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