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                 vma_init_lock(new);
 478                 dup_anon_vma_name(orig, new);
 479         }
 480         return new;
 481 }
 482
 483 void __vm_area_free(struct vm_area_struct *vma)
 484 {
 485         free_anon_vma_name(vma);
 486         kmem_cache_free(vm_area_cachep, vma);
 487 }
 488
 489 #ifdef CONFIG_PER_VMA_LOCK
 490 static void vm_area_free_rcu_cb(struct rcu_head *head)
 491 {
 492         struct vm_area_struct *vma = container_of(head, struct vm_area_struct,
 493                                                   vm_rcu);
 494
 495         /* The vma should not be locked while being destroyed. */
 496         VM_BUG_ON_VMA(rwsem_is_locked(&vma->lock), vma);
 497         __vm_area_free(vma);
 498 }
 499 #endif
 500
 501 void vm_area_free(struct vm_area_struct *vma)
 502 {
 503 #ifdef CONFIG_PER_VMA_LOCK
 504         call_rcu(&vma->vm_rcu, vm_area_free_rcu_cb);
 505 #else
 506         __vm_area_free(vma);
 507 #endif
 508 }
 509
 510 static void account_kernel_stack(struct task_struct *tsk, int account)
 511 {
 512         if (IS_ENABLED(CONFIG_VMAP_STACK)) {
 513                 struct vm_struct *vm = task_stack_vm_area(tsk);
 514                 int i;
 515
 516                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 517                         mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
 518                                               account * (PAGE_SIZE / 1024));
 519         } else {
 520                 void *stack = task_stack_page(tsk);
 521
 522                 /* All stack pages are in the same node. */
 523                 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
 524                                       account * (THREAD_SIZE / 1024));
 525         }
 526 }
 527
 528 void exit_task_stack_account(struct task_struct *tsk)
 529 {
 530         account_kernel_stack(tsk, -1);
 531
 532         if (IS_ENABLED(CONFIG_VMAP_STACK)) {
 533                 struct vm_struct *vm;
 534                 int i;
 535
 536                 vm = task_stack_vm_area(tsk);
 537                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 538                         memcg_kmem_uncharge_page(vm->pages[i], 0);
 539         }
 540 }
 541
 542 static void release_task_stack(struct task_struct *tsk)
 543 {
 544         if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
 545                 return;  /* Better to leak the stack than to free prematurely */
 546
 547         free_thread_stack(tsk);
 548 }
 549
 550 #ifdef CONFIG_THREAD_INFO_IN_TASK
 551 void put_task_stack(struct task_struct *tsk)
 552 {
 553         if (refcount_dec_and_test(&tsk->stack_refcount))
 554                 release_task_stack(tsk);
 555 }
 556 #endif
 557
 558 void free_task(struct task_struct *tsk)
 559 {
 560 #ifdef CONFIG_SECCOMP
 561         WARN_ON_ONCE(tsk->seccomp.filter);
 562 #endif
 563         release_user_cpus_ptr(tsk);
 564         scs_release(tsk);
 565
 566 #ifndef CONFIG_THREAD_INFO_IN_TASK
 567         /*
 568          * The task is finally done with both the stack and thread_info,
 569          * so free both.
 570          */
 571         release_task_stack(tsk);
 572 #else
 573         /*
 574          * If the task had a separate stack allocation, it should be gone
 575          * by now.
 576          */
 577         WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
 578 #endif
 579         rt_mutex_debug_task_free(tsk);
 580         ftrace_graph_exit_task(tsk);
 581         arch_release_task_struct(tsk);
 582         if (tsk->flags & PF_KTHREAD)
 583                 free_kthread_struct(tsk);
 584         free_task_struct(tsk);
 585 }
 586 EXPORT_SYMBOL(free_task);
 587
 588 static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
 589 {
 590         struct file *exe_file;
 591
 592         exe_file = get_mm_exe_file(oldmm);
 593         RCU_INIT_POINTER(mm->exe_file, exe_file);
 594         /*
 595          * We depend on the oldmm having properly denied write access to the
 596          * exe_file already.
 597          */
 598         if (exe_file && deny_write_access(exe_file))
 599                 pr_warn_once("deny_write_access() failed in %s\n", __func__);
 600 }
 601
 602 #ifdef CONFIG_MMU
 603 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 604                                         struct mm_struct *oldmm)
 605 {
 606         struct vm_area_struct *mpnt, *tmp;
 607         int retval;
 608         unsigned long charge = 0;
 609         LIST_HEAD(uf);
 610         VMA_ITERATOR(old_vmi, oldmm, 0);
 611         VMA_ITERATOR(vmi, mm, 0);
 612
 613         uprobe_start_dup_mmap();
 614         if (mmap_write_lock_killable(oldmm)) {
 615                 retval = -EINTR;
 616                 goto fail_uprobe_end;
 617         }
 618         flush_cache_dup_mm(oldmm);
 619         uprobe_dup_mmap(oldmm, mm);
 620         /*
 621          * Not linked in yet - no deadlock potential:
 622          */
 623         mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
 624
 625         /* No ordering required: file already has been exposed. */
 626         dup_mm_exe_file(mm, oldmm);
 627
 628         mm->total_vm = oldmm->total_vm;
 629         mm->data_vm = oldmm->data_vm;
 630         mm->exec_vm = oldmm->exec_vm;
 631         mm->stack_vm = oldmm->stack_vm;
 632
 633         retval = ksm_fork(mm, oldmm);
 634         if (retval)
 635                 goto out;
 636         khugepaged_fork(mm, oldmm);
 637
 638         retval = vma_iter_bulk_alloc(&vmi, oldmm->map_count);
 639         if (retval)
 640                 goto out;
 641
 642         for_each_vma(old_vmi, mpnt) {
 643                 struct file *file;
 644
 645                 if (mpnt->vm_flags & VM_DONTCOPY) {
 646                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 647                         continue;
 648                 }
 649                 charge = 0;
 650                 /*
 651                  * Don't duplicate many vmas if we've been oom-killed (for
 652                  * example)
 653                  */
 654                 if (fatal_signal_pending(current)) {
 655                         retval = -EINTR;
 656                         goto loop_out;
 657                 }
 658                 if (mpnt->vm_flags & VM_ACCOUNT) {
 659                         unsigned long len = vma_pages(mpnt);
 660
 661                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 662                                 goto fail_nomem;
 663                         charge = len;
 664                 }
 665                 tmp = vm_area_dup(mpnt);
 666                 if (!tmp)
 667                         goto fail_nomem;
 668                 retval = vma_dup_policy(mpnt, tmp);
 669                 if (retval)
 670                         goto fail_nomem_policy;
 671                 tmp->vm_mm = mm;
 672                 retval = dup_userfaultfd(tmp, &uf);
 673                 if (retval)
 674                         goto fail_nomem_anon_vma_fork;
 675                 if (tmp->vm_flags & VM_WIPEONFORK) {
 676                         /*
 677                          * VM_WIPEONFORK gets a clean slate in the child.
 678                          * Don't prepare anon_vma until fault since we don't
 679                          * copy page for current vma.
 680                          */
 681                         tmp->anon_vma = NULL;
 682                 } else if (anon_vma_fork(tmp, mpnt))
 683                         goto fail_nomem_anon_vma_fork;
 684                 vm_flags_clear(tmp, VM_LOCKED_MASK);
 685                 file = tmp->vm_file;
 686                 if (file) {
 687                         struct address_space *mapping = file->f_mapping;
 688
 689                         get_file(file);
 690                         i_mmap_lock_write(mapping);
 691                         if (tmp->vm_flags & VM_SHARED)
 692                                 mapping_allow_writable(mapping);
 693                         flush_dcache_mmap_lock(mapping);
 694                         /* insert tmp into the share list, just after mpnt */
 695                         vma_interval_tree_insert_after(tmp, mpnt,
 696                                         &mapping->i_mmap);
 697                         flush_dcache_mmap_unlock(mapping);
 698                         i_mmap_unlock_write(mapping);
 699                 }
 700
 701                 /*
 702                  * Copy/update hugetlb private vma information.
 703                  */
 704                 if (is_vm_hugetlb_page(tmp))
 705                         hugetlb_dup_vma_private(tmp);
 706
 707                 /* Link the vma into the MT */
 708                 if (vma_iter_bulk_store(&vmi, tmp))
 709                         goto fail_nomem_vmi_store;
 710
 711                 mm->map_count++;
 712                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 713                         retval = copy_page_range(tmp, mpnt);
 714
 715                 if (tmp->vm_ops && tmp->vm_ops->open)
 716                         tmp->vm_ops->open(tmp);
 717
 718                 if (retval)
 719                         goto loop_out;
 720         }
 721         /* a new mm has just been created */
 722         retval = arch_dup_mmap(oldmm, mm);
 723 loop_out:
 724         vma_iter_free(&vmi);
 725 out:
 726         mmap_write_unlock(mm);
 727         flush_tlb_mm(oldmm);
 728         mmap_write_unlock(oldmm);
 729         dup_userfaultfd_complete(&uf);
 730 fail_uprobe_end:
 731         uprobe_end_dup_mmap();
 732         return retval;
 733
 734 fail_nomem_vmi_store:
 735         unlink_anon_vmas(tmp);
 736 fail_nomem_anon_vma_fork:
 737         mpol_put(vma_policy(tmp));
 738 fail_nomem_policy:
 739         vm_area_free(tmp);
 740 fail_nomem:
 741         retval = -ENOMEM;
 742         vm_unacct_memory(charge);
 743         goto loop_out;
 744 }
 745
 746 static inline int mm_alloc_pgd(struct mm_struct *mm)
 747 {
 748         mm->pgd = pgd_alloc(mm);
 749         if (unlikely(!mm->pgd))
 750                 return -ENOMEM;
 751         return 0;
 752 }
 753
 754 static inline void mm_free_pgd(struct mm_struct *mm)
 755 {
 756         pgd_free(mm, mm->pgd);
 757 }
 758 #else
 759 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 760 {
 761         mmap_write_lock(oldmm);
 762         dup_mm_exe_file(mm, oldmm);
 763         mmap_write_unlock(oldmm);
 764         return 0;
 765 }
 766 #define mm_alloc_pgd(mm)        (0)
 767 #define mm_free_pgd(mm)
 768 #endif /* CONFIG_MMU */
 769
 770 static void check_mm(struct mm_struct *mm)
 771 {
 772         int i;
 773
 774         BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
 775                          "Please make sure 'struct resident_page_types[]' is updated as well");
 776
 777         for (i = 0; i < NR_MM_COUNTERS; i++) {
 778                 long x = percpu_counter_sum(&mm->rss_stat[i]);
 779
 780                 if (unlikely(x))
 781                         pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
 782                                  mm, resident_page_types[i], x);
 783         }
 784
 785         if (mm_pgtables_bytes(mm))
 786                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 787                                 mm_pgtables_bytes(mm));
 788
 789 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 790         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 791 #endif
 792 }
 793
 794 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 795 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 796
 797 static void do_check_lazy_tlb(void *arg)
 798 {
 799         struct mm_struct *mm = arg;
 800
 801         WARN_ON_ONCE(current->active_mm == mm);
 802 }
 803
 804 static void do_shoot_lazy_tlb(void *arg)
 805 {
 806         struct mm_struct *mm = arg;
 807
 808         if (current->active_mm == mm) {
 809                 WARN_ON_ONCE(current->mm);
 810                 current->active_mm = &init_mm;
 811                 switch_mm(mm, &init_mm, current);
 812         }
 813 }
 814
 815 static void cleanup_lazy_tlbs(struct mm_struct *mm)
 816 {
 817         if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) {
 818                 /*
 819                  * In this case, lazy tlb mms are refounted and would not reach
 820                  * __mmdrop until all CPUs have switched away and mmdrop()ed.
 821                  */
 822                 return;
 823         }
 824
 825         /*
 826          * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
 827          * requires lazy mm users to switch to another mm when the refcount
 828          * drops to zero, before the mm is freed. This requires IPIs here to
 829          * switch kernel threads to init_mm.
 830          *
 831          * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
 832          * switch with the final userspace teardown TLB flush which leaves the
 833          * mm lazy on this CPU but no others, reducing the need for additional
 834          * IPIs here. There are cases where a final IPI is still required here,
 835          * such as the final mmdrop being performed on a different CPU than the
 836          * one exiting, or kernel threads using the mm when userspace exits.
 837          *
 838          * IPI overheads have not found to be expensive, but they could be
 839          * reduced in a number of possible ways, for example (roughly
 840          * increasing order of complexity):
 841          * - The last lazy reference created by exit_mm() could instead switch
 842          *   to init_mm, however it's probable this will run on the same CPU
 843          *   immediately afterwards, so this may not reduce IPIs much.
 844          * - A batch of mms requiring IPIs could be gathered and freed at once.
 845          * - CPUs store active_mm where it can be remotely checked without a
 846          *   lock, to filter out false-positives in the cpumask.
 847          * - After mm_users or mm_count reaches zero, switching away from the
 848          *   mm could clear mm_cpumask to reduce some IPIs, perhaps together
 849          *   with some batching or delaying of the final IPIs.
 850          * - A delayed freeing and RCU-like quiescing sequence based on mm
 851          *   switching to avoid IPIs completely.
 852          */
 853         on_each_cpu_mask(mm_cpumask(mm), do_shoot_lazy_tlb, (void *)mm, 1);
 854         if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES))
 855                 on_each_cpu(do_check_lazy_tlb, (void *)mm, 1);
 856 }
 857
 858 /*
 859  * Called when the last reference to the mm
 860  * is dropped: either by a lazy thread or by
 861  * mmput. Free the page directory and the mm.
 862  */
 863 void __mmdrop(struct mm_struct *mm)
 864 {
 865         int i;
 866
 867         BUG_ON(mm == &init_mm);
 868         WARN_ON_ONCE(mm == current->mm);
 869
 870         /* Ensure no CPUs are using this as their lazy tlb mm */
 871         cleanup_lazy_tlbs(mm);
 872
 873         WARN_ON_ONCE(mm == current->active_mm);
 874         mm_free_pgd(mm);
 875         destroy_context(mm);
 876         mmu_notifier_subscriptions_destroy(mm);
 877         check_mm(mm);
 878         put_user_ns(mm->user_ns);
 879         mm_pasid_drop(mm);
 880
 881         for (i = 0; i < NR_MM_COUNTERS; i++)
 882                 percpu_counter_destroy(&mm->rss_stat[i]);
 883         free_mm(mm);
 884 }
 885 EXPORT_SYMBOL_GPL(__mmdrop);
 886
 887 static void mmdrop_async_fn(struct work_struct *work)
 888 {
 889         struct mm_struct *mm;
 890
 891         mm = container_of(work, struct mm_struct, async_put_work);
 892         __mmdrop(mm);
 893 }
 894
 895 static void mmdrop_async(struct mm_struct *mm)
 896 {
 897         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 898                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 899                 schedule_work(&mm->async_put_work);
 900         }
 901 }
 902
 903 static inline void free_signal_struct(struct signal_struct *sig)
 904 {
 905         taskstats_tgid_free(sig);
 906         sched_autogroup_exit(sig);
 907         /*
 908          * __mmdrop is not safe to call from softirq context on x86 due to
 909          * pgd_dtor so postpone it to the async context
 910          */
 911         if (sig->oom_mm)
 912                 mmdrop_async(sig->oom_mm);
 913         kmem_cache_free(signal_cachep, sig);
 914 }
 915
 916 static inline void put_signal_struct(struct signal_struct *sig)
 917 {
 918         if (refcount_dec_and_test(&sig->sigcnt))
 919                 free_signal_struct(sig);
 920 }
 921
 922 void __put_task_struct(struct task_struct *tsk)
 923 {
 924         WARN_ON(!tsk->exit_state);
 925         WARN_ON(refcount_read(&tsk->usage));
 926         WARN_ON(tsk == current);
 927
 928         io_uring_free(tsk);
 929         cgroup_free(tsk);
 930         task_numa_free(tsk, true);
 931         security_task_free(tsk);
 932         bpf_task_storage_free(tsk);
 933         exit_creds(tsk);
 934         delayacct_tsk_free(tsk);
 935         put_signal_struct(tsk->signal);
 936         sched_core_free(tsk);
 937         free_task(tsk);
 938 }
 939 EXPORT_SYMBOL_GPL(__put_task_struct);
 940
 941 void __init __weak arch_task_cache_init(void) { }
 942
 943 /*
 944  * set_max_threads
 945  */
 946 static void set_max_threads(unsigned int max_threads_suggested)
 947 {
 948         u64 threads;
 949         unsigned long nr_pages = totalram_pages();
 950
 951         /*
 952          * The number of threads shall be limited such that the thread
 953          * structures may only consume a small part of the available memory.
 954          */
 955         if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
 956                 threads = MAX_THREADS;
 957         else
 958                 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
 959                                     (u64) THREAD_SIZE * 8UL);
 960
 961         if (threads > max_threads_suggested)
 962                 threads = max_threads_suggested;
 963
 964         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 965 }
 966
 967 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 968 /* Initialized by the architecture: */
 969 int arch_task_struct_size __read_mostly;
 970 #endif
 971
 972 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 973 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 974 {
 975         /* Fetch thread_struct whitelist for the architecture. */
 976         arch_thread_struct_whitelist(offset, size);
 977
 978         /*
 979          * Handle zero-sized whitelist or empty thread_struct, otherwise
 980          * adjust offset to position of thread_struct in task_struct.
 981          */
 982         if (unlikely(*size == 0))
 983                 *offset = 0;
 984         else
 985                 *offset += offsetof(struct task_struct, thread);
 986 }
 987 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
 988
 989 void __init fork_init(void)
 990 {
 991         int i;
 992 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 993 #ifndef ARCH_MIN_TASKALIGN
 994 #define ARCH_MIN_TASKALIGN      0
 995 #endif
 996         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 997         unsigned long useroffset, usersize;
 998
 999         /* create a slab on which task_structs can be allocated */
1000         task_struct_whitelist(&useroffset, &usersize);
1001         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
1002                         arch_task_struct_size, align,
1003                         SLAB_PANIC|SLAB_ACCOUNT,
1004                         useroffset, usersize, NULL);
1005 #endif
1006
1007         /* do the arch specific task caches init */
1008         arch_task_cache_init();
1009
1010         set_max_threads(MAX_THREADS);
1011
1012         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
1013         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
1014         init_task.signal->rlim[RLIMIT_SIGPENDING] =
1015                 init_task.signal->rlim[RLIMIT_NPROC];
1016
1017         for (i = 0; i < UCOUNT_COUNTS; i++)
1018                 init_user_ns.ucount_max[i] = max_threads/2;
1019
1020         set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC,      RLIM_INFINITY);
1021         set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE,   RLIM_INFINITY);
1022         set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
1023         set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK,    RLIM_INFINITY);
1024
1025 #ifdef CONFIG_VMAP_STACK
1026         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
1027                           NULL, free_vm_stack_cache);
1028 #endif
1029
1030         scs_init();
1031
1032         lockdep_init_task(&init_task);
1033         uprobes_init();
1034 }
1035
1036 int __weak arch_dup_task_struct(struct task_struct *dst,
1037                                                struct task_struct *src)
1038 {
1039         *dst = *src;
1040         return 0;
1041 }
1042
1043 void set_task_stack_end_magic(struct task_struct *tsk)
1044 {
1045         unsigned long *stackend;
1046
1047         stackend = end_of_stack(tsk);
1048         *stackend = STACK_END_MAGIC;    /* for overflow detection */
1049 }
1050
1051 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
1052 {
1053         struct task_struct *tsk;
1054         int err;
1055
1056         if (node == NUMA_NO_NODE)
1057                 node = tsk_fork_get_node(orig);
1058         tsk = alloc_task_struct_node(node);
1059         if (!tsk)
1060                 return NULL;
1061
1062         err = arch_dup_task_struct(tsk, orig);
1063         if (err)
1064                 goto free_tsk;
1065
1066         err = alloc_thread_stack_node(tsk, node);
1067         if (err)
1068                 goto free_tsk;
1069
1070 #ifdef CONFIG_THREAD_INFO_IN_TASK
1071         refcount_set(&tsk->stack_refcount, 1);
1072 #endif
1073         account_kernel_stack(tsk, 1);
1074
1075         err = scs_prepare(tsk, node);
1076         if (err)
1077                 goto free_stack;
1078
1079 #ifdef CONFIG_SECCOMP
1080         /*
1081          * We must handle setting up seccomp filters once we're under
1082          * the sighand lock in case orig has changed between now and
1083          * then. Until then, filter must be NULL to avoid messing up
1084          * the usage counts on the error path calling free_task.
1085          */
1086         tsk->seccomp.filter = NULL;
1087 #endif
1088
1089         setup_thread_stack(tsk, orig);
1090         clear_user_return_notifier(tsk);
1091         clear_tsk_need_resched(tsk);
1092         set_task_stack_end_magic(tsk);
1093         clear_syscall_work_syscall_user_dispatch(tsk);
1094
1095 #ifdef CONFIG_STACKPROTECTOR
1096         tsk->stack_canary = get_random_canary();
1097 #endif
1098         if (orig->cpus_ptr == &orig->cpus_mask)
1099                 tsk->cpus_ptr = &tsk->cpus_mask;
1100         dup_user_cpus_ptr(tsk, orig, node);
1101
1102         /*
1103          * One for the user space visible state that goes away when reaped.
1104          * One for the scheduler.
1105          */
1106         refcount_set(&tsk->rcu_users, 2);
1107         /* One for the rcu users */
1108         refcount_set(&tsk->usage, 1);
1109 #ifdef CONFIG_BLK_DEV_IO_TRACE
1110         tsk->btrace_seq = 0;
1111 #endif
1112         tsk->splice_pipe = NULL;
1113         tsk->task_frag.page = NULL;
1114         tsk->wake_q.next = NULL;
1115         tsk->worker_private = NULL;
1116
1117         kcov_task_init(tsk);
1118         kmsan_task_create(tsk);
1119         kmap_local_fork(tsk);
1120
1121 #ifdef CONFIG_FAULT_INJECTION
1122         tsk->fail_nth = 0;
1123 #endif
1124
1125 #ifdef CONFIG_BLK_CGROUP
1126         tsk->throttle_disk = NULL;
1127         tsk->use_memdelay = 0;
1128 #endif
1129
1130 #ifdef CONFIG_IOMMU_SVA
1131         tsk->pasid_activated = 0;
1132 #endif
1133
1134 #ifdef CONFIG_MEMCG
1135         tsk->active_memcg = NULL;
1136 #endif
1137
1138 #ifdef CONFIG_CPU_SUP_INTEL
1139         tsk->reported_split_lock = 0;
1140 #endif
1141
1142 #ifdef CONFIG_SCHED_MM_CID
1143         tsk->mm_cid = -1;
1144         tsk->mm_cid_active = 0;
1145 #endif
1146         return tsk;
1147
1148 free_stack:
1149         exit_task_stack_account(tsk);
1150         free_thread_stack(tsk);
1151 free_tsk:
1152         free_task_struct(tsk);
1153         return NULL;
1154 }
1155
1156 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1157
1158 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1159
1160 static int __init coredump_filter_setup(char *s)
1161 {
1162         default_dump_filter =
1163                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1164                 MMF_DUMP_FILTER_MASK;
1165         return 1;
1166 }
1167
1168 __setup("coredump_filter=", coredump_filter_setup);
1169
1170 #include <linux/init_task.h>
1171
1172 static void mm_init_aio(struct mm_struct *mm)
1173 {
1174 #ifdef CONFIG_AIO
1175         spin_lock_init(&mm->ioctx_lock);
1176         mm->ioctx_table = NULL;
1177 #endif
1178 }
1179
1180 static __always_inline void mm_clear_owner(struct mm_struct *mm,
1181                                            struct task_struct *p)
1182 {
1183 #ifdef CONFIG_MEMCG
1184         if (mm->owner == p)
1185                 WRITE_ONCE(mm->owner, NULL);
1186 #endif
1187 }
1188
1189 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1190 {
1191 #ifdef CONFIG_MEMCG
1192         mm->owner = p;
1193 #endif
1194 }
1195
1196 static void mm_init_uprobes_state(struct mm_struct *mm)
1197 {
1198 #ifdef CONFIG_UPROBES
1199         mm->uprobes_state.xol_area = NULL;
1200 #endif
1201 }
1202
1203 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1204         struct user_namespace *user_ns)
1205 {
1206         int i;
1207
1208         mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1209         mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1210         atomic_set(&mm->mm_users, 1);
1211         atomic_set(&mm->mm_count, 1);
1212         seqcount_init(&mm->write_protect_seq);
1213         mmap_init_lock(mm);
1214         INIT_LIST_HEAD(&mm->mmlist);
1215 #ifdef CONFIG_PER_VMA_LOCK
1216         mm->mm_lock_seq = 0;
1217 #endif
1218         mm_pgtables_bytes_init(mm);
1219         mm->map_count = 0;
1220         mm->locked_vm = 0;
1221         atomic64_set(&mm->pinned_vm, 0);
1222         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1223         spin_lock_init(&mm->page_table_lock);
1224         spin_lock_init(&mm->arg_lock);
1225         mm_init_cpumask(mm);
1226         mm_init_aio(mm);
1227         mm_init_owner(mm, p);
1228         mm_pasid_init(mm);
1229         RCU_INIT_POINTER(mm->exe_file, NULL);
1230         mmu_notifier_subscriptions_init(mm);
1231         init_tlb_flush_pending(mm);
1232 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1233         mm->pmd_huge_pte = NULL;
1234 #endif
1235         mm_init_uprobes_state(mm);
1236         hugetlb_count_init(mm);
1237
1238         if (current->mm) {
1239                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1240                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1241         } else {
1242                 mm->flags = default_dump_filter;
1243                 mm->def_flags = 0;
1244         }
1245
1246         if (mm_alloc_pgd(mm))
1247                 goto fail_nopgd;
1248
1249         if (init_new_context(p, mm))
1250                 goto fail_nocontext;
1251
1252         for (i = 0; i < NR_MM_COUNTERS; i++)
1253                 if (percpu_counter_init(&mm->rss_stat[i], 0, GFP_KERNEL_ACCOUNT))
1254                         goto fail_pcpu;
1255
1256         mm->user_ns = get_user_ns(user_ns);
1257         lru_gen_init_mm(mm);
1258         mm_init_cid(mm);
1259         return mm;
1260
1261 fail_pcpu:
1262         while (i > 0)
1263                 percpu_counter_destroy(&mm->rss_stat[--i]);
1264 fail_nocontext:
1265         mm_free_pgd(mm);
1266 fail_nopgd:
1267         free_mm(mm);
1268         return NULL;
1269 }
1270
1271 /*
1272  * Allocate and initialize an mm_struct.
1273  */
1274 struct mm_struct *mm_alloc(void)
1275 {
1276         struct mm_struct *mm;
1277
1278         mm = allocate_mm();
1279         if (!mm)
1280                 return NULL;
1281
1282         memset(mm, 0, sizeof(*mm));
1283         return mm_init(mm, current, current_user_ns());
1284 }
1285
1286 static inline void __mmput(struct mm_struct *mm)
1287 {
1288         VM_BUG_ON(atomic_read(&mm->mm_users));
1289
1290         uprobe_clear_state(mm);
1291         exit_aio(mm);
1292         ksm_exit(mm);
1293         khugepaged_exit(mm); /* must run before exit_mmap */
1294         exit_mmap(mm);
1295         mm_put_huge_zero_page(mm);
1296         set_mm_exe_file(mm, NULL);
1297         if (!list_empty(&mm->mmlist)) {
1298                 spin_lock(&mmlist_lock);
1299                 list_del(&mm->mmlist);
1300                 spin_unlock(&mmlist_lock);
1301         }
1302         if (mm->binfmt)
1303                 module_put(mm->binfmt->module);
1304         lru_gen_del_mm(mm);
1305         mmdrop(mm);
1306 }
1307
1308 /*
1309  * Decrement the use count and release all resources for an mm.
1310  */
1311 void mmput(struct mm_struct *mm)
1312 {
1313         might_sleep();
1314
1315         if (atomic_dec_and_test(&mm->mm_users))
1316                 __mmput(mm);
1317 }
1318 EXPORT_SYMBOL_GPL(mmput);
1319
1320 #ifdef CONFIG_MMU
1321 static void mmput_async_fn(struct work_struct *work)
1322 {
1323         struct mm_struct *mm = container_of(work, struct mm_struct,
1324                                             async_put_work);
1325
1326         __mmput(mm);
1327 }
1328
1329 void mmput_async(struct mm_struct *mm)
1330 {
1331         if (atomic_dec_and_test(&mm->mm_users)) {
1332                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1333                 schedule_work(&mm->async_put_work);
1334         }
1335 }
1336 EXPORT_SYMBOL_GPL(mmput_async);
1337 #endif
1338
1339 /**
1340  * set_mm_exe_file - change a reference to the mm's executable file
1341  *
1342  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1343  *
1344  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1345  * invocations: in mmput() nobody alive left, in execve task is single
1346  * threaded.
1347  *
1348  * Can only fail if new_exe_file != NULL.
1349  */
1350 int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1351 {
1352         struct file *old_exe_file;
1353
1354         /*
1355          * It is safe to dereference the exe_file without RCU as
1356          * this function is only called if nobody else can access
1357          * this mm -- see comment above for justification.
1358          */
1359         old_exe_file = rcu_dereference_raw(mm->exe_file);
1360
1361         if (new_exe_file) {
1362                 /*
1363                  * We expect the caller (i.e., sys_execve) to already denied
1364                  * write access, so this is unlikely to fail.
1365                  */
1366                 if (unlikely(deny_write_access(new_exe_file)))
1367                         return -EACCES;
1368                 get_file(new_exe_file);
1369         }
1370         rcu_assign_pointer(mm->exe_file, new_exe_file);
1371         if (old_exe_file) {
1372                 allow_write_access(old_exe_file);
1373                 fput(old_exe_file);
1374         }
1375         return 0;
1376 }
1377
1378 /**
1379  * replace_mm_exe_file - replace a reference to the mm's executable file
1380  *
1381  * This changes mm's executable file (shown as symlink /proc/[pid]/exe),
1382  * dealing with concurrent invocation and without grabbing the mmap lock in
1383  * write mode.
1384  *
1385  * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1386  */
1387 int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1388 {
1389         struct vm_area_struct *vma;
1390         struct file *old_exe_file;
1391         int ret = 0;
1392
1393         /* Forbid mm->exe_file change if old file still mapped. */
1394         old_exe_file = get_mm_exe_file(mm);
1395         if (old_exe_file) {
1396                 VMA_ITERATOR(vmi, mm, 0);
1397                 mmap_read_lock(mm);
1398                 for_each_vma(vmi, vma) {
1399                         if (!vma->vm_file)
1400                                 continue;
1401                         if (path_equal(&vma->vm_file->f_path,
1402                                        &old_exe_file->f_path)) {
1403                                 ret = -EBUSY;
1404                                 break;
1405                         }
1406                 }
1407                 mmap_read_unlock(mm);
1408                 fput(old_exe_file);
1409                 if (ret)
1410                         return ret;
1411         }
1412
1413         /* set the new file, lockless */
1414         ret = deny_write_access(new_exe_file);
1415         if (ret)
1416                 return -EACCES;
1417         get_file(new_exe_file);
1418
1419         old_exe_file = xchg(&mm->exe_file, new_exe_file);
1420         if (old_exe_file) {
1421                 /*
1422                  * Don't race with dup_mmap() getting the file and disallowing
1423                  * write access while someone might open the file writable.
1424                  */
1425                 mmap_read_lock(mm);
1426                 allow_write_access(old_exe_file);
1427                 fput(old_exe_file);
1428                 mmap_read_unlock(mm);
1429         }
1430         return 0;
1431 }
1432
1433 /**
1434  * get_mm_exe_file - acquire a reference to the mm's executable file
1435  *
1436  * Returns %NULL if mm has no associated executable file.
1437  * User must release file via fput().
1438  */
1439 struct file *get_mm_exe_file(struct mm_struct *mm)
1440 {
1441         struct file *exe_file;
1442
1443         rcu_read_lock();
1444         exe_file = rcu_dereference(mm->exe_file);
1445         if (exe_file && !get_file_rcu(exe_file))
1446                 exe_file = NULL;
1447         rcu_read_unlock();
1448         return exe_file;
1449 }
1450
1451 /**
1452  * get_task_exe_file - acquire a reference to the task's executable file
1453  *
1454  * Returns %NULL if task's mm (if any) has no associated executable file or
1455  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1456  * User must release file via fput().
1457  */
1458 struct file *get_task_exe_file(struct task_struct *task)
1459 {
1460         struct file *exe_file = NULL;
1461         struct mm_struct *mm;
1462
1463         task_lock(task);
1464         mm = task->mm;
1465         if (mm) {
1466                 if (!(task->flags & PF_KTHREAD))
1467                         exe_file = get_mm_exe_file(mm);
1468         }
1469         task_unlock(task);
1470         return exe_file;
1471 }
1472
1473 /**
1474  * get_task_mm - acquire a reference to the task's mm
1475  *
1476  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1477  * this kernel workthread has transiently adopted a user mm with use_mm,
1478  * to do its AIO) is not set and if so returns a reference to it, after
1479  * bumping up the use count.  User must release the mm via mmput()
1480  * after use.  Typically used by /proc and ptrace.
1481  */
1482 struct mm_struct *get_task_mm(struct task_struct *task)
1483 {
1484         struct mm_struct *mm;
1485
1486         task_lock(task);
1487         mm = task->mm;
1488         if (mm) {
1489                 if (task->flags & PF_KTHREAD)
1490                         mm = NULL;
1491                 else
1492                         mmget(mm);
1493         }
1494         task_unlock(task);
1495         return mm;
1496 }
1497 EXPORT_SYMBOL_GPL(get_task_mm);
1498
1499 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1500 {
1501         struct mm_struct *mm;
1502         int err;
1503
1504         err =  down_read_killable(&task->signal->exec_update_lock);
1505         if (err)
1506                 return ERR_PTR(err);
1507
1508         mm = get_task_mm(task);
1509         if (mm && mm != current->mm &&
1510                         !ptrace_may_access(task, mode)) {
1511                 mmput(mm);
1512                 mm = ERR_PTR(-EACCES);
1513         }
1514         up_read(&task->signal->exec_update_lock);
1515
1516         return mm;
1517 }
1518
1519 static void complete_vfork_done(struct task_struct *tsk)
1520 {
1521         struct completion *vfork;
1522
1523         task_lock(tsk);
1524         vfork = tsk->vfork_done;
1525         if (likely(vfork)) {
1526                 tsk->vfork_done = NULL;
1527                 complete(vfork);
1528         }
1529         task_unlock(tsk);
1530 }
1531
1532 static int wait_for_vfork_done(struct task_struct *child,
1533                                 struct completion *vfork)
1534 {
1535         unsigned int state = TASK_UNINTERRUPTIBLE|TASK_KILLABLE|TASK_FREEZABLE;
1536         int killed;
1537
1538         cgroup_enter_frozen();
1539         killed = wait_for_completion_state(vfork, state);
1540         cgroup_leave_frozen(false);
1541
1542         if (killed) {
1543                 task_lock(child);
1544                 child->vfork_done = NULL;
1545                 task_unlock(child);
1546         }
1547
1548         put_task_struct(child);
1549         return killed;
1550 }
1551
1552 /* Please note the differences between mmput and mm_release.
1553  * mmput is called whenever we stop holding onto a mm_struct,
1554  * error success whatever.
1555  *
1556  * mm_release is called after a mm_struct has been removed
1557  * from the current process.
1558  *
1559  * This difference is important for error handling, when we
1560  * only half set up a mm_struct for a new process and need to restore
1561  * the old one.  Because we mmput the new mm_struct before
1562  * restoring the old one. . .
1563  * Eric Biederman 10 January 1998
1564  */
1565 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1566 {
1567         uprobe_free_utask(tsk);
1568
1569         /* Get rid of any cached register state */
1570         deactivate_mm(tsk, mm);
1571
1572         /*
1573          * Signal userspace if we're not exiting with a core dump
1574          * because we want to leave the value intact for debugging
1575          * purposes.
1576          */
1577         if (tsk->clear_child_tid) {
1578                 if (atomic_read(&mm->mm_users) > 1) {
1579                         /*
1580                          * We don't check the error code - if userspace has
1581                          * not set up a proper pointer then tough luck.
1582                          */
1583                         put_user(0, tsk->clear_child_tid);
1584                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1585                                         1, NULL, NULL, 0, 0);
1586                 }
1587                 tsk->clear_child_tid = NULL;
1588         }
1589
1590         /*
1591          * All done, finally we can wake up parent and return this mm to him.
1592          * Also kthread_stop() uses this completion for synchronization.
1593          */
1594         if (tsk->vfork_done)
1595                 complete_vfork_done(tsk);
1596 }
1597
1598 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1599 {
1600         futex_exit_release(tsk);
1601         mm_release(tsk, mm);
1602 }
1603
1604 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1605 {
1606         futex_exec_release(tsk);
1607         mm_release(tsk, mm);
1608 }
1609
1610 /**
1611  * dup_mm() - duplicates an existing mm structure
1612  * @tsk: the task_struct with which the new mm will be associated.
1613  * @oldmm: the mm to duplicate.
1614  *
1615  * Allocates a new mm structure and duplicates the provided @oldmm structure
1616  * content into it.
1617  *
1618  * Return: the duplicated mm or NULL on failure.
1619  */
1620 static struct mm_struct *dup_mm(struct task_struct *tsk,
1621                                 struct mm_struct *oldmm)
1622 {
1623         struct mm_struct *mm;
1624         int err;
1625
1626         mm = allocate_mm();
1627         if (!mm)
1628                 goto fail_nomem;
1629
1630         memcpy(mm, oldmm, sizeof(*mm));
1631
1632         if (!mm_init(mm, tsk, mm->user_ns))
1633                 goto fail_nomem;
1634
1635         err = dup_mmap(mm, oldmm);
1636         if (err)
1637                 goto free_pt;
1638
1639         mm->hiwater_rss = get_mm_rss(mm);
1640         mm->hiwater_vm = mm->total_vm;
1641
1642         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1643                 goto free_pt;
1644
1645         return mm;
1646
1647 free_pt:
1648         /* don't put binfmt in mmput, we haven't got module yet */
1649         mm->binfmt = NULL;
1650         mm_init_owner(mm, NULL);
1651         mmput(mm);
1652
1653 fail_nomem:
1654         return NULL;
1655 }
1656
1657 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1658 {
1659         struct mm_struct *mm, *oldmm;
1660
1661         tsk->min_flt = tsk->maj_flt = 0;
1662         tsk->nvcsw = tsk->nivcsw = 0;
1663 #ifdef CONFIG_DETECT_HUNG_TASK
1664         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1665         tsk->last_switch_time = 0;
1666 #endif
1667
1668         tsk->mm = NULL;
1669         tsk->active_mm = NULL;
1670
1671         /*
1672          * Are we cloning a kernel thread?
1673          *
1674          * We need to steal a active VM for that..
1675          */
1676         oldmm = current->mm;
1677         if (!oldmm)
1678                 return 0;
1679
1680         if (clone_flags & CLONE_VM) {
1681                 mmget(oldmm);
1682                 mm = oldmm;
1683         } else {
1684                 mm = dup_mm(tsk, current->mm);
1685                 if (!mm)
1686                         return -ENOMEM;
1687         }
1688
1689         tsk->mm = mm;
1690         tsk->active_mm = mm;
1691         sched_mm_cid_fork(tsk);
1692         return 0;
1693 }
1694
1695 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1696 {
1697         struct fs_struct *fs = current->fs;
1698         if (clone_flags & CLONE_FS) {
1699                 /* tsk->fs is already what we want */
1700                 spin_lock(&fs->lock);
1701                 if (fs->in_exec) {
1702                         spin_unlock(&fs->lock);
1703                         return -EAGAIN;
1704                 }
1705                 fs->users++;
1706                 spin_unlock(&fs->lock);
1707                 return 0;
1708         }
1709         tsk->fs = copy_fs_struct(fs);
1710         if (!tsk->fs)
1711                 return -ENOMEM;
1712         return 0;
1713 }
1714
1715 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1716 {
1717         struct files_struct *oldf, *newf;
1718         int error = 0;
1719
1720         /*
1721          * A background process may not have any files ...
1722          */
1723         oldf = current->files;
1724         if (!oldf)
1725                 goto out;
1726
1727         if (clone_flags & CLONE_FILES) {
1728                 atomic_inc(&oldf->count);
1729                 goto out;
1730         }
1731
1732         newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1733         if (!newf)
1734                 goto out;
1735
1736         tsk->files = newf;
1737         error = 0;
1738 out:
1739         return error;
1740 }
1741
1742 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1743 {
1744         struct sighand_struct *sig;
1745
1746         if (clone_flags & CLONE_SIGHAND) {
1747                 refcount_inc(&current->sighand->count);
1748                 return 0;
1749         }
1750         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1751         RCU_INIT_POINTER(tsk->sighand, sig);
1752         if (!sig)
1753                 return -ENOMEM;
1754
1755         refcount_set(&sig->count, 1);
1756         spin_lock_irq(&current->sighand->siglock);
1757         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1758         spin_unlock_irq(&current->sighand->siglock);
1759
1760         /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1761         if (clone_flags & CLONE_CLEAR_SIGHAND)
1762                 flush_signal_handlers(tsk, 0);
1763
1764         return 0;
1765 }
1766
1767 void __cleanup_sighand(struct sighand_struct *sighand)
1768 {
1769         if (refcount_dec_and_test(&sighand->count)) {
1770                 signalfd_cleanup(sighand);
1771                 /*
1772                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1773                  * without an RCU grace period, see __lock_task_sighand().
1774                  */
1775                 kmem_cache_free(sighand_cachep, sighand);
1776         }
1777 }
1778
1779 /*
1780  * Initialize POSIX timer handling for a thread group.
1781  */
1782 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1783 {
1784         struct posix_cputimers *pct = &sig->posix_cputimers;
1785         unsigned long cpu_limit;
1786
1787         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1788         posix_cputimers_group_init(pct, cpu_limit);
1789 }
1790
1791 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1792 {
1793         struct signal_struct *sig;
1794
1795         if (clone_flags & CLONE_THREAD)
1796                 return 0;
1797
1798         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1799         tsk->signal = sig;
1800         if (!sig)
1801                 return -ENOMEM;
1802
1803         sig->nr_threads = 1;
1804         sig->quick_threads = 1;
1805         atomic_set(&sig->live, 1);
1806         refcount_set(&sig->sigcnt, 1);
1807
1808         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1809         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1810         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1811
1812         init_waitqueue_head(&sig->wait_chldexit);
1813         sig->curr_target = tsk;
1814         init_sigpending(&sig->shared_pending);
1815         INIT_HLIST_HEAD(&sig->multiprocess);
1816         seqlock_init(&sig->stats_lock);
1817         prev_cputime_init(&sig->prev_cputime);
1818
1819 #ifdef CONFIG_POSIX_TIMERS
1820         INIT_LIST_HEAD(&sig->posix_timers);
1821         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1822         sig->real_timer.function = it_real_fn;
1823 #endif
1824
1825         task_lock(current->group_leader);
1826         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1827         task_unlock(current->group_leader);
1828
1829         posix_cpu_timers_init_group(sig);
1830
1831         tty_audit_fork(sig);
1832         sched_autogroup_fork(sig);
1833
1834         sig->oom_score_adj = current->signal->oom_score_adj;
1835         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1836
1837         mutex_init(&sig->cred_guard_mutex);
1838         init_rwsem(&sig->exec_update_lock);
1839
1840         return 0;
1841 }
1842
1843 static void copy_seccomp(struct task_struct *p)
1844 {
1845 #ifdef CONFIG_SECCOMP
1846         /*
1847          * Must be called with sighand->lock held, which is common to
1848          * all threads in the group. Holding cred_guard_mutex is not
1849          * needed because this new task is not yet running and cannot
1850          * be racing exec.
1851          */
1852         assert_spin_locked(&current->sighand->siglock);
1853
1854         /* Ref-count the new filter user, and assign it. */
1855         get_seccomp_filter(current);
1856         p->seccomp = current->seccomp;
1857
1858         /*
1859          * Explicitly enable no_new_privs here in case it got set
1860          * between the task_struct being duplicated and holding the
1861          * sighand lock. The seccomp state and nnp must be in sync.
1862          */
1863         if (task_no_new_privs(current))
1864                 task_set_no_new_privs(p);
1865
1866         /*
1867          * If the parent gained a seccomp mode after copying thread
1868          * flags and between before we held the sighand lock, we have
1869          * to manually enable the seccomp thread flag here.
1870          */
1871         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1872                 set_task_syscall_work(p, SECCOMP);
1873 #endif
1874 }
1875
1876 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1877 {
1878         current->clear_child_tid = tidptr;
1879
1880         return task_pid_vnr(current);
1881 }
1882
1883 static void rt_mutex_init_task(struct task_struct *p)
1884 {
1885         raw_spin_lock_init(&p->pi_lock);
1886 #ifdef CONFIG_RT_MUTEXES
1887         p->pi_waiters = RB_ROOT_CACHED;
1888         p->pi_top_task = NULL;
1889         p->pi_blocked_on = NULL;
1890 #endif
1891 }
1892
1893 static inline void init_task_pid_links(struct task_struct *task)
1894 {
1895         enum pid_type type;
1896
1897         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1898                 INIT_HLIST_NODE(&task->pid_links[type]);
1899 }
1900
1901 static inline void
1902 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1903 {
1904         if (type == PIDTYPE_PID)
1905                 task->thread_pid = pid;
1906         else
1907                 task->signal->pids[type] = pid;
1908 }
1909
1910 static inline void rcu_copy_process(struct task_struct *p)
1911 {
1912 #ifdef CONFIG_PREEMPT_RCU
1913         p->rcu_read_lock_nesting = 0;
1914         p->rcu_read_unlock_special.s = 0;
1915         p->rcu_blocked_node = NULL;
1916         INIT_LIST_HEAD(&p->rcu_node_entry);
1917 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1918 #ifdef CONFIG_TASKS_RCU
1919         p->rcu_tasks_holdout = false;
1920         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1921         p->rcu_tasks_idle_cpu = -1;
1922 #endif /* #ifdef CONFIG_TASKS_RCU */
1923 #ifdef CONFIG_TASKS_TRACE_RCU
1924         p->trc_reader_nesting = 0;
1925         p->trc_reader_special.s = 0;
1926         INIT_LIST_HEAD(&p->trc_holdout_list);
1927         INIT_LIST_HEAD(&p->trc_blkd_node);
1928 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1929 }
1930
1931 struct pid *pidfd_pid(const struct file *file)
1932 {
1933         if (file->f_op == &pidfd_fops)
1934                 return file->private_data;
1935
1936         return ERR_PTR(-EBADF);
1937 }
1938
1939 static int pidfd_release(struct inode *inode, struct file *file)
1940 {
1941         struct pid *pid = file->private_data;
1942
1943         file->private_data = NULL;
1944         put_pid(pid);
1945         return 0;
1946 }
1947
1948 #ifdef CONFIG_PROC_FS
1949 /**
1950  * pidfd_show_fdinfo - print information about a pidfd
1951  * @m: proc fdinfo file
1952  * @f: file referencing a pidfd
1953  *
1954  * Pid:
1955  * This function will print the pid that a given pidfd refers to in the
1956  * pid namespace of the procfs instance.
1957  * If the pid namespace of the process is not a descendant of the pid
1958  * namespace of the procfs instance 0 will be shown as its pid. This is
1959  * similar to calling getppid() on a process whose parent is outside of
1960  * its pid namespace.
1961  *
1962  * NSpid:
1963  * If pid namespaces are supported then this function will also print
1964  * the pid of a given pidfd refers to for all descendant pid namespaces
1965  * starting from the current pid namespace of the instance, i.e. the
1966  * Pid field and the first entry in the NSpid field will be identical.
1967  * If the pid namespace of the process is not a descendant of the pid
1968  * namespace of the procfs instance 0 will be shown as its first NSpid
1969  * entry and no others will be shown.
1970  * Note that this differs from the Pid and NSpid fields in
1971  * /proc/<pid>/status where Pid and NSpid are always shown relative to
1972  * the  pid namespace of the procfs instance. The difference becomes
1973  * obvious when sending around a pidfd between pid namespaces from a
1974  * different branch of the tree, i.e. where no ancestral relation is
1975  * present between the pid namespaces:
1976  * - create two new pid namespaces ns1 and ns2 in the initial pid
1977  *   namespace (also take care to create new mount namespaces in the
1978  *   new pid namespace and mount procfs)
1979  * - create a process with a pidfd in ns1
1980  * - send pidfd from ns1 to ns2
1981  * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1982  *   have exactly one entry, which is 0
1983  */
1984 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1985 {
1986         struct pid *pid = f->private_data;
1987         struct pid_namespace *ns;
1988         pid_t nr = -1;
1989
1990         if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1991                 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1992                 nr = pid_nr_ns(pid, ns);
1993         }
1994
1995         seq_put_decimal_ll(m, "Pid:\t", nr);
1996
1997 #ifdef CONFIG_PID_NS
1998         seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1999         if (nr > 0) {
2000                 int i;
2001
2002                 /* If nr is non-zero it means that 'pid' is valid and that
2003                  * ns, i.e. the pid namespace associated with the procfs
2004                  * instance, is in the pid namespace hierarchy of pid.
2005                  * Start at one below the already printed level.
2006                  */
2007                 for (i = ns->level + 1; i <= pid->level; i++)
2008                         seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
2009         }
2010 #endif
2011         seq_putc(m, '\n');
2012 }
2013 #endif
2014
2015 /*
2016  * Poll support for process exit notification.
2017  */
2018 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
2019 {
2020         struct pid *pid = file->private_data;
2021         __poll_t poll_flags = 0;
2022
2023         poll_wait(file, &pid->wait_pidfd, pts);
2024
2025         /*
2026          * Inform pollers only when the whole thread group exits.
2027          * If the thread group leader exits before all other threads in the
2028          * group, then poll(2) should block, similar to the wait(2) family.
2029          */
2030         if (thread_group_exited(pid))
2031                 poll_flags = EPOLLIN | EPOLLRDNORM;
2032
2033         return poll_flags;
2034 }
2035
2036 const struct file_operations pidfd_fops = {
2037         .release = pidfd_release,
2038         .poll = pidfd_poll,
2039 #ifdef CONFIG_PROC_FS
2040         .show_fdinfo = pidfd_show_fdinfo,
2041 #endif
2042 };
2043
2044 static void __delayed_free_task(struct rcu_head *rhp)
2045 {
2046         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
2047
2048         free_task(tsk);
2049 }
2050
2051 static __always_inline void delayed_free_task(struct task_struct *tsk)
2052 {
2053         if (IS_ENABLED(CONFIG_MEMCG))
2054                 call_rcu(&tsk->rcu, __delayed_free_task);
2055         else
2056                 free_task(tsk);
2057 }
2058
2059 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
2060 {
2061         /* Skip if kernel thread */
2062         if (!tsk->mm)
2063                 return;
2064
2065         /* Skip if spawning a thread or using vfork */
2066         if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
2067                 return;
2068
2069         /* We need to synchronize with __set_oom_adj */
2070         mutex_lock(&oom_adj_mutex);
2071         set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
2072         /* Update the values in case they were changed after copy_signal */
2073         tsk->signal->oom_score_adj = current->signal->oom_score_adj;
2074         tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
2075         mutex_unlock(&oom_adj_mutex);
2076 }
2077
2078 #ifdef CONFIG_RV
2079 static void rv_task_fork(struct task_struct *p)
2080 {
2081         int i;
2082
2083         for (i = 0; i < RV_PER_TASK_MONITORS; i++)
2084                 p->rv[i].da_mon.monitoring = false;
2085 }
2086 #else
2087 #define rv_task_fork(p) do {} while (0)
2088 #endif
2089
2090 /*
2091  * This creates a new process as a copy of the old one,
2092  * but does not actually start it yet.
2093  *
2094  * It copies the registers, and all the appropriate
2095  * parts of the process environment (as per the clone
2096  * flags). The actual kick-off is left to the caller.
2097  */
2098 static __latent_entropy struct task_struct *copy_process(
2099                                         struct pid *pid,
2100                                         int trace,
2101                                         int node,
2102                                         struct kernel_clone_args *args)
2103 {
2104         int pidfd = -1, retval;
2105         struct task_struct *p;
2106         struct multiprocess_signals delayed;
2107         struct file *pidfile = NULL;
2108         const u64 clone_flags = args->flags;
2109         struct nsproxy *nsp = current->nsproxy;
2110
2111         /*
2112          * Don't allow sharing the root directory with processes in a different
2113          * namespace
2114          */
2115         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2116                 return ERR_PTR(-EINVAL);
2117
2118         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2119                 return ERR_PTR(-EINVAL);
2120
2121         /*
2122          * Thread groups must share signals as well, and detached threads
2123          * can only be started up within the thread group.
2124          */
2125         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2126                 return ERR_PTR(-EINVAL);
2127
2128         /*
2129          * Shared signal handlers imply shared VM. By way of the above,
2130          * thread groups also imply shared VM. Blocking this case allows
2131          * for various simplifications in other code.
2132          */
2133         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2134                 return ERR_PTR(-EINVAL);
2135
2136         /*
2137          * Siblings of global init remain as zombies on exit since they are
2138          * not reaped by their parent (swapper). To solve this and to avoid
2139          * multi-rooted process trees, prevent global and container-inits
2140          * from creating siblings.
2141          */
2142         if ((clone_flags & CLONE_PARENT) &&
2143                                 current->signal->flags & SIGNAL_UNKILLABLE)
2144                 return ERR_PTR(-EINVAL);
2145
2146         /*
2147          * If the new process will be in a different pid or user namespace
2148          * do not allow it to share a thread group with the forking task.
2149          */
2150         if (clone_flags & CLONE_THREAD) {
2151                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2152                     (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2153                         return ERR_PTR(-EINVAL);
2154         }
2155
2156         if (clone_flags & CLONE_PIDFD) {
2157                 /*
2158                  * - CLONE_DETACHED is blocked so that we can potentially
2159                  *   reuse it later for CLONE_PIDFD.
2160                  * - CLONE_THREAD is blocked until someone really needs it.
2161                  */
2162                 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
2163                         return ERR_PTR(-EINVAL);
2164         }
2165
2166         /*
2167          * Force any signals received before this point to be delivered
2168          * before the fork happens.  Collect up signals sent to multiple
2169          * processes that happen during the fork and delay them so that
2170          * they appear to happen after the fork.
2171          */
2172         sigemptyset(&delayed.signal);
2173         INIT_HLIST_NODE(&delayed.node);
2174
2175         spin_lock_irq(&current->sighand->siglock);
2176         if (!(clone_flags & CLONE_THREAD))
2177                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
2178         recalc_sigpending();
2179         spin_unlock_irq(&current->sighand->siglock);
2180         retval = -ERESTARTNOINTR;
2181         if (task_sigpending(current))
2182                 goto fork_out;
2183
2184         retval = -ENOMEM;
2185         p = dup_task_struct(current, node);
2186         if (!p)
2187                 goto fork_out;
2188         p->flags &= ~PF_KTHREAD;
2189         if (args->kthread)
2190                 p->flags |= PF_KTHREAD;
2191         if (args->io_thread) {
2192                 /*
2193                  * Mark us an IO worker, and block any signal that isn't
2194                  * fatal or STOP
2195                  */
2196                 p->flags |= PF_IO_WORKER;
2197                 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2198         }
2199
2200         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2201         /*
2202          * Clear TID on mm_release()?
2203          */
2204         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2205
2206         ftrace_graph_init_task(p);
2207
2208         rt_mutex_init_task(p);
2209
2210         lockdep_assert_irqs_enabled();
2211 #ifdef CONFIG_PROVE_LOCKING
2212         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2213 #endif
2214         retval = copy_creds(p, clone_flags);
2215         if (retval < 0)
2216                 goto bad_fork_free;
2217
2218         retval = -EAGAIN;
2219         if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2220                 if (p->real_cred->user != INIT_USER &&
2221                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2222                         goto bad_fork_cleanup_count;
2223         }
2224         current->flags &= ~PF_NPROC_EXCEEDED;
2225
2226         /*
2227          * If multiple threads are within copy_process(), then this check
2228          * triggers too late. This doesn't hurt, the check is only there
2229          * to stop root fork bombs.
2230          */
2231         retval = -EAGAIN;
2232         if (data_race(nr_threads >= max_threads))
2233                 goto bad_fork_cleanup_count;
2234
2235         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
2236         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2237         p->flags |= PF_FORKNOEXEC;
2238         INIT_LIST_HEAD(&p->children);
2239         INIT_LIST_HEAD(&p->sibling);
2240         rcu_copy_process(p);
2241         p->vfork_done = NULL;
2242         spin_lock_init(&p->alloc_lock);
2243
2244         init_sigpending(&p->pending);
2245
2246         p->utime = p->stime = p->gtime = 0;
2247 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2248         p->utimescaled = p->stimescaled = 0;
2249 #endif
2250         prev_cputime_init(&p->prev_cputime);
2251
2252 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2253         seqcount_init(&p->vtime.seqcount);
2254         p->vtime.starttime = 0;
2255         p->vtime.state = VTIME_INACTIVE;
2256 #endif
2257
2258 #ifdef CONFIG_IO_URING
2259         p->io_uring = NULL;
2260 #endif
2261
2262 #if defined(SPLIT_RSS_COUNTING)
2263         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2264 #endif
2265
2266         p->default_timer_slack_ns = current->timer_slack_ns;
2267
2268 #ifdef CONFIG_PSI
2269         p->psi_flags = 0;
2270 #endif
2271
2272         task_io_accounting_init(&p->ioac);
2273         acct_clear_integrals(p);
2274
2275         posix_cputimers_init(&p->posix_cputimers);
2276
2277         p->io_context = NULL;
2278         audit_set_context(p, NULL);
2279         cgroup_fork(p);
2280         if (args->kthread) {
2281                 if (!set_kthread_struct(p))
2282                         goto bad_fork_cleanup_delayacct;
2283         }
2284 #ifdef CONFIG_NUMA
2285         p->mempolicy = mpol_dup(p->mempolicy);
2286         if (IS_ERR(p->mempolicy)) {
2287                 retval = PTR_ERR(p->mempolicy);
2288                 p->mempolicy = NULL;
2289                 goto bad_fork_cleanup_delayacct;
2290         }
2291 #endif
2292 #ifdef CONFIG_CPUSETS
2293         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2294         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2295         seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2296 #endif
2297 #ifdef CONFIG_TRACE_IRQFLAGS
2298         memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2299         p->irqtrace.hardirq_disable_ip  = _THIS_IP_;
2300         p->irqtrace.softirq_enable_ip   = _THIS_IP_;
2301         p->softirqs_enabled             = 1;
2302         p->softirq_context              = 0;
2303 #endif
2304
2305         p->pagefault_disabled = 0;
2306
2307 #ifdef CONFIG_LOCKDEP
2308         lockdep_init_task(p);
2309 #endif
2310
2311 #ifdef CONFIG_DEBUG_MUTEXES
2312         p->blocked_on = NULL; /* not blocked yet */
2313 #endif
2314 #ifdef CONFIG_BCACHE
2315         p->sequential_io        = 0;
2316         p->sequential_io_avg    = 0;
2317 #endif
2318 #ifdef CONFIG_BPF_SYSCALL
2319         RCU_INIT_POINTER(p->bpf_storage, NULL);
2320         p->bpf_ctx = NULL;
2321 #endif
2322
2323         /* Perform scheduler related setup. Assign this task to a CPU. */
2324         retval = sched_fork(clone_flags, p);
2325         if (retval)
2326                 goto bad_fork_cleanup_policy;
2327
2328         retval = perf_event_init_task(p, clone_flags);
2329         if (retval)
2330                 goto bad_fork_cleanup_policy;
2331         retval = audit_alloc(p);
2332         if (retval)
2333                 goto bad_fork_cleanup_perf;
2334         /* copy all the process information */
2335         shm_init_task(p);
2336         retval = security_task_alloc(p, clone_flags);
2337         if (retval)
2338                 goto bad_fork_cleanup_audit;
2339         retval = copy_semundo(clone_flags, p);
2340         if (retval)
2341                 goto bad_fork_cleanup_security;
2342         retval = copy_files(clone_flags, p);
2343         if (retval)
2344                 goto bad_fork_cleanup_semundo;
2345         retval = copy_fs(clone_flags, p);
2346         if (retval)
2347                 goto bad_fork_cleanup_files;
2348         retval = copy_sighand(clone_flags, p);
2349         if (retval)
2350                 goto bad_fork_cleanup_fs;
2351         retval = copy_signal(clone_flags, p);
2352         if (retval)
2353                 goto bad_fork_cleanup_sighand;
2354         retval = copy_mm(clone_flags, p);
2355         if (retval)
2356                 goto bad_fork_cleanup_signal;
2357         retval = copy_namespaces(clone_flags, p);
2358         if (retval)
2359                 goto bad_fork_cleanup_mm;
2360         retval = copy_io(clone_flags, p);
2361         if (retval)
2362                 goto bad_fork_cleanup_namespaces;
2363         retval = copy_thread(p, args);
2364         if (retval)
2365                 goto bad_fork_cleanup_io;
2366
2367         stackleak_task_init(p);
2368
2369         if (pid != &init_struct_pid) {
2370                 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2371                                 args->set_tid_size);
2372                 if (IS_ERR(pid)) {
2373                         retval = PTR_ERR(pid);
2374                         goto bad_fork_cleanup_thread;
2375                 }
2376         }
2377
2378         /*
2379          * This has to happen after we've potentially unshared the file
2380          * descriptor table (so that the pidfd doesn't leak into the child
2381          * if the fd table isn't shared).
2382          */
2383         if (clone_flags & CLONE_PIDFD) {
2384                 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2385                 if (retval < 0)
2386                         goto bad_fork_free_pid;
2387
2388                 pidfd = retval;
2389
2390                 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2391                                               O_RDWR | O_CLOEXEC);
2392                 if (IS_ERR(pidfile)) {
2393                         put_unused_fd(pidfd);
2394                         retval = PTR_ERR(pidfile);
2395                         goto bad_fork_free_pid;
2396                 }
2397                 get_pid(pid);   /* held by pidfile now */
2398
2399                 retval = put_user(pidfd, args->pidfd);
2400                 if (retval)
2401                         goto bad_fork_put_pidfd;
2402         }
2403
2404 #ifdef CONFIG_BLOCK
2405         p->plug = NULL;
2406 #endif
2407         futex_init_task(p);
2408
2409         /*
2410          * sigaltstack should be cleared when sharing the same VM
2411          */
2412         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2413                 sas_ss_reset(p);
2414
2415         /*
2416          * Syscall tracing and stepping should be turned off in the
2417          * child regardless of CLONE_PTRACE.
2418          */
2419         user_disable_single_step(p);
2420         clear_task_syscall_work(p, SYSCALL_TRACE);
2421 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2422         clear_task_syscall_work(p, SYSCALL_EMU);
2423 #endif
2424         clear_tsk_latency_tracing(p);
2425
2426         /* ok, now we should be set up.. */
2427         p->pid = pid_nr(pid);
2428         if (clone_flags & CLONE_THREAD) {
2429                 p->group_leader = current->group_leader;
2430                 p->tgid = current->tgid;
2431         } else {
2432                 p->group_leader = p;
2433                 p->tgid = p->pid;
2434         }
2435
2436         p->nr_dirtied = 0;
2437         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2438         p->dirty_paused_when = 0;
2439
2440         p->pdeath_signal = 0;
2441         INIT_LIST_HEAD(&p->thread_group);
2442         p->task_works = NULL;
2443         clear_posix_cputimers_work(p);
2444
2445 #ifdef CONFIG_KRETPROBES
2446         p->kretprobe_instances.first = NULL;
2447 #endif
2448 #ifdef CONFIG_RETHOOK
2449         p->rethooks.first = NULL;
2450 #endif
2451
2452         /*
2453          * Ensure that the cgroup subsystem policies allow the new process to be
2454          * forked. It should be noted that the new process's css_set can be changed
2455          * between here and cgroup_post_fork() if an organisation operation is in
2456          * progress.
2457          */
2458         retval = cgroup_can_fork(p, args);
2459         if (retval)
2460                 goto bad_fork_put_pidfd;
2461
2462         /*
2463          * Now that the cgroups are pinned, re-clone the parent cgroup and put
2464          * the new task on the correct runqueue. All this *before* the task
2465          * becomes visible.
2466          *
2467          * This isn't part of ->can_fork() because while the re-cloning is
2468          * cgroup specific, it unconditionally needs to place the task on a
2469          * runqueue.
2470          */
2471         sched_cgroup_fork(p, args);
2472
2473         /*
2474          * From this point on we must avoid any synchronous user-space
2475          * communication until we take the tasklist-lock. In particular, we do
2476          * not want user-space to be able to predict the process start-time by
2477          * stalling fork(2) after we recorded the start_time but before it is
2478          * visible to the system.
2479          */
2480
2481         p->start_time = ktime_get_ns();
2482         p->start_boottime = ktime_get_boottime_ns();
2483
2484         /*
2485          * Make it visible to the rest of the system, but dont wake it up yet.
2486          * Need tasklist lock for parent etc handling!
2487          */
2488         write_lock_irq(&tasklist_lock);
2489
2490         /* CLONE_PARENT re-uses the old parent */
2491         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2492                 p->real_parent = current->real_parent;
2493                 p->parent_exec_id = current->parent_exec_id;
2494                 if (clone_flags & CLONE_THREAD)
2495                         p->exit_signal = -1;
2496                 else
2497                         p->exit_signal = current->group_leader->exit_signal;
2498         } else {
2499                 p->real_parent = current;
2500                 p->parent_exec_id = current->self_exec_id;
2501                 p->exit_signal = args->exit_signal;
2502         }
2503
2504         klp_copy_process(p);
2505
2506         sched_core_fork(p);
2507
2508         spin_lock(&current->sighand->siglock);
2509
2510         rv_task_fork(p);
2511
2512         rseq_fork(p, clone_flags);
2513
2514         /* Don't start children in a dying pid namespace */
2515         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2516                 retval = -ENOMEM;
2517                 goto bad_fork_cancel_cgroup;
2518         }
2519
2520         /* Let kill terminate clone/fork in the middle */
2521         if (fatal_signal_pending(current)) {
2522                 retval = -EINTR;
2523                 goto bad_fork_cancel_cgroup;
2524         }
2525
2526         /* No more failure paths after this point. */
2527
2528         /*
2529          * Copy seccomp details explicitly here, in case they were changed
2530          * before holding sighand lock.
2531          */
2532         copy_seccomp(p);
2533
2534         init_task_pid_links(p);
2535         if (likely(p->pid)) {
2536                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2537
2538                 init_task_pid(p, PIDTYPE_PID, pid);
2539                 if (thread_group_leader(p)) {
2540                         init_task_pid(p, PIDTYPE_TGID, pid);
2541                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2542                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2543
2544                         if (is_child_reaper(pid)) {
2545                                 ns_of_pid(pid)->child_reaper = p;
2546                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2547                         }
2548                         p->signal->shared_pending.signal = delayed.signal;
2549                         p->signal->tty = tty_kref_get(current->signal->tty);
2550                         /*
2551                          * Inherit has_child_subreaper flag under the same
2552                          * tasklist_lock with adding child to the process tree
2553                          * for propagate_has_child_subreaper optimization.
2554                          */
2555                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2556                                                          p->real_parent->signal->is_child_subreaper;
2557                         list_add_tail(&p->sibling, &p->real_parent->children);
2558                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2559                         attach_pid(p, PIDTYPE_TGID);
2560                         attach_pid(p, PIDTYPE_PGID);
2561                         attach_pid(p, PIDTYPE_SID);
2562                         __this_cpu_inc(process_counts);
2563                 } else {
2564                         current->signal->nr_threads++;
2565                         current->signal->quick_threads++;
2566                         atomic_inc(&current->signal->live);
2567                         refcount_inc(&current->signal->sigcnt);
2568                         task_join_group_stop(p);
2569                         list_add_tail_rcu(&p->thread_group,
2570                                           &p->group_leader->thread_group);
2571                         list_add_tail_rcu(&p->thread_node,
2572                                           &p->signal->thread_head);
2573                 }
2574                 attach_pid(p, PIDTYPE_PID);
2575                 nr_threads++;
2576         }
2577         total_forks++;
2578         hlist_del_init(&delayed.node);
2579         spin_unlock(&current->sighand->siglock);
2580         syscall_tracepoint_update(p);
2581         write_unlock_irq(&tasklist_lock);
2582
2583         if (pidfile)
2584                 fd_install(pidfd, pidfile);
2585
2586         proc_fork_connector(p);
2587         sched_post_fork(p);
2588         cgroup_post_fork(p, args);
2589         perf_event_fork(p);
2590
2591         trace_task_newtask(p, clone_flags);
2592         uprobe_copy_process(p, clone_flags);
2593
2594         copy_oom_score_adj(clone_flags, p);
2595
2596         return p;
2597
2598 bad_fork_cancel_cgroup:
2599         sched_core_free(p);
2600         spin_unlock(&current->sighand->siglock);
2601         write_unlock_irq(&tasklist_lock);
2602         cgroup_cancel_fork(p, args);
2603 bad_fork_put_pidfd:
2604         if (clone_flags & CLONE_PIDFD) {
2605                 fput(pidfile);
2606                 put_unused_fd(pidfd);
2607         }
2608 bad_fork_free_pid:
2609         if (pid != &init_struct_pid)
2610                 free_pid(pid);
2611 bad_fork_cleanup_thread:
2612         exit_thread(p);
2613 bad_fork_cleanup_io:
2614         if (p->io_context)
2615                 exit_io_context(p);
2616 bad_fork_cleanup_namespaces:
2617         exit_task_namespaces(p);
2618 bad_fork_cleanup_mm:
2619         if (p->mm) {
2620                 mm_clear_owner(p->mm, p);
2621                 mmput(p->mm);
2622         }
2623 bad_fork_cleanup_signal:
2624         if (!(clone_flags & CLONE_THREAD))
2625                 free_signal_struct(p->signal);
2626 bad_fork_cleanup_sighand:
2627         __cleanup_sighand(p->sighand);
2628 bad_fork_cleanup_fs:
2629         exit_fs(p); /* blocking */
2630 bad_fork_cleanup_files:
2631         exit_files(p); /* blocking */
2632 bad_fork_cleanup_semundo:
2633         exit_sem(p);
2634 bad_fork_cleanup_security:
2635         security_task_free(p);
2636 bad_fork_cleanup_audit:
2637         audit_free(p);
2638 bad_fork_cleanup_perf:
2639         perf_event_free_task(p);
2640 bad_fork_cleanup_policy:
2641         lockdep_free_task(p);
2642 #ifdef CONFIG_NUMA
2643         mpol_put(p->mempolicy);
2644 #endif
2645 bad_fork_cleanup_delayacct:
2646         delayacct_tsk_free(p);
2647 bad_fork_cleanup_count:
2648         dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2649         exit_creds(p);
2650 bad_fork_free:
2651         WRITE_ONCE(p->__state, TASK_DEAD);
2652         exit_task_stack_account(p);
2653         put_task_stack(p);
2654         delayed_free_task(p);
2655 fork_out:
2656         spin_lock_irq(&current->sighand->siglock);
2657         hlist_del_init(&delayed.node);
2658         spin_unlock_irq(&current->sighand->siglock);
2659         return ERR_PTR(retval);
2660 }
2661
2662 static inline void init_idle_pids(struct task_struct *idle)
2663 {
2664         enum pid_type type;
2665
2666         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2667                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2668                 init_task_pid(idle, type, &init_struct_pid);
2669         }
2670 }
2671
2672 static int idle_dummy(void *dummy)
2673 {
2674         /* This function is never called */
2675         return 0;
2676 }
2677
2678 struct task_struct * __init fork_idle(int cpu)
2679 {
2680         struct task_struct *task;
2681         struct kernel_clone_args args = {
2682                 .flags          = CLONE_VM,
2683                 .fn             = &idle_dummy,
2684                 .fn_arg         = NULL,
2685                 .kthread        = 1,
2686                 .idle           = 1,
2687         };
2688
2689         task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2690         if (!IS_ERR(task)) {
2691                 init_idle_pids(task);
2692                 init_idle(task, cpu);
2693         }
2694
2695         return task;
2696 }
2697
2698 /*
2699  * This is like kernel_clone(), but shaved down and tailored to just
2700  * creating io_uring workers. It returns a created task, or an error pointer.
2701  * The returned task is inactive, and the caller must fire it up through
2702  * wake_up_new_task(p). All signals are blocked in the created task.
2703  */
2704 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2705 {
2706         unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2707                                 CLONE_IO;
2708         struct kernel_clone_args args = {
2709                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2710                                     CLONE_UNTRACED) & ~CSIGNAL),
2711                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2712                 .fn             = fn,
2713                 .fn_arg         = arg,
2714                 .io_thread      = 1,
2715         };
2716
2717         return copy_process(NULL, 0, node, &args);
2718 }
2719
2720 /*
2721  *  Ok, this is the main fork-routine.
2722  *
2723  * It copies the process, and if successful kick-starts
2724  * it and waits for it to finish using the VM if required.
2725  *
2726  * args->exit_signal is expected to be checked for sanity by the caller.
2727  */
2728 pid_t kernel_clone(struct kernel_clone_args *args)
2729 {
2730         u64 clone_flags = args->flags;
2731         struct completion vfork;
2732         struct pid *pid;
2733         struct task_struct *p;
2734         int trace = 0;
2735         pid_t nr;
2736
2737         /*
2738          * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2739          * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2740          * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2741          * field in struct clone_args and it still doesn't make sense to have
2742          * them both point at the same memory location. Performing this check
2743          * here has the advantage that we don't need to have a separate helper
2744          * to check for legacy clone().
2745          */
2746         if ((args->flags & CLONE_PIDFD) &&
2747             (args->flags & CLONE_PARENT_SETTID) &&
2748             (args->pidfd == args->parent_tid))
2749                 return -EINVAL;
2750
2751         /*
2752          * Determine whether and which event to report to ptracer.  When
2753          * called from kernel_thread or CLONE_UNTRACED is explicitly
2754          * requested, no event is reported; otherwise, report if the event
2755          * for the type of forking is enabled.
2756          */
2757         if (!(clone_flags & CLONE_UNTRACED)) {
2758                 if (clone_flags & CLONE_VFORK)
2759                         trace = PTRACE_EVENT_VFORK;
2760                 else if (args->exit_signal != SIGCHLD)
2761                         trace = PTRACE_EVENT_CLONE;
2762                 else
2763                         trace = PTRACE_EVENT_FORK;
2764
2765                 if (likely(!ptrace_event_enabled(current, trace)))
2766                         trace = 0;
2767         }
2768
2769         p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2770         add_latent_entropy();
2771
2772         if (IS_ERR(p))
2773                 return PTR_ERR(p);
2774
2775         /*
2776          * Do this prior waking up the new thread - the thread pointer
2777          * might get invalid after that point, if the thread exits quickly.
2778          */
2779         trace_sched_process_fork(current, p);
2780
2781         pid = get_task_pid(p, PIDTYPE_PID);
2782         nr = pid_vnr(pid);
2783
2784         if (clone_flags & CLONE_PARENT_SETTID)
2785                 put_user(nr, args->parent_tid);
2786
2787         if (clone_flags & CLONE_VFORK) {
2788                 p->vfork_done = &vfork;
2789                 init_completion(&vfork);
2790                 get_task_struct(p);
2791         }
2792
2793         if (IS_ENABLED(CONFIG_LRU_GEN) && !(clone_flags & CLONE_VM)) {
2794                 /* lock the task to synchronize with memcg migration */
2795                 task_lock(p);
2796                 lru_gen_add_mm(p->mm);
2797                 task_unlock(p);
2798         }
2799
2800         wake_up_new_task(p);
2801
2802         /* forking complete and child started to run, tell ptracer */
2803         if (unlikely(trace))
2804                 ptrace_event_pid(trace, pid);
2805
2806         if (clone_flags & CLONE_VFORK) {
2807                 if (!wait_for_vfork_done(p, &vfork))
2808                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2809         }
2810
2811         put_pid(pid);
2812         return nr;
2813 }
2814
2815 /*
2816  * Create a kernel thread.
2817  */
2818 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2819 {
2820         struct kernel_clone_args args = {
2821                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2822                                     CLONE_UNTRACED) & ~CSIGNAL),
2823                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2824                 .fn             = fn,
2825                 .fn_arg         = arg,
2826                 .kthread        = 1,
2827         };
2828
2829         return kernel_clone(&args);
2830 }
2831
2832 /*
2833  * Create a user mode thread.
2834  */
2835 pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2836 {
2837         struct kernel_clone_args args = {
2838                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2839                                     CLONE_UNTRACED) & ~CSIGNAL),
2840                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2841                 .fn             = fn,
2842                 .fn_arg         = arg,
2843         };
2844
2845         return kernel_clone(&args);
2846 }
2847
2848 #ifdef __ARCH_WANT_SYS_FORK
2849 SYSCALL_DEFINE0(fork)
2850 {
2851 #ifdef CONFIG_MMU
2852         struct kernel_clone_args args = {
2853                 .exit_signal = SIGCHLD,
2854         };
2855
2856         return kernel_clone(&args);
2857 #else
2858         /* can not support in nommu mode */
2859         return -EINVAL;
2860 #endif
2861 }
2862 #endif
2863
2864 #ifdef __ARCH_WANT_SYS_VFORK
2865 SYSCALL_DEFINE0(vfork)
2866 {
2867         struct kernel_clone_args args = {
2868                 .flags          = CLONE_VFORK | CLONE_VM,
2869                 .exit_signal    = SIGCHLD,
2870         };
2871
2872         return kernel_clone(&args);
2873 }
2874 #endif
2875
2876 #ifdef __ARCH_WANT_SYS_CLONE
2877 #ifdef CONFIG_CLONE_BACKWARDS
2878 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2879                  int __user *, parent_tidptr,
2880                  unsigned long, tls,
2881                  int __user *, child_tidptr)
2882 #elif defined(CONFIG_CLONE_BACKWARDS2)
2883 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2884                  int __user *, parent_tidptr,
2885                  int __user *, child_tidptr,
2886                  unsigned long, tls)
2887 #elif defined(CONFIG_CLONE_BACKWARDS3)
2888 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2889                 int, stack_size,
2890                 int __user *, parent_tidptr,
2891                 int __user *, child_tidptr,
2892                 unsigned long, tls)
2893 #else
2894 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2895                  int __user *, parent_tidptr,
2896                  int __user *, child_tidptr,
2897                  unsigned long, tls)
2898 #endif
2899 {
2900         struct kernel_clone_args args = {
2901                 .flags          = (lower_32_bits(clone_flags) & ~CSIGNAL),
2902                 .pidfd          = parent_tidptr,
2903                 .child_tid      = child_tidptr,
2904                 .parent_tid     = parent_tidptr,
2905                 .exit_signal    = (lower_32_bits(clone_flags) & CSIGNAL),
2906                 .stack          = newsp,
2907                 .tls            = tls,
2908         };
2909
2910         return kernel_clone(&args);
2911 }
2912 #endif
2913
2914 #ifdef __ARCH_WANT_SYS_CLONE3
2915
2916 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2917                                               struct clone_args __user *uargs,
2918                                               size_t usize)
2919 {
2920         int err;
2921         struct clone_args args;
2922         pid_t *kset_tid = kargs->set_tid;
2923
2924         BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2925                      CLONE_ARGS_SIZE_VER0);
2926         BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2927                      CLONE_ARGS_SIZE_VER1);
2928         BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2929                      CLONE_ARGS_SIZE_VER2);
2930         BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2931
2932         if (unlikely(usize > PAGE_SIZE))
2933                 return -E2BIG;
2934         if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2935                 return -EINVAL;
2936
2937         err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2938         if (err)
2939                 return err;
2940
2941         if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2942                 return -EINVAL;
2943
2944         if (unlikely(!args.set_tid && args.set_tid_size > 0))
2945                 return -EINVAL;
2946
2947         if (unlikely(args.set_tid && args.set_tid_size == 0))
2948                 return -EINVAL;
2949
2950         /*
2951          * Verify that higher 32bits of exit_signal are unset and that
2952          * it is a valid signal
2953          */
2954         if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2955                      !valid_signal(args.exit_signal)))
2956                 return -EINVAL;
2957
2958         if ((args.flags & CLONE_INTO_CGROUP) &&
2959             (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2960                 return -EINVAL;
2961
2962         *kargs = (struct kernel_clone_args){
2963                 .flags          = args.flags,
2964                 .pidfd          = u64_to_user_ptr(args.pidfd),
2965                 .child_tid      = u64_to_user_ptr(args.child_tid),
2966                 .parent_tid     = u64_to_user_ptr(args.parent_tid),
2967                 .exit_signal    = args.exit_signal,
2968                 .stack          = args.stack,
2969                 .stack_size     = args.stack_size,
2970                 .tls            = args.tls,
2971                 .set_tid_size   = args.set_tid_size,
2972                 .cgroup         = args.cgroup,
2973         };
2974
2975         if (args.set_tid &&
2976                 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2977                         (kargs->set_tid_size * sizeof(pid_t))))
2978                 return -EFAULT;
2979
2980         kargs->set_tid = kset_tid;
2981
2982         return 0;
2983 }
2984
2985 /**
2986  * clone3_stack_valid - check and prepare stack
2987  * @kargs: kernel clone args
2988  *
2989  * Verify that the stack arguments userspace gave us are sane.
2990  * In addition, set the stack direction for userspace since it's easy for us to
2991  * determine.
2992  */
2993 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2994 {
2995         if (kargs->stack == 0) {
2996                 if (kargs->stack_size > 0)
2997                         return false;
2998         } else {
2999                 if (kargs->stack_size == 0)
3000                         return false;
3001
3002                 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
3003                         return false;
3004
3005 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
3006                 kargs->stack += kargs->stack_size;
3007 #endif
3008         }
3009
3010         return true;
3011 }
3012
3013 static bool clone3_args_valid(struct kernel_clone_args *kargs)
3014 {
3015         /* Verify that no unknown flags are passed along. */
3016         if (kargs->flags &
3017             ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
3018                 return false;
3019
3020         /*
3021          * - make the CLONE_DETACHED bit reusable for clone3
3022          * - make the CSIGNAL bits reusable for clone3
3023          */
3024         if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
3025                 return false;
3026
3027         if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
3028             (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
3029                 return false;
3030
3031         if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
3032             kargs->exit_signal)
3033                 return false;
3034
3035         if (!clone3_stack_valid(kargs))
3036                 return false;
3037
3038         return true;
3039 }
3040
3041 /**
3042  * clone3 - create a new process with specific properties
3043  * @uargs: argument structure
3044  * @size:  size of @uargs
3045  *
3046  * clone3() is the extensible successor to clone()/clone2().
3047  * It takes a struct as argument that is versioned by its size.
3048  *
3049  * Return: On success, a positive PID for the child process.
3050  *         On error, a negative errno number.
3051  */
3052 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
3053 {
3054         int err;
3055
3056         struct kernel_clone_args kargs;
3057         pid_t set_tid[MAX_PID_NS_LEVEL];
3058
3059         kargs.set_tid = set_tid;
3060
3061         err = copy_clone_args_from_user(&kargs, uargs, size);
3062         if (err)
3063                 return err;
3064
3065         if (!clone3_args_valid(&kargs))
3066                 return -EINVAL;
3067
3068         return kernel_clone(&kargs);
3069 }
3070 #endif
3071
3072 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
3073 {
3074         struct task_struct *leader, *parent, *child;
3075         int res;
3076
3077         read_lock(&tasklist_lock);
3078         leader = top = top->group_leader;
3079 down:
3080         for_each_thread(leader, parent) {
3081                 list_for_each_entry(child, &parent->children, sibling) {
3082                         res = visitor(child, data);
3083                         if (res) {
3084                                 if (res < 0)
3085                                         goto out;
3086                                 leader = child;
3087                                 goto down;
3088                         }
3089 up:
3090                         ;
3091                 }
3092         }
3093
3094         if (leader != top) {
3095                 child = leader;
3096                 parent = child->real_parent;
3097                 leader = parent->group_leader;
3098                 goto up;
3099         }
3100 out:
3101         read_unlock(&tasklist_lock);
3102 }
3103
3104 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
3105 #define ARCH_MIN_MMSTRUCT_ALIGN 0
3106 #endif
3107
3108 static void sighand_ctor(void *data)
3109 {
3110         struct sighand_struct *sighand = data;
3111
3112         spin_lock_init(&sighand->siglock);
3113         init_waitqueue_head(&sighand->signalfd_wqh);
3114 }
3115
3116 void __init mm_cache_init(void)
3117 {
3118         unsigned int mm_size;
3119
3120         /*
3121          * The mm_cpumask is located at the end of mm_struct, and is
3122          * dynamically sized based on the maximum CPU number this system
3123          * can have, taking hotplug into account (nr_cpu_ids).
3124          */
3125         mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
3126
3127         mm_cachep = kmem_cache_create_usercopy("mm_struct",
3128                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3129                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3130                         offsetof(struct mm_struct, saved_auxv),
3131                         sizeof_field(struct mm_struct, saved_auxv),
3132                         NULL);
3133 }
3134
3135 void __init proc_caches_init(void)
3136 {
3137         sighand_cachep = kmem_cache_create("sighand_cache",
3138                         sizeof(struct sighand_struct), 0,
3139                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3140                         SLAB_ACCOUNT, sighand_ctor);
3141         signal_cachep = kmem_cache_create("signal_cache",
3142                         sizeof(struct signal_struct), 0,
3143                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3144                         NULL);
3145         files_cachep = kmem_cache_create("files_cache",
3146                         sizeof(struct files_struct), 0,
3147                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3148                         NULL);
3149         fs_cachep = kmem_cache_create("fs_cache",
3150                         sizeof(struct fs_struct), 0,
3151                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3152                         NULL);
3153
3154         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3155         mmap_init();
3156         nsproxy_cache_init();
3157 }
3158
3159 /*
3160  * Check constraints on flags passed to the unshare system call.
3161  */
3162 static int check_unshare_flags(unsigned long unshare_flags)
3163 {
3164         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3165                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3166                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3167                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3168                                 CLONE_NEWTIME))
3169                 return -EINVAL;
3170         /*
3171          * Not implemented, but pretend it works if there is nothing
3172          * to unshare.  Note that unsharing the address space or the
3173          * signal handlers also need to unshare the signal queues (aka
3174          * CLONE_THREAD).
3175          */
3176         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3177                 if (!thread_group_empty(current))
3178                         return -EINVAL;
3179         }
3180         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3181                 if (refcount_read(&current->sighand->count) > 1)
3182                         return -EINVAL;
3183         }
3184         if (unshare_flags & CLONE_VM) {
3185                 if (!current_is_single_threaded())
3186                         return -EINVAL;
3187         }
3188
3189         return 0;
3190 }
3191
3192 /*
3193  * Unshare the filesystem structure if it is being shared
3194  */
3195 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3196 {
3197         struct fs_struct *fs = current->fs;
3198
3199         if (!(unshare_flags & CLONE_FS) || !fs)
3200                 return 0;
3201
3202         /* don't need lock here; in the worst case we'll do useless copy */
3203         if (fs->users == 1)
3204                 return 0;
3205
3206         *new_fsp = copy_fs_struct(fs);
3207         if (!*new_fsp)
3208                 return -ENOMEM;
3209
3210         return 0;
3211 }
3212
3213 /*
3214  * Unshare file descriptor table if it is being shared
3215  */
3216 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3217                struct files_struct **new_fdp)
3218 {
3219         struct files_struct *fd = current->files;
3220         int error = 0;
3221
3222         if ((unshare_flags & CLONE_FILES) &&
3223             (fd && atomic_read(&fd->count) > 1)) {
3224                 *new_fdp = dup_fd(fd, max_fds, &error);
3225                 if (!*new_fdp)
3226                         return error;
3227         }
3228
3229         return 0;
3230 }
3231
3232 /*
3233  * unshare allows a process to 'unshare' part of the process
3234  * context which was originally shared using clone.  copy_*
3235  * functions used by kernel_clone() cannot be used here directly
3236  * because they modify an inactive task_struct that is being
3237  * constructed. Here we are modifying the current, active,
3238  * task_struct.
3239  */
3240 int ksys_unshare(unsigned long unshare_flags)
3241 {
3242         struct fs_struct *fs, *new_fs = NULL;
3243         struct files_struct *new_fd = NULL;
3244         struct cred *new_cred = NULL;
3245         struct nsproxy *new_nsproxy = NULL;
3246         int do_sysvsem = 0;
3247         int err;
3248
3249         /*
3250          * If unsharing a user namespace must also unshare the thread group
3251          * and unshare the filesystem root and working directories.
3252          */
3253         if (unshare_flags & CLONE_NEWUSER)
3254                 unshare_flags |= CLONE_THREAD | CLONE_FS;
3255         /*
3256          * If unsharing vm, must also unshare signal handlers.
3257          */
3258         if (unshare_flags & CLONE_VM)
3259                 unshare_flags |= CLONE_SIGHAND;
3260         /*
3261          * If unsharing a signal handlers, must also unshare the signal queues.
3262          */
3263         if (unshare_flags & CLONE_SIGHAND)
3264                 unshare_flags |= CLONE_THREAD;
3265         /*
3266          * If unsharing namespace, must also unshare filesystem information.
3267          */
3268         if (unshare_flags & CLONE_NEWNS)
3269                 unshare_flags |= CLONE_FS;
3270
3271         err = check_unshare_flags(unshare_flags);
3272         if (err)
3273                 goto bad_unshare_out;
3274         /*
3275          * CLONE_NEWIPC must also detach from the undolist: after switching
3276          * to a new ipc namespace, the semaphore arrays from the old
3277          * namespace are unreachable.
3278          */
3279         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3280                 do_sysvsem = 1;
3281         err = unshare_fs(unshare_flags, &new_fs);
3282         if (err)
3283                 goto bad_unshare_out;
3284         err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3285         if (err)
3286                 goto bad_unshare_cleanup_fs;
3287         err = unshare_userns(unshare_flags, &new_cred);
3288         if (err)
3289                 goto bad_unshare_cleanup_fd;
3290         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3291                                          new_cred, new_fs);
3292         if (err)
3293                 goto bad_unshare_cleanup_cred;
3294
3295         if (new_cred) {
3296                 err = set_cred_ucounts(new_cred);
3297                 if (err)
3298                         goto bad_unshare_cleanup_cred;
3299         }
3300
3301         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3302                 if (do_sysvsem) {
3303                         /*
3304                          * CLONE_SYSVSEM is equivalent to sys_exit().
3305                          */
3306                         exit_sem(current);
3307                 }
3308                 if (unshare_flags & CLONE_NEWIPC) {
3309                         /* Orphan segments in old ns (see sem above). */
3310                         exit_shm(current);
3311                         shm_init_task(current);
3312                 }
3313
3314                 if (new_nsproxy)
3315                         switch_task_namespaces(current, new_nsproxy);
3316
3317                 task_lock(current);
3318
3319                 if (new_fs) {
3320                         fs = current->fs;
3321                         spin_lock(&fs->lock);
3322                         current->fs = new_fs;
3323                         if (--fs->users)
3324                                 new_fs = NULL;
3325                         else
3326                                 new_fs = fs;
3327                         spin_unlock(&fs->lock);
3328                 }
3329
3330                 if (new_fd)
3331                         swap(current->files, new_fd);
3332
3333                 task_unlock(current);
3334
3335                 if (new_cred) {
3336                         /* Install the new user namespace */
3337                         commit_creds(new_cred);
3338                         new_cred = NULL;
3339                 }
3340         }
3341
3342         perf_event_namespaces(current);
3343
3344 bad_unshare_cleanup_cred:
3345         if (new_cred)
3346                 put_cred(new_cred);
3347 bad_unshare_cleanup_fd:
3348         if (new_fd)
3349                 put_files_struct(new_fd);
3350
3351 bad_unshare_cleanup_fs:
3352         if (new_fs)
3353                 free_fs_struct(new_fs);
3354
3355 bad_unshare_out:
3356         return err;
3357 }
3358
3359 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3360 {
3361         return ksys_unshare(unshare_flags);
3362 }
3363
3364 /*
3365  *      Helper to unshare the files of the current task.
3366  *      We don't want to expose copy_files internals to
3367  *      the exec layer of the kernel.
3368  */
3369
3370 int unshare_files(void)
3371 {
3372         struct task_struct *task = current;
3373         struct files_struct *old, *copy = NULL;
3374         int error;
3375
3376         error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, &copy);
3377         if (error || !copy)
3378                 return error;
3379
3380         old = task->files;
3381         task_lock(task);
3382         task->files = copy;
3383         task_unlock(task);
3384         put_files_struct(old);
3385         return 0;
3386 }
3387
3388 int sysctl_max_threads(struct ctl_table *table, int write,
3389                        void *buffer, size_t *lenp, loff_t *ppos)
3390 {
3391         struct ctl_table t;
3392         int ret;
3393         int threads = max_threads;
3394         int min = 1;
3395         int max = MAX_THREADS;
3396
3397         t = *table;
3398         t.data = &threads;
3399         t.extra1 = &min;
3400         t.extra2 = &max;
3401
3402         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3403         if (ret || !write)
3404                 return ret;
3405
3406         max_threads = threads;
3407
3408         return 0;
3409 }