kernel/fork.c

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