kernel/fork.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/kernel/fork.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  */
   7
   8 /*
   9  *  'fork.c' contains the help-routines for the 'fork' system call
  10  * (see also entry.S and others).
  11  * Fork is rather simple, once you get the hang of it, but the memory
  12  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  13  */
  14
  15 #include <linux/anon_inodes.h>
  16 #include <linux/slab.h>
  17 #include <linux/sched/autogroup.h>
  18 #include <linux/sched/mm.h>
  19 #include <linux/sched/coredump.h>
  20 #include <linux/sched/user.h>
  21 #include <linux/sched/numa_balancing.h>
  22 #include <linux/sched/stat.h>
  23 #include <linux/sched/task.h>
  24 #include <linux/sched/task_stack.h>
  25 #include <linux/sched/cputime.h>
  26 #include <linux/seq_file.h>
  27 #include <linux/rtmutex.h>
  28 #include <linux/init.h>
  29 #include <linux/unistd.h>
  30 #include <linux/module.h>
  31 #include <linux/vmalloc.h>
  32 #include <linux/completion.h>
  33 #include <linux/personality.h>
  34 #include <linux/mempolicy.h>
  35 #include <linux/sem.h>
  36 #include <linux/file.h>
  37 #include <linux/fdtable.h>
  38 #include <linux/iocontext.h>
  39 #include <linux/key.h>
  40 #include <linux/binfmts.h>
  41 #include <linux/mman.h>
  42 #include <linux/mmu_notifier.h>
  43 #include <linux/fs.h>
  44 #include <linux/mm.h>
  45 #include <linux/vmacache.h>
  46 #include <linux/nsproxy.h>
  47 #include <linux/capability.h>
  48 #include <linux/cpu.h>
  49 #include <linux/cgroup.h>
  50 #include <linux/security.h>
  51 #include <linux/hugetlb.h>
  52 #include <linux/seccomp.h>
  53 #include <linux/swap.h>
  54 #include <linux/syscalls.h>
  55 #include <linux/jiffies.h>
  56 #include <linux/futex.h>
  57 #include <linux/compat.h>
  58 #include <linux/kthread.h>
  59 #include <linux/task_io_accounting_ops.h>
  60 #include <linux/rcupdate.h>
  61 #include <linux/ptrace.h>
  62 #include <linux/mount.h>
  63 #include <linux/audit.h>
  64 #include <linux/memcontrol.h>
  65 #include <linux/ftrace.h>
  66 #include <linux/proc_fs.h>
  67 #include <linux/profile.h>
  68 #include <linux/rmap.h>
  69 #include <linux/ksm.h>
  70 #include <linux/acct.h>
  71 #include <linux/userfaultfd_k.h>
  72 #include <linux/tsacct_kern.h>
  73 #include <linux/cn_proc.h>
  74 #include <linux/freezer.h>
  75 #include <linux/delayacct.h>
  76 #include <linux/taskstats_kern.h>
  77 #include <linux/random.h>
  78 #include <linux/tty.h>
  79 #include <linux/blkdev.h>
  80 #include <linux/fs_struct.h>
  81 #include <linux/magic.h>
  82 #include <linux/perf_event.h>
  83 #include <linux/posix-timers.h>
  84 #include <linux/user-return-notifier.h>
  85 #include <linux/oom.h>
  86 #include <linux/khugepaged.h>
  87 #include <linux/signalfd.h>
  88 #include <linux/uprobes.h>
  89 #include <linux/aio.h>
  90 #include <linux/compiler.h>
  91 #include <linux/sysctl.h>
  92 #include <linux/kcov.h>
  93 #include <linux/livepatch.h>
  94 #include <linux/thread_info.h>
  95 #include <linux/stackleak.h>
  96 #include <linux/kasan.h>
  97 #include <linux/scs.h>
  98
  99 #include <asm/pgalloc.h>
 100 #include <linux/uaccess.h>
 101 #include <asm/mmu_context.h>
 102 #include <asm/cacheflush.h>
 103 #include <asm/tlbflush.h>
 104
 105 #include <trace/events/sched.h>
 106
 107 #define CREATE_TRACE_POINTS
 108 #include <trace/events/task.h>
 109
 110 /*
 111  * Minimum number of threads to boot the kernel
 112  */
 113 #define MIN_THREADS 20
 114
 115 /*
 116  * Maximum number of threads
 117  */
 118 #define MAX_THREADS FUTEX_TID_MASK
 119
 120 /*
 121  * Protected counters by write_lock_irq(&tasklist_lock)
 122  */
 123 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 124 int nr_threads;                 /* The idle threads do not count.. */
 125
 126 static int max_threads;         /* tunable limit on nr_threads */
 127
 128 #define NAMED_ARRAY_INDEX(x)    [x] = __stringify(x)
 129
 130 static const char * const resident_page_types[] = {
 131         NAMED_ARRAY_INDEX(MM_FILEPAGES),
 132         NAMED_ARRAY_INDEX(MM_ANONPAGES),
 133         NAMED_ARRAY_INDEX(MM_SWAPENTS),
 134         NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
 135 };
 136
 137 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 138
 139 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 140
 141 #ifdef CONFIG_PROVE_RCU
 142 int lockdep_tasklist_lock_is_held(void)
 143 {
 144         return lockdep_is_held(&tasklist_lock);
 145 }
 146 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 147 #endif /* #ifdef CONFIG_PROVE_RCU */
 148
 149 int nr_processes(void)
 150 {
 151         int cpu;
 152         int total = 0;
 153
 154         for_each_possible_cpu(cpu)
 155                 total += per_cpu(process_counts, cpu);
 156
 157         return total;
 158 }
 159
 160 void __weak arch_release_task_struct(struct task_struct *tsk)
 161 {
 162 }
 163
 164 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 165 static struct kmem_cache *task_struct_cachep;
 166
 167 static inline struct task_struct *alloc_task_struct_node(int node)
 168 {
 169         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 170 }
 171
 172 static inline void free_task_struct(struct task_struct *tsk)
 173 {
 174         kmem_cache_free(task_struct_cachep, tsk);
 175 }
 176 #endif
 177
 178 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 179
 180 /*
 181  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 182  * kmemcache based allocator.
 183  */
 184 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 185
 186 #ifdef CONFIG_VMAP_STACK
 187 /*
 188  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 189  * flush.  Try to minimize the number of calls by caching stacks.
 190  */
 191 #define NR_CACHED_STACKS 2
 192 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 193
 194 static int free_vm_stack_cache(unsigned int cpu)
 195 {
 196         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 197         int i;
 198
 199         for (i = 0; i < NR_CACHED_STACKS; i++) {
 200                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 201
 202                 if (!vm_stack)
 203                         continue;
 204
 205                 vfree(vm_stack->addr);
 206                 cached_vm_stacks[i] = NULL;
 207         }
 208
 209         return 0;
 210 }
 211 #endif
 212
 213 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 214 {
 215 #ifdef CONFIG_VMAP_STACK
 216         void *stack;
 217         int i;
 218
 219         for (i = 0; i < NR_CACHED_STACKS; i++) {
 220                 struct vm_struct *s;
 221
 222                 s = this_cpu_xchg(cached_stacks[i], NULL);
 223
 224                 if (!s)
 225                         continue;
 226
 227                 /* Clear the KASAN shadow of the stack. */
 228                 kasan_unpoison_shadow(s->addr, THREAD_SIZE);
 229
 230                 /* Clear stale pointers from reused stack. */
 231                 memset(s->addr, 0, THREAD_SIZE);
 232
 233                 tsk->stack_vm_area = s;
 234                 tsk->stack = s->addr;
 235                 return s->addr;
 236         }
 237
 238         /*
 239          * Allocated stacks are cached and later reused by new threads,
 240          * so memcg accounting is performed manually on assigning/releasing
 241          * stacks to tasks. Drop __GFP_ACCOUNT.
 242          */
 243         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 244                                      VMALLOC_START, VMALLOC_END,
 245                                      THREADINFO_GFP & ~__GFP_ACCOUNT,
 246                                      PAGE_KERNEL,
 247                                      0, node, __builtin_return_address(0));
 248
 249         /*
 250          * We can't call find_vm_area() in interrupt context, and
 251          * free_thread_stack() can be called in interrupt context,
 252          * so cache the vm_struct.
 253          */
 254         if (stack) {
 255                 tsk->stack_vm_area = find_vm_area(stack);
 256                 tsk->stack = stack;
 257         }
 258         return stack;
 259 #else
 260         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 261                                              THREAD_SIZE_ORDER);
 262
 263         if (likely(page)) {
 264                 tsk->stack = kasan_reset_tag(page_address(page));
 265                 return tsk->stack;
 266         }
 267         return NULL;
 268 #endif
 269 }
 270
 271 static inline void free_thread_stack(struct task_struct *tsk)
 272 {
 273 #ifdef CONFIG_VMAP_STACK
 274         struct vm_struct *vm = task_stack_vm_area(tsk);
 275
 276         if (vm) {
 277                 int i;
 278
 279                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 280                         memcg_kmem_uncharge_page(vm->pages[i], 0);
 281
 282                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 283                         if (this_cpu_cmpxchg(cached_stacks[i],
 284                                         NULL, tsk->stack_vm_area) != NULL)
 285                                 continue;
 286
 287                         return;
 288                 }
 289
 290                 vfree_atomic(tsk->stack);
 291                 return;
 292         }
 293 #endif
 294
 295         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 296 }
 297 # else
 298 static struct kmem_cache *thread_stack_cache;
 299
 300 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 301                                                   int node)
 302 {
 303         unsigned long *stack;
 304         stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 305         stack = kasan_reset_tag(stack);
 306         tsk->stack = stack;
 307         return stack;
 308 }
 309
 310 static void free_thread_stack(struct task_struct *tsk)
 311 {
 312         kmem_cache_free(thread_stack_cache, tsk->stack);
 313 }
 314
 315 void thread_stack_cache_init(void)
 316 {
 317         thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 318                                         THREAD_SIZE, THREAD_SIZE, 0, 0,
 319                                         THREAD_SIZE, NULL);
 320         BUG_ON(thread_stack_cache == NULL);
 321 }
 322 # endif
 323 #endif
 324
 325 /* SLAB cache for signal_struct structures (tsk->signal) */
 326 static struct kmem_cache *signal_cachep;
 327
 328 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 329 struct kmem_cache *sighand_cachep;
 330
 331 /* SLAB cache for files_struct structures (tsk->files) */
 332 struct kmem_cache *files_cachep;
 333
 334 /* SLAB cache for fs_struct structures (tsk->fs) */
 335 struct kmem_cache *fs_cachep;
 336
 337 /* SLAB cache for vm_area_struct structures */
 338 static struct kmem_cache *vm_area_cachep;
 339
 340 /* SLAB cache for mm_struct structures (tsk->mm) */
 341 static struct kmem_cache *mm_cachep;
 342
 343 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 344 {
 345         struct vm_area_struct *vma;
 346
 347         vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 348         if (vma)
 349                 vma_init(vma, mm);
 350         return vma;
 351 }
 352
 353 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 354 {
 355         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 356
 357         if (new) {
 358                 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
 359                 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
 360                 /*
 361                  * orig->shared.rb may be modified concurrently, but the clone
 362                  * will be reinitialized.
 363                  */
 364                 *new = data_race(*orig);
 365                 INIT_LIST_HEAD(&new->anon_vma_chain);
 366                 new->vm_next = new->vm_prev = NULL;
 367         }
 368         return new;
 369 }
 370
 371 void vm_area_free(struct vm_area_struct *vma)
 372 {
 373         kmem_cache_free(vm_area_cachep, vma);
 374 }
 375
 376 static void account_kernel_stack(struct task_struct *tsk, int account)
 377 {
 378         void *stack = task_stack_page(tsk);
 379         struct vm_struct *vm = task_stack_vm_area(tsk);
 380
 381
 382         /* All stack pages are in the same node. */
 383         if (vm)
 384                 mod_lruvec_page_state(vm->pages[0], NR_KERNEL_STACK_KB,
 385                                       account * (THREAD_SIZE / 1024));
 386         else
 387                 mod_lruvec_slab_state(stack, NR_KERNEL_STACK_KB,
 388                                       account * (THREAD_SIZE / 1024));
 389 }
 390
 391 static int memcg_charge_kernel_stack(struct task_struct *tsk)
 392 {
 393 #ifdef CONFIG_VMAP_STACK
 394         struct vm_struct *vm = task_stack_vm_area(tsk);
 395         int ret;
 396
 397         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 398
 399         if (vm) {
 400                 int i;
 401
 402                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 403
 404                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 405                         /*
 406                          * If memcg_kmem_charge_page() fails, page->mem_cgroup
 407                          * pointer is NULL, and memcg_kmem_uncharge_page() in
 408                          * free_thread_stack() will ignore this page.
 409                          */
 410                         ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL,
 411                                                      0);
 412                         if (ret)
 413                                 return ret;
 414                 }
 415         }
 416 #endif
 417         return 0;
 418 }
 419
 420 static void release_task_stack(struct task_struct *tsk)
 421 {
 422         if (WARN_ON(tsk->state != TASK_DEAD))
 423                 return;  /* Better to leak the stack than to free prematurely */
 424
 425         account_kernel_stack(tsk, -1);
 426         free_thread_stack(tsk);
 427         tsk->stack = NULL;
 428 #ifdef CONFIG_VMAP_STACK
 429         tsk->stack_vm_area = NULL;
 430 #endif
 431 }
 432
 433 #ifdef CONFIG_THREAD_INFO_IN_TASK
 434 void put_task_stack(struct task_struct *tsk)
 435 {
 436         if (refcount_dec_and_test(&tsk->stack_refcount))
 437                 release_task_stack(tsk);
 438 }
 439 #endif
 440
 441 void free_task(struct task_struct *tsk)
 442 {
 443         scs_release(tsk);
 444
 445 #ifndef CONFIG_THREAD_INFO_IN_TASK
 446         /*
 447          * The task is finally done with both the stack and thread_info,
 448          * so free both.
 449          */
 450         release_task_stack(tsk);
 451 #else
 452         /*
 453          * If the task had a separate stack allocation, it should be gone
 454          * by now.
 455          */
 456         WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
 457 #endif
 458         rt_mutex_debug_task_free(tsk);
 459         ftrace_graph_exit_task(tsk);
 460         arch_release_task_struct(tsk);
 461         if (tsk->flags & PF_KTHREAD)
 462                 free_kthread_struct(tsk);
 463         free_task_struct(tsk);
 464 }
 465 EXPORT_SYMBOL(free_task);
 466
 467 #ifdef CONFIG_MMU
 468 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 469                                         struct mm_struct *oldmm)
 470 {
 471         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 472         struct rb_node **rb_link, *rb_parent;
 473         int retval;
 474         unsigned long charge;
 475         LIST_HEAD(uf);
 476
 477         uprobe_start_dup_mmap();
 478         if (mmap_write_lock_killable(oldmm)) {
 479                 retval = -EINTR;
 480                 goto fail_uprobe_end;
 481         }
 482         flush_cache_dup_mm(oldmm);
 483         uprobe_dup_mmap(oldmm, mm);
 484         /*
 485          * Not linked in yet - no deadlock potential:
 486          */
 487         mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
 488
 489         /* No ordering required: file already has been exposed. */
 490         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 491
 492         mm->total_vm = oldmm->total_vm;
 493         mm->data_vm = oldmm->data_vm;
 494         mm->exec_vm = oldmm->exec_vm;
 495         mm->stack_vm = oldmm->stack_vm;
 496
 497         rb_link = &mm->mm_rb.rb_node;
 498         rb_parent = NULL;
 499         pprev = &mm->mmap;
 500         retval = ksm_fork(mm, oldmm);
 501         if (retval)
 502                 goto out;
 503         retval = khugepaged_fork(mm, oldmm);
 504         if (retval)
 505                 goto out;
 506
 507         prev = NULL;
 508         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 509                 struct file *file;
 510
 511                 if (mpnt->vm_flags & VM_DONTCOPY) {
 512                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 513                         continue;
 514                 }
 515                 charge = 0;
 516                 /*
 517                  * Don't duplicate many vmas if we've been oom-killed (for
 518                  * example)
 519                  */
 520                 if (fatal_signal_pending(current)) {
 521                         retval = -EINTR;
 522                         goto out;
 523                 }
 524                 if (mpnt->vm_flags & VM_ACCOUNT) {
 525                         unsigned long len = vma_pages(mpnt);
 526
 527                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 528                                 goto fail_nomem;
 529                         charge = len;
 530                 }
 531                 tmp = vm_area_dup(mpnt);
 532                 if (!tmp)
 533                         goto fail_nomem;
 534                 retval = vma_dup_policy(mpnt, tmp);
 535                 if (retval)
 536                         goto fail_nomem_policy;
 537                 tmp->vm_mm = mm;
 538                 retval = dup_userfaultfd(tmp, &uf);
 539                 if (retval)
 540                         goto fail_nomem_anon_vma_fork;
 541                 if (tmp->vm_flags & VM_WIPEONFORK) {
 542                         /*
 543                          * VM_WIPEONFORK gets a clean slate in the child.
 544                          * Don't prepare anon_vma until fault since we don't
 545                          * copy page for current vma.
 546                          */
 547                         tmp->anon_vma = NULL;
 548                 } else if (anon_vma_fork(tmp, mpnt))
 549                         goto fail_nomem_anon_vma_fork;
 550                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 551                 file = tmp->vm_file;
 552                 if (file) {
 553                         struct inode *inode = file_inode(file);
 554                         struct address_space *mapping = file->f_mapping;
 555
 556                         get_file(file);
 557                         if (tmp->vm_flags & VM_DENYWRITE)
 558                                 atomic_dec(&inode->i_writecount);
 559                         i_mmap_lock_write(mapping);
 560                         if (tmp->vm_flags & VM_SHARED)
 561                                 atomic_inc(&mapping->i_mmap_writable);
 562                         flush_dcache_mmap_lock(mapping);
 563                         /* insert tmp into the share list, just after mpnt */
 564                         vma_interval_tree_insert_after(tmp, mpnt,
 565                                         &mapping->i_mmap);
 566                         flush_dcache_mmap_unlock(mapping);
 567                         i_mmap_unlock_write(mapping);
 568                 }
 569
 570                 /*
 571                  * Clear hugetlb-related page reserves for children. This only
 572                  * affects MAP_PRIVATE mappings. Faults generated by the child
 573                  * are not guaranteed to succeed, even if read-only
 574                  */
 575                 if (is_vm_hugetlb_page(tmp))
 576                         reset_vma_resv_huge_pages(tmp);
 577
 578                 /*
 579                  * Link in the new vma and copy the page table entries.
 580                  */
 581                 *pprev = tmp;
 582                 pprev = &tmp->vm_next;
 583                 tmp->vm_prev = prev;
 584                 prev = tmp;
 585
 586                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 587                 rb_link = &tmp->vm_rb.rb_right;
 588                 rb_parent = &tmp->vm_rb;
 589
 590                 mm->map_count++;
 591                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 592                         retval = copy_page_range(mm, oldmm, mpnt, tmp);
 593
 594                 if (tmp->vm_ops && tmp->vm_ops->open)
 595                         tmp->vm_ops->open(tmp);
 596
 597                 if (retval)
 598                         goto out;
 599         }
 600         /* a new mm has just been created */
 601         retval = arch_dup_mmap(oldmm, mm);
 602 out:
 603         mmap_write_unlock(mm);
 604         flush_tlb_mm(oldmm);
 605         mmap_write_unlock(oldmm);
 606         dup_userfaultfd_complete(&uf);
 607 fail_uprobe_end:
 608         uprobe_end_dup_mmap();
 609         return retval;
 610 fail_nomem_anon_vma_fork:
 611         mpol_put(vma_policy(tmp));
 612 fail_nomem_policy:
 613         vm_area_free(tmp);
 614 fail_nomem:
 615         retval = -ENOMEM;
 616         vm_unacct_memory(charge);
 617         goto out;
 618 }
 619
 620 static inline int mm_alloc_pgd(struct mm_struct *mm)
 621 {
 622         mm->pgd = pgd_alloc(mm);
 623         if (unlikely(!mm->pgd))
 624                 return -ENOMEM;
 625         return 0;
 626 }
 627
 628 static inline void mm_free_pgd(struct mm_struct *mm)
 629 {
 630         pgd_free(mm, mm->pgd);
 631 }
 632 #else
 633 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 634 {
 635         mmap_write_lock(oldmm);
 636         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 637         mmap_write_unlock(oldmm);
 638         return 0;
 639 }
 640 #define mm_alloc_pgd(mm)        (0)
 641 #define mm_free_pgd(mm)
 642 #endif /* CONFIG_MMU */
 643
 644 static void check_mm(struct mm_struct *mm)
 645 {
 646         int i;
 647
 648         BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
 649                          "Please make sure 'struct resident_page_types[]' is updated as well");
 650
 651         for (i = 0; i < NR_MM_COUNTERS; i++) {
 652                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 653
 654                 if (unlikely(x))
 655                         pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
 656                                  mm, resident_page_types[i], x);
 657         }
 658
 659         if (mm_pgtables_bytes(mm))
 660                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 661                                 mm_pgtables_bytes(mm));
 662
 663 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 664         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 665 #endif
 666 }
 667
 668 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 669 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 670
 671 /*
 672  * Called when the last reference to the mm
 673  * is dropped: either by a lazy thread or by
 674  * mmput. Free the page directory and the mm.
 675  */
 676 void __mmdrop(struct mm_struct *mm)
 677 {
 678         BUG_ON(mm == &init_mm);
 679         WARN_ON_ONCE(mm == current->mm);
 680         WARN_ON_ONCE(mm == current->active_mm);
 681         mm_free_pgd(mm);
 682         destroy_context(mm);
 683         mmu_notifier_subscriptions_destroy(mm);
 684         check_mm(mm);
 685         put_user_ns(mm->user_ns);
 686         free_mm(mm);
 687 }
 688 EXPORT_SYMBOL_GPL(__mmdrop);
 689
 690 static void mmdrop_async_fn(struct work_struct *work)
 691 {
 692         struct mm_struct *mm;
 693
 694         mm = container_of(work, struct mm_struct, async_put_work);
 695         __mmdrop(mm);
 696 }
 697
 698 static void mmdrop_async(struct mm_struct *mm)
 699 {
 700         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 701                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 702                 schedule_work(&mm->async_put_work);
 703         }
 704 }
 705
 706 static inline void free_signal_struct(struct signal_struct *sig)
 707 {
 708         taskstats_tgid_free(sig);
 709         sched_autogroup_exit(sig);
 710         /*
 711          * __mmdrop is not safe to call from softirq context on x86 due to
 712          * pgd_dtor so postpone it to the async context
 713          */
 714         if (sig->oom_mm)
 715                 mmdrop_async(sig->oom_mm);
 716         kmem_cache_free(signal_cachep, sig);
 717 }
 718
 719 static inline void put_signal_struct(struct signal_struct *sig)
 720 {
 721         if (refcount_dec_and_test(&sig->sigcnt))
 722                 free_signal_struct(sig);
 723 }
 724
 725 void __put_task_struct(struct task_struct *tsk)
 726 {
 727         WARN_ON(!tsk->exit_state);
 728         WARN_ON(refcount_read(&tsk->usage));
 729         WARN_ON(tsk == current);
 730
 731         cgroup_free(tsk);
 732         task_numa_free(tsk, true);
 733         security_task_free(tsk);
 734         exit_creds(tsk);
 735         delayacct_tsk_free(tsk);
 736         put_signal_struct(tsk->signal);
 737
 738         if (!profile_handoff_task(tsk))
 739                 free_task(tsk);
 740 }
 741 EXPORT_SYMBOL_GPL(__put_task_struct);
 742
 743 void __init __weak arch_task_cache_init(void) { }
 744
 745 /*
 746  * set_max_threads
 747  */
 748 static void set_max_threads(unsigned int max_threads_suggested)
 749 {
 750         u64 threads;
 751         unsigned long nr_pages = totalram_pages();
 752
 753         /*
 754          * The number of threads shall be limited such that the thread
 755          * structures may only consume a small part of the available memory.
 756          */
 757         if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
 758                 threads = MAX_THREADS;
 759         else
 760                 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
 761                                     (u64) THREAD_SIZE * 8UL);
 762
 763         if (threads > max_threads_suggested)
 764                 threads = max_threads_suggested;
 765
 766         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 767 }
 768
 769 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 770 /* Initialized by the architecture: */
 771 int arch_task_struct_size __read_mostly;
 772 #endif
 773
 774 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 775 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 776 {
 777         /* Fetch thread_struct whitelist for the architecture. */
 778         arch_thread_struct_whitelist(offset, size);
 779
 780         /*
 781          * Handle zero-sized whitelist or empty thread_struct, otherwise
 782          * adjust offset to position of thread_struct in task_struct.
 783          */
 784         if (unlikely(*size == 0))
 785                 *offset = 0;
 786         else
 787                 *offset += offsetof(struct task_struct, thread);
 788 }
 789 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
 790
 791 void __init fork_init(void)
 792 {
 793         int i;
 794 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 795 #ifndef ARCH_MIN_TASKALIGN
 796 #define ARCH_MIN_TASKALIGN      0
 797 #endif
 798         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 799         unsigned long useroffset, usersize;
 800
 801         /* create a slab on which task_structs can be allocated */
 802         task_struct_whitelist(&useroffset, &usersize);
 803         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 804                         arch_task_struct_size, align,
 805                         SLAB_PANIC|SLAB_ACCOUNT,
 806                         useroffset, usersize, NULL);
 807 #endif
 808
 809         /* do the arch specific task caches init */
 810         arch_task_cache_init();
 811
 812         set_max_threads(MAX_THREADS);
 813
 814         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 815         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 816         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 817                 init_task.signal->rlim[RLIMIT_NPROC];
 818
 819         for (i = 0; i < UCOUNT_COUNTS; i++) {
 820                 init_user_ns.ucount_max[i] = max_threads/2;
 821         }
 822
 823 #ifdef CONFIG_VMAP_STACK
 824         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 825                           NULL, free_vm_stack_cache);
 826 #endif
 827
 828         scs_init();
 829
 830         lockdep_init_task(&init_task);
 831         uprobes_init();
 832 }
 833
 834 int __weak arch_dup_task_struct(struct task_struct *dst,
 835                                                struct task_struct *src)
 836 {
 837         *dst = *src;
 838         return 0;
 839 }
 840
 841 void set_task_stack_end_magic(struct task_struct *tsk)
 842 {
 843         unsigned long *stackend;
 844
 845         stackend = end_of_stack(tsk);
 846         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 847 }
 848
 849 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 850 {
 851         struct task_struct *tsk;
 852         unsigned long *stack;
 853         struct vm_struct *stack_vm_area __maybe_unused;
 854         int err;
 855
 856         if (node == NUMA_NO_NODE)
 857                 node = tsk_fork_get_node(orig);
 858         tsk = alloc_task_struct_node(node);
 859         if (!tsk)
 860                 return NULL;
 861
 862         stack = alloc_thread_stack_node(tsk, node);
 863         if (!stack)
 864                 goto free_tsk;
 865
 866         if (memcg_charge_kernel_stack(tsk))
 867                 goto free_stack;
 868
 869         stack_vm_area = task_stack_vm_area(tsk);
 870
 871         err = arch_dup_task_struct(tsk, orig);
 872
 873         /*
 874          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 875          * sure they're properly initialized before using any stack-related
 876          * functions again.
 877          */
 878         tsk->stack = stack;
 879 #ifdef CONFIG_VMAP_STACK
 880         tsk->stack_vm_area = stack_vm_area;
 881 #endif
 882 #ifdef CONFIG_THREAD_INFO_IN_TASK
 883         refcount_set(&tsk->stack_refcount, 1);
 884 #endif
 885
 886         if (err)
 887                 goto free_stack;
 888
 889         err = scs_prepare(tsk, node);
 890         if (err)
 891                 goto free_stack;
 892
 893 #ifdef CONFIG_SECCOMP
 894         /*
 895          * We must handle setting up seccomp filters once we're under
 896          * the sighand lock in case orig has changed between now and
 897          * then. Until then, filter must be NULL to avoid messing up
 898          * the usage counts on the error path calling free_task.
 899          */
 900         tsk->seccomp.filter = NULL;
 901 #endif
 902
 903         setup_thread_stack(tsk, orig);
 904         clear_user_return_notifier(tsk);
 905         clear_tsk_need_resched(tsk);
 906         set_task_stack_end_magic(tsk);
 907
 908 #ifdef CONFIG_STACKPROTECTOR
 909         tsk->stack_canary = get_random_canary();
 910 #endif
 911         if (orig->cpus_ptr == &orig->cpus_mask)
 912                 tsk->cpus_ptr = &tsk->cpus_mask;
 913
 914         /*
 915          * One for the user space visible state that goes away when reaped.
 916          * One for the scheduler.
 917          */
 918         refcount_set(&tsk->rcu_users, 2);
 919         /* One for the rcu users */
 920         refcount_set(&tsk->usage, 1);
 921 #ifdef CONFIG_BLK_DEV_IO_TRACE
 922         tsk->btrace_seq = 0;
 923 #endif
 924         tsk->splice_pipe = NULL;
 925         tsk->task_frag.page = NULL;
 926         tsk->wake_q.next = NULL;
 927
 928         account_kernel_stack(tsk, 1);
 929
 930         kcov_task_init(tsk);
 931
 932 #ifdef CONFIG_FAULT_INJECTION
 933         tsk->fail_nth = 0;
 934 #endif
 935
 936 #ifdef CONFIG_BLK_CGROUP
 937         tsk->throttle_queue = NULL;
 938         tsk->use_memdelay = 0;
 939 #endif
 940
 941 #ifdef CONFIG_MEMCG
 942         tsk->active_memcg = NULL;
 943 #endif
 944         return tsk;
 945
 946 free_stack:
 947         free_thread_stack(tsk);
 948 free_tsk:
 949         free_task_struct(tsk);
 950         return NULL;
 951 }
 952
 953 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 954
 955 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 956
 957 static int __init coredump_filter_setup(char *s)
 958 {
 959         default_dump_filter =
 960                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 961                 MMF_DUMP_FILTER_MASK;
 962         return 1;
 963 }
 964
 965 __setup("coredump_filter=", coredump_filter_setup);
 966
 967 #include <linux/init_task.h>
 968
 969 static void mm_init_aio(struct mm_struct *mm)
 970 {
 971 #ifdef CONFIG_AIO
 972         spin_lock_init(&mm->ioctx_lock);
 973         mm->ioctx_table = NULL;
 974 #endif
 975 }
 976
 977 static __always_inline void mm_clear_owner(struct mm_struct *mm,
 978                                            struct task_struct *p)
 979 {
 980 #ifdef CONFIG_MEMCG
 981         if (mm->owner == p)
 982                 WRITE_ONCE(mm->owner, NULL);
 983 #endif
 984 }
 985
 986 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 987 {
 988 #ifdef CONFIG_MEMCG
 989         mm->owner = p;
 990 #endif
 991 }
 992
 993 static void mm_init_uprobes_state(struct mm_struct *mm)
 994 {
 995 #ifdef CONFIG_UPROBES
 996         mm->uprobes_state.xol_area = NULL;
 997 #endif
 998 }
 999
1000 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1001         struct user_namespace *user_ns)
1002 {
1003         mm->mmap = NULL;
1004         mm->mm_rb = RB_ROOT;
1005         mm->vmacache_seqnum = 0;
1006         atomic_set(&mm->mm_users, 1);
1007         atomic_set(&mm->mm_count, 1);
1008         mmap_init_lock(mm);
1009         INIT_LIST_HEAD(&mm->mmlist);
1010         mm->core_state = NULL;
1011         mm_pgtables_bytes_init(mm);
1012         mm->map_count = 0;
1013         mm->locked_vm = 0;
1014         atomic_set(&mm->has_pinned, 0);
1015         atomic64_set(&mm->pinned_vm, 0);
1016         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1017         spin_lock_init(&mm->page_table_lock);
1018         spin_lock_init(&mm->arg_lock);
1019         mm_init_cpumask(mm);
1020         mm_init_aio(mm);
1021         mm_init_owner(mm, p);
1022         RCU_INIT_POINTER(mm->exe_file, NULL);
1023         mmu_notifier_subscriptions_init(mm);
1024         init_tlb_flush_pending(mm);
1025 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1026         mm->pmd_huge_pte = NULL;
1027 #endif
1028         mm_init_uprobes_state(mm);
1029
1030         if (current->mm) {
1031                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1032                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1033         } else {
1034                 mm->flags = default_dump_filter;
1035                 mm->def_flags = 0;
1036         }
1037
1038         if (mm_alloc_pgd(mm))
1039                 goto fail_nopgd;
1040
1041         if (init_new_context(p, mm))
1042                 goto fail_nocontext;
1043
1044         mm->user_ns = get_user_ns(user_ns);
1045         return mm;
1046
1047 fail_nocontext:
1048         mm_free_pgd(mm);
1049 fail_nopgd:
1050         free_mm(mm);
1051         return NULL;
1052 }
1053
1054 /*
1055  * Allocate and initialize an mm_struct.
1056  */
1057 struct mm_struct *mm_alloc(void)
1058 {
1059         struct mm_struct *mm;
1060
1061         mm = allocate_mm();
1062         if (!mm)
1063                 return NULL;
1064
1065         memset(mm, 0, sizeof(*mm));
1066         return mm_init(mm, current, current_user_ns());
1067 }
1068
1069 static inline void __mmput(struct mm_struct *mm)
1070 {
1071         VM_BUG_ON(atomic_read(&mm->mm_users));
1072
1073         uprobe_clear_state(mm);
1074         exit_aio(mm);
1075         ksm_exit(mm);
1076         khugepaged_exit(mm); /* must run before exit_mmap */
1077         exit_mmap(mm);
1078         mm_put_huge_zero_page(mm);
1079         set_mm_exe_file(mm, NULL);
1080         if (!list_empty(&mm->mmlist)) {
1081                 spin_lock(&mmlist_lock);
1082                 list_del(&mm->mmlist);
1083                 spin_unlock(&mmlist_lock);
1084         }
1085         if (mm->binfmt)
1086                 module_put(mm->binfmt->module);
1087         mmdrop(mm);
1088 }
1089
1090 /*
1091  * Decrement the use count and release all resources for an mm.
1092  */
1093 void mmput(struct mm_struct *mm)
1094 {
1095         might_sleep();
1096
1097         if (atomic_dec_and_test(&mm->mm_users))
1098                 __mmput(mm);
1099 }
1100 EXPORT_SYMBOL_GPL(mmput);
1101
1102 #ifdef CONFIG_MMU
1103 static void mmput_async_fn(struct work_struct *work)
1104 {
1105         struct mm_struct *mm = container_of(work, struct mm_struct,
1106                                             async_put_work);
1107
1108         __mmput(mm);
1109 }
1110
1111 void mmput_async(struct mm_struct *mm)
1112 {
1113         if (atomic_dec_and_test(&mm->mm_users)) {
1114                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1115                 schedule_work(&mm->async_put_work);
1116         }
1117 }
1118 #endif
1119
1120 /**
1121  * set_mm_exe_file - change a reference to the mm's executable file
1122  *
1123  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1124  *
1125  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1126  * invocations: in mmput() nobody alive left, in execve task is single
1127  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1128  * mm->exe_file, but does so without using set_mm_exe_file() in order
1129  * to do avoid the need for any locks.
1130  */
1131 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1132 {
1133         struct file *old_exe_file;
1134
1135         /*
1136          * It is safe to dereference the exe_file without RCU as
1137          * this function is only called if nobody else can access
1138          * this mm -- see comment above for justification.
1139          */
1140         old_exe_file = rcu_dereference_raw(mm->exe_file);
1141
1142         if (new_exe_file)
1143                 get_file(new_exe_file);
1144         rcu_assign_pointer(mm->exe_file, new_exe_file);
1145         if (old_exe_file)
1146                 fput(old_exe_file);
1147 }
1148
1149 /**
1150  * get_mm_exe_file - acquire a reference to the mm's executable file
1151  *
1152  * Returns %NULL if mm has no associated executable file.
1153  * User must release file via fput().
1154  */
1155 struct file *get_mm_exe_file(struct mm_struct *mm)
1156 {
1157         struct file *exe_file;
1158
1159         rcu_read_lock();
1160         exe_file = rcu_dereference(mm->exe_file);
1161         if (exe_file && !get_file_rcu(exe_file))
1162                 exe_file = NULL;
1163         rcu_read_unlock();
1164         return exe_file;
1165 }
1166 EXPORT_SYMBOL(get_mm_exe_file);
1167
1168 /**
1169  * get_task_exe_file - acquire a reference to the task's executable file
1170  *
1171  * Returns %NULL if task's mm (if any) has no associated executable file or
1172  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1173  * User must release file via fput().
1174  */
1175 struct file *get_task_exe_file(struct task_struct *task)
1176 {
1177         struct file *exe_file = NULL;
1178         struct mm_struct *mm;
1179
1180         task_lock(task);
1181         mm = task->mm;
1182         if (mm) {
1183                 if (!(task->flags & PF_KTHREAD))
1184                         exe_file = get_mm_exe_file(mm);
1185         }
1186         task_unlock(task);
1187         return exe_file;
1188 }
1189 EXPORT_SYMBOL(get_task_exe_file);
1190
1191 /**
1192  * get_task_mm - acquire a reference to the task's mm
1193  *
1194  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1195  * this kernel workthread has transiently adopted a user mm with use_mm,
1196  * to do its AIO) is not set and if so returns a reference to it, after
1197  * bumping up the use count.  User must release the mm via mmput()
1198  * after use.  Typically used by /proc and ptrace.
1199  */
1200 struct mm_struct *get_task_mm(struct task_struct *task)
1201 {
1202         struct mm_struct *mm;
1203
1204         task_lock(task);
1205         mm = task->mm;
1206         if (mm) {
1207                 if (task->flags & PF_KTHREAD)
1208                         mm = NULL;
1209                 else
1210                         mmget(mm);
1211         }
1212         task_unlock(task);
1213         return mm;
1214 }
1215 EXPORT_SYMBOL_GPL(get_task_mm);
1216
1217 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1218 {
1219         struct mm_struct *mm;
1220         int err;
1221
1222         err =  mutex_lock_killable(&task->signal->exec_update_mutex);
1223         if (err)
1224                 return ERR_PTR(err);
1225
1226         mm = get_task_mm(task);
1227         if (mm && mm != current->mm &&
1228                         !ptrace_may_access(task, mode)) {
1229                 mmput(mm);
1230                 mm = ERR_PTR(-EACCES);
1231         }
1232         mutex_unlock(&task->signal->exec_update_mutex);
1233
1234         return mm;
1235 }
1236
1237 static void complete_vfork_done(struct task_struct *tsk)
1238 {
1239         struct completion *vfork;
1240
1241         task_lock(tsk);
1242         vfork = tsk->vfork_done;
1243         if (likely(vfork)) {
1244                 tsk->vfork_done = NULL;
1245                 complete(vfork);
1246         }
1247         task_unlock(tsk);
1248 }
1249
1250 static int wait_for_vfork_done(struct task_struct *child,
1251                                 struct completion *vfork)
1252 {
1253         int killed;
1254
1255         freezer_do_not_count();
1256         cgroup_enter_frozen();
1257         killed = wait_for_completion_killable(vfork);
1258         cgroup_leave_frozen(false);
1259         freezer_count();
1260
1261         if (killed) {
1262                 task_lock(child);
1263                 child->vfork_done = NULL;
1264                 task_unlock(child);
1265         }
1266
1267         put_task_struct(child);
1268         return killed;
1269 }
1270
1271 /* Please note the differences between mmput and mm_release.
1272  * mmput is called whenever we stop holding onto a mm_struct,
1273  * error success whatever.
1274  *
1275  * mm_release is called after a mm_struct has been removed
1276  * from the current process.
1277  *
1278  * This difference is important for error handling, when we
1279  * only half set up a mm_struct for a new process and need to restore
1280  * the old one.  Because we mmput the new mm_struct before
1281  * restoring the old one. . .
1282  * Eric Biederman 10 January 1998
1283  */
1284 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1285 {
1286         uprobe_free_utask(tsk);
1287
1288         /* Get rid of any cached register state */
1289         deactivate_mm(tsk, mm);
1290
1291         /*
1292          * Signal userspace if we're not exiting with a core dump
1293          * because we want to leave the value intact for debugging
1294          * purposes.
1295          */
1296         if (tsk->clear_child_tid) {
1297                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1298                     atomic_read(&mm->mm_users) > 1) {
1299                         /*
1300                          * We don't check the error code - if userspace has
1301                          * not set up a proper pointer then tough luck.
1302                          */
1303                         put_user(0, tsk->clear_child_tid);
1304                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1305                                         1, NULL, NULL, 0, 0);
1306                 }
1307                 tsk->clear_child_tid = NULL;
1308         }
1309
1310         /*
1311          * All done, finally we can wake up parent and return this mm to him.
1312          * Also kthread_stop() uses this completion for synchronization.
1313          */
1314         if (tsk->vfork_done)
1315                 complete_vfork_done(tsk);
1316 }
1317
1318 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1319 {
1320         futex_exit_release(tsk);
1321         mm_release(tsk, mm);
1322 }
1323
1324 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1325 {
1326         futex_exec_release(tsk);
1327         mm_release(tsk, mm);
1328 }
1329
1330 /**
1331  * dup_mm() - duplicates an existing mm structure
1332  * @tsk: the task_struct with which the new mm will be associated.
1333  * @oldmm: the mm to duplicate.
1334  *
1335  * Allocates a new mm structure and duplicates the provided @oldmm structure
1336  * content into it.
1337  *
1338  * Return: the duplicated mm or NULL on failure.
1339  */
1340 static struct mm_struct *dup_mm(struct task_struct *tsk,
1341                                 struct mm_struct *oldmm)
1342 {
1343         struct mm_struct *mm;
1344         int err;
1345
1346         mm = allocate_mm();
1347         if (!mm)
1348                 goto fail_nomem;
1349
1350         memcpy(mm, oldmm, sizeof(*mm));
1351
1352         if (!mm_init(mm, tsk, mm->user_ns))
1353                 goto fail_nomem;
1354
1355         err = dup_mmap(mm, oldmm);
1356         if (err)
1357                 goto free_pt;
1358
1359         mm->hiwater_rss = get_mm_rss(mm);
1360         mm->hiwater_vm = mm->total_vm;
1361
1362         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1363                 goto free_pt;
1364
1365         return mm;
1366
1367 free_pt:
1368         /* don't put binfmt in mmput, we haven't got module yet */
1369         mm->binfmt = NULL;
1370         mm_init_owner(mm, NULL);
1371         mmput(mm);
1372
1373 fail_nomem:
1374         return NULL;
1375 }
1376
1377 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1378 {
1379         struct mm_struct *mm, *oldmm;
1380         int retval;
1381
1382         tsk->min_flt = tsk->maj_flt = 0;
1383         tsk->nvcsw = tsk->nivcsw = 0;
1384 #ifdef CONFIG_DETECT_HUNG_TASK
1385         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1386         tsk->last_switch_time = 0;
1387 #endif
1388
1389         tsk->mm = NULL;
1390         tsk->active_mm = NULL;
1391
1392         /*
1393          * Are we cloning a kernel thread?
1394          *
1395          * We need to steal a active VM for that..
1396          */
1397         oldmm = current->mm;
1398         if (!oldmm)
1399                 return 0;
1400
1401         /* initialize the new vmacache entries */
1402         vmacache_flush(tsk);
1403
1404         if (clone_flags & CLONE_VM) {
1405                 mmget(oldmm);
1406                 mm = oldmm;
1407                 goto good_mm;
1408         }
1409
1410         retval = -ENOMEM;
1411         mm = dup_mm(tsk, current->mm);
1412         if (!mm)
1413                 goto fail_nomem;
1414
1415 good_mm:
1416         tsk->mm = mm;
1417         tsk->active_mm = mm;
1418         return 0;
1419
1420 fail_nomem:
1421         return retval;
1422 }
1423
1424 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1425 {
1426         struct fs_struct *fs = current->fs;
1427         if (clone_flags & CLONE_FS) {
1428                 /* tsk->fs is already what we want */
1429                 spin_lock(&fs->lock);
1430                 if (fs->in_exec) {
1431                         spin_unlock(&fs->lock);
1432                         return -EAGAIN;
1433                 }
1434                 fs->users++;
1435                 spin_unlock(&fs->lock);
1436                 return 0;
1437         }
1438         tsk->fs = copy_fs_struct(fs);
1439         if (!tsk->fs)
1440                 return -ENOMEM;
1441         return 0;
1442 }
1443
1444 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1445 {
1446         struct files_struct *oldf, *newf;
1447         int error = 0;
1448
1449         /*
1450          * A background process may not have any files ...
1451          */
1452         oldf = current->files;
1453         if (!oldf)
1454                 goto out;
1455
1456         if (clone_flags & CLONE_FILES) {
1457                 atomic_inc(&oldf->count);
1458                 goto out;
1459         }
1460
1461         newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1462         if (!newf)
1463                 goto out;
1464
1465         tsk->files = newf;
1466         error = 0;
1467 out:
1468         return error;
1469 }
1470
1471 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1472 {
1473 #ifdef CONFIG_BLOCK
1474         struct io_context *ioc = current->io_context;
1475         struct io_context *new_ioc;
1476
1477         if (!ioc)
1478                 return 0;
1479         /*
1480          * Share io context with parent, if CLONE_IO is set
1481          */
1482         if (clone_flags & CLONE_IO) {
1483                 ioc_task_link(ioc);
1484                 tsk->io_context = ioc;
1485         } else if (ioprio_valid(ioc->ioprio)) {
1486                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1487                 if (unlikely(!new_ioc))
1488                         return -ENOMEM;
1489
1490                 new_ioc->ioprio = ioc->ioprio;
1491                 put_io_context(new_ioc);
1492         }
1493 #endif
1494         return 0;
1495 }
1496
1497 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1498 {
1499         struct sighand_struct *sig;
1500
1501         if (clone_flags & CLONE_SIGHAND) {
1502                 refcount_inc(&current->sighand->count);
1503                 return 0;
1504         }
1505         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1506         RCU_INIT_POINTER(tsk->sighand, sig);
1507         if (!sig)
1508                 return -ENOMEM;
1509
1510         refcount_set(&sig->count, 1);
1511         spin_lock_irq(&current->sighand->siglock);
1512         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1513         spin_unlock_irq(&current->sighand->siglock);
1514
1515         /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1516         if (clone_flags & CLONE_CLEAR_SIGHAND)
1517                 flush_signal_handlers(tsk, 0);
1518
1519         return 0;
1520 }
1521
1522 void __cleanup_sighand(struct sighand_struct *sighand)
1523 {
1524         if (refcount_dec_and_test(&sighand->count)) {
1525                 signalfd_cleanup(sighand);
1526                 /*
1527                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1528                  * without an RCU grace period, see __lock_task_sighand().
1529                  */
1530                 kmem_cache_free(sighand_cachep, sighand);
1531         }
1532 }
1533
1534 /*
1535  * Initialize POSIX timer handling for a thread group.
1536  */
1537 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1538 {
1539         struct posix_cputimers *pct = &sig->posix_cputimers;
1540         unsigned long cpu_limit;
1541
1542         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1543         posix_cputimers_group_init(pct, cpu_limit);
1544 }
1545
1546 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1547 {
1548         struct signal_struct *sig;
1549
1550         if (clone_flags & CLONE_THREAD)
1551                 return 0;
1552
1553         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1554         tsk->signal = sig;
1555         if (!sig)
1556                 return -ENOMEM;
1557
1558         sig->nr_threads = 1;
1559         atomic_set(&sig->live, 1);
1560         refcount_set(&sig->sigcnt, 1);
1561
1562         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1563         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1564         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1565
1566         init_waitqueue_head(&sig->wait_chldexit);
1567         sig->curr_target = tsk;
1568         init_sigpending(&sig->shared_pending);
1569         INIT_HLIST_HEAD(&sig->multiprocess);
1570         seqlock_init(&sig->stats_lock);
1571         prev_cputime_init(&sig->prev_cputime);
1572
1573 #ifdef CONFIG_POSIX_TIMERS
1574         INIT_LIST_HEAD(&sig->posix_timers);
1575         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1576         sig->real_timer.function = it_real_fn;
1577 #endif
1578
1579         task_lock(current->group_leader);
1580         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1581         task_unlock(current->group_leader);
1582
1583         posix_cpu_timers_init_group(sig);
1584
1585         tty_audit_fork(sig);
1586         sched_autogroup_fork(sig);
1587
1588         sig->oom_score_adj = current->signal->oom_score_adj;
1589         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1590
1591         mutex_init(&sig->cred_guard_mutex);
1592         mutex_init(&sig->exec_update_mutex);
1593
1594         return 0;
1595 }
1596
1597 static void copy_seccomp(struct task_struct *p)
1598 {
1599 #ifdef CONFIG_SECCOMP
1600         /*
1601          * Must be called with sighand->lock held, which is common to
1602          * all threads in the group. Holding cred_guard_mutex is not
1603          * needed because this new task is not yet running and cannot
1604          * be racing exec.
1605          */
1606         assert_spin_locked(&current->sighand->siglock);
1607
1608         /* Ref-count the new filter user, and assign it. */
1609         get_seccomp_filter(current);
1610         p->seccomp = current->seccomp;
1611
1612         /*
1613          * Explicitly enable no_new_privs here in case it got set
1614          * between the task_struct being duplicated and holding the
1615          * sighand lock. The seccomp state and nnp must be in sync.
1616          */
1617         if (task_no_new_privs(current))
1618                 task_set_no_new_privs(p);
1619
1620         /*
1621          * If the parent gained a seccomp mode after copying thread
1622          * flags and between before we held the sighand lock, we have
1623          * to manually enable the seccomp thread flag here.
1624          */
1625         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1626                 set_tsk_thread_flag(p, TIF_SECCOMP);
1627 #endif
1628 }
1629
1630 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1631 {
1632         current->clear_child_tid = tidptr;
1633
1634         return task_pid_vnr(current);
1635 }
1636
1637 static void rt_mutex_init_task(struct task_struct *p)
1638 {
1639         raw_spin_lock_init(&p->pi_lock);
1640 #ifdef CONFIG_RT_MUTEXES
1641         p->pi_waiters = RB_ROOT_CACHED;
1642         p->pi_top_task = NULL;
1643         p->pi_blocked_on = NULL;
1644 #endif
1645 }
1646
1647 static inline void init_task_pid_links(struct task_struct *task)
1648 {
1649         enum pid_type type;
1650
1651         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1652                 INIT_HLIST_NODE(&task->pid_links[type]);
1653         }
1654 }
1655
1656 static inline void
1657 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1658 {
1659         if (type == PIDTYPE_PID)
1660                 task->thread_pid = pid;
1661         else
1662                 task->signal->pids[type] = pid;
1663 }
1664
1665 static inline void rcu_copy_process(struct task_struct *p)
1666 {
1667 #ifdef CONFIG_PREEMPT_RCU
1668         p->rcu_read_lock_nesting = 0;
1669         p->rcu_read_unlock_special.s = 0;
1670         p->rcu_blocked_node = NULL;
1671         INIT_LIST_HEAD(&p->rcu_node_entry);
1672 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1673 #ifdef CONFIG_TASKS_RCU
1674         p->rcu_tasks_holdout = false;
1675         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1676         p->rcu_tasks_idle_cpu = -1;
1677 #endif /* #ifdef CONFIG_TASKS_RCU */
1678 #ifdef CONFIG_TASKS_TRACE_RCU
1679         p->trc_reader_nesting = 0;
1680         p->trc_reader_special.s = 0;
1681         INIT_LIST_HEAD(&p->trc_holdout_list);
1682 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1683 }
1684
1685 struct pid *pidfd_pid(const struct file *file)
1686 {
1687         if (file->f_op == &pidfd_fops)
1688                 return file->private_data;
1689
1690         return ERR_PTR(-EBADF);
1691 }
1692
1693 static int pidfd_release(struct inode *inode, struct file *file)
1694 {
1695         struct pid *pid = file->private_data;
1696
1697         file->private_data = NULL;
1698         put_pid(pid);
1699         return 0;
1700 }
1701
1702 #ifdef CONFIG_PROC_FS
1703 /**
1704  * pidfd_show_fdinfo - print information about a pidfd
1705  * @m: proc fdinfo file
1706  * @f: file referencing a pidfd
1707  *
1708  * Pid:
1709  * This function will print the pid that a given pidfd refers to in the
1710  * pid namespace of the procfs instance.
1711  * If the pid namespace of the process is not a descendant of the pid
1712  * namespace of the procfs instance 0 will be shown as its pid. This is
1713  * similar to calling getppid() on a process whose parent is outside of
1714  * its pid namespace.
1715  *
1716  * NSpid:
1717  * If pid namespaces are supported then this function will also print
1718  * the pid of a given pidfd refers to for all descendant pid namespaces
1719  * starting from the current pid namespace of the instance, i.e. the
1720  * Pid field and the first entry in the NSpid field will be identical.
1721  * If the pid namespace of the process is not a descendant of the pid
1722  * namespace of the procfs instance 0 will be shown as its first NSpid
1723  * entry and no others will be shown.
1724  * Note that this differs from the Pid and NSpid fields in
1725  * /proc/<pid>/status where Pid and NSpid are always shown relative to
1726  * the  pid namespace of the procfs instance. The difference becomes
1727  * obvious when sending around a pidfd between pid namespaces from a
1728  * different branch of the tree, i.e. where no ancestoral relation is
1729  * present between the pid namespaces:
1730  * - create two new pid namespaces ns1 and ns2 in the initial pid
1731  *   namespace (also take care to create new mount namespaces in the
1732  *   new pid namespace and mount procfs)
1733  * - create a process with a pidfd in ns1
1734  * - send pidfd from ns1 to ns2
1735  * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1736  *   have exactly one entry, which is 0
1737  */
1738 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1739 {
1740         struct pid *pid = f->private_data;
1741         struct pid_namespace *ns;
1742         pid_t nr = -1;
1743
1744         if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1745                 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1746                 nr = pid_nr_ns(pid, ns);
1747         }
1748
1749         seq_put_decimal_ll(m, "Pid:\t", nr);
1750
1751 #ifdef CONFIG_PID_NS
1752         seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1753         if (nr > 0) {
1754                 int i;
1755
1756                 /* If nr is non-zero it means that 'pid' is valid and that
1757                  * ns, i.e. the pid namespace associated with the procfs
1758                  * instance, is in the pid namespace hierarchy of pid.
1759                  * Start at one below the already printed level.
1760                  */
1761                 for (i = ns->level + 1; i <= pid->level; i++)
1762                         seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1763         }
1764 #endif
1765         seq_putc(m, '\n');
1766 }
1767 #endif
1768
1769 /*
1770  * Poll support for process exit notification.
1771  */
1772 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1773 {
1774         struct pid *pid = file->private_data;
1775         __poll_t poll_flags = 0;
1776
1777         poll_wait(file, &pid->wait_pidfd, pts);
1778
1779         /*
1780          * Inform pollers only when the whole thread group exits.
1781          * If the thread group leader exits before all other threads in the
1782          * group, then poll(2) should block, similar to the wait(2) family.
1783          */
1784         if (thread_group_exited(pid))
1785                 poll_flags = EPOLLIN | EPOLLRDNORM;
1786
1787         return poll_flags;
1788 }
1789
1790 const struct file_operations pidfd_fops = {
1791         .release = pidfd_release,
1792         .poll = pidfd_poll,
1793 #ifdef CONFIG_PROC_FS
1794         .show_fdinfo = pidfd_show_fdinfo,
1795 #endif
1796 };
1797
1798 static void __delayed_free_task(struct rcu_head *rhp)
1799 {
1800         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1801
1802         free_task(tsk);
1803 }
1804
1805 static __always_inline void delayed_free_task(struct task_struct *tsk)
1806 {
1807         if (IS_ENABLED(CONFIG_MEMCG))
1808                 call_rcu(&tsk->rcu, __delayed_free_task);
1809         else
1810                 free_task(tsk);
1811 }
1812
1813 /*
1814  * This creates a new process as a copy of the old one,
1815  * but does not actually start it yet.
1816  *
1817  * It copies the registers, and all the appropriate
1818  * parts of the process environment (as per the clone
1819  * flags). The actual kick-off is left to the caller.
1820  */
1821 static __latent_entropy struct task_struct *copy_process(
1822                                         struct pid *pid,
1823                                         int trace,
1824                                         int node,
1825                                         struct kernel_clone_args *args)
1826 {
1827         int pidfd = -1, retval;
1828         struct task_struct *p;
1829         struct multiprocess_signals delayed;
1830         struct file *pidfile = NULL;
1831         u64 clone_flags = args->flags;
1832         struct nsproxy *nsp = current->nsproxy;
1833
1834         /*
1835          * Don't allow sharing the root directory with processes in a different
1836          * namespace
1837          */
1838         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1839                 return ERR_PTR(-EINVAL);
1840
1841         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1842                 return ERR_PTR(-EINVAL);
1843
1844         /*
1845          * Thread groups must share signals as well, and detached threads
1846          * can only be started up within the thread group.
1847          */
1848         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1849                 return ERR_PTR(-EINVAL);
1850
1851         /*
1852          * Shared signal handlers imply shared VM. By way of the above,
1853          * thread groups also imply shared VM. Blocking this case allows
1854          * for various simplifications in other code.
1855          */
1856         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1857                 return ERR_PTR(-EINVAL);
1858
1859         /*
1860          * Siblings of global init remain as zombies on exit since they are
1861          * not reaped by their parent (swapper). To solve this and to avoid
1862          * multi-rooted process trees, prevent global and container-inits
1863          * from creating siblings.
1864          */
1865         if ((clone_flags & CLONE_PARENT) &&
1866                                 current->signal->flags & SIGNAL_UNKILLABLE)
1867                 return ERR_PTR(-EINVAL);
1868
1869         /*
1870          * If the new process will be in a different pid or user namespace
1871          * do not allow it to share a thread group with the forking task.
1872          */
1873         if (clone_flags & CLONE_THREAD) {
1874                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1875                     (task_active_pid_ns(current) != nsp->pid_ns_for_children))
1876                         return ERR_PTR(-EINVAL);
1877         }
1878
1879         /*
1880          * If the new process will be in a different time namespace
1881          * do not allow it to share VM or a thread group with the forking task.
1882          */
1883         if (clone_flags & (CLONE_THREAD | CLONE_VM)) {
1884                 if (nsp->time_ns != nsp->time_ns_for_children)
1885                         return ERR_PTR(-EINVAL);
1886         }
1887
1888         if (clone_flags & CLONE_PIDFD) {
1889                 /*
1890                  * - CLONE_DETACHED is blocked so that we can potentially
1891                  *   reuse it later for CLONE_PIDFD.
1892                  * - CLONE_THREAD is blocked until someone really needs it.
1893                  */
1894                 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1895                         return ERR_PTR(-EINVAL);
1896         }
1897
1898         /*
1899          * Force any signals received before this point to be delivered
1900          * before the fork happens.  Collect up signals sent to multiple
1901          * processes that happen during the fork and delay them so that
1902          * they appear to happen after the fork.
1903          */
1904         sigemptyset(&delayed.signal);
1905         INIT_HLIST_NODE(&delayed.node);
1906
1907         spin_lock_irq(&current->sighand->siglock);
1908         if (!(clone_flags & CLONE_THREAD))
1909                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1910         recalc_sigpending();
1911         spin_unlock_irq(&current->sighand->siglock);
1912         retval = -ERESTARTNOINTR;
1913         if (signal_pending(current))
1914                 goto fork_out;
1915
1916         retval = -ENOMEM;
1917         p = dup_task_struct(current, node);
1918         if (!p)
1919                 goto fork_out;
1920
1921         /*
1922          * This _must_ happen before we call free_task(), i.e. before we jump
1923          * to any of the bad_fork_* labels. This is to avoid freeing
1924          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1925          * kernel threads (PF_KTHREAD).
1926          */
1927         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1928         /*
1929          * Clear TID on mm_release()?
1930          */
1931         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1932
1933         ftrace_graph_init_task(p);
1934
1935         rt_mutex_init_task(p);
1936
1937         lockdep_assert_irqs_enabled();
1938 #ifdef CONFIG_PROVE_LOCKING
1939         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1940 #endif
1941         retval = -EAGAIN;
1942         if (atomic_read(&p->real_cred->user->processes) >=
1943                         task_rlimit(p, RLIMIT_NPROC)) {
1944                 if (p->real_cred->user != INIT_USER &&
1945                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1946                         goto bad_fork_free;
1947         }
1948         current->flags &= ~PF_NPROC_EXCEEDED;
1949
1950         retval = copy_creds(p, clone_flags);
1951         if (retval < 0)
1952                 goto bad_fork_free;
1953
1954         /*
1955          * If multiple threads are within copy_process(), then this check
1956          * triggers too late. This doesn't hurt, the check is only there
1957          * to stop root fork bombs.
1958          */
1959         retval = -EAGAIN;
1960         if (data_race(nr_threads >= max_threads))
1961                 goto bad_fork_cleanup_count;
1962
1963         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1964         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1965         p->flags |= PF_FORKNOEXEC;
1966         INIT_LIST_HEAD(&p->children);
1967         INIT_LIST_HEAD(&p->sibling);
1968         rcu_copy_process(p);
1969         p->vfork_done = NULL;
1970         spin_lock_init(&p->alloc_lock);
1971
1972         init_sigpending(&p->pending);
1973
1974         p->utime = p->stime = p->gtime = 0;
1975 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1976         p->utimescaled = p->stimescaled = 0;
1977 #endif
1978         prev_cputime_init(&p->prev_cputime);
1979
1980 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1981         seqcount_init(&p->vtime.seqcount);
1982         p->vtime.starttime = 0;
1983         p->vtime.state = VTIME_INACTIVE;
1984 #endif
1985
1986 #if defined(SPLIT_RSS_COUNTING)
1987         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1988 #endif
1989
1990         p->default_timer_slack_ns = current->timer_slack_ns;
1991
1992 #ifdef CONFIG_PSI
1993         p->psi_flags = 0;
1994 #endif
1995
1996         task_io_accounting_init(&p->ioac);
1997         acct_clear_integrals(p);
1998
1999         posix_cputimers_init(&p->posix_cputimers);
2000
2001         p->io_context = NULL;
2002         audit_set_context(p, NULL);
2003         cgroup_fork(p);
2004 #ifdef CONFIG_NUMA
2005         p->mempolicy = mpol_dup(p->mempolicy);
2006         if (IS_ERR(p->mempolicy)) {
2007                 retval = PTR_ERR(p->mempolicy);
2008                 p->mempolicy = NULL;
2009                 goto bad_fork_cleanup_threadgroup_lock;
2010         }
2011 #endif
2012 #ifdef CONFIG_CPUSETS
2013         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2014         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2015         seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2016 #endif
2017 #ifdef CONFIG_TRACE_IRQFLAGS
2018         memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2019         p->irqtrace.hardirq_disable_ip  = _THIS_IP_;
2020         p->irqtrace.softirq_enable_ip   = _THIS_IP_;
2021         p->softirqs_enabled             = 1;
2022         p->softirq_context              = 0;
2023 #endif
2024
2025         p->pagefault_disabled = 0;
2026
2027 #ifdef CONFIG_LOCKDEP
2028         lockdep_init_task(p);
2029 #endif
2030
2031 #ifdef CONFIG_DEBUG_MUTEXES
2032         p->blocked_on = NULL; /* not blocked yet */
2033 #endif
2034 #ifdef CONFIG_BCACHE
2035         p->sequential_io        = 0;
2036         p->sequential_io_avg    = 0;
2037 #endif
2038
2039         /* Perform scheduler related setup. Assign this task to a CPU. */
2040         retval = sched_fork(clone_flags, p);
2041         if (retval)
2042                 goto bad_fork_cleanup_policy;
2043
2044         retval = perf_event_init_task(p);
2045         if (retval)
2046                 goto bad_fork_cleanup_policy;
2047         retval = audit_alloc(p);
2048         if (retval)
2049                 goto bad_fork_cleanup_perf;
2050         /* copy all the process information */
2051         shm_init_task(p);
2052         retval = security_task_alloc(p, clone_flags);
2053         if (retval)
2054                 goto bad_fork_cleanup_audit;
2055         retval = copy_semundo(clone_flags, p);
2056         if (retval)
2057                 goto bad_fork_cleanup_security;
2058         retval = copy_files(clone_flags, p);
2059         if (retval)
2060                 goto bad_fork_cleanup_semundo;
2061         retval = copy_fs(clone_flags, p);
2062         if (retval)
2063                 goto bad_fork_cleanup_files;
2064         retval = copy_sighand(clone_flags, p);
2065         if (retval)
2066                 goto bad_fork_cleanup_fs;
2067         retval = copy_signal(clone_flags, p);
2068         if (retval)
2069                 goto bad_fork_cleanup_sighand;
2070         retval = copy_mm(clone_flags, p);
2071         if (retval)
2072                 goto bad_fork_cleanup_signal;
2073         retval = copy_namespaces(clone_flags, p);
2074         if (retval)
2075                 goto bad_fork_cleanup_mm;
2076         retval = copy_io(clone_flags, p);
2077         if (retval)
2078                 goto bad_fork_cleanup_namespaces;
2079         retval = copy_thread(clone_flags, args->stack, args->stack_size, p, args->tls);
2080         if (retval)
2081                 goto bad_fork_cleanup_io;
2082
2083         stackleak_task_init(p);
2084
2085         if (pid != &init_struct_pid) {
2086                 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2087                                 args->set_tid_size);
2088                 if (IS_ERR(pid)) {
2089                         retval = PTR_ERR(pid);
2090                         goto bad_fork_cleanup_thread;
2091                 }
2092         }
2093
2094         /*
2095          * This has to happen after we've potentially unshared the file
2096          * descriptor table (so that the pidfd doesn't leak into the child
2097          * if the fd table isn't shared).
2098          */
2099         if (clone_flags & CLONE_PIDFD) {
2100                 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2101                 if (retval < 0)
2102                         goto bad_fork_free_pid;
2103
2104                 pidfd = retval;
2105
2106                 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2107                                               O_RDWR | O_CLOEXEC);
2108                 if (IS_ERR(pidfile)) {
2109                         put_unused_fd(pidfd);
2110                         retval = PTR_ERR(pidfile);
2111                         goto bad_fork_free_pid;
2112                 }
2113                 get_pid(pid);   /* held by pidfile now */
2114
2115                 retval = put_user(pidfd, args->pidfd);
2116                 if (retval)
2117                         goto bad_fork_put_pidfd;
2118         }
2119
2120 #ifdef CONFIG_BLOCK
2121         p->plug = NULL;
2122 #endif
2123         futex_init_task(p);
2124
2125         /*
2126          * sigaltstack should be cleared when sharing the same VM
2127          */
2128         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2129                 sas_ss_reset(p);
2130
2131         /*
2132          * Syscall tracing and stepping should be turned off in the
2133          * child regardless of CLONE_PTRACE.
2134          */
2135         user_disable_single_step(p);
2136         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
2137 #ifdef TIF_SYSCALL_EMU
2138         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
2139 #endif
2140         clear_tsk_latency_tracing(p);
2141
2142         /* ok, now we should be set up.. */
2143         p->pid = pid_nr(pid);
2144         if (clone_flags & CLONE_THREAD) {
2145                 p->exit_signal = -1;
2146                 p->group_leader = current->group_leader;
2147                 p->tgid = current->tgid;
2148         } else {
2149                 if (clone_flags & CLONE_PARENT)
2150                         p->exit_signal = current->group_leader->exit_signal;
2151                 else
2152                         p->exit_signal = args->exit_signal;
2153                 p->group_leader = p;
2154                 p->tgid = p->pid;
2155         }
2156
2157         p->nr_dirtied = 0;
2158         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2159         p->dirty_paused_when = 0;
2160
2161         p->pdeath_signal = 0;
2162         INIT_LIST_HEAD(&p->thread_group);
2163         p->task_works = NULL;
2164
2165         /*
2166          * Ensure that the cgroup subsystem policies allow the new process to be
2167          * forked. It should be noted the the new process's css_set can be changed
2168          * between here and cgroup_post_fork() if an organisation operation is in
2169          * progress.
2170          */
2171         retval = cgroup_can_fork(p, args);
2172         if (retval)
2173                 goto bad_fork_put_pidfd;
2174
2175         /*
2176          * From this point on we must avoid any synchronous user-space
2177          * communication until we take the tasklist-lock. In particular, we do
2178          * not want user-space to be able to predict the process start-time by
2179          * stalling fork(2) after we recorded the start_time but before it is
2180          * visible to the system.
2181          */
2182
2183         p->start_time = ktime_get_ns();
2184         p->start_boottime = ktime_get_boottime_ns();
2185
2186         /*
2187          * Make it visible to the rest of the system, but dont wake it up yet.
2188          * Need tasklist lock for parent etc handling!
2189          */
2190         write_lock_irq(&tasklist_lock);
2191
2192         /* CLONE_PARENT re-uses the old parent */
2193         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2194                 p->real_parent = current->real_parent;
2195                 p->parent_exec_id = current->parent_exec_id;
2196         } else {
2197                 p->real_parent = current;
2198                 p->parent_exec_id = current->self_exec_id;
2199         }
2200
2201         klp_copy_process(p);
2202
2203         spin_lock(&current->sighand->siglock);
2204
2205         /*
2206          * Copy seccomp details explicitly here, in case they were changed
2207          * before holding sighand lock.
2208          */
2209         copy_seccomp(p);
2210
2211         rseq_fork(p, clone_flags);
2212
2213         /* Don't start children in a dying pid namespace */
2214         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2215                 retval = -ENOMEM;
2216                 goto bad_fork_cancel_cgroup;
2217         }
2218
2219         /* Let kill terminate clone/fork in the middle */
2220         if (fatal_signal_pending(current)) {
2221                 retval = -EINTR;
2222                 goto bad_fork_cancel_cgroup;
2223         }
2224
2225         /* past the last point of failure */
2226         if (pidfile)
2227                 fd_install(pidfd, pidfile);
2228
2229         init_task_pid_links(p);
2230         if (likely(p->pid)) {
2231                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2232
2233                 init_task_pid(p, PIDTYPE_PID, pid);
2234                 if (thread_group_leader(p)) {
2235                         init_task_pid(p, PIDTYPE_TGID, pid);
2236                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2237                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2238
2239                         if (is_child_reaper(pid)) {
2240                                 ns_of_pid(pid)->child_reaper = p;
2241                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2242                         }
2243                         p->signal->shared_pending.signal = delayed.signal;
2244                         p->signal->tty = tty_kref_get(current->signal->tty);
2245                         /*
2246                          * Inherit has_child_subreaper flag under the same
2247                          * tasklist_lock with adding child to the process tree
2248                          * for propagate_has_child_subreaper optimization.
2249                          */
2250                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2251                                                          p->real_parent->signal->is_child_subreaper;
2252                         list_add_tail(&p->sibling, &p->real_parent->children);
2253                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2254                         attach_pid(p, PIDTYPE_TGID);
2255                         attach_pid(p, PIDTYPE_PGID);
2256                         attach_pid(p, PIDTYPE_SID);
2257                         __this_cpu_inc(process_counts);
2258                 } else {
2259                         current->signal->nr_threads++;
2260                         atomic_inc(&current->signal->live);
2261                         refcount_inc(&current->signal->sigcnt);
2262                         task_join_group_stop(p);
2263                         list_add_tail_rcu(&p->thread_group,
2264                                           &p->group_leader->thread_group);
2265                         list_add_tail_rcu(&p->thread_node,
2266                                           &p->signal->thread_head);
2267                 }
2268                 attach_pid(p, PIDTYPE_PID);
2269                 nr_threads++;
2270         }
2271         total_forks++;
2272         hlist_del_init(&delayed.node);
2273         spin_unlock(&current->sighand->siglock);
2274         syscall_tracepoint_update(p);
2275         write_unlock_irq(&tasklist_lock);
2276
2277         proc_fork_connector(p);
2278         sched_post_fork(p);
2279         cgroup_post_fork(p, args);
2280         perf_event_fork(p);
2281
2282         trace_task_newtask(p, clone_flags);
2283         uprobe_copy_process(p, clone_flags);
2284
2285         return p;
2286
2287 bad_fork_cancel_cgroup:
2288         spin_unlock(&current->sighand->siglock);
2289         write_unlock_irq(&tasklist_lock);
2290         cgroup_cancel_fork(p, args);
2291 bad_fork_put_pidfd:
2292         if (clone_flags & CLONE_PIDFD) {
2293                 fput(pidfile);
2294                 put_unused_fd(pidfd);
2295         }
2296 bad_fork_free_pid:
2297         if (pid != &init_struct_pid)
2298                 free_pid(pid);
2299 bad_fork_cleanup_thread:
2300         exit_thread(p);
2301 bad_fork_cleanup_io:
2302         if (p->io_context)
2303                 exit_io_context(p);
2304 bad_fork_cleanup_namespaces:
2305         exit_task_namespaces(p);
2306 bad_fork_cleanup_mm:
2307         if (p->mm) {
2308                 mm_clear_owner(p->mm, p);
2309                 mmput(p->mm);
2310         }
2311 bad_fork_cleanup_signal:
2312         if (!(clone_flags & CLONE_THREAD))
2313                 free_signal_struct(p->signal);
2314 bad_fork_cleanup_sighand:
2315         __cleanup_sighand(p->sighand);
2316 bad_fork_cleanup_fs:
2317         exit_fs(p); /* blocking */
2318 bad_fork_cleanup_files:
2319         exit_files(p); /* blocking */
2320 bad_fork_cleanup_semundo:
2321         exit_sem(p);
2322 bad_fork_cleanup_security:
2323         security_task_free(p);
2324 bad_fork_cleanup_audit:
2325         audit_free(p);
2326 bad_fork_cleanup_perf:
2327         perf_event_free_task(p);
2328 bad_fork_cleanup_policy:
2329         lockdep_free_task(p);
2330 #ifdef CONFIG_NUMA
2331         mpol_put(p->mempolicy);
2332 bad_fork_cleanup_threadgroup_lock:
2333 #endif
2334         delayacct_tsk_free(p);
2335 bad_fork_cleanup_count:
2336         atomic_dec(&p->cred->user->processes);
2337         exit_creds(p);
2338 bad_fork_free:
2339         p->state = TASK_DEAD;
2340         put_task_stack(p);
2341         delayed_free_task(p);
2342 fork_out:
2343         spin_lock_irq(&current->sighand->siglock);
2344         hlist_del_init(&delayed.node);
2345         spin_unlock_irq(&current->sighand->siglock);
2346         return ERR_PTR(retval);
2347 }
2348
2349 static inline void init_idle_pids(struct task_struct *idle)
2350 {
2351         enum pid_type type;
2352
2353         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2354                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2355                 init_task_pid(idle, type, &init_struct_pid);
2356         }
2357 }
2358
2359 struct task_struct *fork_idle(int cpu)
2360 {
2361         struct task_struct *task;
2362         struct kernel_clone_args args = {
2363                 .flags = CLONE_VM,
2364         };
2365
2366         task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2367         if (!IS_ERR(task)) {
2368                 init_idle_pids(task);
2369                 init_idle(task, cpu);
2370         }
2371
2372         return task;
2373 }
2374
2375 struct mm_struct *copy_init_mm(void)
2376 {
2377         return dup_mm(NULL, &init_mm);
2378 }
2379
2380 /*
2381  *  Ok, this is the main fork-routine.
2382  *
2383  * It copies the process, and if successful kick-starts
2384  * it and waits for it to finish using the VM if required.
2385  *
2386  * args->exit_signal is expected to be checked for sanity by the caller.
2387  */
2388 long _do_fork(struct kernel_clone_args *args)
2389 {
2390         u64 clone_flags = args->flags;
2391         struct completion vfork;
2392         struct pid *pid;
2393         struct task_struct *p;
2394         int trace = 0;
2395         long nr;
2396
2397         /*
2398          * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2399          * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2400          * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2401          * field in struct clone_args and it still doesn't make sense to have
2402          * them both point at the same memory location. Performing this check
2403          * here has the advantage that we don't need to have a separate helper
2404          * to check for legacy clone().
2405          */
2406         if ((args->flags & CLONE_PIDFD) &&
2407             (args->flags & CLONE_PARENT_SETTID) &&
2408             (args->pidfd == args->parent_tid))
2409                 return -EINVAL;
2410
2411         /*
2412          * Determine whether and which event to report to ptracer.  When
2413          * called from kernel_thread or CLONE_UNTRACED is explicitly
2414          * requested, no event is reported; otherwise, report if the event
2415          * for the type of forking is enabled.
2416          */
2417         if (!(clone_flags & CLONE_UNTRACED)) {
2418                 if (clone_flags & CLONE_VFORK)
2419                         trace = PTRACE_EVENT_VFORK;
2420                 else if (args->exit_signal != SIGCHLD)
2421                         trace = PTRACE_EVENT_CLONE;
2422                 else
2423                         trace = PTRACE_EVENT_FORK;
2424
2425                 if (likely(!ptrace_event_enabled(current, trace)))
2426                         trace = 0;
2427         }
2428
2429         p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2430         add_latent_entropy();
2431
2432         if (IS_ERR(p))
2433                 return PTR_ERR(p);
2434
2435         /*
2436          * Do this prior waking up the new thread - the thread pointer
2437          * might get invalid after that point, if the thread exits quickly.
2438          */
2439         trace_sched_process_fork(current, p);
2440
2441         pid = get_task_pid(p, PIDTYPE_PID);
2442         nr = pid_vnr(pid);
2443
2444         if (clone_flags & CLONE_PARENT_SETTID)
2445                 put_user(nr, args->parent_tid);
2446
2447         if (clone_flags & CLONE_VFORK) {
2448                 p->vfork_done = &vfork;
2449                 init_completion(&vfork);
2450                 get_task_struct(p);
2451         }
2452
2453         wake_up_new_task(p);
2454
2455         /* forking complete and child started to run, tell ptracer */
2456         if (unlikely(trace))
2457                 ptrace_event_pid(trace, pid);
2458
2459         if (clone_flags & CLONE_VFORK) {
2460                 if (!wait_for_vfork_done(p, &vfork))
2461                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2462         }
2463
2464         put_pid(pid);
2465         return nr;
2466 }
2467
2468 /*
2469  * Create a kernel thread.
2470  */
2471 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2472 {
2473         struct kernel_clone_args args = {
2474                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2475                                     CLONE_UNTRACED) & ~CSIGNAL),
2476                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2477                 .stack          = (unsigned long)fn,
2478                 .stack_size     = (unsigned long)arg,
2479         };
2480
2481         return _do_fork(&args);
2482 }
2483
2484 #ifdef __ARCH_WANT_SYS_FORK
2485 SYSCALL_DEFINE0(fork)
2486 {
2487 #ifdef CONFIG_MMU
2488         struct kernel_clone_args args = {
2489                 .exit_signal = SIGCHLD,
2490         };
2491
2492         return _do_fork(&args);
2493 #else
2494         /* can not support in nommu mode */
2495         return -EINVAL;
2496 #endif
2497 }
2498 #endif
2499
2500 #ifdef __ARCH_WANT_SYS_VFORK
2501 SYSCALL_DEFINE0(vfork)
2502 {
2503         struct kernel_clone_args args = {
2504                 .flags          = CLONE_VFORK | CLONE_VM,
2505                 .exit_signal    = SIGCHLD,
2506         };
2507
2508         return _do_fork(&args);
2509 }
2510 #endif
2511
2512 #ifdef __ARCH_WANT_SYS_CLONE
2513 #ifdef CONFIG_CLONE_BACKWARDS
2514 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2515                  int __user *, parent_tidptr,
2516                  unsigned long, tls,
2517                  int __user *, child_tidptr)
2518 #elif defined(CONFIG_CLONE_BACKWARDS2)
2519 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2520                  int __user *, parent_tidptr,
2521                  int __user *, child_tidptr,
2522                  unsigned long, tls)
2523 #elif defined(CONFIG_CLONE_BACKWARDS3)
2524 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2525                 int, stack_size,
2526                 int __user *, parent_tidptr,
2527                 int __user *, child_tidptr,
2528                 unsigned long, tls)
2529 #else
2530 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2531                  int __user *, parent_tidptr,
2532                  int __user *, child_tidptr,
2533                  unsigned long, tls)
2534 #endif
2535 {
2536         struct kernel_clone_args args = {
2537                 .flags          = (lower_32_bits(clone_flags) & ~CSIGNAL),
2538                 .pidfd          = parent_tidptr,
2539                 .child_tid      = child_tidptr,
2540                 .parent_tid     = parent_tidptr,
2541                 .exit_signal    = (lower_32_bits(clone_flags) & CSIGNAL),
2542                 .stack          = newsp,
2543                 .tls            = tls,
2544         };
2545
2546         return _do_fork(&args);
2547 }
2548 #endif
2549
2550 #ifdef __ARCH_WANT_SYS_CLONE3
2551
2552 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2553                                               struct clone_args __user *uargs,
2554                                               size_t usize)
2555 {
2556         int err;
2557         struct clone_args args;
2558         pid_t *kset_tid = kargs->set_tid;
2559
2560         BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2561                      CLONE_ARGS_SIZE_VER0);
2562         BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2563                      CLONE_ARGS_SIZE_VER1);
2564         BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2565                      CLONE_ARGS_SIZE_VER2);
2566         BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2567
2568         if (unlikely(usize > PAGE_SIZE))
2569                 return -E2BIG;
2570         if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2571                 return -EINVAL;
2572
2573         err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2574         if (err)
2575                 return err;
2576
2577         if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2578                 return -EINVAL;
2579
2580         if (unlikely(!args.set_tid && args.set_tid_size > 0))
2581                 return -EINVAL;
2582
2583         if (unlikely(args.set_tid && args.set_tid_size == 0))
2584                 return -EINVAL;
2585
2586         /*
2587          * Verify that higher 32bits of exit_signal are unset and that
2588          * it is a valid signal
2589          */
2590         if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2591                      !valid_signal(args.exit_signal)))
2592                 return -EINVAL;
2593
2594         if ((args.flags & CLONE_INTO_CGROUP) &&
2595             (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2596                 return -EINVAL;
2597
2598         *kargs = (struct kernel_clone_args){
2599                 .flags          = args.flags,
2600                 .pidfd          = u64_to_user_ptr(args.pidfd),
2601                 .child_tid      = u64_to_user_ptr(args.child_tid),
2602                 .parent_tid     = u64_to_user_ptr(args.parent_tid),
2603                 .exit_signal    = args.exit_signal,
2604                 .stack          = args.stack,
2605                 .stack_size     = args.stack_size,
2606                 .tls            = args.tls,
2607                 .set_tid_size   = args.set_tid_size,
2608                 .cgroup         = args.cgroup,
2609         };
2610
2611         if (args.set_tid &&
2612                 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2613                         (kargs->set_tid_size * sizeof(pid_t))))
2614                 return -EFAULT;
2615
2616         kargs->set_tid = kset_tid;
2617
2618         return 0;
2619 }
2620
2621 /**
2622  * clone3_stack_valid - check and prepare stack
2623  * @kargs: kernel clone args
2624  *
2625  * Verify that the stack arguments userspace gave us are sane.
2626  * In addition, set the stack direction for userspace since it's easy for us to
2627  * determine.
2628  */
2629 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2630 {
2631         if (kargs->stack == 0) {
2632                 if (kargs->stack_size > 0)
2633                         return false;
2634         } else {
2635                 if (kargs->stack_size == 0)
2636                         return false;
2637
2638                 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2639                         return false;
2640
2641 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2642                 kargs->stack += kargs->stack_size;
2643 #endif
2644         }
2645
2646         return true;
2647 }
2648
2649 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2650 {
2651         /* Verify that no unknown flags are passed along. */
2652         if (kargs->flags &
2653             ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2654                 return false;
2655
2656         /*
2657          * - make the CLONE_DETACHED bit reuseable for clone3
2658          * - make the CSIGNAL bits reuseable for clone3
2659          */
2660         if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2661                 return false;
2662
2663         if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2664             (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2665                 return false;
2666
2667         if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2668             kargs->exit_signal)
2669                 return false;
2670
2671         if (!clone3_stack_valid(kargs))
2672                 return false;
2673
2674         return true;
2675 }
2676
2677 /**
2678  * clone3 - create a new process with specific properties
2679  * @uargs: argument structure
2680  * @size:  size of @uargs
2681  *
2682  * clone3() is the extensible successor to clone()/clone2().
2683  * It takes a struct as argument that is versioned by its size.
2684  *
2685  * Return: On success, a positive PID for the child process.
2686  *         On error, a negative errno number.
2687  */
2688 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2689 {
2690         int err;
2691
2692         struct kernel_clone_args kargs;
2693         pid_t set_tid[MAX_PID_NS_LEVEL];
2694
2695         kargs.set_tid = set_tid;
2696
2697         err = copy_clone_args_from_user(&kargs, uargs, size);
2698         if (err)
2699                 return err;
2700
2701         if (!clone3_args_valid(&kargs))
2702                 return -EINVAL;
2703
2704         return _do_fork(&kargs);
2705 }
2706 #endif
2707
2708 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2709 {
2710         struct task_struct *leader, *parent, *child;
2711         int res;
2712
2713         read_lock(&tasklist_lock);
2714         leader = top = top->group_leader;
2715 down:
2716         for_each_thread(leader, parent) {
2717                 list_for_each_entry(child, &parent->children, sibling) {
2718                         res = visitor(child, data);
2719                         if (res) {
2720                                 if (res < 0)
2721                                         goto out;
2722                                 leader = child;
2723                                 goto down;
2724                         }
2725 up:
2726                         ;
2727                 }
2728         }
2729
2730         if (leader != top) {
2731                 child = leader;
2732                 parent = child->real_parent;
2733                 leader = parent->group_leader;
2734                 goto up;
2735         }
2736 out:
2737         read_unlock(&tasklist_lock);
2738 }
2739
2740 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2741 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2742 #endif
2743
2744 static void sighand_ctor(void *data)
2745 {
2746         struct sighand_struct *sighand = data;
2747
2748         spin_lock_init(&sighand->siglock);
2749         init_waitqueue_head(&sighand->signalfd_wqh);
2750 }
2751
2752 void __init proc_caches_init(void)
2753 {
2754         unsigned int mm_size;
2755
2756         sighand_cachep = kmem_cache_create("sighand_cache",
2757                         sizeof(struct sighand_struct), 0,
2758                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2759                         SLAB_ACCOUNT, sighand_ctor);
2760         signal_cachep = kmem_cache_create("signal_cache",
2761                         sizeof(struct signal_struct), 0,
2762                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2763                         NULL);
2764         files_cachep = kmem_cache_create("files_cache",
2765                         sizeof(struct files_struct), 0,
2766                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2767                         NULL);
2768         fs_cachep = kmem_cache_create("fs_cache",
2769                         sizeof(struct fs_struct), 0,
2770                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2771                         NULL);
2772
2773         /*
2774          * The mm_cpumask is located at the end of mm_struct, and is
2775          * dynamically sized based on the maximum CPU number this system
2776          * can have, taking hotplug into account (nr_cpu_ids).
2777          */
2778         mm_size = sizeof(struct mm_struct) + cpumask_size();
2779
2780         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2781                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2782                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2783                         offsetof(struct mm_struct, saved_auxv),
2784                         sizeof_field(struct mm_struct, saved_auxv),
2785                         NULL);
2786         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2787         mmap_init();
2788         nsproxy_cache_init();
2789 }
2790
2791 /*
2792  * Check constraints on flags passed to the unshare system call.
2793  */
2794 static int check_unshare_flags(unsigned long unshare_flags)
2795 {
2796         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2797                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2798                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2799                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
2800                                 CLONE_NEWTIME))
2801                 return -EINVAL;
2802         /*
2803          * Not implemented, but pretend it works if there is nothing
2804          * to unshare.  Note that unsharing the address space or the
2805          * signal handlers also need to unshare the signal queues (aka
2806          * CLONE_THREAD).
2807          */
2808         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2809                 if (!thread_group_empty(current))
2810                         return -EINVAL;
2811         }
2812         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2813                 if (refcount_read(&current->sighand->count) > 1)
2814                         return -EINVAL;
2815         }
2816         if (unshare_flags & CLONE_VM) {
2817                 if (!current_is_single_threaded())
2818                         return -EINVAL;
2819         }
2820
2821         return 0;
2822 }
2823
2824 /*
2825  * Unshare the filesystem structure if it is being shared
2826  */
2827 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2828 {
2829         struct fs_struct *fs = current->fs;
2830
2831         if (!(unshare_flags & CLONE_FS) || !fs)
2832                 return 0;
2833
2834         /* don't need lock here; in the worst case we'll do useless copy */
2835         if (fs->users == 1)
2836                 return 0;
2837
2838         *new_fsp = copy_fs_struct(fs);
2839         if (!*new_fsp)
2840                 return -ENOMEM;
2841
2842         return 0;
2843 }
2844
2845 /*
2846  * Unshare file descriptor table if it is being shared
2847  */
2848 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
2849                struct files_struct **new_fdp)
2850 {
2851         struct files_struct *fd = current->files;
2852         int error = 0;
2853
2854         if ((unshare_flags & CLONE_FILES) &&
2855             (fd && atomic_read(&fd->count) > 1)) {
2856                 *new_fdp = dup_fd(fd, max_fds, &error);
2857                 if (!*new_fdp)
2858                         return error;
2859         }
2860
2861         return 0;
2862 }
2863
2864 /*
2865  * unshare allows a process to 'unshare' part of the process
2866  * context which was originally shared using clone.  copy_*
2867  * functions used by _do_fork() cannot be used here directly
2868  * because they modify an inactive task_struct that is being
2869  * constructed. Here we are modifying the current, active,
2870  * task_struct.
2871  */
2872 int ksys_unshare(unsigned long unshare_flags)
2873 {
2874         struct fs_struct *fs, *new_fs = NULL;
2875         struct files_struct *fd, *new_fd = NULL;
2876         struct cred *new_cred = NULL;
2877         struct nsproxy *new_nsproxy = NULL;
2878         int do_sysvsem = 0;
2879         int err;
2880
2881         /*
2882          * If unsharing a user namespace must also unshare the thread group
2883          * and unshare the filesystem root and working directories.
2884          */
2885         if (unshare_flags & CLONE_NEWUSER)
2886                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2887         /*
2888          * If unsharing vm, must also unshare signal handlers.
2889          */
2890         if (unshare_flags & CLONE_VM)
2891                 unshare_flags |= CLONE_SIGHAND;
2892         /*
2893          * If unsharing a signal handlers, must also unshare the signal queues.
2894          */
2895         if (unshare_flags & CLONE_SIGHAND)
2896                 unshare_flags |= CLONE_THREAD;
2897         /*
2898          * If unsharing namespace, must also unshare filesystem information.
2899          */
2900         if (unshare_flags & CLONE_NEWNS)
2901                 unshare_flags |= CLONE_FS;
2902
2903         err = check_unshare_flags(unshare_flags);
2904         if (err)
2905                 goto bad_unshare_out;
2906         /*
2907          * CLONE_NEWIPC must also detach from the undolist: after switching
2908          * to a new ipc namespace, the semaphore arrays from the old
2909          * namespace are unreachable.
2910          */
2911         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2912                 do_sysvsem = 1;
2913         err = unshare_fs(unshare_flags, &new_fs);
2914         if (err)
2915                 goto bad_unshare_out;
2916         err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
2917         if (err)
2918                 goto bad_unshare_cleanup_fs;
2919         err = unshare_userns(unshare_flags, &new_cred);
2920         if (err)
2921                 goto bad_unshare_cleanup_fd;
2922         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2923                                          new_cred, new_fs);
2924         if (err)
2925                 goto bad_unshare_cleanup_cred;
2926
2927         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2928                 if (do_sysvsem) {
2929                         /*
2930                          * CLONE_SYSVSEM is equivalent to sys_exit().
2931                          */
2932                         exit_sem(current);
2933                 }
2934                 if (unshare_flags & CLONE_NEWIPC) {
2935                         /* Orphan segments in old ns (see sem above). */
2936                         exit_shm(current);
2937                         shm_init_task(current);
2938                 }
2939
2940                 if (new_nsproxy)
2941                         switch_task_namespaces(current, new_nsproxy);
2942
2943                 task_lock(current);
2944
2945                 if (new_fs) {
2946                         fs = current->fs;
2947                         spin_lock(&fs->lock);
2948                         current->fs = new_fs;
2949                         if (--fs->users)
2950                                 new_fs = NULL;
2951                         else
2952                                 new_fs = fs;
2953                         spin_unlock(&fs->lock);
2954                 }
2955
2956                 if (new_fd) {
2957                         fd = current->files;
2958                         current->files = new_fd;
2959                         new_fd = fd;
2960                 }
2961
2962                 task_unlock(current);
2963
2964                 if (new_cred) {
2965                         /* Install the new user namespace */
2966                         commit_creds(new_cred);
2967                         new_cred = NULL;
2968                 }
2969         }
2970
2971         perf_event_namespaces(current);
2972
2973 bad_unshare_cleanup_cred:
2974         if (new_cred)
2975                 put_cred(new_cred);
2976 bad_unshare_cleanup_fd:
2977         if (new_fd)
2978                 put_files_struct(new_fd);
2979
2980 bad_unshare_cleanup_fs:
2981         if (new_fs)
2982                 free_fs_struct(new_fs);
2983
2984 bad_unshare_out:
2985         return err;
2986 }
2987
2988 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2989 {
2990         return ksys_unshare(unshare_flags);
2991 }
2992
2993 /*
2994  *      Helper to unshare the files of the current task.
2995  *      We don't want to expose copy_files internals to
2996  *      the exec layer of the kernel.
2997  */
2998
2999 int unshare_files(struct files_struct **displaced)
3000 {
3001         struct task_struct *task = current;
3002         struct files_struct *copy = NULL;
3003         int error;
3004
3005         error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, &copy);
3006         if (error || !copy) {
3007                 *displaced = NULL;
3008                 return error;
3009         }
3010         *displaced = task->files;
3011         task_lock(task);
3012         task->files = copy;
3013         task_unlock(task);
3014         return 0;
3015 }
3016
3017 int sysctl_max_threads(struct ctl_table *table, int write,
3018                        void *buffer, size_t *lenp, loff_t *ppos)
3019 {
3020         struct ctl_table t;
3021         int ret;
3022         int threads = max_threads;
3023         int min = 1;
3024         int max = MAX_THREADS;
3025
3026         t = *table;
3027         t.data = &threads;
3028         t.extra1 = &min;
3029         t.extra2 = &max;
3030
3031         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3032         if (ret || !write)
3033                 return ret;
3034
3035         max_threads = threads;
3036
3037         return 0;
3038 }