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