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