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