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