kernel/fork.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/kernel/fork.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  */
   7
   8 /*
   9  *  'fork.c' contains the help-routines for the 'fork' system call
  10  * (see also entry.S and others).
  11  * Fork is rather simple, once you get the hang of it, but the memory
  12  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  13  */
  14
  15 #include <linux/anon_inodes.h>
  16 #include <linux/slab.h>
  17 #include <linux/sched/autogroup.h>
  18 #include <linux/sched/mm.h>
  19 #include <linux/sched/coredump.h>
  20 #include <linux/sched/user.h>
  21 #include <linux/sched/numa_balancing.h>
  22 #include <linux/sched/stat.h>
  23 #include <linux/sched/task.h>
  24 #include <linux/sched/task_stack.h>
  25 #include <linux/sched/cputime.h>
  26 #include <linux/seq_file.h>
  27 #include <linux/rtmutex.h>
  28 #include <linux/init.h>
  29 #include <linux/unistd.h>
  30 #include <linux/module.h>
  31 #include <linux/vmalloc.h>
  32 #include <linux/completion.h>
  33 #include <linux/personality.h>
  34 #include <linux/mempolicy.h>
  35 #include <linux/sem.h>
  36 #include <linux/file.h>
  37 #include <linux/fdtable.h>
  38 #include <linux/iocontext.h>
  39 #include <linux/key.h>
  40 #include <linux/binfmts.h>
  41 #include <linux/mman.h>
  42 #include <linux/mmu_notifier.h>
  43 #include <linux/fs.h>
  44 #include <linux/mm.h>
  45 #include <linux/mm_inline.h>
  46 #include <linux/vmacache.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/random.h>
  79 #include <linux/tty.h>
  80 #include <linux/fs_struct.h>
  81 #include <linux/magic.h>
  82 #include <linux/perf_event.h>
  83 #include <linux/posix-timers.h>
  84 #include <linux/user-return-notifier.h>
  85 #include <linux/oom.h>
  86 #include <linux/khugepaged.h>
  87 #include <linux/signalfd.h>
  88 #include <linux/uprobes.h>
  89 #include <linux/aio.h>
  90 #include <linux/compiler.h>
  91 #include <linux/sysctl.h>
  92 #include <linux/kcov.h>
  93 #include <linux/livepatch.h>
  94 #include <linux/thread_info.h>
  95 #include <linux/stackleak.h>
  96 #include <linux/kasan.h>
  97 #include <linux/scs.h>
  98 #include <linux/io_uring.h>
  99 #include <linux/bpf.h>
 100 #include <linux/sched/mm.h>
 101
 102 #include <asm/pgalloc.h>
 103 #include <linux/uaccess.h>
 104 #include <asm/mmu_context.h>
 105 #include <asm/cacheflush.h>
 106 #include <asm/tlbflush.h>
 107
 108 #include <trace/events/sched.h>
 109
 110 #define CREATE_TRACE_POINTS
 111 #include <trace/events/task.h>
 112
 113 /*
 114  * Minimum number of threads to boot the kernel
 115  */
 116 #define MIN_THREADS 20
 117
 118 /*
 119  * Maximum number of threads
 120  */
 121 #define MAX_THREADS FUTEX_TID_MASK
 122
 123 /*
 124  * Protected counters by write_lock_irq(&tasklist_lock)
 125  */
 126 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 127 int nr_threads;                 /* The idle threads do not count.. */
 128
 129 static int max_threads;         /* tunable limit on nr_threads */
 130
 131 #define NAMED_ARRAY_INDEX(x)    [x] = __stringify(x)
 132
 133 static const char * const resident_page_types[] = {
 134         NAMED_ARRAY_INDEX(MM_FILEPAGES),
 135         NAMED_ARRAY_INDEX(MM_ANONPAGES),
 136         NAMED_ARRAY_INDEX(MM_SWAPENTS),
 137         NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
 138 };
 139
 140 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 141
 142 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 143
 144 #ifdef CONFIG_PROVE_RCU
 145 int lockdep_tasklist_lock_is_held(void)
 146 {
 147         return lockdep_is_held(&tasklist_lock);
 148 }
 149 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 150 #endif /* #ifdef CONFIG_PROVE_RCU */
 151
 152 int nr_processes(void)
 153 {
 154         int cpu;
 155         int total = 0;
 156
 157         for_each_possible_cpu(cpu)
 158                 total += per_cpu(process_counts, cpu);
 159
 160         return total;
 161 }
 162
 163 void __weak arch_release_task_struct(struct task_struct *tsk)
 164 {
 165 }
 166
 167 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 168 static struct kmem_cache *task_struct_cachep;
 169
 170 static inline struct task_struct *alloc_task_struct_node(int node)
 171 {
 172         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 173 }
 174
 175 static inline void free_task_struct(struct task_struct *tsk)
 176 {
 177         kmem_cache_free(task_struct_cachep, tsk);
 178 }
 179 #endif
 180
 181 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 182
 183 /*
 184  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 185  * kmemcache based allocator.
 186  */
 187 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 188
 189 #  ifdef CONFIG_VMAP_STACK
 190 /*
 191  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 192  * flush.  Try to minimize the number of calls by caching stacks.
 193  */
 194 #define NR_CACHED_STACKS 2
 195 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 196
 197 struct vm_stack {
 198         struct rcu_head rcu;
 199         struct vm_struct *stack_vm_area;
 200 };
 201
 202 static bool try_release_thread_stack_to_cache(struct vm_struct *vm)
 203 {
 204         unsigned int i;
 205
 206         for (i = 0; i < NR_CACHED_STACKS; i++) {
 207                 if (this_cpu_cmpxchg(cached_stacks[i], NULL, vm) != NULL)
 208                         continue;
 209                 return true;
 210         }
 211         return false;
 212 }
 213
 214 static void thread_stack_free_rcu(struct rcu_head *rh)
 215 {
 216         struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
 217
 218         if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
 219                 return;
 220
 221         vfree(vm_stack);
 222 }
 223
 224 static void thread_stack_delayed_free(struct task_struct *tsk)
 225 {
 226         struct vm_stack *vm_stack = tsk->stack;
 227
 228         vm_stack->stack_vm_area = tsk->stack_vm_area;
 229         call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
 230 }
 231
 232 static int free_vm_stack_cache(unsigned int cpu)
 233 {
 234         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 235         int i;
 236
 237         for (i = 0; i < NR_CACHED_STACKS; i++) {
 238                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 239
 240                 if (!vm_stack)
 241                         continue;
 242
 243                 vfree(vm_stack->addr);
 244                 cached_vm_stacks[i] = NULL;
 245         }
 246
 247         return 0;
 248 }
 249
 250 static int memcg_charge_kernel_stack(struct vm_struct *vm)
 251 {
 252         int i;
 253         int ret;
 254
 255         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 256         BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 257
 258         for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 259                 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0);
 260                 if (ret)
 261                         goto err;
 262         }
 263         return 0;
 264 err:
 265         /*
 266          * If memcg_kmem_charge_page() fails, page's memory cgroup pointer is
 267          * NULL, and memcg_kmem_uncharge_page() in free_thread_stack() will
 268          * ignore this page.
 269          */
 270         for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 271                 memcg_kmem_uncharge_page(vm->pages[i], 0);
 272         return ret;
 273 }
 274
 275 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 276 {
 277         struct vm_struct *vm;
 278         void *stack;
 279         int i;
 280
 281         for (i = 0; i < NR_CACHED_STACKS; i++) {
 282                 struct vm_struct *s;
 283
 284                 s = this_cpu_xchg(cached_stacks[i], NULL);
 285
 286                 if (!s)
 287                         continue;
 288
 289                 /* Mark stack accessible for KASAN. */
 290                 kasan_unpoison_range(s->addr, THREAD_SIZE);
 291
 292                 /* Clear stale pointers from reused stack. */
 293                 memset(s->addr, 0, THREAD_SIZE);
 294
 295                 if (memcg_charge_kernel_stack(s)) {
 296                         vfree(s->addr);
 297                         return -ENOMEM;
 298                 }
 299
 300                 tsk->stack_vm_area = s;
 301                 tsk->stack = s->addr;
 302                 return 0;
 303         }
 304
 305         /*
 306          * Allocated stacks are cached and later reused by new threads,
 307          * so memcg accounting is performed manually on assigning/releasing
 308          * stacks to tasks. Drop __GFP_ACCOUNT.
 309          */
 310         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 311                                      VMALLOC_START, VMALLOC_END,
 312                                      THREADINFO_GFP & ~__GFP_ACCOUNT,
 313                                      PAGE_KERNEL,
 314                                      0, node, __builtin_return_address(0));
 315         if (!stack)
 316                 return -ENOMEM;
 317
 318         vm = find_vm_area(stack);
 319         if (memcg_charge_kernel_stack(vm)) {
 320                 vfree(stack);
 321                 return -ENOMEM;
 322         }
 323         /*
 324          * We can't call find_vm_area() in interrupt context, and
 325          * free_thread_stack() can be called in interrupt context,
 326          * so cache the vm_struct.
 327          */
 328         tsk->stack_vm_area = vm;
 329         tsk->stack = stack;
 330         return 0;
 331 }
 332
 333 static void free_thread_stack(struct task_struct *tsk)
 334 {
 335         if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
 336                 thread_stack_delayed_free(tsk);
 337
 338         tsk->stack = NULL;
 339         tsk->stack_vm_area = NULL;
 340 }
 341
 342 #  else /* !CONFIG_VMAP_STACK */
 343
 344 static void thread_stack_free_rcu(struct rcu_head *rh)
 345 {
 346         __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
 347 }
 348
 349 static void thread_stack_delayed_free(struct task_struct *tsk)
 350 {
 351         struct rcu_head *rh = tsk->stack;
 352
 353         call_rcu(rh, thread_stack_free_rcu);
 354 }
 355
 356 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 357 {
 358         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 359                                              THREAD_SIZE_ORDER);
 360
 361         if (likely(page)) {
 362                 tsk->stack = kasan_reset_tag(page_address(page));
 363                 return 0;
 364         }
 365         return -ENOMEM;
 366 }
 367
 368 static void free_thread_stack(struct task_struct *tsk)
 369 {
 370         thread_stack_delayed_free(tsk);
 371         tsk->stack = NULL;
 372 }
 373
 374 #  endif /* CONFIG_VMAP_STACK */
 375 # else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
 376
 377 static struct kmem_cache *thread_stack_cache;
 378
 379 static void thread_stack_free_rcu(struct rcu_head *rh)
 380 {
 381         kmem_cache_free(thread_stack_cache, rh);
 382 }
 383
 384 static void thread_stack_delayed_free(struct task_struct *tsk)
 385 {
 386         struct rcu_head *rh = tsk->stack;
 387
 388         call_rcu(rh, thread_stack_free_rcu);
 389 }
 390
 391 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 392 {
 393         unsigned long *stack;
 394         stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 395         stack = kasan_reset_tag(stack);
 396         tsk->stack = stack;
 397         return stack ? 0 : -ENOMEM;
 398 }
 399
 400 static void free_thread_stack(struct task_struct *tsk)
 401 {
 402         thread_stack_delayed_free(tsk);
 403         tsk->stack = NULL;
 404 }
 405
 406 void thread_stack_cache_init(void)
 407 {
 408         thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 409                                         THREAD_SIZE, THREAD_SIZE, 0, 0,
 410                                         THREAD_SIZE, NULL);
 411         BUG_ON(thread_stack_cache == NULL);
 412 }
 413
 414 # endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
 415 #else /* CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
 416
 417 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 418 {
 419         unsigned long *stack;
 420
 421         stack = arch_alloc_thread_stack_node(tsk, node);
 422         tsk->stack = stack;
 423         return stack ? 0 : -ENOMEM;
 424 }
 425
 426 static void free_thread_stack(struct task_struct *tsk)
 427 {
 428         arch_free_thread_stack(tsk);
 429         tsk->stack = NULL;
 430 }
 431
 432 #endif /* !CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
 433
 434 /* SLAB cache for signal_struct structures (tsk->signal) */
 435 static struct kmem_cache *signal_cachep;
 436
 437 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 438 struct kmem_cache *sighand_cachep;
 439
 440 /* SLAB cache for files_struct structures (tsk->files) */
 441 struct kmem_cache *files_cachep;
 442
 443 /* SLAB cache for fs_struct structures (tsk->fs) */
 444 struct kmem_cache *fs_cachep;
 445
 446 /* SLAB cache for vm_area_struct structures */
 447 static struct kmem_cache *vm_area_cachep;
 448
 449 /* SLAB cache for mm_struct structures (tsk->mm) */
 450 static struct kmem_cache *mm_cachep;
 451
 452 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 453 {
 454         struct vm_area_struct *vma;
 455
 456         vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 457         if (vma)
 458                 vma_init(vma, mm);
 459         return vma;
 460 }
 461
 462 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 463 {
 464         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 465
 466         if (new) {
 467                 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
 468                 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
 469                 /*
 470                  * orig->shared.rb may be modified concurrently, but the clone
 471                  * will be reinitialized.
 472                  */
 473                 *new = data_race(*orig);
 474                 INIT_LIST_HEAD(&new->anon_vma_chain);
 475                 new->vm_next = new->vm_prev = NULL;
 476                 dup_anon_vma_name(orig, new);
 477         }
 478         return new;
 479 }
 480
 481 void vm_area_free(struct vm_area_struct *vma)
 482 {
 483         free_anon_vma_name(vma);
 484         kmem_cache_free(vm_area_cachep, vma);
 485 }
 486
 487 static void account_kernel_stack(struct task_struct *tsk, int account)
 488 {
 489         if (IS_ENABLED(CONFIG_VMAP_STACK)) {
 490                 struct vm_struct *vm = task_stack_vm_area(tsk);
 491                 int i;
 492
 493                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 494                         mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
 495                                               account * (PAGE_SIZE / 1024));
 496         } else {
 497                 void *stack = task_stack_page(tsk);
 498
 499                 /* All stack pages are in the same node. */
 500                 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
 501                                       account * (THREAD_SIZE / 1024));
 502         }
 503 }
 504
 505 void exit_task_stack_account(struct task_struct *tsk)
 506 {
 507         account_kernel_stack(tsk, -1);
 508
 509         if (IS_ENABLED(CONFIG_VMAP_STACK)) {
 510                 struct vm_struct *vm;
 511                 int i;
 512
 513                 vm = task_stack_vm_area(tsk);
 514                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 515                         memcg_kmem_uncharge_page(vm->pages[i], 0);
 516         }
 517 }
 518
 519 static void release_task_stack(struct task_struct *tsk)
 520 {
 521         if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
 522                 return;  /* Better to leak the stack than to free prematurely */
 523
 524         free_thread_stack(tsk);
 525 }
 526
 527 #ifdef CONFIG_THREAD_INFO_IN_TASK
 528 void put_task_stack(struct task_struct *tsk)
 529 {
 530         if (refcount_dec_and_test(&tsk->stack_refcount))
 531                 release_task_stack(tsk);
 532 }
 533 #endif
 534
 535 void free_task(struct task_struct *tsk)
 536 {
 537         release_user_cpus_ptr(tsk);
 538         scs_release(tsk);
 539
 540 #ifndef CONFIG_THREAD_INFO_IN_TASK
 541         /*
 542          * The task is finally done with both the stack and thread_info,
 543          * so free both.
 544          */
 545         release_task_stack(tsk);
 546 #else
 547         /*
 548          * If the task had a separate stack allocation, it should be gone
 549          * by now.
 550          */
 551         WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
 552 #endif
 553         rt_mutex_debug_task_free(tsk);
 554         ftrace_graph_exit_task(tsk);
 555         arch_release_task_struct(tsk);
 556         if (tsk->flags & PF_KTHREAD)
 557                 free_kthread_struct(tsk);
 558         free_task_struct(tsk);
 559 }
 560 EXPORT_SYMBOL(free_task);
 561
 562 static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
 563 {
 564         struct file *exe_file;
 565
 566         exe_file = get_mm_exe_file(oldmm);
 567         RCU_INIT_POINTER(mm->exe_file, exe_file);
 568         /*
 569          * We depend on the oldmm having properly denied write access to the
 570          * exe_file already.
 571          */
 572         if (exe_file && deny_write_access(exe_file))
 573                 pr_warn_once("deny_write_access() failed in %s\n", __func__);
 574 }
 575
 576 #ifdef CONFIG_MMU
 577 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 578                                         struct mm_struct *oldmm)
 579 {
 580         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 581         struct rb_node **rb_link, *rb_parent;
 582         int retval;
 583         unsigned long charge;
 584         LIST_HEAD(uf);
 585
 586         uprobe_start_dup_mmap();
 587         if (mmap_write_lock_killable(oldmm)) {
 588                 retval = -EINTR;
 589                 goto fail_uprobe_end;
 590         }
 591         flush_cache_dup_mm(oldmm);
 592         uprobe_dup_mmap(oldmm, mm);
 593         /*
 594          * Not linked in yet - no deadlock potential:
 595          */
 596         mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
 597
 598         /* No ordering required: file already has been exposed. */
 599         dup_mm_exe_file(mm, oldmm);
 600
 601         mm->total_vm = oldmm->total_vm;
 602         mm->data_vm = oldmm->data_vm;
 603         mm->exec_vm = oldmm->exec_vm;
 604         mm->stack_vm = oldmm->stack_vm;
 605
 606         rb_link = &mm->mm_rb.rb_node;
 607         rb_parent = NULL;
 608         pprev = &mm->mmap;
 609         retval = ksm_fork(mm, oldmm);
 610         if (retval)
 611                 goto out;
 612         retval = khugepaged_fork(mm, oldmm);
 613         if (retval)
 614                 goto out;
 615
 616         prev = NULL;
 617         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 618                 struct file *file;
 619
 620                 if (mpnt->vm_flags & VM_DONTCOPY) {
 621                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 622                         continue;
 623                 }
 624                 charge = 0;
 625                 /*
 626                  * Don't duplicate many vmas if we've been oom-killed (for
 627                  * example)
 628                  */
 629                 if (fatal_signal_pending(current)) {
 630                         retval = -EINTR;
 631                         goto out;
 632                 }
 633                 if (mpnt->vm_flags & VM_ACCOUNT) {
 634                         unsigned long len = vma_pages(mpnt);
 635
 636                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 637                                 goto fail_nomem;
 638                         charge = len;
 639                 }
 640                 tmp = vm_area_dup(mpnt);
 641                 if (!tmp)
 642                         goto fail_nomem;
 643                 retval = vma_dup_policy(mpnt, tmp);
 644                 if (retval)
 645                         goto fail_nomem_policy;
 646                 tmp->vm_mm = mm;
 647                 retval = dup_userfaultfd(tmp, &uf);
 648                 if (retval)
 649                         goto fail_nomem_anon_vma_fork;
 650                 if (tmp->vm_flags & VM_WIPEONFORK) {
 651                         /*
 652                          * VM_WIPEONFORK gets a clean slate in the child.
 653                          * Don't prepare anon_vma until fault since we don't
 654                          * copy page for current vma.
 655                          */
 656                         tmp->anon_vma = NULL;
 657                 } else if (anon_vma_fork(tmp, mpnt))
 658                         goto fail_nomem_anon_vma_fork;
 659                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 660                 file = tmp->vm_file;
 661                 if (file) {
 662                         struct address_space *mapping = file->f_mapping;
 663
 664                         get_file(file);
 665                         i_mmap_lock_write(mapping);
 666                         if (tmp->vm_flags & VM_SHARED)
 667                                 mapping_allow_writable(mapping);
 668                         flush_dcache_mmap_lock(mapping);
 669                         /* insert tmp into the share list, just after mpnt */
 670                         vma_interval_tree_insert_after(tmp, mpnt,
 671                                         &mapping->i_mmap);
 672                         flush_dcache_mmap_unlock(mapping);
 673                         i_mmap_unlock_write(mapping);
 674                 }
 675
 676                 /*
 677                  * Clear hugetlb-related page reserves for children. This only
 678                  * affects MAP_PRIVATE mappings. Faults generated by the child
 679                  * are not guaranteed to succeed, even if read-only
 680                  */
 681                 if (is_vm_hugetlb_page(tmp))
 682                         reset_vma_resv_huge_pages(tmp);
 683
 684                 /*
 685                  * Link in the new vma and copy the page table entries.
 686                  */
 687                 *pprev = tmp;
 688                 pprev = &tmp->vm_next;
 689                 tmp->vm_prev = prev;
 690                 prev = tmp;
 691
 692                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 693                 rb_link = &tmp->vm_rb.rb_right;
 694                 rb_parent = &tmp->vm_rb;
 695
 696                 mm->map_count++;
 697                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 698                         retval = copy_page_range(tmp, mpnt);
 699
 700                 if (tmp->vm_ops && tmp->vm_ops->open)
 701                         tmp->vm_ops->open(tmp);
 702
 703                 if (retval)
 704                         goto out;
 705         }
 706         /* a new mm has just been created */
 707         retval = arch_dup_mmap(oldmm, mm);
 708 out:
 709         mmap_write_unlock(mm);
 710         flush_tlb_mm(oldmm);
 711         mmap_write_unlock(oldmm);
 712         dup_userfaultfd_complete(&uf);
 713 fail_uprobe_end:
 714         uprobe_end_dup_mmap();
 715         return retval;
 716 fail_nomem_anon_vma_fork:
 717         mpol_put(vma_policy(tmp));
 718 fail_nomem_policy:
 719         vm_area_free(tmp);
 720 fail_nomem:
 721         retval = -ENOMEM;
 722         vm_unacct_memory(charge);
 723         goto out;
 724 }
 725
 726 static inline int mm_alloc_pgd(struct mm_struct *mm)
 727 {
 728         mm->pgd = pgd_alloc(mm);
 729         if (unlikely(!mm->pgd))
 730                 return -ENOMEM;
 731         return 0;
 732 }
 733
 734 static inline void mm_free_pgd(struct mm_struct *mm)
 735 {
 736         pgd_free(mm, mm->pgd);
 737 }
 738 #else
 739 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 740 {
 741         mmap_write_lock(oldmm);
 742         dup_mm_exe_file(mm, oldmm);
 743         mmap_write_unlock(oldmm);
 744         return 0;
 745 }
 746 #define mm_alloc_pgd(mm)        (0)
 747 #define mm_free_pgd(mm)
 748 #endif /* CONFIG_MMU */
 749
 750 static void check_mm(struct mm_struct *mm)
 751 {
 752         int i;
 753
 754         BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
 755                          "Please make sure 'struct resident_page_types[]' is updated as well");
 756
 757         for (i = 0; i < NR_MM_COUNTERS; i++) {
 758                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 759
 760                 if (unlikely(x))
 761                         pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
 762                                  mm, resident_page_types[i], x);
 763         }
 764
 765         if (mm_pgtables_bytes(mm))
 766                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 767                                 mm_pgtables_bytes(mm));
 768
 769 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 770         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 771 #endif
 772 }
 773
 774 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 775 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 776
 777 /*
 778  * Called when the last reference to the mm
 779  * is dropped: either by a lazy thread or by
 780  * mmput. Free the page directory and the mm.
 781  */
 782 void __mmdrop(struct mm_struct *mm)
 783 {
 784         BUG_ON(mm == &init_mm);
 785         WARN_ON_ONCE(mm == current->mm);
 786         WARN_ON_ONCE(mm == current->active_mm);
 787         mm_free_pgd(mm);
 788         destroy_context(mm);
 789         mmu_notifier_subscriptions_destroy(mm);
 790         check_mm(mm);
 791         put_user_ns(mm->user_ns);
 792         free_mm(mm);
 793 }
 794 EXPORT_SYMBOL_GPL(__mmdrop);
 795
 796 static void mmdrop_async_fn(struct work_struct *work)
 797 {
 798         struct mm_struct *mm;
 799
 800         mm = container_of(work, struct mm_struct, async_put_work);
 801         __mmdrop(mm);
 802 }
 803
 804 static void mmdrop_async(struct mm_struct *mm)
 805 {
 806         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 807                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 808                 schedule_work(&mm->async_put_work);
 809         }
 810 }
 811
 812 static inline void free_signal_struct(struct signal_struct *sig)
 813 {
 814         taskstats_tgid_free(sig);
 815         sched_autogroup_exit(sig);
 816         /*
 817          * __mmdrop is not safe to call from softirq context on x86 due to
 818          * pgd_dtor so postpone it to the async context
 819          */
 820         if (sig->oom_mm)
 821                 mmdrop_async(sig->oom_mm);
 822         kmem_cache_free(signal_cachep, sig);
 823 }
 824
 825 static inline void put_signal_struct(struct signal_struct *sig)
 826 {
 827         if (refcount_dec_and_test(&sig->sigcnt))
 828                 free_signal_struct(sig);
 829 }
 830
 831 void __put_task_struct(struct task_struct *tsk)
 832 {
 833         WARN_ON(!tsk->exit_state);
 834         WARN_ON(refcount_read(&tsk->usage));
 835         WARN_ON(tsk == current);
 836
 837         io_uring_free(tsk);
 838         cgroup_free(tsk);
 839         task_numa_free(tsk, true);
 840         security_task_free(tsk);
 841         bpf_task_storage_free(tsk);
 842         exit_creds(tsk);
 843         delayacct_tsk_free(tsk);
 844         put_signal_struct(tsk->signal);
 845         sched_core_free(tsk);
 846         free_task(tsk);
 847 }
 848 EXPORT_SYMBOL_GPL(__put_task_struct);
 849
 850 void __init __weak arch_task_cache_init(void) { }
 851
 852 /*
 853  * set_max_threads
 854  */
 855 static void set_max_threads(unsigned int max_threads_suggested)
 856 {
 857         u64 threads;
 858         unsigned long nr_pages = totalram_pages();
 859
 860         /*
 861          * The number of threads shall be limited such that the thread
 862          * structures may only consume a small part of the available memory.
 863          */
 864         if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
 865                 threads = MAX_THREADS;
 866         else
 867                 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
 868                                     (u64) THREAD_SIZE * 8UL);
 869
 870         if (threads > max_threads_suggested)
 871                 threads = max_threads_suggested;
 872
 873         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 874 }
 875
 876 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 877 /* Initialized by the architecture: */
 878 int arch_task_struct_size __read_mostly;
 879 #endif
 880
 881 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 882 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 883 {
 884         /* Fetch thread_struct whitelist for the architecture. */
 885         arch_thread_struct_whitelist(offset, size);
 886
 887         /*
 888          * Handle zero-sized whitelist or empty thread_struct, otherwise
 889          * adjust offset to position of thread_struct in task_struct.
 890          */
 891         if (unlikely(*size == 0))
 892                 *offset = 0;
 893         else
 894                 *offset += offsetof(struct task_struct, thread);
 895 }
 896 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
 897
 898 void __init fork_init(void)
 899 {
 900         int i;
 901 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 902 #ifndef ARCH_MIN_TASKALIGN
 903 #define ARCH_MIN_TASKALIGN      0
 904 #endif
 905         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 906         unsigned long useroffset, usersize;
 907
 908         /* create a slab on which task_structs can be allocated */
 909         task_struct_whitelist(&useroffset, &usersize);
 910         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 911                         arch_task_struct_size, align,
 912                         SLAB_PANIC|SLAB_ACCOUNT,
 913                         useroffset, usersize, NULL);
 914 #endif
 915
 916         /* do the arch specific task caches init */
 917         arch_task_cache_init();
 918
 919         set_max_threads(MAX_THREADS);
 920
 921         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 922         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 923         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 924                 init_task.signal->rlim[RLIMIT_NPROC];
 925
 926         for (i = 0; i < MAX_PER_NAMESPACE_UCOUNTS; i++)
 927                 init_user_ns.ucount_max[i] = max_threads/2;
 928
 929         set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_NPROC,      RLIM_INFINITY);
 930         set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE,   RLIM_INFINITY);
 931         set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
 932         set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK,    RLIM_INFINITY);
 933
 934 #ifdef CONFIG_VMAP_STACK
 935         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 936                           NULL, free_vm_stack_cache);
 937 #endif
 938
 939         scs_init();
 940
 941         lockdep_init_task(&init_task);
 942         uprobes_init();
 943 }
 944
 945 int __weak arch_dup_task_struct(struct task_struct *dst,
 946                                                struct task_struct *src)
 947 {
 948         *dst = *src;
 949         return 0;
 950 }
 951
 952 void set_task_stack_end_magic(struct task_struct *tsk)
 953 {
 954         unsigned long *stackend;
 955
 956         stackend = end_of_stack(tsk);
 957         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 958 }
 959
 960 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 961 {
 962         struct task_struct *tsk;
 963         int err;
 964
 965         if (node == NUMA_NO_NODE)
 966                 node = tsk_fork_get_node(orig);
 967         tsk = alloc_task_struct_node(node);
 968         if (!tsk)
 969                 return NULL;
 970
 971         err = arch_dup_task_struct(tsk, orig);
 972         if (err)
 973                 goto free_tsk;
 974
 975         err = alloc_thread_stack_node(tsk, node);
 976         if (err)
 977                 goto free_tsk;
 978
 979 #ifdef CONFIG_THREAD_INFO_IN_TASK
 980         refcount_set(&tsk->stack_refcount, 1);
 981 #endif
 982         account_kernel_stack(tsk, 1);
 983
 984         err = scs_prepare(tsk, node);
 985         if (err)
 986                 goto free_stack;
 987
 988 #ifdef CONFIG_SECCOMP
 989         /*
 990          * We must handle setting up seccomp filters once we're under
 991          * the sighand lock in case orig has changed between now and
 992          * then. Until then, filter must be NULL to avoid messing up
 993          * the usage counts on the error path calling free_task.
 994          */
 995         tsk->seccomp.filter = NULL;
 996 #endif
 997
 998         setup_thread_stack(tsk, orig);
 999         clear_user_return_notifier(tsk);
1000         clear_tsk_need_resched(tsk);
1001         set_task_stack_end_magic(tsk);
1002         clear_syscall_work_syscall_user_dispatch(tsk);
1003
1004 #ifdef CONFIG_STACKPROTECTOR
1005         tsk->stack_canary = get_random_canary();
1006 #endif
1007         if (orig->cpus_ptr == &orig->cpus_mask)
1008                 tsk->cpus_ptr = &tsk->cpus_mask;
1009         dup_user_cpus_ptr(tsk, orig, node);
1010
1011         /*
1012          * One for the user space visible state that goes away when reaped.
1013          * One for the scheduler.
1014          */
1015         refcount_set(&tsk->rcu_users, 2);
1016         /* One for the rcu users */
1017         refcount_set(&tsk->usage, 1);
1018 #ifdef CONFIG_BLK_DEV_IO_TRACE
1019         tsk->btrace_seq = 0;
1020 #endif
1021         tsk->splice_pipe = NULL;
1022         tsk->task_frag.page = NULL;
1023         tsk->wake_q.next = NULL;
1024         tsk->worker_private = NULL;
1025
1026         kcov_task_init(tsk);
1027         kmap_local_fork(tsk);
1028
1029 #ifdef CONFIG_FAULT_INJECTION
1030         tsk->fail_nth = 0;
1031 #endif
1032
1033 #ifdef CONFIG_BLK_CGROUP
1034         tsk->throttle_queue = NULL;
1035         tsk->use_memdelay = 0;
1036 #endif
1037
1038 #ifdef CONFIG_IOMMU_SVA
1039         tsk->pasid_activated = 0;
1040 #endif
1041
1042 #ifdef CONFIG_MEMCG
1043         tsk->active_memcg = NULL;
1044 #endif
1045         return tsk;
1046
1047 free_stack:
1048         exit_task_stack_account(tsk);
1049         free_thread_stack(tsk);
1050 free_tsk:
1051         free_task_struct(tsk);
1052         return NULL;
1053 }
1054
1055 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1056
1057 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1058
1059 static int __init coredump_filter_setup(char *s)
1060 {
1061         default_dump_filter =
1062                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1063                 MMF_DUMP_FILTER_MASK;
1064         return 1;
1065 }
1066
1067 __setup("coredump_filter=", coredump_filter_setup);
1068
1069 #include <linux/init_task.h>
1070
1071 static void mm_init_aio(struct mm_struct *mm)
1072 {
1073 #ifdef CONFIG_AIO
1074         spin_lock_init(&mm->ioctx_lock);
1075         mm->ioctx_table = NULL;
1076 #endif
1077 }
1078
1079 static __always_inline void mm_clear_owner(struct mm_struct *mm,
1080                                            struct task_struct *p)
1081 {
1082 #ifdef CONFIG_MEMCG
1083         if (mm->owner == p)
1084                 WRITE_ONCE(mm->owner, NULL);
1085 #endif
1086 }
1087
1088 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1089 {
1090 #ifdef CONFIG_MEMCG
1091         mm->owner = p;
1092 #endif
1093 }
1094
1095 static void mm_init_uprobes_state(struct mm_struct *mm)
1096 {
1097 #ifdef CONFIG_UPROBES
1098         mm->uprobes_state.xol_area = NULL;
1099 #endif
1100 }
1101
1102 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1103         struct user_namespace *user_ns)
1104 {
1105         mm->mmap = NULL;
1106         mm->mm_rb = RB_ROOT;
1107         mm->vmacache_seqnum = 0;
1108         atomic_set(&mm->mm_users, 1);
1109         atomic_set(&mm->mm_count, 1);
1110         seqcount_init(&mm->write_protect_seq);
1111         mmap_init_lock(mm);
1112         INIT_LIST_HEAD(&mm->mmlist);
1113         mm_pgtables_bytes_init(mm);
1114         mm->map_count = 0;
1115         mm->locked_vm = 0;
1116         atomic64_set(&mm->pinned_vm, 0);
1117         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1118         spin_lock_init(&mm->page_table_lock);
1119         spin_lock_init(&mm->arg_lock);
1120         mm_init_cpumask(mm);
1121         mm_init_aio(mm);
1122         mm_init_owner(mm, p);
1123         mm_pasid_init(mm);
1124         RCU_INIT_POINTER(mm->exe_file, NULL);
1125         mmu_notifier_subscriptions_init(mm);
1126         init_tlb_flush_pending(mm);
1127 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1128         mm->pmd_huge_pte = NULL;
1129 #endif
1130         mm_init_uprobes_state(mm);
1131         hugetlb_count_init(mm);
1132
1133         if (current->mm) {
1134                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1135                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1136         } else {
1137                 mm->flags = default_dump_filter;
1138                 mm->def_flags = 0;
1139         }
1140
1141         if (mm_alloc_pgd(mm))
1142                 goto fail_nopgd;
1143
1144         if (init_new_context(p, mm))
1145                 goto fail_nocontext;
1146
1147         mm->user_ns = get_user_ns(user_ns);
1148         return mm;
1149
1150 fail_nocontext:
1151         mm_free_pgd(mm);
1152 fail_nopgd:
1153         free_mm(mm);
1154         return NULL;
1155 }
1156
1157 /*
1158  * Allocate and initialize an mm_struct.
1159  */
1160 struct mm_struct *mm_alloc(void)
1161 {
1162         struct mm_struct *mm;
1163
1164         mm = allocate_mm();
1165         if (!mm)
1166                 return NULL;
1167
1168         memset(mm, 0, sizeof(*mm));
1169         return mm_init(mm, current, current_user_ns());
1170 }
1171
1172 static inline void __mmput(struct mm_struct *mm)
1173 {
1174         VM_BUG_ON(atomic_read(&mm->mm_users));
1175
1176         uprobe_clear_state(mm);
1177         exit_aio(mm);
1178         ksm_exit(mm);
1179         khugepaged_exit(mm); /* must run before exit_mmap */
1180         exit_mmap(mm);
1181         mm_put_huge_zero_page(mm);
1182         set_mm_exe_file(mm, NULL);
1183         if (!list_empty(&mm->mmlist)) {
1184                 spin_lock(&mmlist_lock);
1185                 list_del(&mm->mmlist);
1186                 spin_unlock(&mmlist_lock);
1187         }
1188         if (mm->binfmt)
1189                 module_put(mm->binfmt->module);
1190         mm_pasid_drop(mm);
1191         mmdrop(mm);
1192 }
1193
1194 /*
1195  * Decrement the use count and release all resources for an mm.
1196  */
1197 void mmput(struct mm_struct *mm)
1198 {
1199         might_sleep();
1200
1201         if (atomic_dec_and_test(&mm->mm_users))
1202                 __mmput(mm);
1203 }
1204 EXPORT_SYMBOL_GPL(mmput);
1205
1206 #ifdef CONFIG_MMU
1207 static void mmput_async_fn(struct work_struct *work)
1208 {
1209         struct mm_struct *mm = container_of(work, struct mm_struct,
1210                                             async_put_work);
1211
1212         __mmput(mm);
1213 }
1214
1215 void mmput_async(struct mm_struct *mm)
1216 {
1217         if (atomic_dec_and_test(&mm->mm_users)) {
1218                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1219                 schedule_work(&mm->async_put_work);
1220         }
1221 }
1222 #endif
1223
1224 /**
1225  * set_mm_exe_file - change a reference to the mm's executable file
1226  *
1227  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1228  *
1229  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1230  * invocations: in mmput() nobody alive left, in execve task is single
1231  * threaded.
1232  *
1233  * Can only fail if new_exe_file != NULL.
1234  */
1235 int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1236 {
1237         struct file *old_exe_file;
1238
1239         /*
1240          * It is safe to dereference the exe_file without RCU as
1241          * this function is only called if nobody else can access
1242          * this mm -- see comment above for justification.
1243          */
1244         old_exe_file = rcu_dereference_raw(mm->exe_file);
1245
1246         if (new_exe_file) {
1247                 /*
1248                  * We expect the caller (i.e., sys_execve) to already denied
1249                  * write access, so this is unlikely to fail.
1250                  */
1251                 if (unlikely(deny_write_access(new_exe_file)))
1252                         return -EACCES;
1253                 get_file(new_exe_file);
1254         }
1255         rcu_assign_pointer(mm->exe_file, new_exe_file);
1256         if (old_exe_file) {
1257                 allow_write_access(old_exe_file);
1258                 fput(old_exe_file);
1259         }
1260         return 0;
1261 }
1262
1263 /**
1264  * replace_mm_exe_file - replace a reference to the mm's executable file
1265  *
1266  * This changes mm's executable file (shown as symlink /proc/[pid]/exe),
1267  * dealing with concurrent invocation and without grabbing the mmap lock in
1268  * write mode.
1269  *
1270  * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1271  */
1272 int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1273 {
1274         struct vm_area_struct *vma;
1275         struct file *old_exe_file;
1276         int ret = 0;
1277
1278         /* Forbid mm->exe_file change if old file still mapped. */
1279         old_exe_file = get_mm_exe_file(mm);
1280         if (old_exe_file) {
1281                 mmap_read_lock(mm);
1282                 for (vma = mm->mmap; vma && !ret; vma = vma->vm_next) {
1283                         if (!vma->vm_file)
1284                                 continue;
1285                         if (path_equal(&vma->vm_file->f_path,
1286                                        &old_exe_file->f_path))
1287                                 ret = -EBUSY;
1288                 }
1289                 mmap_read_unlock(mm);
1290                 fput(old_exe_file);
1291                 if (ret)
1292                         return ret;
1293         }
1294
1295         /* set the new file, lockless */
1296         ret = deny_write_access(new_exe_file);
1297         if (ret)
1298                 return -EACCES;
1299         get_file(new_exe_file);
1300
1301         old_exe_file = xchg(&mm->exe_file, new_exe_file);
1302         if (old_exe_file) {
1303                 /*
1304                  * Don't race with dup_mmap() getting the file and disallowing
1305                  * write access while someone might open the file writable.
1306                  */
1307                 mmap_read_lock(mm);
1308                 allow_write_access(old_exe_file);
1309                 fput(old_exe_file);
1310                 mmap_read_unlock(mm);
1311         }
1312         return 0;
1313 }
1314
1315 /**
1316  * get_mm_exe_file - acquire a reference to the mm's executable file
1317  *
1318  * Returns %NULL if mm has no associated executable file.
1319  * User must release file via fput().
1320  */
1321 struct file *get_mm_exe_file(struct mm_struct *mm)
1322 {
1323         struct file *exe_file;
1324
1325         rcu_read_lock();
1326         exe_file = rcu_dereference(mm->exe_file);
1327         if (exe_file && !get_file_rcu(exe_file))
1328                 exe_file = NULL;
1329         rcu_read_unlock();
1330         return exe_file;
1331 }
1332
1333 /**
1334  * get_task_exe_file - acquire a reference to the task's executable file
1335  *
1336  * Returns %NULL if task's mm (if any) has no associated executable file or
1337  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1338  * User must release file via fput().
1339  */
1340 struct file *get_task_exe_file(struct task_struct *task)
1341 {
1342         struct file *exe_file = NULL;
1343         struct mm_struct *mm;
1344
1345         task_lock(task);
1346         mm = task->mm;
1347         if (mm) {
1348                 if (!(task->flags & PF_KTHREAD))
1349                         exe_file = get_mm_exe_file(mm);
1350         }
1351         task_unlock(task);
1352         return exe_file;
1353 }
1354
1355 /**
1356  * get_task_mm - acquire a reference to the task's mm
1357  *
1358  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1359  * this kernel workthread has transiently adopted a user mm with use_mm,
1360  * to do its AIO) is not set and if so returns a reference to it, after
1361  * bumping up the use count.  User must release the mm via mmput()
1362  * after use.  Typically used by /proc and ptrace.
1363  */
1364 struct mm_struct *get_task_mm(struct task_struct *task)
1365 {
1366         struct mm_struct *mm;
1367
1368         task_lock(task);
1369         mm = task->mm;
1370         if (mm) {
1371                 if (task->flags & PF_KTHREAD)
1372                         mm = NULL;
1373                 else
1374                         mmget(mm);
1375         }
1376         task_unlock(task);
1377         return mm;
1378 }
1379 EXPORT_SYMBOL_GPL(get_task_mm);
1380
1381 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1382 {
1383         struct mm_struct *mm;
1384         int err;
1385
1386         err =  down_read_killable(&task->signal->exec_update_lock);
1387         if (err)
1388                 return ERR_PTR(err);
1389
1390         mm = get_task_mm(task);
1391         if (mm && mm != current->mm &&
1392                         !ptrace_may_access(task, mode)) {
1393                 mmput(mm);
1394                 mm = ERR_PTR(-EACCES);
1395         }
1396         up_read(&task->signal->exec_update_lock);
1397
1398         return mm;
1399 }
1400
1401 static void complete_vfork_done(struct task_struct *tsk)
1402 {
1403         struct completion *vfork;
1404
1405         task_lock(tsk);
1406         vfork = tsk->vfork_done;
1407         if (likely(vfork)) {
1408                 tsk->vfork_done = NULL;
1409                 complete(vfork);
1410         }
1411         task_unlock(tsk);
1412 }
1413
1414 static int wait_for_vfork_done(struct task_struct *child,
1415                                 struct completion *vfork)
1416 {
1417         int killed;
1418
1419         freezer_do_not_count();
1420         cgroup_enter_frozen();
1421         killed = wait_for_completion_killable(vfork);
1422         cgroup_leave_frozen(false);
1423         freezer_count();
1424
1425         if (killed) {
1426                 task_lock(child);
1427                 child->vfork_done = NULL;
1428                 task_unlock(child);
1429         }
1430
1431         put_task_struct(child);
1432         return killed;
1433 }
1434
1435 /* Please note the differences between mmput and mm_release.
1436  * mmput is called whenever we stop holding onto a mm_struct,
1437  * error success whatever.
1438  *
1439  * mm_release is called after a mm_struct has been removed
1440  * from the current process.
1441  *
1442  * This difference is important for error handling, when we
1443  * only half set up a mm_struct for a new process and need to restore
1444  * the old one.  Because we mmput the new mm_struct before
1445  * restoring the old one. . .
1446  * Eric Biederman 10 January 1998
1447  */
1448 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1449 {
1450         uprobe_free_utask(tsk);
1451
1452         /* Get rid of any cached register state */
1453         deactivate_mm(tsk, mm);
1454
1455         /*
1456          * Signal userspace if we're not exiting with a core dump
1457          * because we want to leave the value intact for debugging
1458          * purposes.
1459          */
1460         if (tsk->clear_child_tid) {
1461                 if (atomic_read(&mm->mm_users) > 1) {
1462                         /*
1463                          * We don't check the error code - if userspace has
1464                          * not set up a proper pointer then tough luck.
1465                          */
1466                         put_user(0, tsk->clear_child_tid);
1467                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1468                                         1, NULL, NULL, 0, 0);
1469                 }
1470                 tsk->clear_child_tid = NULL;
1471         }
1472
1473         /*
1474          * All done, finally we can wake up parent and return this mm to him.
1475          * Also kthread_stop() uses this completion for synchronization.
1476          */
1477         if (tsk->vfork_done)
1478                 complete_vfork_done(tsk);
1479 }
1480
1481 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1482 {
1483         futex_exit_release(tsk);
1484         mm_release(tsk, mm);
1485 }
1486
1487 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1488 {
1489         futex_exec_release(tsk);
1490         mm_release(tsk, mm);
1491 }
1492
1493 /**
1494  * dup_mm() - duplicates an existing mm structure
1495  * @tsk: the task_struct with which the new mm will be associated.
1496  * @oldmm: the mm to duplicate.
1497  *
1498  * Allocates a new mm structure and duplicates the provided @oldmm structure
1499  * content into it.
1500  *
1501  * Return: the duplicated mm or NULL on failure.
1502  */
1503 static struct mm_struct *dup_mm(struct task_struct *tsk,
1504                                 struct mm_struct *oldmm)
1505 {
1506         struct mm_struct *mm;
1507         int err;
1508
1509         mm = allocate_mm();
1510         if (!mm)
1511                 goto fail_nomem;
1512
1513         memcpy(mm, oldmm, sizeof(*mm));
1514
1515         if (!mm_init(mm, tsk, mm->user_ns))
1516                 goto fail_nomem;
1517
1518         err = dup_mmap(mm, oldmm);
1519         if (err)
1520                 goto free_pt;
1521
1522         mm->hiwater_rss = get_mm_rss(mm);
1523         mm->hiwater_vm = mm->total_vm;
1524
1525         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1526                 goto free_pt;
1527
1528         return mm;
1529
1530 free_pt:
1531         /* don't put binfmt in mmput, we haven't got module yet */
1532         mm->binfmt = NULL;
1533         mm_init_owner(mm, NULL);
1534         mmput(mm);
1535
1536 fail_nomem:
1537         return NULL;
1538 }
1539
1540 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1541 {
1542         struct mm_struct *mm, *oldmm;
1543
1544         tsk->min_flt = tsk->maj_flt = 0;
1545         tsk->nvcsw = tsk->nivcsw = 0;
1546 #ifdef CONFIG_DETECT_HUNG_TASK
1547         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1548         tsk->last_switch_time = 0;
1549 #endif
1550
1551         tsk->mm = NULL;
1552         tsk->active_mm = NULL;
1553
1554         /*
1555          * Are we cloning a kernel thread?
1556          *
1557          * We need to steal a active VM for that..
1558          */
1559         oldmm = current->mm;
1560         if (!oldmm)
1561                 return 0;
1562
1563         /* initialize the new vmacache entries */
1564         vmacache_flush(tsk);
1565
1566         if (clone_flags & CLONE_VM) {
1567                 mmget(oldmm);
1568                 mm = oldmm;
1569         } else {
1570                 mm = dup_mm(tsk, current->mm);
1571                 if (!mm)
1572                         return -ENOMEM;
1573         }
1574
1575         tsk->mm = mm;
1576         tsk->active_mm = mm;
1577         return 0;
1578 }
1579
1580 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1581 {
1582         struct fs_struct *fs = current->fs;
1583         if (clone_flags & CLONE_FS) {
1584                 /* tsk->fs is already what we want */
1585                 spin_lock(&fs->lock);
1586                 if (fs->in_exec) {
1587                         spin_unlock(&fs->lock);
1588                         return -EAGAIN;
1589                 }
1590                 fs->users++;
1591                 spin_unlock(&fs->lock);
1592                 return 0;
1593         }
1594         tsk->fs = copy_fs_struct(fs);
1595         if (!tsk->fs)
1596                 return -ENOMEM;
1597         return 0;
1598 }
1599
1600 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1601 {
1602         struct files_struct *oldf, *newf;
1603         int error = 0;
1604
1605         /*
1606          * A background process may not have any files ...
1607          */
1608         oldf = current->files;
1609         if (!oldf)
1610                 goto out;
1611
1612         if (clone_flags & CLONE_FILES) {
1613                 atomic_inc(&oldf->count);
1614                 goto out;
1615         }
1616
1617         newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1618         if (!newf)
1619                 goto out;
1620
1621         tsk->files = newf;
1622         error = 0;
1623 out:
1624         return error;
1625 }
1626
1627 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1628 {
1629         struct sighand_struct *sig;
1630
1631         if (clone_flags & CLONE_SIGHAND) {
1632                 refcount_inc(&current->sighand->count);
1633                 return 0;
1634         }
1635         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1636         RCU_INIT_POINTER(tsk->sighand, sig);
1637         if (!sig)
1638                 return -ENOMEM;
1639
1640         refcount_set(&sig->count, 1);
1641         spin_lock_irq(&current->sighand->siglock);
1642         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1643         spin_unlock_irq(&current->sighand->siglock);
1644
1645         /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1646         if (clone_flags & CLONE_CLEAR_SIGHAND)
1647                 flush_signal_handlers(tsk, 0);
1648
1649         return 0;
1650 }
1651
1652 void __cleanup_sighand(struct sighand_struct *sighand)
1653 {
1654         if (refcount_dec_and_test(&sighand->count)) {
1655                 signalfd_cleanup(sighand);
1656                 /*
1657                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1658                  * without an RCU grace period, see __lock_task_sighand().
1659                  */
1660                 kmem_cache_free(sighand_cachep, sighand);
1661         }
1662 }
1663
1664 /*
1665  * Initialize POSIX timer handling for a thread group.
1666  */
1667 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1668 {
1669         struct posix_cputimers *pct = &sig->posix_cputimers;
1670         unsigned long cpu_limit;
1671
1672         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1673         posix_cputimers_group_init(pct, cpu_limit);
1674 }
1675
1676 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1677 {
1678         struct signal_struct *sig;
1679
1680         if (clone_flags & CLONE_THREAD)
1681                 return 0;
1682
1683         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1684         tsk->signal = sig;
1685         if (!sig)
1686                 return -ENOMEM;
1687
1688         sig->nr_threads = 1;
1689         atomic_set(&sig->live, 1);
1690         refcount_set(&sig->sigcnt, 1);
1691
1692         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1693         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1694         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1695
1696         init_waitqueue_head(&sig->wait_chldexit);
1697         sig->curr_target = tsk;
1698         init_sigpending(&sig->shared_pending);
1699         INIT_HLIST_HEAD(&sig->multiprocess);
1700         seqlock_init(&sig->stats_lock);
1701         prev_cputime_init(&sig->prev_cputime);
1702
1703 #ifdef CONFIG_POSIX_TIMERS
1704         INIT_LIST_HEAD(&sig->posix_timers);
1705         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1706         sig->real_timer.function = it_real_fn;
1707 #endif
1708
1709         task_lock(current->group_leader);
1710         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1711         task_unlock(current->group_leader);
1712
1713         posix_cpu_timers_init_group(sig);
1714
1715         tty_audit_fork(sig);
1716         sched_autogroup_fork(sig);
1717
1718         sig->oom_score_adj = current->signal->oom_score_adj;
1719         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1720
1721         mutex_init(&sig->cred_guard_mutex);
1722         init_rwsem(&sig->exec_update_lock);
1723
1724         return 0;
1725 }
1726
1727 static void copy_seccomp(struct task_struct *p)
1728 {
1729 #ifdef CONFIG_SECCOMP
1730         /*
1731          * Must be called with sighand->lock held, which is common to
1732          * all threads in the group. Holding cred_guard_mutex is not
1733          * needed because this new task is not yet running and cannot
1734          * be racing exec.
1735          */
1736         assert_spin_locked(&current->sighand->siglock);
1737
1738         /* Ref-count the new filter user, and assign it. */
1739         get_seccomp_filter(current);
1740         p->seccomp = current->seccomp;
1741
1742         /*
1743          * Explicitly enable no_new_privs here in case it got set
1744          * between the task_struct being duplicated and holding the
1745          * sighand lock. The seccomp state and nnp must be in sync.
1746          */
1747         if (task_no_new_privs(current))
1748                 task_set_no_new_privs(p);
1749
1750         /*
1751          * If the parent gained a seccomp mode after copying thread
1752          * flags and between before we held the sighand lock, we have
1753          * to manually enable the seccomp thread flag here.
1754          */
1755         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1756                 set_task_syscall_work(p, SECCOMP);
1757 #endif
1758 }
1759
1760 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1761 {
1762         current->clear_child_tid = tidptr;
1763
1764         return task_pid_vnr(current);
1765 }
1766
1767 static void rt_mutex_init_task(struct task_struct *p)
1768 {
1769         raw_spin_lock_init(&p->pi_lock);
1770 #ifdef CONFIG_RT_MUTEXES
1771         p->pi_waiters = RB_ROOT_CACHED;
1772         p->pi_top_task = NULL;
1773         p->pi_blocked_on = NULL;
1774 #endif
1775 }
1776
1777 static inline void init_task_pid_links(struct task_struct *task)
1778 {
1779         enum pid_type type;
1780
1781         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1782                 INIT_HLIST_NODE(&task->pid_links[type]);
1783 }
1784
1785 static inline void
1786 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1787 {
1788         if (type == PIDTYPE_PID)
1789                 task->thread_pid = pid;
1790         else
1791                 task->signal->pids[type] = pid;
1792 }
1793
1794 static inline void rcu_copy_process(struct task_struct *p)
1795 {
1796 #ifdef CONFIG_PREEMPT_RCU
1797         p->rcu_read_lock_nesting = 0;
1798         p->rcu_read_unlock_special.s = 0;
1799         p->rcu_blocked_node = NULL;
1800         INIT_LIST_HEAD(&p->rcu_node_entry);
1801 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1802 #ifdef CONFIG_TASKS_RCU
1803         p->rcu_tasks_holdout = false;
1804         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1805         p->rcu_tasks_idle_cpu = -1;
1806 #endif /* #ifdef CONFIG_TASKS_RCU */
1807 #ifdef CONFIG_TASKS_TRACE_RCU
1808         p->trc_reader_nesting = 0;
1809         p->trc_reader_special.s = 0;
1810         INIT_LIST_HEAD(&p->trc_holdout_list);
1811 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1812 }
1813
1814 struct pid *pidfd_pid(const struct file *file)
1815 {
1816         if (file->f_op == &pidfd_fops)
1817                 return file->private_data;
1818
1819         return ERR_PTR(-EBADF);
1820 }
1821
1822 static int pidfd_release(struct inode *inode, struct file *file)
1823 {
1824         struct pid *pid = file->private_data;
1825
1826         file->private_data = NULL;
1827         put_pid(pid);
1828         return 0;
1829 }
1830
1831 #ifdef CONFIG_PROC_FS
1832 /**
1833  * pidfd_show_fdinfo - print information about a pidfd
1834  * @m: proc fdinfo file
1835  * @f: file referencing a pidfd
1836  *
1837  * Pid:
1838  * This function will print the pid that a given pidfd refers to in the
1839  * pid namespace of the procfs instance.
1840  * If the pid namespace of the process is not a descendant of the pid
1841  * namespace of the procfs instance 0 will be shown as its pid. This is
1842  * similar to calling getppid() on a process whose parent is outside of
1843  * its pid namespace.
1844  *
1845  * NSpid:
1846  * If pid namespaces are supported then this function will also print
1847  * the pid of a given pidfd refers to for all descendant pid namespaces
1848  * starting from the current pid namespace of the instance, i.e. the
1849  * Pid field and the first entry in the NSpid field will be identical.
1850  * If the pid namespace of the process is not a descendant of the pid
1851  * namespace of the procfs instance 0 will be shown as its first NSpid
1852  * entry and no others will be shown.
1853  * Note that this differs from the Pid and NSpid fields in
1854  * /proc/<pid>/status where Pid and NSpid are always shown relative to
1855  * the  pid namespace of the procfs instance. The difference becomes
1856  * obvious when sending around a pidfd between pid namespaces from a
1857  * different branch of the tree, i.e. where no ancestral relation is
1858  * present between the pid namespaces:
1859  * - create two new pid namespaces ns1 and ns2 in the initial pid
1860  *   namespace (also take care to create new mount namespaces in the
1861  *   new pid namespace and mount procfs)
1862  * - create a process with a pidfd in ns1
1863  * - send pidfd from ns1 to ns2
1864  * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1865  *   have exactly one entry, which is 0
1866  */
1867 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1868 {
1869         struct pid *pid = f->private_data;
1870         struct pid_namespace *ns;
1871         pid_t nr = -1;
1872
1873         if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1874                 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1875                 nr = pid_nr_ns(pid, ns);
1876         }
1877
1878         seq_put_decimal_ll(m, "Pid:\t", nr);
1879
1880 #ifdef CONFIG_PID_NS
1881         seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1882         if (nr > 0) {
1883                 int i;
1884
1885                 /* If nr is non-zero it means that 'pid' is valid and that
1886                  * ns, i.e. the pid namespace associated with the procfs
1887                  * instance, is in the pid namespace hierarchy of pid.
1888                  * Start at one below the already printed level.
1889                  */
1890                 for (i = ns->level + 1; i <= pid->level; i++)
1891                         seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1892         }
1893 #endif
1894         seq_putc(m, '\n');
1895 }
1896 #endif
1897
1898 /*
1899  * Poll support for process exit notification.
1900  */
1901 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1902 {
1903         struct pid *pid = file->private_data;
1904         __poll_t poll_flags = 0;
1905
1906         poll_wait(file, &pid->wait_pidfd, pts);
1907
1908         /*
1909          * Inform pollers only when the whole thread group exits.
1910          * If the thread group leader exits before all other threads in the
1911          * group, then poll(2) should block, similar to the wait(2) family.
1912          */
1913         if (thread_group_exited(pid))
1914                 poll_flags = EPOLLIN | EPOLLRDNORM;
1915
1916         return poll_flags;
1917 }
1918
1919 const struct file_operations pidfd_fops = {
1920         .release = pidfd_release,
1921         .poll = pidfd_poll,
1922 #ifdef CONFIG_PROC_FS
1923         .show_fdinfo = pidfd_show_fdinfo,
1924 #endif
1925 };
1926
1927 static void __delayed_free_task(struct rcu_head *rhp)
1928 {
1929         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1930
1931         free_task(tsk);
1932 }
1933
1934 static __always_inline void delayed_free_task(struct task_struct *tsk)
1935 {
1936         if (IS_ENABLED(CONFIG_MEMCG))
1937                 call_rcu(&tsk->rcu, __delayed_free_task);
1938         else
1939                 free_task(tsk);
1940 }
1941
1942 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1943 {
1944         /* Skip if kernel thread */
1945         if (!tsk->mm)
1946                 return;
1947
1948         /* Skip if spawning a thread or using vfork */
1949         if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1950                 return;
1951
1952         /* We need to synchronize with __set_oom_adj */
1953         mutex_lock(&oom_adj_mutex);
1954         set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1955         /* Update the values in case they were changed after copy_signal */
1956         tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1957         tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1958         mutex_unlock(&oom_adj_mutex);
1959 }
1960
1961 /*
1962  * This creates a new process as a copy of the old one,
1963  * but does not actually start it yet.
1964  *
1965  * It copies the registers, and all the appropriate
1966  * parts of the process environment (as per the clone
1967  * flags). The actual kick-off is left to the caller.
1968  */
1969 static __latent_entropy struct task_struct *copy_process(
1970                                         struct pid *pid,
1971                                         int trace,
1972                                         int node,
1973                                         struct kernel_clone_args *args)
1974 {
1975         int pidfd = -1, retval;
1976         struct task_struct *p;
1977         struct multiprocess_signals delayed;
1978         struct file *pidfile = NULL;
1979         u64 clone_flags = args->flags;
1980         struct nsproxy *nsp = current->nsproxy;
1981
1982         /*
1983          * Don't allow sharing the root directory with processes in a different
1984          * namespace
1985          */
1986         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1987                 return ERR_PTR(-EINVAL);
1988
1989         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1990                 return ERR_PTR(-EINVAL);
1991
1992         /*
1993          * Thread groups must share signals as well, and detached threads
1994          * can only be started up within the thread group.
1995          */
1996         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1997                 return ERR_PTR(-EINVAL);
1998
1999         /*
2000          * Shared signal handlers imply shared VM. By way of the above,
2001          * thread groups also imply shared VM. Blocking this case allows
2002          * for various simplifications in other code.
2003          */
2004         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2005                 return ERR_PTR(-EINVAL);
2006
2007         /*
2008          * Siblings of global init remain as zombies on exit since they are
2009          * not reaped by their parent (swapper). To solve this and to avoid
2010          * multi-rooted process trees, prevent global and container-inits
2011          * from creating siblings.
2012          */
2013         if ((clone_flags & CLONE_PARENT) &&
2014                                 current->signal->flags & SIGNAL_UNKILLABLE)
2015                 return ERR_PTR(-EINVAL);
2016
2017         /*
2018          * If the new process will be in a different pid or user namespace
2019          * do not allow it to share a thread group with the forking task.
2020          */
2021         if (clone_flags & CLONE_THREAD) {
2022                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2023                     (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2024                         return ERR_PTR(-EINVAL);
2025         }
2026
2027         /*
2028          * If the new process will be in a different time namespace
2029          * do not allow it to share VM or a thread group with the forking task.
2030          */
2031         if (clone_flags & (CLONE_THREAD | CLONE_VM)) {
2032                 if (nsp->time_ns != nsp->time_ns_for_children)
2033                         return ERR_PTR(-EINVAL);
2034         }
2035
2036         if (clone_flags & CLONE_PIDFD) {
2037                 /*
2038                  * - CLONE_DETACHED is blocked so that we can potentially
2039                  *   reuse it later for CLONE_PIDFD.
2040                  * - CLONE_THREAD is blocked until someone really needs it.
2041                  */
2042                 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
2043                         return ERR_PTR(-EINVAL);
2044         }
2045
2046         /*
2047          * Force any signals received before this point to be delivered
2048          * before the fork happens.  Collect up signals sent to multiple
2049          * processes that happen during the fork and delay them so that
2050          * they appear to happen after the fork.
2051          */
2052         sigemptyset(&delayed.signal);
2053         INIT_HLIST_NODE(&delayed.node);
2054
2055         spin_lock_irq(&current->sighand->siglock);
2056         if (!(clone_flags & CLONE_THREAD))
2057                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
2058         recalc_sigpending();
2059         spin_unlock_irq(&current->sighand->siglock);
2060         retval = -ERESTARTNOINTR;
2061         if (task_sigpending(current))
2062                 goto fork_out;
2063
2064         retval = -ENOMEM;
2065         p = dup_task_struct(current, node);
2066         if (!p)
2067                 goto fork_out;
2068         if (args->io_thread) {
2069                 /*
2070                  * Mark us an IO worker, and block any signal that isn't
2071                  * fatal or STOP
2072                  */
2073                 p->flags |= PF_IO_WORKER;
2074                 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2075         }
2076
2077         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2078         /*
2079          * Clear TID on mm_release()?
2080          */
2081         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2082
2083         ftrace_graph_init_task(p);
2084
2085         rt_mutex_init_task(p);
2086
2087         lockdep_assert_irqs_enabled();
2088 #ifdef CONFIG_PROVE_LOCKING
2089         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2090 #endif
2091         retval = copy_creds(p, clone_flags);
2092         if (retval < 0)
2093                 goto bad_fork_free;
2094
2095         retval = -EAGAIN;
2096         if (is_ucounts_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2097                 if (p->real_cred->user != INIT_USER &&
2098                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2099                         goto bad_fork_cleanup_count;
2100         }
2101         current->flags &= ~PF_NPROC_EXCEEDED;
2102
2103         /*
2104          * If multiple threads are within copy_process(), then this check
2105          * triggers too late. This doesn't hurt, the check is only there
2106          * to stop root fork bombs.
2107          */
2108         retval = -EAGAIN;
2109         if (data_race(nr_threads >= max_threads))
2110                 goto bad_fork_cleanup_count;
2111
2112         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
2113         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2114         p->flags |= PF_FORKNOEXEC;
2115         INIT_LIST_HEAD(&p->children);
2116         INIT_LIST_HEAD(&p->sibling);
2117         rcu_copy_process(p);
2118         p->vfork_done = NULL;
2119         spin_lock_init(&p->alloc_lock);
2120
2121         init_sigpending(&p->pending);
2122
2123         p->utime = p->stime = p->gtime = 0;
2124 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2125         p->utimescaled = p->stimescaled = 0;
2126 #endif
2127         prev_cputime_init(&p->prev_cputime);
2128
2129 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2130         seqcount_init(&p->vtime.seqcount);
2131         p->vtime.starttime = 0;
2132         p->vtime.state = VTIME_INACTIVE;
2133 #endif
2134
2135 #ifdef CONFIG_IO_URING
2136         p->io_uring = NULL;
2137 #endif
2138
2139 #if defined(SPLIT_RSS_COUNTING)
2140         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2141 #endif
2142
2143         p->default_timer_slack_ns = current->timer_slack_ns;
2144
2145 #ifdef CONFIG_PSI
2146         p->psi_flags = 0;
2147 #endif
2148
2149         task_io_accounting_init(&p->ioac);
2150         acct_clear_integrals(p);
2151
2152         posix_cputimers_init(&p->posix_cputimers);
2153
2154         p->io_context = NULL;
2155         audit_set_context(p, NULL);
2156         cgroup_fork(p);
2157         if (p->flags & PF_KTHREAD) {
2158                 if (!set_kthread_struct(p))
2159                         goto bad_fork_cleanup_delayacct;
2160         }
2161 #ifdef CONFIG_NUMA
2162         p->mempolicy = mpol_dup(p->mempolicy);
2163         if (IS_ERR(p->mempolicy)) {
2164                 retval = PTR_ERR(p->mempolicy);
2165                 p->mempolicy = NULL;
2166                 goto bad_fork_cleanup_delayacct;
2167         }
2168 #endif
2169 #ifdef CONFIG_CPUSETS
2170         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2171         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2172         seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2173 #endif
2174 #ifdef CONFIG_TRACE_IRQFLAGS
2175         memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2176         p->irqtrace.hardirq_disable_ip  = _THIS_IP_;
2177         p->irqtrace.softirq_enable_ip   = _THIS_IP_;
2178         p->softirqs_enabled             = 1;
2179         p->softirq_context              = 0;
2180 #endif
2181
2182         p->pagefault_disabled = 0;
2183
2184 #ifdef CONFIG_LOCKDEP
2185         lockdep_init_task(p);
2186 #endif
2187
2188 #ifdef CONFIG_DEBUG_MUTEXES
2189         p->blocked_on = NULL; /* not blocked yet */
2190 #endif
2191 #ifdef CONFIG_BCACHE
2192         p->sequential_io        = 0;
2193         p->sequential_io_avg    = 0;
2194 #endif
2195 #ifdef CONFIG_BPF_SYSCALL
2196         RCU_INIT_POINTER(p->bpf_storage, NULL);
2197         p->bpf_ctx = NULL;
2198 #endif
2199
2200         /* Perform scheduler related setup. Assign this task to a CPU. */
2201         retval = sched_fork(clone_flags, p);
2202         if (retval)
2203                 goto bad_fork_cleanup_policy;
2204
2205         retval = perf_event_init_task(p, clone_flags);
2206         if (retval)
2207                 goto bad_fork_cleanup_policy;
2208         retval = audit_alloc(p);
2209         if (retval)
2210                 goto bad_fork_cleanup_perf;
2211         /* copy all the process information */
2212         shm_init_task(p);
2213         retval = security_task_alloc(p, clone_flags);
2214         if (retval)
2215                 goto bad_fork_cleanup_audit;
2216         retval = copy_semundo(clone_flags, p);
2217         if (retval)
2218                 goto bad_fork_cleanup_security;
2219         retval = copy_files(clone_flags, p);
2220         if (retval)
2221                 goto bad_fork_cleanup_semundo;
2222         retval = copy_fs(clone_flags, p);
2223         if (retval)
2224                 goto bad_fork_cleanup_files;
2225         retval = copy_sighand(clone_flags, p);
2226         if (retval)
2227                 goto bad_fork_cleanup_fs;
2228         retval = copy_signal(clone_flags, p);
2229         if (retval)
2230                 goto bad_fork_cleanup_sighand;
2231         retval = copy_mm(clone_flags, p);
2232         if (retval)
2233                 goto bad_fork_cleanup_signal;
2234         retval = copy_namespaces(clone_flags, p);
2235         if (retval)
2236                 goto bad_fork_cleanup_mm;
2237         retval = copy_io(clone_flags, p);
2238         if (retval)
2239                 goto bad_fork_cleanup_namespaces;
2240         retval = copy_thread(clone_flags, args->stack, args->stack_size, p, args->tls);
2241         if (retval)
2242                 goto bad_fork_cleanup_io;
2243
2244         stackleak_task_init(p);
2245
2246         if (pid != &init_struct_pid) {
2247                 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2248                                 args->set_tid_size);
2249                 if (IS_ERR(pid)) {
2250                         retval = PTR_ERR(pid);
2251                         goto bad_fork_cleanup_thread;
2252                 }
2253         }
2254
2255         /*
2256          * This has to happen after we've potentially unshared the file
2257          * descriptor table (so that the pidfd doesn't leak into the child
2258          * if the fd table isn't shared).
2259          */
2260         if (clone_flags & CLONE_PIDFD) {
2261                 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2262                 if (retval < 0)
2263                         goto bad_fork_free_pid;
2264
2265                 pidfd = retval;
2266
2267                 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2268                                               O_RDWR | O_CLOEXEC);
2269                 if (IS_ERR(pidfile)) {
2270                         put_unused_fd(pidfd);
2271                         retval = PTR_ERR(pidfile);
2272                         goto bad_fork_free_pid;
2273                 }
2274                 get_pid(pid);   /* held by pidfile now */
2275
2276                 retval = put_user(pidfd, args->pidfd);
2277                 if (retval)
2278                         goto bad_fork_put_pidfd;
2279         }
2280
2281 #ifdef CONFIG_BLOCK
2282         p->plug = NULL;
2283 #endif
2284         futex_init_task(p);
2285
2286         /*
2287          * sigaltstack should be cleared when sharing the same VM
2288          */
2289         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2290                 sas_ss_reset(p);
2291
2292         /*
2293          * Syscall tracing and stepping should be turned off in the
2294          * child regardless of CLONE_PTRACE.
2295          */
2296         user_disable_single_step(p);
2297         clear_task_syscall_work(p, SYSCALL_TRACE);
2298 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2299         clear_task_syscall_work(p, SYSCALL_EMU);
2300 #endif
2301         clear_tsk_latency_tracing(p);
2302
2303         /* ok, now we should be set up.. */
2304         p->pid = pid_nr(pid);
2305         if (clone_flags & CLONE_THREAD) {
2306                 p->group_leader = current->group_leader;
2307                 p->tgid = current->tgid;
2308         } else {
2309                 p->group_leader = p;
2310                 p->tgid = p->pid;
2311         }
2312
2313         p->nr_dirtied = 0;
2314         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2315         p->dirty_paused_when = 0;
2316
2317         p->pdeath_signal = 0;
2318         INIT_LIST_HEAD(&p->thread_group);
2319         p->task_works = NULL;
2320         clear_posix_cputimers_work(p);
2321
2322 #ifdef CONFIG_KRETPROBES
2323         p->kretprobe_instances.first = NULL;
2324 #endif
2325 #ifdef CONFIG_RETHOOK
2326         p->rethooks.first = NULL;
2327 #endif
2328
2329         /*
2330          * Ensure that the cgroup subsystem policies allow the new process to be
2331          * forked. It should be noted that the new process's css_set can be changed
2332          * between here and cgroup_post_fork() if an organisation operation is in
2333          * progress.
2334          */
2335         retval = cgroup_can_fork(p, args);
2336         if (retval)
2337                 goto bad_fork_put_pidfd;
2338
2339         /*
2340          * Now that the cgroups are pinned, re-clone the parent cgroup and put
2341          * the new task on the correct runqueue. All this *before* the task
2342          * becomes visible.
2343          *
2344          * This isn't part of ->can_fork() because while the re-cloning is
2345          * cgroup specific, it unconditionally needs to place the task on a
2346          * runqueue.
2347          */
2348         sched_cgroup_fork(p, args);
2349
2350         /*
2351          * From this point on we must avoid any synchronous user-space
2352          * communication until we take the tasklist-lock. In particular, we do
2353          * not want user-space to be able to predict the process start-time by
2354          * stalling fork(2) after we recorded the start_time but before it is
2355          * visible to the system.
2356          */
2357
2358         p->start_time = ktime_get_ns();
2359         p->start_boottime = ktime_get_boottime_ns();
2360
2361         /*
2362          * Make it visible to the rest of the system, but dont wake it up yet.
2363          * Need tasklist lock for parent etc handling!
2364          */
2365         write_lock_irq(&tasklist_lock);
2366
2367         /* CLONE_PARENT re-uses the old parent */
2368         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2369                 p->real_parent = current->real_parent;
2370                 p->parent_exec_id = current->parent_exec_id;
2371                 if (clone_flags & CLONE_THREAD)
2372                         p->exit_signal = -1;
2373                 else
2374                         p->exit_signal = current->group_leader->exit_signal;
2375         } else {
2376                 p->real_parent = current;
2377                 p->parent_exec_id = current->self_exec_id;
2378                 p->exit_signal = args->exit_signal;
2379         }
2380
2381         klp_copy_process(p);
2382
2383         sched_core_fork(p);
2384
2385         spin_lock(&current->sighand->siglock);
2386
2387         /*
2388          * Copy seccomp details explicitly here, in case they were changed
2389          * before holding sighand lock.
2390          */
2391         copy_seccomp(p);
2392
2393         rseq_fork(p, clone_flags);
2394
2395         /* Don't start children in a dying pid namespace */
2396         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2397                 retval = -ENOMEM;
2398                 goto bad_fork_cancel_cgroup;
2399         }
2400
2401         /* Let kill terminate clone/fork in the middle */
2402         if (fatal_signal_pending(current)) {
2403                 retval = -EINTR;
2404                 goto bad_fork_cancel_cgroup;
2405         }
2406
2407         init_task_pid_links(p);
2408         if (likely(p->pid)) {
2409                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2410
2411                 init_task_pid(p, PIDTYPE_PID, pid);
2412                 if (thread_group_leader(p)) {
2413                         init_task_pid(p, PIDTYPE_TGID, pid);
2414                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2415                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2416
2417                         if (is_child_reaper(pid)) {
2418                                 ns_of_pid(pid)->child_reaper = p;
2419                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2420                         }
2421                         p->signal->shared_pending.signal = delayed.signal;
2422                         p->signal->tty = tty_kref_get(current->signal->tty);
2423                         /*
2424                          * Inherit has_child_subreaper flag under the same
2425                          * tasklist_lock with adding child to the process tree
2426                          * for propagate_has_child_subreaper optimization.
2427                          */
2428                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2429                                                          p->real_parent->signal->is_child_subreaper;
2430                         list_add_tail(&p->sibling, &p->real_parent->children);
2431                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2432                         attach_pid(p, PIDTYPE_TGID);
2433                         attach_pid(p, PIDTYPE_PGID);
2434                         attach_pid(p, PIDTYPE_SID);
2435                         __this_cpu_inc(process_counts);
2436                 } else {
2437                         current->signal->nr_threads++;
2438                         atomic_inc(&current->signal->live);
2439                         refcount_inc(&current->signal->sigcnt);
2440                         task_join_group_stop(p);
2441                         list_add_tail_rcu(&p->thread_group,
2442                                           &p->group_leader->thread_group);
2443                         list_add_tail_rcu(&p->thread_node,
2444                                           &p->signal->thread_head);
2445                 }
2446                 attach_pid(p, PIDTYPE_PID);
2447                 nr_threads++;
2448         }
2449         total_forks++;
2450         hlist_del_init(&delayed.node);
2451         spin_unlock(&current->sighand->siglock);
2452         syscall_tracepoint_update(p);
2453         write_unlock_irq(&tasklist_lock);
2454
2455         if (pidfile)
2456                 fd_install(pidfd, pidfile);
2457
2458         proc_fork_connector(p);
2459         sched_post_fork(p);
2460         cgroup_post_fork(p, args);
2461         perf_event_fork(p);
2462
2463         trace_task_newtask(p, clone_flags);
2464         uprobe_copy_process(p, clone_flags);
2465
2466         copy_oom_score_adj(clone_flags, p);
2467
2468         return p;
2469
2470 bad_fork_cancel_cgroup:
2471         sched_core_free(p);
2472         spin_unlock(&current->sighand->siglock);
2473         write_unlock_irq(&tasklist_lock);
2474         cgroup_cancel_fork(p, args);
2475 bad_fork_put_pidfd:
2476         if (clone_flags & CLONE_PIDFD) {
2477                 fput(pidfile);
2478                 put_unused_fd(pidfd);
2479         }
2480 bad_fork_free_pid:
2481         if (pid != &init_struct_pid)
2482                 free_pid(pid);
2483 bad_fork_cleanup_thread:
2484         exit_thread(p);
2485 bad_fork_cleanup_io:
2486         if (p->io_context)
2487                 exit_io_context(p);
2488 bad_fork_cleanup_namespaces:
2489         exit_task_namespaces(p);
2490 bad_fork_cleanup_mm:
2491         if (p->mm) {
2492                 mm_clear_owner(p->mm, p);
2493                 mmput(p->mm);
2494         }
2495 bad_fork_cleanup_signal:
2496         if (!(clone_flags & CLONE_THREAD))
2497                 free_signal_struct(p->signal);
2498 bad_fork_cleanup_sighand:
2499         __cleanup_sighand(p->sighand);
2500 bad_fork_cleanup_fs:
2501         exit_fs(p); /* blocking */
2502 bad_fork_cleanup_files:
2503         exit_files(p); /* blocking */
2504 bad_fork_cleanup_semundo:
2505         exit_sem(p);
2506 bad_fork_cleanup_security:
2507         security_task_free(p);
2508 bad_fork_cleanup_audit:
2509         audit_free(p);
2510 bad_fork_cleanup_perf:
2511         perf_event_free_task(p);
2512 bad_fork_cleanup_policy:
2513         lockdep_free_task(p);
2514 #ifdef CONFIG_NUMA
2515         mpol_put(p->mempolicy);
2516 #endif
2517 bad_fork_cleanup_delayacct:
2518         delayacct_tsk_free(p);
2519 bad_fork_cleanup_count:
2520         dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2521         exit_creds(p);
2522 bad_fork_free:
2523         WRITE_ONCE(p->__state, TASK_DEAD);
2524         exit_task_stack_account(p);
2525         put_task_stack(p);
2526         delayed_free_task(p);
2527 fork_out:
2528         spin_lock_irq(&current->sighand->siglock);
2529         hlist_del_init(&delayed.node);
2530         spin_unlock_irq(&current->sighand->siglock);
2531         return ERR_PTR(retval);
2532 }
2533
2534 static inline void init_idle_pids(struct task_struct *idle)
2535 {
2536         enum pid_type type;
2537
2538         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2539                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2540                 init_task_pid(idle, type, &init_struct_pid);
2541         }
2542 }
2543
2544 struct task_struct * __init fork_idle(int cpu)
2545 {
2546         struct task_struct *task;
2547         struct kernel_clone_args args = {
2548                 .flags = CLONE_VM,
2549         };
2550
2551         task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2552         if (!IS_ERR(task)) {
2553                 init_idle_pids(task);
2554                 init_idle(task, cpu);
2555         }
2556
2557         return task;
2558 }
2559
2560 struct mm_struct *copy_init_mm(void)
2561 {
2562         return dup_mm(NULL, &init_mm);
2563 }
2564
2565 /*
2566  * This is like kernel_clone(), but shaved down and tailored to just
2567  * creating io_uring workers. It returns a created task, or an error pointer.
2568  * The returned task is inactive, and the caller must fire it up through
2569  * wake_up_new_task(p). All signals are blocked in the created task.
2570  */
2571 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2572 {
2573         unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2574                                 CLONE_IO;
2575         struct kernel_clone_args args = {
2576                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2577                                     CLONE_UNTRACED) & ~CSIGNAL),
2578                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2579                 .stack          = (unsigned long)fn,
2580                 .stack_size     = (unsigned long)arg,
2581                 .io_thread      = 1,
2582         };
2583
2584         return copy_process(NULL, 0, node, &args);
2585 }
2586
2587 /*
2588  *  Ok, this is the main fork-routine.
2589  *
2590  * It copies the process, and if successful kick-starts
2591  * it and waits for it to finish using the VM if required.
2592  *
2593  * args->exit_signal is expected to be checked for sanity by the caller.
2594  */
2595 pid_t kernel_clone(struct kernel_clone_args *args)
2596 {
2597         u64 clone_flags = args->flags;
2598         struct completion vfork;
2599         struct pid *pid;
2600         struct task_struct *p;
2601         int trace = 0;
2602         pid_t nr;
2603
2604         /*
2605          * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2606          * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2607          * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2608          * field in struct clone_args and it still doesn't make sense to have
2609          * them both point at the same memory location. Performing this check
2610          * here has the advantage that we don't need to have a separate helper
2611          * to check for legacy clone().
2612          */
2613         if ((args->flags & CLONE_PIDFD) &&
2614             (args->flags & CLONE_PARENT_SETTID) &&
2615             (args->pidfd == args->parent_tid))
2616                 return -EINVAL;
2617
2618         /*
2619          * Determine whether and which event to report to ptracer.  When
2620          * called from kernel_thread or CLONE_UNTRACED is explicitly
2621          * requested, no event is reported; otherwise, report if the event
2622          * for the type of forking is enabled.
2623          */
2624         if (!(clone_flags & CLONE_UNTRACED)) {
2625                 if (clone_flags & CLONE_VFORK)
2626                         trace = PTRACE_EVENT_VFORK;
2627                 else if (args->exit_signal != SIGCHLD)
2628                         trace = PTRACE_EVENT_CLONE;
2629                 else
2630                         trace = PTRACE_EVENT_FORK;
2631
2632                 if (likely(!ptrace_event_enabled(current, trace)))
2633                         trace = 0;
2634         }
2635
2636         p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2637         add_latent_entropy();
2638
2639         if (IS_ERR(p))
2640                 return PTR_ERR(p);
2641
2642         /*
2643          * Do this prior waking up the new thread - the thread pointer
2644          * might get invalid after that point, if the thread exits quickly.
2645          */
2646         trace_sched_process_fork(current, p);
2647
2648         pid = get_task_pid(p, PIDTYPE_PID);
2649         nr = pid_vnr(pid);
2650
2651         if (clone_flags & CLONE_PARENT_SETTID)
2652                 put_user(nr, args->parent_tid);
2653
2654         if (clone_flags & CLONE_VFORK) {
2655                 p->vfork_done = &vfork;
2656                 init_completion(&vfork);
2657                 get_task_struct(p);
2658         }
2659
2660         wake_up_new_task(p);
2661
2662         /* forking complete and child started to run, tell ptracer */
2663         if (unlikely(trace))
2664                 ptrace_event_pid(trace, pid);
2665
2666         if (clone_flags & CLONE_VFORK) {
2667                 if (!wait_for_vfork_done(p, &vfork))
2668                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2669         }
2670
2671         put_pid(pid);
2672         return nr;
2673 }
2674
2675 /*
2676  * Create a kernel thread.
2677  */
2678 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2679 {
2680         struct kernel_clone_args args = {
2681                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2682                                     CLONE_UNTRACED) & ~CSIGNAL),
2683                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2684                 .stack          = (unsigned long)fn,
2685                 .stack_size     = (unsigned long)arg,
2686         };
2687
2688         return kernel_clone(&args);
2689 }
2690
2691 #ifdef __ARCH_WANT_SYS_FORK
2692 SYSCALL_DEFINE0(fork)
2693 {
2694 #ifdef CONFIG_MMU
2695         struct kernel_clone_args args = {
2696                 .exit_signal = SIGCHLD,
2697         };
2698
2699         return kernel_clone(&args);
2700 #else
2701         /* can not support in nommu mode */
2702         return -EINVAL;
2703 #endif
2704 }
2705 #endif
2706
2707 #ifdef __ARCH_WANT_SYS_VFORK
2708 SYSCALL_DEFINE0(vfork)
2709 {
2710         struct kernel_clone_args args = {
2711                 .flags          = CLONE_VFORK | CLONE_VM,
2712                 .exit_signal    = SIGCHLD,
2713         };
2714
2715         return kernel_clone(&args);
2716 }
2717 #endif
2718
2719 #ifdef __ARCH_WANT_SYS_CLONE
2720 #ifdef CONFIG_CLONE_BACKWARDS
2721 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2722                  int __user *, parent_tidptr,
2723                  unsigned long, tls,
2724                  int __user *, child_tidptr)
2725 #elif defined(CONFIG_CLONE_BACKWARDS2)
2726 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2727                  int __user *, parent_tidptr,
2728                  int __user *, child_tidptr,
2729                  unsigned long, tls)
2730 #elif defined(CONFIG_CLONE_BACKWARDS3)
2731 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2732                 int, stack_size,
2733                 int __user *, parent_tidptr,
2734                 int __user *, child_tidptr,
2735                 unsigned long, tls)
2736 #else
2737 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2738                  int __user *, parent_tidptr,
2739                  int __user *, child_tidptr,
2740                  unsigned long, tls)
2741 #endif
2742 {
2743         struct kernel_clone_args args = {
2744                 .flags          = (lower_32_bits(clone_flags) & ~CSIGNAL),
2745                 .pidfd          = parent_tidptr,
2746                 .child_tid      = child_tidptr,
2747                 .parent_tid     = parent_tidptr,
2748                 .exit_signal    = (lower_32_bits(clone_flags) & CSIGNAL),
2749                 .stack          = newsp,
2750                 .tls            = tls,
2751         };
2752
2753         return kernel_clone(&args);
2754 }
2755 #endif
2756
2757 #ifdef __ARCH_WANT_SYS_CLONE3
2758
2759 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2760                                               struct clone_args __user *uargs,
2761                                               size_t usize)
2762 {
2763         int err;
2764         struct clone_args args;
2765         pid_t *kset_tid = kargs->set_tid;
2766
2767         BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2768                      CLONE_ARGS_SIZE_VER0);
2769         BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2770                      CLONE_ARGS_SIZE_VER1);
2771         BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2772                      CLONE_ARGS_SIZE_VER2);
2773         BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2774
2775         if (unlikely(usize > PAGE_SIZE))
2776                 return -E2BIG;
2777         if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2778                 return -EINVAL;
2779
2780         err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2781         if (err)
2782                 return err;
2783
2784         if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2785                 return -EINVAL;
2786
2787         if (unlikely(!args.set_tid && args.set_tid_size > 0))
2788                 return -EINVAL;
2789
2790         if (unlikely(args.set_tid && args.set_tid_size == 0))
2791                 return -EINVAL;
2792
2793         /*
2794          * Verify that higher 32bits of exit_signal are unset and that
2795          * it is a valid signal
2796          */
2797         if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2798                      !valid_signal(args.exit_signal)))
2799                 return -EINVAL;
2800
2801         if ((args.flags & CLONE_INTO_CGROUP) &&
2802             (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2803                 return -EINVAL;
2804
2805         *kargs = (struct kernel_clone_args){
2806                 .flags          = args.flags,
2807                 .pidfd          = u64_to_user_ptr(args.pidfd),
2808                 .child_tid      = u64_to_user_ptr(args.child_tid),
2809                 .parent_tid     = u64_to_user_ptr(args.parent_tid),
2810                 .exit_signal    = args.exit_signal,
2811                 .stack          = args.stack,
2812                 .stack_size     = args.stack_size,
2813                 .tls            = args.tls,
2814                 .set_tid_size   = args.set_tid_size,
2815                 .cgroup         = args.cgroup,
2816         };
2817
2818         if (args.set_tid &&
2819                 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2820                         (kargs->set_tid_size * sizeof(pid_t))))
2821                 return -EFAULT;
2822
2823         kargs->set_tid = kset_tid;
2824
2825         return 0;
2826 }
2827
2828 /**
2829  * clone3_stack_valid - check and prepare stack
2830  * @kargs: kernel clone args
2831  *
2832  * Verify that the stack arguments userspace gave us are sane.
2833  * In addition, set the stack direction for userspace since it's easy for us to
2834  * determine.
2835  */
2836 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2837 {
2838         if (kargs->stack == 0) {
2839                 if (kargs->stack_size > 0)
2840                         return false;
2841         } else {
2842                 if (kargs->stack_size == 0)
2843                         return false;
2844
2845                 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2846                         return false;
2847
2848 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2849                 kargs->stack += kargs->stack_size;
2850 #endif
2851         }
2852
2853         return true;
2854 }
2855
2856 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2857 {
2858         /* Verify that no unknown flags are passed along. */
2859         if (kargs->flags &
2860             ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2861                 return false;
2862
2863         /*
2864          * - make the CLONE_DETACHED bit reusable for clone3
2865          * - make the CSIGNAL bits reusable for clone3
2866          */
2867         if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2868                 return false;
2869
2870         if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2871             (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2872                 return false;
2873
2874         if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2875             kargs->exit_signal)
2876                 return false;
2877
2878         if (!clone3_stack_valid(kargs))
2879                 return false;
2880
2881         return true;
2882 }
2883
2884 /**
2885  * clone3 - create a new process with specific properties
2886  * @uargs: argument structure
2887  * @size:  size of @uargs
2888  *
2889  * clone3() is the extensible successor to clone()/clone2().
2890  * It takes a struct as argument that is versioned by its size.
2891  *
2892  * Return: On success, a positive PID for the child process.
2893  *         On error, a negative errno number.
2894  */
2895 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2896 {
2897         int err;
2898
2899         struct kernel_clone_args kargs;
2900         pid_t set_tid[MAX_PID_NS_LEVEL];
2901
2902         kargs.set_tid = set_tid;
2903
2904         err = copy_clone_args_from_user(&kargs, uargs, size);
2905         if (err)
2906                 return err;
2907
2908         if (!clone3_args_valid(&kargs))
2909                 return -EINVAL;
2910
2911         return kernel_clone(&kargs);
2912 }
2913 #endif
2914
2915 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2916 {
2917         struct task_struct *leader, *parent, *child;
2918         int res;
2919
2920         read_lock(&tasklist_lock);
2921         leader = top = top->group_leader;
2922 down:
2923         for_each_thread(leader, parent) {
2924                 list_for_each_entry(child, &parent->children, sibling) {
2925                         res = visitor(child, data);
2926                         if (res) {
2927                                 if (res < 0)
2928                                         goto out;
2929                                 leader = child;
2930                                 goto down;
2931                         }
2932 up:
2933                         ;
2934                 }
2935         }
2936
2937         if (leader != top) {
2938                 child = leader;
2939                 parent = child->real_parent;
2940                 leader = parent->group_leader;
2941                 goto up;
2942         }
2943 out:
2944         read_unlock(&tasklist_lock);
2945 }
2946
2947 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2948 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2949 #endif
2950
2951 static void sighand_ctor(void *data)
2952 {
2953         struct sighand_struct *sighand = data;
2954
2955         spin_lock_init(&sighand->siglock);
2956         init_waitqueue_head(&sighand->signalfd_wqh);
2957 }
2958
2959 void __init proc_caches_init(void)
2960 {
2961         unsigned int mm_size;
2962
2963         sighand_cachep = kmem_cache_create("sighand_cache",
2964                         sizeof(struct sighand_struct), 0,
2965                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2966                         SLAB_ACCOUNT, sighand_ctor);
2967         signal_cachep = kmem_cache_create("signal_cache",
2968                         sizeof(struct signal_struct), 0,
2969                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2970                         NULL);
2971         files_cachep = kmem_cache_create("files_cache",
2972                         sizeof(struct files_struct), 0,
2973                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2974                         NULL);
2975         fs_cachep = kmem_cache_create("fs_cache",
2976                         sizeof(struct fs_struct), 0,
2977                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2978                         NULL);
2979
2980         /*
2981          * The mm_cpumask is located at the end of mm_struct, and is
2982          * dynamically sized based on the maximum CPU number this system
2983          * can have, taking hotplug into account (nr_cpu_ids).
2984          */
2985         mm_size = sizeof(struct mm_struct) + cpumask_size();
2986
2987         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2988                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2989                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2990                         offsetof(struct mm_struct, saved_auxv),
2991                         sizeof_field(struct mm_struct, saved_auxv),
2992                         NULL);
2993         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2994         mmap_init();
2995         nsproxy_cache_init();
2996 }
2997
2998 /*
2999  * Check constraints on flags passed to the unshare system call.
3000  */
3001 static int check_unshare_flags(unsigned long unshare_flags)
3002 {
3003         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3004                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3005                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3006                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3007                                 CLONE_NEWTIME))
3008                 return -EINVAL;
3009         /*
3010          * Not implemented, but pretend it works if there is nothing
3011          * to unshare.  Note that unsharing the address space or the
3012          * signal handlers also need to unshare the signal queues (aka
3013          * CLONE_THREAD).
3014          */
3015         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3016                 if (!thread_group_empty(current))
3017                         return -EINVAL;
3018         }
3019         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3020                 if (refcount_read(&current->sighand->count) > 1)
3021                         return -EINVAL;
3022         }
3023         if (unshare_flags & CLONE_VM) {
3024                 if (!current_is_single_threaded())
3025                         return -EINVAL;
3026         }
3027
3028         return 0;
3029 }
3030
3031 /*
3032  * Unshare the filesystem structure if it is being shared
3033  */
3034 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3035 {
3036         struct fs_struct *fs = current->fs;
3037
3038         if (!(unshare_flags & CLONE_FS) || !fs)
3039                 return 0;
3040
3041         /* don't need lock here; in the worst case we'll do useless copy */
3042         if (fs->users == 1)
3043                 return 0;
3044
3045         *new_fsp = copy_fs_struct(fs);
3046         if (!*new_fsp)
3047                 return -ENOMEM;
3048
3049         return 0;
3050 }
3051
3052 /*
3053  * Unshare file descriptor table if it is being shared
3054  */
3055 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3056                struct files_struct **new_fdp)
3057 {
3058         struct files_struct *fd = current->files;
3059         int error = 0;
3060
3061         if ((unshare_flags & CLONE_FILES) &&
3062             (fd && atomic_read(&fd->count) > 1)) {
3063                 *new_fdp = dup_fd(fd, max_fds, &error);
3064                 if (!*new_fdp)
3065                         return error;
3066         }
3067
3068         return 0;
3069 }
3070
3071 /*
3072  * unshare allows a process to 'unshare' part of the process
3073  * context which was originally shared using clone.  copy_*
3074  * functions used by kernel_clone() cannot be used here directly
3075  * because they modify an inactive task_struct that is being
3076  * constructed. Here we are modifying the current, active,
3077  * task_struct.
3078  */
3079 int ksys_unshare(unsigned long unshare_flags)
3080 {
3081         struct fs_struct *fs, *new_fs = NULL;
3082         struct files_struct *new_fd = NULL;
3083         struct cred *new_cred = NULL;
3084         struct nsproxy *new_nsproxy = NULL;
3085         int do_sysvsem = 0;
3086         int err;
3087
3088         /*
3089          * If unsharing a user namespace must also unshare the thread group
3090          * and unshare the filesystem root and working directories.
3091          */
3092         if (unshare_flags & CLONE_NEWUSER)
3093                 unshare_flags |= CLONE_THREAD | CLONE_FS;
3094         /*
3095          * If unsharing vm, must also unshare signal handlers.
3096          */
3097         if (unshare_flags & CLONE_VM)
3098                 unshare_flags |= CLONE_SIGHAND;
3099         /*
3100          * If unsharing a signal handlers, must also unshare the signal queues.
3101          */
3102         if (unshare_flags & CLONE_SIGHAND)
3103                 unshare_flags |= CLONE_THREAD;
3104         /*
3105          * If unsharing namespace, must also unshare filesystem information.
3106          */
3107         if (unshare_flags & CLONE_NEWNS)
3108                 unshare_flags |= CLONE_FS;
3109
3110         err = check_unshare_flags(unshare_flags);
3111         if (err)
3112                 goto bad_unshare_out;
3113         /*
3114          * CLONE_NEWIPC must also detach from the undolist: after switching
3115          * to a new ipc namespace, the semaphore arrays from the old
3116          * namespace are unreachable.
3117          */
3118         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3119                 do_sysvsem = 1;
3120         err = unshare_fs(unshare_flags, &new_fs);
3121         if (err)
3122                 goto bad_unshare_out;
3123         err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3124         if (err)
3125                 goto bad_unshare_cleanup_fs;
3126         err = unshare_userns(unshare_flags, &new_cred);
3127         if (err)
3128                 goto bad_unshare_cleanup_fd;
3129         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3130                                          new_cred, new_fs);
3131         if (err)
3132                 goto bad_unshare_cleanup_cred;
3133
3134         if (new_cred) {
3135                 err = set_cred_ucounts(new_cred);
3136                 if (err)
3137                         goto bad_unshare_cleanup_cred;
3138         }
3139
3140         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3141                 if (do_sysvsem) {
3142                         /*
3143                          * CLONE_SYSVSEM is equivalent to sys_exit().
3144                          */
3145                         exit_sem(current);
3146                 }
3147                 if (unshare_flags & CLONE_NEWIPC) {
3148                         /* Orphan segments in old ns (see sem above). */
3149                         exit_shm(current);
3150                         shm_init_task(current);
3151                 }
3152
3153                 if (new_nsproxy)
3154                         switch_task_namespaces(current, new_nsproxy);
3155
3156                 task_lock(current);
3157
3158                 if (new_fs) {
3159                         fs = current->fs;
3160                         spin_lock(&fs->lock);
3161                         current->fs = new_fs;
3162                         if (--fs->users)
3163                                 new_fs = NULL;
3164                         else
3165                                 new_fs = fs;
3166                         spin_unlock(&fs->lock);
3167                 }
3168
3169                 if (new_fd)
3170                         swap(current->files, new_fd);
3171
3172                 task_unlock(current);
3173
3174                 if (new_cred) {
3175                         /* Install the new user namespace */
3176                         commit_creds(new_cred);
3177                         new_cred = NULL;
3178                 }
3179         }
3180
3181         perf_event_namespaces(current);
3182
3183 bad_unshare_cleanup_cred:
3184         if (new_cred)
3185                 put_cred(new_cred);
3186 bad_unshare_cleanup_fd:
3187         if (new_fd)
3188                 put_files_struct(new_fd);
3189
3190 bad_unshare_cleanup_fs:
3191         if (new_fs)
3192                 free_fs_struct(new_fs);
3193
3194 bad_unshare_out:
3195         return err;
3196 }
3197
3198 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3199 {
3200         return ksys_unshare(unshare_flags);
3201 }
3202
3203 /*
3204  *      Helper to unshare the files of the current task.
3205  *      We don't want to expose copy_files internals to
3206  *      the exec layer of the kernel.
3207  */
3208
3209 int unshare_files(void)
3210 {
3211         struct task_struct *task = current;
3212         struct files_struct *old, *copy = NULL;
3213         int error;
3214
3215         error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, &copy);
3216         if (error || !copy)
3217                 return error;
3218
3219         old = task->files;
3220         task_lock(task);
3221         task->files = copy;
3222         task_unlock(task);
3223         put_files_struct(old);
3224         return 0;
3225 }
3226
3227 int sysctl_max_threads(struct ctl_table *table, int write,
3228                        void *buffer, size_t *lenp, loff_t *ppos)
3229 {
3230         struct ctl_table t;
3231         int ret;
3232         int threads = max_threads;
3233         int min = 1;
3234         int max = MAX_THREADS;
3235
3236         t = *table;
3237         t.data = &threads;
3238         t.extra1 = &min;
3239         t.extra2 = &max;
3240
3241         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3242         if (ret || !write)
3243                 return ret;
3244
3245         max_threads = threads;
3246
3247         return 0;
3248 }