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