kernel/fork.c

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