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