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