kernel/fork.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/kernel/fork.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  */
   7
   8 /*
   9  *  'fork.c' contains the help-routines for the 'fork' system call
  10  * (see also entry.S and others).
  11  * Fork is rather simple, once you get the hang of it, but the memory
  12  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  13  */
  14
  15 #include <linux/anon_inodes.h>
  16 #include <linux/slab.h>
  17 #include <linux/sched/autogroup.h>
  18 #include <linux/sched/mm.h>
  19 #include <linux/sched/coredump.h>
  20 #include <linux/sched/user.h>
  21 #include <linux/sched/numa_balancing.h>
  22 #include <linux/sched/stat.h>
  23 #include <linux/sched/task.h>
  24 #include <linux/sched/task_stack.h>
  25 #include <linux/sched/cputime.h>
  26 #include <linux/seq_file.h>
  27 #include <linux/rtmutex.h>
  28 #include <linux/init.h>
  29 #include <linux/unistd.h>
  30 #include <linux/module.h>
  31 #include <linux/vmalloc.h>
  32 #include <linux/completion.h>
  33 #include <linux/personality.h>
  34 #include <linux/mempolicy.h>
  35 #include <linux/sem.h>
  36 #include <linux/file.h>
  37 #include <linux/fdtable.h>
  38 #include <linux/iocontext.h>
  39 #include <linux/key.h>
  40 #include <linux/kmsan.h>
  41 #include <linux/binfmts.h>
  42 #include <linux/mman.h>
  43 #include <linux/mmu_notifier.h>
  44 #include <linux/fs.h>
  45 #include <linux/mm.h>
  46 #include <linux/mm_inline.h>
  47 #include <linux/nsproxy.h>
  48 #include <linux/capability.h>
  49 #include <linux/cpu.h>
  50 #include <linux/cgroup.h>
  51 #include <linux/security.h>
  52 #include <linux/hugetlb.h>
  53 #include <linux/seccomp.h>
  54 #include <linux/swap.h>
  55 #include <linux/syscalls.h>
  56 #include <linux/syscall_user_dispatch.h>
  57 #include <linux/jiffies.h>
  58 #include <linux/futex.h>
  59 #include <linux/compat.h>
  60 #include <linux/kthread.h>
  61 #include <linux/task_io_accounting_ops.h>
  62 #include <linux/rcupdate.h>
  63 #include <linux/ptrace.h>
  64 #include <linux/mount.h>
  65 #include <linux/audit.h>
  66 #include <linux/memcontrol.h>
  67 #include <linux/ftrace.h>
  68 #include <linux/proc_fs.h>
  69 #include <linux/profile.h>
  70 #include <linux/rmap.h>
  71 #include <linux/ksm.h>
  72 #include <linux/acct.h>
  73 #include <linux/userfaultfd_k.h>
  74 #include <linux/tsacct_kern.h>
  75 #include <linux/cn_proc.h>
  76 #include <linux/freezer.h>
  77 #include <linux/delayacct.h>
  78 #include <linux/taskstats_kern.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/stackprotector.h>
 101 #include <linux/user_events.h>
 102 #include <linux/iommu.h>
 103 #include <linux/rseq.h>
 104 #include <uapi/linux/pidfd.h>
 105 #include <linux/pidfs.h>
 106
 107 #include <asm/pgalloc.h>
 108 #include <linux/uaccess.h>
 109 #include <asm/mmu_context.h>
 110 #include <asm/cacheflush.h>
 111 #include <asm/tlbflush.h>
 112
 113 #include <trace/events/sched.h>
 114
 115 #define CREATE_TRACE_POINTS
 116 #include <trace/events/task.h>
 117
 118 /*
 119  * Minimum number of threads to boot the kernel
 120  */
 121 #define MIN_THREADS 20
 122
 123 /*
 124  * Maximum number of threads
 125  */
 126 #define MAX_THREADS FUTEX_TID_MASK
 127
 128 /*
 129  * Protected counters by write_lock_irq(&tasklist_lock)
 130  */
 131 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 132 int nr_threads;                 /* The idle threads do not count.. */
 133
 134 static int max_threads;         /* tunable limit on nr_threads */
 135
 136 #define NAMED_ARRAY_INDEX(x)    [x] = __stringify(x)
 137
 138 static const char * const resident_page_types[] = {
 139         NAMED_ARRAY_INDEX(MM_FILEPAGES),
 140         NAMED_ARRAY_INDEX(MM_ANONPAGES),
 141         NAMED_ARRAY_INDEX(MM_SWAPENTS),
 142         NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
 143 };
 144
 145 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 146
 147 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 148
 149 #ifdef CONFIG_PROVE_RCU
 150 int lockdep_tasklist_lock_is_held(void)
 151 {
 152         return lockdep_is_held(&tasklist_lock);
 153 }
 154 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 155 #endif /* #ifdef CONFIG_PROVE_RCU */
 156
 157 int nr_processes(void)
 158 {
 159         int cpu;
 160         int total = 0;
 161
 162         for_each_possible_cpu(cpu)
 163                 total += per_cpu(process_counts, cpu);
 164
 165         return total;
 166 }
 167
 168 void __weak arch_release_task_struct(struct task_struct *tsk)
 169 {
 170 }
 171
 172 static struct kmem_cache *task_struct_cachep;
 173
 174 static inline struct task_struct *alloc_task_struct_node(int node)
 175 {
 176         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 177 }
 178
 179 static inline void free_task_struct(struct task_struct *tsk)
 180 {
 181         kmem_cache_free(task_struct_cachep, tsk);
 182 }
 183
 184 /*
 185  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 186  * kmemcache based allocator.
 187  */
 188 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 189
 190 #  ifdef CONFIG_VMAP_STACK
 191 /*
 192  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 193  * flush.  Try to minimize the number of calls by caching stacks.
 194  */
 195 #define NR_CACHED_STACKS 2
 196 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 197
 198 struct vm_stack {
 199         struct rcu_head rcu;
 200         struct vm_struct *stack_vm_area;
 201 };
 202
 203 static bool try_release_thread_stack_to_cache(struct vm_struct *vm)
 204 {
 205         unsigned int i;
 206
 207         for (i = 0; i < NR_CACHED_STACKS; i++) {
 208                 if (this_cpu_cmpxchg(cached_stacks[i], NULL, vm) != NULL)
 209                         continue;
 210                 return true;
 211         }
 212         return false;
 213 }
 214
 215 static void thread_stack_free_rcu(struct rcu_head *rh)
 216 {
 217         struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
 218
 219         if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
 220                 return;
 221
 222         vfree(vm_stack);
 223 }
 224
 225 static void thread_stack_delayed_free(struct task_struct *tsk)
 226 {
 227         struct vm_stack *vm_stack = tsk->stack;
 228
 229         vm_stack->stack_vm_area = tsk->stack_vm_area;
 230         call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
 231 }
 232
 233 static int free_vm_stack_cache(unsigned int cpu)
 234 {
 235         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 236         int i;
 237
 238         for (i = 0; i < NR_CACHED_STACKS; i++) {
 239                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 240
 241                 if (!vm_stack)
 242                         continue;
 243
 244                 vfree(vm_stack->addr);
 245                 cached_vm_stacks[i] = NULL;
 246         }
 247
 248         return 0;
 249 }
 250
 251 static int memcg_charge_kernel_stack(struct vm_struct *vm)
 252 {
 253         int i;
 254         int ret;
 255         int nr_charged = 0;
 256
 257         BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 258
 259         for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 260                 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0);
 261                 if (ret)
 262                         goto err;
 263                 nr_charged++;
 264         }
 265         return 0;
 266 err:
 267         for (i = 0; i < nr_charged; i++)
 268                 memcg_kmem_uncharge_page(vm->pages[i], 0);
 269         return ret;
 270 }
 271
 272 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 273 {
 274         struct vm_struct *vm;
 275         void *stack;
 276         int i;
 277
 278         for (i = 0; i < NR_CACHED_STACKS; i++) {
 279                 struct vm_struct *s;
 280
 281                 s = this_cpu_xchg(cached_stacks[i], NULL);
 282
 283                 if (!s)
 284                         continue;
 285
 286                 /* Reset stack metadata. */
 287                 kasan_unpoison_range(s->addr, THREAD_SIZE);
 288
 289                 stack = kasan_reset_tag(s->addr);
 290
 291                 /* Clear stale pointers from reused stack. */
 292                 memset(stack, 0, THREAD_SIZE);
 293
 294                 if (memcg_charge_kernel_stack(s)) {
 295                         vfree(s->addr);
 296                         return -ENOMEM;
 297                 }
 298
 299                 tsk->stack_vm_area = s;
 300                 tsk->stack = stack;
 301                 return 0;
 302         }
 303
 304         /*
 305          * Allocated stacks are cached and later reused by new threads,
 306          * so memcg accounting is performed manually on assigning/releasing
 307          * stacks to tasks. Drop __GFP_ACCOUNT.
 308          */
 309         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 310                                      VMALLOC_START, VMALLOC_END,
 311                                      THREADINFO_GFP & ~__GFP_ACCOUNT,
 312                                      PAGE_KERNEL,
 313                                      0, node, __builtin_return_address(0));
 314         if (!stack)
 315                 return -ENOMEM;
 316
 317         vm = find_vm_area(stack);
 318         if (memcg_charge_kernel_stack(vm)) {
 319                 vfree(stack);
 320                 return -ENOMEM;
 321         }
 322         /*
 323          * We can't call find_vm_area() in interrupt context, and
 324          * free_thread_stack() can be called in interrupt context,
 325          * so cache the vm_struct.
 326          */
 327         tsk->stack_vm_area = vm;
 328         stack = kasan_reset_tag(stack);
 329         tsk->stack = stack;
 330         return 0;
 331 }
 332
 333 static void free_thread_stack(struct task_struct *tsk)
 334 {
 335         if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
 336                 thread_stack_delayed_free(tsk);
 337
 338         tsk->stack = NULL;
 339         tsk->stack_vm_area = NULL;
 340 }
 341
 342 #  else /* !CONFIG_VMAP_STACK */
 343
 344 static void thread_stack_free_rcu(struct rcu_head *rh)
 345 {
 346         __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
 347 }
 348
 349 static void thread_stack_delayed_free(struct task_struct *tsk)
 350 {
 351         struct rcu_head *rh = tsk->stack;
 352
 353         call_rcu(rh, thread_stack_free_rcu);
 354 }
 355
 356 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 357 {
 358         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 359                                              THREAD_SIZE_ORDER);
 360
 361         if (likely(page)) {
 362                 tsk->stack = kasan_reset_tag(page_address(page));
 363                 return 0;
 364         }
 365         return -ENOMEM;
 366 }
 367
 368 static void free_thread_stack(struct task_struct *tsk)
 369 {
 370         thread_stack_delayed_free(tsk);
 371         tsk->stack = NULL;
 372 }
 373
 374 #  endif /* CONFIG_VMAP_STACK */
 375 # else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
 376
 377 static struct kmem_cache *thread_stack_cache;
 378
 379 static void thread_stack_free_rcu(struct rcu_head *rh)
 380 {
 381         kmem_cache_free(thread_stack_cache, rh);
 382 }
 383
 384 static void thread_stack_delayed_free(struct task_struct *tsk)
 385 {
 386         struct rcu_head *rh = tsk->stack;
 387
 388         call_rcu(rh, thread_stack_free_rcu);
 389 }
 390
 391 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 392 {
 393         unsigned long *stack;
 394         stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 395         stack = kasan_reset_tag(stack);
 396         tsk->stack = stack;
 397         return stack ? 0 : -ENOMEM;
 398 }
 399
 400 static void free_thread_stack(struct task_struct *tsk)
 401 {
 402         thread_stack_delayed_free(tsk);
 403         tsk->stack = NULL;
 404 }
 405
 406 void thread_stack_cache_init(void)
 407 {
 408         thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 409                                         THREAD_SIZE, THREAD_SIZE, 0, 0,
 410                                         THREAD_SIZE, NULL);
 411         BUG_ON(thread_stack_cache == NULL);
 412 }
 413
 414 # endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
 415
 416 /* SLAB cache for signal_struct structures (tsk->signal) */
 417 static struct kmem_cache *signal_cachep;
 418
 419 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 420 struct kmem_cache *sighand_cachep;
 421
 422 /* SLAB cache for files_struct structures (tsk->files) */
 423 struct kmem_cache *files_cachep;
 424
 425 /* SLAB cache for fs_struct structures (tsk->fs) */
 426 struct kmem_cache *fs_cachep;
 427
 428 /* SLAB cache for vm_area_struct structures */
 429 static struct kmem_cache *vm_area_cachep;
 430
 431 /* SLAB cache for mm_struct structures (tsk->mm) */
 432 static struct kmem_cache *mm_cachep;
 433
 434 #ifdef CONFIG_PER_VMA_LOCK
 435
 436 /* SLAB cache for vm_area_struct.lock */
 437 static struct kmem_cache *vma_lock_cachep;
 438
 439 static bool vma_lock_alloc(struct vm_area_struct *vma)
 440 {
 441         vma->vm_lock = kmem_cache_alloc(vma_lock_cachep, GFP_KERNEL);
 442         if (!vma->vm_lock)
 443                 return false;
 444
 445         init_rwsem(&vma->vm_lock->lock);
 446         vma->vm_lock_seq = -1;
 447
 448         return true;
 449 }
 450
 451 static inline void vma_lock_free(struct vm_area_struct *vma)
 452 {
 453         kmem_cache_free(vma_lock_cachep, vma->vm_lock);
 454 }
 455
 456 #else /* CONFIG_PER_VMA_LOCK */
 457
 458 static inline bool vma_lock_alloc(struct vm_area_struct *vma) { return true; }
 459 static inline void vma_lock_free(struct vm_area_struct *vma) {}
 460
 461 #endif /* CONFIG_PER_VMA_LOCK */
 462
 463 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 464 {
 465         struct vm_area_struct *vma;
 466
 467         vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 468         if (!vma)
 469                 return NULL;
 470
 471         vma_init(vma, mm);
 472         if (!vma_lock_alloc(vma)) {
 473                 kmem_cache_free(vm_area_cachep, vma);
 474                 return NULL;
 475         }
 476
 477         return vma;
 478 }
 479
 480 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 481 {
 482         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 483
 484         if (!new)
 485                 return NULL;
 486
 487         ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
 488         ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
 489         /*
 490          * orig->shared.rb may be modified concurrently, but the clone
 491          * will be reinitialized.
 492          */
 493         data_race(memcpy(new, orig, sizeof(*new)));
 494         if (!vma_lock_alloc(new)) {
 495                 kmem_cache_free(vm_area_cachep, new);
 496                 return NULL;
 497         }
 498         INIT_LIST_HEAD(&new->anon_vma_chain);
 499         vma_numab_state_init(new);
 500         dup_anon_vma_name(orig, new);
 501
 502         return new;
 503 }
 504
 505 void __vm_area_free(struct vm_area_struct *vma)
 506 {
 507         vma_numab_state_free(vma);
 508         free_anon_vma_name(vma);
 509         vma_lock_free(vma);
 510         kmem_cache_free(vm_area_cachep, vma);
 511 }
 512
 513 #ifdef CONFIG_PER_VMA_LOCK
 514 static void vm_area_free_rcu_cb(struct rcu_head *head)
 515 {
 516         struct vm_area_struct *vma = container_of(head, struct vm_area_struct,
 517                                                   vm_rcu);
 518
 519         /* The vma should not be locked while being destroyed. */
 520         VM_BUG_ON_VMA(rwsem_is_locked(&vma->vm_lock->lock), vma);
 521         __vm_area_free(vma);
 522 }
 523 #endif
 524
 525 void vm_area_free(struct vm_area_struct *vma)
 526 {
 527 #ifdef CONFIG_PER_VMA_LOCK
 528         call_rcu(&vma->vm_rcu, vm_area_free_rcu_cb);
 529 #else
 530         __vm_area_free(vma);
 531 #endif
 532 }
 533
 534 static void account_kernel_stack(struct task_struct *tsk, int account)
 535 {
 536         if (IS_ENABLED(CONFIG_VMAP_STACK)) {
 537                 struct vm_struct *vm = task_stack_vm_area(tsk);
 538                 int i;
 539
 540                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 541                         mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
 542                                               account * (PAGE_SIZE / 1024));
 543         } else {
 544                 void *stack = task_stack_page(tsk);
 545
 546                 /* All stack pages are in the same node. */
 547                 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
 548                                       account * (THREAD_SIZE / 1024));
 549         }
 550 }
 551
 552 void exit_task_stack_account(struct task_struct *tsk)
 553 {
 554         account_kernel_stack(tsk, -1);
 555
 556         if (IS_ENABLED(CONFIG_VMAP_STACK)) {
 557                 struct vm_struct *vm;
 558                 int i;
 559
 560                 vm = task_stack_vm_area(tsk);
 561                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 562                         memcg_kmem_uncharge_page(vm->pages[i], 0);
 563         }
 564 }
 565
 566 static void release_task_stack(struct task_struct *tsk)
 567 {
 568         if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
 569                 return;  /* Better to leak the stack than to free prematurely */
 570
 571         free_thread_stack(tsk);
 572 }
 573
 574 #ifdef CONFIG_THREAD_INFO_IN_TASK
 575 void put_task_stack(struct task_struct *tsk)
 576 {
 577         if (refcount_dec_and_test(&tsk->stack_refcount))
 578                 release_task_stack(tsk);
 579 }
 580 #endif
 581
 582 void free_task(struct task_struct *tsk)
 583 {
 584 #ifdef CONFIG_SECCOMP
 585         WARN_ON_ONCE(tsk->seccomp.filter);
 586 #endif
 587         release_user_cpus_ptr(tsk);
 588         scs_release(tsk);
 589
 590 #ifndef CONFIG_THREAD_INFO_IN_TASK
 591         /*
 592          * The task is finally done with both the stack and thread_info,
 593          * so free both.
 594          */
 595         release_task_stack(tsk);
 596 #else
 597         /*
 598          * If the task had a separate stack allocation, it should be gone
 599          * by now.
 600          */
 601         WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
 602 #endif
 603         rt_mutex_debug_task_free(tsk);
 604         ftrace_graph_exit_task(tsk);
 605         arch_release_task_struct(tsk);
 606         if (tsk->flags & PF_KTHREAD)
 607                 free_kthread_struct(tsk);
 608         bpf_task_storage_free(tsk);
 609         free_task_struct(tsk);
 610 }
 611 EXPORT_SYMBOL(free_task);
 612
 613 static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
 614 {
 615         struct file *exe_file;
 616
 617         exe_file = get_mm_exe_file(oldmm);
 618         RCU_INIT_POINTER(mm->exe_file, exe_file);
 619         /*
 620          * We depend on the oldmm having properly denied write access to the
 621          * exe_file already.
 622          */
 623         if (exe_file && deny_write_access(exe_file))
 624                 pr_warn_once("deny_write_access() failed in %s\n", __func__);
 625 }
 626
 627 #ifdef CONFIG_MMU
 628 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 629                                         struct mm_struct *oldmm)
 630 {
 631         struct vm_area_struct *mpnt, *tmp;
 632         int retval;
 633         unsigned long charge = 0;
 634         LIST_HEAD(uf);
 635         VMA_ITERATOR(vmi, mm, 0);
 636
 637         uprobe_start_dup_mmap();
 638         if (mmap_write_lock_killable(oldmm)) {
 639                 retval = -EINTR;
 640                 goto fail_uprobe_end;
 641         }
 642         flush_cache_dup_mm(oldmm);
 643         uprobe_dup_mmap(oldmm, mm);
 644         /*
 645          * Not linked in yet - no deadlock potential:
 646          */
 647         mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
 648
 649         /* No ordering required: file already has been exposed. */
 650         dup_mm_exe_file(mm, oldmm);
 651
 652         mm->total_vm = oldmm->total_vm;
 653         mm->data_vm = oldmm->data_vm;
 654         mm->exec_vm = oldmm->exec_vm;
 655         mm->stack_vm = oldmm->stack_vm;
 656
 657         retval = ksm_fork(mm, oldmm);
 658         if (retval)
 659                 goto out;
 660         khugepaged_fork(mm, oldmm);
 661
 662         /* Use __mt_dup() to efficiently build an identical maple tree. */
 663         retval = __mt_dup(&oldmm->mm_mt, &mm->mm_mt, GFP_KERNEL);
 664         if (unlikely(retval))
 665                 goto out;
 666
 667         mt_clear_in_rcu(vmi.mas.tree);
 668         for_each_vma(vmi, mpnt) {
 669                 struct file *file;
 670
 671                 vma_start_write(mpnt);
 672                 if (mpnt->vm_flags & VM_DONTCOPY) {
 673                         retval = vma_iter_clear_gfp(&vmi, mpnt->vm_start,
 674                                                     mpnt->vm_end, GFP_KERNEL);
 675                         if (retval)
 676                                 goto loop_out;
 677
 678                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 679                         continue;
 680                 }
 681                 charge = 0;
 682                 /*
 683                  * Don't duplicate many vmas if we've been oom-killed (for
 684                  * example)
 685                  */
 686                 if (fatal_signal_pending(current)) {
 687                         retval = -EINTR;
 688                         goto loop_out;
 689                 }
 690                 if (mpnt->vm_flags & VM_ACCOUNT) {
 691                         unsigned long len = vma_pages(mpnt);
 692
 693                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 694                                 goto fail_nomem;
 695                         charge = len;
 696                 }
 697                 tmp = vm_area_dup(mpnt);
 698                 if (!tmp)
 699                         goto fail_nomem;
 700                 retval = vma_dup_policy(mpnt, tmp);
 701                 if (retval)
 702                         goto fail_nomem_policy;
 703                 tmp->vm_mm = mm;
 704                 retval = dup_userfaultfd(tmp, &uf);
 705                 if (retval)
 706                         goto fail_nomem_anon_vma_fork;
 707                 if (tmp->vm_flags & VM_WIPEONFORK) {
 708                         /*
 709                          * VM_WIPEONFORK gets a clean slate in the child.
 710                          * Don't prepare anon_vma until fault since we don't
 711                          * copy page for current vma.
 712                          */
 713                         tmp->anon_vma = NULL;
 714                 } else if (anon_vma_fork(tmp, mpnt))
 715                         goto fail_nomem_anon_vma_fork;
 716                 vm_flags_clear(tmp, VM_LOCKED_MASK);
 717                 file = tmp->vm_file;
 718                 if (file) {
 719                         struct address_space *mapping = file->f_mapping;
 720
 721                         get_file(file);
 722                         i_mmap_lock_write(mapping);
 723                         if (vma_is_shared_maywrite(tmp))
 724                                 mapping_allow_writable(mapping);
 725                         flush_dcache_mmap_lock(mapping);
 726                         /* insert tmp into the share list, just after mpnt */
 727                         vma_interval_tree_insert_after(tmp, mpnt,
 728                                         &mapping->i_mmap);
 729                         flush_dcache_mmap_unlock(mapping);
 730                         i_mmap_unlock_write(mapping);
 731                 }
 732
 733                 /*
 734                  * Copy/update hugetlb private vma information.
 735                  */
 736                 if (is_vm_hugetlb_page(tmp))
 737                         hugetlb_dup_vma_private(tmp);
 738
 739                 /*
 740                  * Link the vma into the MT. After using __mt_dup(), memory
 741                  * allocation is not necessary here, so it cannot fail.
 742                  */
 743                 vma_iter_bulk_store(&vmi, tmp);
 744
 745                 mm->map_count++;
 746                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 747                         retval = copy_page_range(tmp, mpnt);
 748
 749                 if (tmp->vm_ops && tmp->vm_ops->open)
 750                         tmp->vm_ops->open(tmp);
 751
 752                 if (retval) {
 753                         mpnt = vma_next(&vmi);
 754                         goto loop_out;
 755                 }
 756         }
 757         /* a new mm has just been created */
 758         retval = arch_dup_mmap(oldmm, mm);
 759 loop_out:
 760         vma_iter_free(&vmi);
 761         if (!retval) {
 762                 mt_set_in_rcu(vmi.mas.tree);
 763         } else if (mpnt) {
 764                 /*
 765                  * The entire maple tree has already been duplicated. If the
 766                  * mmap duplication fails, mark the failure point with
 767                  * XA_ZERO_ENTRY. In exit_mmap(), if this marker is encountered,
 768                  * stop releasing VMAs that have not been duplicated after this
 769                  * point.
 770                  */
 771                 mas_set_range(&vmi.mas, mpnt->vm_start, mpnt->vm_end - 1);
 772                 mas_store(&vmi.mas, XA_ZERO_ENTRY);
 773         }
 774 out:
 775         mmap_write_unlock(mm);
 776         flush_tlb_mm(oldmm);
 777         mmap_write_unlock(oldmm);
 778         dup_userfaultfd_complete(&uf);
 779 fail_uprobe_end:
 780         uprobe_end_dup_mmap();
 781         return retval;
 782
 783 fail_nomem_anon_vma_fork:
 784         mpol_put(vma_policy(tmp));
 785 fail_nomem_policy:
 786         vm_area_free(tmp);
 787 fail_nomem:
 788         retval = -ENOMEM;
 789         vm_unacct_memory(charge);
 790         goto loop_out;
 791 }
 792
 793 static inline int mm_alloc_pgd(struct mm_struct *mm)
 794 {
 795         mm->pgd = pgd_alloc(mm);
 796         if (unlikely(!mm->pgd))
 797                 return -ENOMEM;
 798         return 0;
 799 }
 800
 801 static inline void mm_free_pgd(struct mm_struct *mm)
 802 {
 803         pgd_free(mm, mm->pgd);
 804 }
 805 #else
 806 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 807 {
 808         mmap_write_lock(oldmm);
 809         dup_mm_exe_file(mm, oldmm);
 810         mmap_write_unlock(oldmm);
 811         return 0;
 812 }
 813 #define mm_alloc_pgd(mm)        (0)
 814 #define mm_free_pgd(mm)
 815 #endif /* CONFIG_MMU */
 816
 817 static void check_mm(struct mm_struct *mm)
 818 {
 819         int i;
 820
 821         BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
 822                          "Please make sure 'struct resident_page_types[]' is updated as well");
 823
 824         for (i = 0; i < NR_MM_COUNTERS; i++) {
 825                 long x = percpu_counter_sum(&mm->rss_stat[i]);
 826
 827                 if (unlikely(x))
 828                         pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
 829                                  mm, resident_page_types[i], x);
 830         }
 831
 832         if (mm_pgtables_bytes(mm))
 833                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 834                                 mm_pgtables_bytes(mm));
 835
 836 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 837         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 838 #endif
 839 }
 840
 841 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 842 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 843
 844 static void do_check_lazy_tlb(void *arg)
 845 {
 846         struct mm_struct *mm = arg;
 847
 848         WARN_ON_ONCE(current->active_mm == mm);
 849 }
 850
 851 static void do_shoot_lazy_tlb(void *arg)
 852 {
 853         struct mm_struct *mm = arg;
 854
 855         if (current->active_mm == mm) {
 856                 WARN_ON_ONCE(current->mm);
 857                 current->active_mm = &init_mm;
 858                 switch_mm(mm, &init_mm, current);
 859         }
 860 }
 861
 862 static void cleanup_lazy_tlbs(struct mm_struct *mm)
 863 {
 864         if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) {
 865                 /*
 866                  * In this case, lazy tlb mms are refounted and would not reach
 867                  * __mmdrop until all CPUs have switched away and mmdrop()ed.
 868                  */
 869                 return;
 870         }
 871
 872         /*
 873          * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
 874          * requires lazy mm users to switch to another mm when the refcount
 875          * drops to zero, before the mm is freed. This requires IPIs here to
 876          * switch kernel threads to init_mm.
 877          *
 878          * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
 879          * switch with the final userspace teardown TLB flush which leaves the
 880          * mm lazy on this CPU but no others, reducing the need for additional
 881          * IPIs here. There are cases where a final IPI is still required here,
 882          * such as the final mmdrop being performed on a different CPU than the
 883          * one exiting, or kernel threads using the mm when userspace exits.
 884          *
 885          * IPI overheads have not found to be expensive, but they could be
 886          * reduced in a number of possible ways, for example (roughly
 887          * increasing order of complexity):
 888          * - The last lazy reference created by exit_mm() could instead switch
 889          *   to init_mm, however it's probable this will run on the same CPU
 890          *   immediately afterwards, so this may not reduce IPIs much.
 891          * - A batch of mms requiring IPIs could be gathered and freed at once.
 892          * - CPUs store active_mm where it can be remotely checked without a
 893          *   lock, to filter out false-positives in the cpumask.
 894          * - After mm_users or mm_count reaches zero, switching away from the
 895          *   mm could clear mm_cpumask to reduce some IPIs, perhaps together
 896          *   with some batching or delaying of the final IPIs.
 897          * - A delayed freeing and RCU-like quiescing sequence based on mm
 898          *   switching to avoid IPIs completely.
 899          */
 900         on_each_cpu_mask(mm_cpumask(mm), do_shoot_lazy_tlb, (void *)mm, 1);
 901         if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES))
 902                 on_each_cpu(do_check_lazy_tlb, (void *)mm, 1);
 903 }
 904
 905 /*
 906  * Called when the last reference to the mm
 907  * is dropped: either by a lazy thread or by
 908  * mmput. Free the page directory and the mm.
 909  */
 910 void __mmdrop(struct mm_struct *mm)
 911 {
 912         BUG_ON(mm == &init_mm);
 913         WARN_ON_ONCE(mm == current->mm);
 914
 915         /* Ensure no CPUs are using this as their lazy tlb mm */
 916         cleanup_lazy_tlbs(mm);
 917
 918         WARN_ON_ONCE(mm == current->active_mm);
 919         mm_free_pgd(mm);
 920         destroy_context(mm);
 921         mmu_notifier_subscriptions_destroy(mm);
 922         check_mm(mm);
 923         put_user_ns(mm->user_ns);
 924         mm_pasid_drop(mm);
 925         mm_destroy_cid(mm);
 926         percpu_counter_destroy_many(mm->rss_stat, NR_MM_COUNTERS);
 927
 928         free_mm(mm);
 929 }
 930 EXPORT_SYMBOL_GPL(__mmdrop);
 931
 932 static void mmdrop_async_fn(struct work_struct *work)
 933 {
 934         struct mm_struct *mm;
 935
 936         mm = container_of(work, struct mm_struct, async_put_work);
 937         __mmdrop(mm);
 938 }
 939
 940 static void mmdrop_async(struct mm_struct *mm)
 941 {
 942         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 943                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 944                 schedule_work(&mm->async_put_work);
 945         }
 946 }
 947
 948 static inline void free_signal_struct(struct signal_struct *sig)
 949 {
 950         taskstats_tgid_free(sig);
 951         sched_autogroup_exit(sig);
 952         /*
 953          * __mmdrop is not safe to call from softirq context on x86 due to
 954          * pgd_dtor so postpone it to the async context
 955          */
 956         if (sig->oom_mm)
 957                 mmdrop_async(sig->oom_mm);
 958         kmem_cache_free(signal_cachep, sig);
 959 }
 960
 961 static inline void put_signal_struct(struct signal_struct *sig)
 962 {
 963         if (refcount_dec_and_test(&sig->sigcnt))
 964                 free_signal_struct(sig);
 965 }
 966
 967 void __put_task_struct(struct task_struct *tsk)
 968 {
 969         WARN_ON(!tsk->exit_state);
 970         WARN_ON(refcount_read(&tsk->usage));
 971         WARN_ON(tsk == current);
 972
 973         io_uring_free(tsk);
 974         cgroup_free(tsk);
 975         task_numa_free(tsk, true);
 976         security_task_free(tsk);
 977         exit_creds(tsk);
 978         delayacct_tsk_free(tsk);
 979         put_signal_struct(tsk->signal);
 980         sched_core_free(tsk);
 981         free_task(tsk);
 982 }
 983 EXPORT_SYMBOL_GPL(__put_task_struct);
 984
 985 void __put_task_struct_rcu_cb(struct rcu_head *rhp)
 986 {
 987         struct task_struct *task = container_of(rhp, struct task_struct, rcu);
 988
 989         __put_task_struct(task);
 990 }
 991 EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb);
 992
 993 void __init __weak arch_task_cache_init(void) { }
 994
 995 /*
 996  * set_max_threads
 997  */
 998 static void set_max_threads(unsigned int max_threads_suggested)
 999 {
1000         u64 threads;
1001         unsigned long nr_pages = totalram_pages();
1002
1003         /*
1004          * The number of threads shall be limited such that the thread
1005          * structures may only consume a small part of the available memory.
1006          */
1007         if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
1008                 threads = MAX_THREADS;
1009         else
1010                 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
1011                                     (u64) THREAD_SIZE * 8UL);
1012
1013         if (threads > max_threads_suggested)
1014                 threads = max_threads_suggested;
1015
1016         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
1017 }
1018
1019 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
1020 /* Initialized by the architecture: */
1021 int arch_task_struct_size __read_mostly;
1022 #endif
1023
1024 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
1025 {
1026         /* Fetch thread_struct whitelist for the architecture. */
1027         arch_thread_struct_whitelist(offset, size);
1028
1029         /*
1030          * Handle zero-sized whitelist or empty thread_struct, otherwise
1031          * adjust offset to position of thread_struct in task_struct.
1032          */
1033         if (unlikely(*size == 0))
1034                 *offset = 0;
1035         else
1036                 *offset += offsetof(struct task_struct, thread);
1037 }
1038
1039 void __init fork_init(void)
1040 {
1041         int i;
1042 #ifndef ARCH_MIN_TASKALIGN
1043 #define ARCH_MIN_TASKALIGN      0
1044 #endif
1045         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
1046         unsigned long useroffset, usersize;
1047
1048         /* create a slab on which task_structs can be allocated */
1049         task_struct_whitelist(&useroffset, &usersize);
1050         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
1051                         arch_task_struct_size, align,
1052                         SLAB_PANIC|SLAB_ACCOUNT,
1053                         useroffset, usersize, NULL);
1054
1055         /* do the arch specific task caches init */
1056         arch_task_cache_init();
1057
1058         set_max_threads(MAX_THREADS);
1059
1060         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
1061         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
1062         init_task.signal->rlim[RLIMIT_SIGPENDING] =
1063                 init_task.signal->rlim[RLIMIT_NPROC];
1064
1065         for (i = 0; i < UCOUNT_COUNTS; i++)
1066                 init_user_ns.ucount_max[i] = max_threads/2;
1067
1068         set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC,      RLIM_INFINITY);
1069         set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE,   RLIM_INFINITY);
1070         set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
1071         set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK,    RLIM_INFINITY);
1072
1073 #ifdef CONFIG_VMAP_STACK
1074         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
1075                           NULL, free_vm_stack_cache);
1076 #endif
1077
1078         scs_init();
1079
1080         lockdep_init_task(&init_task);
1081         uprobes_init();
1082 }
1083
1084 int __weak arch_dup_task_struct(struct task_struct *dst,
1085                                                struct task_struct *src)
1086 {
1087         *dst = *src;
1088         return 0;
1089 }
1090
1091 void set_task_stack_end_magic(struct task_struct *tsk)
1092 {
1093         unsigned long *stackend;
1094
1095         stackend = end_of_stack(tsk);
1096         *stackend = STACK_END_MAGIC;    /* for overflow detection */
1097 }
1098
1099 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
1100 {
1101         struct task_struct *tsk;
1102         int err;
1103
1104         if (node == NUMA_NO_NODE)
1105                 node = tsk_fork_get_node(orig);
1106         tsk = alloc_task_struct_node(node);
1107         if (!tsk)
1108                 return NULL;
1109
1110         err = arch_dup_task_struct(tsk, orig);
1111         if (err)
1112                 goto free_tsk;
1113
1114         err = alloc_thread_stack_node(tsk, node);
1115         if (err)
1116                 goto free_tsk;
1117
1118 #ifdef CONFIG_THREAD_INFO_IN_TASK
1119         refcount_set(&tsk->stack_refcount, 1);
1120 #endif
1121         account_kernel_stack(tsk, 1);
1122
1123         err = scs_prepare(tsk, node);
1124         if (err)
1125                 goto free_stack;
1126
1127 #ifdef CONFIG_SECCOMP
1128         /*
1129          * We must handle setting up seccomp filters once we're under
1130          * the sighand lock in case orig has changed between now and
1131          * then. Until then, filter must be NULL to avoid messing up
1132          * the usage counts on the error path calling free_task.
1133          */
1134         tsk->seccomp.filter = NULL;
1135 #endif
1136
1137         setup_thread_stack(tsk, orig);
1138         clear_user_return_notifier(tsk);
1139         clear_tsk_need_resched(tsk);
1140         set_task_stack_end_magic(tsk);
1141         clear_syscall_work_syscall_user_dispatch(tsk);
1142
1143 #ifdef CONFIG_STACKPROTECTOR
1144         tsk->stack_canary = get_random_canary();
1145 #endif
1146         if (orig->cpus_ptr == &orig->cpus_mask)
1147                 tsk->cpus_ptr = &tsk->cpus_mask;
1148         dup_user_cpus_ptr(tsk, orig, node);
1149
1150         /*
1151          * One for the user space visible state that goes away when reaped.
1152          * One for the scheduler.
1153          */
1154         refcount_set(&tsk->rcu_users, 2);
1155         /* One for the rcu users */
1156         refcount_set(&tsk->usage, 1);
1157 #ifdef CONFIG_BLK_DEV_IO_TRACE
1158         tsk->btrace_seq = 0;
1159 #endif
1160         tsk->splice_pipe = NULL;
1161         tsk->task_frag.page = NULL;
1162         tsk->wake_q.next = NULL;
1163         tsk->worker_private = NULL;
1164
1165         kcov_task_init(tsk);
1166         kmsan_task_create(tsk);
1167         kmap_local_fork(tsk);
1168
1169 #ifdef CONFIG_FAULT_INJECTION
1170         tsk->fail_nth = 0;
1171 #endif
1172
1173 #ifdef CONFIG_BLK_CGROUP
1174         tsk->throttle_disk = NULL;
1175         tsk->use_memdelay = 0;
1176 #endif
1177
1178 #ifdef CONFIG_ARCH_HAS_CPU_PASID
1179         tsk->pasid_activated = 0;
1180 #endif
1181
1182 #ifdef CONFIG_MEMCG
1183         tsk->active_memcg = NULL;
1184 #endif
1185
1186 #ifdef CONFIG_CPU_SUP_INTEL
1187         tsk->reported_split_lock = 0;
1188 #endif
1189
1190 #ifdef CONFIG_SCHED_MM_CID
1191         tsk->mm_cid = -1;
1192         tsk->last_mm_cid = -1;
1193         tsk->mm_cid_active = 0;
1194         tsk->migrate_from_cpu = -1;
1195 #endif
1196         return tsk;
1197
1198 free_stack:
1199         exit_task_stack_account(tsk);
1200         free_thread_stack(tsk);
1201 free_tsk:
1202         free_task_struct(tsk);
1203         return NULL;
1204 }
1205
1206 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1207
1208 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1209
1210 static int __init coredump_filter_setup(char *s)
1211 {
1212         default_dump_filter =
1213                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1214                 MMF_DUMP_FILTER_MASK;
1215         return 1;
1216 }
1217
1218 __setup("coredump_filter=", coredump_filter_setup);
1219
1220 #include <linux/init_task.h>
1221
1222 static void mm_init_aio(struct mm_struct *mm)
1223 {
1224 #ifdef CONFIG_AIO
1225         spin_lock_init(&mm->ioctx_lock);
1226         mm->ioctx_table = NULL;
1227 #endif
1228 }
1229
1230 static __always_inline void mm_clear_owner(struct mm_struct *mm,
1231                                            struct task_struct *p)
1232 {
1233 #ifdef CONFIG_MEMCG
1234         if (mm->owner == p)
1235                 WRITE_ONCE(mm->owner, NULL);
1236 #endif
1237 }
1238
1239 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1240 {
1241 #ifdef CONFIG_MEMCG
1242         mm->owner = p;
1243 #endif
1244 }
1245
1246 static void mm_init_uprobes_state(struct mm_struct *mm)
1247 {
1248 #ifdef CONFIG_UPROBES
1249         mm->uprobes_state.xol_area = NULL;
1250 #endif
1251 }
1252
1253 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1254         struct user_namespace *user_ns)
1255 {
1256         mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1257         mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1258         atomic_set(&mm->mm_users, 1);
1259         atomic_set(&mm->mm_count, 1);
1260         seqcount_init(&mm->write_protect_seq);
1261         mmap_init_lock(mm);
1262         INIT_LIST_HEAD(&mm->mmlist);
1263 #ifdef CONFIG_PER_VMA_LOCK
1264         mm->mm_lock_seq = 0;
1265 #endif
1266         mm_pgtables_bytes_init(mm);
1267         mm->map_count = 0;
1268         mm->locked_vm = 0;
1269         atomic64_set(&mm->pinned_vm, 0);
1270         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1271         spin_lock_init(&mm->page_table_lock);
1272         spin_lock_init(&mm->arg_lock);
1273         mm_init_cpumask(mm);
1274         mm_init_aio(mm);
1275         mm_init_owner(mm, p);
1276         mm_pasid_init(mm);
1277         RCU_INIT_POINTER(mm->exe_file, NULL);
1278         mmu_notifier_subscriptions_init(mm);
1279         init_tlb_flush_pending(mm);
1280 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1281         mm->pmd_huge_pte = NULL;
1282 #endif
1283         mm_init_uprobes_state(mm);
1284         hugetlb_count_init(mm);
1285
1286         if (current->mm) {
1287                 mm->flags = mmf_init_flags(current->mm->flags);
1288                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1289         } else {
1290                 mm->flags = default_dump_filter;
1291                 mm->def_flags = 0;
1292         }
1293
1294         if (mm_alloc_pgd(mm))
1295                 goto fail_nopgd;
1296
1297         if (init_new_context(p, mm))
1298                 goto fail_nocontext;
1299
1300         if (mm_alloc_cid(mm))
1301                 goto fail_cid;
1302
1303         if (percpu_counter_init_many(mm->rss_stat, 0, GFP_KERNEL_ACCOUNT,
1304                                      NR_MM_COUNTERS))
1305                 goto fail_pcpu;
1306
1307         mm->user_ns = get_user_ns(user_ns);
1308         lru_gen_init_mm(mm);
1309         return mm;
1310
1311 fail_pcpu:
1312         mm_destroy_cid(mm);
1313 fail_cid:
1314         destroy_context(mm);
1315 fail_nocontext:
1316         mm_free_pgd(mm);
1317 fail_nopgd:
1318         free_mm(mm);
1319         return NULL;
1320 }
1321
1322 /*
1323  * Allocate and initialize an mm_struct.
1324  */
1325 struct mm_struct *mm_alloc(void)
1326 {
1327         struct mm_struct *mm;
1328
1329         mm = allocate_mm();
1330         if (!mm)
1331                 return NULL;
1332
1333         memset(mm, 0, sizeof(*mm));
1334         return mm_init(mm, current, current_user_ns());
1335 }
1336
1337 static inline void __mmput(struct mm_struct *mm)
1338 {
1339         VM_BUG_ON(atomic_read(&mm->mm_users));
1340
1341         uprobe_clear_state(mm);
1342         exit_aio(mm);
1343         ksm_exit(mm);
1344         khugepaged_exit(mm); /* must run before exit_mmap */
1345         exit_mmap(mm);
1346         mm_put_huge_zero_page(mm);
1347         set_mm_exe_file(mm, NULL);
1348         if (!list_empty(&mm->mmlist)) {
1349                 spin_lock(&mmlist_lock);
1350                 list_del(&mm->mmlist);
1351                 spin_unlock(&mmlist_lock);
1352         }
1353         if (mm->binfmt)
1354                 module_put(mm->binfmt->module);
1355         lru_gen_del_mm(mm);
1356         mmdrop(mm);
1357 }
1358
1359 /*
1360  * Decrement the use count and release all resources for an mm.
1361  */
1362 void mmput(struct mm_struct *mm)
1363 {
1364         might_sleep();
1365
1366         if (atomic_dec_and_test(&mm->mm_users))
1367                 __mmput(mm);
1368 }
1369 EXPORT_SYMBOL_GPL(mmput);
1370
1371 #ifdef CONFIG_MMU
1372 static void mmput_async_fn(struct work_struct *work)
1373 {
1374         struct mm_struct *mm = container_of(work, struct mm_struct,
1375                                             async_put_work);
1376
1377         __mmput(mm);
1378 }
1379
1380 void mmput_async(struct mm_struct *mm)
1381 {
1382         if (atomic_dec_and_test(&mm->mm_users)) {
1383                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1384                 schedule_work(&mm->async_put_work);
1385         }
1386 }
1387 EXPORT_SYMBOL_GPL(mmput_async);
1388 #endif
1389
1390 /**
1391  * set_mm_exe_file - change a reference to the mm's executable file
1392  * @mm: The mm to change.
1393  * @new_exe_file: The new file to use.
1394  *
1395  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1396  *
1397  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1398  * invocations: in mmput() nobody alive left, in execve it happens before
1399  * the new mm is made visible to anyone.
1400  *
1401  * Can only fail if new_exe_file != NULL.
1402  */
1403 int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1404 {
1405         struct file *old_exe_file;
1406
1407         /*
1408          * It is safe to dereference the exe_file without RCU as
1409          * this function is only called if nobody else can access
1410          * this mm -- see comment above for justification.
1411          */
1412         old_exe_file = rcu_dereference_raw(mm->exe_file);
1413
1414         if (new_exe_file) {
1415                 /*
1416                  * We expect the caller (i.e., sys_execve) to already denied
1417                  * write access, so this is unlikely to fail.
1418                  */
1419                 if (unlikely(deny_write_access(new_exe_file)))
1420                         return -EACCES;
1421                 get_file(new_exe_file);
1422         }
1423         rcu_assign_pointer(mm->exe_file, new_exe_file);
1424         if (old_exe_file) {
1425                 allow_write_access(old_exe_file);
1426                 fput(old_exe_file);
1427         }
1428         return 0;
1429 }
1430
1431 /**
1432  * replace_mm_exe_file - replace a reference to the mm's executable file
1433  * @mm: The mm to change.
1434  * @new_exe_file: The new file to use.
1435  *
1436  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1437  *
1438  * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1439  */
1440 int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1441 {
1442         struct vm_area_struct *vma;
1443         struct file *old_exe_file;
1444         int ret = 0;
1445
1446         /* Forbid mm->exe_file change if old file still mapped. */
1447         old_exe_file = get_mm_exe_file(mm);
1448         if (old_exe_file) {
1449                 VMA_ITERATOR(vmi, mm, 0);
1450                 mmap_read_lock(mm);
1451                 for_each_vma(vmi, vma) {
1452                         if (!vma->vm_file)
1453                                 continue;
1454                         if (path_equal(&vma->vm_file->f_path,
1455                                        &old_exe_file->f_path)) {
1456                                 ret = -EBUSY;
1457                                 break;
1458                         }
1459                 }
1460                 mmap_read_unlock(mm);
1461                 fput(old_exe_file);
1462                 if (ret)
1463                         return ret;
1464         }
1465
1466         ret = deny_write_access(new_exe_file);
1467         if (ret)
1468                 return -EACCES;
1469         get_file(new_exe_file);
1470
1471         /* set the new file */
1472         mmap_write_lock(mm);
1473         old_exe_file = rcu_dereference_raw(mm->exe_file);
1474         rcu_assign_pointer(mm->exe_file, new_exe_file);
1475         mmap_write_unlock(mm);
1476
1477         if (old_exe_file) {
1478                 allow_write_access(old_exe_file);
1479                 fput(old_exe_file);
1480         }
1481         return 0;
1482 }
1483
1484 /**
1485  * get_mm_exe_file - acquire a reference to the mm's executable file
1486  * @mm: The mm of interest.
1487  *
1488  * Returns %NULL if mm has no associated executable file.
1489  * User must release file via fput().
1490  */
1491 struct file *get_mm_exe_file(struct mm_struct *mm)
1492 {
1493         struct file *exe_file;
1494
1495         rcu_read_lock();
1496         exe_file = get_file_rcu(&mm->exe_file);
1497         rcu_read_unlock();
1498         return exe_file;
1499 }
1500
1501 /**
1502  * get_task_exe_file - acquire a reference to the task's executable file
1503  * @task: The task.
1504  *
1505  * Returns %NULL if task's mm (if any) has no associated executable file or
1506  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1507  * User must release file via fput().
1508  */
1509 struct file *get_task_exe_file(struct task_struct *task)
1510 {
1511         struct file *exe_file = NULL;
1512         struct mm_struct *mm;
1513
1514         task_lock(task);
1515         mm = task->mm;
1516         if (mm) {
1517                 if (!(task->flags & PF_KTHREAD))
1518                         exe_file = get_mm_exe_file(mm);
1519         }
1520         task_unlock(task);
1521         return exe_file;
1522 }
1523
1524 /**
1525  * get_task_mm - acquire a reference to the task's mm
1526  * @task: The task.
1527  *
1528  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1529  * this kernel workthread has transiently adopted a user mm with use_mm,
1530  * to do its AIO) is not set and if so returns a reference to it, after
1531  * bumping up the use count.  User must release the mm via mmput()
1532  * after use.  Typically used by /proc and ptrace.
1533  */
1534 struct mm_struct *get_task_mm(struct task_struct *task)
1535 {
1536         struct mm_struct *mm;
1537
1538         task_lock(task);
1539         mm = task->mm;
1540         if (mm) {
1541                 if (task->flags & PF_KTHREAD)
1542                         mm = NULL;
1543                 else
1544                         mmget(mm);
1545         }
1546         task_unlock(task);
1547         return mm;
1548 }
1549 EXPORT_SYMBOL_GPL(get_task_mm);
1550
1551 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1552 {
1553         struct mm_struct *mm;
1554         int err;
1555
1556         err =  down_read_killable(&task->signal->exec_update_lock);
1557         if (err)
1558                 return ERR_PTR(err);
1559
1560         mm = get_task_mm(task);
1561         if (mm && mm != current->mm &&
1562                         !ptrace_may_access(task, mode)) {
1563                 mmput(mm);
1564                 mm = ERR_PTR(-EACCES);
1565         }
1566         up_read(&task->signal->exec_update_lock);
1567
1568         return mm;
1569 }
1570
1571 static void complete_vfork_done(struct task_struct *tsk)
1572 {
1573         struct completion *vfork;
1574
1575         task_lock(tsk);
1576         vfork = tsk->vfork_done;
1577         if (likely(vfork)) {
1578                 tsk->vfork_done = NULL;
1579                 complete(vfork);
1580         }
1581         task_unlock(tsk);
1582 }
1583
1584 static int wait_for_vfork_done(struct task_struct *child,
1585                                 struct completion *vfork)
1586 {
1587         unsigned int state = TASK_KILLABLE|TASK_FREEZABLE;
1588         int killed;
1589
1590         cgroup_enter_frozen();
1591         killed = wait_for_completion_state(vfork, state);
1592         cgroup_leave_frozen(false);
1593
1594         if (killed) {
1595                 task_lock(child);
1596                 child->vfork_done = NULL;
1597                 task_unlock(child);
1598         }
1599
1600         put_task_struct(child);
1601         return killed;
1602 }
1603
1604 /* Please note the differences between mmput and mm_release.
1605  * mmput is called whenever we stop holding onto a mm_struct,
1606  * error success whatever.
1607  *
1608  * mm_release is called after a mm_struct has been removed
1609  * from the current process.
1610  *
1611  * This difference is important for error handling, when we
1612  * only half set up a mm_struct for a new process and need to restore
1613  * the old one.  Because we mmput the new mm_struct before
1614  * restoring the old one. . .
1615  * Eric Biederman 10 January 1998
1616  */
1617 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1618 {
1619         uprobe_free_utask(tsk);
1620
1621         /* Get rid of any cached register state */
1622         deactivate_mm(tsk, mm);
1623
1624         /*
1625          * Signal userspace if we're not exiting with a core dump
1626          * because we want to leave the value intact for debugging
1627          * purposes.
1628          */
1629         if (tsk->clear_child_tid) {
1630                 if (atomic_read(&mm->mm_users) > 1) {
1631                         /*
1632                          * We don't check the error code - if userspace has
1633                          * not set up a proper pointer then tough luck.
1634                          */
1635                         put_user(0, tsk->clear_child_tid);
1636                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1637                                         1, NULL, NULL, 0, 0);
1638                 }
1639                 tsk->clear_child_tid = NULL;
1640         }
1641
1642         /*
1643          * All done, finally we can wake up parent and return this mm to him.
1644          * Also kthread_stop() uses this completion for synchronization.
1645          */
1646         if (tsk->vfork_done)
1647                 complete_vfork_done(tsk);
1648 }
1649
1650 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1651 {
1652         futex_exit_release(tsk);
1653         mm_release(tsk, mm);
1654 }
1655
1656 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1657 {
1658         futex_exec_release(tsk);
1659         mm_release(tsk, mm);
1660 }
1661
1662 /**
1663  * dup_mm() - duplicates an existing mm structure
1664  * @tsk: the task_struct with which the new mm will be associated.
1665  * @oldmm: the mm to duplicate.
1666  *
1667  * Allocates a new mm structure and duplicates the provided @oldmm structure
1668  * content into it.
1669  *
1670  * Return: the duplicated mm or NULL on failure.
1671  */
1672 static struct mm_struct *dup_mm(struct task_struct *tsk,
1673                                 struct mm_struct *oldmm)
1674 {
1675         struct mm_struct *mm;
1676         int err;
1677
1678         mm = allocate_mm();
1679         if (!mm)
1680                 goto fail_nomem;
1681
1682         memcpy(mm, oldmm, sizeof(*mm));
1683
1684         if (!mm_init(mm, tsk, mm->user_ns))
1685                 goto fail_nomem;
1686
1687         err = dup_mmap(mm, oldmm);
1688         if (err)
1689                 goto free_pt;
1690
1691         mm->hiwater_rss = get_mm_rss(mm);
1692         mm->hiwater_vm = mm->total_vm;
1693
1694         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1695                 goto free_pt;
1696
1697         return mm;
1698
1699 free_pt:
1700         /* don't put binfmt in mmput, we haven't got module yet */
1701         mm->binfmt = NULL;
1702         mm_init_owner(mm, NULL);
1703         mmput(mm);
1704
1705 fail_nomem:
1706         return NULL;
1707 }
1708
1709 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1710 {
1711         struct mm_struct *mm, *oldmm;
1712
1713         tsk->min_flt = tsk->maj_flt = 0;
1714         tsk->nvcsw = tsk->nivcsw = 0;
1715 #ifdef CONFIG_DETECT_HUNG_TASK
1716         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1717         tsk->last_switch_time = 0;
1718 #endif
1719
1720         tsk->mm = NULL;
1721         tsk->active_mm = NULL;
1722
1723         /*
1724          * Are we cloning a kernel thread?
1725          *
1726          * We need to steal a active VM for that..
1727          */
1728         oldmm = current->mm;
1729         if (!oldmm)
1730                 return 0;
1731
1732         if (clone_flags & CLONE_VM) {
1733                 mmget(oldmm);
1734                 mm = oldmm;
1735         } else {
1736                 mm = dup_mm(tsk, current->mm);
1737                 if (!mm)
1738                         return -ENOMEM;
1739         }
1740
1741         tsk->mm = mm;
1742         tsk->active_mm = mm;
1743         sched_mm_cid_fork(tsk);
1744         return 0;
1745 }
1746
1747 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1748 {
1749         struct fs_struct *fs = current->fs;
1750         if (clone_flags & CLONE_FS) {
1751                 /* tsk->fs is already what we want */
1752                 spin_lock(&fs->lock);
1753                 /* "users" and "in_exec" locked for check_unsafe_exec() */
1754                 if (fs->in_exec) {
1755                         spin_unlock(&fs->lock);
1756                         return -EAGAIN;
1757                 }
1758                 fs->users++;
1759                 spin_unlock(&fs->lock);
1760                 return 0;
1761         }
1762         tsk->fs = copy_fs_struct(fs);
1763         if (!tsk->fs)
1764                 return -ENOMEM;
1765         return 0;
1766 }
1767
1768 static int copy_files(unsigned long clone_flags, struct task_struct *tsk,
1769                       int no_files)
1770 {
1771         struct files_struct *oldf, *newf;
1772         int error = 0;
1773
1774         /*
1775          * A background process may not have any files ...
1776          */
1777         oldf = current->files;
1778         if (!oldf)
1779                 goto out;
1780
1781         if (no_files) {
1782                 tsk->files = NULL;
1783                 goto out;
1784         }
1785
1786         if (clone_flags & CLONE_FILES) {
1787                 atomic_inc(&oldf->count);
1788                 goto out;
1789         }
1790
1791         newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1792         if (!newf)
1793                 goto out;
1794
1795         tsk->files = newf;
1796         error = 0;
1797 out:
1798         return error;
1799 }
1800
1801 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1802 {
1803         struct sighand_struct *sig;
1804
1805         if (clone_flags & CLONE_SIGHAND) {
1806                 refcount_inc(&current->sighand->count);
1807                 return 0;
1808         }
1809         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1810         RCU_INIT_POINTER(tsk->sighand, sig);
1811         if (!sig)
1812                 return -ENOMEM;
1813
1814         refcount_set(&sig->count, 1);
1815         spin_lock_irq(&current->sighand->siglock);
1816         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1817         spin_unlock_irq(&current->sighand->siglock);
1818
1819         /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1820         if (clone_flags & CLONE_CLEAR_SIGHAND)
1821                 flush_signal_handlers(tsk, 0);
1822
1823         return 0;
1824 }
1825
1826 void __cleanup_sighand(struct sighand_struct *sighand)
1827 {
1828         if (refcount_dec_and_test(&sighand->count)) {
1829                 signalfd_cleanup(sighand);
1830                 /*
1831                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1832                  * without an RCU grace period, see __lock_task_sighand().
1833                  */
1834                 kmem_cache_free(sighand_cachep, sighand);
1835         }
1836 }
1837
1838 /*
1839  * Initialize POSIX timer handling for a thread group.
1840  */
1841 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1842 {
1843         struct posix_cputimers *pct = &sig->posix_cputimers;
1844         unsigned long cpu_limit;
1845
1846         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1847         posix_cputimers_group_init(pct, cpu_limit);
1848 }
1849
1850 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1851 {
1852         struct signal_struct *sig;
1853
1854         if (clone_flags & CLONE_THREAD)
1855                 return 0;
1856
1857         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1858         tsk->signal = sig;
1859         if (!sig)
1860                 return -ENOMEM;
1861
1862         sig->nr_threads = 1;
1863         sig->quick_threads = 1;
1864         atomic_set(&sig->live, 1);
1865         refcount_set(&sig->sigcnt, 1);
1866
1867         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1868         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1869         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1870
1871         init_waitqueue_head(&sig->wait_chldexit);
1872         sig->curr_target = tsk;
1873         init_sigpending(&sig->shared_pending);
1874         INIT_HLIST_HEAD(&sig->multiprocess);
1875         seqlock_init(&sig->stats_lock);
1876         prev_cputime_init(&sig->prev_cputime);
1877
1878 #ifdef CONFIG_POSIX_TIMERS
1879         INIT_LIST_HEAD(&sig->posix_timers);
1880         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1881         sig->real_timer.function = it_real_fn;
1882 #endif
1883
1884         task_lock(current->group_leader);
1885         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1886         task_unlock(current->group_leader);
1887
1888         posix_cpu_timers_init_group(sig);
1889
1890         tty_audit_fork(sig);
1891         sched_autogroup_fork(sig);
1892
1893         sig->oom_score_adj = current->signal->oom_score_adj;
1894         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1895
1896         mutex_init(&sig->cred_guard_mutex);
1897         init_rwsem(&sig->exec_update_lock);
1898
1899         return 0;
1900 }
1901
1902 static void copy_seccomp(struct task_struct *p)
1903 {
1904 #ifdef CONFIG_SECCOMP
1905         /*
1906          * Must be called with sighand->lock held, which is common to
1907          * all threads in the group. Holding cred_guard_mutex is not
1908          * needed because this new task is not yet running and cannot
1909          * be racing exec.
1910          */
1911         assert_spin_locked(&current->sighand->siglock);
1912
1913         /* Ref-count the new filter user, and assign it. */
1914         get_seccomp_filter(current);
1915         p->seccomp = current->seccomp;
1916
1917         /*
1918          * Explicitly enable no_new_privs here in case it got set
1919          * between the task_struct being duplicated and holding the
1920          * sighand lock. The seccomp state and nnp must be in sync.
1921          */
1922         if (task_no_new_privs(current))
1923                 task_set_no_new_privs(p);
1924
1925         /*
1926          * If the parent gained a seccomp mode after copying thread
1927          * flags and between before we held the sighand lock, we have
1928          * to manually enable the seccomp thread flag here.
1929          */
1930         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1931                 set_task_syscall_work(p, SECCOMP);
1932 #endif
1933 }
1934
1935 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1936 {
1937         current->clear_child_tid = tidptr;
1938
1939         return task_pid_vnr(current);
1940 }
1941
1942 static void rt_mutex_init_task(struct task_struct *p)
1943 {
1944         raw_spin_lock_init(&p->pi_lock);
1945 #ifdef CONFIG_RT_MUTEXES
1946         p->pi_waiters = RB_ROOT_CACHED;
1947         p->pi_top_task = NULL;
1948         p->pi_blocked_on = NULL;
1949 #endif
1950 }
1951
1952 static inline void init_task_pid_links(struct task_struct *task)
1953 {
1954         enum pid_type type;
1955
1956         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1957                 INIT_HLIST_NODE(&task->pid_links[type]);
1958 }
1959
1960 static inline void
1961 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1962 {
1963         if (type == PIDTYPE_PID)
1964                 task->thread_pid = pid;
1965         else
1966                 task->signal->pids[type] = pid;
1967 }
1968
1969 static inline void rcu_copy_process(struct task_struct *p)
1970 {
1971 #ifdef CONFIG_PREEMPT_RCU
1972         p->rcu_read_lock_nesting = 0;
1973         p->rcu_read_unlock_special.s = 0;
1974         p->rcu_blocked_node = NULL;
1975         INIT_LIST_HEAD(&p->rcu_node_entry);
1976 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1977 #ifdef CONFIG_TASKS_RCU
1978         p->rcu_tasks_holdout = false;
1979         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1980         p->rcu_tasks_idle_cpu = -1;
1981         INIT_LIST_HEAD(&p->rcu_tasks_exit_list);
1982 #endif /* #ifdef CONFIG_TASKS_RCU */
1983 #ifdef CONFIG_TASKS_TRACE_RCU
1984         p->trc_reader_nesting = 0;
1985         p->trc_reader_special.s = 0;
1986         INIT_LIST_HEAD(&p->trc_holdout_list);
1987         INIT_LIST_HEAD(&p->trc_blkd_node);
1988 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1989 }
1990
1991 /**
1992  * __pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
1993  * @pid:   the struct pid for which to create a pidfd
1994  * @flags: flags of the new @pidfd
1995  * @ret: Where to return the file for the pidfd.
1996  *
1997  * Allocate a new file that stashes @pid and reserve a new pidfd number in the
1998  * caller's file descriptor table. The pidfd is reserved but not installed yet.
1999  *
2000  * The helper doesn't perform checks on @pid which makes it useful for pidfds
2001  * created via CLONE_PIDFD where @pid has no task attached when the pidfd and
2002  * pidfd file are prepared.
2003  *
2004  * If this function returns successfully the caller is responsible to either
2005  * call fd_install() passing the returned pidfd and pidfd file as arguments in
2006  * order to install the pidfd into its file descriptor table or they must use
2007  * put_unused_fd() and fput() on the returned pidfd and pidfd file
2008  * respectively.
2009  *
2010  * This function is useful when a pidfd must already be reserved but there
2011  * might still be points of failure afterwards and the caller wants to ensure
2012  * that no pidfd is leaked into its file descriptor table.
2013  *
2014  * Return: On success, a reserved pidfd is returned from the function and a new
2015  *         pidfd file is returned in the last argument to the function. On
2016  *         error, a negative error code is returned from the function and the
2017  *         last argument remains unchanged.
2018  */
2019 static int __pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2020 {
2021         int pidfd;
2022         struct file *pidfd_file;
2023
2024         pidfd = get_unused_fd_flags(O_CLOEXEC);
2025         if (pidfd < 0)
2026                 return pidfd;
2027
2028         pidfd_file = pidfs_alloc_file(pid, flags | O_RDWR);
2029         if (IS_ERR(pidfd_file)) {
2030                 put_unused_fd(pidfd);
2031                 return PTR_ERR(pidfd_file);
2032         }
2033         /*
2034          * anon_inode_getfile() ignores everything outside of the
2035          * O_ACCMODE | O_NONBLOCK mask, set PIDFD_THREAD manually.
2036          */
2037         pidfd_file->f_flags |= (flags & PIDFD_THREAD);
2038         *ret = pidfd_file;
2039         return pidfd;
2040 }
2041
2042 /**
2043  * pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2044  * @pid:   the struct pid for which to create a pidfd
2045  * @flags: flags of the new @pidfd
2046  * @ret: Where to return the pidfd.
2047  *
2048  * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2049  * caller's file descriptor table. The pidfd is reserved but not installed yet.
2050  *
2051  * The helper verifies that @pid is still in use, without PIDFD_THREAD the
2052  * task identified by @pid must be a thread-group leader.
2053  *
2054  * If this function returns successfully the caller is responsible to either
2055  * call fd_install() passing the returned pidfd and pidfd file as arguments in
2056  * order to install the pidfd into its file descriptor table or they must use
2057  * put_unused_fd() and fput() on the returned pidfd and pidfd file
2058  * respectively.
2059  *
2060  * This function is useful when a pidfd must already be reserved but there
2061  * might still be points of failure afterwards and the caller wants to ensure
2062  * that no pidfd is leaked into its file descriptor table.
2063  *
2064  * Return: On success, a reserved pidfd is returned from the function and a new
2065  *         pidfd file is returned in the last argument to the function. On
2066  *         error, a negative error code is returned from the function and the
2067  *         last argument remains unchanged.
2068  */
2069 int pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2070 {
2071         bool thread = flags & PIDFD_THREAD;
2072
2073         if (!pid || !pid_has_task(pid, thread ? PIDTYPE_PID : PIDTYPE_TGID))
2074                 return -EINVAL;
2075
2076         return __pidfd_prepare(pid, flags, ret);
2077 }
2078
2079 static void __delayed_free_task(struct rcu_head *rhp)
2080 {
2081         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
2082
2083         free_task(tsk);
2084 }
2085
2086 static __always_inline void delayed_free_task(struct task_struct *tsk)
2087 {
2088         if (IS_ENABLED(CONFIG_MEMCG))
2089                 call_rcu(&tsk->rcu, __delayed_free_task);
2090         else
2091                 free_task(tsk);
2092 }
2093
2094 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
2095 {
2096         /* Skip if kernel thread */
2097         if (!tsk->mm)
2098                 return;
2099
2100         /* Skip if spawning a thread or using vfork */
2101         if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
2102                 return;
2103
2104         /* We need to synchronize with __set_oom_adj */
2105         mutex_lock(&oom_adj_mutex);
2106         set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
2107         /* Update the values in case they were changed after copy_signal */
2108         tsk->signal->oom_score_adj = current->signal->oom_score_adj;
2109         tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
2110         mutex_unlock(&oom_adj_mutex);
2111 }
2112
2113 #ifdef CONFIG_RV
2114 static void rv_task_fork(struct task_struct *p)
2115 {
2116         int i;
2117
2118         for (i = 0; i < RV_PER_TASK_MONITORS; i++)
2119                 p->rv[i].da_mon.monitoring = false;
2120 }
2121 #else
2122 #define rv_task_fork(p) do {} while (0)
2123 #endif
2124
2125 /*
2126  * This creates a new process as a copy of the old one,
2127  * but does not actually start it yet.
2128  *
2129  * It copies the registers, and all the appropriate
2130  * parts of the process environment (as per the clone
2131  * flags). The actual kick-off is left to the caller.
2132  */
2133 __latent_entropy struct task_struct *copy_process(
2134                                         struct pid *pid,
2135                                         int trace,
2136                                         int node,
2137                                         struct kernel_clone_args *args)
2138 {
2139         int pidfd = -1, retval;
2140         struct task_struct *p;
2141         struct multiprocess_signals delayed;
2142         struct file *pidfile = NULL;
2143         const u64 clone_flags = args->flags;
2144         struct nsproxy *nsp = current->nsproxy;
2145
2146         /*
2147          * Don't allow sharing the root directory with processes in a different
2148          * namespace
2149          */
2150         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2151                 return ERR_PTR(-EINVAL);
2152
2153         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2154                 return ERR_PTR(-EINVAL);
2155
2156         /*
2157          * Thread groups must share signals as well, and detached threads
2158          * can only be started up within the thread group.
2159          */
2160         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2161                 return ERR_PTR(-EINVAL);
2162
2163         /*
2164          * Shared signal handlers imply shared VM. By way of the above,
2165          * thread groups also imply shared VM. Blocking this case allows
2166          * for various simplifications in other code.
2167          */
2168         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2169                 return ERR_PTR(-EINVAL);
2170
2171         /*
2172          * Siblings of global init remain as zombies on exit since they are
2173          * not reaped by their parent (swapper). To solve this and to avoid
2174          * multi-rooted process trees, prevent global and container-inits
2175          * from creating siblings.
2176          */
2177         if ((clone_flags & CLONE_PARENT) &&
2178                                 current->signal->flags & SIGNAL_UNKILLABLE)
2179                 return ERR_PTR(-EINVAL);
2180
2181         /*
2182          * If the new process will be in a different pid or user namespace
2183          * do not allow it to share a thread group with the forking task.
2184          */
2185         if (clone_flags & CLONE_THREAD) {
2186                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2187                     (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2188                         return ERR_PTR(-EINVAL);
2189         }
2190
2191         if (clone_flags & CLONE_PIDFD) {
2192                 /*
2193                  * - CLONE_DETACHED is blocked so that we can potentially
2194                  *   reuse it later for CLONE_PIDFD.
2195                  */
2196                 if (clone_flags & CLONE_DETACHED)
2197                         return ERR_PTR(-EINVAL);
2198         }
2199
2200         /*
2201          * Force any signals received before this point to be delivered
2202          * before the fork happens.  Collect up signals sent to multiple
2203          * processes that happen during the fork and delay them so that
2204          * they appear to happen after the fork.
2205          */
2206         sigemptyset(&delayed.signal);
2207         INIT_HLIST_NODE(&delayed.node);
2208
2209         spin_lock_irq(&current->sighand->siglock);
2210         if (!(clone_flags & CLONE_THREAD))
2211                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
2212         recalc_sigpending();
2213         spin_unlock_irq(&current->sighand->siglock);
2214         retval = -ERESTARTNOINTR;
2215         if (task_sigpending(current))
2216                 goto fork_out;
2217
2218         retval = -ENOMEM;
2219         p = dup_task_struct(current, node);
2220         if (!p)
2221                 goto fork_out;
2222         p->flags &= ~PF_KTHREAD;
2223         if (args->kthread)
2224                 p->flags |= PF_KTHREAD;
2225         if (args->user_worker) {
2226                 /*
2227                  * Mark us a user worker, and block any signal that isn't
2228                  * fatal or STOP
2229                  */
2230                 p->flags |= PF_USER_WORKER;
2231                 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2232         }
2233         if (args->io_thread)
2234                 p->flags |= PF_IO_WORKER;
2235
2236         if (args->name)
2237                 strscpy_pad(p->comm, args->name, sizeof(p->comm));
2238
2239         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2240         /*
2241          * Clear TID on mm_release()?
2242          */
2243         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2244
2245         ftrace_graph_init_task(p);
2246
2247         rt_mutex_init_task(p);
2248
2249         lockdep_assert_irqs_enabled();
2250 #ifdef CONFIG_PROVE_LOCKING
2251         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2252 #endif
2253         retval = copy_creds(p, clone_flags);
2254         if (retval < 0)
2255                 goto bad_fork_free;
2256
2257         retval = -EAGAIN;
2258         if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2259                 if (p->real_cred->user != INIT_USER &&
2260                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2261                         goto bad_fork_cleanup_count;
2262         }
2263         current->flags &= ~PF_NPROC_EXCEEDED;
2264
2265         /*
2266          * If multiple threads are within copy_process(), then this check
2267          * triggers too late. This doesn't hurt, the check is only there
2268          * to stop root fork bombs.
2269          */
2270         retval = -EAGAIN;
2271         if (data_race(nr_threads >= max_threads))
2272                 goto bad_fork_cleanup_count;
2273
2274         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
2275         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2276         p->flags |= PF_FORKNOEXEC;
2277         INIT_LIST_HEAD(&p->children);
2278         INIT_LIST_HEAD(&p->sibling);
2279         rcu_copy_process(p);
2280         p->vfork_done = NULL;
2281         spin_lock_init(&p->alloc_lock);
2282
2283         init_sigpending(&p->pending);
2284
2285         p->utime = p->stime = p->gtime = 0;
2286 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2287         p->utimescaled = p->stimescaled = 0;
2288 #endif
2289         prev_cputime_init(&p->prev_cputime);
2290
2291 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2292         seqcount_init(&p->vtime.seqcount);
2293         p->vtime.starttime = 0;
2294         p->vtime.state = VTIME_INACTIVE;
2295 #endif
2296
2297 #ifdef CONFIG_IO_URING
2298         p->io_uring = NULL;
2299 #endif
2300
2301         p->default_timer_slack_ns = current->timer_slack_ns;
2302
2303 #ifdef CONFIG_PSI
2304         p->psi_flags = 0;
2305 #endif
2306
2307         task_io_accounting_init(&p->ioac);
2308         acct_clear_integrals(p);
2309
2310         posix_cputimers_init(&p->posix_cputimers);
2311
2312         p->io_context = NULL;
2313         audit_set_context(p, NULL);
2314         cgroup_fork(p);
2315         if (args->kthread) {
2316                 if (!set_kthread_struct(p))
2317                         goto bad_fork_cleanup_delayacct;
2318         }
2319 #ifdef CONFIG_NUMA
2320         p->mempolicy = mpol_dup(p->mempolicy);
2321         if (IS_ERR(p->mempolicy)) {
2322                 retval = PTR_ERR(p->mempolicy);
2323                 p->mempolicy = NULL;
2324                 goto bad_fork_cleanup_delayacct;
2325         }
2326 #endif
2327 #ifdef CONFIG_CPUSETS
2328         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2329         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2330         seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2331 #endif
2332 #ifdef CONFIG_TRACE_IRQFLAGS
2333         memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2334         p->irqtrace.hardirq_disable_ip  = _THIS_IP_;
2335         p->irqtrace.softirq_enable_ip   = _THIS_IP_;
2336         p->softirqs_enabled             = 1;
2337         p->softirq_context              = 0;
2338 #endif
2339
2340         p->pagefault_disabled = 0;
2341
2342 #ifdef CONFIG_LOCKDEP
2343         lockdep_init_task(p);
2344 #endif
2345
2346 #ifdef CONFIG_DEBUG_MUTEXES
2347         p->blocked_on = NULL; /* not blocked yet */
2348 #endif
2349 #ifdef CONFIG_BCACHE
2350         p->sequential_io        = 0;
2351         p->sequential_io_avg    = 0;
2352 #endif
2353 #ifdef CONFIG_BPF_SYSCALL
2354         RCU_INIT_POINTER(p->bpf_storage, NULL);
2355         p->bpf_ctx = NULL;
2356 #endif
2357
2358         /* Perform scheduler related setup. Assign this task to a CPU. */
2359         retval = sched_fork(clone_flags, p);
2360         if (retval)
2361                 goto bad_fork_cleanup_policy;
2362
2363         retval = perf_event_init_task(p, clone_flags);
2364         if (retval)
2365                 goto bad_fork_cleanup_policy;
2366         retval = audit_alloc(p);
2367         if (retval)
2368                 goto bad_fork_cleanup_perf;
2369         /* copy all the process information */
2370         shm_init_task(p);
2371         retval = security_task_alloc(p, clone_flags);
2372         if (retval)
2373                 goto bad_fork_cleanup_audit;
2374         retval = copy_semundo(clone_flags, p);
2375         if (retval)
2376                 goto bad_fork_cleanup_security;
2377         retval = copy_files(clone_flags, p, args->no_files);
2378         if (retval)
2379                 goto bad_fork_cleanup_semundo;
2380         retval = copy_fs(clone_flags, p);
2381         if (retval)
2382                 goto bad_fork_cleanup_files;
2383         retval = copy_sighand(clone_flags, p);
2384         if (retval)
2385                 goto bad_fork_cleanup_fs;
2386         retval = copy_signal(clone_flags, p);
2387         if (retval)
2388                 goto bad_fork_cleanup_sighand;
2389         retval = copy_mm(clone_flags, p);
2390         if (retval)
2391                 goto bad_fork_cleanup_signal;
2392         retval = copy_namespaces(clone_flags, p);
2393         if (retval)
2394                 goto bad_fork_cleanup_mm;
2395         retval = copy_io(clone_flags, p);
2396         if (retval)
2397                 goto bad_fork_cleanup_namespaces;
2398         retval = copy_thread(p, args);
2399         if (retval)
2400                 goto bad_fork_cleanup_io;
2401
2402         stackleak_task_init(p);
2403
2404         if (pid != &init_struct_pid) {
2405                 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2406                                 args->set_tid_size);
2407                 if (IS_ERR(pid)) {
2408                         retval = PTR_ERR(pid);
2409                         goto bad_fork_cleanup_thread;
2410                 }
2411         }
2412
2413         /*
2414          * This has to happen after we've potentially unshared the file
2415          * descriptor table (so that the pidfd doesn't leak into the child
2416          * if the fd table isn't shared).
2417          */
2418         if (clone_flags & CLONE_PIDFD) {
2419                 int flags = (clone_flags & CLONE_THREAD) ? PIDFD_THREAD : 0;
2420
2421                 /* Note that no task has been attached to @pid yet. */
2422                 retval = __pidfd_prepare(pid, flags, &pidfile);
2423                 if (retval < 0)
2424                         goto bad_fork_free_pid;
2425                 pidfd = retval;
2426
2427                 retval = put_user(pidfd, args->pidfd);
2428                 if (retval)
2429                         goto bad_fork_put_pidfd;
2430         }
2431
2432 #ifdef CONFIG_BLOCK
2433         p->plug = NULL;
2434 #endif
2435         futex_init_task(p);
2436
2437         /*
2438          * sigaltstack should be cleared when sharing the same VM
2439          */
2440         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2441                 sas_ss_reset(p);
2442
2443         /*
2444          * Syscall tracing and stepping should be turned off in the
2445          * child regardless of CLONE_PTRACE.
2446          */
2447         user_disable_single_step(p);
2448         clear_task_syscall_work(p, SYSCALL_TRACE);
2449 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2450         clear_task_syscall_work(p, SYSCALL_EMU);
2451 #endif
2452         clear_tsk_latency_tracing(p);
2453
2454         /* ok, now we should be set up.. */
2455         p->pid = pid_nr(pid);
2456         if (clone_flags & CLONE_THREAD) {
2457                 p->group_leader = current->group_leader;
2458                 p->tgid = current->tgid;
2459         } else {
2460                 p->group_leader = p;
2461                 p->tgid = p->pid;
2462         }
2463
2464         p->nr_dirtied = 0;
2465         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2466         p->dirty_paused_when = 0;
2467
2468         p->pdeath_signal = 0;
2469         p->task_works = NULL;
2470         clear_posix_cputimers_work(p);
2471
2472 #ifdef CONFIG_KRETPROBES
2473         p->kretprobe_instances.first = NULL;
2474 #endif
2475 #ifdef CONFIG_RETHOOK
2476         p->rethooks.first = NULL;
2477 #endif
2478
2479         /*
2480          * Ensure that the cgroup subsystem policies allow the new process to be
2481          * forked. It should be noted that the new process's css_set can be changed
2482          * between here and cgroup_post_fork() if an organisation operation is in
2483          * progress.
2484          */
2485         retval = cgroup_can_fork(p, args);
2486         if (retval)
2487                 goto bad_fork_put_pidfd;
2488
2489         /*
2490          * Now that the cgroups are pinned, re-clone the parent cgroup and put
2491          * the new task on the correct runqueue. All this *before* the task
2492          * becomes visible.
2493          *
2494          * This isn't part of ->can_fork() because while the re-cloning is
2495          * cgroup specific, it unconditionally needs to place the task on a
2496          * runqueue.
2497          */
2498         sched_cgroup_fork(p, args);
2499
2500         /*
2501          * From this point on we must avoid any synchronous user-space
2502          * communication until we take the tasklist-lock. In particular, we do
2503          * not want user-space to be able to predict the process start-time by
2504          * stalling fork(2) after we recorded the start_time but before it is
2505          * visible to the system.
2506          */
2507
2508         p->start_time = ktime_get_ns();
2509         p->start_boottime = ktime_get_boottime_ns();
2510
2511         /*
2512          * Make it visible to the rest of the system, but dont wake it up yet.
2513          * Need tasklist lock for parent etc handling!
2514          */
2515         write_lock_irq(&tasklist_lock);
2516
2517         /* CLONE_PARENT re-uses the old parent */
2518         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2519                 p->real_parent = current->real_parent;
2520                 p->parent_exec_id = current->parent_exec_id;
2521                 if (clone_flags & CLONE_THREAD)
2522                         p->exit_signal = -1;
2523                 else
2524                         p->exit_signal = current->group_leader->exit_signal;
2525         } else {
2526                 p->real_parent = current;
2527                 p->parent_exec_id = current->self_exec_id;
2528                 p->exit_signal = args->exit_signal;
2529         }
2530
2531         klp_copy_process(p);
2532
2533         sched_core_fork(p);
2534
2535         spin_lock(&current->sighand->siglock);
2536
2537         rv_task_fork(p);
2538
2539         rseq_fork(p, clone_flags);
2540
2541         /* Don't start children in a dying pid namespace */
2542         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2543                 retval = -ENOMEM;
2544                 goto bad_fork_cancel_cgroup;
2545         }
2546
2547         /* Let kill terminate clone/fork in the middle */
2548         if (fatal_signal_pending(current)) {
2549                 retval = -EINTR;
2550                 goto bad_fork_cancel_cgroup;
2551         }
2552
2553         /* No more failure paths after this point. */
2554
2555         /*
2556          * Copy seccomp details explicitly here, in case they were changed
2557          * before holding sighand lock.
2558          */
2559         copy_seccomp(p);
2560
2561         init_task_pid_links(p);
2562         if (likely(p->pid)) {
2563                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2564
2565                 init_task_pid(p, PIDTYPE_PID, pid);
2566                 if (thread_group_leader(p)) {
2567                         init_task_pid(p, PIDTYPE_TGID, pid);
2568                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2569                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2570
2571                         if (is_child_reaper(pid)) {
2572                                 ns_of_pid(pid)->child_reaper = p;
2573                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2574                         }
2575                         p->signal->shared_pending.signal = delayed.signal;
2576                         p->signal->tty = tty_kref_get(current->signal->tty);
2577                         /*
2578                          * Inherit has_child_subreaper flag under the same
2579                          * tasklist_lock with adding child to the process tree
2580                          * for propagate_has_child_subreaper optimization.
2581                          */
2582                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2583                                                          p->real_parent->signal->is_child_subreaper;
2584                         list_add_tail(&p->sibling, &p->real_parent->children);
2585                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2586                         attach_pid(p, PIDTYPE_TGID);
2587                         attach_pid(p, PIDTYPE_PGID);
2588                         attach_pid(p, PIDTYPE_SID);
2589                         __this_cpu_inc(process_counts);
2590                 } else {
2591                         current->signal->nr_threads++;
2592                         current->signal->quick_threads++;
2593                         atomic_inc(&current->signal->live);
2594                         refcount_inc(&current->signal->sigcnt);
2595                         task_join_group_stop(p);
2596                         list_add_tail_rcu(&p->thread_node,
2597                                           &p->signal->thread_head);
2598                 }
2599                 attach_pid(p, PIDTYPE_PID);
2600                 nr_threads++;
2601         }
2602         total_forks++;
2603         hlist_del_init(&delayed.node);
2604         spin_unlock(&current->sighand->siglock);
2605         syscall_tracepoint_update(p);
2606         write_unlock_irq(&tasklist_lock);
2607
2608         if (pidfile)
2609                 fd_install(pidfd, pidfile);
2610
2611         proc_fork_connector(p);
2612         sched_post_fork(p);
2613         cgroup_post_fork(p, args);
2614         perf_event_fork(p);
2615
2616         trace_task_newtask(p, clone_flags);
2617         uprobe_copy_process(p, clone_flags);
2618         user_events_fork(p, clone_flags);
2619
2620         copy_oom_score_adj(clone_flags, p);
2621
2622         return p;
2623
2624 bad_fork_cancel_cgroup:
2625         sched_core_free(p);
2626         spin_unlock(&current->sighand->siglock);
2627         write_unlock_irq(&tasklist_lock);
2628         cgroup_cancel_fork(p, args);
2629 bad_fork_put_pidfd:
2630         if (clone_flags & CLONE_PIDFD) {
2631                 fput(pidfile);
2632                 put_unused_fd(pidfd);
2633         }
2634 bad_fork_free_pid:
2635         if (pid != &init_struct_pid)
2636                 free_pid(pid);
2637 bad_fork_cleanup_thread:
2638         exit_thread(p);
2639 bad_fork_cleanup_io:
2640         if (p->io_context)
2641                 exit_io_context(p);
2642 bad_fork_cleanup_namespaces:
2643         exit_task_namespaces(p);
2644 bad_fork_cleanup_mm:
2645         if (p->mm) {
2646                 mm_clear_owner(p->mm, p);
2647                 mmput(p->mm);
2648         }
2649 bad_fork_cleanup_signal:
2650         if (!(clone_flags & CLONE_THREAD))
2651                 free_signal_struct(p->signal);
2652 bad_fork_cleanup_sighand:
2653         __cleanup_sighand(p->sighand);
2654 bad_fork_cleanup_fs:
2655         exit_fs(p); /* blocking */
2656 bad_fork_cleanup_files:
2657         exit_files(p); /* blocking */
2658 bad_fork_cleanup_semundo:
2659         exit_sem(p);
2660 bad_fork_cleanup_security:
2661         security_task_free(p);
2662 bad_fork_cleanup_audit:
2663         audit_free(p);
2664 bad_fork_cleanup_perf:
2665         perf_event_free_task(p);
2666 bad_fork_cleanup_policy:
2667         lockdep_free_task(p);
2668 #ifdef CONFIG_NUMA
2669         mpol_put(p->mempolicy);
2670 #endif
2671 bad_fork_cleanup_delayacct:
2672         delayacct_tsk_free(p);
2673 bad_fork_cleanup_count:
2674         dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2675         exit_creds(p);
2676 bad_fork_free:
2677         WRITE_ONCE(p->__state, TASK_DEAD);
2678         exit_task_stack_account(p);
2679         put_task_stack(p);
2680         delayed_free_task(p);
2681 fork_out:
2682         spin_lock_irq(&current->sighand->siglock);
2683         hlist_del_init(&delayed.node);
2684         spin_unlock_irq(&current->sighand->siglock);
2685         return ERR_PTR(retval);
2686 }
2687
2688 static inline void init_idle_pids(struct task_struct *idle)
2689 {
2690         enum pid_type type;
2691
2692         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2693                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2694                 init_task_pid(idle, type, &init_struct_pid);
2695         }
2696 }
2697
2698 static int idle_dummy(void *dummy)
2699 {
2700         /* This function is never called */
2701         return 0;
2702 }
2703
2704 struct task_struct * __init fork_idle(int cpu)
2705 {
2706         struct task_struct *task;
2707         struct kernel_clone_args args = {
2708                 .flags          = CLONE_VM,
2709                 .fn             = &idle_dummy,
2710                 .fn_arg         = NULL,
2711                 .kthread        = 1,
2712                 .idle           = 1,
2713         };
2714
2715         task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2716         if (!IS_ERR(task)) {
2717                 init_idle_pids(task);
2718                 init_idle(task, cpu);
2719         }
2720
2721         return task;
2722 }
2723
2724 /*
2725  * This is like kernel_clone(), but shaved down and tailored to just
2726  * creating io_uring workers. It returns a created task, or an error pointer.
2727  * The returned task is inactive, and the caller must fire it up through
2728  * wake_up_new_task(p). All signals are blocked in the created task.
2729  */
2730 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2731 {
2732         unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2733                                 CLONE_IO;
2734         struct kernel_clone_args args = {
2735                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2736                                     CLONE_UNTRACED) & ~CSIGNAL),
2737                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2738                 .fn             = fn,
2739                 .fn_arg         = arg,
2740                 .io_thread      = 1,
2741                 .user_worker    = 1,
2742         };
2743
2744         return copy_process(NULL, 0, node, &args);
2745 }
2746
2747 /*
2748  *  Ok, this is the main fork-routine.
2749  *
2750  * It copies the process, and if successful kick-starts
2751  * it and waits for it to finish using the VM if required.
2752  *
2753  * args->exit_signal is expected to be checked for sanity by the caller.
2754  */
2755 pid_t kernel_clone(struct kernel_clone_args *args)
2756 {
2757         u64 clone_flags = args->flags;
2758         struct completion vfork;
2759         struct pid *pid;
2760         struct task_struct *p;
2761         int trace = 0;
2762         pid_t nr;
2763
2764         /*
2765          * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2766          * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2767          * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2768          * field in struct clone_args and it still doesn't make sense to have
2769          * them both point at the same memory location. Performing this check
2770          * here has the advantage that we don't need to have a separate helper
2771          * to check for legacy clone().
2772          */
2773         if ((clone_flags & CLONE_PIDFD) &&
2774             (clone_flags & CLONE_PARENT_SETTID) &&
2775             (args->pidfd == args->parent_tid))
2776                 return -EINVAL;
2777
2778         /*
2779          * Determine whether and which event to report to ptracer.  When
2780          * called from kernel_thread or CLONE_UNTRACED is explicitly
2781          * requested, no event is reported; otherwise, report if the event
2782          * for the type of forking is enabled.
2783          */
2784         if (!(clone_flags & CLONE_UNTRACED)) {
2785                 if (clone_flags & CLONE_VFORK)
2786                         trace = PTRACE_EVENT_VFORK;
2787                 else if (args->exit_signal != SIGCHLD)
2788                         trace = PTRACE_EVENT_CLONE;
2789                 else
2790                         trace = PTRACE_EVENT_FORK;
2791
2792                 if (likely(!ptrace_event_enabled(current, trace)))
2793                         trace = 0;
2794         }
2795
2796         p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2797         add_latent_entropy();
2798
2799         if (IS_ERR(p))
2800                 return PTR_ERR(p);
2801
2802         /*
2803          * Do this prior waking up the new thread - the thread pointer
2804          * might get invalid after that point, if the thread exits quickly.
2805          */
2806         trace_sched_process_fork(current, p);
2807
2808         pid = get_task_pid(p, PIDTYPE_PID);
2809         nr = pid_vnr(pid);
2810
2811         if (clone_flags & CLONE_PARENT_SETTID)
2812                 put_user(nr, args->parent_tid);
2813
2814         if (clone_flags & CLONE_VFORK) {
2815                 p->vfork_done = &vfork;
2816                 init_completion(&vfork);
2817                 get_task_struct(p);
2818         }
2819
2820         if (IS_ENABLED(CONFIG_LRU_GEN_WALKS_MMU) && !(clone_flags & CLONE_VM)) {
2821                 /* lock the task to synchronize with memcg migration */
2822                 task_lock(p);
2823                 lru_gen_add_mm(p->mm);
2824                 task_unlock(p);
2825         }
2826
2827         wake_up_new_task(p);
2828
2829         /* forking complete and child started to run, tell ptracer */
2830         if (unlikely(trace))
2831                 ptrace_event_pid(trace, pid);
2832
2833         if (clone_flags & CLONE_VFORK) {
2834                 if (!wait_for_vfork_done(p, &vfork))
2835                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2836         }
2837
2838         put_pid(pid);
2839         return nr;
2840 }
2841
2842 /*
2843  * Create a kernel thread.
2844  */
2845 pid_t kernel_thread(int (*fn)(void *), void *arg, const char *name,
2846                     unsigned long flags)
2847 {
2848         struct kernel_clone_args args = {
2849                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2850                                     CLONE_UNTRACED) & ~CSIGNAL),
2851                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2852                 .fn             = fn,
2853                 .fn_arg         = arg,
2854                 .name           = name,
2855                 .kthread        = 1,
2856         };
2857
2858         return kernel_clone(&args);
2859 }
2860
2861 /*
2862  * Create a user mode thread.
2863  */
2864 pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2865 {
2866         struct kernel_clone_args args = {
2867                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2868                                     CLONE_UNTRACED) & ~CSIGNAL),
2869                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2870                 .fn             = fn,
2871                 .fn_arg         = arg,
2872         };
2873
2874         return kernel_clone(&args);
2875 }
2876
2877 #ifdef __ARCH_WANT_SYS_FORK
2878 SYSCALL_DEFINE0(fork)
2879 {
2880 #ifdef CONFIG_MMU
2881         struct kernel_clone_args args = {
2882                 .exit_signal = SIGCHLD,
2883         };
2884
2885         return kernel_clone(&args);
2886 #else
2887         /* can not support in nommu mode */
2888         return -EINVAL;
2889 #endif
2890 }
2891 #endif
2892
2893 #ifdef __ARCH_WANT_SYS_VFORK
2894 SYSCALL_DEFINE0(vfork)
2895 {
2896         struct kernel_clone_args args = {
2897                 .flags          = CLONE_VFORK | CLONE_VM,
2898                 .exit_signal    = SIGCHLD,
2899         };
2900
2901         return kernel_clone(&args);
2902 }
2903 #endif
2904
2905 #ifdef __ARCH_WANT_SYS_CLONE
2906 #ifdef CONFIG_CLONE_BACKWARDS
2907 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2908                  int __user *, parent_tidptr,
2909                  unsigned long, tls,
2910                  int __user *, child_tidptr)
2911 #elif defined(CONFIG_CLONE_BACKWARDS2)
2912 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2913                  int __user *, parent_tidptr,
2914                  int __user *, child_tidptr,
2915                  unsigned long, tls)
2916 #elif defined(CONFIG_CLONE_BACKWARDS3)
2917 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2918                 int, stack_size,
2919                 int __user *, parent_tidptr,
2920                 int __user *, child_tidptr,
2921                 unsigned long, tls)
2922 #else
2923 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2924                  int __user *, parent_tidptr,
2925                  int __user *, child_tidptr,
2926                  unsigned long, tls)
2927 #endif
2928 {
2929         struct kernel_clone_args args = {
2930                 .flags          = (lower_32_bits(clone_flags) & ~CSIGNAL),
2931                 .pidfd          = parent_tidptr,
2932                 .child_tid      = child_tidptr,
2933                 .parent_tid     = parent_tidptr,
2934                 .exit_signal    = (lower_32_bits(clone_flags) & CSIGNAL),
2935                 .stack          = newsp,
2936                 .tls            = tls,
2937         };
2938
2939         return kernel_clone(&args);
2940 }
2941 #endif
2942
2943 #ifdef __ARCH_WANT_SYS_CLONE3
2944
2945 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2946                                               struct clone_args __user *uargs,
2947                                               size_t usize)
2948 {
2949         int err;
2950         struct clone_args args;
2951         pid_t *kset_tid = kargs->set_tid;
2952
2953         BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2954                      CLONE_ARGS_SIZE_VER0);
2955         BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2956                      CLONE_ARGS_SIZE_VER1);
2957         BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2958                      CLONE_ARGS_SIZE_VER2);
2959         BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2960
2961         if (unlikely(usize > PAGE_SIZE))
2962                 return -E2BIG;
2963         if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2964                 return -EINVAL;
2965
2966         err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2967         if (err)
2968                 return err;
2969
2970         if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2971                 return -EINVAL;
2972
2973         if (unlikely(!args.set_tid && args.set_tid_size > 0))
2974                 return -EINVAL;
2975
2976         if (unlikely(args.set_tid && args.set_tid_size == 0))
2977                 return -EINVAL;
2978
2979         /*
2980          * Verify that higher 32bits of exit_signal are unset and that
2981          * it is a valid signal
2982          */
2983         if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2984                      !valid_signal(args.exit_signal)))
2985                 return -EINVAL;
2986
2987         if ((args.flags & CLONE_INTO_CGROUP) &&
2988             (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2989                 return -EINVAL;
2990
2991         *kargs = (struct kernel_clone_args){
2992                 .flags          = args.flags,
2993                 .pidfd          = u64_to_user_ptr(args.pidfd),
2994                 .child_tid      = u64_to_user_ptr(args.child_tid),
2995                 .parent_tid     = u64_to_user_ptr(args.parent_tid),
2996                 .exit_signal    = args.exit_signal,
2997                 .stack          = args.stack,
2998                 .stack_size     = args.stack_size,
2999                 .tls            = args.tls,
3000                 .set_tid_size   = args.set_tid_size,
3001                 .cgroup         = args.cgroup,
3002         };
3003
3004         if (args.set_tid &&
3005                 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
3006                         (kargs->set_tid_size * sizeof(pid_t))))
3007                 return -EFAULT;
3008
3009         kargs->set_tid = kset_tid;
3010
3011         return 0;
3012 }
3013
3014 /**
3015  * clone3_stack_valid - check and prepare stack
3016  * @kargs: kernel clone args
3017  *
3018  * Verify that the stack arguments userspace gave us are sane.
3019  * In addition, set the stack direction for userspace since it's easy for us to
3020  * determine.
3021  */
3022 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
3023 {
3024         if (kargs->stack == 0) {
3025                 if (kargs->stack_size > 0)
3026                         return false;
3027         } else {
3028                 if (kargs->stack_size == 0)
3029                         return false;
3030
3031                 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
3032                         return false;
3033
3034 #if !defined(CONFIG_STACK_GROWSUP)
3035                 kargs->stack += kargs->stack_size;
3036 #endif
3037         }
3038
3039         return true;
3040 }
3041
3042 static bool clone3_args_valid(struct kernel_clone_args *kargs)
3043 {
3044         /* Verify that no unknown flags are passed along. */
3045         if (kargs->flags &
3046             ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
3047                 return false;
3048
3049         /*
3050          * - make the CLONE_DETACHED bit reusable for clone3
3051          * - make the CSIGNAL bits reusable for clone3
3052          */
3053         if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
3054                 return false;
3055
3056         if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
3057             (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
3058                 return false;
3059
3060         if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
3061             kargs->exit_signal)
3062                 return false;
3063
3064         if (!clone3_stack_valid(kargs))
3065                 return false;
3066
3067         return true;
3068 }
3069
3070 /**
3071  * sys_clone3 - create a new process with specific properties
3072  * @uargs: argument structure
3073  * @size:  size of @uargs
3074  *
3075  * clone3() is the extensible successor to clone()/clone2().
3076  * It takes a struct as argument that is versioned by its size.
3077  *
3078  * Return: On success, a positive PID for the child process.
3079  *         On error, a negative errno number.
3080  */
3081 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
3082 {
3083         int err;
3084
3085         struct kernel_clone_args kargs;
3086         pid_t set_tid[MAX_PID_NS_LEVEL];
3087
3088         kargs.set_tid = set_tid;
3089
3090         err = copy_clone_args_from_user(&kargs, uargs, size);
3091         if (err)
3092                 return err;
3093
3094         if (!clone3_args_valid(&kargs))
3095                 return -EINVAL;
3096
3097         return kernel_clone(&kargs);
3098 }
3099 #endif
3100
3101 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
3102 {
3103         struct task_struct *leader, *parent, *child;
3104         int res;
3105
3106         read_lock(&tasklist_lock);
3107         leader = top = top->group_leader;
3108 down:
3109         for_each_thread(leader, parent) {
3110                 list_for_each_entry(child, &parent->children, sibling) {
3111                         res = visitor(child, data);
3112                         if (res) {
3113                                 if (res < 0)
3114                                         goto out;
3115                                 leader = child;
3116                                 goto down;
3117                         }
3118 up:
3119                         ;
3120                 }
3121         }
3122
3123         if (leader != top) {
3124                 child = leader;
3125                 parent = child->real_parent;
3126                 leader = parent->group_leader;
3127                 goto up;
3128         }
3129 out:
3130         read_unlock(&tasklist_lock);
3131 }
3132
3133 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
3134 #define ARCH_MIN_MMSTRUCT_ALIGN 0
3135 #endif
3136
3137 static void sighand_ctor(void *data)
3138 {
3139         struct sighand_struct *sighand = data;
3140
3141         spin_lock_init(&sighand->siglock);
3142         init_waitqueue_head(&sighand->signalfd_wqh);
3143 }
3144
3145 void __init mm_cache_init(void)
3146 {
3147         unsigned int mm_size;
3148
3149         /*
3150          * The mm_cpumask is located at the end of mm_struct, and is
3151          * dynamically sized based on the maximum CPU number this system
3152          * can have, taking hotplug into account (nr_cpu_ids).
3153          */
3154         mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
3155
3156         mm_cachep = kmem_cache_create_usercopy("mm_struct",
3157                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3158                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3159                         offsetof(struct mm_struct, saved_auxv),
3160                         sizeof_field(struct mm_struct, saved_auxv),
3161                         NULL);
3162 }
3163
3164 void __init proc_caches_init(void)
3165 {
3166         sighand_cachep = kmem_cache_create("sighand_cache",
3167                         sizeof(struct sighand_struct), 0,
3168                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3169                         SLAB_ACCOUNT, sighand_ctor);
3170         signal_cachep = kmem_cache_create("signal_cache",
3171                         sizeof(struct signal_struct), 0,
3172                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3173                         NULL);
3174         files_cachep = kmem_cache_create("files_cache",
3175                         sizeof(struct files_struct), 0,
3176                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3177                         NULL);
3178         fs_cachep = kmem_cache_create("fs_cache",
3179                         sizeof(struct fs_struct), 0,
3180                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3181                         NULL);
3182
3183         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3184 #ifdef CONFIG_PER_VMA_LOCK
3185         vma_lock_cachep = KMEM_CACHE(vma_lock, SLAB_PANIC|SLAB_ACCOUNT);
3186 #endif
3187         mmap_init();
3188         nsproxy_cache_init();
3189 }
3190
3191 /*
3192  * Check constraints on flags passed to the unshare system call.
3193  */
3194 static int check_unshare_flags(unsigned long unshare_flags)
3195 {
3196         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3197                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3198                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3199                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3200                                 CLONE_NEWTIME))
3201                 return -EINVAL;
3202         /*
3203          * Not implemented, but pretend it works if there is nothing
3204          * to unshare.  Note that unsharing the address space or the
3205          * signal handlers also need to unshare the signal queues (aka
3206          * CLONE_THREAD).
3207          */
3208         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3209                 if (!thread_group_empty(current))
3210                         return -EINVAL;
3211         }
3212         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3213                 if (refcount_read(&current->sighand->count) > 1)
3214                         return -EINVAL;
3215         }
3216         if (unshare_flags & CLONE_VM) {
3217                 if (!current_is_single_threaded())
3218                         return -EINVAL;
3219         }
3220
3221         return 0;
3222 }
3223
3224 /*
3225  * Unshare the filesystem structure if it is being shared
3226  */
3227 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3228 {
3229         struct fs_struct *fs = current->fs;
3230
3231         if (!(unshare_flags & CLONE_FS) || !fs)
3232                 return 0;
3233
3234         /* don't need lock here; in the worst case we'll do useless copy */
3235         if (fs->users == 1)
3236                 return 0;
3237
3238         *new_fsp = copy_fs_struct(fs);
3239         if (!*new_fsp)
3240                 return -ENOMEM;
3241
3242         return 0;
3243 }
3244
3245 /*
3246  * Unshare file descriptor table if it is being shared
3247  */
3248 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3249                struct files_struct **new_fdp)
3250 {
3251         struct files_struct *fd = current->files;
3252         int error = 0;
3253
3254         if ((unshare_flags & CLONE_FILES) &&
3255             (fd && atomic_read(&fd->count) > 1)) {
3256                 *new_fdp = dup_fd(fd, max_fds, &error);
3257                 if (!*new_fdp)
3258                         return error;
3259         }
3260
3261         return 0;
3262 }
3263
3264 /*
3265  * unshare allows a process to 'unshare' part of the process
3266  * context which was originally shared using clone.  copy_*
3267  * functions used by kernel_clone() cannot be used here directly
3268  * because they modify an inactive task_struct that is being
3269  * constructed. Here we are modifying the current, active,
3270  * task_struct.
3271  */
3272 int ksys_unshare(unsigned long unshare_flags)
3273 {
3274         struct fs_struct *fs, *new_fs = NULL;
3275         struct files_struct *new_fd = NULL;
3276         struct cred *new_cred = NULL;
3277         struct nsproxy *new_nsproxy = NULL;
3278         int do_sysvsem = 0;
3279         int err;
3280
3281         /*
3282          * If unsharing a user namespace must also unshare the thread group
3283          * and unshare the filesystem root and working directories.
3284          */
3285         if (unshare_flags & CLONE_NEWUSER)
3286                 unshare_flags |= CLONE_THREAD | CLONE_FS;
3287         /*
3288          * If unsharing vm, must also unshare signal handlers.
3289          */
3290         if (unshare_flags & CLONE_VM)
3291                 unshare_flags |= CLONE_SIGHAND;
3292         /*
3293          * If unsharing a signal handlers, must also unshare the signal queues.
3294          */
3295         if (unshare_flags & CLONE_SIGHAND)
3296                 unshare_flags |= CLONE_THREAD;
3297         /*
3298          * If unsharing namespace, must also unshare filesystem information.
3299          */
3300         if (unshare_flags & CLONE_NEWNS)
3301                 unshare_flags |= CLONE_FS;
3302
3303         err = check_unshare_flags(unshare_flags);
3304         if (err)
3305                 goto bad_unshare_out;
3306         /*
3307          * CLONE_NEWIPC must also detach from the undolist: after switching
3308          * to a new ipc namespace, the semaphore arrays from the old
3309          * namespace are unreachable.
3310          */
3311         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3312                 do_sysvsem = 1;
3313         err = unshare_fs(unshare_flags, &new_fs);
3314         if (err)
3315                 goto bad_unshare_out;
3316         err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3317         if (err)
3318                 goto bad_unshare_cleanup_fs;
3319         err = unshare_userns(unshare_flags, &new_cred);
3320         if (err)
3321                 goto bad_unshare_cleanup_fd;
3322         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3323                                          new_cred, new_fs);
3324         if (err)
3325                 goto bad_unshare_cleanup_cred;
3326
3327         if (new_cred) {
3328                 err = set_cred_ucounts(new_cred);
3329                 if (err)
3330                         goto bad_unshare_cleanup_cred;
3331         }
3332
3333         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3334                 if (do_sysvsem) {
3335                         /*
3336                          * CLONE_SYSVSEM is equivalent to sys_exit().
3337                          */
3338                         exit_sem(current);
3339                 }
3340                 if (unshare_flags & CLONE_NEWIPC) {
3341                         /* Orphan segments in old ns (see sem above). */
3342                         exit_shm(current);
3343                         shm_init_task(current);
3344                 }
3345
3346                 if (new_nsproxy)
3347                         switch_task_namespaces(current, new_nsproxy);
3348
3349                 task_lock(current);
3350
3351                 if (new_fs) {
3352                         fs = current->fs;
3353                         spin_lock(&fs->lock);
3354                         current->fs = new_fs;
3355                         if (--fs->users)
3356                                 new_fs = NULL;
3357                         else
3358                                 new_fs = fs;
3359                         spin_unlock(&fs->lock);
3360                 }
3361
3362                 if (new_fd)
3363                         swap(current->files, new_fd);
3364
3365                 task_unlock(current);
3366
3367                 if (new_cred) {
3368                         /* Install the new user namespace */
3369                         commit_creds(new_cred);
3370                         new_cred = NULL;
3371                 }
3372         }
3373
3374         perf_event_namespaces(current);
3375
3376 bad_unshare_cleanup_cred:
3377         if (new_cred)
3378                 put_cred(new_cred);
3379 bad_unshare_cleanup_fd:
3380         if (new_fd)
3381                 put_files_struct(new_fd);
3382
3383 bad_unshare_cleanup_fs:
3384         if (new_fs)
3385                 free_fs_struct(new_fs);
3386
3387 bad_unshare_out:
3388         return err;
3389 }
3390
3391 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3392 {
3393         return ksys_unshare(unshare_flags);
3394 }
3395
3396 /*
3397  *      Helper to unshare the files of the current task.
3398  *      We don't want to expose copy_files internals to
3399  *      the exec layer of the kernel.
3400  */
3401
3402 int unshare_files(void)
3403 {
3404         struct task_struct *task = current;
3405         struct files_struct *old, *copy = NULL;
3406         int error;
3407
3408         error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, &copy);
3409         if (error || !copy)
3410                 return error;
3411
3412         old = task->files;
3413         task_lock(task);
3414         task->files = copy;
3415         task_unlock(task);
3416         put_files_struct(old);
3417         return 0;
3418 }
3419
3420 int sysctl_max_threads(struct ctl_table *table, int write,
3421                        void *buffer, size_t *lenp, loff_t *ppos)
3422 {
3423         struct ctl_table t;
3424         int ret;
3425         int threads = max_threads;
3426         int min = 1;
3427         int max = MAX_THREADS;
3428
3429         t = *table;
3430         t.data = &threads;
3431         t.extra1 = &min;
3432         t.extra2 = &max;
3433
3434         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3435         if (ret || !write)
3436                 return ret;
3437
3438         max_threads = threads;
3439
3440         return 0;
3441 }