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