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                 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
 293                         vfree_atomic(tsk->stack);
 294                 else
 295                         vfree(tsk->stack);
 296                 return;
 297         }
 298 #endif
 299
 300         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 301 }
 302 # else
 303 static struct kmem_cache *thread_stack_cache;
 304
 305 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 306                                                   int node)
 307 {
 308         unsigned long *stack;
 309         stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 310         stack = kasan_reset_tag(stack);
 311         tsk->stack = stack;
 312         return stack;
 313 }
 314
 315 static void free_thread_stack(struct task_struct *tsk)
 316 {
 317         kmem_cache_free(thread_stack_cache, tsk->stack);
 318 }
 319
 320 void thread_stack_cache_init(void)
 321 {
 322         thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 323                                         THREAD_SIZE, THREAD_SIZE, 0, 0,
 324                                         THREAD_SIZE, NULL);
 325         BUG_ON(thread_stack_cache == NULL);
 326 }
 327 # endif
 328 #endif
 329
 330 /* SLAB cache for signal_struct structures (tsk->signal) */
 331 static struct kmem_cache *signal_cachep;
 332
 333 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 334 struct kmem_cache *sighand_cachep;
 335
 336 /* SLAB cache for files_struct structures (tsk->files) */
 337 struct kmem_cache *files_cachep;
 338
 339 /* SLAB cache for fs_struct structures (tsk->fs) */
 340 struct kmem_cache *fs_cachep;
 341
 342 /* SLAB cache for vm_area_struct structures */
 343 static struct kmem_cache *vm_area_cachep;
 344
 345 /* SLAB cache for mm_struct structures (tsk->mm) */
 346 static struct kmem_cache *mm_cachep;
 347
 348 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 349 {
 350         struct vm_area_struct *vma;
 351
 352         vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 353         if (vma)
 354                 vma_init(vma, mm);
 355         return vma;
 356 }
 357
 358 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 359 {
 360         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 361
 362         if (new) {
 363                 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
 364                 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
 365                 /*
 366                  * orig->shared.rb may be modified concurrently, but the clone
 367                  * will be reinitialized.
 368                  */
 369                 *new = data_race(*orig);
 370                 INIT_LIST_HEAD(&new->anon_vma_chain);
 371                 new->vm_next = new->vm_prev = NULL;
 372         }
 373         return new;
 374 }
 375
 376 void vm_area_free(struct vm_area_struct *vma)
 377 {
 378         kmem_cache_free(vm_area_cachep, vma);
 379 }
 380
 381 static void account_kernel_stack(struct task_struct *tsk, int account)
 382 {
 383         void *stack = task_stack_page(tsk);
 384         struct vm_struct *vm = task_stack_vm_area(tsk);
 385
 386         if (vm) {
 387                 int i;
 388
 389                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 390                         mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
 391                                               account * (PAGE_SIZE / 1024));
 392         } else {
 393                 /* All stack pages are in the same node. */
 394                 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
 395                                       account * (THREAD_SIZE / 1024));
 396         }
 397 }
 398
 399 static int memcg_charge_kernel_stack(struct task_struct *tsk)
 400 {
 401 #ifdef CONFIG_VMAP_STACK
 402         struct vm_struct *vm = task_stack_vm_area(tsk);
 403         int ret;
 404
 405         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 406
 407         if (vm) {
 408                 int i;
 409
 410                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 411
 412                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 413                         /*
 414                          * If memcg_kmem_charge_page() fails, page's
 415                          * memory cgroup pointer is NULL, and
 416                          * memcg_kmem_uncharge_page() in free_thread_stack()
 417                          * will ignore this page.
 418                          */
 419                         ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL,
 420                                                      0);
 421                         if (ret)
 422                                 return ret;
 423                 }
 424         }
 425 #endif
 426         return 0;
 427 }
 428
 429 static void release_task_stack(struct task_struct *tsk)
 430 {
 431         if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
 432                 return;  /* Better to leak the stack than to free prematurely */
 433
 434         account_kernel_stack(tsk, -1);
 435         free_thread_stack(tsk);
 436         tsk->stack = NULL;
 437 #ifdef CONFIG_VMAP_STACK
 438         tsk->stack_vm_area = NULL;
 439 #endif
 440 }
 441
 442 #ifdef CONFIG_THREAD_INFO_IN_TASK
 443 void put_task_stack(struct task_struct *tsk)
 444 {
 445         if (refcount_dec_and_test(&tsk->stack_refcount))
 446                 release_task_stack(tsk);
 447 }
 448 #endif
 449
 450 void free_task(struct task_struct *tsk)
 451 {
 452         release_user_cpus_ptr(tsk);
 453         scs_release(tsk);
 454
 455 #ifndef CONFIG_THREAD_INFO_IN_TASK
 456         /*
 457          * The task is finally done with both the stack and thread_info,
 458          * so free both.
 459          */
 460         release_task_stack(tsk);
 461 #else
 462         /*
 463          * If the task had a separate stack allocation, it should be gone
 464          * by now.
 465          */
 466         WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
 467 #endif
 468         rt_mutex_debug_task_free(tsk);
 469         ftrace_graph_exit_task(tsk);
 470         arch_release_task_struct(tsk);
 471         if (tsk->flags & PF_KTHREAD)
 472                 free_kthread_struct(tsk);
 473         free_task_struct(tsk);
 474 }
 475 EXPORT_SYMBOL(free_task);
 476
 477 static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
 478 {
 479         struct file *exe_file;
 480
 481         exe_file = get_mm_exe_file(oldmm);
 482         RCU_INIT_POINTER(mm->exe_file, exe_file);
 483         /*
 484          * We depend on the oldmm having properly denied write access to the
 485          * exe_file already.
 486          */
 487         if (exe_file && deny_write_access(exe_file))
 488                 pr_warn_once("deny_write_access() failed in %s\n", __func__);
 489 }
 490
 491 #ifdef CONFIG_MMU
 492 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 493                                         struct mm_struct *oldmm)
 494 {
 495         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 496         struct rb_node **rb_link, *rb_parent;
 497         int retval;
 498         unsigned long charge;
 499         LIST_HEAD(uf);
 500
 501         uprobe_start_dup_mmap();
 502         if (mmap_write_lock_killable(oldmm)) {
 503                 retval = -EINTR;
 504                 goto fail_uprobe_end;
 505         }
 506         flush_cache_dup_mm(oldmm);
 507         uprobe_dup_mmap(oldmm, mm);
 508         /*
 509          * Not linked in yet - no deadlock potential:
 510          */
 511         mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
 512
 513         /* No ordering required: file already has been exposed. */
 514         dup_mm_exe_file(mm, oldmm);
 515
 516         mm->total_vm = oldmm->total_vm;
 517         mm->data_vm = oldmm->data_vm;
 518         mm->exec_vm = oldmm->exec_vm;
 519         mm->stack_vm = oldmm->stack_vm;
 520
 521         rb_link = &mm->mm_rb.rb_node;
 522         rb_parent = NULL;
 523         pprev = &mm->mmap;
 524         retval = ksm_fork(mm, oldmm);
 525         if (retval)
 526                 goto out;
 527         retval = khugepaged_fork(mm, oldmm);
 528         if (retval)
 529                 goto out;
 530
 531         prev = NULL;
 532         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 533                 struct file *file;
 534
 535                 if (mpnt->vm_flags & VM_DONTCOPY) {
 536                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 537                         continue;
 538                 }
 539                 charge = 0;
 540                 /*
 541                  * Don't duplicate many vmas if we've been oom-killed (for
 542                  * example)
 543                  */
 544                 if (fatal_signal_pending(current)) {
 545                         retval = -EINTR;
 546                         goto out;
 547                 }
 548                 if (mpnt->vm_flags & VM_ACCOUNT) {
 549                         unsigned long len = vma_pages(mpnt);
 550
 551                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 552                                 goto fail_nomem;
 553                         charge = len;
 554                 }
 555                 tmp = vm_area_dup(mpnt);
 556                 if (!tmp)
 557                         goto fail_nomem;
 558                 retval = vma_dup_policy(mpnt, tmp);
 559                 if (retval)
 560                         goto fail_nomem_policy;
 561                 tmp->vm_mm = mm;
 562                 retval = dup_userfaultfd(tmp, &uf);
 563                 if (retval)
 564                         goto fail_nomem_anon_vma_fork;
 565                 if (tmp->vm_flags & VM_WIPEONFORK) {
 566                         /*
 567                          * VM_WIPEONFORK gets a clean slate in the child.
 568                          * Don't prepare anon_vma until fault since we don't
 569                          * copy page for current vma.
 570                          */
 571                         tmp->anon_vma = NULL;
 572                 } else if (anon_vma_fork(tmp, mpnt))
 573                         goto fail_nomem_anon_vma_fork;
 574                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 575                 file = tmp->vm_file;
 576                 if (file) {
 577                         struct address_space *mapping = file->f_mapping;
 578
 579                         get_file(file);
 580                         i_mmap_lock_write(mapping);
 581                         if (tmp->vm_flags & VM_SHARED)
 582                                 mapping_allow_writable(mapping);
 583                         flush_dcache_mmap_lock(mapping);
 584                         /* insert tmp into the share list, just after mpnt */
 585                         vma_interval_tree_insert_after(tmp, mpnt,
 586                                         &mapping->i_mmap);
 587                         flush_dcache_mmap_unlock(mapping);
 588                         i_mmap_unlock_write(mapping);
 589                 }
 590
 591                 /*
 592                  * Clear hugetlb-related page reserves for children. This only
 593                  * affects MAP_PRIVATE mappings. Faults generated by the child
 594                  * are not guaranteed to succeed, even if read-only
 595                  */
 596                 if (is_vm_hugetlb_page(tmp))
 597                         reset_vma_resv_huge_pages(tmp);
 598
 599                 /*
 600                  * Link in the new vma and copy the page table entries.
 601                  */
 602                 *pprev = tmp;
 603                 pprev = &tmp->vm_next;
 604                 tmp->vm_prev = prev;
 605                 prev = tmp;
 606
 607                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 608                 rb_link = &tmp->vm_rb.rb_right;
 609                 rb_parent = &tmp->vm_rb;
 610
 611                 mm->map_count++;
 612                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 613                         retval = copy_page_range(tmp, mpnt);
 614
 615                 if (tmp->vm_ops && tmp->vm_ops->open)
 616                         tmp->vm_ops->open(tmp);
 617
 618                 if (retval)
 619                         goto out;
 620         }
 621         /* a new mm has just been created */
 622         retval = arch_dup_mmap(oldmm, mm);
 623 out:
 624         mmap_write_unlock(mm);
 625         flush_tlb_mm(oldmm);
 626         mmap_write_unlock(oldmm);
 627         dup_userfaultfd_complete(&uf);
 628 fail_uprobe_end:
 629         uprobe_end_dup_mmap();
 630         return retval;
 631 fail_nomem_anon_vma_fork:
 632         mpol_put(vma_policy(tmp));
 633 fail_nomem_policy:
 634         vm_area_free(tmp);
 635 fail_nomem:
 636         retval = -ENOMEM;
 637         vm_unacct_memory(charge);
 638         goto out;
 639 }
 640
 641 static inline int mm_alloc_pgd(struct mm_struct *mm)
 642 {
 643         mm->pgd = pgd_alloc(mm);
 644         if (unlikely(!mm->pgd))
 645                 return -ENOMEM;
 646         return 0;
 647 }
 648
 649 static inline void mm_free_pgd(struct mm_struct *mm)
 650 {
 651         pgd_free(mm, mm->pgd);
 652 }
 653 #else
 654 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 655 {
 656         mmap_write_lock(oldmm);
 657         dup_mm_exe_file(mm, oldmm);
 658         mmap_write_unlock(oldmm);
 659         return 0;
 660 }
 661 #define mm_alloc_pgd(mm)        (0)
 662 #define mm_free_pgd(mm)
 663 #endif /* CONFIG_MMU */
 664
 665 static void check_mm(struct mm_struct *mm)
 666 {
 667         int i;
 668
 669         BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
 670                          "Please make sure 'struct resident_page_types[]' is updated as well");
 671
 672         for (i = 0; i < NR_MM_COUNTERS; i++) {
 673                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 674
 675                 if (unlikely(x))
 676                         pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
 677                                  mm, resident_page_types[i], x);
 678         }
 679
 680         if (mm_pgtables_bytes(mm))
 681                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 682                                 mm_pgtables_bytes(mm));
 683
 684 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 685         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 686 #endif
 687 }
 688
 689 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 690 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 691
 692 /*
 693  * Called when the last reference to the mm
 694  * is dropped: either by a lazy thread or by
 695  * mmput. Free the page directory and the mm.
 696  */
 697 void __mmdrop(struct mm_struct *mm)
 698 {
 699         BUG_ON(mm == &init_mm);
 700         WARN_ON_ONCE(mm == current->mm);
 701         WARN_ON_ONCE(mm == current->active_mm);
 702         mm_free_pgd(mm);
 703         destroy_context(mm);
 704         mmu_notifier_subscriptions_destroy(mm);
 705         check_mm(mm);
 706         put_user_ns(mm->user_ns);
 707         free_mm(mm);
 708 }
 709 EXPORT_SYMBOL_GPL(__mmdrop);
 710
 711 #ifdef CONFIG_PREEMPT_RT
 712 /*
 713  * RCU callback for delayed mm drop. Not strictly RCU, but call_rcu() is
 714  * by far the least expensive way to do that.
 715  */
 716 void __mmdrop_delayed(struct rcu_head *rhp)
 717 {
 718         struct mm_struct *mm = container_of(rhp, struct mm_struct, delayed_drop);
 719
 720         __mmdrop(mm);
 721 }
 722 #endif
 723
 724 static void mmdrop_async_fn(struct work_struct *work)
 725 {
 726         struct mm_struct *mm;
 727
 728         mm = container_of(work, struct mm_struct, async_put_work);
 729         __mmdrop(mm);
 730 }
 731
 732 static void mmdrop_async(struct mm_struct *mm)
 733 {
 734         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 735                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 736                 schedule_work(&mm->async_put_work);
 737         }
 738 }
 739
 740 static inline void free_signal_struct(struct signal_struct *sig)
 741 {
 742         taskstats_tgid_free(sig);
 743         sched_autogroup_exit(sig);
 744         /*
 745          * __mmdrop is not safe to call from softirq context on x86 due to
 746          * pgd_dtor so postpone it to the async context
 747          */
 748         if (sig->oom_mm)
 749                 mmdrop_async(sig->oom_mm);
 750         kmem_cache_free(signal_cachep, sig);
 751 }
 752
 753 static inline void put_signal_struct(struct signal_struct *sig)
 754 {
 755         if (refcount_dec_and_test(&sig->sigcnt))
 756                 free_signal_struct(sig);
 757 }
 758
 759 void __put_task_struct(struct task_struct *tsk)
 760 {
 761         WARN_ON(!tsk->exit_state);
 762         WARN_ON(refcount_read(&tsk->usage));
 763         WARN_ON(tsk == current);
 764
 765         io_uring_free(tsk);
 766         cgroup_free(tsk);
 767         task_numa_free(tsk, true);
 768         security_task_free(tsk);
 769         bpf_task_storage_free(tsk);
 770         exit_creds(tsk);
 771         delayacct_tsk_free(tsk);
 772         put_signal_struct(tsk->signal);
 773         sched_core_free(tsk);
 774
 775         if (!profile_handoff_task(tsk))
 776                 free_task(tsk);
 777 }
 778 EXPORT_SYMBOL_GPL(__put_task_struct);
 779
 780 void __init __weak arch_task_cache_init(void) { }
 781
 782 /*
 783  * set_max_threads
 784  */
 785 static void set_max_threads(unsigned int max_threads_suggested)
 786 {
 787         u64 threads;
 788         unsigned long nr_pages = totalram_pages();
 789
 790         /*
 791          * The number of threads shall be limited such that the thread
 792          * structures may only consume a small part of the available memory.
 793          */
 794         if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
 795                 threads = MAX_THREADS;
 796         else
 797                 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
 798                                     (u64) THREAD_SIZE * 8UL);
 799
 800         if (threads > max_threads_suggested)
 801                 threads = max_threads_suggested;
 802
 803         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 804 }
 805
 806 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 807 /* Initialized by the architecture: */
 808 int arch_task_struct_size __read_mostly;
 809 #endif
 810
 811 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 812 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 813 {
 814         /* Fetch thread_struct whitelist for the architecture. */
 815         arch_thread_struct_whitelist(offset, size);
 816
 817         /*
 818          * Handle zero-sized whitelist or empty thread_struct, otherwise
 819          * adjust offset to position of thread_struct in task_struct.
 820          */
 821         if (unlikely(*size == 0))
 822                 *offset = 0;
 823         else
 824                 *offset += offsetof(struct task_struct, thread);
 825 }
 826 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
 827
 828 void __init fork_init(void)
 829 {
 830         int i;
 831 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 832 #ifndef ARCH_MIN_TASKALIGN
 833 #define ARCH_MIN_TASKALIGN      0
 834 #endif
 835         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 836         unsigned long useroffset, usersize;
 837
 838         /* create a slab on which task_structs can be allocated */
 839         task_struct_whitelist(&useroffset, &usersize);
 840         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 841                         arch_task_struct_size, align,
 842                         SLAB_PANIC|SLAB_ACCOUNT,
 843                         useroffset, usersize, NULL);
 844 #endif
 845
 846         /* do the arch specific task caches init */
 847         arch_task_cache_init();
 848
 849         set_max_threads(MAX_THREADS);
 850
 851         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 852         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 853         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 854                 init_task.signal->rlim[RLIMIT_NPROC];
 855
 856         for (i = 0; i < MAX_PER_NAMESPACE_UCOUNTS; i++)
 857                 init_user_ns.ucount_max[i] = max_threads/2;
 858
 859         set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_NPROC,      RLIM_INFINITY);
 860         set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE,   RLIM_INFINITY);
 861         set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
 862         set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK,    RLIM_INFINITY);
 863
 864 #ifdef CONFIG_VMAP_STACK
 865         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 866                           NULL, free_vm_stack_cache);
 867 #endif
 868
 869         scs_init();
 870
 871         lockdep_init_task(&init_task);
 872         uprobes_init();
 873 }
 874
 875 int __weak arch_dup_task_struct(struct task_struct *dst,
 876                                                struct task_struct *src)
 877 {
 878         *dst = *src;
 879         return 0;
 880 }
 881
 882 void set_task_stack_end_magic(struct task_struct *tsk)
 883 {
 884         unsigned long *stackend;
 885
 886         stackend = end_of_stack(tsk);
 887         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 888 }
 889
 890 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 891 {
 892         struct task_struct *tsk;
 893         unsigned long *stack;
 894         struct vm_struct *stack_vm_area __maybe_unused;
 895         int err;
 896
 897         if (node == NUMA_NO_NODE)
 898                 node = tsk_fork_get_node(orig);
 899         tsk = alloc_task_struct_node(node);
 900         if (!tsk)
 901                 return NULL;
 902
 903         stack = alloc_thread_stack_node(tsk, node);
 904         if (!stack)
 905                 goto free_tsk;
 906
 907         if (memcg_charge_kernel_stack(tsk))
 908                 goto free_stack;
 909
 910         stack_vm_area = task_stack_vm_area(tsk);
 911
 912         err = arch_dup_task_struct(tsk, orig);
 913
 914         /*
 915          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 916          * sure they're properly initialized before using any stack-related
 917          * functions again.
 918          */
 919         tsk->stack = stack;
 920 #ifdef CONFIG_VMAP_STACK
 921         tsk->stack_vm_area = stack_vm_area;
 922 #endif
 923 #ifdef CONFIG_THREAD_INFO_IN_TASK
 924         refcount_set(&tsk->stack_refcount, 1);
 925 #endif
 926
 927         if (err)
 928                 goto free_stack;
 929
 930         err = scs_prepare(tsk, node);
 931         if (err)
 932                 goto free_stack;
 933
 934 #ifdef CONFIG_SECCOMP
 935         /*
 936          * We must handle setting up seccomp filters once we're under
 937          * the sighand lock in case orig has changed between now and
 938          * then. Until then, filter must be NULL to avoid messing up
 939          * the usage counts on the error path calling free_task.
 940          */
 941         tsk->seccomp.filter = NULL;
 942 #endif
 943
 944         setup_thread_stack(tsk, orig);
 945         clear_user_return_notifier(tsk);
 946         clear_tsk_need_resched(tsk);
 947         set_task_stack_end_magic(tsk);
 948         clear_syscall_work_syscall_user_dispatch(tsk);
 949
 950 #ifdef CONFIG_STACKPROTECTOR
 951         tsk->stack_canary = get_random_canary();
 952 #endif
 953         if (orig->cpus_ptr == &orig->cpus_mask)
 954                 tsk->cpus_ptr = &tsk->cpus_mask;
 955         dup_user_cpus_ptr(tsk, orig, node);
 956
 957         /*
 958          * One for the user space visible state that goes away when reaped.
 959          * One for the scheduler.
 960          */
 961         refcount_set(&tsk->rcu_users, 2);
 962         /* One for the rcu users */
 963         refcount_set(&tsk->usage, 1);
 964 #ifdef CONFIG_BLK_DEV_IO_TRACE
 965         tsk->btrace_seq = 0;
 966 #endif
 967         tsk->splice_pipe = NULL;
 968         tsk->task_frag.page = NULL;
 969         tsk->wake_q.next = NULL;
 970         tsk->pf_io_worker = NULL;
 971
 972         account_kernel_stack(tsk, 1);
 973
 974         kcov_task_init(tsk);
 975         kmap_local_fork(tsk);
 976
 977 #ifdef CONFIG_FAULT_INJECTION
 978         tsk->fail_nth = 0;
 979 #endif
 980
 981 #ifdef CONFIG_BLK_CGROUP
 982         tsk->throttle_queue = NULL;
 983         tsk->use_memdelay = 0;
 984 #endif
 985
 986 #ifdef CONFIG_MEMCG
 987         tsk->active_memcg = NULL;
 988 #endif
 989         return tsk;
 990
 991 free_stack:
 992         free_thread_stack(tsk);
 993 free_tsk:
 994         free_task_struct(tsk);
 995         return NULL;
 996 }
 997
 998 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 999
1000 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1001
1002 static int __init coredump_filter_setup(char *s)
1003 {
1004         default_dump_filter =
1005                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1006                 MMF_DUMP_FILTER_MASK;
1007         return 1;
1008 }
1009
1010 __setup("coredump_filter=", coredump_filter_setup);
1011
1012 #include <linux/init_task.h>
1013
1014 static void mm_init_aio(struct mm_struct *mm)
1015 {
1016 #ifdef CONFIG_AIO
1017         spin_lock_init(&mm->ioctx_lock);
1018         mm->ioctx_table = NULL;
1019 #endif
1020 }
1021
1022 static __always_inline void mm_clear_owner(struct mm_struct *mm,
1023                                            struct task_struct *p)
1024 {
1025 #ifdef CONFIG_MEMCG
1026         if (mm->owner == p)
1027                 WRITE_ONCE(mm->owner, NULL);
1028 #endif
1029 }
1030
1031 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1032 {
1033 #ifdef CONFIG_MEMCG
1034         mm->owner = p;
1035 #endif
1036 }
1037
1038 static void mm_init_pasid(struct mm_struct *mm)
1039 {
1040 #ifdef CONFIG_IOMMU_SUPPORT
1041         mm->pasid = INIT_PASID;
1042 #endif
1043 }
1044
1045 static void mm_init_uprobes_state(struct mm_struct *mm)
1046 {
1047 #ifdef CONFIG_UPROBES
1048         mm->uprobes_state.xol_area = NULL;
1049 #endif
1050 }
1051
1052 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1053         struct user_namespace *user_ns)
1054 {
1055         mm->mmap = NULL;
1056         mm->mm_rb = RB_ROOT;
1057         mm->vmacache_seqnum = 0;
1058         atomic_set(&mm->mm_users, 1);
1059         atomic_set(&mm->mm_count, 1);
1060         seqcount_init(&mm->write_protect_seq);
1061         mmap_init_lock(mm);
1062         INIT_LIST_HEAD(&mm->mmlist);
1063         mm->core_state = NULL;
1064         mm_pgtables_bytes_init(mm);
1065         mm->map_count = 0;
1066         mm->locked_vm = 0;
1067         atomic64_set(&mm->pinned_vm, 0);
1068         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1069         spin_lock_init(&mm->page_table_lock);
1070         spin_lock_init(&mm->arg_lock);
1071         mm_init_cpumask(mm);
1072         mm_init_aio(mm);
1073         mm_init_owner(mm, p);
1074         mm_init_pasid(mm);
1075         RCU_INIT_POINTER(mm->exe_file, NULL);
1076         mmu_notifier_subscriptions_init(mm);
1077         init_tlb_flush_pending(mm);
1078 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1079         mm->pmd_huge_pte = NULL;
1080 #endif
1081         mm_init_uprobes_state(mm);
1082         hugetlb_count_init(mm);
1083
1084         if (current->mm) {
1085                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1086                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1087         } else {
1088                 mm->flags = default_dump_filter;
1089                 mm->def_flags = 0;
1090         }
1091
1092         if (mm_alloc_pgd(mm))
1093                 goto fail_nopgd;
1094
1095         if (init_new_context(p, mm))
1096                 goto fail_nocontext;
1097
1098         mm->user_ns = get_user_ns(user_ns);
1099         return mm;
1100
1101 fail_nocontext:
1102         mm_free_pgd(mm);
1103 fail_nopgd:
1104         free_mm(mm);
1105         return NULL;
1106 }
1107
1108 /*
1109  * Allocate and initialize an mm_struct.
1110  */
1111 struct mm_struct *mm_alloc(void)
1112 {
1113         struct mm_struct *mm;
1114
1115         mm = allocate_mm();
1116         if (!mm)
1117                 return NULL;
1118
1119         memset(mm, 0, sizeof(*mm));
1120         return mm_init(mm, current, current_user_ns());
1121 }
1122
1123 static inline void __mmput(struct mm_struct *mm)
1124 {
1125         VM_BUG_ON(atomic_read(&mm->mm_users));
1126
1127         uprobe_clear_state(mm);
1128         exit_aio(mm);
1129         ksm_exit(mm);
1130         khugepaged_exit(mm); /* must run before exit_mmap */
1131         exit_mmap(mm);
1132         mm_put_huge_zero_page(mm);
1133         set_mm_exe_file(mm, NULL);
1134         if (!list_empty(&mm->mmlist)) {
1135                 spin_lock(&mmlist_lock);
1136                 list_del(&mm->mmlist);
1137                 spin_unlock(&mmlist_lock);
1138         }
1139         if (mm->binfmt)
1140                 module_put(mm->binfmt->module);
1141         mmdrop(mm);
1142 }
1143
1144 /*
1145  * Decrement the use count and release all resources for an mm.
1146  */
1147 void mmput(struct mm_struct *mm)
1148 {
1149         might_sleep();
1150
1151         if (atomic_dec_and_test(&mm->mm_users))
1152                 __mmput(mm);
1153 }
1154 EXPORT_SYMBOL_GPL(mmput);
1155
1156 #ifdef CONFIG_MMU
1157 static void mmput_async_fn(struct work_struct *work)
1158 {
1159         struct mm_struct *mm = container_of(work, struct mm_struct,
1160                                             async_put_work);
1161
1162         __mmput(mm);
1163 }
1164
1165 void mmput_async(struct mm_struct *mm)
1166 {
1167         if (atomic_dec_and_test(&mm->mm_users)) {
1168                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1169                 schedule_work(&mm->async_put_work);
1170         }
1171 }
1172 EXPORT_SYMBOL_GPL(mmput_async);
1173 #endif
1174
1175 /**
1176  * set_mm_exe_file - change a reference to the mm's executable file
1177  *
1178  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1179  *
1180  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1181  * invocations: in mmput() nobody alive left, in execve task is single
1182  * threaded.
1183  *
1184  * Can only fail if new_exe_file != NULL.
1185  */
1186 int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1187 {
1188         struct file *old_exe_file;
1189
1190         /*
1191          * It is safe to dereference the exe_file without RCU as
1192          * this function is only called if nobody else can access
1193          * this mm -- see comment above for justification.
1194          */
1195         old_exe_file = rcu_dereference_raw(mm->exe_file);
1196
1197         if (new_exe_file) {
1198                 /*
1199                  * We expect the caller (i.e., sys_execve) to already denied
1200                  * write access, so this is unlikely to fail.
1201                  */
1202                 if (unlikely(deny_write_access(new_exe_file)))
1203                         return -EACCES;
1204                 get_file(new_exe_file);
1205         }
1206         rcu_assign_pointer(mm->exe_file, new_exe_file);
1207         if (old_exe_file) {
1208                 allow_write_access(old_exe_file);
1209                 fput(old_exe_file);
1210         }
1211         return 0;
1212 }
1213
1214 /**
1215  * replace_mm_exe_file - replace a reference to the mm's executable file
1216  *
1217  * This changes mm's executable file (shown as symlink /proc/[pid]/exe),
1218  * dealing with concurrent invocation and without grabbing the mmap lock in
1219  * write mode.
1220  *
1221  * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1222  */
1223 int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1224 {
1225         struct vm_area_struct *vma;
1226         struct file *old_exe_file;
1227         int ret = 0;
1228
1229         /* Forbid mm->exe_file change if old file still mapped. */
1230         old_exe_file = get_mm_exe_file(mm);
1231         if (old_exe_file) {
1232                 mmap_read_lock(mm);
1233                 for (vma = mm->mmap; vma && !ret; vma = vma->vm_next) {
1234                         if (!vma->vm_file)
1235                                 continue;
1236                         if (path_equal(&vma->vm_file->f_path,
1237                                        &old_exe_file->f_path))
1238                                 ret = -EBUSY;
1239                 }
1240                 mmap_read_unlock(mm);
1241                 fput(old_exe_file);
1242                 if (ret)
1243                         return ret;
1244         }
1245
1246         /* set the new file, lockless */
1247         ret = deny_write_access(new_exe_file);
1248         if (ret)
1249                 return -EACCES;
1250         get_file(new_exe_file);
1251
1252         old_exe_file = xchg(&mm->exe_file, new_exe_file);
1253         if (old_exe_file) {
1254                 /*
1255                  * Don't race with dup_mmap() getting the file and disallowing
1256                  * write access while someone might open the file writable.
1257                  */
1258                 mmap_read_lock(mm);
1259                 allow_write_access(old_exe_file);
1260                 fput(old_exe_file);
1261                 mmap_read_unlock(mm);
1262         }
1263         return 0;
1264 }
1265
1266 /**
1267  * get_mm_exe_file - acquire a reference to the mm's executable file
1268  *
1269  * Returns %NULL if mm has no associated executable file.
1270  * User must release file via fput().
1271  */
1272 struct file *get_mm_exe_file(struct mm_struct *mm)
1273 {
1274         struct file *exe_file;
1275
1276         rcu_read_lock();
1277         exe_file = rcu_dereference(mm->exe_file);
1278         if (exe_file && !get_file_rcu(exe_file))
1279                 exe_file = NULL;
1280         rcu_read_unlock();
1281         return exe_file;
1282 }
1283
1284 /**
1285  * get_task_exe_file - acquire a reference to the task's executable file
1286  *
1287  * Returns %NULL if task's mm (if any) has no associated executable file or
1288  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1289  * User must release file via fput().
1290  */
1291 struct file *get_task_exe_file(struct task_struct *task)
1292 {
1293         struct file *exe_file = NULL;
1294         struct mm_struct *mm;
1295
1296         task_lock(task);
1297         mm = task->mm;
1298         if (mm) {
1299                 if (!(task->flags & PF_KTHREAD))
1300                         exe_file = get_mm_exe_file(mm);
1301         }
1302         task_unlock(task);
1303         return exe_file;
1304 }
1305
1306 /**
1307  * get_task_mm - acquire a reference to the task's mm
1308  *
1309  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1310  * this kernel workthread has transiently adopted a user mm with use_mm,
1311  * to do its AIO) is not set and if so returns a reference to it, after
1312  * bumping up the use count.  User must release the mm via mmput()
1313  * after use.  Typically used by /proc and ptrace.
1314  */
1315 struct mm_struct *get_task_mm(struct task_struct *task)
1316 {
1317         struct mm_struct *mm;
1318
1319         task_lock(task);
1320         mm = task->mm;
1321         if (mm) {
1322                 if (task->flags & PF_KTHREAD)
1323                         mm = NULL;
1324                 else
1325                         mmget(mm);
1326         }
1327         task_unlock(task);
1328         return mm;
1329 }
1330 EXPORT_SYMBOL_GPL(get_task_mm);
1331
1332 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1333 {
1334         struct mm_struct *mm;
1335         int err;
1336
1337         err =  down_read_killable(&task->signal->exec_update_lock);
1338         if (err)
1339                 return ERR_PTR(err);
1340
1341         mm = get_task_mm(task);
1342         if (mm && mm != current->mm &&
1343                         !ptrace_may_access(task, mode)) {
1344                 mmput(mm);
1345                 mm = ERR_PTR(-EACCES);
1346         }
1347         up_read(&task->signal->exec_update_lock);
1348
1349         return mm;
1350 }
1351
1352 static void complete_vfork_done(struct task_struct *tsk)
1353 {
1354         struct completion *vfork;
1355
1356         task_lock(tsk);
1357         vfork = tsk->vfork_done;
1358         if (likely(vfork)) {
1359                 tsk->vfork_done = NULL;
1360                 complete(vfork);
1361         }
1362         task_unlock(tsk);
1363 }
1364
1365 static int wait_for_vfork_done(struct task_struct *child,
1366                                 struct completion *vfork)
1367 {
1368         int killed;
1369
1370         freezer_do_not_count();
1371         cgroup_enter_frozen();
1372         killed = wait_for_completion_killable(vfork);
1373         cgroup_leave_frozen(false);
1374         freezer_count();
1375
1376         if (killed) {
1377                 task_lock(child);
1378                 child->vfork_done = NULL;
1379                 task_unlock(child);
1380         }
1381
1382         put_task_struct(child);
1383         return killed;
1384 }
1385
1386 /* Please note the differences between mmput and mm_release.
1387  * mmput is called whenever we stop holding onto a mm_struct,
1388  * error success whatever.
1389  *
1390  * mm_release is called after a mm_struct has been removed
1391  * from the current process.
1392  *
1393  * This difference is important for error handling, when we
1394  * only half set up a mm_struct for a new process and need to restore
1395  * the old one.  Because we mmput the new mm_struct before
1396  * restoring the old one. . .
1397  * Eric Biederman 10 January 1998
1398  */
1399 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1400 {
1401         uprobe_free_utask(tsk);
1402
1403         /* Get rid of any cached register state */
1404         deactivate_mm(tsk, mm);
1405
1406         /*
1407          * Signal userspace if we're not exiting with a core dump
1408          * because we want to leave the value intact for debugging
1409          * purposes.
1410          */
1411         if (tsk->clear_child_tid) {
1412                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1413                     atomic_read(&mm->mm_users) > 1) {
1414                         /*
1415                          * We don't check the error code - if userspace has
1416                          * not set up a proper pointer then tough luck.
1417                          */
1418                         put_user(0, tsk->clear_child_tid);
1419                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1420                                         1, NULL, NULL, 0, 0);
1421                 }
1422                 tsk->clear_child_tid = NULL;
1423         }
1424
1425         /*
1426          * All done, finally we can wake up parent and return this mm to him.
1427          * Also kthread_stop() uses this completion for synchronization.
1428          */
1429         if (tsk->vfork_done)
1430                 complete_vfork_done(tsk);
1431 }
1432
1433 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1434 {
1435         futex_exit_release(tsk);
1436         mm_release(tsk, mm);
1437 }
1438
1439 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1440 {
1441         futex_exec_release(tsk);
1442         mm_release(tsk, mm);
1443 }
1444
1445 /**
1446  * dup_mm() - duplicates an existing mm structure
1447  * @tsk: the task_struct with which the new mm will be associated.
1448  * @oldmm: the mm to duplicate.
1449  *
1450  * Allocates a new mm structure and duplicates the provided @oldmm structure
1451  * content into it.
1452  *
1453  * Return: the duplicated mm or NULL on failure.
1454  */
1455 static struct mm_struct *dup_mm(struct task_struct *tsk,
1456                                 struct mm_struct *oldmm)
1457 {
1458         struct mm_struct *mm;
1459         int err;
1460
1461         mm = allocate_mm();
1462         if (!mm)
1463                 goto fail_nomem;
1464
1465         memcpy(mm, oldmm, sizeof(*mm));
1466
1467         if (!mm_init(mm, tsk, mm->user_ns))
1468                 goto fail_nomem;
1469
1470         err = dup_mmap(mm, oldmm);
1471         if (err)
1472                 goto free_pt;
1473
1474         mm->hiwater_rss = get_mm_rss(mm);
1475         mm->hiwater_vm = mm->total_vm;
1476
1477         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1478                 goto free_pt;
1479
1480         return mm;
1481
1482 free_pt:
1483         /* don't put binfmt in mmput, we haven't got module yet */
1484         mm->binfmt = NULL;
1485         mm_init_owner(mm, NULL);
1486         mmput(mm);
1487
1488 fail_nomem:
1489         return NULL;
1490 }
1491
1492 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1493 {
1494         struct mm_struct *mm, *oldmm;
1495
1496         tsk->min_flt = tsk->maj_flt = 0;
1497         tsk->nvcsw = tsk->nivcsw = 0;
1498 #ifdef CONFIG_DETECT_HUNG_TASK
1499         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1500         tsk->last_switch_time = 0;
1501 #endif
1502
1503         tsk->mm = NULL;
1504         tsk->active_mm = NULL;
1505
1506         /*
1507          * Are we cloning a kernel thread?
1508          *
1509          * We need to steal a active VM for that..
1510          */
1511         oldmm = current->mm;
1512         if (!oldmm)
1513                 return 0;
1514
1515         /* initialize the new vmacache entries */
1516         vmacache_flush(tsk);
1517
1518         if (clone_flags & CLONE_VM) {
1519                 mmget(oldmm);
1520                 mm = oldmm;
1521         } else {
1522                 mm = dup_mm(tsk, current->mm);
1523                 if (!mm)
1524                         return -ENOMEM;
1525         }
1526
1527         tsk->mm = mm;
1528         tsk->active_mm = mm;
1529         return 0;
1530 }
1531
1532 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1533 {
1534         struct fs_struct *fs = current->fs;
1535         if (clone_flags & CLONE_FS) {
1536                 /* tsk->fs is already what we want */
1537                 spin_lock(&fs->lock);
1538                 if (fs->in_exec) {
1539                         spin_unlock(&fs->lock);
1540                         return -EAGAIN;
1541                 }
1542                 fs->users++;
1543                 spin_unlock(&fs->lock);
1544                 return 0;
1545         }
1546         tsk->fs = copy_fs_struct(fs);
1547         if (!tsk->fs)
1548                 return -ENOMEM;
1549         return 0;
1550 }
1551
1552 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1553 {
1554         struct files_struct *oldf, *newf;
1555         int error = 0;
1556
1557         /*
1558          * A background process may not have any files ...
1559          */
1560         oldf = current->files;
1561         if (!oldf)
1562                 goto out;
1563
1564         if (clone_flags & CLONE_FILES) {
1565                 atomic_inc(&oldf->count);
1566                 goto out;
1567         }
1568
1569         newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1570         if (!newf)
1571                 goto out;
1572
1573         tsk->files = newf;
1574         error = 0;
1575 out:
1576         return error;
1577 }
1578
1579 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1580 {
1581 #ifdef CONFIG_BLOCK
1582         struct io_context *ioc = current->io_context;
1583         struct io_context *new_ioc;
1584
1585         if (!ioc)
1586                 return 0;
1587         /*
1588          * Share io context with parent, if CLONE_IO is set
1589          */
1590         if (clone_flags & CLONE_IO) {
1591                 ioc_task_link(ioc);
1592                 tsk->io_context = ioc;
1593         } else if (ioprio_valid(ioc->ioprio)) {
1594                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1595                 if (unlikely(!new_ioc))
1596                         return -ENOMEM;
1597
1598                 new_ioc->ioprio = ioc->ioprio;
1599                 put_io_context(new_ioc);
1600         }
1601 #endif
1602         return 0;
1603 }
1604
1605 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1606 {
1607         struct sighand_struct *sig;
1608
1609         if (clone_flags & CLONE_SIGHAND) {
1610                 refcount_inc(&current->sighand->count);
1611                 return 0;
1612         }
1613         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1614         RCU_INIT_POINTER(tsk->sighand, sig);
1615         if (!sig)
1616                 return -ENOMEM;
1617
1618         refcount_set(&sig->count, 1);
1619         spin_lock_irq(&current->sighand->siglock);
1620         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1621         spin_unlock_irq(&current->sighand->siglock);
1622
1623         /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1624         if (clone_flags & CLONE_CLEAR_SIGHAND)
1625                 flush_signal_handlers(tsk, 0);
1626
1627         return 0;
1628 }
1629
1630 void __cleanup_sighand(struct sighand_struct *sighand)
1631 {
1632         if (refcount_dec_and_test(&sighand->count)) {
1633                 signalfd_cleanup(sighand);
1634                 /*
1635                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1636                  * without an RCU grace period, see __lock_task_sighand().
1637                  */
1638                 kmem_cache_free(sighand_cachep, sighand);
1639         }
1640 }
1641
1642 /*
1643  * Initialize POSIX timer handling for a thread group.
1644  */
1645 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1646 {
1647         struct posix_cputimers *pct = &sig->posix_cputimers;
1648         unsigned long cpu_limit;
1649
1650         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1651         posix_cputimers_group_init(pct, cpu_limit);
1652 }
1653
1654 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1655 {
1656         struct signal_struct *sig;
1657
1658         if (clone_flags & CLONE_THREAD)
1659                 return 0;
1660
1661         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1662         tsk->signal = sig;
1663         if (!sig)
1664                 return -ENOMEM;
1665
1666         sig->nr_threads = 1;
1667         atomic_set(&sig->live, 1);
1668         refcount_set(&sig->sigcnt, 1);
1669
1670         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1671         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1672         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1673
1674         init_waitqueue_head(&sig->wait_chldexit);
1675         sig->curr_target = tsk;
1676         init_sigpending(&sig->shared_pending);
1677         INIT_HLIST_HEAD(&sig->multiprocess);
1678         seqlock_init(&sig->stats_lock);
1679         prev_cputime_init(&sig->prev_cputime);
1680
1681 #ifdef CONFIG_POSIX_TIMERS
1682         INIT_LIST_HEAD(&sig->posix_timers);
1683         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1684         sig->real_timer.function = it_real_fn;
1685 #endif
1686
1687         task_lock(current->group_leader);
1688         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1689         task_unlock(current->group_leader);
1690
1691         posix_cpu_timers_init_group(sig);
1692
1693         tty_audit_fork(sig);
1694         sched_autogroup_fork(sig);
1695
1696         sig->oom_score_adj = current->signal->oom_score_adj;
1697         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1698
1699         mutex_init(&sig->cred_guard_mutex);
1700         init_rwsem(&sig->exec_update_lock);
1701
1702         return 0;
1703 }
1704
1705 static void copy_seccomp(struct task_struct *p)
1706 {
1707 #ifdef CONFIG_SECCOMP
1708         /*
1709          * Must be called with sighand->lock held, which is common to
1710          * all threads in the group. Holding cred_guard_mutex is not
1711          * needed because this new task is not yet running and cannot
1712          * be racing exec.
1713          */
1714         assert_spin_locked(&current->sighand->siglock);
1715
1716         /* Ref-count the new filter user, and assign it. */
1717         get_seccomp_filter(current);
1718         p->seccomp = current->seccomp;
1719
1720         /*
1721          * Explicitly enable no_new_privs here in case it got set
1722          * between the task_struct being duplicated and holding the
1723          * sighand lock. The seccomp state and nnp must be in sync.
1724          */
1725         if (task_no_new_privs(current))
1726                 task_set_no_new_privs(p);
1727
1728         /*
1729          * If the parent gained a seccomp mode after copying thread
1730          * flags and between before we held the sighand lock, we have
1731          * to manually enable the seccomp thread flag here.
1732          */
1733         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1734                 set_task_syscall_work(p, SECCOMP);
1735 #endif
1736 }
1737
1738 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1739 {
1740         current->clear_child_tid = tidptr;
1741
1742         return task_pid_vnr(current);
1743 }
1744
1745 static void rt_mutex_init_task(struct task_struct *p)
1746 {
1747         raw_spin_lock_init(&p->pi_lock);
1748 #ifdef CONFIG_RT_MUTEXES
1749         p->pi_waiters = RB_ROOT_CACHED;
1750         p->pi_top_task = NULL;
1751         p->pi_blocked_on = NULL;
1752 #endif
1753 }
1754
1755 static inline void init_task_pid_links(struct task_struct *task)
1756 {
1757         enum pid_type type;
1758
1759         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1760                 INIT_HLIST_NODE(&task->pid_links[type]);
1761 }
1762
1763 static inline void
1764 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1765 {
1766         if (type == PIDTYPE_PID)
1767                 task->thread_pid = pid;
1768         else
1769                 task->signal->pids[type] = pid;
1770 }
1771
1772 static inline void rcu_copy_process(struct task_struct *p)
1773 {
1774 #ifdef CONFIG_PREEMPT_RCU
1775         p->rcu_read_lock_nesting = 0;
1776         p->rcu_read_unlock_special.s = 0;
1777         p->rcu_blocked_node = NULL;
1778         INIT_LIST_HEAD(&p->rcu_node_entry);
1779 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1780 #ifdef CONFIG_TASKS_RCU
1781         p->rcu_tasks_holdout = false;
1782         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1783         p->rcu_tasks_idle_cpu = -1;
1784 #endif /* #ifdef CONFIG_TASKS_RCU */
1785 #ifdef CONFIG_TASKS_TRACE_RCU
1786         p->trc_reader_nesting = 0;
1787         p->trc_reader_special.s = 0;
1788         INIT_LIST_HEAD(&p->trc_holdout_list);
1789 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1790 }
1791
1792 struct pid *pidfd_pid(const struct file *file)
1793 {
1794         if (file->f_op == &pidfd_fops)
1795                 return file->private_data;
1796
1797         return ERR_PTR(-EBADF);
1798 }
1799
1800 static int pidfd_release(struct inode *inode, struct file *file)
1801 {
1802         struct pid *pid = file->private_data;
1803
1804         file->private_data = NULL;
1805         put_pid(pid);
1806         return 0;
1807 }
1808
1809 #ifdef CONFIG_PROC_FS
1810 /**
1811  * pidfd_show_fdinfo - print information about a pidfd
1812  * @m: proc fdinfo file
1813  * @f: file referencing a pidfd
1814  *
1815  * Pid:
1816  * This function will print the pid that a given pidfd refers to in the
1817  * pid namespace of the procfs instance.
1818  * If the pid namespace of the process is not a descendant of the pid
1819  * namespace of the procfs instance 0 will be shown as its pid. This is
1820  * similar to calling getppid() on a process whose parent is outside of
1821  * its pid namespace.
1822  *
1823  * NSpid:
1824  * If pid namespaces are supported then this function will also print
1825  * the pid of a given pidfd refers to for all descendant pid namespaces
1826  * starting from the current pid namespace of the instance, i.e. the
1827  * Pid field and the first entry in the NSpid field will be identical.
1828  * If the pid namespace of the process is not a descendant of the pid
1829  * namespace of the procfs instance 0 will be shown as its first NSpid
1830  * entry and no others will be shown.
1831  * Note that this differs from the Pid and NSpid fields in
1832  * /proc/<pid>/status where Pid and NSpid are always shown relative to
1833  * the  pid namespace of the procfs instance. The difference becomes
1834  * obvious when sending around a pidfd between pid namespaces from a
1835  * different branch of the tree, i.e. where no ancestral relation is
1836  * present between the pid namespaces:
1837  * - create two new pid namespaces ns1 and ns2 in the initial pid
1838  *   namespace (also take care to create new mount namespaces in the
1839  *   new pid namespace and mount procfs)
1840  * - create a process with a pidfd in ns1
1841  * - send pidfd from ns1 to ns2
1842  * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1843  *   have exactly one entry, which is 0
1844  */
1845 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1846 {
1847         struct pid *pid = f->private_data;
1848         struct pid_namespace *ns;
1849         pid_t nr = -1;
1850
1851         if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1852                 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1853                 nr = pid_nr_ns(pid, ns);
1854         }
1855
1856         seq_put_decimal_ll(m, "Pid:\t", nr);
1857
1858 #ifdef CONFIG_PID_NS
1859         seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1860         if (nr > 0) {
1861                 int i;
1862
1863                 /* If nr is non-zero it means that 'pid' is valid and that
1864                  * ns, i.e. the pid namespace associated with the procfs
1865                  * instance, is in the pid namespace hierarchy of pid.
1866                  * Start at one below the already printed level.
1867                  */
1868                 for (i = ns->level + 1; i <= pid->level; i++)
1869                         seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1870         }
1871 #endif
1872         seq_putc(m, '\n');
1873 }
1874 #endif
1875
1876 /*
1877  * Poll support for process exit notification.
1878  */
1879 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1880 {
1881         struct pid *pid = file->private_data;
1882         __poll_t poll_flags = 0;
1883
1884         poll_wait(file, &pid->wait_pidfd, pts);
1885
1886         /*
1887          * Inform pollers only when the whole thread group exits.
1888          * If the thread group leader exits before all other threads in the
1889          * group, then poll(2) should block, similar to the wait(2) family.
1890          */
1891         if (thread_group_exited(pid))
1892                 poll_flags = EPOLLIN | EPOLLRDNORM;
1893
1894         return poll_flags;
1895 }
1896
1897 const struct file_operations pidfd_fops = {
1898         .release = pidfd_release,
1899         .poll = pidfd_poll,
1900 #ifdef CONFIG_PROC_FS
1901         .show_fdinfo = pidfd_show_fdinfo,
1902 #endif
1903 };
1904
1905 static void __delayed_free_task(struct rcu_head *rhp)
1906 {
1907         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1908
1909         free_task(tsk);
1910 }
1911
1912 static __always_inline void delayed_free_task(struct task_struct *tsk)
1913 {
1914         if (IS_ENABLED(CONFIG_MEMCG))
1915                 call_rcu(&tsk->rcu, __delayed_free_task);
1916         else
1917                 free_task(tsk);
1918 }
1919
1920 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1921 {
1922         /* Skip if kernel thread */
1923         if (!tsk->mm)
1924                 return;
1925
1926         /* Skip if spawning a thread or using vfork */
1927         if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1928                 return;
1929
1930         /* We need to synchronize with __set_oom_adj */
1931         mutex_lock(&oom_adj_mutex);
1932         set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1933         /* Update the values in case they were changed after copy_signal */
1934         tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1935         tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1936         mutex_unlock(&oom_adj_mutex);
1937 }
1938
1939 /*
1940  * This creates a new process as a copy of the old one,
1941  * but does not actually start it yet.
1942  *
1943  * It copies the registers, and all the appropriate
1944  * parts of the process environment (as per the clone
1945  * flags). The actual kick-off is left to the caller.
1946  */
1947 static __latent_entropy struct task_struct *copy_process(
1948                                         struct pid *pid,
1949                                         int trace,
1950                                         int node,
1951                                         struct kernel_clone_args *args)
1952 {
1953         int pidfd = -1, retval;
1954         struct task_struct *p;
1955         struct multiprocess_signals delayed;
1956         struct file *pidfile = NULL;
1957         u64 clone_flags = args->flags;
1958         struct nsproxy *nsp = current->nsproxy;
1959
1960         /*
1961          * Don't allow sharing the root directory with processes in a different
1962          * namespace
1963          */
1964         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1965                 return ERR_PTR(-EINVAL);
1966
1967         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1968                 return ERR_PTR(-EINVAL);
1969
1970         /*
1971          * Thread groups must share signals as well, and detached threads
1972          * can only be started up within the thread group.
1973          */
1974         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1975                 return ERR_PTR(-EINVAL);
1976
1977         /*
1978          * Shared signal handlers imply shared VM. By way of the above,
1979          * thread groups also imply shared VM. Blocking this case allows
1980          * for various simplifications in other code.
1981          */
1982         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1983                 return ERR_PTR(-EINVAL);
1984
1985         /*
1986          * Siblings of global init remain as zombies on exit since they are
1987          * not reaped by their parent (swapper). To solve this and to avoid
1988          * multi-rooted process trees, prevent global and container-inits
1989          * from creating siblings.
1990          */
1991         if ((clone_flags & CLONE_PARENT) &&
1992                                 current->signal->flags & SIGNAL_UNKILLABLE)
1993                 return ERR_PTR(-EINVAL);
1994
1995         /*
1996          * If the new process will be in a different pid or user namespace
1997          * do not allow it to share a thread group with the forking task.
1998          */
1999         if (clone_flags & CLONE_THREAD) {
2000                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2001                     (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2002                         return ERR_PTR(-EINVAL);
2003         }
2004
2005         /*
2006          * If the new process will be in a different time namespace
2007          * do not allow it to share VM or a thread group with the forking task.
2008          */
2009         if (clone_flags & (CLONE_THREAD | CLONE_VM)) {
2010                 if (nsp->time_ns != nsp->time_ns_for_children)
2011                         return ERR_PTR(-EINVAL);
2012         }
2013
2014         if (clone_flags & CLONE_PIDFD) {
2015                 /*
2016                  * - CLONE_DETACHED is blocked so that we can potentially
2017                  *   reuse it later for CLONE_PIDFD.
2018                  * - CLONE_THREAD is blocked until someone really needs it.
2019                  */
2020                 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
2021                         return ERR_PTR(-EINVAL);
2022         }
2023
2024         /*
2025          * Force any signals received before this point to be delivered
2026          * before the fork happens.  Collect up signals sent to multiple
2027          * processes that happen during the fork and delay them so that
2028          * they appear to happen after the fork.
2029          */
2030         sigemptyset(&delayed.signal);
2031         INIT_HLIST_NODE(&delayed.node);
2032
2033         spin_lock_irq(&current->sighand->siglock);
2034         if (!(clone_flags & CLONE_THREAD))
2035                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
2036         recalc_sigpending();
2037         spin_unlock_irq(&current->sighand->siglock);
2038         retval = -ERESTARTNOINTR;
2039         if (task_sigpending(current))
2040                 goto fork_out;
2041
2042         retval = -ENOMEM;
2043         p = dup_task_struct(current, node);
2044         if (!p)
2045                 goto fork_out;
2046         if (args->io_thread) {
2047                 /*
2048                  * Mark us an IO worker, and block any signal that isn't
2049                  * fatal or STOP
2050                  */
2051                 p->flags |= PF_IO_WORKER;
2052                 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2053         }
2054
2055         /*
2056          * This _must_ happen before we call free_task(), i.e. before we jump
2057          * to any of the bad_fork_* labels. This is to avoid freeing
2058          * p->set_child_tid which is (ab)used as a kthread's data pointer for
2059          * kernel threads (PF_KTHREAD).
2060          */
2061         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2062         /*
2063          * Clear TID on mm_release()?
2064          */
2065         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2066
2067         ftrace_graph_init_task(p);
2068
2069         rt_mutex_init_task(p);
2070
2071         lockdep_assert_irqs_enabled();
2072 #ifdef CONFIG_PROVE_LOCKING
2073         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2074 #endif
2075         retval = copy_creds(p, clone_flags);
2076         if (retval < 0)
2077                 goto bad_fork_free;
2078
2079         retval = -EAGAIN;
2080         if (is_ucounts_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2081                 if (p->real_cred->user != INIT_USER &&
2082                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2083                         goto bad_fork_cleanup_count;
2084         }
2085         current->flags &= ~PF_NPROC_EXCEEDED;
2086
2087         /*
2088          * If multiple threads are within copy_process(), then this check
2089          * triggers too late. This doesn't hurt, the check is only there
2090          * to stop root fork bombs.
2091          */
2092         retval = -EAGAIN;
2093         if (data_race(nr_threads >= max_threads))
2094                 goto bad_fork_cleanup_count;
2095
2096         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
2097         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2098         p->flags |= PF_FORKNOEXEC;
2099         INIT_LIST_HEAD(&p->children);
2100         INIT_LIST_HEAD(&p->sibling);
2101         rcu_copy_process(p);
2102         p->vfork_done = NULL;
2103         spin_lock_init(&p->alloc_lock);
2104
2105         init_sigpending(&p->pending);
2106
2107         p->utime = p->stime = p->gtime = 0;
2108 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2109         p->utimescaled = p->stimescaled = 0;
2110 #endif
2111         prev_cputime_init(&p->prev_cputime);
2112
2113 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2114         seqcount_init(&p->vtime.seqcount);
2115         p->vtime.starttime = 0;
2116         p->vtime.state = VTIME_INACTIVE;
2117 #endif
2118
2119 #ifdef CONFIG_IO_URING
2120         p->io_uring = NULL;
2121 #endif
2122
2123 #if defined(SPLIT_RSS_COUNTING)
2124         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2125 #endif
2126
2127         p->default_timer_slack_ns = current->timer_slack_ns;
2128
2129 #ifdef CONFIG_PSI
2130         p->psi_flags = 0;
2131 #endif
2132
2133         task_io_accounting_init(&p->ioac);
2134         acct_clear_integrals(p);
2135
2136         posix_cputimers_init(&p->posix_cputimers);
2137
2138         p->io_context = NULL;
2139         audit_set_context(p, NULL);
2140         cgroup_fork(p);
2141 #ifdef CONFIG_NUMA
2142         p->mempolicy = mpol_dup(p->mempolicy);
2143         if (IS_ERR(p->mempolicy)) {
2144                 retval = PTR_ERR(p->mempolicy);
2145                 p->mempolicy = NULL;
2146                 goto bad_fork_cleanup_threadgroup_lock;
2147         }
2148 #endif
2149 #ifdef CONFIG_CPUSETS
2150         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2151         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2152         seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2153 #endif
2154 #ifdef CONFIG_TRACE_IRQFLAGS
2155         memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2156         p->irqtrace.hardirq_disable_ip  = _THIS_IP_;
2157         p->irqtrace.softirq_enable_ip   = _THIS_IP_;
2158         p->softirqs_enabled             = 1;
2159         p->softirq_context              = 0;
2160 #endif
2161
2162         p->pagefault_disabled = 0;
2163
2164 #ifdef CONFIG_LOCKDEP
2165         lockdep_init_task(p);
2166 #endif
2167
2168 #ifdef CONFIG_DEBUG_MUTEXES
2169         p->blocked_on = NULL; /* not blocked yet */
2170 #endif
2171 #ifdef CONFIG_BCACHE
2172         p->sequential_io        = 0;
2173         p->sequential_io_avg    = 0;
2174 #endif
2175 #ifdef CONFIG_BPF_SYSCALL
2176         RCU_INIT_POINTER(p->bpf_storage, NULL);
2177         p->bpf_ctx = NULL;
2178 #endif
2179
2180         /* Perform scheduler related setup. Assign this task to a CPU. */
2181         retval = sched_fork(clone_flags, p);
2182         if (retval)
2183                 goto bad_fork_cleanup_policy;
2184
2185         retval = perf_event_init_task(p, clone_flags);
2186         if (retval)
2187                 goto bad_fork_cleanup_policy;
2188         retval = audit_alloc(p);
2189         if (retval)
2190                 goto bad_fork_cleanup_perf;
2191         /* copy all the process information */
2192         shm_init_task(p);
2193         retval = security_task_alloc(p, clone_flags);
2194         if (retval)
2195                 goto bad_fork_cleanup_audit;
2196         retval = copy_semundo(clone_flags, p);
2197         if (retval)
2198                 goto bad_fork_cleanup_security;
2199         retval = copy_files(clone_flags, p);
2200         if (retval)
2201                 goto bad_fork_cleanup_semundo;
2202         retval = copy_fs(clone_flags, p);
2203         if (retval)
2204                 goto bad_fork_cleanup_files;
2205         retval = copy_sighand(clone_flags, p);
2206         if (retval)
2207                 goto bad_fork_cleanup_fs;
2208         retval = copy_signal(clone_flags, p);
2209         if (retval)
2210                 goto bad_fork_cleanup_sighand;
2211         retval = copy_mm(clone_flags, p);
2212         if (retval)
2213                 goto bad_fork_cleanup_signal;
2214         retval = copy_namespaces(clone_flags, p);
2215         if (retval)
2216                 goto bad_fork_cleanup_mm;
2217         retval = copy_io(clone_flags, p);
2218         if (retval)
2219                 goto bad_fork_cleanup_namespaces;
2220         retval = copy_thread(clone_flags, args->stack, args->stack_size, p, args->tls);
2221         if (retval)
2222                 goto bad_fork_cleanup_io;
2223
2224         stackleak_task_init(p);
2225
2226         if (pid != &init_struct_pid) {
2227                 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2228                                 args->set_tid_size);
2229                 if (IS_ERR(pid)) {
2230                         retval = PTR_ERR(pid);
2231                         goto bad_fork_cleanup_thread;
2232                 }
2233         }
2234
2235         /*
2236          * This has to happen after we've potentially unshared the file
2237          * descriptor table (so that the pidfd doesn't leak into the child
2238          * if the fd table isn't shared).
2239          */
2240         if (clone_flags & CLONE_PIDFD) {
2241                 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2242                 if (retval < 0)
2243                         goto bad_fork_free_pid;
2244
2245                 pidfd = retval;
2246
2247                 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2248                                               O_RDWR | O_CLOEXEC);
2249                 if (IS_ERR(pidfile)) {
2250                         put_unused_fd(pidfd);
2251                         retval = PTR_ERR(pidfile);
2252                         goto bad_fork_free_pid;
2253                 }
2254                 get_pid(pid);   /* held by pidfile now */
2255
2256                 retval = put_user(pidfd, args->pidfd);
2257                 if (retval)
2258                         goto bad_fork_put_pidfd;
2259         }
2260
2261 #ifdef CONFIG_BLOCK
2262         p->plug = NULL;
2263 #endif
2264         futex_init_task(p);
2265
2266         /*
2267          * sigaltstack should be cleared when sharing the same VM
2268          */
2269         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2270                 sas_ss_reset(p);
2271
2272         /*
2273          * Syscall tracing and stepping should be turned off in the
2274          * child regardless of CLONE_PTRACE.
2275          */
2276         user_disable_single_step(p);
2277         clear_task_syscall_work(p, SYSCALL_TRACE);
2278 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2279         clear_task_syscall_work(p, SYSCALL_EMU);
2280 #endif
2281         clear_tsk_latency_tracing(p);
2282
2283         /* ok, now we should be set up.. */
2284         p->pid = pid_nr(pid);
2285         if (clone_flags & CLONE_THREAD) {
2286                 p->group_leader = current->group_leader;
2287                 p->tgid = current->tgid;
2288         } else {
2289                 p->group_leader = p;
2290                 p->tgid = p->pid;
2291         }
2292
2293         p->nr_dirtied = 0;
2294         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2295         p->dirty_paused_when = 0;
2296
2297         p->pdeath_signal = 0;
2298         INIT_LIST_HEAD(&p->thread_group);
2299         p->task_works = NULL;
2300         clear_posix_cputimers_work(p);
2301
2302 #ifdef CONFIG_KRETPROBES
2303         p->kretprobe_instances.first = NULL;
2304 #endif
2305
2306         /*
2307          * Ensure that the cgroup subsystem policies allow the new process to be
2308          * forked. It should be noted that the new process's css_set can be changed
2309          * between here and cgroup_post_fork() if an organisation operation is in
2310          * progress.
2311          */
2312         retval = cgroup_can_fork(p, args);
2313         if (retval)
2314                 goto bad_fork_put_pidfd;
2315
2316         /*
2317          * Now that the cgroups are pinned, re-clone the parent cgroup and put
2318          * the new task on the correct runqueue. All this *before* the task
2319          * becomes visible.
2320          *
2321          * This isn't part of ->can_fork() because while the re-cloning is
2322          * cgroup specific, it unconditionally needs to place the task on a
2323          * runqueue.
2324          */
2325         sched_cgroup_fork(p, args);
2326
2327         /*
2328          * From this point on we must avoid any synchronous user-space
2329          * communication until we take the tasklist-lock. In particular, we do
2330          * not want user-space to be able to predict the process start-time by
2331          * stalling fork(2) after we recorded the start_time but before it is
2332          * visible to the system.
2333          */
2334
2335         p->start_time = ktime_get_ns();
2336         p->start_boottime = ktime_get_boottime_ns();
2337
2338         /*
2339          * Make it visible to the rest of the system, but dont wake it up yet.
2340          * Need tasklist lock for parent etc handling!
2341          */
2342         write_lock_irq(&tasklist_lock);
2343
2344         /* CLONE_PARENT re-uses the old parent */
2345         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2346                 p->real_parent = current->real_parent;
2347                 p->parent_exec_id = current->parent_exec_id;
2348                 if (clone_flags & CLONE_THREAD)
2349                         p->exit_signal = -1;
2350                 else
2351                         p->exit_signal = current->group_leader->exit_signal;
2352         } else {
2353                 p->real_parent = current;
2354                 p->parent_exec_id = current->self_exec_id;
2355                 p->exit_signal = args->exit_signal;
2356         }
2357
2358         klp_copy_process(p);
2359
2360         sched_core_fork(p);
2361
2362         spin_lock(&current->sighand->siglock);
2363
2364         /*
2365          * Copy seccomp details explicitly here, in case they were changed
2366          * before holding sighand lock.
2367          */
2368         copy_seccomp(p);
2369
2370         rseq_fork(p, clone_flags);
2371
2372         /* Don't start children in a dying pid namespace */
2373         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2374                 retval = -ENOMEM;
2375                 goto bad_fork_cancel_cgroup;
2376         }
2377
2378         /* Let kill terminate clone/fork in the middle */
2379         if (fatal_signal_pending(current)) {
2380                 retval = -EINTR;
2381                 goto bad_fork_cancel_cgroup;
2382         }
2383
2384         init_task_pid_links(p);
2385         if (likely(p->pid)) {
2386                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2387
2388                 init_task_pid(p, PIDTYPE_PID, pid);
2389                 if (thread_group_leader(p)) {
2390                         init_task_pid(p, PIDTYPE_TGID, pid);
2391                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2392                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2393
2394                         if (is_child_reaper(pid)) {
2395                                 ns_of_pid(pid)->child_reaper = p;
2396                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2397                         }
2398                         p->signal->shared_pending.signal = delayed.signal;
2399                         p->signal->tty = tty_kref_get(current->signal->tty);
2400                         /*
2401                          * Inherit has_child_subreaper flag under the same
2402                          * tasklist_lock with adding child to the process tree
2403                          * for propagate_has_child_subreaper optimization.
2404                          */
2405                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2406                                                          p->real_parent->signal->is_child_subreaper;
2407                         list_add_tail(&p->sibling, &p->real_parent->children);
2408                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2409                         attach_pid(p, PIDTYPE_TGID);
2410                         attach_pid(p, PIDTYPE_PGID);
2411                         attach_pid(p, PIDTYPE_SID);
2412                         __this_cpu_inc(process_counts);
2413                 } else {
2414                         current->signal->nr_threads++;
2415                         atomic_inc(&current->signal->live);
2416                         refcount_inc(&current->signal->sigcnt);
2417                         task_join_group_stop(p);
2418                         list_add_tail_rcu(&p->thread_group,
2419                                           &p->group_leader->thread_group);
2420                         list_add_tail_rcu(&p->thread_node,
2421                                           &p->signal->thread_head);
2422                 }
2423                 attach_pid(p, PIDTYPE_PID);
2424                 nr_threads++;
2425         }
2426         total_forks++;
2427         hlist_del_init(&delayed.node);
2428         spin_unlock(&current->sighand->siglock);
2429         syscall_tracepoint_update(p);
2430         write_unlock_irq(&tasklist_lock);
2431
2432         if (pidfile)
2433                 fd_install(pidfd, pidfile);
2434
2435         proc_fork_connector(p);
2436         sched_post_fork(p);
2437         cgroup_post_fork(p, args);
2438         perf_event_fork(p);
2439
2440         trace_task_newtask(p, clone_flags);
2441         uprobe_copy_process(p, clone_flags);
2442
2443         copy_oom_score_adj(clone_flags, p);
2444
2445         return p;
2446
2447 bad_fork_cancel_cgroup:
2448         sched_core_free(p);
2449         spin_unlock(&current->sighand->siglock);
2450         write_unlock_irq(&tasklist_lock);
2451         cgroup_cancel_fork(p, args);
2452 bad_fork_put_pidfd:
2453         if (clone_flags & CLONE_PIDFD) {
2454                 fput(pidfile);
2455                 put_unused_fd(pidfd);
2456         }
2457 bad_fork_free_pid:
2458         if (pid != &init_struct_pid)
2459                 free_pid(pid);
2460 bad_fork_cleanup_thread:
2461         exit_thread(p);
2462 bad_fork_cleanup_io:
2463         if (p->io_context)
2464                 exit_io_context(p);
2465 bad_fork_cleanup_namespaces:
2466         exit_task_namespaces(p);
2467 bad_fork_cleanup_mm:
2468         if (p->mm) {
2469                 mm_clear_owner(p->mm, p);
2470                 mmput(p->mm);
2471         }
2472 bad_fork_cleanup_signal:
2473         if (!(clone_flags & CLONE_THREAD))
2474                 free_signal_struct(p->signal);
2475 bad_fork_cleanup_sighand:
2476         __cleanup_sighand(p->sighand);
2477 bad_fork_cleanup_fs:
2478         exit_fs(p); /* blocking */
2479 bad_fork_cleanup_files:
2480         exit_files(p); /* blocking */
2481 bad_fork_cleanup_semundo:
2482         exit_sem(p);
2483 bad_fork_cleanup_security:
2484         security_task_free(p);
2485 bad_fork_cleanup_audit:
2486         audit_free(p);
2487 bad_fork_cleanup_perf:
2488         perf_event_free_task(p);
2489 bad_fork_cleanup_policy:
2490         lockdep_free_task(p);
2491 #ifdef CONFIG_NUMA
2492         mpol_put(p->mempolicy);
2493 bad_fork_cleanup_threadgroup_lock:
2494 #endif
2495         delayacct_tsk_free(p);
2496 bad_fork_cleanup_count:
2497         dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2498         exit_creds(p);
2499 bad_fork_free:
2500         WRITE_ONCE(p->__state, TASK_DEAD);
2501         put_task_stack(p);
2502         delayed_free_task(p);
2503 fork_out:
2504         spin_lock_irq(&current->sighand->siglock);
2505         hlist_del_init(&delayed.node);
2506         spin_unlock_irq(&current->sighand->siglock);
2507         return ERR_PTR(retval);
2508 }
2509
2510 static inline void init_idle_pids(struct task_struct *idle)
2511 {
2512         enum pid_type type;
2513
2514         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2515                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2516                 init_task_pid(idle, type, &init_struct_pid);
2517         }
2518 }
2519
2520 struct task_struct * __init fork_idle(int cpu)
2521 {
2522         struct task_struct *task;
2523         struct kernel_clone_args args = {
2524                 .flags = CLONE_VM,
2525         };
2526
2527         task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2528         if (!IS_ERR(task)) {
2529                 init_idle_pids(task);
2530                 init_idle(task, cpu);
2531         }
2532
2533         return task;
2534 }
2535
2536 struct mm_struct *copy_init_mm(void)
2537 {
2538         return dup_mm(NULL, &init_mm);
2539 }
2540
2541 /*
2542  * This is like kernel_clone(), but shaved down and tailored to just
2543  * creating io_uring workers. It returns a created task, or an error pointer.
2544  * The returned task is inactive, and the caller must fire it up through
2545  * wake_up_new_task(p). All signals are blocked in the created task.
2546  */
2547 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2548 {
2549         unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2550                                 CLONE_IO;
2551         struct kernel_clone_args args = {
2552                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2553                                     CLONE_UNTRACED) & ~CSIGNAL),
2554                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2555                 .stack          = (unsigned long)fn,
2556                 .stack_size     = (unsigned long)arg,
2557                 .io_thread      = 1,
2558         };
2559
2560         return copy_process(NULL, 0, node, &args);
2561 }
2562
2563 /*
2564  *  Ok, this is the main fork-routine.
2565  *
2566  * It copies the process, and if successful kick-starts
2567  * it and waits for it to finish using the VM if required.
2568  *
2569  * args->exit_signal is expected to be checked for sanity by the caller.
2570  */
2571 pid_t kernel_clone(struct kernel_clone_args *args)
2572 {
2573         u64 clone_flags = args->flags;
2574         struct completion vfork;
2575         struct pid *pid;
2576         struct task_struct *p;
2577         int trace = 0;
2578         pid_t nr;
2579
2580         /*
2581          * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2582          * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2583          * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2584          * field in struct clone_args and it still doesn't make sense to have
2585          * them both point at the same memory location. Performing this check
2586          * here has the advantage that we don't need to have a separate helper
2587          * to check for legacy clone().
2588          */
2589         if ((args->flags & CLONE_PIDFD) &&
2590             (args->flags & CLONE_PARENT_SETTID) &&
2591             (args->pidfd == args->parent_tid))
2592                 return -EINVAL;
2593
2594         /*
2595          * Determine whether and which event to report to ptracer.  When
2596          * called from kernel_thread or CLONE_UNTRACED is explicitly
2597          * requested, no event is reported; otherwise, report if the event
2598          * for the type of forking is enabled.
2599          */
2600         if (!(clone_flags & CLONE_UNTRACED)) {
2601                 if (clone_flags & CLONE_VFORK)
2602                         trace = PTRACE_EVENT_VFORK;
2603                 else if (args->exit_signal != SIGCHLD)
2604                         trace = PTRACE_EVENT_CLONE;
2605                 else
2606                         trace = PTRACE_EVENT_FORK;
2607
2608                 if (likely(!ptrace_event_enabled(current, trace)))
2609                         trace = 0;
2610         }
2611
2612         p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2613         add_latent_entropy();
2614
2615         if (IS_ERR(p))
2616                 return PTR_ERR(p);
2617
2618         /*
2619          * Do this prior waking up the new thread - the thread pointer
2620          * might get invalid after that point, if the thread exits quickly.
2621          */
2622         trace_sched_process_fork(current, p);
2623
2624         pid = get_task_pid(p, PIDTYPE_PID);
2625         nr = pid_vnr(pid);
2626
2627         if (clone_flags & CLONE_PARENT_SETTID)
2628                 put_user(nr, args->parent_tid);
2629
2630         if (clone_flags & CLONE_VFORK) {
2631                 p->vfork_done = &vfork;
2632                 init_completion(&vfork);
2633                 get_task_struct(p);
2634         }
2635
2636         wake_up_new_task(p);
2637
2638         /* forking complete and child started to run, tell ptracer */
2639         if (unlikely(trace))
2640                 ptrace_event_pid(trace, pid);
2641
2642         if (clone_flags & CLONE_VFORK) {
2643                 if (!wait_for_vfork_done(p, &vfork))
2644                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2645         }
2646
2647         put_pid(pid);
2648         return nr;
2649 }
2650
2651 /*
2652  * Create a kernel thread.
2653  */
2654 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2655 {
2656         struct kernel_clone_args args = {
2657                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2658                                     CLONE_UNTRACED) & ~CSIGNAL),
2659                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2660                 .stack          = (unsigned long)fn,
2661                 .stack_size     = (unsigned long)arg,
2662         };
2663
2664         return kernel_clone(&args);
2665 }
2666
2667 #ifdef __ARCH_WANT_SYS_FORK
2668 SYSCALL_DEFINE0(fork)
2669 {
2670 #ifdef CONFIG_MMU
2671         struct kernel_clone_args args = {
2672                 .exit_signal = SIGCHLD,
2673         };
2674
2675         return kernel_clone(&args);
2676 #else
2677         /* can not support in nommu mode */
2678         return -EINVAL;
2679 #endif
2680 }
2681 #endif
2682
2683 #ifdef __ARCH_WANT_SYS_VFORK
2684 SYSCALL_DEFINE0(vfork)
2685 {
2686         struct kernel_clone_args args = {
2687                 .flags          = CLONE_VFORK | CLONE_VM,
2688                 .exit_signal    = SIGCHLD,
2689         };
2690
2691         return kernel_clone(&args);
2692 }
2693 #endif
2694
2695 #ifdef __ARCH_WANT_SYS_CLONE
2696 #ifdef CONFIG_CLONE_BACKWARDS
2697 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2698                  int __user *, parent_tidptr,
2699                  unsigned long, tls,
2700                  int __user *, child_tidptr)
2701 #elif defined(CONFIG_CLONE_BACKWARDS2)
2702 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2703                  int __user *, parent_tidptr,
2704                  int __user *, child_tidptr,
2705                  unsigned long, tls)
2706 #elif defined(CONFIG_CLONE_BACKWARDS3)
2707 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2708                 int, stack_size,
2709                 int __user *, parent_tidptr,
2710                 int __user *, child_tidptr,
2711                 unsigned long, tls)
2712 #else
2713 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2714                  int __user *, parent_tidptr,
2715                  int __user *, child_tidptr,
2716                  unsigned long, tls)
2717 #endif
2718 {
2719         struct kernel_clone_args args = {
2720                 .flags          = (lower_32_bits(clone_flags) & ~CSIGNAL),
2721                 .pidfd          = parent_tidptr,
2722                 .child_tid      = child_tidptr,
2723                 .parent_tid     = parent_tidptr,
2724                 .exit_signal    = (lower_32_bits(clone_flags) & CSIGNAL),
2725                 .stack          = newsp,
2726                 .tls            = tls,
2727         };
2728
2729         return kernel_clone(&args);
2730 }
2731 #endif
2732
2733 #ifdef __ARCH_WANT_SYS_CLONE3
2734
2735 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2736                                               struct clone_args __user *uargs,
2737                                               size_t usize)
2738 {
2739         int err;
2740         struct clone_args args;
2741         pid_t *kset_tid = kargs->set_tid;
2742
2743         BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2744                      CLONE_ARGS_SIZE_VER0);
2745         BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2746                      CLONE_ARGS_SIZE_VER1);
2747         BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2748                      CLONE_ARGS_SIZE_VER2);
2749         BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2750
2751         if (unlikely(usize > PAGE_SIZE))
2752                 return -E2BIG;
2753         if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2754                 return -EINVAL;
2755
2756         err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2757         if (err)
2758                 return err;
2759
2760         if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2761                 return -EINVAL;
2762
2763         if (unlikely(!args.set_tid && args.set_tid_size > 0))
2764                 return -EINVAL;
2765
2766         if (unlikely(args.set_tid && args.set_tid_size == 0))
2767                 return -EINVAL;
2768
2769         /*
2770          * Verify that higher 32bits of exit_signal are unset and that
2771          * it is a valid signal
2772          */
2773         if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2774                      !valid_signal(args.exit_signal)))
2775                 return -EINVAL;
2776
2777         if ((args.flags & CLONE_INTO_CGROUP) &&
2778             (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2779                 return -EINVAL;
2780
2781         *kargs = (struct kernel_clone_args){
2782                 .flags          = args.flags,
2783                 .pidfd          = u64_to_user_ptr(args.pidfd),
2784                 .child_tid      = u64_to_user_ptr(args.child_tid),
2785                 .parent_tid     = u64_to_user_ptr(args.parent_tid),
2786                 .exit_signal    = args.exit_signal,
2787                 .stack          = args.stack,
2788                 .stack_size     = args.stack_size,
2789                 .tls            = args.tls,
2790                 .set_tid_size   = args.set_tid_size,
2791                 .cgroup         = args.cgroup,
2792         };
2793
2794         if (args.set_tid &&
2795                 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2796                         (kargs->set_tid_size * sizeof(pid_t))))
2797                 return -EFAULT;
2798
2799         kargs->set_tid = kset_tid;
2800
2801         return 0;
2802 }
2803
2804 /**
2805  * clone3_stack_valid - check and prepare stack
2806  * @kargs: kernel clone args
2807  *
2808  * Verify that the stack arguments userspace gave us are sane.
2809  * In addition, set the stack direction for userspace since it's easy for us to
2810  * determine.
2811  */
2812 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2813 {
2814         if (kargs->stack == 0) {
2815                 if (kargs->stack_size > 0)
2816                         return false;
2817         } else {
2818                 if (kargs->stack_size == 0)
2819                         return false;
2820
2821                 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2822                         return false;
2823
2824 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2825                 kargs->stack += kargs->stack_size;
2826 #endif
2827         }
2828
2829         return true;
2830 }
2831
2832 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2833 {
2834         /* Verify that no unknown flags are passed along. */
2835         if (kargs->flags &
2836             ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2837                 return false;
2838
2839         /*
2840          * - make the CLONE_DETACHED bit reusable for clone3
2841          * - make the CSIGNAL bits reusable for clone3
2842          */
2843         if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2844                 return false;
2845
2846         if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2847             (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2848                 return false;
2849
2850         if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2851             kargs->exit_signal)
2852                 return false;
2853
2854         if (!clone3_stack_valid(kargs))
2855                 return false;
2856
2857         return true;
2858 }
2859
2860 /**
2861  * clone3 - create a new process with specific properties
2862  * @uargs: argument structure
2863  * @size:  size of @uargs
2864  *
2865  * clone3() is the extensible successor to clone()/clone2().
2866  * It takes a struct as argument that is versioned by its size.
2867  *
2868  * Return: On success, a positive PID for the child process.
2869  *         On error, a negative errno number.
2870  */
2871 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2872 {
2873         int err;
2874
2875         struct kernel_clone_args kargs;
2876         pid_t set_tid[MAX_PID_NS_LEVEL];
2877
2878         kargs.set_tid = set_tid;
2879
2880         err = copy_clone_args_from_user(&kargs, uargs, size);
2881         if (err)
2882                 return err;
2883
2884         if (!clone3_args_valid(&kargs))
2885                 return -EINVAL;
2886
2887         return kernel_clone(&kargs);
2888 }
2889 #endif
2890
2891 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2892 {
2893         struct task_struct *leader, *parent, *child;
2894         int res;
2895
2896         read_lock(&tasklist_lock);
2897         leader = top = top->group_leader;
2898 down:
2899         for_each_thread(leader, parent) {
2900                 list_for_each_entry(child, &parent->children, sibling) {
2901                         res = visitor(child, data);
2902                         if (res) {
2903                                 if (res < 0)
2904                                         goto out;
2905                                 leader = child;
2906                                 goto down;
2907                         }
2908 up:
2909                         ;
2910                 }
2911         }
2912
2913         if (leader != top) {
2914                 child = leader;
2915                 parent = child->real_parent;
2916                 leader = parent->group_leader;
2917                 goto up;
2918         }
2919 out:
2920         read_unlock(&tasklist_lock);
2921 }
2922
2923 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2924 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2925 #endif
2926
2927 static void sighand_ctor(void *data)
2928 {
2929         struct sighand_struct *sighand = data;
2930
2931         spin_lock_init(&sighand->siglock);
2932         init_waitqueue_head(&sighand->signalfd_wqh);
2933 }
2934
2935 void __init proc_caches_init(void)
2936 {
2937         unsigned int mm_size;
2938
2939         sighand_cachep = kmem_cache_create("sighand_cache",
2940                         sizeof(struct sighand_struct), 0,
2941                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2942                         SLAB_ACCOUNT, sighand_ctor);
2943         signal_cachep = kmem_cache_create("signal_cache",
2944                         sizeof(struct signal_struct), 0,
2945                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2946                         NULL);
2947         files_cachep = kmem_cache_create("files_cache",
2948                         sizeof(struct files_struct), 0,
2949                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2950                         NULL);
2951         fs_cachep = kmem_cache_create("fs_cache",
2952                         sizeof(struct fs_struct), 0,
2953                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2954                         NULL);
2955
2956         /*
2957          * The mm_cpumask is located at the end of mm_struct, and is
2958          * dynamically sized based on the maximum CPU number this system
2959          * can have, taking hotplug into account (nr_cpu_ids).
2960          */
2961         mm_size = sizeof(struct mm_struct) + cpumask_size();
2962
2963         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2964                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2965                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2966                         offsetof(struct mm_struct, saved_auxv),
2967                         sizeof_field(struct mm_struct, saved_auxv),
2968                         NULL);
2969         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2970         mmap_init();
2971         nsproxy_cache_init();
2972 }
2973
2974 /*
2975  * Check constraints on flags passed to the unshare system call.
2976  */
2977 static int check_unshare_flags(unsigned long unshare_flags)
2978 {
2979         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2980                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2981                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2982                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
2983                                 CLONE_NEWTIME))
2984                 return -EINVAL;
2985         /*
2986          * Not implemented, but pretend it works if there is nothing
2987          * to unshare.  Note that unsharing the address space or the
2988          * signal handlers also need to unshare the signal queues (aka
2989          * CLONE_THREAD).
2990          */
2991         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2992                 if (!thread_group_empty(current))
2993                         return -EINVAL;
2994         }
2995         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2996                 if (refcount_read(&current->sighand->count) > 1)
2997                         return -EINVAL;
2998         }
2999         if (unshare_flags & CLONE_VM) {
3000                 if (!current_is_single_threaded())
3001                         return -EINVAL;
3002         }
3003
3004         return 0;
3005 }
3006
3007 /*
3008  * Unshare the filesystem structure if it is being shared
3009  */
3010 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3011 {
3012         struct fs_struct *fs = current->fs;
3013
3014         if (!(unshare_flags & CLONE_FS) || !fs)
3015                 return 0;
3016
3017         /* don't need lock here; in the worst case we'll do useless copy */
3018         if (fs->users == 1)
3019                 return 0;
3020
3021         *new_fsp = copy_fs_struct(fs);
3022         if (!*new_fsp)
3023                 return -ENOMEM;
3024
3025         return 0;
3026 }
3027
3028 /*
3029  * Unshare file descriptor table if it is being shared
3030  */
3031 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3032                struct files_struct **new_fdp)
3033 {
3034         struct files_struct *fd = current->files;
3035         int error = 0;
3036
3037         if ((unshare_flags & CLONE_FILES) &&
3038             (fd && atomic_read(&fd->count) > 1)) {
3039                 *new_fdp = dup_fd(fd, max_fds, &error);
3040                 if (!*new_fdp)
3041                         return error;
3042         }
3043
3044         return 0;
3045 }
3046
3047 /*
3048  * unshare allows a process to 'unshare' part of the process
3049  * context which was originally shared using clone.  copy_*
3050  * functions used by kernel_clone() cannot be used here directly
3051  * because they modify an inactive task_struct that is being
3052  * constructed. Here we are modifying the current, active,
3053  * task_struct.
3054  */
3055 int ksys_unshare(unsigned long unshare_flags)
3056 {
3057         struct fs_struct *fs, *new_fs = NULL;
3058         struct files_struct *fd, *new_fd = NULL;
3059         struct cred *new_cred = NULL;
3060         struct nsproxy *new_nsproxy = NULL;
3061         int do_sysvsem = 0;
3062         int err;
3063
3064         /*
3065          * If unsharing a user namespace must also unshare the thread group
3066          * and unshare the filesystem root and working directories.
3067          */
3068         if (unshare_flags & CLONE_NEWUSER)
3069                 unshare_flags |= CLONE_THREAD | CLONE_FS;
3070         /*
3071          * If unsharing vm, must also unshare signal handlers.
3072          */
3073         if (unshare_flags & CLONE_VM)
3074                 unshare_flags |= CLONE_SIGHAND;
3075         /*
3076          * If unsharing a signal handlers, must also unshare the signal queues.
3077          */
3078         if (unshare_flags & CLONE_SIGHAND)
3079                 unshare_flags |= CLONE_THREAD;
3080         /*
3081          * If unsharing namespace, must also unshare filesystem information.
3082          */
3083         if (unshare_flags & CLONE_NEWNS)
3084                 unshare_flags |= CLONE_FS;
3085
3086         err = check_unshare_flags(unshare_flags);
3087         if (err)
3088                 goto bad_unshare_out;
3089         /*
3090          * CLONE_NEWIPC must also detach from the undolist: after switching
3091          * to a new ipc namespace, the semaphore arrays from the old
3092          * namespace are unreachable.
3093          */
3094         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3095                 do_sysvsem = 1;
3096         err = unshare_fs(unshare_flags, &new_fs);
3097         if (err)
3098                 goto bad_unshare_out;
3099         err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3100         if (err)
3101                 goto bad_unshare_cleanup_fs;
3102         err = unshare_userns(unshare_flags, &new_cred);
3103         if (err)
3104                 goto bad_unshare_cleanup_fd;
3105         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3106                                          new_cred, new_fs);
3107         if (err)
3108                 goto bad_unshare_cleanup_cred;
3109
3110         if (new_cred) {
3111                 err = set_cred_ucounts(new_cred);
3112                 if (err)
3113                         goto bad_unshare_cleanup_cred;
3114         }
3115
3116         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3117                 if (do_sysvsem) {
3118                         /*
3119                          * CLONE_SYSVSEM is equivalent to sys_exit().
3120                          */
3121                         exit_sem(current);
3122                 }
3123                 if (unshare_flags & CLONE_NEWIPC) {
3124                         /* Orphan segments in old ns (see sem above). */
3125                         exit_shm(current);
3126                         shm_init_task(current);
3127                 }
3128
3129                 if (new_nsproxy)
3130                         switch_task_namespaces(current, new_nsproxy);
3131
3132                 task_lock(current);
3133
3134                 if (new_fs) {
3135                         fs = current->fs;
3136                         spin_lock(&fs->lock);
3137                         current->fs = new_fs;
3138                         if (--fs->users)
3139                                 new_fs = NULL;
3140                         else
3141                                 new_fs = fs;
3142                         spin_unlock(&fs->lock);
3143                 }
3144
3145                 if (new_fd) {
3146                         fd = current->files;
3147                         current->files = new_fd;
3148                         new_fd = fd;
3149                 }
3150
3151                 task_unlock(current);
3152
3153                 if (new_cred) {
3154                         /* Install the new user namespace */
3155                         commit_creds(new_cred);
3156                         new_cred = NULL;
3157                 }
3158         }
3159
3160         perf_event_namespaces(current);
3161
3162 bad_unshare_cleanup_cred:
3163         if (new_cred)
3164                 put_cred(new_cred);
3165 bad_unshare_cleanup_fd:
3166         if (new_fd)
3167                 put_files_struct(new_fd);
3168
3169 bad_unshare_cleanup_fs:
3170         if (new_fs)
3171                 free_fs_struct(new_fs);
3172
3173 bad_unshare_out:
3174         return err;
3175 }
3176
3177 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3178 {
3179         return ksys_unshare(unshare_flags);
3180 }
3181
3182 /*
3183  *      Helper to unshare the files of the current task.
3184  *      We don't want to expose copy_files internals to
3185  *      the exec layer of the kernel.
3186  */
3187
3188 int unshare_files(void)
3189 {
3190         struct task_struct *task = current;
3191         struct files_struct *old, *copy = NULL;
3192         int error;
3193
3194         error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, &copy);
3195         if (error || !copy)
3196                 return error;
3197
3198         old = task->files;
3199         task_lock(task);
3200         task->files = copy;
3201         task_unlock(task);
3202         put_files_struct(old);
3203         return 0;
3204 }
3205
3206 int sysctl_max_threads(struct ctl_table *table, int write,
3207                        void *buffer, size_t *lenp, loff_t *ppos)
3208 {
3209         struct ctl_table t;
3210         int ret;
3211         int threads = max_threads;
3212         int min = 1;
3213         int max = MAX_THREADS;
3214
3215         t = *table;
3216         t.data = &threads;
3217         t.extra1 = &min;
3218         t.extra2 = &max;
3219
3220         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3221         if (ret || !write)
3222                 return ret;
3223
3224         max_threads = threads;
3225
3226         return 0;
3227 }