kernel/fork.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/kernel/fork.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  */
   7
   8 /*
   9  *  'fork.c' contains the help-routines for the 'fork' system call
  10  * (see also entry.S and others).
  11  * Fork is rather simple, once you get the hang of it, but the memory
  12  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  13  */
  14
  15 #include <linux/anon_inodes.h>
  16 #include <linux/slab.h>
  17 #include <linux/sched/autogroup.h>
  18 #include <linux/sched/mm.h>
  19 #include <linux/sched/coredump.h>
  20 #include <linux/sched/user.h>
  21 #include <linux/sched/numa_balancing.h>
  22 #include <linux/sched/stat.h>
  23 #include <linux/sched/task.h>
  24 #include <linux/sched/task_stack.h>
  25 #include <linux/sched/cputime.h>
  26 #include <linux/seq_file.h>
  27 #include <linux/rtmutex.h>
  28 #include <linux/init.h>
  29 #include <linux/unistd.h>
  30 #include <linux/module.h>
  31 #include <linux/vmalloc.h>
  32 #include <linux/completion.h>
  33 #include <linux/personality.h>
  34 #include <linux/mempolicy.h>
  35 #include <linux/sem.h>
  36 #include <linux/file.h>
  37 #include <linux/fdtable.h>
  38 #include <linux/iocontext.h>
  39 #include <linux/key.h>
  40 #include <linux/binfmts.h>
  41 #include <linux/mman.h>
  42 #include <linux/mmu_notifier.h>
  43 #include <linux/fs.h>
  44 #include <linux/mm.h>
  45 #include <linux/vmacache.h>
  46 #include <linux/nsproxy.h>
  47 #include <linux/capability.h>
  48 #include <linux/cpu.h>
  49 #include <linux/cgroup.h>
  50 #include <linux/security.h>
  51 #include <linux/hugetlb.h>
  52 #include <linux/seccomp.h>
  53 #include <linux/swap.h>
  54 #include <linux/syscalls.h>
  55 #include <linux/jiffies.h>
  56 #include <linux/futex.h>
  57 #include <linux/compat.h>
  58 #include <linux/kthread.h>
  59 #include <linux/task_io_accounting_ops.h>
  60 #include <linux/rcupdate.h>
  61 #include <linux/ptrace.h>
  62 #include <linux/mount.h>
  63 #include <linux/audit.h>
  64 #include <linux/memcontrol.h>
  65 #include <linux/ftrace.h>
  66 #include <linux/proc_fs.h>
  67 #include <linux/profile.h>
  68 #include <linux/rmap.h>
  69 #include <linux/ksm.h>
  70 #include <linux/acct.h>
  71 #include <linux/userfaultfd_k.h>
  72 #include <linux/tsacct_kern.h>
  73 #include <linux/cn_proc.h>
  74 #include <linux/freezer.h>
  75 #include <linux/delayacct.h>
  76 #include <linux/taskstats_kern.h>
  77 #include <linux/random.h>
  78 #include <linux/tty.h>
  79 #include <linux/blkdev.h>
  80 #include <linux/fs_struct.h>
  81 #include <linux/magic.h>
  82 #include <linux/perf_event.h>
  83 #include <linux/posix-timers.h>
  84 #include <linux/user-return-notifier.h>
  85 #include <linux/oom.h>
  86 #include <linux/khugepaged.h>
  87 #include <linux/signalfd.h>
  88 #include <linux/uprobes.h>
  89 #include <linux/aio.h>
  90 #include <linux/compiler.h>
  91 #include <linux/sysctl.h>
  92 #include <linux/kcov.h>
  93 #include <linux/livepatch.h>
  94 #include <linux/thread_info.h>
  95 #include <linux/stackleak.h>
  96 #include <linux/kasan.h>
  97 #include <linux/scs.h>
  98 #include <linux/io_uring.h>
  99
 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                 /* Clear the KASAN shadow of the stack. */
 229                 kasan_unpoison_shadow(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
 383         /* All stack pages are in the same node. */
 384         if (vm)
 385                 mod_lruvec_page_state(vm->pages[0], NR_KERNEL_STACK_KB,
 386                                       account * (THREAD_SIZE / 1024));
 387         else
 388                 mod_lruvec_slab_state(stack, NR_KERNEL_STACK_KB,
 389                                       account * (THREAD_SIZE / 1024));
 390 }
 391
 392 static int memcg_charge_kernel_stack(struct task_struct *tsk)
 393 {
 394 #ifdef CONFIG_VMAP_STACK
 395         struct vm_struct *vm = task_stack_vm_area(tsk);
 396         int ret;
 397
 398         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 399
 400         if (vm) {
 401                 int i;
 402
 403                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 404
 405                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 406                         /*
 407                          * If memcg_kmem_charge_page() fails, page->mem_cgroup
 408                          * pointer is NULL, and memcg_kmem_uncharge_page() in
 409                          * free_thread_stack() will ignore this page.
 410                          */
 411                         ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL,
 412                                                      0);
 413                         if (ret)
 414                                 return ret;
 415                 }
 416         }
 417 #endif
 418         return 0;
 419 }
 420
 421 static void release_task_stack(struct task_struct *tsk)
 422 {
 423         if (WARN_ON(tsk->state != TASK_DEAD))
 424                 return;  /* Better to leak the stack than to free prematurely */
 425
 426         account_kernel_stack(tsk, -1);
 427         free_thread_stack(tsk);
 428         tsk->stack = NULL;
 429 #ifdef CONFIG_VMAP_STACK
 430         tsk->stack_vm_area = NULL;
 431 #endif
 432 }
 433
 434 #ifdef CONFIG_THREAD_INFO_IN_TASK
 435 void put_task_stack(struct task_struct *tsk)
 436 {
 437         if (refcount_dec_and_test(&tsk->stack_refcount))
 438                 release_task_stack(tsk);
 439 }
 440 #endif
 441
 442 void free_task(struct task_struct *tsk)
 443 {
 444         scs_release(tsk);
 445
 446 #ifndef CONFIG_THREAD_INFO_IN_TASK
 447         /*
 448          * The task is finally done with both the stack and thread_info,
 449          * so free both.
 450          */
 451         release_task_stack(tsk);
 452 #else
 453         /*
 454          * If the task had a separate stack allocation, it should be gone
 455          * by now.
 456          */
 457         WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
 458 #endif
 459         rt_mutex_debug_task_free(tsk);
 460         ftrace_graph_exit_task(tsk);
 461         arch_release_task_struct(tsk);
 462         if (tsk->flags & PF_KTHREAD)
 463                 free_kthread_struct(tsk);
 464         free_task_struct(tsk);
 465 }
 466 EXPORT_SYMBOL(free_task);
 467
 468 #ifdef CONFIG_MMU
 469 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 470                                         struct mm_struct *oldmm)
 471 {
 472         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 473         struct rb_node **rb_link, *rb_parent;
 474         int retval;
 475         unsigned long charge;
 476         LIST_HEAD(uf);
 477
 478         uprobe_start_dup_mmap();
 479         if (mmap_write_lock_killable(oldmm)) {
 480                 retval = -EINTR;
 481                 goto fail_uprobe_end;
 482         }
 483         flush_cache_dup_mm(oldmm);
 484         uprobe_dup_mmap(oldmm, mm);
 485         /*
 486          * Not linked in yet - no deadlock potential:
 487          */
 488         mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
 489
 490         /* No ordering required: file already has been exposed. */
 491         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 492
 493         mm->total_vm = oldmm->total_vm;
 494         mm->data_vm = oldmm->data_vm;
 495         mm->exec_vm = oldmm->exec_vm;
 496         mm->stack_vm = oldmm->stack_vm;
 497
 498         rb_link = &mm->mm_rb.rb_node;
 499         rb_parent = NULL;
 500         pprev = &mm->mmap;
 501         retval = ksm_fork(mm, oldmm);
 502         if (retval)
 503                 goto out;
 504         retval = khugepaged_fork(mm, oldmm);
 505         if (retval)
 506                 goto out;
 507
 508         prev = NULL;
 509         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 510                 struct file *file;
 511
 512                 if (mpnt->vm_flags & VM_DONTCOPY) {
 513                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 514                         continue;
 515                 }
 516                 charge = 0;
 517                 /*
 518                  * Don't duplicate many vmas if we've been oom-killed (for
 519                  * example)
 520                  */
 521                 if (fatal_signal_pending(current)) {
 522                         retval = -EINTR;
 523                         goto out;
 524                 }
 525                 if (mpnt->vm_flags & VM_ACCOUNT) {
 526                         unsigned long len = vma_pages(mpnt);
 527
 528                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 529                                 goto fail_nomem;
 530                         charge = len;
 531                 }
 532                 tmp = vm_area_dup(mpnt);
 533                 if (!tmp)
 534                         goto fail_nomem;
 535                 retval = vma_dup_policy(mpnt, tmp);
 536                 if (retval)
 537                         goto fail_nomem_policy;
 538                 tmp->vm_mm = mm;
 539                 retval = dup_userfaultfd(tmp, &uf);
 540                 if (retval)
 541                         goto fail_nomem_anon_vma_fork;
 542                 if (tmp->vm_flags & VM_WIPEONFORK) {
 543                         /*
 544                          * VM_WIPEONFORK gets a clean slate in the child.
 545                          * Don't prepare anon_vma until fault since we don't
 546                          * copy page for current vma.
 547                          */
 548                         tmp->anon_vma = NULL;
 549                 } else if (anon_vma_fork(tmp, mpnt))
 550                         goto fail_nomem_anon_vma_fork;
 551                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 552                 file = tmp->vm_file;
 553                 if (file) {
 554                         struct inode *inode = file_inode(file);
 555                         struct address_space *mapping = file->f_mapping;
 556
 557                         get_file(file);
 558                         if (tmp->vm_flags & VM_DENYWRITE)
 559                                 put_write_access(inode);
 560                         i_mmap_lock_write(mapping);
 561                         if (tmp->vm_flags & VM_SHARED)
 562                                 mapping_allow_writable(mapping);
 563                         flush_dcache_mmap_lock(mapping);
 564                         /* insert tmp into the share list, just after mpnt */
 565                         vma_interval_tree_insert_after(tmp, mpnt,
 566                                         &mapping->i_mmap);
 567                         flush_dcache_mmap_unlock(mapping);
 568                         i_mmap_unlock_write(mapping);
 569                 }
 570
 571                 /*
 572                  * Clear hugetlb-related page reserves for children. This only
 573                  * affects MAP_PRIVATE mappings. Faults generated by the child
 574                  * are not guaranteed to succeed, even if read-only
 575                  */
 576                 if (is_vm_hugetlb_page(tmp))
 577                         reset_vma_resv_huge_pages(tmp);
 578
 579                 /*
 580                  * Link in the new vma and copy the page table entries.
 581                  */
 582                 *pprev = tmp;
 583                 pprev = &tmp->vm_next;
 584                 tmp->vm_prev = prev;
 585                 prev = tmp;
 586
 587                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 588                 rb_link = &tmp->vm_rb.rb_right;
 589                 rb_parent = &tmp->vm_rb;
 590
 591                 mm->map_count++;
 592                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 593                         retval = copy_page_range(tmp, mpnt);
 594
 595                 if (tmp->vm_ops && tmp->vm_ops->open)
 596                         tmp->vm_ops->open(tmp);
 597
 598                 if (retval)
 599                         goto out;
 600         }
 601         /* a new mm has just been created */
 602         retval = arch_dup_mmap(oldmm, mm);
 603 out:
 604         mmap_write_unlock(mm);
 605         flush_tlb_mm(oldmm);
 606         mmap_write_unlock(oldmm);
 607         dup_userfaultfd_complete(&uf);
 608 fail_uprobe_end:
 609         uprobe_end_dup_mmap();
 610         return retval;
 611 fail_nomem_anon_vma_fork:
 612         mpol_put(vma_policy(tmp));
 613 fail_nomem_policy:
 614         vm_area_free(tmp);
 615 fail_nomem:
 616         retval = -ENOMEM;
 617         vm_unacct_memory(charge);
 618         goto out;
 619 }
 620
 621 static inline int mm_alloc_pgd(struct mm_struct *mm)
 622 {
 623         mm->pgd = pgd_alloc(mm);
 624         if (unlikely(!mm->pgd))
 625                 return -ENOMEM;
 626         return 0;
 627 }
 628
 629 static inline void mm_free_pgd(struct mm_struct *mm)
 630 {
 631         pgd_free(mm, mm->pgd);
 632 }
 633 #else
 634 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 635 {
 636         mmap_write_lock(oldmm);
 637         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 638         mmap_write_unlock(oldmm);
 639         return 0;
 640 }
 641 #define mm_alloc_pgd(mm)        (0)
 642 #define mm_free_pgd(mm)
 643 #endif /* CONFIG_MMU */
 644
 645 static void check_mm(struct mm_struct *mm)
 646 {
 647         int i;
 648
 649         BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
 650                          "Please make sure 'struct resident_page_types[]' is updated as well");
 651
 652         for (i = 0; i < NR_MM_COUNTERS; i++) {
 653                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 654
 655                 if (unlikely(x))
 656                         pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
 657                                  mm, resident_page_types[i], x);
 658         }
 659
 660         if (mm_pgtables_bytes(mm))
 661                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 662                                 mm_pgtables_bytes(mm));
 663
 664 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 665         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 666 #endif
 667 }
 668
 669 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 670 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 671
 672 /*
 673  * Called when the last reference to the mm
 674  * is dropped: either by a lazy thread or by
 675  * mmput. Free the page directory and the mm.
 676  */
 677 void __mmdrop(struct mm_struct *mm)
 678 {
 679         BUG_ON(mm == &init_mm);
 680         WARN_ON_ONCE(mm == current->mm);
 681         WARN_ON_ONCE(mm == current->active_mm);
 682         mm_free_pgd(mm);
 683         destroy_context(mm);
 684         mmu_notifier_subscriptions_destroy(mm);
 685         check_mm(mm);
 686         put_user_ns(mm->user_ns);
 687         free_mm(mm);
 688 }
 689 EXPORT_SYMBOL_GPL(__mmdrop);
 690
 691 static void mmdrop_async_fn(struct work_struct *work)
 692 {
 693         struct mm_struct *mm;
 694
 695         mm = container_of(work, struct mm_struct, async_put_work);
 696         __mmdrop(mm);
 697 }
 698
 699 static void mmdrop_async(struct mm_struct *mm)
 700 {
 701         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 702                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 703                 schedule_work(&mm->async_put_work);
 704         }
 705 }
 706
 707 static inline void free_signal_struct(struct signal_struct *sig)
 708 {
 709         taskstats_tgid_free(sig);
 710         sched_autogroup_exit(sig);
 711         /*
 712          * __mmdrop is not safe to call from softirq context on x86 due to
 713          * pgd_dtor so postpone it to the async context
 714          */
 715         if (sig->oom_mm)
 716                 mmdrop_async(sig->oom_mm);
 717         kmem_cache_free(signal_cachep, sig);
 718 }
 719
 720 static inline void put_signal_struct(struct signal_struct *sig)
 721 {
 722         if (refcount_dec_and_test(&sig->sigcnt))
 723                 free_signal_struct(sig);
 724 }
 725
 726 void __put_task_struct(struct task_struct *tsk)
 727 {
 728         WARN_ON(!tsk->exit_state);
 729         WARN_ON(refcount_read(&tsk->usage));
 730         WARN_ON(tsk == current);
 731
 732         io_uring_free(tsk);
 733         cgroup_free(tsk);
 734         task_numa_free(tsk, true);
 735         security_task_free(tsk);
 736         exit_creds(tsk);
 737         delayacct_tsk_free(tsk);
 738         put_signal_struct(tsk->signal);
 739
 740         if (!profile_handoff_task(tsk))
 741                 free_task(tsk);
 742 }
 743 EXPORT_SYMBOL_GPL(__put_task_struct);
 744
 745 void __init __weak arch_task_cache_init(void) { }
 746
 747 /*
 748  * set_max_threads
 749  */
 750 static void set_max_threads(unsigned int max_threads_suggested)
 751 {
 752         u64 threads;
 753         unsigned long nr_pages = totalram_pages();
 754
 755         /*
 756          * The number of threads shall be limited such that the thread
 757          * structures may only consume a small part of the available memory.
 758          */
 759         if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
 760                 threads = MAX_THREADS;
 761         else
 762                 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
 763                                     (u64) THREAD_SIZE * 8UL);
 764
 765         if (threads > max_threads_suggested)
 766                 threads = max_threads_suggested;
 767
 768         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 769 }
 770
 771 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 772 /* Initialized by the architecture: */
 773 int arch_task_struct_size __read_mostly;
 774 #endif
 775
 776 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 777 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 778 {
 779         /* Fetch thread_struct whitelist for the architecture. */
 780         arch_thread_struct_whitelist(offset, size);
 781
 782         /*
 783          * Handle zero-sized whitelist or empty thread_struct, otherwise
 784          * adjust offset to position of thread_struct in task_struct.
 785          */
 786         if (unlikely(*size == 0))
 787                 *offset = 0;
 788         else
 789                 *offset += offsetof(struct task_struct, thread);
 790 }
 791 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
 792
 793 void __init fork_init(void)
 794 {
 795         int i;
 796 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 797 #ifndef ARCH_MIN_TASKALIGN
 798 #define ARCH_MIN_TASKALIGN      0
 799 #endif
 800         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 801         unsigned long useroffset, usersize;
 802
 803         /* create a slab on which task_structs can be allocated */
 804         task_struct_whitelist(&useroffset, &usersize);
 805         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 806                         arch_task_struct_size, align,
 807                         SLAB_PANIC|SLAB_ACCOUNT,
 808                         useroffset, usersize, NULL);
 809 #endif
 810
 811         /* do the arch specific task caches init */
 812         arch_task_cache_init();
 813
 814         set_max_threads(MAX_THREADS);
 815
 816         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 817         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 818         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 819                 init_task.signal->rlim[RLIMIT_NPROC];
 820
 821         for (i = 0; i < UCOUNT_COUNTS; i++) {
 822                 init_user_ns.ucount_max[i] = max_threads/2;
 823         }
 824
 825 #ifdef CONFIG_VMAP_STACK
 826         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 827                           NULL, free_vm_stack_cache);
 828 #endif
 829
 830         scs_init();
 831
 832         lockdep_init_task(&init_task);
 833         uprobes_init();
 834 }
 835
 836 int __weak arch_dup_task_struct(struct task_struct *dst,
 837                                                struct task_struct *src)
 838 {
 839         *dst = *src;
 840         return 0;
 841 }
 842
 843 void set_task_stack_end_magic(struct task_struct *tsk)
 844 {
 845         unsigned long *stackend;
 846
 847         stackend = end_of_stack(tsk);
 848         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 849 }
 850
 851 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 852 {
 853         struct task_struct *tsk;
 854         unsigned long *stack;
 855         struct vm_struct *stack_vm_area __maybe_unused;
 856         int err;
 857
 858         if (node == NUMA_NO_NODE)
 859                 node = tsk_fork_get_node(orig);
 860         tsk = alloc_task_struct_node(node);
 861         if (!tsk)
 862                 return NULL;
 863
 864         stack = alloc_thread_stack_node(tsk, node);
 865         if (!stack)
 866                 goto free_tsk;
 867
 868         if (memcg_charge_kernel_stack(tsk))
 869                 goto free_stack;
 870
 871         stack_vm_area = task_stack_vm_area(tsk);
 872
 873         err = arch_dup_task_struct(tsk, orig);
 874
 875         /*
 876          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 877          * sure they're properly initialized before using any stack-related
 878          * functions again.
 879          */
 880         tsk->stack = stack;
 881 #ifdef CONFIG_VMAP_STACK
 882         tsk->stack_vm_area = stack_vm_area;
 883 #endif
 884 #ifdef CONFIG_THREAD_INFO_IN_TASK
 885         refcount_set(&tsk->stack_refcount, 1);
 886 #endif
 887
 888         if (err)
 889                 goto free_stack;
 890
 891         err = scs_prepare(tsk, node);
 892         if (err)
 893                 goto free_stack;
 894
 895 #ifdef CONFIG_SECCOMP
 896         /*
 897          * We must handle setting up seccomp filters once we're under
 898          * the sighand lock in case orig has changed between now and
 899          * then. Until then, filter must be NULL to avoid messing up
 900          * the usage counts on the error path calling free_task.
 901          */
 902         tsk->seccomp.filter = NULL;
 903 #endif
 904
 905         setup_thread_stack(tsk, orig);
 906         clear_user_return_notifier(tsk);
 907         clear_tsk_need_resched(tsk);
 908         set_task_stack_end_magic(tsk);
 909         clear_syscall_work_syscall_user_dispatch(tsk);
 910
 911 #ifdef CONFIG_STACKPROTECTOR
 912         tsk->stack_canary = get_random_canary();
 913 #endif
 914         if (orig->cpus_ptr == &orig->cpus_mask)
 915                 tsk->cpus_ptr = &tsk->cpus_mask;
 916
 917         /*
 918          * One for the user space visible state that goes away when reaped.
 919          * One for the scheduler.
 920          */
 921         refcount_set(&tsk->rcu_users, 2);
 922         /* One for the rcu users */
 923         refcount_set(&tsk->usage, 1);
 924 #ifdef CONFIG_BLK_DEV_IO_TRACE
 925         tsk->btrace_seq = 0;
 926 #endif
 927         tsk->splice_pipe = NULL;
 928         tsk->task_frag.page = NULL;
 929         tsk->wake_q.next = NULL;
 930
 931         account_kernel_stack(tsk, 1);
 932
 933         kcov_task_init(tsk);
 934
 935 #ifdef CONFIG_FAULT_INJECTION
 936         tsk->fail_nth = 0;
 937 #endif
 938
 939 #ifdef CONFIG_BLK_CGROUP
 940         tsk->throttle_queue = NULL;
 941         tsk->use_memdelay = 0;
 942 #endif
 943
 944 #ifdef CONFIG_MEMCG
 945         tsk->active_memcg = NULL;
 946 #endif
 947         return tsk;
 948
 949 free_stack:
 950         free_thread_stack(tsk);
 951 free_tsk:
 952         free_task_struct(tsk);
 953         return NULL;
 954 }
 955
 956 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 957
 958 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 959
 960 static int __init coredump_filter_setup(char *s)
 961 {
 962         default_dump_filter =
 963                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 964                 MMF_DUMP_FILTER_MASK;
 965         return 1;
 966 }
 967
 968 __setup("coredump_filter=", coredump_filter_setup);
 969
 970 #include <linux/init_task.h>
 971
 972 static void mm_init_aio(struct mm_struct *mm)
 973 {
 974 #ifdef CONFIG_AIO
 975         spin_lock_init(&mm->ioctx_lock);
 976         mm->ioctx_table = NULL;
 977 #endif
 978 }
 979
 980 static __always_inline void mm_clear_owner(struct mm_struct *mm,
 981                                            struct task_struct *p)
 982 {
 983 #ifdef CONFIG_MEMCG
 984         if (mm->owner == p)
 985                 WRITE_ONCE(mm->owner, NULL);
 986 #endif
 987 }
 988
 989 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 990 {
 991 #ifdef CONFIG_MEMCG
 992         mm->owner = p;
 993 #endif
 994 }
 995
 996 static void mm_init_uprobes_state(struct mm_struct *mm)
 997 {
 998 #ifdef CONFIG_UPROBES
 999         mm->uprobes_state.xol_area = NULL;
1000 #endif
1001 }
1002
1003 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1004         struct user_namespace *user_ns)
1005 {
1006         mm->mmap = NULL;
1007         mm->mm_rb = RB_ROOT;
1008         mm->vmacache_seqnum = 0;
1009         atomic_set(&mm->mm_users, 1);
1010         atomic_set(&mm->mm_count, 1);
1011         mmap_init_lock(mm);
1012         INIT_LIST_HEAD(&mm->mmlist);
1013         mm->core_state = NULL;
1014         mm_pgtables_bytes_init(mm);
1015         mm->map_count = 0;
1016         mm->locked_vm = 0;
1017         atomic_set(&mm->has_pinned, 0);
1018         atomic64_set(&mm->pinned_vm, 0);
1019         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1020         spin_lock_init(&mm->page_table_lock);
1021         spin_lock_init(&mm->arg_lock);
1022         mm_init_cpumask(mm);
1023         mm_init_aio(mm);
1024         mm_init_owner(mm, p);
1025         RCU_INIT_POINTER(mm->exe_file, NULL);
1026         mmu_notifier_subscriptions_init(mm);
1027         init_tlb_flush_pending(mm);
1028 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1029         mm->pmd_huge_pte = NULL;
1030 #endif
1031         mm_init_uprobes_state(mm);
1032
1033         if (current->mm) {
1034                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1035                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1036         } else {
1037                 mm->flags = default_dump_filter;
1038                 mm->def_flags = 0;
1039         }
1040
1041         if (mm_alloc_pgd(mm))
1042                 goto fail_nopgd;
1043
1044         if (init_new_context(p, mm))
1045                 goto fail_nocontext;
1046
1047         mm->user_ns = get_user_ns(user_ns);
1048         return mm;
1049
1050 fail_nocontext:
1051         mm_free_pgd(mm);
1052 fail_nopgd:
1053         free_mm(mm);
1054         return NULL;
1055 }
1056
1057 /*
1058  * Allocate and initialize an mm_struct.
1059  */
1060 struct mm_struct *mm_alloc(void)
1061 {
1062         struct mm_struct *mm;
1063
1064         mm = allocate_mm();
1065         if (!mm)
1066                 return NULL;
1067
1068         memset(mm, 0, sizeof(*mm));
1069         return mm_init(mm, current, current_user_ns());
1070 }
1071
1072 static inline void __mmput(struct mm_struct *mm)
1073 {
1074         VM_BUG_ON(atomic_read(&mm->mm_users));
1075
1076         uprobe_clear_state(mm);
1077         exit_aio(mm);
1078         ksm_exit(mm);
1079         khugepaged_exit(mm); /* must run before exit_mmap */
1080         exit_mmap(mm);
1081         mm_put_huge_zero_page(mm);
1082         set_mm_exe_file(mm, NULL);
1083         if (!list_empty(&mm->mmlist)) {
1084                 spin_lock(&mmlist_lock);
1085                 list_del(&mm->mmlist);
1086                 spin_unlock(&mmlist_lock);
1087         }
1088         if (mm->binfmt)
1089                 module_put(mm->binfmt->module);
1090         mmdrop(mm);
1091 }
1092
1093 /*
1094  * Decrement the use count and release all resources for an mm.
1095  */
1096 void mmput(struct mm_struct *mm)
1097 {
1098         might_sleep();
1099
1100         if (atomic_dec_and_test(&mm->mm_users))
1101                 __mmput(mm);
1102 }
1103 EXPORT_SYMBOL_GPL(mmput);
1104
1105 #ifdef CONFIG_MMU
1106 static void mmput_async_fn(struct work_struct *work)
1107 {
1108         struct mm_struct *mm = container_of(work, struct mm_struct,
1109                                             async_put_work);
1110
1111         __mmput(mm);
1112 }
1113
1114 void mmput_async(struct mm_struct *mm)
1115 {
1116         if (atomic_dec_and_test(&mm->mm_users)) {
1117                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1118                 schedule_work(&mm->async_put_work);
1119         }
1120 }
1121 #endif
1122
1123 /**
1124  * set_mm_exe_file - change a reference to the mm's executable file
1125  *
1126  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1127  *
1128  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1129  * invocations: in mmput() nobody alive left, in execve task is single
1130  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1131  * mm->exe_file, but does so without using set_mm_exe_file() in order
1132  * to do avoid the need for any locks.
1133  */
1134 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1135 {
1136         struct file *old_exe_file;
1137
1138         /*
1139          * It is safe to dereference the exe_file without RCU as
1140          * this function is only called if nobody else can access
1141          * this mm -- see comment above for justification.
1142          */
1143         old_exe_file = rcu_dereference_raw(mm->exe_file);
1144
1145         if (new_exe_file)
1146                 get_file(new_exe_file);
1147         rcu_assign_pointer(mm->exe_file, new_exe_file);
1148         if (old_exe_file)
1149                 fput(old_exe_file);
1150 }
1151
1152 /**
1153  * get_mm_exe_file - acquire a reference to the mm's executable file
1154  *
1155  * Returns %NULL if mm has no associated executable file.
1156  * User must release file via fput().
1157  */
1158 struct file *get_mm_exe_file(struct mm_struct *mm)
1159 {
1160         struct file *exe_file;
1161
1162         rcu_read_lock();
1163         exe_file = rcu_dereference(mm->exe_file);
1164         if (exe_file && !get_file_rcu(exe_file))
1165                 exe_file = NULL;
1166         rcu_read_unlock();
1167         return exe_file;
1168 }
1169 EXPORT_SYMBOL(get_mm_exe_file);
1170
1171 /**
1172  * get_task_exe_file - acquire a reference to the task's executable file
1173  *
1174  * Returns %NULL if task's mm (if any) has no associated executable file or
1175  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1176  * User must release file via fput().
1177  */
1178 struct file *get_task_exe_file(struct task_struct *task)
1179 {
1180         struct file *exe_file = NULL;
1181         struct mm_struct *mm;
1182
1183         task_lock(task);
1184         mm = task->mm;
1185         if (mm) {
1186                 if (!(task->flags & PF_KTHREAD))
1187                         exe_file = get_mm_exe_file(mm);
1188         }
1189         task_unlock(task);
1190         return exe_file;
1191 }
1192 EXPORT_SYMBOL(get_task_exe_file);
1193
1194 /**
1195  * get_task_mm - acquire a reference to the task's mm
1196  *
1197  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1198  * this kernel workthread has transiently adopted a user mm with use_mm,
1199  * to do its AIO) is not set and if so returns a reference to it, after
1200  * bumping up the use count.  User must release the mm via mmput()
1201  * after use.  Typically used by /proc and ptrace.
1202  */
1203 struct mm_struct *get_task_mm(struct task_struct *task)
1204 {
1205         struct mm_struct *mm;
1206
1207         task_lock(task);
1208         mm = task->mm;
1209         if (mm) {
1210                 if (task->flags & PF_KTHREAD)
1211                         mm = NULL;
1212                 else
1213                         mmget(mm);
1214         }
1215         task_unlock(task);
1216         return mm;
1217 }
1218 EXPORT_SYMBOL_GPL(get_task_mm);
1219
1220 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1221 {
1222         struct mm_struct *mm;
1223         int err;
1224
1225         err =  mutex_lock_killable(&task->signal->exec_update_mutex);
1226         if (err)
1227                 return ERR_PTR(err);
1228
1229         mm = get_task_mm(task);
1230         if (mm && mm != current->mm &&
1231                         !ptrace_may_access(task, mode)) {
1232                 mmput(mm);
1233                 mm = ERR_PTR(-EACCES);
1234         }
1235         mutex_unlock(&task->signal->exec_update_mutex);
1236
1237         return mm;
1238 }
1239
1240 static void complete_vfork_done(struct task_struct *tsk)
1241 {
1242         struct completion *vfork;
1243
1244         task_lock(tsk);
1245         vfork = tsk->vfork_done;
1246         if (likely(vfork)) {
1247                 tsk->vfork_done = NULL;
1248                 complete(vfork);
1249         }
1250         task_unlock(tsk);
1251 }
1252
1253 static int wait_for_vfork_done(struct task_struct *child,
1254                                 struct completion *vfork)
1255 {
1256         int killed;
1257
1258         freezer_do_not_count();
1259         cgroup_enter_frozen();
1260         killed = wait_for_completion_killable(vfork);
1261         cgroup_leave_frozen(false);
1262         freezer_count();
1263
1264         if (killed) {
1265                 task_lock(child);
1266                 child->vfork_done = NULL;
1267                 task_unlock(child);
1268         }
1269
1270         put_task_struct(child);
1271         return killed;
1272 }
1273
1274 /* Please note the differences between mmput and mm_release.
1275  * mmput is called whenever we stop holding onto a mm_struct,
1276  * error success whatever.
1277  *
1278  * mm_release is called after a mm_struct has been removed
1279  * from the current process.
1280  *
1281  * This difference is important for error handling, when we
1282  * only half set up a mm_struct for a new process and need to restore
1283  * the old one.  Because we mmput the new mm_struct before
1284  * restoring the old one. . .
1285  * Eric Biederman 10 January 1998
1286  */
1287 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1288 {
1289         uprobe_free_utask(tsk);
1290
1291         /* Get rid of any cached register state */
1292         deactivate_mm(tsk, mm);
1293
1294         /*
1295          * Signal userspace if we're not exiting with a core dump
1296          * because we want to leave the value intact for debugging
1297          * purposes.
1298          */
1299         if (tsk->clear_child_tid) {
1300                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1301                     atomic_read(&mm->mm_users) > 1) {
1302                         /*
1303                          * We don't check the error code - if userspace has
1304                          * not set up a proper pointer then tough luck.
1305                          */
1306                         put_user(0, tsk->clear_child_tid);
1307                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1308                                         1, NULL, NULL, 0, 0);
1309                 }
1310                 tsk->clear_child_tid = NULL;
1311         }
1312
1313         /*
1314          * All done, finally we can wake up parent and return this mm to him.
1315          * Also kthread_stop() uses this completion for synchronization.
1316          */
1317         if (tsk->vfork_done)
1318                 complete_vfork_done(tsk);
1319 }
1320
1321 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1322 {
1323         futex_exit_release(tsk);
1324         mm_release(tsk, mm);
1325 }
1326
1327 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1328 {
1329         futex_exec_release(tsk);
1330         mm_release(tsk, mm);
1331 }
1332
1333 /**
1334  * dup_mm() - duplicates an existing mm structure
1335  * @tsk: the task_struct with which the new mm will be associated.
1336  * @oldmm: the mm to duplicate.
1337  *
1338  * Allocates a new mm structure and duplicates the provided @oldmm structure
1339  * content into it.
1340  *
1341  * Return: the duplicated mm or NULL on failure.
1342  */
1343 static struct mm_struct *dup_mm(struct task_struct *tsk,
1344                                 struct mm_struct *oldmm)
1345 {
1346         struct mm_struct *mm;
1347         int err;
1348
1349         mm = allocate_mm();
1350         if (!mm)
1351                 goto fail_nomem;
1352
1353         memcpy(mm, oldmm, sizeof(*mm));
1354
1355         if (!mm_init(mm, tsk, mm->user_ns))
1356                 goto fail_nomem;
1357
1358         err = dup_mmap(mm, oldmm);
1359         if (err)
1360                 goto free_pt;
1361
1362         mm->hiwater_rss = get_mm_rss(mm);
1363         mm->hiwater_vm = mm->total_vm;
1364
1365         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1366                 goto free_pt;
1367
1368         return mm;
1369
1370 free_pt:
1371         /* don't put binfmt in mmput, we haven't got module yet */
1372         mm->binfmt = NULL;
1373         mm_init_owner(mm, NULL);
1374         mmput(mm);
1375
1376 fail_nomem:
1377         return NULL;
1378 }
1379
1380 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1381 {
1382         struct mm_struct *mm, *oldmm;
1383         int retval;
1384
1385         tsk->min_flt = tsk->maj_flt = 0;
1386         tsk->nvcsw = tsk->nivcsw = 0;
1387 #ifdef CONFIG_DETECT_HUNG_TASK
1388         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1389         tsk->last_switch_time = 0;
1390 #endif
1391
1392         tsk->mm = NULL;
1393         tsk->active_mm = NULL;
1394
1395         /*
1396          * Are we cloning a kernel thread?
1397          *
1398          * We need to steal a active VM for that..
1399          */
1400         oldmm = current->mm;
1401         if (!oldmm)
1402                 return 0;
1403
1404         /* initialize the new vmacache entries */
1405         vmacache_flush(tsk);
1406
1407         if (clone_flags & CLONE_VM) {
1408                 mmget(oldmm);
1409                 mm = oldmm;
1410                 goto good_mm;
1411         }
1412
1413         retval = -ENOMEM;
1414         mm = dup_mm(tsk, current->mm);
1415         if (!mm)
1416                 goto fail_nomem;
1417
1418 good_mm:
1419         tsk->mm = mm;
1420         tsk->active_mm = mm;
1421         return 0;
1422
1423 fail_nomem:
1424         return retval;
1425 }
1426
1427 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1428 {
1429         struct fs_struct *fs = current->fs;
1430         if (clone_flags & CLONE_FS) {
1431                 /* tsk->fs is already what we want */
1432                 spin_lock(&fs->lock);
1433                 if (fs->in_exec) {
1434                         spin_unlock(&fs->lock);
1435                         return -EAGAIN;
1436                 }
1437                 fs->users++;
1438                 spin_unlock(&fs->lock);
1439                 return 0;
1440         }
1441         tsk->fs = copy_fs_struct(fs);
1442         if (!tsk->fs)
1443                 return -ENOMEM;
1444         return 0;
1445 }
1446
1447 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1448 {
1449         struct files_struct *oldf, *newf;
1450         int error = 0;
1451
1452         /*
1453          * A background process may not have any files ...
1454          */
1455         oldf = current->files;
1456         if (!oldf)
1457                 goto out;
1458
1459         if (clone_flags & CLONE_FILES) {
1460                 atomic_inc(&oldf->count);
1461                 goto out;
1462         }
1463
1464         newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1465         if (!newf)
1466                 goto out;
1467
1468         tsk->files = newf;
1469         error = 0;
1470 out:
1471         return error;
1472 }
1473
1474 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1475 {
1476 #ifdef CONFIG_BLOCK
1477         struct io_context *ioc = current->io_context;
1478         struct io_context *new_ioc;
1479
1480         if (!ioc)
1481                 return 0;
1482         /*
1483          * Share io context with parent, if CLONE_IO is set
1484          */
1485         if (clone_flags & CLONE_IO) {
1486                 ioc_task_link(ioc);
1487                 tsk->io_context = ioc;
1488         } else if (ioprio_valid(ioc->ioprio)) {
1489                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1490                 if (unlikely(!new_ioc))
1491                         return -ENOMEM;
1492
1493                 new_ioc->ioprio = ioc->ioprio;
1494                 put_io_context(new_ioc);
1495         }
1496 #endif
1497         return 0;
1498 }
1499
1500 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1501 {
1502         struct sighand_struct *sig;
1503
1504         if (clone_flags & CLONE_SIGHAND) {
1505                 refcount_inc(&current->sighand->count);
1506                 return 0;
1507         }
1508         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1509         RCU_INIT_POINTER(tsk->sighand, sig);
1510         if (!sig)
1511                 return -ENOMEM;
1512
1513         refcount_set(&sig->count, 1);
1514         spin_lock_irq(&current->sighand->siglock);
1515         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1516         spin_unlock_irq(&current->sighand->siglock);
1517
1518         /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1519         if (clone_flags & CLONE_CLEAR_SIGHAND)
1520                 flush_signal_handlers(tsk, 0);
1521
1522         return 0;
1523 }
1524
1525 void __cleanup_sighand(struct sighand_struct *sighand)
1526 {
1527         if (refcount_dec_and_test(&sighand->count)) {
1528                 signalfd_cleanup(sighand);
1529                 /*
1530                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1531                  * without an RCU grace period, see __lock_task_sighand().
1532                  */
1533                 kmem_cache_free(sighand_cachep, sighand);
1534         }
1535 }
1536
1537 /*
1538  * Initialize POSIX timer handling for a thread group.
1539  */
1540 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1541 {
1542         struct posix_cputimers *pct = &sig->posix_cputimers;
1543         unsigned long cpu_limit;
1544
1545         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1546         posix_cputimers_group_init(pct, cpu_limit);
1547 }
1548
1549 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1550 {
1551         struct signal_struct *sig;
1552
1553         if (clone_flags & CLONE_THREAD)
1554                 return 0;
1555
1556         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1557         tsk->signal = sig;
1558         if (!sig)
1559                 return -ENOMEM;
1560
1561         sig->nr_threads = 1;
1562         atomic_set(&sig->live, 1);
1563         refcount_set(&sig->sigcnt, 1);
1564
1565         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1566         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1567         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1568
1569         init_waitqueue_head(&sig->wait_chldexit);
1570         sig->curr_target = tsk;
1571         init_sigpending(&sig->shared_pending);
1572         INIT_HLIST_HEAD(&sig->multiprocess);
1573         seqlock_init(&sig->stats_lock);
1574         prev_cputime_init(&sig->prev_cputime);
1575
1576 #ifdef CONFIG_POSIX_TIMERS
1577         INIT_LIST_HEAD(&sig->posix_timers);
1578         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1579         sig->real_timer.function = it_real_fn;
1580 #endif
1581
1582         task_lock(current->group_leader);
1583         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1584         task_unlock(current->group_leader);
1585
1586         posix_cpu_timers_init_group(sig);
1587
1588         tty_audit_fork(sig);
1589         sched_autogroup_fork(sig);
1590
1591         sig->oom_score_adj = current->signal->oom_score_adj;
1592         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1593
1594         mutex_init(&sig->cred_guard_mutex);
1595         mutex_init(&sig->exec_update_mutex);
1596
1597         return 0;
1598 }
1599
1600 static void copy_seccomp(struct task_struct *p)
1601 {
1602 #ifdef CONFIG_SECCOMP
1603         /*
1604          * Must be called with sighand->lock held, which is common to
1605          * all threads in the group. Holding cred_guard_mutex is not
1606          * needed because this new task is not yet running and cannot
1607          * be racing exec.
1608          */
1609         assert_spin_locked(&current->sighand->siglock);
1610
1611         /* Ref-count the new filter user, and assign it. */
1612         get_seccomp_filter(current);
1613         p->seccomp = current->seccomp;
1614
1615         /*
1616          * Explicitly enable no_new_privs here in case it got set
1617          * between the task_struct being duplicated and holding the
1618          * sighand lock. The seccomp state and nnp must be in sync.
1619          */
1620         if (task_no_new_privs(current))
1621                 task_set_no_new_privs(p);
1622
1623         /*
1624          * If the parent gained a seccomp mode after copying thread
1625          * flags and between before we held the sighand lock, we have
1626          * to manually enable the seccomp thread flag here.
1627          */
1628         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1629                 set_task_syscall_work(p, SECCOMP);
1630 #endif
1631 }
1632
1633 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1634 {
1635         current->clear_child_tid = tidptr;
1636
1637         return task_pid_vnr(current);
1638 }
1639
1640 static void rt_mutex_init_task(struct task_struct *p)
1641 {
1642         raw_spin_lock_init(&p->pi_lock);
1643 #ifdef CONFIG_RT_MUTEXES
1644         p->pi_waiters = RB_ROOT_CACHED;
1645         p->pi_top_task = NULL;
1646         p->pi_blocked_on = NULL;
1647 #endif
1648 }
1649
1650 static inline void init_task_pid_links(struct task_struct *task)
1651 {
1652         enum pid_type type;
1653
1654         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1655                 INIT_HLIST_NODE(&task->pid_links[type]);
1656         }
1657 }
1658
1659 static inline void
1660 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1661 {
1662         if (type == PIDTYPE_PID)
1663                 task->thread_pid = pid;
1664         else
1665                 task->signal->pids[type] = pid;
1666 }
1667
1668 static inline void rcu_copy_process(struct task_struct *p)
1669 {
1670 #ifdef CONFIG_PREEMPT_RCU
1671         p->rcu_read_lock_nesting = 0;
1672         p->rcu_read_unlock_special.s = 0;
1673         p->rcu_blocked_node = NULL;
1674         INIT_LIST_HEAD(&p->rcu_node_entry);
1675 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1676 #ifdef CONFIG_TASKS_RCU
1677         p->rcu_tasks_holdout = false;
1678         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1679         p->rcu_tasks_idle_cpu = -1;
1680 #endif /* #ifdef CONFIG_TASKS_RCU */
1681 #ifdef CONFIG_TASKS_TRACE_RCU
1682         p->trc_reader_nesting = 0;
1683         p->trc_reader_special.s = 0;
1684         INIT_LIST_HEAD(&p->trc_holdout_list);
1685 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1686 }
1687
1688 struct pid *pidfd_pid(const struct file *file)
1689 {
1690         if (file->f_op == &pidfd_fops)
1691                 return file->private_data;
1692
1693         return ERR_PTR(-EBADF);
1694 }
1695
1696 static int pidfd_release(struct inode *inode, struct file *file)
1697 {
1698         struct pid *pid = file->private_data;
1699
1700         file->private_data = NULL;
1701         put_pid(pid);
1702         return 0;
1703 }
1704
1705 #ifdef CONFIG_PROC_FS
1706 /**
1707  * pidfd_show_fdinfo - print information about a pidfd
1708  * @m: proc fdinfo file
1709  * @f: file referencing a pidfd
1710  *
1711  * Pid:
1712  * This function will print the pid that a given pidfd refers to in the
1713  * pid namespace of the procfs instance.
1714  * If the pid namespace of the process is not a descendant of the pid
1715  * namespace of the procfs instance 0 will be shown as its pid. This is
1716  * similar to calling getppid() on a process whose parent is outside of
1717  * its pid namespace.
1718  *
1719  * NSpid:
1720  * If pid namespaces are supported then this function will also print
1721  * the pid of a given pidfd refers to for all descendant pid namespaces
1722  * starting from the current pid namespace of the instance, i.e. the
1723  * Pid field and the first entry in the NSpid field will be identical.
1724  * If the pid namespace of the process is not a descendant of the pid
1725  * namespace of the procfs instance 0 will be shown as its first NSpid
1726  * entry and no others will be shown.
1727  * Note that this differs from the Pid and NSpid fields in
1728  * /proc/<pid>/status where Pid and NSpid are always shown relative to
1729  * the  pid namespace of the procfs instance. The difference becomes
1730  * obvious when sending around a pidfd between pid namespaces from a
1731  * different branch of the tree, i.e. where no ancestoral relation is
1732  * present between the pid namespaces:
1733  * - create two new pid namespaces ns1 and ns2 in the initial pid
1734  *   namespace (also take care to create new mount namespaces in the
1735  *   new pid namespace and mount procfs)
1736  * - create a process with a pidfd in ns1
1737  * - send pidfd from ns1 to ns2
1738  * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1739  *   have exactly one entry, which is 0
1740  */
1741 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1742 {
1743         struct pid *pid = f->private_data;
1744         struct pid_namespace *ns;
1745         pid_t nr = -1;
1746
1747         if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1748                 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1749                 nr = pid_nr_ns(pid, ns);
1750         }
1751
1752         seq_put_decimal_ll(m, "Pid:\t", nr);
1753
1754 #ifdef CONFIG_PID_NS
1755         seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1756         if (nr > 0) {
1757                 int i;
1758
1759                 /* If nr is non-zero it means that 'pid' is valid and that
1760                  * ns, i.e. the pid namespace associated with the procfs
1761                  * instance, is in the pid namespace hierarchy of pid.
1762                  * Start at one below the already printed level.
1763                  */
1764                 for (i = ns->level + 1; i <= pid->level; i++)
1765                         seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1766         }
1767 #endif
1768         seq_putc(m, '\n');
1769 }
1770 #endif
1771
1772 /*
1773  * Poll support for process exit notification.
1774  */
1775 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1776 {
1777         struct pid *pid = file->private_data;
1778         __poll_t poll_flags = 0;
1779
1780         poll_wait(file, &pid->wait_pidfd, pts);
1781
1782         /*
1783          * Inform pollers only when the whole thread group exits.
1784          * If the thread group leader exits before all other threads in the
1785          * group, then poll(2) should block, similar to the wait(2) family.
1786          */
1787         if (thread_group_exited(pid))
1788                 poll_flags = EPOLLIN | EPOLLRDNORM;
1789
1790         return poll_flags;
1791 }
1792
1793 const struct file_operations pidfd_fops = {
1794         .release = pidfd_release,
1795         .poll = pidfd_poll,
1796 #ifdef CONFIG_PROC_FS
1797         .show_fdinfo = pidfd_show_fdinfo,
1798 #endif
1799 };
1800
1801 static void __delayed_free_task(struct rcu_head *rhp)
1802 {
1803         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1804
1805         free_task(tsk);
1806 }
1807
1808 static __always_inline void delayed_free_task(struct task_struct *tsk)
1809 {
1810         if (IS_ENABLED(CONFIG_MEMCG))
1811                 call_rcu(&tsk->rcu, __delayed_free_task);
1812         else
1813                 free_task(tsk);
1814 }
1815
1816 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1817 {
1818         /* Skip if kernel thread */
1819         if (!tsk->mm)
1820                 return;
1821
1822         /* Skip if spawning a thread or using vfork */
1823         if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1824                 return;
1825
1826         /* We need to synchronize with __set_oom_adj */
1827         mutex_lock(&oom_adj_mutex);
1828         set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1829         /* Update the values in case they were changed after copy_signal */
1830         tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1831         tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1832         mutex_unlock(&oom_adj_mutex);
1833 }
1834
1835 /*
1836  * This creates a new process as a copy of the old one,
1837  * but does not actually start it yet.
1838  *
1839  * It copies the registers, and all the appropriate
1840  * parts of the process environment (as per the clone
1841  * flags). The actual kick-off is left to the caller.
1842  */
1843 static __latent_entropy struct task_struct *copy_process(
1844                                         struct pid *pid,
1845                                         int trace,
1846                                         int node,
1847                                         struct kernel_clone_args *args)
1848 {
1849         int pidfd = -1, retval;
1850         struct task_struct *p;
1851         struct multiprocess_signals delayed;
1852         struct file *pidfile = NULL;
1853         u64 clone_flags = args->flags;
1854         struct nsproxy *nsp = current->nsproxy;
1855
1856         /*
1857          * Don't allow sharing the root directory with processes in a different
1858          * namespace
1859          */
1860         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1861                 return ERR_PTR(-EINVAL);
1862
1863         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1864                 return ERR_PTR(-EINVAL);
1865
1866         /*
1867          * Thread groups must share signals as well, and detached threads
1868          * can only be started up within the thread group.
1869          */
1870         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1871                 return ERR_PTR(-EINVAL);
1872
1873         /*
1874          * Shared signal handlers imply shared VM. By way of the above,
1875          * thread groups also imply shared VM. Blocking this case allows
1876          * for various simplifications in other code.
1877          */
1878         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1879                 return ERR_PTR(-EINVAL);
1880
1881         /*
1882          * Siblings of global init remain as zombies on exit since they are
1883          * not reaped by their parent (swapper). To solve this and to avoid
1884          * multi-rooted process trees, prevent global and container-inits
1885          * from creating siblings.
1886          */
1887         if ((clone_flags & CLONE_PARENT) &&
1888                                 current->signal->flags & SIGNAL_UNKILLABLE)
1889                 return ERR_PTR(-EINVAL);
1890
1891         /*
1892          * If the new process will be in a different pid or user namespace
1893          * do not allow it to share a thread group with the forking task.
1894          */
1895         if (clone_flags & CLONE_THREAD) {
1896                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1897                     (task_active_pid_ns(current) != nsp->pid_ns_for_children))
1898                         return ERR_PTR(-EINVAL);
1899         }
1900
1901         /*
1902          * If the new process will be in a different time namespace
1903          * do not allow it to share VM or a thread group with the forking task.
1904          */
1905         if (clone_flags & (CLONE_THREAD | CLONE_VM)) {
1906                 if (nsp->time_ns != nsp->time_ns_for_children)
1907                         return ERR_PTR(-EINVAL);
1908         }
1909
1910         if (clone_flags & CLONE_PIDFD) {
1911                 /*
1912                  * - CLONE_DETACHED is blocked so that we can potentially
1913                  *   reuse it later for CLONE_PIDFD.
1914                  * - CLONE_THREAD is blocked until someone really needs it.
1915                  */
1916                 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1917                         return ERR_PTR(-EINVAL);
1918         }
1919
1920         /*
1921          * Force any signals received before this point to be delivered
1922          * before the fork happens.  Collect up signals sent to multiple
1923          * processes that happen during the fork and delay them so that
1924          * they appear to happen after the fork.
1925          */
1926         sigemptyset(&delayed.signal);
1927         INIT_HLIST_NODE(&delayed.node);
1928
1929         spin_lock_irq(&current->sighand->siglock);
1930         if (!(clone_flags & CLONE_THREAD))
1931                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1932         recalc_sigpending();
1933         spin_unlock_irq(&current->sighand->siglock);
1934         retval = -ERESTARTNOINTR;
1935         if (signal_pending(current))
1936                 goto fork_out;
1937
1938         retval = -ENOMEM;
1939         p = dup_task_struct(current, node);
1940         if (!p)
1941                 goto fork_out;
1942
1943         /*
1944          * This _must_ happen before we call free_task(), i.e. before we jump
1945          * to any of the bad_fork_* labels. This is to avoid freeing
1946          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1947          * kernel threads (PF_KTHREAD).
1948          */
1949         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1950         /*
1951          * Clear TID on mm_release()?
1952          */
1953         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1954
1955         ftrace_graph_init_task(p);
1956
1957         rt_mutex_init_task(p);
1958
1959         lockdep_assert_irqs_enabled();
1960 #ifdef CONFIG_PROVE_LOCKING
1961         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1962 #endif
1963         retval = -EAGAIN;
1964         if (atomic_read(&p->real_cred->user->processes) >=
1965                         task_rlimit(p, RLIMIT_NPROC)) {
1966                 if (p->real_cred->user != INIT_USER &&
1967                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1968                         goto bad_fork_free;
1969         }
1970         current->flags &= ~PF_NPROC_EXCEEDED;
1971
1972         retval = copy_creds(p, clone_flags);
1973         if (retval < 0)
1974                 goto bad_fork_free;
1975
1976         /*
1977          * If multiple threads are within copy_process(), then this check
1978          * triggers too late. This doesn't hurt, the check is only there
1979          * to stop root fork bombs.
1980          */
1981         retval = -EAGAIN;
1982         if (data_race(nr_threads >= max_threads))
1983                 goto bad_fork_cleanup_count;
1984
1985         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1986         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1987         p->flags |= PF_FORKNOEXEC;
1988         INIT_LIST_HEAD(&p->children);
1989         INIT_LIST_HEAD(&p->sibling);
1990         rcu_copy_process(p);
1991         p->vfork_done = NULL;
1992         spin_lock_init(&p->alloc_lock);
1993
1994         init_sigpending(&p->pending);
1995
1996         p->utime = p->stime = p->gtime = 0;
1997 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1998         p->utimescaled = p->stimescaled = 0;
1999 #endif
2000         prev_cputime_init(&p->prev_cputime);
2001
2002 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2003         seqcount_init(&p->vtime.seqcount);
2004         p->vtime.starttime = 0;
2005         p->vtime.state = VTIME_INACTIVE;
2006 #endif
2007
2008 #ifdef CONFIG_IO_URING
2009         p->io_uring = NULL;
2010 #endif
2011
2012 #if defined(SPLIT_RSS_COUNTING)
2013         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2014 #endif
2015
2016         p->default_timer_slack_ns = current->timer_slack_ns;
2017
2018 #ifdef CONFIG_PSI
2019         p->psi_flags = 0;
2020 #endif
2021
2022         task_io_accounting_init(&p->ioac);
2023         acct_clear_integrals(p);
2024
2025         posix_cputimers_init(&p->posix_cputimers);
2026
2027         p->io_context = NULL;
2028         audit_set_context(p, NULL);
2029         cgroup_fork(p);
2030 #ifdef CONFIG_NUMA
2031         p->mempolicy = mpol_dup(p->mempolicy);
2032         if (IS_ERR(p->mempolicy)) {
2033                 retval = PTR_ERR(p->mempolicy);
2034                 p->mempolicy = NULL;
2035                 goto bad_fork_cleanup_threadgroup_lock;
2036         }
2037 #endif
2038 #ifdef CONFIG_CPUSETS
2039         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2040         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2041         seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2042 #endif
2043 #ifdef CONFIG_TRACE_IRQFLAGS
2044         memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2045         p->irqtrace.hardirq_disable_ip  = _THIS_IP_;
2046         p->irqtrace.softirq_enable_ip   = _THIS_IP_;
2047         p->softirqs_enabled             = 1;
2048         p->softirq_context              = 0;
2049 #endif
2050
2051         p->pagefault_disabled = 0;
2052
2053 #ifdef CONFIG_LOCKDEP
2054         lockdep_init_task(p);
2055 #endif
2056
2057 #ifdef CONFIG_DEBUG_MUTEXES
2058         p->blocked_on = NULL; /* not blocked yet */
2059 #endif
2060 #ifdef CONFIG_BCACHE
2061         p->sequential_io        = 0;
2062         p->sequential_io_avg    = 0;
2063 #endif
2064
2065         /* Perform scheduler related setup. Assign this task to a CPU. */
2066         retval = sched_fork(clone_flags, p);
2067         if (retval)
2068                 goto bad_fork_cleanup_policy;
2069
2070         retval = perf_event_init_task(p);
2071         if (retval)
2072                 goto bad_fork_cleanup_policy;
2073         retval = audit_alloc(p);
2074         if (retval)
2075                 goto bad_fork_cleanup_perf;
2076         /* copy all the process information */
2077         shm_init_task(p);
2078         retval = security_task_alloc(p, clone_flags);
2079         if (retval)
2080                 goto bad_fork_cleanup_audit;
2081         retval = copy_semundo(clone_flags, p);
2082         if (retval)
2083                 goto bad_fork_cleanup_security;
2084         retval = copy_files(clone_flags, p);
2085         if (retval)
2086                 goto bad_fork_cleanup_semundo;
2087         retval = copy_fs(clone_flags, p);
2088         if (retval)
2089                 goto bad_fork_cleanup_files;
2090         retval = copy_sighand(clone_flags, p);
2091         if (retval)
2092                 goto bad_fork_cleanup_fs;
2093         retval = copy_signal(clone_flags, p);
2094         if (retval)
2095                 goto bad_fork_cleanup_sighand;
2096         retval = copy_mm(clone_flags, p);
2097         if (retval)
2098                 goto bad_fork_cleanup_signal;
2099         retval = copy_namespaces(clone_flags, p);
2100         if (retval)
2101                 goto bad_fork_cleanup_mm;
2102         retval = copy_io(clone_flags, p);
2103         if (retval)
2104                 goto bad_fork_cleanup_namespaces;
2105         retval = copy_thread(clone_flags, args->stack, args->stack_size, p, args->tls);
2106         if (retval)
2107                 goto bad_fork_cleanup_io;
2108
2109         stackleak_task_init(p);
2110
2111         if (pid != &init_struct_pid) {
2112                 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2113                                 args->set_tid_size);
2114                 if (IS_ERR(pid)) {
2115                         retval = PTR_ERR(pid);
2116                         goto bad_fork_cleanup_thread;
2117                 }
2118         }
2119
2120         /*
2121          * This has to happen after we've potentially unshared the file
2122          * descriptor table (so that the pidfd doesn't leak into the child
2123          * if the fd table isn't shared).
2124          */
2125         if (clone_flags & CLONE_PIDFD) {
2126                 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2127                 if (retval < 0)
2128                         goto bad_fork_free_pid;
2129
2130                 pidfd = retval;
2131
2132                 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2133                                               O_RDWR | O_CLOEXEC);
2134                 if (IS_ERR(pidfile)) {
2135                         put_unused_fd(pidfd);
2136                         retval = PTR_ERR(pidfile);
2137                         goto bad_fork_free_pid;
2138                 }
2139                 get_pid(pid);   /* held by pidfile now */
2140
2141                 retval = put_user(pidfd, args->pidfd);
2142                 if (retval)
2143                         goto bad_fork_put_pidfd;
2144         }
2145
2146 #ifdef CONFIG_BLOCK
2147         p->plug = NULL;
2148 #endif
2149         futex_init_task(p);
2150
2151         /*
2152          * sigaltstack should be cleared when sharing the same VM
2153          */
2154         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2155                 sas_ss_reset(p);
2156
2157         /*
2158          * Syscall tracing and stepping should be turned off in the
2159          * child regardless of CLONE_PTRACE.
2160          */
2161         user_disable_single_step(p);
2162         clear_task_syscall_work(p, SYSCALL_TRACE);
2163 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2164         clear_task_syscall_work(p, SYSCALL_EMU);
2165 #endif
2166         clear_tsk_latency_tracing(p);
2167
2168         /* ok, now we should be set up.. */
2169         p->pid = pid_nr(pid);
2170         if (clone_flags & CLONE_THREAD) {
2171                 p->group_leader = current->group_leader;
2172                 p->tgid = current->tgid;
2173         } else {
2174                 p->group_leader = p;
2175                 p->tgid = p->pid;
2176         }
2177
2178         p->nr_dirtied = 0;
2179         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2180         p->dirty_paused_when = 0;
2181
2182         p->pdeath_signal = 0;
2183         INIT_LIST_HEAD(&p->thread_group);
2184         p->task_works = NULL;
2185
2186         /*
2187          * Ensure that the cgroup subsystem policies allow the new process to be
2188          * forked. It should be noted that the new process's css_set can be changed
2189          * between here and cgroup_post_fork() if an organisation operation is in
2190          * progress.
2191          */
2192         retval = cgroup_can_fork(p, args);
2193         if (retval)
2194                 goto bad_fork_put_pidfd;
2195
2196         /*
2197          * From this point on we must avoid any synchronous user-space
2198          * communication until we take the tasklist-lock. In particular, we do
2199          * not want user-space to be able to predict the process start-time by
2200          * stalling fork(2) after we recorded the start_time but before it is
2201          * visible to the system.
2202          */
2203
2204         p->start_time = ktime_get_ns();
2205         p->start_boottime = ktime_get_boottime_ns();
2206
2207         /*
2208          * Make it visible to the rest of the system, but dont wake it up yet.
2209          * Need tasklist lock for parent etc handling!
2210          */
2211         write_lock_irq(&tasklist_lock);
2212
2213         /* CLONE_PARENT re-uses the old parent */
2214         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2215                 p->real_parent = current->real_parent;
2216                 p->parent_exec_id = current->parent_exec_id;
2217                 if (clone_flags & CLONE_THREAD)
2218                         p->exit_signal = -1;
2219                 else
2220                         p->exit_signal = current->group_leader->exit_signal;
2221         } else {
2222                 p->real_parent = current;
2223                 p->parent_exec_id = current->self_exec_id;
2224                 p->exit_signal = args->exit_signal;
2225         }
2226
2227         klp_copy_process(p);
2228
2229         spin_lock(&current->sighand->siglock);
2230
2231         /*
2232          * Copy seccomp details explicitly here, in case they were changed
2233          * before holding sighand lock.
2234          */
2235         copy_seccomp(p);
2236
2237         rseq_fork(p, clone_flags);
2238
2239         /* Don't start children in a dying pid namespace */
2240         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2241                 retval = -ENOMEM;
2242                 goto bad_fork_cancel_cgroup;
2243         }
2244
2245         /* Let kill terminate clone/fork in the middle */
2246         if (fatal_signal_pending(current)) {
2247                 retval = -EINTR;
2248                 goto bad_fork_cancel_cgroup;
2249         }
2250
2251         /* past the last point of failure */
2252         if (pidfile)
2253                 fd_install(pidfd, pidfile);
2254
2255         init_task_pid_links(p);
2256         if (likely(p->pid)) {
2257                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2258
2259                 init_task_pid(p, PIDTYPE_PID, pid);
2260                 if (thread_group_leader(p)) {
2261                         init_task_pid(p, PIDTYPE_TGID, pid);
2262                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2263                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2264
2265                         if (is_child_reaper(pid)) {
2266                                 ns_of_pid(pid)->child_reaper = p;
2267                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2268                         }
2269                         p->signal->shared_pending.signal = delayed.signal;
2270                         p->signal->tty = tty_kref_get(current->signal->tty);
2271                         /*
2272                          * Inherit has_child_subreaper flag under the same
2273                          * tasklist_lock with adding child to the process tree
2274                          * for propagate_has_child_subreaper optimization.
2275                          */
2276                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2277                                                          p->real_parent->signal->is_child_subreaper;
2278                         list_add_tail(&p->sibling, &p->real_parent->children);
2279                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2280                         attach_pid(p, PIDTYPE_TGID);
2281                         attach_pid(p, PIDTYPE_PGID);
2282                         attach_pid(p, PIDTYPE_SID);
2283                         __this_cpu_inc(process_counts);
2284                 } else {
2285                         current->signal->nr_threads++;
2286                         atomic_inc(&current->signal->live);
2287                         refcount_inc(&current->signal->sigcnt);
2288                         task_join_group_stop(p);
2289                         list_add_tail_rcu(&p->thread_group,
2290                                           &p->group_leader->thread_group);
2291                         list_add_tail_rcu(&p->thread_node,
2292                                           &p->signal->thread_head);
2293                 }
2294                 attach_pid(p, PIDTYPE_PID);
2295                 nr_threads++;
2296         }
2297         total_forks++;
2298         hlist_del_init(&delayed.node);
2299         spin_unlock(&current->sighand->siglock);
2300         syscall_tracepoint_update(p);
2301         write_unlock_irq(&tasklist_lock);
2302
2303         proc_fork_connector(p);
2304         sched_post_fork(p);
2305         cgroup_post_fork(p, args);
2306         perf_event_fork(p);
2307
2308         trace_task_newtask(p, clone_flags);
2309         uprobe_copy_process(p, clone_flags);
2310
2311         copy_oom_score_adj(clone_flags, p);
2312
2313         return p;
2314
2315 bad_fork_cancel_cgroup:
2316         spin_unlock(&current->sighand->siglock);
2317         write_unlock_irq(&tasklist_lock);
2318         cgroup_cancel_fork(p, args);
2319 bad_fork_put_pidfd:
2320         if (clone_flags & CLONE_PIDFD) {
2321                 fput(pidfile);
2322                 put_unused_fd(pidfd);
2323         }
2324 bad_fork_free_pid:
2325         if (pid != &init_struct_pid)
2326                 free_pid(pid);
2327 bad_fork_cleanup_thread:
2328         exit_thread(p);
2329 bad_fork_cleanup_io:
2330         if (p->io_context)
2331                 exit_io_context(p);
2332 bad_fork_cleanup_namespaces:
2333         exit_task_namespaces(p);
2334 bad_fork_cleanup_mm:
2335         if (p->mm) {
2336                 mm_clear_owner(p->mm, p);
2337                 mmput(p->mm);
2338         }
2339 bad_fork_cleanup_signal:
2340         if (!(clone_flags & CLONE_THREAD))
2341                 free_signal_struct(p->signal);
2342 bad_fork_cleanup_sighand:
2343         __cleanup_sighand(p->sighand);
2344 bad_fork_cleanup_fs:
2345         exit_fs(p); /* blocking */
2346 bad_fork_cleanup_files:
2347         exit_files(p); /* blocking */
2348 bad_fork_cleanup_semundo:
2349         exit_sem(p);
2350 bad_fork_cleanup_security:
2351         security_task_free(p);
2352 bad_fork_cleanup_audit:
2353         audit_free(p);
2354 bad_fork_cleanup_perf:
2355         perf_event_free_task(p);
2356 bad_fork_cleanup_policy:
2357         lockdep_free_task(p);
2358 #ifdef CONFIG_NUMA
2359         mpol_put(p->mempolicy);
2360 bad_fork_cleanup_threadgroup_lock:
2361 #endif
2362         delayacct_tsk_free(p);
2363 bad_fork_cleanup_count:
2364         atomic_dec(&p->cred->user->processes);
2365         exit_creds(p);
2366 bad_fork_free:
2367         p->state = TASK_DEAD;
2368         put_task_stack(p);
2369         delayed_free_task(p);
2370 fork_out:
2371         spin_lock_irq(&current->sighand->siglock);
2372         hlist_del_init(&delayed.node);
2373         spin_unlock_irq(&current->sighand->siglock);
2374         return ERR_PTR(retval);
2375 }
2376
2377 static inline void init_idle_pids(struct task_struct *idle)
2378 {
2379         enum pid_type type;
2380
2381         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2382                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2383                 init_task_pid(idle, type, &init_struct_pid);
2384         }
2385 }
2386
2387 struct task_struct *fork_idle(int cpu)
2388 {
2389         struct task_struct *task;
2390         struct kernel_clone_args args = {
2391                 .flags = CLONE_VM,
2392         };
2393
2394         task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2395         if (!IS_ERR(task)) {
2396                 init_idle_pids(task);
2397                 init_idle(task, cpu);
2398         }
2399
2400         return task;
2401 }
2402
2403 struct mm_struct *copy_init_mm(void)
2404 {
2405         return dup_mm(NULL, &init_mm);
2406 }
2407
2408 /*
2409  *  Ok, this is the main fork-routine.
2410  *
2411  * It copies the process, and if successful kick-starts
2412  * it and waits for it to finish using the VM if required.
2413  *
2414  * args->exit_signal is expected to be checked for sanity by the caller.
2415  */
2416 pid_t kernel_clone(struct kernel_clone_args *args)
2417 {
2418         u64 clone_flags = args->flags;
2419         struct completion vfork;
2420         struct pid *pid;
2421         struct task_struct *p;
2422         int trace = 0;
2423         pid_t nr;
2424
2425         /*
2426          * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2427          * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2428          * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2429          * field in struct clone_args and it still doesn't make sense to have
2430          * them both point at the same memory location. Performing this check
2431          * here has the advantage that we don't need to have a separate helper
2432          * to check for legacy clone().
2433          */
2434         if ((args->flags & CLONE_PIDFD) &&
2435             (args->flags & CLONE_PARENT_SETTID) &&
2436             (args->pidfd == args->parent_tid))
2437                 return -EINVAL;
2438
2439         /*
2440          * Determine whether and which event to report to ptracer.  When
2441          * called from kernel_thread or CLONE_UNTRACED is explicitly
2442          * requested, no event is reported; otherwise, report if the event
2443          * for the type of forking is enabled.
2444          */
2445         if (!(clone_flags & CLONE_UNTRACED)) {
2446                 if (clone_flags & CLONE_VFORK)
2447                         trace = PTRACE_EVENT_VFORK;
2448                 else if (args->exit_signal != SIGCHLD)
2449                         trace = PTRACE_EVENT_CLONE;
2450                 else
2451                         trace = PTRACE_EVENT_FORK;
2452
2453                 if (likely(!ptrace_event_enabled(current, trace)))
2454                         trace = 0;
2455         }
2456
2457         p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2458         add_latent_entropy();
2459
2460         if (IS_ERR(p))
2461                 return PTR_ERR(p);
2462
2463         /*
2464          * Do this prior waking up the new thread - the thread pointer
2465          * might get invalid after that point, if the thread exits quickly.
2466          */
2467         trace_sched_process_fork(current, p);
2468
2469         pid = get_task_pid(p, PIDTYPE_PID);
2470         nr = pid_vnr(pid);
2471
2472         if (clone_flags & CLONE_PARENT_SETTID)
2473                 put_user(nr, args->parent_tid);
2474
2475         if (clone_flags & CLONE_VFORK) {
2476                 p->vfork_done = &vfork;
2477                 init_completion(&vfork);
2478                 get_task_struct(p);
2479         }
2480
2481         wake_up_new_task(p);
2482
2483         /* forking complete and child started to run, tell ptracer */
2484         if (unlikely(trace))
2485                 ptrace_event_pid(trace, pid);
2486
2487         if (clone_flags & CLONE_VFORK) {
2488                 if (!wait_for_vfork_done(p, &vfork))
2489                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2490         }
2491
2492         put_pid(pid);
2493         return nr;
2494 }
2495
2496 /*
2497  * Create a kernel thread.
2498  */
2499 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2500 {
2501         struct kernel_clone_args args = {
2502                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2503                                     CLONE_UNTRACED) & ~CSIGNAL),
2504                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2505                 .stack          = (unsigned long)fn,
2506                 .stack_size     = (unsigned long)arg,
2507         };
2508
2509         return kernel_clone(&args);
2510 }
2511
2512 #ifdef __ARCH_WANT_SYS_FORK
2513 SYSCALL_DEFINE0(fork)
2514 {
2515 #ifdef CONFIG_MMU
2516         struct kernel_clone_args args = {
2517                 .exit_signal = SIGCHLD,
2518         };
2519
2520         return kernel_clone(&args);
2521 #else
2522         /* can not support in nommu mode */
2523         return -EINVAL;
2524 #endif
2525 }
2526 #endif
2527
2528 #ifdef __ARCH_WANT_SYS_VFORK
2529 SYSCALL_DEFINE0(vfork)
2530 {
2531         struct kernel_clone_args args = {
2532                 .flags          = CLONE_VFORK | CLONE_VM,
2533                 .exit_signal    = SIGCHLD,
2534         };
2535
2536         return kernel_clone(&args);
2537 }
2538 #endif
2539
2540 #ifdef __ARCH_WANT_SYS_CLONE
2541 #ifdef CONFIG_CLONE_BACKWARDS
2542 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2543                  int __user *, parent_tidptr,
2544                  unsigned long, tls,
2545                  int __user *, child_tidptr)
2546 #elif defined(CONFIG_CLONE_BACKWARDS2)
2547 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2548                  int __user *, parent_tidptr,
2549                  int __user *, child_tidptr,
2550                  unsigned long, tls)
2551 #elif defined(CONFIG_CLONE_BACKWARDS3)
2552 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2553                 int, stack_size,
2554                 int __user *, parent_tidptr,
2555                 int __user *, child_tidptr,
2556                 unsigned long, tls)
2557 #else
2558 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2559                  int __user *, parent_tidptr,
2560                  int __user *, child_tidptr,
2561                  unsigned long, tls)
2562 #endif
2563 {
2564         struct kernel_clone_args args = {
2565                 .flags          = (lower_32_bits(clone_flags) & ~CSIGNAL),
2566                 .pidfd          = parent_tidptr,
2567                 .child_tid      = child_tidptr,
2568                 .parent_tid     = parent_tidptr,
2569                 .exit_signal    = (lower_32_bits(clone_flags) & CSIGNAL),
2570                 .stack          = newsp,
2571                 .tls            = tls,
2572         };
2573
2574         return kernel_clone(&args);
2575 }
2576 #endif
2577
2578 #ifdef __ARCH_WANT_SYS_CLONE3
2579
2580 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2581                                               struct clone_args __user *uargs,
2582                                               size_t usize)
2583 {
2584         int err;
2585         struct clone_args args;
2586         pid_t *kset_tid = kargs->set_tid;
2587
2588         BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2589                      CLONE_ARGS_SIZE_VER0);
2590         BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2591                      CLONE_ARGS_SIZE_VER1);
2592         BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2593                      CLONE_ARGS_SIZE_VER2);
2594         BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2595
2596         if (unlikely(usize > PAGE_SIZE))
2597                 return -E2BIG;
2598         if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2599                 return -EINVAL;
2600
2601         err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2602         if (err)
2603                 return err;
2604
2605         if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2606                 return -EINVAL;
2607
2608         if (unlikely(!args.set_tid && args.set_tid_size > 0))
2609                 return -EINVAL;
2610
2611         if (unlikely(args.set_tid && args.set_tid_size == 0))
2612                 return -EINVAL;
2613
2614         /*
2615          * Verify that higher 32bits of exit_signal are unset and that
2616          * it is a valid signal
2617          */
2618         if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2619                      !valid_signal(args.exit_signal)))
2620                 return -EINVAL;
2621
2622         if ((args.flags & CLONE_INTO_CGROUP) &&
2623             (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2624                 return -EINVAL;
2625
2626         *kargs = (struct kernel_clone_args){
2627                 .flags          = args.flags,
2628                 .pidfd          = u64_to_user_ptr(args.pidfd),
2629                 .child_tid      = u64_to_user_ptr(args.child_tid),
2630                 .parent_tid     = u64_to_user_ptr(args.parent_tid),
2631                 .exit_signal    = args.exit_signal,
2632                 .stack          = args.stack,
2633                 .stack_size     = args.stack_size,
2634                 .tls            = args.tls,
2635                 .set_tid_size   = args.set_tid_size,
2636                 .cgroup         = args.cgroup,
2637         };
2638
2639         if (args.set_tid &&
2640                 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2641                         (kargs->set_tid_size * sizeof(pid_t))))
2642                 return -EFAULT;
2643
2644         kargs->set_tid = kset_tid;
2645
2646         return 0;
2647 }
2648
2649 /**
2650  * clone3_stack_valid - check and prepare stack
2651  * @kargs: kernel clone args
2652  *
2653  * Verify that the stack arguments userspace gave us are sane.
2654  * In addition, set the stack direction for userspace since it's easy for us to
2655  * determine.
2656  */
2657 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2658 {
2659         if (kargs->stack == 0) {
2660                 if (kargs->stack_size > 0)
2661                         return false;
2662         } else {
2663                 if (kargs->stack_size == 0)
2664                         return false;
2665
2666                 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2667                         return false;
2668
2669 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2670                 kargs->stack += kargs->stack_size;
2671 #endif
2672         }
2673
2674         return true;
2675 }
2676
2677 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2678 {
2679         /* Verify that no unknown flags are passed along. */
2680         if (kargs->flags &
2681             ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2682                 return false;
2683
2684         /*
2685          * - make the CLONE_DETACHED bit reuseable for clone3
2686          * - make the CSIGNAL bits reuseable for clone3
2687          */
2688         if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2689                 return false;
2690
2691         if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2692             (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2693                 return false;
2694
2695         if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2696             kargs->exit_signal)
2697                 return false;
2698
2699         if (!clone3_stack_valid(kargs))
2700                 return false;
2701
2702         return true;
2703 }
2704
2705 /**
2706  * clone3 - create a new process with specific properties
2707  * @uargs: argument structure
2708  * @size:  size of @uargs
2709  *
2710  * clone3() is the extensible successor to clone()/clone2().
2711  * It takes a struct as argument that is versioned by its size.
2712  *
2713  * Return: On success, a positive PID for the child process.
2714  *         On error, a negative errno number.
2715  */
2716 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2717 {
2718         int err;
2719
2720         struct kernel_clone_args kargs;
2721         pid_t set_tid[MAX_PID_NS_LEVEL];
2722
2723         kargs.set_tid = set_tid;
2724
2725         err = copy_clone_args_from_user(&kargs, uargs, size);
2726         if (err)
2727                 return err;
2728
2729         if (!clone3_args_valid(&kargs))
2730                 return -EINVAL;
2731
2732         return kernel_clone(&kargs);
2733 }
2734 #endif
2735
2736 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2737 {
2738         struct task_struct *leader, *parent, *child;
2739         int res;
2740
2741         read_lock(&tasklist_lock);
2742         leader = top = top->group_leader;
2743 down:
2744         for_each_thread(leader, parent) {
2745                 list_for_each_entry(child, &parent->children, sibling) {
2746                         res = visitor(child, data);
2747                         if (res) {
2748                                 if (res < 0)
2749                                         goto out;
2750                                 leader = child;
2751                                 goto down;
2752                         }
2753 up:
2754                         ;
2755                 }
2756         }
2757
2758         if (leader != top) {
2759                 child = leader;
2760                 parent = child->real_parent;
2761                 leader = parent->group_leader;
2762                 goto up;
2763         }
2764 out:
2765         read_unlock(&tasklist_lock);
2766 }
2767
2768 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2769 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2770 #endif
2771
2772 static void sighand_ctor(void *data)
2773 {
2774         struct sighand_struct *sighand = data;
2775
2776         spin_lock_init(&sighand->siglock);
2777         init_waitqueue_head(&sighand->signalfd_wqh);
2778 }
2779
2780 void __init proc_caches_init(void)
2781 {
2782         unsigned int mm_size;
2783
2784         sighand_cachep = kmem_cache_create("sighand_cache",
2785                         sizeof(struct sighand_struct), 0,
2786                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2787                         SLAB_ACCOUNT, sighand_ctor);
2788         signal_cachep = kmem_cache_create("signal_cache",
2789                         sizeof(struct signal_struct), 0,
2790                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2791                         NULL);
2792         files_cachep = kmem_cache_create("files_cache",
2793                         sizeof(struct files_struct), 0,
2794                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2795                         NULL);
2796         fs_cachep = kmem_cache_create("fs_cache",
2797                         sizeof(struct fs_struct), 0,
2798                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2799                         NULL);
2800
2801         /*
2802          * The mm_cpumask is located at the end of mm_struct, and is
2803          * dynamically sized based on the maximum CPU number this system
2804          * can have, taking hotplug into account (nr_cpu_ids).
2805          */
2806         mm_size = sizeof(struct mm_struct) + cpumask_size();
2807
2808         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2809                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2810                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2811                         offsetof(struct mm_struct, saved_auxv),
2812                         sizeof_field(struct mm_struct, saved_auxv),
2813                         NULL);
2814         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2815         mmap_init();
2816         nsproxy_cache_init();
2817 }
2818
2819 /*
2820  * Check constraints on flags passed to the unshare system call.
2821  */
2822 static int check_unshare_flags(unsigned long unshare_flags)
2823 {
2824         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2825                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2826                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2827                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
2828                                 CLONE_NEWTIME))
2829                 return -EINVAL;
2830         /*
2831          * Not implemented, but pretend it works if there is nothing
2832          * to unshare.  Note that unsharing the address space or the
2833          * signal handlers also need to unshare the signal queues (aka
2834          * CLONE_THREAD).
2835          */
2836         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2837                 if (!thread_group_empty(current))
2838                         return -EINVAL;
2839         }
2840         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2841                 if (refcount_read(&current->sighand->count) > 1)
2842                         return -EINVAL;
2843         }
2844         if (unshare_flags & CLONE_VM) {
2845                 if (!current_is_single_threaded())
2846                         return -EINVAL;
2847         }
2848
2849         return 0;
2850 }
2851
2852 /*
2853  * Unshare the filesystem structure if it is being shared
2854  */
2855 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2856 {
2857         struct fs_struct *fs = current->fs;
2858
2859         if (!(unshare_flags & CLONE_FS) || !fs)
2860                 return 0;
2861
2862         /* don't need lock here; in the worst case we'll do useless copy */
2863         if (fs->users == 1)
2864                 return 0;
2865
2866         *new_fsp = copy_fs_struct(fs);
2867         if (!*new_fsp)
2868                 return -ENOMEM;
2869
2870         return 0;
2871 }
2872
2873 /*
2874  * Unshare file descriptor table if it is being shared
2875  */
2876 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
2877                struct files_struct **new_fdp)
2878 {
2879         struct files_struct *fd = current->files;
2880         int error = 0;
2881
2882         if ((unshare_flags & CLONE_FILES) &&
2883             (fd && atomic_read(&fd->count) > 1)) {
2884                 *new_fdp = dup_fd(fd, max_fds, &error);
2885                 if (!*new_fdp)
2886                         return error;
2887         }
2888
2889         return 0;
2890 }
2891
2892 /*
2893  * unshare allows a process to 'unshare' part of the process
2894  * context which was originally shared using clone.  copy_*
2895  * functions used by kernel_clone() cannot be used here directly
2896  * because they modify an inactive task_struct that is being
2897  * constructed. Here we are modifying the current, active,
2898  * task_struct.
2899  */
2900 int ksys_unshare(unsigned long unshare_flags)
2901 {
2902         struct fs_struct *fs, *new_fs = NULL;
2903         struct files_struct *fd, *new_fd = NULL;
2904         struct cred *new_cred = NULL;
2905         struct nsproxy *new_nsproxy = NULL;
2906         int do_sysvsem = 0;
2907         int err;
2908
2909         /*
2910          * If unsharing a user namespace must also unshare the thread group
2911          * and unshare the filesystem root and working directories.
2912          */
2913         if (unshare_flags & CLONE_NEWUSER)
2914                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2915         /*
2916          * If unsharing vm, must also unshare signal handlers.
2917          */
2918         if (unshare_flags & CLONE_VM)
2919                 unshare_flags |= CLONE_SIGHAND;
2920         /*
2921          * If unsharing a signal handlers, must also unshare the signal queues.
2922          */
2923         if (unshare_flags & CLONE_SIGHAND)
2924                 unshare_flags |= CLONE_THREAD;
2925         /*
2926          * If unsharing namespace, must also unshare filesystem information.
2927          */
2928         if (unshare_flags & CLONE_NEWNS)
2929                 unshare_flags |= CLONE_FS;
2930
2931         err = check_unshare_flags(unshare_flags);
2932         if (err)
2933                 goto bad_unshare_out;
2934         /*
2935          * CLONE_NEWIPC must also detach from the undolist: after switching
2936          * to a new ipc namespace, the semaphore arrays from the old
2937          * namespace are unreachable.
2938          */
2939         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2940                 do_sysvsem = 1;
2941         err = unshare_fs(unshare_flags, &new_fs);
2942         if (err)
2943                 goto bad_unshare_out;
2944         err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
2945         if (err)
2946                 goto bad_unshare_cleanup_fs;
2947         err = unshare_userns(unshare_flags, &new_cred);
2948         if (err)
2949                 goto bad_unshare_cleanup_fd;
2950         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2951                                          new_cred, new_fs);
2952         if (err)
2953                 goto bad_unshare_cleanup_cred;
2954
2955         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2956                 if (do_sysvsem) {
2957                         /*
2958                          * CLONE_SYSVSEM is equivalent to sys_exit().
2959                          */
2960                         exit_sem(current);
2961                 }
2962                 if (unshare_flags & CLONE_NEWIPC) {
2963                         /* Orphan segments in old ns (see sem above). */
2964                         exit_shm(current);
2965                         shm_init_task(current);
2966                 }
2967
2968                 if (new_nsproxy)
2969                         switch_task_namespaces(current, new_nsproxy);
2970
2971                 task_lock(current);
2972
2973                 if (new_fs) {
2974                         fs = current->fs;
2975                         spin_lock(&fs->lock);
2976                         current->fs = new_fs;
2977                         if (--fs->users)
2978                                 new_fs = NULL;
2979                         else
2980                                 new_fs = fs;
2981                         spin_unlock(&fs->lock);
2982                 }
2983
2984                 if (new_fd) {
2985                         fd = current->files;
2986                         current->files = new_fd;
2987                         new_fd = fd;
2988                 }
2989
2990                 task_unlock(current);
2991
2992                 if (new_cred) {
2993                         /* Install the new user namespace */
2994                         commit_creds(new_cred);
2995                         new_cred = NULL;
2996                 }
2997         }
2998
2999         perf_event_namespaces(current);
3000
3001 bad_unshare_cleanup_cred:
3002         if (new_cred)
3003                 put_cred(new_cred);
3004 bad_unshare_cleanup_fd:
3005         if (new_fd)
3006                 put_files_struct(new_fd);
3007
3008 bad_unshare_cleanup_fs:
3009         if (new_fs)
3010                 free_fs_struct(new_fs);
3011
3012 bad_unshare_out:
3013         return err;
3014 }
3015
3016 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3017 {
3018         return ksys_unshare(unshare_flags);
3019 }
3020
3021 /*
3022  *      Helper to unshare the files of the current task.
3023  *      We don't want to expose copy_files internals to
3024  *      the exec layer of the kernel.
3025  */
3026
3027 int unshare_files(struct files_struct **displaced)
3028 {
3029         struct task_struct *task = current;
3030         struct files_struct *copy = NULL;
3031         int error;
3032
3033         error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, &copy);
3034         if (error || !copy) {
3035                 *displaced = NULL;
3036                 return error;
3037         }
3038         *displaced = task->files;
3039         task_lock(task);
3040         task->files = copy;
3041         task_unlock(task);
3042         return 0;
3043 }
3044
3045 int sysctl_max_threads(struct ctl_table *table, int write,
3046                        void *buffer, size_t *lenp, loff_t *ppos)
3047 {
3048         struct ctl_table t;
3049         int ret;
3050         int threads = max_threads;
3051         int min = 1;
3052         int max = MAX_THREADS;
3053
3054         t = *table;
3055         t.data = &threads;
3056         t.extra1 = &min;
3057         t.extra2 = &max;
3058
3059         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3060         if (ret || !write)
3061                 return ret;
3062
3063         max_threads = threads;
3064
3065         return 0;
3066 }