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