vfork, state);
1533         cgroup_leave_frozen(false);
1534
1535         if (killed) {
1536                 task_lock(child);
1537                 child->vfork_done = NULL;
1538                 task_unlock(child);
1539         }
1540
1541         put_task_struct(child);
1542         return killed;
1543 }
1544
1545 /* Please note the differences between mmput and mm_release.
1546  * mmput is called whenever we stop holding onto a mm_struct,
1547  * error success whatever.
1548  *
1549  * mm_release is called after a mm_struct has been removed
1550  * from the current process.
1551  *
1552  * This difference is important for error handling, when we
1553  * only half set up a mm_struct for a new process and need to restore
1554  * the old one.  Because we mmput the new mm_struct before
1555  * restoring the old one. . .
1556  * Eric Biederman 10 January 1998
1557  */
1558 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1559 {
1560         uprobe_free_utask(tsk);
1561
1562         /* Get rid of any cached register state */
1563         deactivate_mm(tsk, mm);
1564
1565         /*
1566          * Signal userspace if we're not exiting with a core dump
1567          * because we want to leave the value intact for debugging
1568          * purposes.
1569          */
1570         if (tsk->clear_child_tid) {
1571                 if (atomic_read(&mm->mm_users) > 1) {
1572                         /*
1573                          * We don't check the error code - if userspace has
1574                          * not set up a proper pointer then tough luck.
1575                          */
1576                         put_user(0, tsk->clear_child_tid);
1577                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1578                                         1, NULL, NULL, 0, 0);
1579                 }
1580                 tsk->clear_child_tid = NULL;
1581         }
1582
1583         /*
1584          * All done, finally we can wake up parent and return this mm to him.
1585          * Also kthread_stop() uses this completion for synchronization.
1586          */
1587         if (tsk->vfork_done)
1588                 complete_vfork_done(tsk);
1589 }
1590
1591 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1592 {
1593         futex_exit_release(tsk);
1594         mm_release(tsk, mm);
1595 }
1596
1597 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1598 {
1599         futex_exec_release(tsk);
1600         mm_release(tsk, mm);
1601 }
1602
1603 /**
1604  * dup_mm() - duplicates an existing mm structure
1605  * @tsk: the task_struct with which the new mm will be associated.
1606  * @oldmm: the mm to duplicate.
1607  *
1608  * Allocates a new mm structure and duplicates the provided @oldmm structure
1609  * content into it.
1610  *
1611  * Return: the duplicated mm or NULL on failure.
1612  */
1613 static struct mm_struct *dup_mm(struct task_struct *tsk,
1614                                 struct mm_struct *oldmm)
1615 {
1616         struct mm_struct *mm;
1617         int err;
1618
1619         mm = allocate_mm();
1620         if (!mm)
1621                 goto fail_nomem;
1622
1623         memcpy(mm, oldmm, sizeof(*mm));
1624
1625         if (!mm_init(mm, tsk, mm->user_ns))
1626                 goto fail_nomem;
1627
1628         err = dup_mmap(mm, oldmm);
1629         if (err)
1630                 goto free_pt;
1631
1632         mm->hiwater_rss = get_mm_rss(mm);
1633         mm->hiwater_vm = mm->total_vm;
1634
1635         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1636                 goto free_pt;
1637
1638         return mm;
1639
1640 free_pt:
1641         /* don't put binfmt in mmput, we haven't got module yet */
1642         mm->binfmt = NULL;
1643         mm_init_owner(mm, NULL);
1644         mmput(mm);
1645
1646 fail_nomem:
1647         return NULL;
1648 }
1649
1650 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1651 {
1652         struct mm_struct *mm, *oldmm;
1653
1654         tsk->min_flt = tsk->maj_flt = 0;
1655         tsk->nvcsw = tsk->nivcsw = 0;
1656 #ifdef CONFIG_DETECT_HUNG_TASK
1657         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1658         tsk->last_switch_time = 0;
1659 #endif
1660
1661         tsk->mm = NULL;
1662         tsk->active_mm = NULL;
1663
1664         /*
1665          * Are we cloning a kernel thread?
1666          *
1667          * We need to steal a active VM for that..
1668          */
1669         oldmm = current->mm;
1670         if (!oldmm)
1671                 return 0;
1672
1673         if (clone_flags & CLONE_VM) {
1674                 mmget(oldmm);
1675                 mm = oldmm;
1676         } else {
1677                 mm = dup_mm(tsk, current->mm);
1678                 if (!mm)
1679                         return -ENOMEM;
1680         }
1681
1682         tsk->mm = mm;
1683         tsk->active_mm = mm;
1684         sched_mm_cid_fork(tsk);
1685         return 0;
1686 }
1687
1688 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1689 {
1690         struct fs_struct *fs = current->fs;
1691         if (clone_flags & CLONE_FS) {
1692                 /* tsk->fs is already what we want */
1693                 spin_lock(&fs->lock);
1694                 if (fs->in_exec) {
1695                         spin_unlock(&fs->lock);
1696                         return -EAGAIN;
1697                 }
1698                 fs->users++;
1699                 spin_unlock(&fs->lock);
1700                 return 0;
1701         }
1702         tsk->fs = copy_fs_struct(fs);
1703         if (!tsk->fs)
1704                 return -ENOMEM;
1705         return 0;
1706 }
1707
1708 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1709 {
1710         struct files_struct *oldf, *newf;
1711         int error = 0;
1712
1713         /*
1714          * A background process may not have any files ...
1715          */
1716         oldf = current->files;
1717         if (!oldf)
1718                 goto out;
1719
1720         if (clone_flags & CLONE_FILES) {
1721                 atomic_inc(&oldf->count);
1722                 goto out;
1723         }
1724
1725         newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1726         if (!newf)
1727                 goto out;
1728
1729         tsk->files = newf;
1730         error = 0;
1731 out:
1732         return error;
1733 }
1734
1735 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1736 {
1737         struct sighand_struct *sig;
1738
1739         if (clone_flags & CLONE_SIGHAND) {
1740                 refcount_inc(&current->sighand->count);
1741                 return 0;
1742         }
1743         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1744         RCU_INIT_POINTER(tsk->sighand, sig);
1745         if (!sig)
1746                 return -ENOMEM;
1747
1748         refcount_set(&sig->count, 1);
1749         spin_lock_irq(&current->sighand->siglock);
1750         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1751         spin_unlock_irq(&current->sighand->siglock);
1752
1753         /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1754         if (clone_flags & CLONE_CLEAR_SIGHAND)
1755                 flush_signal_handlers(tsk, 0);
1756
1757         return 0;
1758 }
1759
1760 void __cleanup_sighand(struct sighand_struct *sighand)
1761 {
1762         if (refcount_dec_and_test(&sighand->count)) {
1763                 signalfd_cleanup(sighand);
1764                 /*
1765                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1766                  * without an RCU grace period, see __lock_task_sighand().
1767                  */
1768                 kmem_cache_free(sighand_cachep, sighand);
1769         }
1770 }
1771
1772 /*
1773  * Initialize POSIX timer handling for a thread group.
1774  */
1775 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1776 {
1777         struct posix_cputimers *pct = &sig->posix_cputimers;
1778         unsigned long cpu_limit;
1779
1780         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1781         posix_cputimers_group_init(pct, cpu_limit);
1782 }
1783
1784 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1785 {
1786         struct signal_struct *sig;
1787
1788         if (clone_flags & CLONE_THREAD)
1789                 return 0;
1790
1791         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1792         tsk->signal = sig;
1793         if (!sig)
1794                 return -ENOMEM;
1795
1796         sig->nr_threads = 1;
1797         sig->quick_threads = 1;
1798         atomic_set(&sig->live, 1);
1799         refcount_set(&sig->sigcnt, 1);
1800
1801         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1802         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1803         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1804
1805         init_waitqueue_head(&sig->wait_chldexit);
1806         sig->curr_target = tsk;
1807         init_sigpending(&sig->shared_pending);
1808         INIT_HLIST_HEAD(&sig->multiprocess);
1809         seqlock_init(&sig->stats_lock);
1810         prev_cputime_init(&sig->prev_cputime);
1811
1812 #ifdef CONFIG_POSIX_TIMERS
1813         INIT_LIST_HEAD(&sig->posix_timers);
1814         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1815         sig->real_timer.function = it_real_fn;
1816 #endif
1817
1818         task_lock(current->group_leader);
1819         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1820         task_unlock(current->group_leader);
1821
1822         posix_cpu_timers_init_group(sig);
1823
1824         tty_audit_fork(sig);
1825         sched_autogroup_fork(sig);
1826
1827         sig->oom_score_adj = current->signal->oom_score_adj;
1828         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1829
1830         mutex_init(&sig->cred_guard_mutex);
1831         init_rwsem(&sig->exec_update_lock);
1832
1833         return 0;
1834 }
1835
1836 static void copy_seccomp(struct task_struct *p)
1837 {
1838 #ifdef CONFIG_SECCOMP
1839         /*
1840          * Must be called with sighand->lock held, which is common to
1841          * all threads in the group. Holding cred_guard_mutex is not
1842          * needed because this new task is not yet running and cannot
1843          * be racing exec.
1844          */
1845         assert_spin_locked(&current->sighand->siglock);
1846
1847         /* Ref-count the new filter user, and assign it. */
1848         get_seccomp_filter(current);
1849         p->seccomp = current->seccomp;
1850
1851         /*
1852          * Explicitly enable no_new_privs here in case it got set
1853          * between the task_struct being duplicated and holding the
1854          * sighand lock. The seccomp state and nnp must be in sync.
1855          */
1856         if (task_no_new_privs(current))
1857                 task_set_no_new_privs(p);
1858
1859         /*
1860          * If the parent gained a seccomp mode after copying thread
1861          * flags and between before we held the sighand lock, we have
1862          * to manually enable the seccomp thread flag here.
1863          */
1864         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1865                 set_task_syscall_work(p, SECCOMP);
1866 #endif
1867 }
1868
1869 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1870 {
1871         current->clear_child_tid = tidptr;
1872
1873         return task_pid_vnr(current);
1874 }
1875
1876 static void rt_mutex_init_task(struct task_struct *p)
1877 {
1878         raw_spin_lock_init(&p->pi_lock);
1879 #ifdef CONFIG_RT_MUTEXES
1880         p->pi_waiters = RB_ROOT_CACHED;
1881         p->pi_top_task = NULL;
1882         p->pi_blocked_on = NULL;
1883 #endif
1884 }
1885
1886 static inline void init_task_pid_links(struct task_struct *task)
1887 {
1888         enum pid_type type;
1889
1890         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1891                 INIT_HLIST_NODE(&task->pid_links[type]);
1892 }
1893
1894 static inline void
1895 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1896 {
1897         if (type == PIDTYPE_PID)
1898                 task->thread_pid = pid;
1899         else
1900                 task->signal->pids[type] = pid;
1901 }
1902
1903 static inline void rcu_copy_process(struct task_struct *p)
1904 {
1905 #ifdef CONFIG_PREEMPT_RCU
1906         p->rcu_read_lock_nesting = 0;
1907         p->rcu_read_unlock_special.s = 0;
1908         p->rcu_blocked_node = NULL;
1909         INIT_LIST_HEAD(&p->rcu_node_entry);
1910 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1911 #ifdef CONFIG_TASKS_RCU
1912         p->rcu_tasks_holdout = false;
1913         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1914         p->rcu_tasks_idle_cpu = -1;
1915 #endif /* #ifdef CONFIG_TASKS_RCU */
1916 #ifdef CONFIG_TASKS_TRACE_RCU
1917         p->trc_reader_nesting = 0;
1918         p->trc_reader_special.s = 0;
1919         INIT_LIST_HEAD(&p->trc_holdout_list);
1920         INIT_LIST_HEAD(&p->trc_blkd_node);
1921 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1922 }
1923
1924 struct pid *pidfd_pid(const struct file *file)
1925 {
1926         if (file->f_op == &pidfd_fops)
1927                 return file->private_data;
1928
1929         return ERR_PTR(-EBADF);
1930 }
1931
1932 static int pidfd_release(struct inode *inode, struct file *file)
1933 {
1934         struct pid *pid = file->private_data;
1935
1936         file->private_data = NULL;
1937         put_pid(pid);
1938         return 0;
1939 }
1940
1941 #ifdef CONFIG_PROC_FS
1942 /**
1943  * pidfd_show_fdinfo - print information about a pidfd
1944  * @m: proc fdinfo file
1945  * @f: file referencing a pidfd
1946  *
1947  * Pid:
1948  * This function will print the pid that a given pidfd refers to in the
1949  * pid namespace of the procfs instance.
1950  * If the pid namespace of the process is not a descendant of the pid
1951  * namespace of the procfs instance 0 will be shown as its pid. This is
1952  * similar to calling getppid() on a process whose parent is outside of
1953  * its pid namespace.
1954  *
1955  * NSpid:
1956  * If pid namespaces are supported then this function will also print
1957  * the pid of a given pidfd refers to for all descendant pid namespaces
1958  * starting from the current pid namespace of the instance, i.e. the
1959  * Pid field and the first entry in the NSpid field will be identical.
1960  * If the pid namespace of the process is not a descendant of the pid
1961  * namespace of the procfs instance 0 will be shown as its first NSpid
1962  * entry and no others will be shown.
1963  * Note that this differs from the Pid and NSpid fields in
1964  * /proc/<pid>/status where Pid and NSpid are always shown relative to
1965  * the  pid namespace of the procfs instance. The difference becomes
1966  * obvious when sending around a pidfd between pid namespaces from a
1967  * different branch of the tree, i.e. where no ancestral relation is
1968  * present between the pid namespaces:
1969  * - create two new pid namespaces ns1 and ns2 in the initial pid
1970  *   namespace (also take care to create new mount namespaces in the
1971  *   new pid namespace and mount procfs)
1972  * - create a process with a pidfd in ns1
1973  * - send pidfd from ns1 to ns2
1974  * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1975  *   have exactly one entry, which is 0
1976  */
1977 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1978 {
1979         struct pid *pid = f->private_data;
1980         struct pid_namespace *ns;
1981         pid_t nr = -1;
1982
1983         if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1984                 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1985                 nr = pid_nr_ns(pid, ns);
1986         }
1987
1988         seq_put_decimal_ll(m, "Pid:\t", nr);
1989
1990 #ifdef CONFIG_PID_NS
1991         seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1992         if (nr > 0) {
1993                 int i;
1994
1995                 /* If nr is non-zero it means that 'pid' is valid and that
1996                  * ns, i.e. the pid namespace associated with the procfs
1997                  * instance, is in the pid namespace hierarchy of pid.
1998                  * Start at one below the already printed level.
1999                  */
2000                 for (i = ns->level + 1; i <= pid->level; i++)
2001                         seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
2002         }
2003 #endif
2004         seq_putc(m, '\n');
2005 }
2006 #endif
2007
2008 /*
2009  * Poll support for process exit notification.
2010  */
2011 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
2012 {
2013         struct pid *pid = file->private_data;
2014         __poll_t poll_flags = 0;
2015
2016         poll_wait(file, &pid->wait_pidfd, pts);
2017
2018         /*
2019          * Inform pollers only when the whole thread group exits.
2020          * If the thread group leader exits before all other threads in the
2021          * group, then poll(2) should block, similar to the wait(2) family.
2022          */
2023         if (thread_group_exited(pid))
2024                 poll_flags = EPOLLIN | EPOLLRDNORM;
2025
2026         return poll_flags;
2027 }
2028
2029 const struct file_operations pidfd_fops = {
2030         .release = pidfd_release,
2031         .poll = pidfd_poll,
2032 #ifdef CONFIG_PROC_FS
2033         .show_fdinfo = pidfd_show_fdinfo,
2034 #endif
2035 };
2036
2037 static void __delayed_free_task(struct rcu_head *rhp)
2038 {
2039         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
2040
2041         free_task(tsk);
2042 }
2043
2044 static __always_inline void delayed_free_task(struct task_struct *tsk)
2045 {
2046         if (IS_ENABLED(CONFIG_MEMCG))
2047                 call_rcu(&tsk->rcu, __delayed_free_task);
2048         else
2049                 free_task(tsk);
2050 }
2051
2052 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
2053 {
2054         /* Skip if kernel thread */
2055         if (!tsk->mm)
2056                 return;
2057
2058         /* Skip if spawning a thread or using vfork */
2059         if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
2060                 return;
2061
2062         /* We need to synchronize with __set_oom_adj */
2063         mutex_lock(&oom_adj_mutex);
2064         set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
2065         /* Update the values in case they were changed after copy_signal */
2066         tsk->signal->oom_score_adj = current->signal->oom_score_adj;
2067         tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
2068         mutex_unlock(&oom_adj_mutex);
2069 }
2070
2071 #ifdef CONFIG_RV
2072 static void rv_task_fork(struct task_struct *p)
2073 {
2074         int i;
2075
2076         for (i = 0; i < RV_PER_TASK_MONITORS; i++)
2077                 p->rv[i].da_mon.monitoring = false;
2078 }
2079 #else
2080 #define rv_task_fork(p) do {} while (0)
2081 #endif
2082
2083 /*
2084  * This creates a new process as a copy of the old one,
2085  * but does not actually start it yet.
2086  *
2087  * It copies the registers, and all the appropriate
2088  * parts of the process environment (as per the clone
2089  * flags). The actual kick-off is left to the caller.
2090  */
2091 static __latent_entropy struct task_struct *copy_process(
2092                                         struct pid *pid,
2093                                         int trace,
2094                                         int node,
2095                                         struct kernel_clone_args *args)
2096 {
2097         int pidfd = -1, retval;
2098         struct task_struct *p;
2099         struct multiprocess_signals delayed;
2100         struct file *pidfile = NULL;
2101         const u64 clone_flags = args->flags;
2102         struct nsproxy *nsp = current->nsproxy;
2103
2104         /*
2105          * Don't allow sharing the root directory with processes in a different
2106          * namespace
2107          */
2108         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2109                 return ERR_PTR(-EINVAL);
2110
2111         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2112                 return ERR_PTR(-EINVAL);
2113
2114         /*
2115          * Thread groups must share signals as well, and detached threads
2116          * can only be started up within the thread group.
2117          */
2118         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2119                 return ERR_PTR(-EINVAL);
2120
2121         /*
2122          * Shared signal handlers imply shared VM. By way of the above,
2123          * thread groups also imply shared VM. Blocking this case allows
2124          * for various simplifications in other code.
2125          */
2126         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2127                 return ERR_PTR(-EINVAL);
2128
2129         /*
2130          * Siblings of global init remain as zombies on exit since they are
2131          * not reaped by their parent (swapper). To solve this and to avoid
2132          * multi-rooted process trees, prevent global and container-inits
2133          * from creating siblings.
2134          */
2135         if ((clone_flags & CLONE_PARENT) &&
2136                                 current->signal->flags & SIGNAL_UNKILLABLE)
2137                 return ERR_PTR(-EINVAL);
2138
2139         /*
2140          * If the new process will be in a different pid or user namespace
2141          * do not allow it to share a thread group with the forking task.
2142          */
2143         if (clone_flags & CLONE_THREAD) {
2144                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2145                     (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2146                         return ERR_PTR(-EINVAL);
2147         }
2148
2149         if (clone_flags & CLONE_PIDFD) {
2150                 /*
2151                  * - CLONE_DETACHED is blocked so that we can potentially
2152                  *   reuse it later for CLONE_PIDFD.
2153                  * - CLONE_THREAD is blocked until someone really needs it.
2154                  */
2155                 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
2156                         return ERR_PTR(-EINVAL);
2157         }
2158
2159         /*
2160          * Force any signals received before this point to be delivered
2161          * before the fork happens.  Collect up signals sent to multiple
2162          * processes that happen during the fork and delay them so that
2163          * they appear to happen after the fork.
2164          */
2165         sigemptyset(&delayed.signal);
2166         INIT_HLIST_NODE(&delayed.node);
2167
2168         spin_lock_irq(&current->sighand->siglock);
2169         if (!(clone_flags & CLONE_THREAD))
2170                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
2171         recalc_sigpending();
2172         spin_unlock_irq(&current->sighand->siglock);
2173         retval = -ERESTARTNOINTR;
2174         if (task_sigpending(current))
2175                 goto fork_out;
2176
2177         retval = -ENOMEM;
2178         p = dup_task_struct(current, node);
2179         if (!p)
2180                 goto fork_out;
2181         p->flags &= ~PF_KTHREAD;
2182         if (args->kthread)
2183                 p->flags |= PF_KTHREAD;
2184         if (args->io_thread) {
2185                 /*
2186                  * Mark us an IO worker, and block any signal that isn't
2187                  * fatal or STOP
2188                  */
2189                 p->flags |= PF_IO_WORKER;
2190                 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2191         }
2192
2193         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2194         /*
2195          * Clear TID on mm_release()?
2196          */
2197         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2198
2199         ftrace_graph_init_task(p);
2200
2201         rt_mutex_init_task(p);
2202
2203         lockdep_assert_irqs_enabled();
2204 #ifdef CONFIG_PROVE_LOCKING
2205         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2206 #endif
2207         retval = copy_creds(p, clone_flags);
2208         if (retval < 0)
2209                 goto bad_fork_free;
2210
2211         retval = -EAGAIN;
2212         if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2213                 if (p->real_cred->user != INIT_USER &&
2214                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2215                         goto bad_fork_cleanup_count;
2216         }
2217         current->flags &= ~PF_NPROC_EXCEEDED;
2218
2219         /*
2220          * If multiple threads are within copy_process(), then this check
2221          * triggers too late. This doesn't hurt, the check is only there
2222          * to stop root fork bombs.
2223          */
2224         retval = -EAGAIN;
2225         if (data_race(nr_threads >= max_threads))
2226                 goto bad_fork_cleanup_count;
2227
2228         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
2229         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2230         p->flags |= PF_FORKNOEXEC;
2231         INIT_LIST_HEAD(&p->children);
2232         INIT_LIST_HEAD(&p->sibling);
2233         rcu_copy_process(p);
2234         p->vfork_done = NULL;
2235         spin_lock_init(&p->alloc_lock);
2236
2237         init_sigpending(&p->pending);
2238
2239         p->utime = p->stime = p->gtime = 0;
2240 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2241         p->utimescaled = p->stimescaled = 0;
2242 #endif
2243         prev_cputime_init(&p->prev_cputime);
2244
2245 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2246         seqcount_init(&p->vtime.seqcount);
2247         p->vtime.starttime = 0;
2248         p->vtime.state = VTIME_INACTIVE;
2249 #endif
2250
2251 #ifdef CONFIG_IO_URING
2252         p->io_uring = NULL;
2253 #endif
2254
2255 #if defined(SPLIT_RSS_COUNTING)
2256         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2257 #endif
2258
2259         p->default_timer_slack_ns = current->timer_slack_ns;
2260
2261 #ifdef CONFIG_PSI
2262         p->psi_flags = 0;
2263 #endif
2264
2265         task_io_accounting_init(&p->ioac);
2266         acct_clear_integrals(p);
2267
2268         posix_cputimers_init(&p->posix_cputimers);
2269
2270         p->io_context = NULL;
2271         audit_set_context(p, NULL);
2272         cgroup_fork(p);
2273         if (args->kthread) {
2274                 if (!set_kthread_struct(p))
2275                         goto bad_fork_cleanup_delayacct;
2276         }
2277 #ifdef CONFIG_NUMA
2278         p->mempolicy = mpol_dup(p->mempolicy);
2279         if (IS_ERR(p->mempolicy)) {
2280                 retval = PTR_ERR(p->mempolicy);
2281                 p->mempolicy = NULL;
2282                 goto bad_fork_cleanup_delayacct;
2283         }
2284 #endif
2285 #ifdef CONFIG_CPUSETS
2286         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2287         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2288         seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2289 #endif
2290 #ifdef CONFIG_TRACE_IRQFLAGS
2291         memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2292         p->irqtrace.hardirq_disable_ip  = _THIS_IP_;
2293         p->irqtrace.softirq_enable_ip   = _THIS_IP_;
2294         p->softirqs_enabled             = 1;
2295         p->softirq_context              = 0;
2296 #endif
2297
2298         p->pagefault_disabled = 0;
2299
2300 #ifdef CONFIG_LOCKDEP
2301         lockdep_init_task(p);
2302 #endif
2303
2304 #ifdef CONFIG_DEBUG_MUTEXES
2305         p->blocked_on = NULL; /* not blocked yet */
2306 #endif
2307 #ifdef CONFIG_BCACHE
2308         p->sequential_io        = 0;
2309         p->sequential_io_avg    = 0;
2310 #endif
2311 #ifdef CONFIG_BPF_SYSCALL
2312         RCU_INIT_POINTER(p->bpf_storage, NULL);
2313         p->bpf_ctx = NULL;
2314 #endif
2315
2316         /* Perform scheduler related setup. Assign this task to a CPU. */
2317         retval = sched_fork(clone_flags, p);
2318         if (retval)
2319                 goto bad_fork_cleanup_policy;
2320
2321         retval = perf_event_init_task(p, clone_flags);
2322         if (retval)
2323                 goto bad_fork_cleanup_policy;
2324         retval = audit_alloc(p);
2325         if (retval)
2326                 goto bad_fork_cleanup_perf;
2327         /* copy all the process information */
2328         shm_init_task(p);
2329         retval = security_task_alloc(p, clone_flags);
2330         if (retval)
2331                 goto bad_fork_cleanup_audit;
2332         retval = copy_semundo(clone_flags, p);
2333         if (retval)
2334                 goto bad_fork_cleanup_security;
2335         retval = copy_files(clone_flags, p);
2336         if (retval)
2337                 goto bad_fork_cleanup_semundo;
2338         retval = copy_fs(clone_flags, p);
2339         if (retval)
2340                 goto bad_fork_cleanup_files;
2341         retval = copy_sighand(clone_flags, p);
2342         if (retval)
2343                 goto bad_fork_cleanup_fs;
2344         retval = copy_signal(clone_flags, p);
2345         if (retval)
2346                 goto bad_fork_cleanup_sighand;
2347         retval = copy_mm(clone_flags, p);
2348         if (retval)
2349                 goto bad_fork_cleanup_signal;
2350         retval = copy_namespaces(clone_flags, p);
2351         if (retval)
2352                 goto bad_fork_cleanup_mm;
2353         retval = copy_io(clone_flags, p);
2354         if (retval)
2355                 goto bad_fork_cleanup_namespaces;
2356         retval = copy_thread(p, args);
2357         if (retval)
2358                 goto bad_fork_cleanup_io;
2359
2360         stackleak_task_init(p);
2361
2362         if (pid != &init_struct_pid) {
2363                 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2364                                 args->set_tid_size);
2365                 if (IS_ERR(pid)) {
2366                         retval = PTR_ERR(pid);
2367                         goto bad_fork_cleanup_thread;
2368                 }
2369         }
2370
2371         /*
2372          * This has to happen after we've potentially unshared the file
2373          * descriptor table (so that the pidfd doesn't leak into the child
2374          * if the fd table isn't shared).
2375          */
2376         if (clone_flags & CLONE_PIDFD) {
2377                 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2378                 if (retval < 0)
2379                         goto bad_fork_free_pid;
2380
2381                 pidfd = retval;
2382
2383                 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2384                                               O_RDWR | O_CLOEXEC);
2385                 if (IS_ERR(pidfile)) {
2386                         put_unused_fd(pidfd);
2387                         retval = PTR_ERR(pidfile);
2388                         goto bad_fork_free_pid;
2389                 }
2390                 get_pid(pid);   /* held by pidfile now */
2391
2392                 retval = put_user(pidfd, args->pidfd);
2393                 if (retval)
2394                         goto bad_fork_put_pidfd;
2395         }
2396
2397 #ifdef CONFIG_BLOCK
2398         p->plug = NULL;
2399 #endif
2400         futex_init_task(p);
2401
2402         /*
2403          * sigaltstack should be cleared when sharing the same VM
2404          */
2405         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2406                 sas_ss_reset(p);
2407
2408         /*
2409          * Syscall tracing and stepping should be turned off in the
2410          * child regardless of CLONE_PTRACE.
2411          */
2412         user_disable_single_step(p);
2413         clear_task_syscall_work(p, SYSCALL_TRACE);
2414 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2415         clear_task_syscall_work(p, SYSCALL_EMU);
2416 #endif
2417         clear_tsk_latency_tracing(p);
2418
2419         /* ok, now we should be set up.. */
2420         p->pid = pid_nr(pid);
2421         if (clone_flags & CLONE_THREAD) {
2422                 p->group_leader = current->group_leader;
2423                 p->tgid = current->tgid;
2424         } else {
2425                 p->group_leader = p;
2426                 p->tgid = p->pid;
2427         }
2428
2429         p->nr_dirtied = 0;
2430         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2431         p->dirty_paused_when = 0;
2432
2433         p->pdeath_signal = 0;
2434         INIT_LIST_HEAD(&p->thread_group);
2435         p->task_works = NULL;
2436         clear_posix_cputimers_work(p);
2437
2438 #ifdef CONFIG_KRETPROBES
2439         p->kretprobe_instances.first = NULL;
2440 #endif
2441 #ifdef CONFIG_RETHOOK
2442         p->rethooks.first = NULL;
2443 #endif
2444
2445         /*
2446          * Ensure that the cgroup subsystem policies allow the new process to be
2447          * forked. It should be noted that the new process's css_set can be changed
2448          * between here and cgroup_post_fork() if an organisation operation is in
2449          * progress.
2450          */
2451         retval = cgroup_can_fork(p, args);
2452         if (retval)
2453                 goto bad_fork_put_pidfd;
2454
2455         /*
2456          * Now that the cgroups are pinned, re-clone the parent cgroup and put
2457          * the new task on the correct runqueue. All this *before* the task
2458          * becomes visible.
2459          *
2460          * This isn't part of ->can_fork() because while the re-cloning is
2461          * cgroup specific, it unconditionally needs to place the task on a
2462          * runqueue.
2463          */
2464         sched_cgroup_fork(p, args);
2465
2466         /*
2467          * From this point on we must avoid any synchronous user-space
2468          * communication until we take the tasklist-lock. In particular, we do
2469          * not want user-space to be able to predict the process start-time by
2470          * stalling fork(2) after we recorded the start_time but before it is
2471          * visible to the system.
2472          */
2473
2474         p->start_time = ktime_get_ns();
2475         p->start_boottime = ktime_get_boottime_ns();
2476
2477         /*
2478          * Make it visible to the rest of the system, but dont wake it up yet.
2479          * Need tasklist lock for parent etc handling!
2480          */
2481         write_lock_irq(&tasklist_lock);
2482
2483         /* CLONE_PARENT re-uses the old parent */
2484         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2485                 p->real_parent = current->real_parent;
2486                 p->parent_exec_id = current->parent_exec_id;
2487                 if (clone_flags & CLONE_THREAD)
2488                         p->exit_signal = -1;
2489                 else
2490                         p->exit_signal = current->group_leader->exit_signal;
2491         } else {
2492                 p->real_parent = current;
2493                 p->parent_exec_id = current->self_exec_id;
2494                 p->exit_signal = args->exit_signal;
2495         }
2496
2497         klp_copy_process(p);
2498
2499         sched_core_fork(p);
2500
2501         spin_lock(&current->sighand->siglock);
2502
2503         rv_task_fork(p);
2504
2505         rseq_fork(p, clone_flags);
2506
2507         /* Don't start children in a dying pid namespace */
2508         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2509                 retval = -ENOMEM;
2510                 goto bad_fork_cancel_cgroup;
2511         }
2512
2513         /* Let kill terminate clone/fork in the middle */
2514         if (fatal_signal_pending(current)) {
2515                 retval = -EINTR;
2516                 goto bad_fork_cancel_cgroup;
2517         }
2518
2519         /* No more failure paths after this point. */
2520
2521         /*
2522          * Copy seccomp details explicitly here, in case they were changed
2523          * before holding sighand lock.
2524          */
2525         copy_seccomp(p);
2526
2527         init_task_pid_links(p);
2528         if (likely(p->pid)) {
2529                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2530
2531                 init_task_pid(p, PIDTYPE_PID, pid);
2532                 if (thread_group_leader(p)) {
2533                         init_task_pid(p, PIDTYPE_TGID, pid);
2534                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2535                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2536
2537                         if (is_child_reaper(pid)) {
2538                                 ns_of_pid(pid)->child_reaper = p;
2539                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2540                         }
2541                         p->signal->shared_pending.signal = delayed.signal;
2542                         p->signal->tty = tty_kref_get(current->signal->tty);
2543                         /*
2544                          * Inherit has_child_subreaper flag under the same
2545                          * tasklist_lock with adding child to the process tree
2546                          * for propagate_has_child_subreaper optimization.
2547                          */
2548                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2549                                                          p->real_parent->signal->is_child_subreaper;
2550                         list_add_tail(&p->sibling, &p->real_parent->children);
2551                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2552                         attach_pid(p, PIDTYPE_TGID);
2553                         attach_pid(p, PIDTYPE_PGID);
2554                         attach_pid(p, PIDTYPE_SID);
2555                         __this_cpu_inc(process_counts);
2556                 } else {
2557                         current->signal->nr_threads++;
2558                         current->signal->quick_threads++;
2559                         atomic_inc(&current->signal->live);
2560                         refcount_inc(&current->signal->sigcnt);
2561                         task_join_group_stop(p);
2562                         list_add_tail_rcu(&p->thread_group,
2563                                           &p->group_leader->thread_group);
2564                         list_add_tail_rcu(&p->thread_node,
2565                                           &p->signal->thread_head);
2566                 }
2567                 attach_pid(p, PIDTYPE_PID);
2568                 nr_threads++;
2569         }
2570         total_forks++;
2571         hlist_del_init(&delayed.node);
2572         spin_unlock(&current->sighand->siglock);
2573         syscall_tracepoint_update(p);
2574         write_unlock_irq(&tasklist_lock);
2575
2576         if (pidfile)
2577                 fd_install(pidfd, pidfile);
2578
2579         proc_fork_connector(p);
2580         sched_post_fork(p);
2581         cgroup_post_fork(p, args);
2582         perf_event_fork(p);
2583
2584         trace_task_newtask(p, clone_flags);
2585         uprobe_copy_process(p, clone_flags);
2586
2587         copy_oom_score_adj(clone_flags, p);
2588
2589         return p;
2590
2591 bad_fork_cancel_cgroup:
2592         sched_core_free(p);
2593         spin_unlock(&current->sighand->siglock);
2594         write_unlock_irq(&tasklist_lock);
2595         cgroup_cancel_fork(p, args);
2596 bad_fork_put_pidfd:
2597         if (clone_flags & CLONE_PIDFD) {
2598                 fput(pidfile);
2599                 put_unused_fd(pidfd);
2600         }
2601 bad_fork_free_pid:
2602         if (pid != &init_struct_pid)
2603                 free_pid(pid);
2604 bad_fork_cleanup_thread:
2605         exit_thread(p);
2606 bad_fork_cleanup_io:
2607         if (p->io_context)
2608                 exit_io_context(p);
2609 bad_fork_cleanup_namespaces:
2610         exit_task_namespaces(p);
2611 bad_fork_cleanup_mm:
2612         if (p->mm) {
2613                 mm_clear_owner(p->mm, p);
2614                 mmput(p->mm);
2615         }
2616 bad_fork_cleanup_signal:
2617         if (!(clone_flags & CLONE_THREAD))
2618                 free_signal_struct(p->signal);
2619 bad_fork_cleanup_sighand:
2620         __cleanup_sighand(p->sighand);
2621 bad_fork_cleanup_fs:
2622         exit_fs(p); /* blocking */
2623 bad_fork_cleanup_files:
2624         exit_files(p); /* blocking */
2625 bad_fork_cleanup_semundo:
2626         exit_sem(p);
2627 bad_fork_cleanup_security:
2628         security_task_free(p);
2629 bad_fork_cleanup_audit:
2630         audit_free(p);
2631 bad_fork_cleanup_perf:
2632         perf_event_free_task(p);
2633 bad_fork_cleanup_policy:
2634         lockdep_free_task(p);
2635 #ifdef CONFIG_NUMA
2636         mpol_put(p->mempolicy);
2637 #endif
2638 bad_fork_cleanup_delayacct:
2639         delayacct_tsk_free(p);
2640 bad_fork_cleanup_count:
2641         dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2642         exit_creds(p);
2643 bad_fork_free:
2644         WRITE_ONCE(p->__state, TASK_DEAD);
2645         exit_task_stack_account(p);
2646         put_task_stack(p);
2647         delayed_free_task(p);
2648 fork_out:
2649         spin_lock_irq(&current->sighand->siglock);
2650         hlist_del_init(&delayed.node);
2651         spin_unlock_irq(&current->sighand->siglock);
2652         return ERR_PTR(retval);
2653 }
2654
2655 static inline void init_idle_pids(struct task_struct *idle)
2656 {
2657         enum pid_type type;
2658
2659         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2660                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2661                 init_task_pid(idle, type, &init_struct_pid);
2662         }
2663 }
2664
2665 static int idle_dummy(void *dummy)
2666 {
2667         /* This function is never called */
2668         return 0;
2669 }
2670
2671 struct task_struct * __init fork_idle(int cpu)
2672 {
2673         struct task_struct *task;
2674         struct kernel_clone_args args = {
2675                 .flags          = CLONE_VM,
2676                 .fn             = &idle_dummy,
2677                 .fn_arg         = NULL,
2678                 .kthread        = 1,
2679                 .idle           = 1,
2680         };
2681
2682         task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2683         if (!IS_ERR(task)) {
2684                 init_idle_pids(task);
2685                 init_idle(task, cpu);
2686         }
2687
2688         return task;
2689 }
2690
2691 /*
2692  * This is like kernel_clone(), but shaved down and tailored to just
2693  * creating io_uring workers. It returns a created task, or an error pointer.
2694  * The returned task is inactive, and the caller must fire it up through
2695  * wake_up_new_task(p). All signals are blocked in the created task.
2696  */
2697 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2698 {
2699         unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2700                                 CLONE_IO;
2701         struct kernel_clone_args args = {
2702                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2703                                     CLONE_UNTRACED) & ~CSIGNAL),
2704                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2705                 .fn             = fn,
2706                 .fn_arg         = arg,
2707                 .io_thread      = 1,
2708         };
2709
2710         return copy_process(NULL, 0, node, &args);
2711 }
2712
2713 /*
2714  *  Ok, this is the main fork-routine.
2715  *
2716  * It copies the process, and if successful kick-starts
2717  * it and waits for it to finish using the VM if required.
2718  *
2719  * args->exit_signal is expected to be checked for sanity by the caller.
2720  */
2721 pid_t kernel_clone(struct kernel_clone_args *args)
2722 {
2723         u64 clone_flags = args->flags;
2724         struct completion vfork;
2725         struct pid *pid;
2726         struct task_struct *p;
2727         int trace = 0;
2728         pid_t nr;
2729
2730         /*
2731          * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2732          * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2733          * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2734          * field in struct clone_args and it still doesn't make sense to have
2735          * them both point at the same memory location. Performing this check
2736          * here has the advantage that we don't need to have a separate helper
2737          * to check for legacy clone().
2738          */
2739         if ((args->flags & CLONE_PIDFD) &&
2740             (args->flags & CLONE_PARENT_SETTID) &&
2741             (args->pidfd == args->parent_tid))
2742                 return -EINVAL;
2743
2744         /*
2745          * Determine whether and which event to report to ptracer.  When
2746          * called from kernel_thread or CLONE_UNTRACED is explicitly
2747          * requested, no event is reported; otherwise, report if the event
2748          * for the type of forking is enabled.
2749          */
2750         if (!(clone_flags & CLONE_UNTRACED)) {
2751                 if (clone_flags & CLONE_VFORK)
2752                         trace = PTRACE_EVENT_VFORK;
2753                 else if (args->exit_signal != SIGCHLD)
2754                         trace = PTRACE_EVENT_CLONE;
2755                 else
2756                         trace = PTRACE_EVENT_FORK;
2757
2758                 if (likely(!ptrace_event_enabled(current, trace)))
2759                         trace = 0;
2760         }
2761
2762         p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2763         add_latent_entropy();
2764
2765         if (IS_ERR(p))
2766                 return PTR_ERR(p);
2767
2768         /*
2769          * Do this prior waking up the new thread - the thread pointer
2770          * might get invalid after that point, if the thread exits quickly.
2771          */
2772         trace_sched_process_fork(current, p);
2773
2774         pid = get_task_pid(p, PIDTYPE_PID);
2775         nr = pid_vnr(pid);
2776
2777         if (clone_flags & CLONE_PARENT_SETTID)
2778                 put_user(nr, args->parent_tid);
2779
2780         if (clone_flags & CLONE_VFORK) {
2781                 p->vfork_done = &vfork;
2782                 init_completion(&vfork);
2783                 get_task_struct(p);
2784         }
2785
2786         if (IS_ENABLED(CONFIG_LRU_GEN) && !(clone_flags & CLONE_VM)) {
2787                 /* lock the task to synchronize with memcg migration */
2788                 task_lock(p);
2789                 lru_gen_add_mm(p->mm);
2790                 task_unlock(p);
2791         }
2792
2793         wake_up_new_task(p);
2794
2795         /* forking complete and child started to run, tell ptracer */
2796         if (unlikely(trace))
2797                 ptrace_event_pid(trace, pid);
2798
2799         if (clone_flags & CLONE_VFORK) {
2800                 if (!wait_for_vfork_done(p, &vfork))
2801                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2802         }
2803
2804         put_pid(pid);
2805         return nr;
2806 }
2807
2808 /*
2809  * Create a kernel thread.
2810  */
2811 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2812 {
2813         struct kernel_clone_args args = {
2814                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2815                                     CLONE_UNTRACED) & ~CSIGNAL),
2816                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2817                 .fn             = fn,
2818                 .fn_arg         = arg,
2819                 .kthread        = 1,
2820         };
2821
2822         return kernel_clone(&args);
2823 }
2824
2825 /*
2826  * Create a user mode thread.
2827  */
2828 pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2829 {
2830         struct kernel_clone_args args = {
2831                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2832                                     CLONE_UNTRACED) & ~CSIGNAL),
2833                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2834                 .fn             = fn,
2835                 .fn_arg         = arg,
2836         };
2837
2838         return kernel_clone(&args);
2839 }
2840
2841 #ifdef __ARCH_WANT_SYS_FORK
2842 SYSCALL_DEFINE0(fork)
2843 {
2844 #ifdef CONFIG_MMU
2845         struct kernel_clone_args args = {
2846                 .exit_signal = SIGCHLD,
2847         };
2848
2849         return kernel_clone(&args);
2850 #else
2851         /* can not support in nommu mode */
2852         return -EINVAL;
2853 #endif
2854 }
2855 #endif
2856
2857 #ifdef __ARCH_WANT_SYS_VFORK
2858 SYSCALL_DEFINE0(vfork)
2859 {
2860         struct kernel_clone_args args = {
2861                 .flags          = CLONE_VFORK | CLONE_VM,
2862                 .exit_signal    = SIGCHLD,
2863         };
2864
2865         return kernel_clone(&args);
2866 }
2867 #endif
2868
2869 #ifdef __ARCH_WANT_SYS_CLONE
2870 #ifdef CONFIG_CLONE_BACKWARDS
2871 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2872                  int __user *, parent_tidptr,
2873                  unsigned long, tls,
2874                  int __user *, child_tidptr)
2875 #elif defined(CONFIG_CLONE_BACKWARDS2)
2876 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2877                  int __user *, parent_tidptr,
2878                  int __user *, child_tidptr,
2879                  unsigned long, tls)
2880 #elif defined(CONFIG_CLONE_BACKWARDS3)
2881 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2882                 int, stack_size,
2883                 int __user *, parent_tidptr,
2884                 int __user *, child_tidptr,
2885                 unsigned long, tls)
2886 #else
2887 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2888                  int __user *, parent_tidptr,
2889                  int __user *, child_tidptr,
2890                  unsigned long, tls)
2891 #endif
2892 {
2893         struct kernel_clone_args args = {
2894                 .flags          = (lower_32_bits(clone_flags) & ~CSIGNAL),
2895                 .pidfd          = parent_tidptr,
2896                 .child_tid      = child_tidptr,
2897                 .parent_tid     = parent_tidptr,
2898                 .exit_signal    = (lower_32_bits(clone_flags) & CSIGNAL),
2899                 .stack          = newsp,
2900                 .tls            = tls,
2901         };
2902
2903         return kernel_clone(&args);
2904 }
2905 #endif
2906
2907 #ifdef __ARCH_WANT_SYS_CLONE3
2908
2909 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2910                                               struct clone_args __user *uargs,
2911                                               size_t usize)
2912 {
2913         int err;
2914         struct clone_args args;
2915         pid_t *kset_tid = kargs->set_tid;
2916
2917         BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2918                      CLONE_ARGS_SIZE_VER0);
2919         BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2920                      CLONE_ARGS_SIZE_VER1);
2921         BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2922                      CLONE_ARGS_SIZE_VER2);
2923         BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2924
2925         if (unlikely(usize > PAGE_SIZE))
2926                 return -E2BIG;
2927         if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2928                 return -EINVAL;
2929
2930         err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2931         if (err)
2932                 return err;
2933
2934         if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2935                 return -EINVAL;
2936
2937         if (unlikely(!args.set_tid && args.set_tid_size > 0))
2938                 return -EINVAL;
2939
2940         if (unlikely(args.set_tid && args.set_tid_size == 0))
2941                 return -EINVAL;
2942
2943         /*
2944          * Verify that higher 32bits of exit_signal are unset and that
2945          * it is a valid signal
2946          */
2947         if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2948                      !valid_signal(args.exit_signal)))
2949                 return -EINVAL;
2950
2951         if ((args.flags & CLONE_INTO_CGROUP) &&
2952             (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2953                 return -EINVAL;
2954
2955         *kargs = (struct kernel_clone_args){
2956                 .flags          = args.flags,
2957                 .pidfd          = u64_to_user_ptr(args.pidfd),
2958                 .child_tid      = u64_to_user_ptr(args.child_tid),
2959                 .parent_tid     = u64_to_user_ptr(args.parent_tid),
2960                 .exit_signal    = args.exit_signal,
2961                 .stack          = args.stack,
2962                 .stack_size     = args.stack_size,
2963                 .tls            = args.tls,
2964                 .set_tid_size   = args.set_tid_size,
2965                 .cgroup         = args.cgroup,
2966         };
2967
2968         if (args.set_tid &&
2969                 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2970                         (kargs->set_tid_size * sizeof(pid_t))))
2971                 return -EFAULT;
2972
2973         kargs->set_tid = kset_tid;
2974
2975         return 0;
2976 }
2977
2978 /**
2979  * clone3_stack_valid - check and prepare stack
2980  * @kargs: kernel clone args
2981  *
2982  * Verify that the stack arguments userspace gave us are sane.
2983  * In addition, set the stack direction for userspace since it's easy for us to
2984  * determine.
2985  */
2986 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2987 {
2988         if (kargs->stack == 0) {
2989                 if (kargs->stack_size > 0)
2990                         return false;
2991         } else {
2992                 if (kargs->stack_size == 0)
2993                         return false;
2994
2995                 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2996                         return false;
2997
2998 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2999                 kargs->stack += kargs->stack_size;
3000 #endif
3001         }
3002
3003         return true;
3004 }
3005
3006 static bool clone3_args_valid(struct kernel_clone_args *kargs)
3007 {
3008         /* Verify that no unknown flags are passed along. */
3009         if (kargs->flags &
3010             ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
3011                 return false;
3012
3013         /*
3014          * - make the CLONE_DETACHED bit reusable for clone3
3015          * - make the CSIGNAL bits reusable for clone3
3016          */
3017         if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
3018                 return false;
3019
3020         if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
3021             (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
3022                 return false;
3023
3024         if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
3025             kargs->exit_signal)
3026                 return false;
3027
3028         if (!clone3_stack_valid(kargs))
3029                 return false;
3030
3031         return true;
3032 }
3033
3034 /**
3035  * clone3 - create a new process with specific properties
3036  * @uargs: argument structure
3037  * @size:  size of @uargs
3038  *
3039  * clone3() is the extensible successor to clone()/clone2().
3040  * It takes a struct as argument that is versioned by its size.
3041  *
3042  * Return: On success, a positive PID for the child process.
3043  *         On error, a negative errno number.
3044  */
3045 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
3046 {
3047         int err;
3048
3049         struct kernel_clone_args kargs;
3050         pid_t set_tid[MAX_PID_NS_LEVEL];
3051
3052         kargs.set_tid = set_tid;
3053
3054         err = copy_clone_args_from_user(&kargs, uargs, size);
3055         if (err)
3056                 return err;
3057
3058         if (!clone3_args_valid(&kargs))
3059                 return -EINVAL;
3060
3061         return kernel_clone(&kargs);
3062 }
3063 #endif
3064
3065 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
3066 {
3067         struct task_struct *leader, *parent, *child;
3068         int res;
3069
3070         read_lock(&tasklist_lock);
3071         leader = top = top->group_leader;
3072 down:
3073         for_each_thread(leader, parent) {
3074                 list_for_each_entry(child, &parent->children, sibling) {
3075                         res = visitor(child, data);
3076                         if (res) {
3077                                 if (res < 0)
3078                                         goto out;
3079                                 leader = child;
3080                                 goto down;
3081                         }
3082 up:
3083                         ;
3084                 }
3085         }
3086
3087         if (leader != top) {
3088                 child = leader;
3089                 parent = child->real_parent;
3090                 leader = parent->group_leader;
3091                 goto up;
3092         }
3093 out:
3094         read_unlock(&tasklist_lock);
3095 }
3096
3097 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
3098 #define ARCH_MIN_MMSTRUCT_ALIGN 0
3099 #endif
3100
3101 static void sighand_ctor(void *data)
3102 {
3103         struct sighand_struct *sighand = data;
3104
3105         spin_lock_init(&sighand->siglock);
3106         init_waitqueue_head(&sighand->signalfd_wqh);
3107 }
3108
3109 void __init mm_cache_init(void)
3110 {
3111         unsigned int mm_size;
3112
3113         /*
3114          * The mm_cpumask is located at the end of mm_struct, and is
3115          * dynamically sized based on the maximum CPU number this system
3116          * can have, taking hotplug into account (nr_cpu_ids).
3117          */
3118         mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
3119
3120         mm_cachep = kmem_cache_create_usercopy("mm_struct",
3121                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3122                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3123                         offsetof(struct mm_struct, saved_auxv),
3124                         sizeof_field(struct mm_struct, saved_auxv),
3125                         NULL);
3126 }
3127
3128 void __init proc_caches_init(void)
3129 {
3130         sighand_cachep = kmem_cache_create("sighand_cache",
3131                         sizeof(struct sighand_struct), 0,
3132                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3133                         SLAB_ACCOUNT, sighand_ctor);
3134         signal_cachep = kmem_cache_create("signal_cache",
3135                         sizeof(struct signal_struct), 0,
3136                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3137                         NULL);
3138         files_cachep = kmem_cache_create("files_cache",
3139                         sizeof(struct files_struct), 0,
3140                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3141                         NULL);
3142         fs_cachep = kmem_cache_create("fs_cache",
3143                         sizeof(struct fs_struct), 0,
3144                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3145                         NULL);
3146
3147         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3148         mmap_init();
3149         nsproxy_cache_init();
3150 }
3151
3152 /*
3153  * Check constraints on flags passed to the unshare system call.
3154  */
3155 static int check_unshare_flags(unsigned long unshare_flags)
3156 {
3157         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3158                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3159                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3160                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3161                                 CLONE_NEWTIME))
3162                 return -EINVAL;
3163         /*
3164          * Not implemented, but pretend it works if there is nothing
3165          * to unshare.  Note that unsharing the address space or the
3166          * signal handlers also need to unshare the signal queues (aka
3167          * CLONE_THREAD).
3168          */
3169         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3170                 if (!thread_group_empty(current))
3171                         return -EINVAL;
3172         }
3173         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3174                 if (refcount_read(&current->sighand->count) > 1)
3175                         return -EINVAL;
3176         }
3177         if (unshare_flags & CLONE_VM) {
3178                 if (!current_is_single_threaded())
3179                         return -EINVAL;
3180         }
3181
3182         return 0;
3183 }
3184
3185 /*
3186  * Unshare the filesystem structure if it is being shared
3187  */
3188 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3189 {
3190         struct fs_struct *fs = current->fs;
3191
3192         if (!(unshare_flags & CLONE_FS) || !fs)
3193                 return 0;
3194
3195         /* don't need lock here; in the worst case we'll do useless copy */
3196         if (fs->users == 1)
3197                 return 0;
3198
3199         *new_fsp = copy_fs_struct(fs);
3200         if (!*new_fsp)
3201                 return -ENOMEM;
3202
3203         return 0;
3204 }
3205
3206 /*
3207  * Unshare file descriptor table if it is being shared
3208  */
3209 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3210                struct files_struct **new_fdp)
3211 {
3212         struct files_struct *fd = current->files;
3213         int error = 0;
3214
3215         if ((unshare_flags & CLONE_FILES) &&
3216             (fd && atomic_read(&fd->count) > 1)) {
3217                 *new_fdp = dup_fd(fd, max_fds, &error);
3218                 if (!*new_fdp)
3219                         return error;
3220         }
3221
3222         return 0;
3223 }
3224
3225 /*
3226  * unshare allows a process to 'unshare' part of the process
3227  * context which was originally shared using clone.  copy_*
3228  * functions used by kernel_clone() cannot be used here directly
3229  * because they modify an inactive task_struct that is being
3230  * constructed. Here we are modifying the current, active,
3231  * task_struct.
3232  */
3233 int ksys_unshare(unsigned long unshare_flags)
3234 {
3235         struct fs_struct *fs, *new_fs = NULL;
3236         struct files_struct *new_fd = NULL;
3237         struct cred *new_cred = NULL;
3238         struct nsproxy *new_nsproxy = NULL;
3239         int do_sysvsem = 0;
3240         int err;
3241
3242         /*
3243          * If unsharing a user namespace must also unshare the thread group
3244          * and unshare the filesystem root and working directories.
3245          */
3246         if (unshare_flags & CLONE_NEWUSER)
3247                 unshare_flags |= CLONE_THREAD | CLONE_FS;
3248         /*
3249          * If unsharing vm, must also unshare signal handlers.
3250          */
3251         if (unshare_flags & CLONE_VM)
3252                 unshare_flags |= CLONE_SIGHAND;
3253         /*
3254          * If unsharing a signal handlers, must also unshare the signal queues.
3255          */
3256         if (unshare_flags & CLONE_SIGHAND)
3257                 unshare_flags |= CLONE_THREAD;
3258         /*
3259          * If unsharing namespace, must also unshare filesystem information.
3260          */
3261         if (unshare_flags & CLONE_NEWNS)
3262                 unshare_flags |= CLONE_FS;
3263
3264         err = check_unshare_flags(unshare_flags);
3265         if (err)
3266                 goto bad_unshare_out;
3267         /*
3268          * CLONE_NEWIPC must also detach from the undolist: after switching
3269          * to a new ipc namespace, the semaphore arrays from the old
3270          * namespace are unreachable.
3271          */
3272         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3273                 do_sysvsem = 1;
3274         err = unshare_fs(unshare_flags, &new_fs);
3275         if (err)
3276                 goto bad_unshare_out;
3277         err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3278         if (err)
3279                 goto bad_unshare_cleanup_fs;
3280         err = unshare_userns(unshare_flags, &new_cred);
3281         if (err)
3282                 goto bad_unshare_cleanup_fd;
3283         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3284                                          new_cred, new_fs);
3285         if (err)
3286                 goto bad_unshare_cleanup_cred;
3287
3288         if (new_cred) {
3289                 err = set_cred_ucounts(new_cred);
3290                 if (err)
3291                         goto bad_unshare_cleanup_cred;
3292         }
3293
3294         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3295                 if (do_sysvsem) {
3296                         /*
3297                          * CLONE_SYSVSEM is equivalent to sys_exit().
3298                          */
3299                         exit_sem(current);
3300                 }
3301                 if (unshare_flags & CLONE_NEWIPC) {
3302                         /* Orphan segments in old ns (see sem above). */
3303                         exit_shm(current);
3304                         shm_init_task(current);
3305                 }
3306
3307                 if (new_nsproxy)
3308                         switch_task_namespaces(current, new_nsproxy);
3309
3310                 task_lock(current);
3311
3312                 if (new_fs) {
3313                         fs = current->fs;
3314                         spin_lock(&fs->lock);
3315                         current->fs = new_fs;
3316                         if (--fs->users)
3317                                 new_fs = NULL;
3318                         else
3319                                 new_fs = fs;
3320                         spin_unlock(&fs->lock);
3321                 }
3322
3323                 if (new_fd)
3324                         swap(current->files, new_fd);
3325
3326                 task_unlock(current);
3327
3328                 if (new_cred) {
3329                         /* Install the new user namespace */
3330                         commit_creds(new_cred);
3331                         new_cred = NULL;
3332                 }
3333         }
3334
3335         perf_event_namespaces(current);
3336
3337 bad_unshare_cleanup_cred:
3338         if (new_cred)
3339                 put_cred(new_cred);
3340 bad_unshare_cleanup_fd:
3341         if (new_fd)
3342                 put_files_struct(new_fd);
3343
3344 bad_unshare_cleanup_fs:
3345         if (new_fs)
3346                 free_fs_struct(new_fs);
3347
3348 bad_unshare_out:
3349         return err;
3350 }
3351
3352 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3353 {
3354         return ksys_unshare(unshare_flags);
3355 }
3356
3357 /*
3358  *      Helper to unshare the files of the current task.
3359  *      We don't want to expose copy_files internals to
3360  *      the exec layer of the kernel.
3361  */
3362
3363 int unshare_files(void)
3364 {
3365         struct task_struct *task = current;
3366         struct files_struct *old, *copy = NULL;
3367         int error;
3368
3369         error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, &copy);
3370         if (error || !copy)
3371                 return error;
3372
3373         old = task->files;
3374         task_lock(task);
3375         task->files = copy;
3376         task_unlock(task);
3377         put_files_struct(old);
3378         return 0;
3379 }
3380
3381 int sysctl_max_threads(struct ctl_table *table, int write,
3382                        void *buffer, size_t *lenp, loff_t *ppos)
3383 {
3384         struct ctl_table t;
3385         int ret;
3386         int threads = max_threads;
3387         int min = 1;
3388         int max = MAX_THREADS;
3389
3390         t = *table;
3391         t.data = &threads;
3392         t.extra1 = &min;
3393         t.extra2 = &max;
3394
3395         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3396         if (ret || !write)
3397                 return ret;
3398
3399         max_threads = threads;
3400
3401         return 0;
3402 }