kernel/fork.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/kernel/fork.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  */
   7
   8 /*
   9  *  'fork.c' contains the help-routines for the 'fork' system call
  10  * (see also entry.S and others).
  11  * Fork is rather simple, once you get the hang of it, but the memory
  12  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  13  */
  14
  15 #include <linux/anon_inodes.h>
  16 #include <linux/slab.h>
  17 #include <linux/sched/autogroup.h>
  18 #include <linux/sched/mm.h>
  19 #include <linux/sched/coredump.h>
  20 #include <linux/sched/user.h>
  21 #include <linux/sched/numa_balancing.h>
  22 #include <linux/sched/stat.h>
  23 #include <linux/sched/task.h>
  24 #include <linux/sched/task_stack.h>
  25 #include <linux/sched/cputime.h>
  26 #include <linux/seq_file.h>
  27 #include <linux/rtmutex.h>
  28 #include <linux/init.h>
  29 #include <linux/unistd.h>
  30 #include <linux/module.h>
  31 #include <linux/vmalloc.h>
  32 #include <linux/completion.h>
  33 #include <linux/personality.h>
  34 #include <linux/mempolicy.h>
  35 #include <linux/sem.h>
  36 #include <linux/file.h>
  37 #include <linux/fdtable.h>
  38 #include <linux/iocontext.h>
  39 #include <linux/key.h>
  40 #include <linux/binfmts.h>
  41 #include <linux/mman.h>
  42 #include <linux/mmu_notifier.h>
  43 #include <linux/fs.h>
  44 #include <linux/mm.h>
  45 #include <linux/vmacache.h>
  46 #include <linux/nsproxy.h>
  47 #include <linux/capability.h>
  48 #include <linux/cpu.h>
  49 #include <linux/cgroup.h>
  50 #include <linux/security.h>
  51 #include <linux/hugetlb.h>
  52 #include <linux/seccomp.h>
  53 #include <linux/swap.h>
  54 #include <linux/syscalls.h>
  55 #include <linux/jiffies.h>
  56 #include <linux/futex.h>
  57 #include <linux/compat.h>
  58 #include <linux/kthread.h>
  59 #include <linux/task_io_accounting_ops.h>
  60 #include <linux/rcupdate.h>
  61 #include <linux/ptrace.h>
  62 #include <linux/mount.h>
  63 #include <linux/audit.h>
  64 #include <linux/memcontrol.h>
  65 #include <linux/ftrace.h>
  66 #include <linux/proc_fs.h>
  67 #include <linux/profile.h>
  68 #include <linux/rmap.h>
  69 #include <linux/ksm.h>
  70 #include <linux/acct.h>
  71 #include <linux/userfaultfd_k.h>
  72 #include <linux/tsacct_kern.h>
  73 #include <linux/cn_proc.h>
  74 #include <linux/freezer.h>
  75 #include <linux/delayacct.h>
  76 #include <linux/taskstats_kern.h>
  77 #include <linux/random.h>
  78 #include <linux/tty.h>
  79 #include <linux/blkdev.h>
  80 #include <linux/fs_struct.h>
  81 #include <linux/magic.h>
  82 #include <linux/perf_event.h>
  83 #include <linux/posix-timers.h>
  84 #include <linux/user-return-notifier.h>
  85 #include <linux/oom.h>
  86 #include <linux/khugepaged.h>
  87 #include <linux/signalfd.h>
  88 #include <linux/uprobes.h>
  89 #include <linux/aio.h>
  90 #include <linux/compiler.h>
  91 #include <linux/sysctl.h>
  92 #include <linux/kcov.h>
  93 #include <linux/livepatch.h>
  94 #include <linux/thread_info.h>
  95 #include <linux/stackleak.h>
  96 #include <linux/kasan.h>
  97 #include <linux/scs.h>
  98 #include <linux/io_uring.h>
  99 #include <linux/bpf.h>
 100
 101 #include <asm/pgalloc.h>
 102 #include <linux/uaccess.h>
 103 #include <asm/mmu_context.h>
 104 #include <asm/cacheflush.h>
 105 #include <asm/tlbflush.h>
 106
 107 #include <trace/events/sched.h>
 108
 109 #define CREATE_TRACE_POINTS
 110 #include <trace/events/task.h>
 111
 112 /*
 113  * Minimum number of threads to boot the kernel
 114  */
 115 #define MIN_THREADS 20
 116
 117 /*
 118  * Maximum number of threads
 119  */
 120 #define MAX_THREADS FUTEX_TID_MASK
 121
 122 /*
 123  * Protected counters by write_lock_irq(&tasklist_lock)
 124  */
 125 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 126 int nr_threads;                 /* The idle threads do not count.. */
 127
 128 static int max_threads;         /* tunable limit on nr_threads */
 129
 130 #define NAMED_ARRAY_INDEX(x)    [x] = __stringify(x)
 131
 132 static const char * const resident_page_types[] = {
 133         NAMED_ARRAY_INDEX(MM_FILEPAGES),
 134         NAMED_ARRAY_INDEX(MM_ANONPAGES),
 135         NAMED_ARRAY_INDEX(MM_SWAPENTS),
 136         NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
 137 };
 138
 139 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 140
 141 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 142
 143 #ifdef CONFIG_PROVE_RCU
 144 int lockdep_tasklist_lock_is_held(void)
 145 {
 146         return lockdep_is_held(&tasklist_lock);
 147 }
 148 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 149 #endif /* #ifdef CONFIG_PROVE_RCU */
 150
 151 int nr_processes(void)
 152 {
 153         int cpu;
 154         int total = 0;
 155
 156         for_each_possible_cpu(cpu)
 157                 total += per_cpu(process_counts, cpu);
 158
 159         return total;
 160 }
 161
 162 void __weak arch_release_task_struct(struct task_struct *tsk)
 163 {
 164 }
 165
 166 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 167 static struct kmem_cache *task_struct_cachep;
 168
 169 static inline struct task_struct *alloc_task_struct_node(int node)
 170 {
 171         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 172 }
 173
 174 static inline void free_task_struct(struct task_struct *tsk)
 175 {
 176         kmem_cache_free(task_struct_cachep, tsk);
 177 }
 178 #endif
 179
 180 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 181
 182 /*
 183  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 184  * kmemcache based allocator.
 185  */
 186 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 187
 188 #ifdef CONFIG_VMAP_STACK
 189 /*
 190  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 191  * flush.  Try to minimize the number of calls by caching stacks.
 192  */
 193 #define NR_CACHED_STACKS 2
 194 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 195
 196 static int free_vm_stack_cache(unsigned int cpu)
 197 {
 198         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 199         int i;
 200
 201         for (i = 0; i < NR_CACHED_STACKS; i++) {
 202                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 203
 204                 if (!vm_stack)
 205                         continue;
 206
 207                 vfree(vm_stack->addr);
 208                 cached_vm_stacks[i] = NULL;
 209         }
 210
 211         return 0;
 212 }
 213 #endif
 214
 215 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 216 {
 217 #ifdef CONFIG_VMAP_STACK
 218         void *stack;
 219         int i;
 220
 221         for (i = 0; i < NR_CACHED_STACKS; i++) {
 222                 struct vm_struct *s;
 223
 224                 s = this_cpu_xchg(cached_stacks[i], NULL);
 225
 226                 if (!s)
 227                         continue;
 228
 229                 /* Mark stack accessible for KASAN. */
 230                 kasan_unpoison_range(s->addr, THREAD_SIZE);
 231
 232                 /* Clear stale pointers from reused stack. */
 233                 memset(s->addr, 0, THREAD_SIZE);
 234
 235                 tsk->stack_vm_area = s;
 236                 tsk->stack = s->addr;
 237                 return s->addr;
 238         }
 239
 240         /*
 241          * Allocated stacks are cached and later reused by new threads,
 242          * so memcg accounting is performed manually on assigning/releasing
 243          * stacks to tasks. Drop __GFP_ACCOUNT.
 244          */
 245         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 246                                      VMALLOC_START, VMALLOC_END,
 247                                      THREADINFO_GFP & ~__GFP_ACCOUNT,
 248                                      PAGE_KERNEL,
 249                                      0, node, __builtin_return_address(0));
 250
 251         /*
 252          * We can't call find_vm_area() in interrupt context, and
 253          * free_thread_stack() can be called in interrupt context,
 254          * so cache the vm_struct.
 255          */
 256         if (stack) {
 257                 tsk->stack_vm_area = find_vm_area(stack);
 258                 tsk->stack = stack;
 259         }
 260         return stack;
 261 #else
 262         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 263                                              THREAD_SIZE_ORDER);
 264
 265         if (likely(page)) {
 266                 tsk->stack = kasan_reset_tag(page_address(page));
 267                 return tsk->stack;
 268         }
 269         return NULL;
 270 #endif
 271 }
 272
 273 static inline void free_thread_stack(struct task_struct *tsk)
 274 {
 275 #ifdef CONFIG_VMAP_STACK
 276         struct vm_struct *vm = task_stack_vm_area(tsk);
 277
 278         if (vm) {
 279                 int i;
 280
 281                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 282                         memcg_kmem_uncharge_page(vm->pages[i], 0);
 283
 284                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 285                         if (this_cpu_cmpxchg(cached_stacks[i],
 286                                         NULL, tsk->stack_vm_area) != NULL)
 287                                 continue;
 288
 289                         return;
 290                 }
 291
 292                 vfree_atomic(tsk->stack);
 293                 return;
 294         }
 295 #endif
 296
 297         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 298 }
 299 # else
 300 static struct kmem_cache *thread_stack_cache;
 301
 302 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 303                                                   int node)
 304 {
 305         unsigned long *stack;
 306         stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 307         stack = kasan_reset_tag(stack);
 308         tsk->stack = stack;
 309         return stack;
 310 }
 311
 312 static void free_thread_stack(struct task_struct *tsk)
 313 {
 314         kmem_cache_free(thread_stack_cache, tsk->stack);
 315 }
 316
 317 void thread_stack_cache_init(void)
 318 {
 319         thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 320                                         THREAD_SIZE, THREAD_SIZE, 0, 0,
 321                                         THREAD_SIZE, NULL);
 322         BUG_ON(thread_stack_cache == NULL);
 323 }
 324 # endif
 325 #endif
 326
 327 /* SLAB cache for signal_struct structures (tsk->signal) */
 328 static struct kmem_cache *signal_cachep;
 329
 330 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 331 struct kmem_cache *sighand_cachep;
 332
 333 /* SLAB cache for files_struct structures (tsk->files) */
 334 struct kmem_cache *files_cachep;
 335
 336 /* SLAB cache for fs_struct structures (tsk->fs) */
 337 struct kmem_cache *fs_cachep;
 338
 339 /* SLAB cache for vm_area_struct structures */
 340 static struct kmem_cache *vm_area_cachep;
 341
 342 /* SLAB cache for mm_struct structures (tsk->mm) */
 343 static struct kmem_cache *mm_cachep;
 344
 345 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 346 {
 347         struct vm_area_struct *vma;
 348
 349         vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 350         if (vma)
 351                 vma_init(vma, mm);
 352         return vma;
 353 }
 354
 355 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 356 {
 357         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 358
 359         if (new) {
 360                 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
 361                 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
 362                 /*
 363                  * orig->shared.rb may be modified concurrently, but the clone
 364                  * will be reinitialized.
 365                  */
 366                 *new = data_race(*orig);
 367                 INIT_LIST_HEAD(&new->anon_vma_chain);
 368                 new->vm_next = new->vm_prev = NULL;
 369         }
 370         return new;
 371 }
 372
 373 void vm_area_free(struct vm_area_struct *vma)
 374 {
 375         kmem_cache_free(vm_area_cachep, vma);
 376 }
 377
 378 static void account_kernel_stack(struct task_struct *tsk, int account)
 379 {
 380         void *stack = task_stack_page(tsk);
 381         struct vm_struct *vm = task_stack_vm_area(tsk);
 382
 383         if (vm) {
 384                 int i;
 385
 386                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 387                         mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
 388                                               account * (PAGE_SIZE / 1024));
 389         } else {
 390                 /* All stack pages are in the same node. */
 391                 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
 392                                       account * (THREAD_SIZE / 1024));
 393         }
 394 }
 395
 396 static int memcg_charge_kernel_stack(struct task_struct *tsk)
 397 {
 398 #ifdef CONFIG_VMAP_STACK
 399         struct vm_struct *vm = task_stack_vm_area(tsk);
 400         int ret;
 401
 402         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 403
 404         if (vm) {
 405                 int i;
 406
 407                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 408
 409                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 410                         /*
 411                          * If memcg_kmem_charge_page() fails, page's
 412                          * memory cgroup pointer is NULL, and
 413                          * memcg_kmem_uncharge_page() in free_thread_stack()
 414                          * will ignore this page.
 415                          */
 416                         ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL,
 417                                                      0);
 418                         if (ret)
 419                                 return ret;
 420                 }
 421         }
 422 #endif
 423         return 0;
 424 }
 425
 426 static void release_task_stack(struct task_struct *tsk)
 427 {
 428         if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
 429                 return;  /* Better to leak the stack than to free prematurely */
 430
 431         account_kernel_stack(tsk, -1);
 432         free_thread_stack(tsk);
 433         tsk->stack = NULL;
 434 #ifdef CONFIG_VMAP_STACK
 435         tsk->stack_vm_area = NULL;
 436 #endif
 437 }
 438
 439 #ifdef CONFIG_THREAD_INFO_IN_TASK
 440 void put_task_stack(struct task_struct *tsk)
 441 {
 442         if (refcount_dec_and_test(&tsk->stack_refcount))
 443                 release_task_stack(tsk);
 444 }
 445 #endif
 446
 447 void free_task(struct task_struct *tsk)
 448 {
 449         release_user_cpus_ptr(tsk);
 450         scs_release(tsk);
 451
 452 #ifndef CONFIG_THREAD_INFO_IN_TASK
 453         /*
 454          * The task is finally done with both the stack and thread_info,
 455          * so free both.
 456          */
 457         release_task_stack(tsk);
 458 #else
 459         /*
 460          * If the task had a separate stack allocation, it should be gone
 461          * by now.
 462          */
 463         WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
 464 #endif
 465         rt_mutex_debug_task_free(tsk);
 466         ftrace_graph_exit_task(tsk);
 467         arch_release_task_struct(tsk);
 468         if (tsk->flags & PF_KTHREAD)
 469                 free_kthread_struct(tsk);
 470         free_task_struct(tsk);
 471 }
 472 EXPORT_SYMBOL(free_task);
 473
 474 #ifdef CONFIG_MMU
 475 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 476                                         struct mm_struct *oldmm)
 477 {
 478         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 479         struct rb_node **rb_link, *rb_parent;
 480         int retval;
 481         unsigned long charge;
 482         LIST_HEAD(uf);
 483
 484         uprobe_start_dup_mmap();
 485         if (mmap_write_lock_killable(oldmm)) {
 486                 retval = -EINTR;
 487                 goto fail_uprobe_end;
 488         }
 489         flush_cache_dup_mm(oldmm);
 490         uprobe_dup_mmap(oldmm, mm);
 491         /*
 492          * Not linked in yet - no deadlock potential:
 493          */
 494         mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
 495
 496         /* No ordering required: file already has been exposed. */
 497         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 498
 499         mm->total_vm = oldmm->total_vm;
 500         mm->data_vm = oldmm->data_vm;
 501         mm->exec_vm = oldmm->exec_vm;
 502         mm->stack_vm = oldmm->stack_vm;
 503
 504         rb_link = &mm->mm_rb.rb_node;
 505         rb_parent = NULL;
 506         pprev = &mm->mmap;
 507         retval = ksm_fork(mm, oldmm);
 508         if (retval)
 509                 goto out;
 510         retval = khugepaged_fork(mm, oldmm);
 511         if (retval)
 512                 goto out;
 513
 514         prev = NULL;
 515         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 516                 struct file *file;
 517
 518                 if (mpnt->vm_flags & VM_DONTCOPY) {
 519                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 520                         continue;
 521                 }
 522                 charge = 0;
 523                 /*
 524                  * Don't duplicate many vmas if we've been oom-killed (for
 525                  * example)
 526                  */
 527                 if (fatal_signal_pending(current)) {
 528                         retval = -EINTR;
 529                         goto out;
 530                 }
 531                 if (mpnt->vm_flags & VM_ACCOUNT) {
 532                         unsigned long len = vma_pages(mpnt);
 533
 534                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 535                                 goto fail_nomem;
 536                         charge = len;
 537                 }
 538                 tmp = vm_area_dup(mpnt);
 539                 if (!tmp)
 540                         goto fail_nomem;
 541                 retval = vma_dup_policy(mpnt, tmp);
 542                 if (retval)
 543                         goto fail_nomem_policy;
 544                 tmp->vm_mm = mm;
 545                 retval = dup_userfaultfd(tmp, &uf);
 546                 if (retval)
 547                         goto fail_nomem_anon_vma_fork;
 548                 if (tmp->vm_flags & VM_WIPEONFORK) {
 549                         /*
 550                          * VM_WIPEONFORK gets a clean slate in the child.
 551                          * Don't prepare anon_vma until fault since we don't
 552                          * copy page for current vma.
 553                          */
 554                         tmp->anon_vma = NULL;
 555                 } else if (anon_vma_fork(tmp, mpnt))
 556                         goto fail_nomem_anon_vma_fork;
 557                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 558                 file = tmp->vm_file;
 559                 if (file) {
 560                         struct inode *inode = file_inode(file);
 561                         struct address_space *mapping = file->f_mapping;
 562
 563                         get_file(file);
 564                         if (tmp->vm_flags & VM_DENYWRITE)
 565                                 put_write_access(inode);
 566                         i_mmap_lock_write(mapping);
 567                         if (tmp->vm_flags & VM_SHARED)
 568                                 mapping_allow_writable(mapping);
 569                         flush_dcache_mmap_lock(mapping);
 570                         /* insert tmp into the share list, just after mpnt */
 571                         vma_interval_tree_insert_after(tmp, mpnt,
 572                                         &mapping->i_mmap);
 573                         flush_dcache_mmap_unlock(mapping);
 574                         i_mmap_unlock_write(mapping);
 575                 }
 576
 577                 /*
 578                  * Clear hugetlb-related page reserves for children. This only
 579                  * affects MAP_PRIVATE mappings. Faults generated by the child
 580                  * are not guaranteed to succeed, even if read-only
 581                  */
 582                 if (is_vm_hugetlb_page(tmp))
 583                         reset_vma_resv_huge_pages(tmp);
 584
 585                 /*
 586                  * Link in the new vma and copy the page table entries.
 587                  */
 588                 *pprev = tmp;
 589                 pprev = &tmp->vm_next;
 590                 tmp->vm_prev = prev;
 591                 prev = tmp;
 592
 593                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 594                 rb_link = &tmp->vm_rb.rb_right;
 595                 rb_parent = &tmp->vm_rb;
 596
 597                 mm->map_count++;
 598                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 599                         retval = copy_page_range(tmp, mpnt);
 600
 601                 if (tmp->vm_ops && tmp->vm_ops->open)
 602                         tmp->vm_ops->open(tmp);
 603
 604                 if (retval)
 605                         goto out;
 606         }
 607         /* a new mm has just been created */
 608         retval = arch_dup_mmap(oldmm, mm);
 609 out:
 610         mmap_write_unlock(mm);
 611         flush_tlb_mm(oldmm);
 612         mmap_write_unlock(oldmm);
 613         dup_userfaultfd_complete(&uf);
 614 fail_uprobe_end:
 615         uprobe_end_dup_mmap();
 616         return retval;
 617 fail_nomem_anon_vma_fork:
 618         mpol_put(vma_policy(tmp));
 619 fail_nomem_policy:
 620         vm_area_free(tmp);
 621 fail_nomem:
 622         retval = -ENOMEM;
 623         vm_unacct_memory(charge);
 624         goto out;
 625 }
 626
 627 static inline int mm_alloc_pgd(struct mm_struct *mm)
 628 {
 629         mm->pgd = pgd_alloc(mm);
 630         if (unlikely(!mm->pgd))
 631                 return -ENOMEM;
 632         return 0;
 633 }
 634
 635 static inline void mm_free_pgd(struct mm_struct *mm)
 636 {
 637         pgd_free(mm, mm->pgd);
 638 }
 639 #else
 640 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 641 {
 642         mmap_write_lock(oldmm);
 643         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 644         mmap_write_unlock(oldmm);
 645         return 0;
 646 }
 647 #define mm_alloc_pgd(mm)        (0)
 648 #define mm_free_pgd(mm)
 649 #endif /* CONFIG_MMU */
 650
 651 static void check_mm(struct mm_struct *mm)
 652 {
 653         int i;
 654
 655         BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
 656                          "Please make sure 'struct resident_page_types[]' is updated as well");
 657
 658         for (i = 0; i < NR_MM_COUNTERS; i++) {
 659                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 660
 661                 if (unlikely(x))
 662                         pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
 663                                  mm, resident_page_types[i], x);
 664         }
 665
 666         if (mm_pgtables_bytes(mm))
 667                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 668                                 mm_pgtables_bytes(mm));
 669
 670 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 671         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 672 #endif
 673 }
 674
 675 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 676 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 677
 678 /*
 679  * Called when the last reference to the mm
 680  * is dropped: either by a lazy thread or by
 681  * mmput. Free the page directory and the mm.
 682  */
 683 void __mmdrop(struct mm_struct *mm)
 684 {
 685         BUG_ON(mm == &init_mm);
 686         WARN_ON_ONCE(mm == current->mm);
 687         WARN_ON_ONCE(mm == current->active_mm);
 688         mm_free_pgd(mm);
 689         destroy_context(mm);
 690         mmu_notifier_subscriptions_destroy(mm);
 691         check_mm(mm);
 692         put_user_ns(mm->user_ns);
 693         free_mm(mm);
 694 }
 695 EXPORT_SYMBOL_GPL(__mmdrop);
 696
 697 static void mmdrop_async_fn(struct work_struct *work)
 698 {
 699         struct mm_struct *mm;
 700
 701         mm = container_of(work, struct mm_struct, async_put_work);
 702         __mmdrop(mm);
 703 }
 704
 705 static void mmdrop_async(struct mm_struct *mm)
 706 {
 707         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 708                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 709                 schedule_work(&mm->async_put_work);
 710         }
 711 }
 712
 713 static inline void free_signal_struct(struct signal_struct *sig)
 714 {
 715         taskstats_tgid_free(sig);
 716         sched_autogroup_exit(sig);
 717         /*
 718          * __mmdrop is not safe to call from softirq context on x86 due to
 719          * pgd_dtor so postpone it to the async context
 720          */
 721         if (sig->oom_mm)
 722                 mmdrop_async(sig->oom_mm);
 723         kmem_cache_free(signal_cachep, sig);
 724 }
 725
 726 static inline void put_signal_struct(struct signal_struct *sig)
 727 {
 728         if (refcount_dec_and_test(&sig->sigcnt))
 729                 free_signal_struct(sig);
 730 }
 731
 732 void __put_task_struct(struct task_struct *tsk)
 733 {
 734         WARN_ON(!tsk->exit_state);
 735         WARN_ON(refcount_read(&tsk->usage));
 736         WARN_ON(tsk == current);
 737
 738         io_uring_free(tsk);
 739         cgroup_free(tsk);
 740         task_numa_free(tsk, true);
 741         security_task_free(tsk);
 742         bpf_task_storage_free(tsk);
 743         exit_creds(tsk);
 744         delayacct_tsk_free(tsk);
 745         put_signal_struct(tsk->signal);
 746         sched_core_free(tsk);
 747
 748         if (!profile_handoff_task(tsk))
 749                 free_task(tsk);
 750 }
 751 EXPORT_SYMBOL_GPL(__put_task_struct);
 752
 753 void __init __weak arch_task_cache_init(void) { }
 754
 755 /*
 756  * set_max_threads
 757  */
 758 static void set_max_threads(unsigned int max_threads_suggested)
 759 {
 760         u64 threads;
 761         unsigned long nr_pages = totalram_pages();
 762
 763         /*
 764          * The number of threads shall be limited such that the thread
 765          * structures may only consume a small part of the available memory.
 766          */
 767         if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
 768                 threads = MAX_THREADS;
 769         else
 770                 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
 771                                     (u64) THREAD_SIZE * 8UL);
 772
 773         if (threads > max_threads_suggested)
 774                 threads = max_threads_suggested;
 775
 776         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 777 }
 778
 779 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 780 /* Initialized by the architecture: */
 781 int arch_task_struct_size __read_mostly;
 782 #endif
 783
 784 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 785 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 786 {
 787         /* Fetch thread_struct whitelist for the architecture. */
 788         arch_thread_struct_whitelist(offset, size);
 789
 790         /*
 791          * Handle zero-sized whitelist or empty thread_struct, otherwise
 792          * adjust offset to position of thread_struct in task_struct.
 793          */
 794         if (unlikely(*size == 0))
 795                 *offset = 0;
 796         else
 797                 *offset += offsetof(struct task_struct, thread);
 798 }
 799 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
 800
 801 void __init fork_init(void)
 802 {
 803         int i;
 804 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 805 #ifndef ARCH_MIN_TASKALIGN
 806 #define ARCH_MIN_TASKALIGN      0
 807 #endif
 808         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 809         unsigned long useroffset, usersize;
 810
 811         /* create a slab on which task_structs can be allocated */
 812         task_struct_whitelist(&useroffset, &usersize);
 813         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 814                         arch_task_struct_size, align,
 815                         SLAB_PANIC|SLAB_ACCOUNT,
 816                         useroffset, usersize, NULL);
 817 #endif
 818
 819         /* do the arch specific task caches init */
 820         arch_task_cache_init();
 821
 822         set_max_threads(MAX_THREADS);
 823
 824         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 825         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 826         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 827                 init_task.signal->rlim[RLIMIT_NPROC];
 828
 829         for (i = 0; i < MAX_PER_NAMESPACE_UCOUNTS; i++)
 830                 init_user_ns.ucount_max[i] = max_threads/2;
 831
 832         set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_NPROC,      RLIM_INFINITY);
 833         set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE,   RLIM_INFINITY);
 834         set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
 835         set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK,    RLIM_INFINITY);
 836
 837 #ifdef CONFIG_VMAP_STACK
 838         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 839                           NULL, free_vm_stack_cache);
 840 #endif
 841
 842         scs_init();
 843
 844         lockdep_init_task(&init_task);
 845         uprobes_init();
 846 }
 847
 848 int __weak arch_dup_task_struct(struct task_struct *dst,
 849                                                struct task_struct *src)
 850 {
 851         *dst = *src;
 852         return 0;
 853 }
 854
 855 void set_task_stack_end_magic(struct task_struct *tsk)
 856 {
 857         unsigned long *stackend;
 858
 859         stackend = end_of_stack(tsk);
 860         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 861 }
 862
 863 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 864 {
 865         struct task_struct *tsk;
 866         unsigned long *stack;
 867         struct vm_struct *stack_vm_area __maybe_unused;
 868         int err;
 869
 870         if (node == NUMA_NO_NODE)
 871                 node = tsk_fork_get_node(orig);
 872         tsk = alloc_task_struct_node(node);
 873         if (!tsk)
 874                 return NULL;
 875
 876         stack = alloc_thread_stack_node(tsk, node);
 877         if (!stack)
 878                 goto free_tsk;
 879
 880         if (memcg_charge_kernel_stack(tsk))
 881                 goto free_stack;
 882
 883         stack_vm_area = task_stack_vm_area(tsk);
 884
 885         err = arch_dup_task_struct(tsk, orig);
 886
 887         /*
 888          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 889          * sure they're properly initialized before using any stack-related
 890          * functions again.
 891          */
 892         tsk->stack = stack;
 893 #ifdef CONFIG_VMAP_STACK
 894         tsk->stack_vm_area = stack_vm_area;
 895 #endif
 896 #ifdef CONFIG_THREAD_INFO_IN_TASK
 897         refcount_set(&tsk->stack_refcount, 1);
 898 #endif
 899
 900         if (err)
 901                 goto free_stack;
 902
 903         err = scs_prepare(tsk, node);
 904         if (err)
 905                 goto free_stack;
 906
 907 #ifdef CONFIG_SECCOMP
 908         /*
 909          * We must handle setting up seccomp filters once we're under
 910          * the sighand lock in case orig has changed between now and
 911          * then. Until then, filter must be NULL to avoid messing up
 912          * the usage counts on the error path calling free_task.
 913          */
 914         tsk->seccomp.filter = NULL;
 915 #endif
 916
 917         setup_thread_stack(tsk, orig);
 918         clear_user_return_notifier(tsk);
 919         clear_tsk_need_resched(tsk);
 920         set_task_stack_end_magic(tsk);
 921         clear_syscall_work_syscall_user_dispatch(tsk);
 922
 923 #ifdef CONFIG_STACKPROTECTOR
 924         tsk->stack_canary = get_random_canary();
 925 #endif
 926         if (orig->cpus_ptr == &orig->cpus_mask)
 927                 tsk->cpus_ptr = &tsk->cpus_mask;
 928         dup_user_cpus_ptr(tsk, orig, node);
 929
 930         /*
 931          * One for the user space visible state that goes away when reaped.
 932          * One for the scheduler.
 933          */
 934         refcount_set(&tsk->rcu_users, 2);
 935         /* One for the rcu users */
 936         refcount_set(&tsk->usage, 1);
 937 #ifdef CONFIG_BLK_DEV_IO_TRACE
 938         tsk->btrace_seq = 0;
 939 #endif
 940         tsk->splice_pipe = NULL;
 941         tsk->task_frag.page = NULL;
 942         tsk->wake_q.next = NULL;
 943         tsk->pf_io_worker = NULL;
 944
 945         account_kernel_stack(tsk, 1);
 946
 947         kcov_task_init(tsk);
 948         kmap_local_fork(tsk);
 949
 950 #ifdef CONFIG_FAULT_INJECTION
 951         tsk->fail_nth = 0;
 952 #endif
 953
 954 #ifdef CONFIG_BLK_CGROUP
 955         tsk->throttle_queue = NULL;
 956         tsk->use_memdelay = 0;
 957 #endif
 958
 959 #ifdef CONFIG_MEMCG
 960         tsk->active_memcg = NULL;
 961 #endif
 962         return tsk;
 963
 964 free_stack:
 965         free_thread_stack(tsk);
 966 free_tsk:
 967         free_task_struct(tsk);
 968         return NULL;
 969 }
 970
 971 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 972
 973 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 974
 975 static int __init coredump_filter_setup(char *s)
 976 {
 977         default_dump_filter =
 978                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 979                 MMF_DUMP_FILTER_MASK;
 980         return 1;
 981 }
 982
 983 __setup("coredump_filter=", coredump_filter_setup);
 984
 985 #include <linux/init_task.h>
 986
 987 static void mm_init_aio(struct mm_struct *mm)
 988 {
 989 #ifdef CONFIG_AIO
 990         spin_lock_init(&mm->ioctx_lock);
 991         mm->ioctx_table = NULL;
 992 #endif
 993 }
 994
 995 static __always_inline void mm_clear_owner(struct mm_struct *mm,
 996                                            struct task_struct *p)
 997 {
 998 #ifdef CONFIG_MEMCG
 999         if (mm->owner == p)
1000                 WRITE_ONCE(mm->owner, NULL);
1001 #endif
1002 }
1003
1004 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1005 {
1006 #ifdef CONFIG_MEMCG
1007         mm->owner = p;
1008 #endif
1009 }
1010
1011 static void mm_init_pasid(struct mm_struct *mm)
1012 {
1013 #ifdef CONFIG_IOMMU_SUPPORT
1014         mm->pasid = INIT_PASID;
1015 #endif
1016 }
1017
1018 static void mm_init_uprobes_state(struct mm_struct *mm)
1019 {
1020 #ifdef CONFIG_UPROBES
1021         mm->uprobes_state.xol_area = NULL;
1022 #endif
1023 }
1024
1025 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1026         struct user_namespace *user_ns)
1027 {
1028         mm->mmap = NULL;
1029         mm->mm_rb = RB_ROOT;
1030         mm->vmacache_seqnum = 0;
1031         atomic_set(&mm->mm_users, 1);
1032         atomic_set(&mm->mm_count, 1);
1033         seqcount_init(&mm->write_protect_seq);
1034         mmap_init_lock(mm);
1035         INIT_LIST_HEAD(&mm->mmlist);
1036         mm->core_state = NULL;
1037         mm_pgtables_bytes_init(mm);
1038         mm->map_count = 0;
1039         mm->locked_vm = 0;
1040         atomic64_set(&mm->pinned_vm, 0);
1041         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1042         spin_lock_init(&mm->page_table_lock);
1043         spin_lock_init(&mm->arg_lock);
1044         mm_init_cpumask(mm);
1045         mm_init_aio(mm);
1046         mm_init_owner(mm, p);
1047         mm_init_pasid(mm);
1048         RCU_INIT_POINTER(mm->exe_file, NULL);
1049         mmu_notifier_subscriptions_init(mm);
1050         init_tlb_flush_pending(mm);
1051 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1052         mm->pmd_huge_pte = NULL;
1053 #endif
1054         mm_init_uprobes_state(mm);
1055
1056         if (current->mm) {
1057                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1058                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1059         } else {
1060                 mm->flags = default_dump_filter;
1061                 mm->def_flags = 0;
1062         }
1063
1064         if (mm_alloc_pgd(mm))
1065                 goto fail_nopgd;
1066
1067         if (init_new_context(p, mm))
1068                 goto fail_nocontext;
1069
1070         mm->user_ns = get_user_ns(user_ns);
1071         return mm;
1072
1073 fail_nocontext:
1074         mm_free_pgd(mm);
1075 fail_nopgd:
1076         free_mm(mm);
1077         return NULL;
1078 }
1079
1080 /*
1081  * Allocate and initialize an mm_struct.
1082  */
1083 struct mm_struct *mm_alloc(void)
1084 {
1085         struct mm_struct *mm;
1086
1087         mm = allocate_mm();
1088         if (!mm)
1089                 return NULL;
1090
1091         memset(mm, 0, sizeof(*mm));
1092         return mm_init(mm, current, current_user_ns());
1093 }
1094
1095 static inline void __mmput(struct mm_struct *mm)
1096 {
1097         VM_BUG_ON(atomic_read(&mm->mm_users));
1098
1099         uprobe_clear_state(mm);
1100         exit_aio(mm);
1101         ksm_exit(mm);
1102         khugepaged_exit(mm); /* must run before exit_mmap */
1103         exit_mmap(mm);
1104         mm_put_huge_zero_page(mm);
1105         set_mm_exe_file(mm, NULL);
1106         if (!list_empty(&mm->mmlist)) {
1107                 spin_lock(&mmlist_lock);
1108                 list_del(&mm->mmlist);
1109                 spin_unlock(&mmlist_lock);
1110         }
1111         if (mm->binfmt)
1112                 module_put(mm->binfmt->module);
1113         mmdrop(mm);
1114 }
1115
1116 /*
1117  * Decrement the use count and release all resources for an mm.
1118  */
1119 void mmput(struct mm_struct *mm)
1120 {
1121         might_sleep();
1122
1123         if (atomic_dec_and_test(&mm->mm_users))
1124                 __mmput(mm);
1125 }
1126 EXPORT_SYMBOL_GPL(mmput);
1127
1128 #ifdef CONFIG_MMU
1129 static void mmput_async_fn(struct work_struct *work)
1130 {
1131         struct mm_struct *mm = container_of(work, struct mm_struct,
1132                                             async_put_work);
1133
1134         __mmput(mm);
1135 }
1136
1137 void mmput_async(struct mm_struct *mm)
1138 {
1139         if (atomic_dec_and_test(&mm->mm_users)) {
1140                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1141                 schedule_work(&mm->async_put_work);
1142         }
1143 }
1144 #endif
1145
1146 /**
1147  * set_mm_exe_file - change a reference to the mm's executable file
1148  *
1149  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1150  *
1151  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1152  * invocations: in mmput() nobody alive left, in execve task is single
1153  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1154  * mm->exe_file, but does so without using set_mm_exe_file() in order
1155  * to avoid the need for any locks.
1156  */
1157 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1158 {
1159         struct file *old_exe_file;
1160
1161         /*
1162          * It is safe to dereference the exe_file without RCU as
1163          * this function is only called if nobody else can access
1164          * this mm -- see comment above for justification.
1165          */
1166         old_exe_file = rcu_dereference_raw(mm->exe_file);
1167
1168         if (new_exe_file)
1169                 get_file(new_exe_file);
1170         rcu_assign_pointer(mm->exe_file, new_exe_file);
1171         if (old_exe_file)
1172                 fput(old_exe_file);
1173 }
1174
1175 /**
1176  * get_mm_exe_file - acquire a reference to the mm's executable file
1177  *
1178  * Returns %NULL if mm has no associated executable file.
1179  * User must release file via fput().
1180  */
1181 struct file *get_mm_exe_file(struct mm_struct *mm)
1182 {
1183         struct file *exe_file;
1184
1185         rcu_read_lock();
1186         exe_file = rcu_dereference(mm->exe_file);
1187         if (exe_file && !get_file_rcu(exe_file))
1188                 exe_file = NULL;
1189         rcu_read_unlock();
1190         return exe_file;
1191 }
1192 EXPORT_SYMBOL(get_mm_exe_file);
1193
1194 /**
1195  * get_task_exe_file - acquire a reference to the task's executable file
1196  *
1197  * Returns %NULL if task's mm (if any) has no associated executable file or
1198  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1199  * User must release file via fput().
1200  */
1201 struct file *get_task_exe_file(struct task_struct *task)
1202 {
1203         struct file *exe_file = NULL;
1204         struct mm_struct *mm;
1205
1206         task_lock(task);
1207         mm = task->mm;
1208         if (mm) {
1209                 if (!(task->flags & PF_KTHREAD))
1210                         exe_file = get_mm_exe_file(mm);
1211         }
1212         task_unlock(task);
1213         return exe_file;
1214 }
1215 EXPORT_SYMBOL(get_task_exe_file);
1216
1217 /**
1218  * get_task_mm - acquire a reference to the task's mm
1219  *
1220  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1221  * this kernel workthread has transiently adopted a user mm with use_mm,
1222  * to do its AIO) is not set and if so returns a reference to it, after
1223  * bumping up the use count.  User must release the mm via mmput()
1224  * after use.  Typically used by /proc and ptrace.
1225  */
1226 struct mm_struct *get_task_mm(struct task_struct *task)
1227 {
1228         struct mm_struct *mm;
1229
1230         task_lock(task);
1231         mm = task->mm;
1232         if (mm) {
1233                 if (task->flags & PF_KTHREAD)
1234                         mm = NULL;
1235                 else
1236                         mmget(mm);
1237         }
1238         task_unlock(task);
1239         return mm;
1240 }
1241 EXPORT_SYMBOL_GPL(get_task_mm);
1242
1243 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1244 {
1245         struct mm_struct *mm;
1246         int err;
1247
1248         err =  down_read_killable(&task->signal->exec_update_lock);
1249         if (err)
1250                 return ERR_PTR(err);
1251
1252         mm = get_task_mm(task);
1253         if (mm && mm != current->mm &&
1254                         !ptrace_may_access(task, mode)) {
1255                 mmput(mm);
1256                 mm = ERR_PTR(-EACCES);
1257         }
1258         up_read(&task->signal->exec_update_lock);
1259
1260         return mm;
1261 }
1262
1263 static void complete_vfork_done(struct task_struct *tsk)
1264 {
1265         struct completion *vfork;
1266
1267         task_lock(tsk);
1268         vfork = tsk->vfork_done;
1269         if (likely(vfork)) {
1270                 tsk->vfork_done = NULL;
1271                 complete(vfork);
1272         }
1273         task_unlock(tsk);
1274 }
1275
1276 static int wait_for_vfork_done(struct task_struct *child,
1277                                 struct completion *vfork)
1278 {
1279         int killed;
1280
1281         freezer_do_not_count();
1282         cgroup_enter_frozen();
1283         killed = wait_for_completion_killable(vfork);
1284         cgroup_leave_frozen(false);
1285         freezer_count();
1286
1287         if (killed) {
1288                 task_lock(child);
1289                 child->vfork_done = NULL;
1290                 task_unlock(child);
1291         }
1292
1293         put_task_struct(child);
1294         return killed;
1295 }
1296
1297 /* Please note the differences between mmput and mm_release.
1298  * mmput is called whenever we stop holding onto a mm_struct,
1299  * error success whatever.
1300  *
1301  * mm_release is called after a mm_struct has been removed
1302  * from the current process.
1303  *
1304  * This difference is important for error handling, when we
1305  * only half set up a mm_struct for a new process and need to restore
1306  * the old one.  Because we mmput the new mm_struct before
1307  * restoring the old one. . .
1308  * Eric Biederman 10 January 1998
1309  */
1310 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1311 {
1312         uprobe_free_utask(tsk);
1313
1314         /* Get rid of any cached register state */
1315         deactivate_mm(tsk, mm);
1316
1317         /*
1318          * Signal userspace if we're not exiting with a core dump
1319          * because we want to leave the value intact for debugging
1320          * purposes.
1321          */
1322         if (tsk->clear_child_tid) {
1323                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1324                     atomic_read(&mm->mm_users) > 1) {
1325                         /*
1326                          * We don't check the error code - if userspace has
1327                          * not set up a proper pointer then tough luck.
1328                          */
1329                         put_user(0, tsk->clear_child_tid);
1330                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1331                                         1, NULL, NULL, 0, 0);
1332                 }
1333                 tsk->clear_child_tid = NULL;
1334         }
1335
1336         /*
1337          * All done, finally we can wake up parent and return this mm to him.
1338          * Also kthread_stop() uses this completion for synchronization.
1339          */
1340         if (tsk->vfork_done)
1341                 complete_vfork_done(tsk);
1342 }
1343
1344 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1345 {
1346         futex_exit_release(tsk);
1347         mm_release(tsk, mm);
1348 }
1349
1350 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1351 {
1352         futex_exec_release(tsk);
1353         mm_release(tsk, mm);
1354 }
1355
1356 /**
1357  * dup_mm() - duplicates an existing mm structure
1358  * @tsk: the task_struct with which the new mm will be associated.
1359  * @oldmm: the mm to duplicate.
1360  *
1361  * Allocates a new mm structure and duplicates the provided @oldmm structure
1362  * content into it.
1363  *
1364  * Return: the duplicated mm or NULL on failure.
1365  */
1366 static struct mm_struct *dup_mm(struct task_struct *tsk,
1367                                 struct mm_struct *oldmm)
1368 {
1369         struct mm_struct *mm;
1370         int err;
1371
1372         mm = allocate_mm();
1373         if (!mm)
1374                 goto fail_nomem;
1375
1376         memcpy(mm, oldmm, sizeof(*mm));
1377
1378         if (!mm_init(mm, tsk, mm->user_ns))
1379                 goto fail_nomem;
1380
1381         err = dup_mmap(mm, oldmm);
1382         if (err)
1383                 goto free_pt;
1384
1385         mm->hiwater_rss = get_mm_rss(mm);
1386         mm->hiwater_vm = mm->total_vm;
1387
1388         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1389                 goto free_pt;
1390
1391         return mm;
1392
1393 free_pt:
1394         /* don't put binfmt in mmput, we haven't got module yet */
1395         mm->binfmt = NULL;
1396         mm_init_owner(mm, NULL);
1397         mmput(mm);
1398
1399 fail_nomem:
1400         return NULL;
1401 }
1402
1403 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1404 {
1405         struct mm_struct *mm, *oldmm;
1406
1407         tsk->min_flt = tsk->maj_flt = 0;
1408         tsk->nvcsw = tsk->nivcsw = 0;
1409 #ifdef CONFIG_DETECT_HUNG_TASK
1410         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1411         tsk->last_switch_time = 0;
1412 #endif
1413
1414         tsk->mm = NULL;
1415         tsk->active_mm = NULL;
1416
1417         /*
1418          * Are we cloning a kernel thread?
1419          *
1420          * We need to steal a active VM for that..
1421          */
1422         oldmm = current->mm;
1423         if (!oldmm)
1424                 return 0;
1425
1426         /* initialize the new vmacache entries */
1427         vmacache_flush(tsk);
1428
1429         if (clone_flags & CLONE_VM) {
1430                 mmget(oldmm);
1431                 mm = oldmm;
1432         } else {
1433                 mm = dup_mm(tsk, current->mm);
1434                 if (!mm)
1435                         return -ENOMEM;
1436         }
1437
1438         tsk->mm = mm;
1439         tsk->active_mm = mm;
1440         return 0;
1441 }
1442
1443 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1444 {
1445         struct fs_struct *fs = current->fs;
1446         if (clone_flags & CLONE_FS) {
1447                 /* tsk->fs is already what we want */
1448                 spin_lock(&fs->lock);
1449                 if (fs->in_exec) {
1450                         spin_unlock(&fs->lock);
1451                         return -EAGAIN;
1452                 }
1453                 fs->users++;
1454                 spin_unlock(&fs->lock);
1455                 return 0;
1456         }
1457         tsk->fs = copy_fs_struct(fs);
1458         if (!tsk->fs)
1459                 return -ENOMEM;
1460         return 0;
1461 }
1462
1463 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1464 {
1465         struct files_struct *oldf, *newf;
1466         int error = 0;
1467
1468         /*
1469          * A background process may not have any files ...
1470          */
1471         oldf = current->files;
1472         if (!oldf)
1473                 goto out;
1474
1475         if (clone_flags & CLONE_FILES) {
1476                 atomic_inc(&oldf->count);
1477                 goto out;
1478         }
1479
1480         newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1481         if (!newf)
1482                 goto out;
1483
1484         tsk->files = newf;
1485         error = 0;
1486 out:
1487         return error;
1488 }
1489
1490 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1491 {
1492 #ifdef CONFIG_BLOCK
1493         struct io_context *ioc = current->io_context;
1494         struct io_context *new_ioc;
1495
1496         if (!ioc)
1497                 return 0;
1498         /*
1499          * Share io context with parent, if CLONE_IO is set
1500          */
1501         if (clone_flags & CLONE_IO) {
1502                 ioc_task_link(ioc);
1503                 tsk->io_context = ioc;
1504         } else if (ioprio_valid(ioc->ioprio)) {
1505                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1506                 if (unlikely(!new_ioc))
1507                         return -ENOMEM;
1508
1509                 new_ioc->ioprio = ioc->ioprio;
1510                 put_io_context(new_ioc);
1511         }
1512 #endif
1513         return 0;
1514 }
1515
1516 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1517 {
1518         struct sighand_struct *sig;
1519
1520         if (clone_flags & CLONE_SIGHAND) {
1521                 refcount_inc(&current->sighand->count);
1522                 return 0;
1523         }
1524         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1525         RCU_INIT_POINTER(tsk->sighand, sig);
1526         if (!sig)
1527                 return -ENOMEM;
1528
1529         refcount_set(&sig->count, 1);
1530         spin_lock_irq(&current->sighand->siglock);
1531         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1532         spin_unlock_irq(&current->sighand->siglock);
1533
1534         /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1535         if (clone_flags & CLONE_CLEAR_SIGHAND)
1536                 flush_signal_handlers(tsk, 0);
1537
1538         return 0;
1539 }
1540
1541 void __cleanup_sighand(struct sighand_struct *sighand)
1542 {
1543         if (refcount_dec_and_test(&sighand->count)) {
1544                 signalfd_cleanup(sighand);
1545                 /*
1546                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1547                  * without an RCU grace period, see __lock_task_sighand().
1548                  */
1549                 kmem_cache_free(sighand_cachep, sighand);
1550         }
1551 }
1552
1553 /*
1554  * Initialize POSIX timer handling for a thread group.
1555  */
1556 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1557 {
1558         struct posix_cputimers *pct = &sig->posix_cputimers;
1559         unsigned long cpu_limit;
1560
1561         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1562         posix_cputimers_group_init(pct, cpu_limit);
1563 }
1564
1565 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1566 {
1567         struct signal_struct *sig;
1568
1569         if (clone_flags & CLONE_THREAD)
1570                 return 0;
1571
1572         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1573         tsk->signal = sig;
1574         if (!sig)
1575                 return -ENOMEM;
1576
1577         sig->nr_threads = 1;
1578         atomic_set(&sig->live, 1);
1579         refcount_set(&sig->sigcnt, 1);
1580
1581         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1582         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1583         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1584
1585         init_waitqueue_head(&sig->wait_chldexit);
1586         sig->curr_target = tsk;
1587         init_sigpending(&sig->shared_pending);
1588         INIT_HLIST_HEAD(&sig->multiprocess);
1589         seqlock_init(&sig->stats_lock);
1590         prev_cputime_init(&sig->prev_cputime);
1591
1592 #ifdef CONFIG_POSIX_TIMERS
1593         INIT_LIST_HEAD(&sig->posix_timers);
1594         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1595         sig->real_timer.function = it_real_fn;
1596 #endif
1597
1598         task_lock(current->group_leader);
1599         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1600         task_unlock(current->group_leader);
1601
1602         posix_cpu_timers_init_group(sig);
1603
1604         tty_audit_fork(sig);
1605         sched_autogroup_fork(sig);
1606
1607         sig->oom_score_adj = current->signal->oom_score_adj;
1608         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1609
1610         mutex_init(&sig->cred_guard_mutex);
1611         init_rwsem(&sig->exec_update_lock);
1612
1613         return 0;
1614 }
1615
1616 static void copy_seccomp(struct task_struct *p)
1617 {
1618 #ifdef CONFIG_SECCOMP
1619         /*
1620          * Must be called with sighand->lock held, which is common to
1621          * all threads in the group. Holding cred_guard_mutex is not
1622          * needed because this new task is not yet running and cannot
1623          * be racing exec.
1624          */
1625         assert_spin_locked(&current->sighand->siglock);
1626
1627         /* Ref-count the new filter user, and assign it. */
1628         get_seccomp_filter(current);
1629         p->seccomp = current->seccomp;
1630
1631         /*
1632          * Explicitly enable no_new_privs here in case it got set
1633          * between the task_struct being duplicated and holding the
1634          * sighand lock. The seccomp state and nnp must be in sync.
1635          */
1636         if (task_no_new_privs(current))
1637                 task_set_no_new_privs(p);
1638
1639         /*
1640          * If the parent gained a seccomp mode after copying thread
1641          * flags and between before we held the sighand lock, we have
1642          * to manually enable the seccomp thread flag here.
1643          */
1644         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1645                 set_task_syscall_work(p, SECCOMP);
1646 #endif
1647 }
1648
1649 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1650 {
1651         current->clear_child_tid = tidptr;
1652
1653         return task_pid_vnr(current);
1654 }
1655
1656 static void rt_mutex_init_task(struct task_struct *p)
1657 {
1658         raw_spin_lock_init(&p->pi_lock);
1659 #ifdef CONFIG_RT_MUTEXES
1660         p->pi_waiters = RB_ROOT_CACHED;
1661         p->pi_top_task = NULL;
1662         p->pi_blocked_on = NULL;
1663 #endif
1664 }
1665
1666 static inline void init_task_pid_links(struct task_struct *task)
1667 {
1668         enum pid_type type;
1669
1670         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1671                 INIT_HLIST_NODE(&task->pid_links[type]);
1672 }
1673
1674 static inline void
1675 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1676 {
1677         if (type == PIDTYPE_PID)
1678                 task->thread_pid = pid;
1679         else
1680                 task->signal->pids[type] = pid;
1681 }
1682
1683 static inline void rcu_copy_process(struct task_struct *p)
1684 {
1685 #ifdef CONFIG_PREEMPT_RCU
1686         p->rcu_read_lock_nesting = 0;
1687         p->rcu_read_unlock_special.s = 0;
1688         p->rcu_blocked_node = NULL;
1689         INIT_LIST_HEAD(&p->rcu_node_entry);
1690 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1691 #ifdef CONFIG_TASKS_RCU
1692         p->rcu_tasks_holdout = false;
1693         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1694         p->rcu_tasks_idle_cpu = -1;
1695 #endif /* #ifdef CONFIG_TASKS_RCU */
1696 #ifdef CONFIG_TASKS_TRACE_RCU
1697         p->trc_reader_nesting = 0;
1698         p->trc_reader_special.s = 0;
1699         INIT_LIST_HEAD(&p->trc_holdout_list);
1700 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1701 }
1702
1703 struct pid *pidfd_pid(const struct file *file)
1704 {
1705         if (file->f_op == &pidfd_fops)
1706                 return file->private_data;
1707
1708         return ERR_PTR(-EBADF);
1709 }
1710
1711 static int pidfd_release(struct inode *inode, struct file *file)
1712 {
1713         struct pid *pid = file->private_data;
1714
1715         file->private_data = NULL;
1716         put_pid(pid);
1717         return 0;
1718 }
1719
1720 #ifdef CONFIG_PROC_FS
1721 /**
1722  * pidfd_show_fdinfo - print information about a pidfd
1723  * @m: proc fdinfo file
1724  * @f: file referencing a pidfd
1725  *
1726  * Pid:
1727  * This function will print the pid that a given pidfd refers to in the
1728  * pid namespace of the procfs instance.
1729  * If the pid namespace of the process is not a descendant of the pid
1730  * namespace of the procfs instance 0 will be shown as its pid. This is
1731  * similar to calling getppid() on a process whose parent is outside of
1732  * its pid namespace.
1733  *
1734  * NSpid:
1735  * If pid namespaces are supported then this function will also print
1736  * the pid of a given pidfd refers to for all descendant pid namespaces
1737  * starting from the current pid namespace of the instance, i.e. the
1738  * Pid field and the first entry in the NSpid field will be identical.
1739  * If the pid namespace of the process is not a descendant of the pid
1740  * namespace of the procfs instance 0 will be shown as its first NSpid
1741  * entry and no others will be shown.
1742  * Note that this differs from the Pid and NSpid fields in
1743  * /proc/<pid>/status where Pid and NSpid are always shown relative to
1744  * the  pid namespace of the procfs instance. The difference becomes
1745  * obvious when sending around a pidfd between pid namespaces from a
1746  * different branch of the tree, i.e. where no ancestral relation is
1747  * present between the pid namespaces:
1748  * - create two new pid namespaces ns1 and ns2 in the initial pid
1749  *   namespace (also take care to create new mount namespaces in the
1750  *   new pid namespace and mount procfs)
1751  * - create a process with a pidfd in ns1
1752  * - send pidfd from ns1 to ns2
1753  * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1754  *   have exactly one entry, which is 0
1755  */
1756 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1757 {
1758         struct pid *pid = f->private_data;
1759         struct pid_namespace *ns;
1760         pid_t nr = -1;
1761
1762         if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1763                 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1764                 nr = pid_nr_ns(pid, ns);
1765         }
1766
1767         seq_put_decimal_ll(m, "Pid:\t", nr);
1768
1769 #ifdef CONFIG_PID_NS
1770         seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1771         if (nr > 0) {
1772                 int i;
1773
1774                 /* If nr is non-zero it means that 'pid' is valid and that
1775                  * ns, i.e. the pid namespace associated with the procfs
1776                  * instance, is in the pid namespace hierarchy of pid.
1777                  * Start at one below the already printed level.
1778                  */
1779                 for (i = ns->level + 1; i <= pid->level; i++)
1780                         seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1781         }
1782 #endif
1783         seq_putc(m, '\n');
1784 }
1785 #endif
1786
1787 /*
1788  * Poll support for process exit notification.
1789  */
1790 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1791 {
1792         struct pid *pid = file->private_data;
1793         __poll_t poll_flags = 0;
1794
1795         poll_wait(file, &pid->wait_pidfd, pts);
1796
1797         /*
1798          * Inform pollers only when the whole thread group exits.
1799          * If the thread group leader exits before all other threads in the
1800          * group, then poll(2) should block, similar to the wait(2) family.
1801          */
1802         if (thread_group_exited(pid))
1803                 poll_flags = EPOLLIN | EPOLLRDNORM;
1804
1805         return poll_flags;
1806 }
1807
1808 const struct file_operations pidfd_fops = {
1809         .release = pidfd_release,
1810         .poll = pidfd_poll,
1811 #ifdef CONFIG_PROC_FS
1812         .show_fdinfo = pidfd_show_fdinfo,
1813 #endif
1814 };
1815
1816 static void __delayed_free_task(struct rcu_head *rhp)
1817 {
1818         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1819
1820         free_task(tsk);
1821 }
1822
1823 static __always_inline void delayed_free_task(struct task_struct *tsk)
1824 {
1825         if (IS_ENABLED(CONFIG_MEMCG))
1826                 call_rcu(&tsk->rcu, __delayed_free_task);
1827         else
1828                 free_task(tsk);
1829 }
1830
1831 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1832 {
1833         /* Skip if kernel thread */
1834         if (!tsk->mm)
1835                 return;
1836
1837         /* Skip if spawning a thread or using vfork */
1838         if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1839                 return;
1840
1841         /* We need to synchronize with __set_oom_adj */
1842         mutex_lock(&oom_adj_mutex);
1843         set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1844         /* Update the values in case they were changed after copy_signal */
1845         tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1846         tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1847         mutex_unlock(&oom_adj_mutex);
1848 }
1849
1850 /*
1851  * This creates a new process as a copy of the old one,
1852  * but does not actually start it yet.
1853  *
1854  * It copies the registers, and all the appropriate
1855  * parts of the process environment (as per the clone
1856  * flags). The actual kick-off is left to the caller.
1857  */
1858 static __latent_entropy struct task_struct *copy_process(
1859                                         struct pid *pid,
1860                                         int trace,
1861                                         int node,
1862                                         struct kernel_clone_args *args)
1863 {
1864         int pidfd = -1, retval;
1865         struct task_struct *p;
1866         struct multiprocess_signals delayed;
1867         struct file *pidfile = NULL;
1868         u64 clone_flags = args->flags;
1869         struct nsproxy *nsp = current->nsproxy;
1870
1871         /*
1872          * Don't allow sharing the root directory with processes in a different
1873          * namespace
1874          */
1875         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1876                 return ERR_PTR(-EINVAL);
1877
1878         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1879                 return ERR_PTR(-EINVAL);
1880
1881         /*
1882          * Thread groups must share signals as well, and detached threads
1883          * can only be started up within the thread group.
1884          */
1885         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1886                 return ERR_PTR(-EINVAL);
1887
1888         /*
1889          * Shared signal handlers imply shared VM. By way of the above,
1890          * thread groups also imply shared VM. Blocking this case allows
1891          * for various simplifications in other code.
1892          */
1893         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1894                 return ERR_PTR(-EINVAL);
1895
1896         /*
1897          * Siblings of global init remain as zombies on exit since they are
1898          * not reaped by their parent (swapper). To solve this and to avoid
1899          * multi-rooted process trees, prevent global and container-inits
1900          * from creating siblings.
1901          */
1902         if ((clone_flags & CLONE_PARENT) &&
1903                                 current->signal->flags & SIGNAL_UNKILLABLE)
1904                 return ERR_PTR(-EINVAL);
1905
1906         /*
1907          * If the new process will be in a different pid or user namespace
1908          * do not allow it to share a thread group with the forking task.
1909          */
1910         if (clone_flags & CLONE_THREAD) {
1911                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1912                     (task_active_pid_ns(current) != nsp->pid_ns_for_children))
1913                         return ERR_PTR(-EINVAL);
1914         }
1915
1916         /*
1917          * If the new process will be in a different time namespace
1918          * do not allow it to share VM or a thread group with the forking task.
1919          */
1920         if (clone_flags & (CLONE_THREAD | CLONE_VM)) {
1921                 if (nsp->time_ns != nsp->time_ns_for_children)
1922                         return ERR_PTR(-EINVAL);
1923         }
1924
1925         if (clone_flags & CLONE_PIDFD) {
1926                 /*
1927                  * - CLONE_DETACHED is blocked so that we can potentially
1928                  *   reuse it later for CLONE_PIDFD.
1929                  * - CLONE_THREAD is blocked until someone really needs it.
1930                  */
1931                 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1932                         return ERR_PTR(-EINVAL);
1933         }
1934
1935         /*
1936          * Force any signals received before this point to be delivered
1937          * before the fork happens.  Collect up signals sent to multiple
1938          * processes that happen during the fork and delay them so that
1939          * they appear to happen after the fork.
1940          */
1941         sigemptyset(&delayed.signal);
1942         INIT_HLIST_NODE(&delayed.node);
1943
1944         spin_lock_irq(&current->sighand->siglock);
1945         if (!(clone_flags & CLONE_THREAD))
1946                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1947         recalc_sigpending();
1948         spin_unlock_irq(&current->sighand->siglock);
1949         retval = -ERESTARTNOINTR;
1950         if (task_sigpending(current))
1951                 goto fork_out;
1952
1953         retval = -ENOMEM;
1954         p = dup_task_struct(current, node);
1955         if (!p)
1956                 goto fork_out;
1957         if (args->io_thread) {
1958                 /*
1959                  * Mark us an IO worker, and block any signal that isn't
1960                  * fatal or STOP
1961                  */
1962                 p->flags |= PF_IO_WORKER;
1963                 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
1964         }
1965
1966         /*
1967          * This _must_ happen before we call free_task(), i.e. before we jump
1968          * to any of the bad_fork_* labels. This is to avoid freeing
1969          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1970          * kernel threads (PF_KTHREAD).
1971          */
1972         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1973         /*
1974          * Clear TID on mm_release()?
1975          */
1976         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1977
1978         ftrace_graph_init_task(p);
1979
1980         rt_mutex_init_task(p);
1981
1982         lockdep_assert_irqs_enabled();
1983 #ifdef CONFIG_PROVE_LOCKING
1984         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1985 #endif
1986         retval = -EAGAIN;
1987         if (is_ucounts_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
1988                 if (p->real_cred->user != INIT_USER &&
1989                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1990                         goto bad_fork_free;
1991         }
1992         current->flags &= ~PF_NPROC_EXCEEDED;
1993
1994         retval = copy_creds(p, clone_flags);
1995         if (retval < 0)
1996                 goto bad_fork_free;
1997
1998         /*
1999          * If multiple threads are within copy_process(), then this check
2000          * triggers too late. This doesn't hurt, the check is only there
2001          * to stop root fork bombs.
2002          */
2003         retval = -EAGAIN;
2004         if (data_race(nr_threads >= max_threads))
2005                 goto bad_fork_cleanup_count;
2006
2007         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
2008         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2009         p->flags |= PF_FORKNOEXEC;
2010         INIT_LIST_HEAD(&p->children);
2011         INIT_LIST_HEAD(&p->sibling);
2012         rcu_copy_process(p);
2013         p->vfork_done = NULL;
2014         spin_lock_init(&p->alloc_lock);
2015
2016         init_sigpending(&p->pending);
2017
2018         p->utime = p->stime = p->gtime = 0;
2019 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2020         p->utimescaled = p->stimescaled = 0;
2021 #endif
2022         prev_cputime_init(&p->prev_cputime);
2023
2024 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2025         seqcount_init(&p->vtime.seqcount);
2026         p->vtime.starttime = 0;
2027         p->vtime.state = VTIME_INACTIVE;
2028 #endif
2029
2030 #ifdef CONFIG_IO_URING
2031         p->io_uring = NULL;
2032 #endif
2033
2034 #if defined(SPLIT_RSS_COUNTING)
2035         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2036 #endif
2037
2038         p->default_timer_slack_ns = current->timer_slack_ns;
2039
2040 #ifdef CONFIG_PSI
2041         p->psi_flags = 0;
2042 #endif
2043
2044         task_io_accounting_init(&p->ioac);
2045         acct_clear_integrals(p);
2046
2047         posix_cputimers_init(&p->posix_cputimers);
2048
2049         p->io_context = NULL;
2050         audit_set_context(p, NULL);
2051         cgroup_fork(p);
2052 #ifdef CONFIG_NUMA
2053         p->mempolicy = mpol_dup(p->mempolicy);
2054         if (IS_ERR(p->mempolicy)) {
2055                 retval = PTR_ERR(p->mempolicy);
2056                 p->mempolicy = NULL;
2057                 goto bad_fork_cleanup_threadgroup_lock;
2058         }
2059 #endif
2060 #ifdef CONFIG_CPUSETS
2061         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2062         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2063         seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2064 #endif
2065 #ifdef CONFIG_TRACE_IRQFLAGS
2066         memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2067         p->irqtrace.hardirq_disable_ip  = _THIS_IP_;
2068         p->irqtrace.softirq_enable_ip   = _THIS_IP_;
2069         p->softirqs_enabled             = 1;
2070         p->softirq_context              = 0;
2071 #endif
2072
2073         p->pagefault_disabled = 0;
2074
2075 #ifdef CONFIG_LOCKDEP
2076         lockdep_init_task(p);
2077 #endif
2078
2079 #ifdef CONFIG_DEBUG_MUTEXES
2080         p->blocked_on = NULL; /* not blocked yet */
2081 #endif
2082 #ifdef CONFIG_BCACHE
2083         p->sequential_io        = 0;
2084         p->sequential_io_avg    = 0;
2085 #endif
2086 #ifdef CONFIG_BPF_SYSCALL
2087         RCU_INIT_POINTER(p->bpf_storage, NULL);
2088         p->bpf_ctx = NULL;
2089 #endif
2090
2091         /* Perform scheduler related setup. Assign this task to a CPU. */
2092         retval = sched_fork(clone_flags, p);
2093         if (retval)
2094                 goto bad_fork_cleanup_policy;
2095
2096         retval = perf_event_init_task(p, clone_flags);
2097         if (retval)
2098                 goto bad_fork_cleanup_policy;
2099         retval = audit_alloc(p);
2100         if (retval)
2101                 goto bad_fork_cleanup_perf;
2102         /* copy all the process information */
2103         shm_init_task(p);
2104         retval = security_task_alloc(p, clone_flags);
2105         if (retval)
2106                 goto bad_fork_cleanup_audit;
2107         retval = copy_semundo(clone_flags, p);
2108         if (retval)
2109                 goto bad_fork_cleanup_security;
2110         retval = copy_files(clone_flags, p);
2111         if (retval)
2112                 goto bad_fork_cleanup_semundo;
2113         retval = copy_fs(clone_flags, p);
2114         if (retval)
2115                 goto bad_fork_cleanup_files;
2116         retval = copy_sighand(clone_flags, p);
2117         if (retval)
2118                 goto bad_fork_cleanup_fs;
2119         retval = copy_signal(clone_flags, p);
2120         if (retval)
2121                 goto bad_fork_cleanup_sighand;
2122         retval = copy_mm(clone_flags, p);
2123         if (retval)
2124                 goto bad_fork_cleanup_signal;
2125         retval = copy_namespaces(clone_flags, p);
2126         if (retval)
2127                 goto bad_fork_cleanup_mm;
2128         retval = copy_io(clone_flags, p);
2129         if (retval)
2130                 goto bad_fork_cleanup_namespaces;
2131         retval = copy_thread(clone_flags, args->stack, args->stack_size, p, args->tls);
2132         if (retval)
2133                 goto bad_fork_cleanup_io;
2134
2135         stackleak_task_init(p);
2136
2137         if (pid != &init_struct_pid) {
2138                 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2139                                 args->set_tid_size);
2140                 if (IS_ERR(pid)) {
2141                         retval = PTR_ERR(pid);
2142                         goto bad_fork_cleanup_thread;
2143                 }
2144         }
2145
2146         /*
2147          * This has to happen after we've potentially unshared the file
2148          * descriptor table (so that the pidfd doesn't leak into the child
2149          * if the fd table isn't shared).
2150          */
2151         if (clone_flags & CLONE_PIDFD) {
2152                 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2153                 if (retval < 0)
2154                         goto bad_fork_free_pid;
2155
2156                 pidfd = retval;
2157
2158                 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2159                                               O_RDWR | O_CLOEXEC);
2160                 if (IS_ERR(pidfile)) {
2161                         put_unused_fd(pidfd);
2162                         retval = PTR_ERR(pidfile);
2163                         goto bad_fork_free_pid;
2164                 }
2165                 get_pid(pid);   /* held by pidfile now */
2166
2167                 retval = put_user(pidfd, args->pidfd);
2168                 if (retval)
2169                         goto bad_fork_put_pidfd;
2170         }
2171
2172 #ifdef CONFIG_BLOCK
2173         p->plug = NULL;
2174 #endif
2175         futex_init_task(p);
2176
2177         /*
2178          * sigaltstack should be cleared when sharing the same VM
2179          */
2180         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2181                 sas_ss_reset(p);
2182
2183         /*
2184          * Syscall tracing and stepping should be turned off in the
2185          * child regardless of CLONE_PTRACE.
2186          */
2187         user_disable_single_step(p);
2188         clear_task_syscall_work(p, SYSCALL_TRACE);
2189 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2190         clear_task_syscall_work(p, SYSCALL_EMU);
2191 #endif
2192         clear_tsk_latency_tracing(p);
2193
2194         /* ok, now we should be set up.. */
2195         p->pid = pid_nr(pid);
2196         if (clone_flags & CLONE_THREAD) {
2197                 p->group_leader = current->group_leader;
2198                 p->tgid = current->tgid;
2199         } else {
2200                 p->group_leader = p;
2201                 p->tgid = p->pid;
2202         }
2203
2204         p->nr_dirtied = 0;
2205         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2206         p->dirty_paused_when = 0;
2207
2208         p->pdeath_signal = 0;
2209         INIT_LIST_HEAD(&p->thread_group);
2210         p->task_works = NULL;
2211
2212 #ifdef CONFIG_KRETPROBES
2213         p->kretprobe_instances.first = NULL;
2214 #endif
2215
2216         /*
2217          * Ensure that the cgroup subsystem policies allow the new process to be
2218          * forked. It should be noted that the new process's css_set can be changed
2219          * between here and cgroup_post_fork() if an organisation operation is in
2220          * progress.
2221          */
2222         retval = cgroup_can_fork(p, args);
2223         if (retval)
2224                 goto bad_fork_put_pidfd;
2225
2226         /*
2227          * From this point on we must avoid any synchronous user-space
2228          * communication until we take the tasklist-lock. In particular, we do
2229          * not want user-space to be able to predict the process start-time by
2230          * stalling fork(2) after we recorded the start_time but before it is
2231          * visible to the system.
2232          */
2233
2234         p->start_time = ktime_get_ns();
2235         p->start_boottime = ktime_get_boottime_ns();
2236
2237         /*
2238          * Make it visible to the rest of the system, but dont wake it up yet.
2239          * Need tasklist lock for parent etc handling!
2240          */
2241         write_lock_irq(&tasklist_lock);
2242
2243         /* CLONE_PARENT re-uses the old parent */
2244         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2245                 p->real_parent = current->real_parent;
2246                 p->parent_exec_id = current->parent_exec_id;
2247                 if (clone_flags & CLONE_THREAD)
2248                         p->exit_signal = -1;
2249                 else
2250                         p->exit_signal = current->group_leader->exit_signal;
2251         } else {
2252                 p->real_parent = current;
2253                 p->parent_exec_id = current->self_exec_id;
2254                 p->exit_signal = args->exit_signal;
2255         }
2256
2257         klp_copy_process(p);
2258
2259         sched_core_fork(p);
2260
2261         spin_lock(&current->sighand->siglock);
2262
2263         /*
2264          * Copy seccomp details explicitly here, in case they were changed
2265          * before holding sighand lock.
2266          */
2267         copy_seccomp(p);
2268
2269         rseq_fork(p, clone_flags);
2270
2271         /* Don't start children in a dying pid namespace */
2272         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2273                 retval = -ENOMEM;
2274                 goto bad_fork_cancel_cgroup;
2275         }
2276
2277         /* Let kill terminate clone/fork in the middle */
2278         if (fatal_signal_pending(current)) {
2279                 retval = -EINTR;
2280                 goto bad_fork_cancel_cgroup;
2281         }
2282
2283         /* past the last point of failure */
2284         if (pidfile)
2285                 fd_install(pidfd, pidfile);
2286
2287         init_task_pid_links(p);
2288         if (likely(p->pid)) {
2289                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2290
2291                 init_task_pid(p, PIDTYPE_PID, pid);
2292                 if (thread_group_leader(p)) {
2293                         init_task_pid(p, PIDTYPE_TGID, pid);
2294                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2295                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2296
2297                         if (is_child_reaper(pid)) {
2298                                 ns_of_pid(pid)->child_reaper = p;
2299                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2300                         }
2301                         p->signal->shared_pending.signal = delayed.signal;
2302                         p->signal->tty = tty_kref_get(current->signal->tty);
2303                         /*
2304                          * Inherit has_child_subreaper flag under the same
2305                          * tasklist_lock with adding child to the process tree
2306                          * for propagate_has_child_subreaper optimization.
2307                          */
2308                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2309                                                          p->real_parent->signal->is_child_subreaper;
2310                         list_add_tail(&p->sibling, &p->real_parent->children);
2311                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2312                         attach_pid(p, PIDTYPE_TGID);
2313                         attach_pid(p, PIDTYPE_PGID);
2314                         attach_pid(p, PIDTYPE_SID);
2315                         __this_cpu_inc(process_counts);
2316                 } else {
2317                         current->signal->nr_threads++;
2318                         atomic_inc(&current->signal->live);
2319                         refcount_inc(&current->signal->sigcnt);
2320                         task_join_group_stop(p);
2321                         list_add_tail_rcu(&p->thread_group,
2322                                           &p->group_leader->thread_group);
2323                         list_add_tail_rcu(&p->thread_node,
2324                                           &p->signal->thread_head);
2325                 }
2326                 attach_pid(p, PIDTYPE_PID);
2327                 nr_threads++;
2328         }
2329         total_forks++;
2330         hlist_del_init(&delayed.node);
2331         spin_unlock(&current->sighand->siglock);
2332         syscall_tracepoint_update(p);
2333         write_unlock_irq(&tasklist_lock);
2334
2335         proc_fork_connector(p);
2336         sched_post_fork(p);
2337         cgroup_post_fork(p, args);
2338         perf_event_fork(p);
2339
2340         trace_task_newtask(p, clone_flags);
2341         uprobe_copy_process(p, clone_flags);
2342
2343         copy_oom_score_adj(clone_flags, p);
2344
2345         return p;
2346
2347 bad_fork_cancel_cgroup:
2348         sched_core_free(p);
2349         spin_unlock(&current->sighand->siglock);
2350         write_unlock_irq(&tasklist_lock);
2351         cgroup_cancel_fork(p, args);
2352 bad_fork_put_pidfd:
2353         if (clone_flags & CLONE_PIDFD) {
2354                 fput(pidfile);
2355                 put_unused_fd(pidfd);
2356         }
2357 bad_fork_free_pid:
2358         if (pid != &init_struct_pid)
2359                 free_pid(pid);
2360 bad_fork_cleanup_thread:
2361         exit_thread(p);
2362 bad_fork_cleanup_io:
2363         if (p->io_context)
2364                 exit_io_context(p);
2365 bad_fork_cleanup_namespaces:
2366         exit_task_namespaces(p);
2367 bad_fork_cleanup_mm:
2368         if (p->mm) {
2369                 mm_clear_owner(p->mm, p);
2370                 mmput(p->mm);
2371         }
2372 bad_fork_cleanup_signal:
2373         if (!(clone_flags & CLONE_THREAD))
2374                 free_signal_struct(p->signal);
2375 bad_fork_cleanup_sighand:
2376         __cleanup_sighand(p->sighand);
2377 bad_fork_cleanup_fs:
2378         exit_fs(p); /* blocking */
2379 bad_fork_cleanup_files:
2380         exit_files(p); /* blocking */
2381 bad_fork_cleanup_semundo:
2382         exit_sem(p);
2383 bad_fork_cleanup_security:
2384         security_task_free(p);
2385 bad_fork_cleanup_audit:
2386         audit_free(p);
2387 bad_fork_cleanup_perf:
2388         perf_event_free_task(p);
2389 bad_fork_cleanup_policy:
2390         lockdep_free_task(p);
2391 #ifdef CONFIG_NUMA
2392         mpol_put(p->mempolicy);
2393 bad_fork_cleanup_threadgroup_lock:
2394 #endif
2395         delayacct_tsk_free(p);
2396 bad_fork_cleanup_count:
2397         dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2398         exit_creds(p);
2399 bad_fork_free:
2400         WRITE_ONCE(p->__state, TASK_DEAD);
2401         put_task_stack(p);
2402         delayed_free_task(p);
2403 fork_out:
2404         spin_lock_irq(&current->sighand->siglock);
2405         hlist_del_init(&delayed.node);
2406         spin_unlock_irq(&current->sighand->siglock);
2407         return ERR_PTR(retval);
2408 }
2409
2410 static inline void init_idle_pids(struct task_struct *idle)
2411 {
2412         enum pid_type type;
2413
2414         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2415                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2416                 init_task_pid(idle, type, &init_struct_pid);
2417         }
2418 }
2419
2420 struct task_struct * __init fork_idle(int cpu)
2421 {
2422         struct task_struct *task;
2423         struct kernel_clone_args args = {
2424                 .flags = CLONE_VM,
2425         };
2426
2427         task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2428         if (!IS_ERR(task)) {
2429                 init_idle_pids(task);
2430                 init_idle(task, cpu);
2431         }
2432
2433         return task;
2434 }
2435
2436 struct mm_struct *copy_init_mm(void)
2437 {
2438         return dup_mm(NULL, &init_mm);
2439 }
2440
2441 /*
2442  * This is like kernel_clone(), but shaved down and tailored to just
2443  * creating io_uring workers. It returns a created task, or an error pointer.
2444  * The returned task is inactive, and the caller must fire it up through
2445  * wake_up_new_task(p). All signals are blocked in the created task.
2446  */
2447 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2448 {
2449         unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2450                                 CLONE_IO;
2451         struct kernel_clone_args args = {
2452                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2453                                     CLONE_UNTRACED) & ~CSIGNAL),
2454                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2455                 .stack          = (unsigned long)fn,
2456                 .stack_size     = (unsigned long)arg,
2457                 .io_thread      = 1,
2458         };
2459
2460         return copy_process(NULL, 0, node, &args);
2461 }
2462
2463 /*
2464  *  Ok, this is the main fork-routine.
2465  *
2466  * It copies the process, and if successful kick-starts
2467  * it and waits for it to finish using the VM if required.
2468  *
2469  * args->exit_signal is expected to be checked for sanity by the caller.
2470  */
2471 pid_t kernel_clone(struct kernel_clone_args *args)
2472 {
2473         u64 clone_flags = args->flags;
2474         struct completion vfork;
2475         struct pid *pid;
2476         struct task_struct *p;
2477         int trace = 0;
2478         pid_t nr;
2479
2480         /*
2481          * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2482          * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2483          * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2484          * field in struct clone_args and it still doesn't make sense to have
2485          * them both point at the same memory location. Performing this check
2486          * here has the advantage that we don't need to have a separate helper
2487          * to check for legacy clone().
2488          */
2489         if ((args->flags & CLONE_PIDFD) &&
2490             (args->flags & CLONE_PARENT_SETTID) &&
2491             (args->pidfd == args->parent_tid))
2492                 return -EINVAL;
2493
2494         /*
2495          * Determine whether and which event to report to ptracer.  When
2496          * called from kernel_thread or CLONE_UNTRACED is explicitly
2497          * requested, no event is reported; otherwise, report if the event
2498          * for the type of forking is enabled.
2499          */
2500         if (!(clone_flags & CLONE_UNTRACED)) {
2501                 if (clone_flags & CLONE_VFORK)
2502                         trace = PTRACE_EVENT_VFORK;
2503                 else if (args->exit_signal != SIGCHLD)
2504                         trace = PTRACE_EVENT_CLONE;
2505                 else
2506                         trace = PTRACE_EVENT_FORK;
2507
2508                 if (likely(!ptrace_event_enabled(current, trace)))
2509                         trace = 0;
2510         }
2511
2512         p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2513         add_latent_entropy();
2514
2515         if (IS_ERR(p))
2516                 return PTR_ERR(p);
2517
2518         /*
2519          * Do this prior waking up the new thread - the thread pointer
2520          * might get invalid after that point, if the thread exits quickly.
2521          */
2522         trace_sched_process_fork(current, p);
2523
2524         pid = get_task_pid(p, PIDTYPE_PID);
2525         nr = pid_vnr(pid);
2526
2527         if (clone_flags & CLONE_PARENT_SETTID)
2528                 put_user(nr, args->parent_tid);
2529
2530         if (clone_flags & CLONE_VFORK) {
2531                 p->vfork_done = &vfork;
2532                 init_completion(&vfork);
2533                 get_task_struct(p);
2534         }
2535
2536         wake_up_new_task(p);
2537
2538         /* forking complete and child started to run, tell ptracer */
2539         if (unlikely(trace))
2540                 ptrace_event_pid(trace, pid);
2541
2542         if (clone_flags & CLONE_VFORK) {
2543                 if (!wait_for_vfork_done(p, &vfork))
2544                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2545         }
2546
2547         put_pid(pid);
2548         return nr;
2549 }
2550
2551 /*
2552  * Create a kernel thread.
2553  */
2554 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2555 {
2556         struct kernel_clone_args args = {
2557                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2558                                     CLONE_UNTRACED) & ~CSIGNAL),
2559                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2560                 .stack          = (unsigned long)fn,
2561                 .stack_size     = (unsigned long)arg,
2562         };
2563
2564         return kernel_clone(&args);
2565 }
2566
2567 #ifdef __ARCH_WANT_SYS_FORK
2568 SYSCALL_DEFINE0(fork)
2569 {
2570 #ifdef CONFIG_MMU
2571         struct kernel_clone_args args = {
2572                 .exit_signal = SIGCHLD,
2573         };
2574
2575         return kernel_clone(&args);
2576 #else
2577         /* can not support in nommu mode */
2578         return -EINVAL;
2579 #endif
2580 }
2581 #endif
2582
2583 #ifdef __ARCH_WANT_SYS_VFORK
2584 SYSCALL_DEFINE0(vfork)
2585 {
2586         struct kernel_clone_args args = {
2587                 .flags          = CLONE_VFORK | CLONE_VM,
2588                 .exit_signal    = SIGCHLD,
2589         };
2590
2591         return kernel_clone(&args);
2592 }
2593 #endif
2594
2595 #ifdef __ARCH_WANT_SYS_CLONE
2596 #ifdef CONFIG_CLONE_BACKWARDS
2597 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2598                  int __user *, parent_tidptr,
2599                  unsigned long, tls,
2600                  int __user *, child_tidptr)
2601 #elif defined(CONFIG_CLONE_BACKWARDS2)
2602 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2603                  int __user *, parent_tidptr,
2604                  int __user *, child_tidptr,
2605                  unsigned long, tls)
2606 #elif defined(CONFIG_CLONE_BACKWARDS3)
2607 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2608                 int, stack_size,
2609                 int __user *, parent_tidptr,
2610                 int __user *, child_tidptr,
2611                 unsigned long, tls)
2612 #else
2613 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2614                  int __user *, parent_tidptr,
2615                  int __user *, child_tidptr,
2616                  unsigned long, tls)
2617 #endif
2618 {
2619         struct kernel_clone_args args = {
2620                 .flags          = (lower_32_bits(clone_flags) & ~CSIGNAL),
2621                 .pidfd          = parent_tidptr,
2622                 .child_tid      = child_tidptr,
2623                 .parent_tid     = parent_tidptr,
2624                 .exit_signal    = (lower_32_bits(clone_flags) & CSIGNAL),
2625                 .stack          = newsp,
2626                 .tls            = tls,
2627         };
2628
2629         return kernel_clone(&args);
2630 }
2631 #endif
2632
2633 #ifdef __ARCH_WANT_SYS_CLONE3
2634
2635 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2636                                               struct clone_args __user *uargs,
2637                                               size_t usize)
2638 {
2639         int err;
2640         struct clone_args args;
2641         pid_t *kset_tid = kargs->set_tid;
2642
2643         BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2644                      CLONE_ARGS_SIZE_VER0);
2645         BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2646                      CLONE_ARGS_SIZE_VER1);
2647         BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2648                      CLONE_ARGS_SIZE_VER2);
2649         BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2650
2651         if (unlikely(usize > PAGE_SIZE))
2652                 return -E2BIG;
2653         if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2654                 return -EINVAL;
2655
2656         err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2657         if (err)
2658                 return err;
2659
2660         if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2661                 return -EINVAL;
2662
2663         if (unlikely(!args.set_tid && args.set_tid_size > 0))
2664                 return -EINVAL;
2665
2666         if (unlikely(args.set_tid && args.set_tid_size == 0))
2667                 return -EINVAL;
2668
2669         /*
2670          * Verify that higher 32bits of exit_signal are unset and that
2671          * it is a valid signal
2672          */
2673         if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2674                      !valid_signal(args.exit_signal)))
2675                 return -EINVAL;
2676
2677         if ((args.flags & CLONE_INTO_CGROUP) &&
2678             (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2679                 return -EINVAL;
2680
2681         *kargs = (struct kernel_clone_args){
2682                 .flags          = args.flags,
2683                 .pidfd          = u64_to_user_ptr(args.pidfd),
2684                 .child_tid      = u64_to_user_ptr(args.child_tid),
2685                 .parent_tid     = u64_to_user_ptr(args.parent_tid),
2686                 .exit_signal    = args.exit_signal,
2687                 .stack          = args.stack,
2688                 .stack_size     = args.stack_size,
2689                 .tls            = args.tls,
2690                 .set_tid_size   = args.set_tid_size,
2691                 .cgroup         = args.cgroup,
2692         };
2693
2694         if (args.set_tid &&
2695                 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2696                         (kargs->set_tid_size * sizeof(pid_t))))
2697                 return -EFAULT;
2698
2699         kargs->set_tid = kset_tid;
2700
2701         return 0;
2702 }
2703
2704 /**
2705  * clone3_stack_valid - check and prepare stack
2706  * @kargs: kernel clone args
2707  *
2708  * Verify that the stack arguments userspace gave us are sane.
2709  * In addition, set the stack direction for userspace since it's easy for us to
2710  * determine.
2711  */
2712 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2713 {
2714         if (kargs->stack == 0) {
2715                 if (kargs->stack_size > 0)
2716                         return false;
2717         } else {
2718                 if (kargs->stack_size == 0)
2719                         return false;
2720
2721                 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2722                         return false;
2723
2724 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2725                 kargs->stack += kargs->stack_size;
2726 #endif
2727         }
2728
2729         return true;
2730 }
2731
2732 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2733 {
2734         /* Verify that no unknown flags are passed along. */
2735         if (kargs->flags &
2736             ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2737                 return false;
2738
2739         /*
2740          * - make the CLONE_DETACHED bit reusable for clone3
2741          * - make the CSIGNAL bits reusable for clone3
2742          */
2743         if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2744                 return false;
2745
2746         if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2747             (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2748                 return false;
2749
2750         if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2751             kargs->exit_signal)
2752                 return false;
2753
2754         if (!clone3_stack_valid(kargs))
2755                 return false;
2756
2757         return true;
2758 }
2759
2760 /**
2761  * clone3 - create a new process with specific properties
2762  * @uargs: argument structure
2763  * @size:  size of @uargs
2764  *
2765  * clone3() is the extensible successor to clone()/clone2().
2766  * It takes a struct as argument that is versioned by its size.
2767  *
2768  * Return: On success, a positive PID for the child process.
2769  *         On error, a negative errno number.
2770  */
2771 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2772 {
2773         int err;
2774
2775         struct kernel_clone_args kargs;
2776         pid_t set_tid[MAX_PID_NS_LEVEL];
2777
2778         kargs.set_tid = set_tid;
2779
2780         err = copy_clone_args_from_user(&kargs, uargs, size);
2781         if (err)
2782                 return err;
2783
2784         if (!clone3_args_valid(&kargs))
2785                 return -EINVAL;
2786
2787         return kernel_clone(&kargs);
2788 }
2789 #endif
2790
2791 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2792 {
2793         struct task_struct *leader, *parent, *child;
2794         int res;
2795
2796         read_lock(&tasklist_lock);
2797         leader = top = top->group_leader;
2798 down:
2799         for_each_thread(leader, parent) {
2800                 list_for_each_entry(child, &parent->children, sibling) {
2801                         res = visitor(child, data);
2802                         if (res) {
2803                                 if (res < 0)
2804                                         goto out;
2805                                 leader = child;
2806                                 goto down;
2807                         }
2808 up:
2809                         ;
2810                 }
2811         }
2812
2813         if (leader != top) {
2814                 child = leader;
2815                 parent = child->real_parent;
2816                 leader = parent->group_leader;
2817                 goto up;
2818         }
2819 out:
2820         read_unlock(&tasklist_lock);
2821 }
2822
2823 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2824 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2825 #endif
2826
2827 static void sighand_ctor(void *data)
2828 {
2829         struct sighand_struct *sighand = data;
2830
2831         spin_lock_init(&sighand->siglock);
2832         init_waitqueue_head(&sighand->signalfd_wqh);
2833 }
2834
2835 void __init proc_caches_init(void)
2836 {
2837         unsigned int mm_size;
2838
2839         sighand_cachep = kmem_cache_create("sighand_cache",
2840                         sizeof(struct sighand_struct), 0,
2841                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2842                         SLAB_ACCOUNT, sighand_ctor);
2843         signal_cachep = kmem_cache_create("signal_cache",
2844                         sizeof(struct signal_struct), 0,
2845                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2846                         NULL);
2847         files_cachep = kmem_cache_create("files_cache",
2848                         sizeof(struct files_struct), 0,
2849                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2850                         NULL);
2851         fs_cachep = kmem_cache_create("fs_cache",
2852                         sizeof(struct fs_struct), 0,
2853                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2854                         NULL);
2855
2856         /*
2857          * The mm_cpumask is located at the end of mm_struct, and is
2858          * dynamically sized based on the maximum CPU number this system
2859          * can have, taking hotplug into account (nr_cpu_ids).
2860          */
2861         mm_size = sizeof(struct mm_struct) + cpumask_size();
2862
2863         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2864                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2865                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2866                         offsetof(struct mm_struct, saved_auxv),
2867                         sizeof_field(struct mm_struct, saved_auxv),
2868                         NULL);
2869         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2870         mmap_init();
2871         nsproxy_cache_init();
2872 }
2873
2874 /*
2875  * Check constraints on flags passed to the unshare system call.
2876  */
2877 static int check_unshare_flags(unsigned long unshare_flags)
2878 {
2879         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2880                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2881                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2882                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
2883                                 CLONE_NEWTIME))
2884                 return -EINVAL;
2885         /*
2886          * Not implemented, but pretend it works if there is nothing
2887          * to unshare.  Note that unsharing the address space or the
2888          * signal handlers also need to unshare the signal queues (aka
2889          * CLONE_THREAD).
2890          */
2891         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2892                 if (!thread_group_empty(current))
2893                         return -EINVAL;
2894         }
2895         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2896                 if (refcount_read(&current->sighand->count) > 1)
2897                         return -EINVAL;
2898         }
2899         if (unshare_flags & CLONE_VM) {
2900                 if (!current_is_single_threaded())
2901                         return -EINVAL;
2902         }
2903
2904         return 0;
2905 }
2906
2907 /*
2908  * Unshare the filesystem structure if it is being shared
2909  */
2910 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2911 {
2912         struct fs_struct *fs = current->fs;
2913
2914         if (!(unshare_flags & CLONE_FS) || !fs)
2915                 return 0;
2916
2917         /* don't need lock here; in the worst case we'll do useless copy */
2918         if (fs->users == 1)
2919                 return 0;
2920
2921         *new_fsp = copy_fs_struct(fs);
2922         if (!*new_fsp)
2923                 return -ENOMEM;
2924
2925         return 0;
2926 }
2927
2928 /*
2929  * Unshare file descriptor table if it is being shared
2930  */
2931 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
2932                struct files_struct **new_fdp)
2933 {
2934         struct files_struct *fd = current->files;
2935         int error = 0;
2936
2937         if ((unshare_flags & CLONE_FILES) &&
2938             (fd && atomic_read(&fd->count) > 1)) {
2939                 *new_fdp = dup_fd(fd, max_fds, &error);
2940                 if (!*new_fdp)
2941                         return error;
2942         }
2943
2944         return 0;
2945 }
2946
2947 /*
2948  * unshare allows a process to 'unshare' part of the process
2949  * context which was originally shared using clone.  copy_*
2950  * functions used by kernel_clone() cannot be used here directly
2951  * because they modify an inactive task_struct that is being
2952  * constructed. Here we are modifying the current, active,
2953  * task_struct.
2954  */
2955 int ksys_unshare(unsigned long unshare_flags)
2956 {
2957         struct fs_struct *fs, *new_fs = NULL;
2958         struct files_struct *fd, *new_fd = NULL;
2959         struct cred *new_cred = NULL;
2960         struct nsproxy *new_nsproxy = NULL;
2961         int do_sysvsem = 0;
2962         int err;
2963
2964         /*
2965          * If unsharing a user namespace must also unshare the thread group
2966          * and unshare the filesystem root and working directories.
2967          */
2968         if (unshare_flags & CLONE_NEWUSER)
2969                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2970         /*
2971          * If unsharing vm, must also unshare signal handlers.
2972          */
2973         if (unshare_flags & CLONE_VM)
2974                 unshare_flags |= CLONE_SIGHAND;
2975         /*
2976          * If unsharing a signal handlers, must also unshare the signal queues.
2977          */
2978         if (unshare_flags & CLONE_SIGHAND)
2979                 unshare_flags |= CLONE_THREAD;
2980         /*
2981          * If unsharing namespace, must also unshare filesystem information.
2982          */
2983         if (unshare_flags & CLONE_NEWNS)
2984                 unshare_flags |= CLONE_FS;
2985
2986         err = check_unshare_flags(unshare_flags);
2987         if (err)
2988                 goto bad_unshare_out;
2989         /*
2990          * CLONE_NEWIPC must also detach from the undolist: after switching
2991          * to a new ipc namespace, the semaphore arrays from the old
2992          * namespace are unreachable.
2993          */
2994         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2995                 do_sysvsem = 1;
2996         err = unshare_fs(unshare_flags, &new_fs);
2997         if (err)
2998                 goto bad_unshare_out;
2999         err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3000         if (err)
3001                 goto bad_unshare_cleanup_fs;
3002         err = unshare_userns(unshare_flags, &new_cred);
3003         if (err)
3004                 goto bad_unshare_cleanup_fd;
3005         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3006                                          new_cred, new_fs);
3007         if (err)
3008                 goto bad_unshare_cleanup_cred;
3009
3010         if (new_cred) {
3011                 err = set_cred_ucounts(new_cred);
3012                 if (err)
3013                         goto bad_unshare_cleanup_cred;
3014         }
3015
3016         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3017                 if (do_sysvsem) {
3018                         /*
3019                          * CLONE_SYSVSEM is equivalent to sys_exit().
3020                          */
3021                         exit_sem(current);
3022                 }
3023                 if (unshare_flags & CLONE_NEWIPC) {
3024                         /* Orphan segments in old ns (see sem above). */
3025                         exit_shm(current);
3026                         shm_init_task(current);
3027                 }
3028
3029                 if (new_nsproxy)
3030                         switch_task_namespaces(current, new_nsproxy);
3031
3032                 task_lock(current);
3033
3034                 if (new_fs) {
3035                         fs = current->fs;
3036                         spin_lock(&fs->lock);
3037                         current->fs = new_fs;
3038                         if (--fs->users)
3039                                 new_fs = NULL;
3040                         else
3041                                 new_fs = fs;
3042                         spin_unlock(&fs->lock);
3043                 }
3044
3045                 if (new_fd) {
3046                         fd = current->files;
3047                         current->files = new_fd;
3048                         new_fd = fd;
3049                 }
3050
3051                 task_unlock(current);
3052
3053                 if (new_cred) {
3054                         /* Install the new user namespace */
3055                         commit_creds(new_cred);
3056                         new_cred = NULL;
3057                 }
3058         }
3059
3060         perf_event_namespaces(current);
3061
3062 bad_unshare_cleanup_cred:
3063         if (new_cred)
3064                 put_cred(new_cred);
3065 bad_unshare_cleanup_fd:
3066         if (new_fd)
3067                 put_files_struct(new_fd);
3068
3069 bad_unshare_cleanup_fs:
3070         if (new_fs)
3071                 free_fs_struct(new_fs);
3072
3073 bad_unshare_out:
3074         return err;
3075 }
3076
3077 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3078 {
3079         return ksys_unshare(unshare_flags);
3080 }
3081
3082 /*
3083  *      Helper to unshare the files of the current task.
3084  *      We don't want to expose copy_files internals to
3085  *      the exec layer of the kernel.
3086  */
3087
3088 int unshare_files(void)
3089 {
3090         struct task_struct *task = current;
3091         struct files_struct *old, *copy = NULL;
3092         int error;
3093
3094         error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, &copy);
3095         if (error || !copy)
3096                 return error;
3097
3098         old = task->files;
3099         task_lock(task);
3100         task->files = copy;
3101         task_unlock(task);
3102         put_files_struct(old);
3103         return 0;
3104 }
3105
3106 int sysctl_max_threads(struct ctl_table *table, int write,
3107                        void *buffer, size_t *lenp, loff_t *ppos)
3108 {
3109         struct ctl_table t;
3110         int ret;
3111         int threads = max_threads;
3112         int min = 1;
3113         int max = MAX_THREADS;
3114
3115         t = *table;
3116         t.data = &threads;
3117         t.extra1 = &min;
3118         t.extra2 = &max;
3119
3120         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3121         if (ret || !write)
3122                 return ret;
3123
3124         max_threads = threads;
3125
3126         return 0;
3127 }