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 __put_task_struct_rcu_cb(struct rcu_head *rhp)
 753 {
 754         struct task_struct *task = container_of(rhp, struct task_struct, rcu);
 755
 756         __put_task_struct(task);
 757 }
 758 EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb);
 759
 760 void __init __weak arch_task_cache_init(void) { }
 761
 762 /*
 763  * set_max_threads
 764  */
 765 static void set_max_threads(unsigned int max_threads_suggested)
 766 {
 767         u64 threads;
 768         unsigned long nr_pages = totalram_pages();
 769
 770         /*
 771          * The number of threads shall be limited such that the thread
 772          * structures may only consume a small part of the available memory.
 773          */
 774         if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
 775                 threads = MAX_THREADS;
 776         else
 777                 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
 778                                     (u64) THREAD_SIZE * 8UL);
 779
 780         if (threads > max_threads_suggested)
 781                 threads = max_threads_suggested;
 782
 783         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 784 }
 785
 786 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 787 /* Initialized by the architecture: */
 788 int arch_task_struct_size __read_mostly;
 789 #endif
 790
 791 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 792 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 793 {
 794         /* Fetch thread_struct whitelist for the architecture. */
 795         arch_thread_struct_whitelist(offset, size);
 796
 797         /*
 798          * Handle zero-sized whitelist or empty thread_struct, otherwise
 799          * adjust offset to position of thread_struct in task_struct.
 800          */
 801         if (unlikely(*size == 0))
 802                 *offset = 0;
 803         else
 804                 *offset += offsetof(struct task_struct, thread);
 805 }
 806 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
 807
 808 void __init fork_init(void)
 809 {
 810         int i;
 811 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 812 #ifndef ARCH_MIN_TASKALIGN
 813 #define ARCH_MIN_TASKALIGN      0
 814 #endif
 815         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 816         unsigned long useroffset, usersize;
 817
 818         /* create a slab on which task_structs can be allocated */
 819         task_struct_whitelist(&useroffset, &usersize);
 820         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 821                         arch_task_struct_size, align,
 822                         SLAB_PANIC|SLAB_ACCOUNT,
 823                         useroffset, usersize, NULL);
 824 #endif
 825
 826         /* do the arch specific task caches init */
 827         arch_task_cache_init();
 828
 829         set_max_threads(MAX_THREADS);
 830
 831         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 832         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 833         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 834                 init_task.signal->rlim[RLIMIT_NPROC];
 835
 836         for (i = 0; i < UCOUNT_COUNTS; i++) {
 837                 init_user_ns.ucount_max[i] = max_threads/2;
 838         }
 839
 840 #ifdef CONFIG_VMAP_STACK
 841         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 842                           NULL, free_vm_stack_cache);
 843 #endif
 844
 845         lockdep_init_task(&init_task);
 846         uprobes_init();
 847 }
 848
 849 int __weak arch_dup_task_struct(struct task_struct *dst,
 850                                                struct task_struct *src)
 851 {
 852         *dst = *src;
 853         return 0;
 854 }
 855
 856 void set_task_stack_end_magic(struct task_struct *tsk)
 857 {
 858         unsigned long *stackend;
 859
 860         stackend = end_of_stack(tsk);
 861         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 862 }
 863
 864 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 865 {
 866         struct task_struct *tsk;
 867         unsigned long *stack;
 868         struct vm_struct *stack_vm_area __maybe_unused;
 869         int err;
 870
 871         if (node == NUMA_NO_NODE)
 872                 node = tsk_fork_get_node(orig);
 873         tsk = alloc_task_struct_node(node);
 874         if (!tsk)
 875                 return NULL;
 876
 877         stack = alloc_thread_stack_node(tsk, node);
 878         if (!stack)
 879                 goto free_tsk;
 880
 881         if (memcg_charge_kernel_stack(tsk))
 882                 goto free_stack;
 883
 884         stack_vm_area = task_stack_vm_area(tsk);
 885
 886         err = arch_dup_task_struct(tsk, orig);
 887
 888         /*
 889          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 890          * sure they're properly initialized before using any stack-related
 891          * functions again.
 892          */
 893         tsk->stack = stack;
 894 #ifdef CONFIG_VMAP_STACK
 895         tsk->stack_vm_area = stack_vm_area;
 896 #endif
 897 #ifdef CONFIG_THREAD_INFO_IN_TASK
 898         refcount_set(&tsk->stack_refcount, 1);
 899 #endif
 900
 901         if (err)
 902                 goto free_stack;
 903
 904 #ifdef CONFIG_SECCOMP
 905         /*
 906          * We must handle setting up seccomp filters once we're under
 907          * the sighand lock in case orig has changed between now and
 908          * then. Until then, filter must be NULL to avoid messing up
 909          * the usage counts on the error path calling free_task.
 910          */
 911         tsk->seccomp.filter = NULL;
 912 #endif
 913
 914         setup_thread_stack(tsk, orig);
 915         clear_user_return_notifier(tsk);
 916         clear_tsk_need_resched(tsk);
 917         set_task_stack_end_magic(tsk);
 918
 919 #ifdef CONFIG_STACKPROTECTOR
 920         tsk->stack_canary = get_random_canary();
 921 #endif
 922         if (orig->cpus_ptr == &orig->cpus_mask)
 923                 tsk->cpus_ptr = &tsk->cpus_mask;
 924
 925         /*
 926          * One for the user space visible state that goes away when reaped.
 927          * One for the scheduler.
 928          */
 929         refcount_set(&tsk->rcu_users, 2);
 930         /* One for the rcu users */
 931         refcount_set(&tsk->usage, 1);
 932 #ifdef CONFIG_BLK_DEV_IO_TRACE
 933         tsk->btrace_seq = 0;
 934 #endif
 935         tsk->splice_pipe = NULL;
 936         tsk->task_frag.page = NULL;
 937         tsk->wake_q.next = NULL;
 938
 939         account_kernel_stack(tsk, 1);
 940
 941         kcov_task_init(tsk);
 942
 943 #ifdef CONFIG_FAULT_INJECTION
 944         tsk->fail_nth = 0;
 945 #endif
 946
 947 #ifdef CONFIG_BLK_CGROUP
 948         tsk->throttle_queue = NULL;
 949         tsk->use_memdelay = 0;
 950 #endif
 951
 952 #ifdef CONFIG_MEMCG
 953         tsk->active_memcg = NULL;
 954 #endif
 955         return tsk;
 956
 957 free_stack:
 958         free_thread_stack(tsk);
 959 free_tsk:
 960         free_task_struct(tsk);
 961         return NULL;
 962 }
 963
 964 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 965
 966 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 967
 968 static int __init coredump_filter_setup(char *s)
 969 {
 970         default_dump_filter =
 971                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 972                 MMF_DUMP_FILTER_MASK;
 973         return 1;
 974 }
 975
 976 __setup("coredump_filter=", coredump_filter_setup);
 977
 978 #include <linux/init_task.h>
 979
 980 static void mm_init_aio(struct mm_struct *mm)
 981 {
 982 #ifdef CONFIG_AIO
 983         spin_lock_init(&mm->ioctx_lock);
 984         mm->ioctx_table = NULL;
 985 #endif
 986 }
 987
 988 static __always_inline void mm_clear_owner(struct mm_struct *mm,
 989                                            struct task_struct *p)
 990 {
 991 #ifdef CONFIG_MEMCG
 992         if (mm->owner == p)
 993                 WRITE_ONCE(mm->owner, NULL);
 994 #endif
 995 }
 996
 997 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 998 {
 999 #ifdef CONFIG_MEMCG
1000         mm->owner = p;
1001 #endif
1002 }
1003
1004 static void mm_init_uprobes_state(struct mm_struct *mm)
1005 {
1006 #ifdef CONFIG_UPROBES
1007         mm->uprobes_state.xol_area = NULL;
1008 #endif
1009 }
1010
1011 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1012         struct user_namespace *user_ns)
1013 {
1014         mm->mmap = NULL;
1015         mm->mm_rb = RB_ROOT;
1016         mm->vmacache_seqnum = 0;
1017         atomic_set(&mm->mm_users, 1);
1018         atomic_set(&mm->mm_count, 1);
1019         init_rwsem(&mm->mmap_sem);
1020         INIT_LIST_HEAD(&mm->mmlist);
1021         mm->core_state = NULL;
1022         mm_pgtables_bytes_init(mm);
1023         mm->map_count = 0;
1024         mm->locked_vm = 0;
1025         atomic64_set(&mm->pinned_vm, 0);
1026         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1027         spin_lock_init(&mm->page_table_lock);
1028         spin_lock_init(&mm->arg_lock);
1029         mm_init_cpumask(mm);
1030         mm_init_aio(mm);
1031         mm_init_owner(mm, p);
1032         RCU_INIT_POINTER(mm->exe_file, NULL);
1033         mmu_notifier_mm_init(mm);
1034         init_tlb_flush_pending(mm);
1035 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1036         mm->pmd_huge_pte = NULL;
1037 #endif
1038         mm_init_uprobes_state(mm);
1039         hugetlb_count_init(mm);
1040
1041         if (current->mm) {
1042                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1043                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1044         } else {
1045                 mm->flags = default_dump_filter;
1046                 mm->def_flags = 0;
1047         }
1048
1049         if (mm_alloc_pgd(mm))
1050                 goto fail_nopgd;
1051
1052         if (init_new_context(p, mm))
1053                 goto fail_nocontext;
1054
1055         mm->user_ns = get_user_ns(user_ns);
1056         return mm;
1057
1058 fail_nocontext:
1059         mm_free_pgd(mm);
1060 fail_nopgd:
1061         free_mm(mm);
1062         return NULL;
1063 }
1064
1065 /*
1066  * Allocate and initialize an mm_struct.
1067  */
1068 struct mm_struct *mm_alloc(void)
1069 {
1070         struct mm_struct *mm;
1071
1072         mm = allocate_mm();
1073         if (!mm)
1074                 return NULL;
1075
1076         memset(mm, 0, sizeof(*mm));
1077         return mm_init(mm, current, current_user_ns());
1078 }
1079
1080 static inline void __mmput(struct mm_struct *mm)
1081 {
1082         VM_BUG_ON(atomic_read(&mm->mm_users));
1083
1084         uprobe_clear_state(mm);
1085         exit_aio(mm);
1086         ksm_exit(mm);
1087         khugepaged_exit(mm); /* must run before exit_mmap */
1088         exit_mmap(mm);
1089         mm_put_huge_zero_page(mm);
1090         set_mm_exe_file(mm, NULL);
1091         if (!list_empty(&mm->mmlist)) {
1092                 spin_lock(&mmlist_lock);
1093                 list_del(&mm->mmlist);
1094                 spin_unlock(&mmlist_lock);
1095         }
1096         if (mm->binfmt)
1097                 module_put(mm->binfmt->module);
1098         mmdrop(mm);
1099 }
1100
1101 /*
1102  * Decrement the use count and release all resources for an mm.
1103  */
1104 void mmput(struct mm_struct *mm)
1105 {
1106         might_sleep();
1107
1108         if (atomic_dec_and_test(&mm->mm_users))
1109                 __mmput(mm);
1110 }
1111 EXPORT_SYMBOL_GPL(mmput);
1112
1113 #ifdef CONFIG_MMU
1114 static void mmput_async_fn(struct work_struct *work)
1115 {
1116         struct mm_struct *mm = container_of(work, struct mm_struct,
1117                                             async_put_work);
1118
1119         __mmput(mm);
1120 }
1121
1122 void mmput_async(struct mm_struct *mm)
1123 {
1124         if (atomic_dec_and_test(&mm->mm_users)) {
1125                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1126                 schedule_work(&mm->async_put_work);
1127         }
1128 }
1129 #endif
1130
1131 /**
1132  * set_mm_exe_file - change a reference to the mm's executable file
1133  *
1134  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1135  *
1136  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1137  * invocations: in mmput() nobody alive left, in execve task is single
1138  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1139  * mm->exe_file, but does so without using set_mm_exe_file() in order
1140  * to do avoid the need for any locks.
1141  */
1142 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1143 {
1144         struct file *old_exe_file;
1145
1146         /*
1147          * It is safe to dereference the exe_file without RCU as
1148          * this function is only called if nobody else can access
1149          * this mm -- see comment above for justification.
1150          */
1151         old_exe_file = rcu_dereference_raw(mm->exe_file);
1152
1153         if (new_exe_file)
1154                 get_file(new_exe_file);
1155         rcu_assign_pointer(mm->exe_file, new_exe_file);
1156         if (old_exe_file)
1157                 fput(old_exe_file);
1158 }
1159
1160 /**
1161  * get_mm_exe_file - acquire a reference to the mm's executable file
1162  *
1163  * Returns %NULL if mm has no associated executable file.
1164  * User must release file via fput().
1165  */
1166 struct file *get_mm_exe_file(struct mm_struct *mm)
1167 {
1168         struct file *exe_file;
1169
1170         rcu_read_lock();
1171         exe_file = rcu_dereference(mm->exe_file);
1172         if (exe_file && !get_file_rcu(exe_file))
1173                 exe_file = NULL;
1174         rcu_read_unlock();
1175         return exe_file;
1176 }
1177 EXPORT_SYMBOL(get_mm_exe_file);
1178
1179 /**
1180  * get_task_exe_file - acquire a reference to the task's executable file
1181  *
1182  * Returns %NULL if task's mm (if any) has no associated executable file or
1183  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1184  * User must release file via fput().
1185  */
1186 struct file *get_task_exe_file(struct task_struct *task)
1187 {
1188         struct file *exe_file = NULL;
1189         struct mm_struct *mm;
1190
1191         task_lock(task);
1192         mm = task->mm;
1193         if (mm) {
1194                 if (!(task->flags & PF_KTHREAD))
1195                         exe_file = get_mm_exe_file(mm);
1196         }
1197         task_unlock(task);
1198         return exe_file;
1199 }
1200 EXPORT_SYMBOL(get_task_exe_file);
1201
1202 /**
1203  * get_task_mm - acquire a reference to the task's mm
1204  *
1205  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1206  * this kernel workthread has transiently adopted a user mm with use_mm,
1207  * to do its AIO) is not set and if so returns a reference to it, after
1208  * bumping up the use count.  User must release the mm via mmput()
1209  * after use.  Typically used by /proc and ptrace.
1210  */
1211 struct mm_struct *get_task_mm(struct task_struct *task)
1212 {
1213         struct mm_struct *mm;
1214
1215         task_lock(task);
1216         mm = task->mm;
1217         if (mm) {
1218                 if (task->flags & PF_KTHREAD)
1219                         mm = NULL;
1220                 else
1221                         mmget(mm);
1222         }
1223         task_unlock(task);
1224         return mm;
1225 }
1226 EXPORT_SYMBOL_GPL(get_task_mm);
1227
1228 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1229 {
1230         struct mm_struct *mm;
1231         int err;
1232
1233         err =  down_read_killable(&task->signal->exec_update_lock);
1234         if (err)
1235                 return ERR_PTR(err);
1236
1237         mm = get_task_mm(task);
1238         if (mm && mm != current->mm &&
1239                         !ptrace_may_access(task, mode)) {
1240                 mmput(mm);
1241                 mm = ERR_PTR(-EACCES);
1242         }
1243         up_read(&task->signal->exec_update_lock);
1244
1245         return mm;
1246 }
1247
1248 static void complete_vfork_done(struct task_struct *tsk)
1249 {
1250         struct completion *vfork;
1251
1252         task_lock(tsk);
1253         vfork = tsk->vfork_done;
1254         if (likely(vfork)) {
1255                 tsk->vfork_done = NULL;
1256                 complete(vfork);
1257         }
1258         task_unlock(tsk);
1259 }
1260
1261 static int wait_for_vfork_done(struct task_struct *child,
1262                                 struct completion *vfork)
1263 {
1264         int killed;
1265
1266         freezer_do_not_count();
1267         cgroup_enter_frozen();
1268         killed = wait_for_completion_killable(vfork);
1269         cgroup_leave_frozen(false);
1270         freezer_count();
1271
1272         if (killed) {
1273                 task_lock(child);
1274                 child->vfork_done = NULL;
1275                 task_unlock(child);
1276         }
1277
1278         put_task_struct(child);
1279         return killed;
1280 }
1281
1282 /* Please note the differences between mmput and mm_release.
1283  * mmput is called whenever we stop holding onto a mm_struct,
1284  * error success whatever.
1285  *
1286  * mm_release is called after a mm_struct has been removed
1287  * from the current process.
1288  *
1289  * This difference is important for error handling, when we
1290  * only half set up a mm_struct for a new process and need to restore
1291  * the old one.  Because we mmput the new mm_struct before
1292  * restoring the old one. . .
1293  * Eric Biederman 10 January 1998
1294  */
1295 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1296 {
1297         uprobe_free_utask(tsk);
1298
1299         /* Get rid of any cached register state */
1300         deactivate_mm(tsk, mm);
1301
1302         /*
1303          * Signal userspace if we're not exiting with a core dump
1304          * because we want to leave the value intact for debugging
1305          * purposes.
1306          */
1307         if (tsk->clear_child_tid) {
1308                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1309                     atomic_read(&mm->mm_users) > 1) {
1310                         /*
1311                          * We don't check the error code - if userspace has
1312                          * not set up a proper pointer then tough luck.
1313                          */
1314                         put_user(0, tsk->clear_child_tid);
1315                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1316                                         1, NULL, NULL, 0, 0);
1317                 }
1318                 tsk->clear_child_tid = NULL;
1319         }
1320
1321         /*
1322          * All done, finally we can wake up parent and return this mm to him.
1323          * Also kthread_stop() uses this completion for synchronization.
1324          */
1325         if (tsk->vfork_done)
1326                 complete_vfork_done(tsk);
1327 }
1328
1329 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1330 {
1331         futex_exit_release(tsk);
1332         mm_release(tsk, mm);
1333 }
1334
1335 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1336 {
1337         futex_exec_release(tsk);
1338         mm_release(tsk, mm);
1339 }
1340
1341 /**
1342  * dup_mm() - duplicates an existing mm structure
1343  * @tsk: the task_struct with which the new mm will be associated.
1344  * @oldmm: the mm to duplicate.
1345  *
1346  * Allocates a new mm structure and duplicates the provided @oldmm structure
1347  * content into it.
1348  *
1349  * Return: the duplicated mm or NULL on failure.
1350  */
1351 static struct mm_struct *dup_mm(struct task_struct *tsk,
1352                                 struct mm_struct *oldmm)
1353 {
1354         struct mm_struct *mm;
1355         int err;
1356
1357         mm = allocate_mm();
1358         if (!mm)
1359                 goto fail_nomem;
1360
1361         memcpy(mm, oldmm, sizeof(*mm));
1362
1363         if (!mm_init(mm, tsk, mm->user_ns))
1364                 goto fail_nomem;
1365
1366         err = dup_mmap(mm, oldmm);
1367         if (err)
1368                 goto free_pt;
1369
1370         mm->hiwater_rss = get_mm_rss(mm);
1371         mm->hiwater_vm = mm->total_vm;
1372
1373         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1374                 goto free_pt;
1375
1376         return mm;
1377
1378 free_pt:
1379         /* don't put binfmt in mmput, we haven't got module yet */
1380         mm->binfmt = NULL;
1381         mm_init_owner(mm, NULL);
1382         mmput(mm);
1383
1384 fail_nomem:
1385         return NULL;
1386 }
1387
1388 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1389 {
1390         struct mm_struct *mm, *oldmm;
1391         int retval;
1392
1393         tsk->min_flt = tsk->maj_flt = 0;
1394         tsk->nvcsw = tsk->nivcsw = 0;
1395 #ifdef CONFIG_DETECT_HUNG_TASK
1396         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1397         tsk->last_switch_time = 0;
1398 #endif
1399
1400         tsk->mm = NULL;
1401         tsk->active_mm = NULL;
1402
1403         /*
1404          * Are we cloning a kernel thread?
1405          *
1406          * We need to steal a active VM for that..
1407          */
1408         oldmm = current->mm;
1409         if (!oldmm)
1410                 return 0;
1411
1412         /* initialize the new vmacache entries */
1413         vmacache_flush(tsk);
1414
1415         if (clone_flags & CLONE_VM) {
1416                 mmget(oldmm);
1417                 mm = oldmm;
1418                 goto good_mm;
1419         }
1420
1421         retval = -ENOMEM;
1422         mm = dup_mm(tsk, current->mm);
1423         if (!mm)
1424                 goto fail_nomem;
1425
1426 good_mm:
1427         tsk->mm = mm;
1428         tsk->active_mm = mm;
1429         return 0;
1430
1431 fail_nomem:
1432         return retval;
1433 }
1434
1435 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1436 {
1437         struct fs_struct *fs = current->fs;
1438         if (clone_flags & CLONE_FS) {
1439                 /* tsk->fs is already what we want */
1440                 spin_lock(&fs->lock);
1441                 if (fs->in_exec) {
1442                         spin_unlock(&fs->lock);
1443                         return -EAGAIN;
1444                 }
1445                 fs->users++;
1446                 spin_unlock(&fs->lock);
1447                 return 0;
1448         }
1449         tsk->fs = copy_fs_struct(fs);
1450         if (!tsk->fs)
1451                 return -ENOMEM;
1452         return 0;
1453 }
1454
1455 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1456 {
1457         struct files_struct *oldf, *newf;
1458         int error = 0;
1459
1460         /*
1461          * A background process may not have any files ...
1462          */
1463         oldf = current->files;
1464         if (!oldf)
1465                 goto out;
1466
1467         if (clone_flags & CLONE_FILES) {
1468                 atomic_inc(&oldf->count);
1469                 goto out;
1470         }
1471
1472         newf = dup_fd(oldf, &error);
1473         if (!newf)
1474                 goto out;
1475
1476         tsk->files = newf;
1477         error = 0;
1478 out:
1479         return error;
1480 }
1481
1482 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1483 {
1484 #ifdef CONFIG_BLOCK
1485         struct io_context *ioc = current->io_context;
1486         struct io_context *new_ioc;
1487
1488         if (!ioc)
1489                 return 0;
1490         /*
1491          * Share io context with parent, if CLONE_IO is set
1492          */
1493         if (clone_flags & CLONE_IO) {
1494                 ioc_task_link(ioc);
1495                 tsk->io_context = ioc;
1496         } else if (ioprio_valid(ioc->ioprio)) {
1497                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1498                 if (unlikely(!new_ioc))
1499                         return -ENOMEM;
1500
1501                 new_ioc->ioprio = ioc->ioprio;
1502                 put_io_context(new_ioc);
1503         }
1504 #endif
1505         return 0;
1506 }
1507
1508 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1509 {
1510         struct sighand_struct *sig;
1511
1512         if (clone_flags & CLONE_SIGHAND) {
1513                 refcount_inc(&current->sighand->count);
1514                 return 0;
1515         }
1516         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1517         rcu_assign_pointer(tsk->sighand, sig);
1518         if (!sig)
1519                 return -ENOMEM;
1520
1521         refcount_set(&sig->count, 1);
1522         spin_lock_irq(&current->sighand->siglock);
1523         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1524         spin_unlock_irq(&current->sighand->siglock);
1525         return 0;
1526 }
1527
1528 void __cleanup_sighand(struct sighand_struct *sighand)
1529 {
1530         if (refcount_dec_and_test(&sighand->count)) {
1531                 signalfd_cleanup(sighand);
1532                 /*
1533                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1534                  * without an RCU grace period, see __lock_task_sighand().
1535                  */
1536                 kmem_cache_free(sighand_cachep, sighand);
1537         }
1538 }
1539
1540 /*
1541  * Initialize POSIX timer handling for a thread group.
1542  */
1543 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1544 {
1545         struct posix_cputimers *pct = &sig->posix_cputimers;
1546         unsigned long cpu_limit;
1547
1548         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1549         posix_cputimers_group_init(pct, cpu_limit);
1550 }
1551
1552 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1553 {
1554         struct signal_struct *sig;
1555
1556         if (clone_flags & CLONE_THREAD)
1557                 return 0;
1558
1559         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1560         tsk->signal = sig;
1561         if (!sig)
1562                 return -ENOMEM;
1563
1564         sig->nr_threads = 1;
1565         atomic_set(&sig->live, 1);
1566         refcount_set(&sig->sigcnt, 1);
1567
1568         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1569         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1570         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1571
1572         init_waitqueue_head(&sig->wait_chldexit);
1573         sig->curr_target = tsk;
1574         init_sigpending(&sig->shared_pending);
1575         INIT_HLIST_HEAD(&sig->multiprocess);
1576         seqlock_init(&sig->stats_lock);
1577         prev_cputime_init(&sig->prev_cputime);
1578
1579 #ifdef CONFIG_POSIX_TIMERS
1580         INIT_LIST_HEAD(&sig->posix_timers);
1581         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1582         sig->real_timer.function = it_real_fn;
1583 #endif
1584
1585         task_lock(current->group_leader);
1586         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1587         task_unlock(current->group_leader);
1588
1589         posix_cpu_timers_init_group(sig);
1590
1591         tty_audit_fork(sig);
1592         sched_autogroup_fork(sig);
1593
1594         sig->oom_score_adj = current->signal->oom_score_adj;
1595         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1596
1597         mutex_init(&sig->cred_guard_mutex);
1598         init_rwsem(&sig->exec_update_lock);
1599
1600         return 0;
1601 }
1602
1603 static void copy_seccomp(struct task_struct *p)
1604 {
1605 #ifdef CONFIG_SECCOMP
1606         /*
1607          * Must be called with sighand->lock held, which is common to
1608          * all threads in the group. Holding cred_guard_mutex is not
1609          * needed because this new task is not yet running and cannot
1610          * be racing exec.
1611          */
1612         assert_spin_locked(&current->sighand->siglock);
1613
1614         /* Ref-count the new filter user, and assign it. */
1615         get_seccomp_filter(current);
1616         p->seccomp = current->seccomp;
1617
1618         /*
1619          * Explicitly enable no_new_privs here in case it got set
1620          * between the task_struct being duplicated and holding the
1621          * sighand lock. The seccomp state and nnp must be in sync.
1622          */
1623         if (task_no_new_privs(current))
1624                 task_set_no_new_privs(p);
1625
1626         /*
1627          * If the parent gained a seccomp mode after copying thread
1628          * flags and between before we held the sighand lock, we have
1629          * to manually enable the seccomp thread flag here.
1630          */
1631         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1632                 set_tsk_thread_flag(p, TIF_SECCOMP);
1633 #endif
1634 }
1635
1636 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1637 {
1638         current->clear_child_tid = tidptr;
1639
1640         return task_pid_vnr(current);
1641 }
1642
1643 static void rt_mutex_init_task(struct task_struct *p)
1644 {
1645         raw_spin_lock_init(&p->pi_lock);
1646 #ifdef CONFIG_RT_MUTEXES
1647         p->pi_waiters = RB_ROOT_CACHED;
1648         p->pi_top_task = NULL;
1649         p->pi_blocked_on = NULL;
1650 #endif
1651 }
1652
1653 static inline void init_task_pid_links(struct task_struct *task)
1654 {
1655         enum pid_type type;
1656
1657         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1658                 INIT_HLIST_NODE(&task->pid_links[type]);
1659         }
1660 }
1661
1662 static inline void
1663 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1664 {
1665         if (type == PIDTYPE_PID)
1666                 task->thread_pid = pid;
1667         else
1668                 task->signal->pids[type] = pid;
1669 }
1670
1671 static inline void rcu_copy_process(struct task_struct *p)
1672 {
1673 #ifdef CONFIG_PREEMPT_RCU
1674         p->rcu_read_lock_nesting = 0;
1675         p->rcu_read_unlock_special.s = 0;
1676         p->rcu_blocked_node = NULL;
1677         INIT_LIST_HEAD(&p->rcu_node_entry);
1678 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1679 #ifdef CONFIG_TASKS_RCU
1680         p->rcu_tasks_holdout = false;
1681         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1682         p->rcu_tasks_idle_cpu = -1;
1683 #endif /* #ifdef CONFIG_TASKS_RCU */
1684 }
1685
1686 struct pid *pidfd_pid(const struct file *file)
1687 {
1688         if (file->f_op == &pidfd_fops)
1689                 return file->private_data;
1690
1691         return ERR_PTR(-EBADF);
1692 }
1693
1694 static int pidfd_release(struct inode *inode, struct file *file)
1695 {
1696         struct pid *pid = file->private_data;
1697
1698         file->private_data = NULL;
1699         put_pid(pid);
1700         return 0;
1701 }
1702
1703 #ifdef CONFIG_PROC_FS
1704 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1705 {
1706         struct pid_namespace *ns = proc_pid_ns(file_inode(m->file));
1707         struct pid *pid = f->private_data;
1708
1709         seq_put_decimal_ull(m, "Pid:\t", pid_nr_ns(pid, ns));
1710         seq_putc(m, '\n');
1711 }
1712 #endif
1713
1714 /*
1715  * Poll support for process exit notification.
1716  */
1717 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1718 {
1719         struct task_struct *task;
1720         struct pid *pid = file->private_data;
1721         __poll_t poll_flags = 0;
1722
1723         poll_wait(file, &pid->wait_pidfd, pts);
1724
1725         rcu_read_lock();
1726         task = pid_task(pid, PIDTYPE_PID);
1727         /*
1728          * Inform pollers only when the whole thread group exits.
1729          * If the thread group leader exits before all other threads in the
1730          * group, then poll(2) should block, similar to the wait(2) family.
1731          */
1732         if (!task || (task->exit_state && thread_group_empty(task)))
1733                 poll_flags = EPOLLIN | EPOLLRDNORM;
1734         rcu_read_unlock();
1735
1736         return poll_flags;
1737 }
1738
1739 const struct file_operations pidfd_fops = {
1740         .release = pidfd_release,
1741         .poll = pidfd_poll,
1742 #ifdef CONFIG_PROC_FS
1743         .show_fdinfo = pidfd_show_fdinfo,
1744 #endif
1745 };
1746
1747 static void __delayed_free_task(struct rcu_head *rhp)
1748 {
1749         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1750
1751         free_task(tsk);
1752 }
1753
1754 static __always_inline void delayed_free_task(struct task_struct *tsk)
1755 {
1756         if (IS_ENABLED(CONFIG_MEMCG))
1757                 call_rcu(&tsk->rcu, __delayed_free_task);
1758         else
1759                 free_task(tsk);
1760 }
1761
1762 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1763 {
1764         /* Skip if kernel thread */
1765         if (!tsk->mm)
1766                 return;
1767
1768         /* Skip if spawning a thread or using vfork */
1769         if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1770                 return;
1771
1772         /* We need to synchronize with __set_oom_adj */
1773         mutex_lock(&oom_adj_mutex);
1774         set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1775         /* Update the values in case they were changed after copy_signal */
1776         tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1777         tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1778         mutex_unlock(&oom_adj_mutex);
1779 }
1780
1781 /*
1782  * This creates a new process as a copy of the old one,
1783  * but does not actually start it yet.
1784  *
1785  * It copies the registers, and all the appropriate
1786  * parts of the process environment (as per the clone
1787  * flags). The actual kick-off is left to the caller.
1788  */
1789 static __latent_entropy struct task_struct *copy_process(
1790                                         struct pid *pid,
1791                                         int trace,
1792                                         int node,
1793                                         struct kernel_clone_args *args)
1794 {
1795         int pidfd = -1, retval;
1796         struct task_struct *p;
1797         struct multiprocess_signals delayed;
1798         struct file *pidfile = NULL;
1799         u64 clone_flags = args->flags;
1800
1801         /*
1802          * Don't allow sharing the root directory with processes in a different
1803          * namespace
1804          */
1805         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1806                 return ERR_PTR(-EINVAL);
1807
1808         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1809                 return ERR_PTR(-EINVAL);
1810
1811         /*
1812          * Thread groups must share signals as well, and detached threads
1813          * can only be started up within the thread group.
1814          */
1815         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1816                 return ERR_PTR(-EINVAL);
1817
1818         /*
1819          * Shared signal handlers imply shared VM. By way of the above,
1820          * thread groups also imply shared VM. Blocking this case allows
1821          * for various simplifications in other code.
1822          */
1823         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1824                 return ERR_PTR(-EINVAL);
1825
1826         /*
1827          * Siblings of global init remain as zombies on exit since they are
1828          * not reaped by their parent (swapper). To solve this and to avoid
1829          * multi-rooted process trees, prevent global and container-inits
1830          * from creating siblings.
1831          */
1832         if ((clone_flags & CLONE_PARENT) &&
1833                                 current->signal->flags & SIGNAL_UNKILLABLE)
1834                 return ERR_PTR(-EINVAL);
1835
1836         /*
1837          * If the new process will be in a different pid or user namespace
1838          * do not allow it to share a thread group with the forking task.
1839          */
1840         if (clone_flags & CLONE_THREAD) {
1841                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1842                     (task_active_pid_ns(current) !=
1843                                 current->nsproxy->pid_ns_for_children))
1844                         return ERR_PTR(-EINVAL);
1845         }
1846
1847         if (clone_flags & CLONE_PIDFD) {
1848                 /*
1849                  * - CLONE_DETACHED is blocked so that we can potentially
1850                  *   reuse it later for CLONE_PIDFD.
1851                  * - CLONE_THREAD is blocked until someone really needs it.
1852                  */
1853                 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1854                         return ERR_PTR(-EINVAL);
1855         }
1856
1857         /*
1858          * Force any signals received before this point to be delivered
1859          * before the fork happens.  Collect up signals sent to multiple
1860          * processes that happen during the fork and delay them so that
1861          * they appear to happen after the fork.
1862          */
1863         sigemptyset(&delayed.signal);
1864         INIT_HLIST_NODE(&delayed.node);
1865
1866         spin_lock_irq(&current->sighand->siglock);
1867         if (!(clone_flags & CLONE_THREAD))
1868                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1869         recalc_sigpending();
1870         spin_unlock_irq(&current->sighand->siglock);
1871         retval = -ERESTARTNOINTR;
1872         if (signal_pending(current))
1873                 goto fork_out;
1874
1875         retval = -ENOMEM;
1876         p = dup_task_struct(current, node);
1877         if (!p)
1878                 goto fork_out;
1879
1880         /*
1881          * This _must_ happen before we call free_task(), i.e. before we jump
1882          * to any of the bad_fork_* labels. This is to avoid freeing
1883          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1884          * kernel threads (PF_KTHREAD).
1885          */
1886         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1887         /*
1888          * Clear TID on mm_release()?
1889          */
1890         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1891
1892         ftrace_graph_init_task(p);
1893
1894         rt_mutex_init_task(p);
1895
1896 #ifdef CONFIG_PROVE_LOCKING
1897         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1898         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1899 #endif
1900         retval = -EAGAIN;
1901         if (atomic_read(&p->real_cred->user->processes) >=
1902                         task_rlimit(p, RLIMIT_NPROC)) {
1903                 if (p->real_cred->user != INIT_USER &&
1904                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1905                         goto bad_fork_free;
1906         }
1907         current->flags &= ~PF_NPROC_EXCEEDED;
1908
1909         retval = copy_creds(p, clone_flags);
1910         if (retval < 0)
1911                 goto bad_fork_free;
1912
1913         /*
1914          * If multiple threads are within copy_process(), then this check
1915          * triggers too late. This doesn't hurt, the check is only there
1916          * to stop root fork bombs.
1917          */
1918         retval = -EAGAIN;
1919         if (nr_threads >= max_threads)
1920                 goto bad_fork_cleanup_count;
1921
1922         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1923         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1924         p->flags |= PF_FORKNOEXEC;
1925         INIT_LIST_HEAD(&p->children);
1926         INIT_LIST_HEAD(&p->sibling);
1927         rcu_copy_process(p);
1928         p->vfork_done = NULL;
1929         spin_lock_init(&p->alloc_lock);
1930
1931         init_sigpending(&p->pending);
1932
1933         p->utime = p->stime = p->gtime = 0;
1934 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1935         p->utimescaled = p->stimescaled = 0;
1936 #endif
1937         prev_cputime_init(&p->prev_cputime);
1938
1939 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1940         seqcount_init(&p->vtime.seqcount);
1941         p->vtime.starttime = 0;
1942         p->vtime.state = VTIME_INACTIVE;
1943 #endif
1944
1945 #if defined(SPLIT_RSS_COUNTING)
1946         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1947 #endif
1948
1949         p->default_timer_slack_ns = current->timer_slack_ns;
1950
1951 #ifdef CONFIG_PSI
1952         p->psi_flags = 0;
1953 #endif
1954
1955         task_io_accounting_init(&p->ioac);
1956         acct_clear_integrals(p);
1957
1958         posix_cputimers_init(&p->posix_cputimers);
1959
1960         p->io_context = NULL;
1961         audit_set_context(p, NULL);
1962         cgroup_fork(p);
1963 #ifdef CONFIG_NUMA
1964         p->mempolicy = mpol_dup(p->mempolicy);
1965         if (IS_ERR(p->mempolicy)) {
1966                 retval = PTR_ERR(p->mempolicy);
1967                 p->mempolicy = NULL;
1968                 goto bad_fork_cleanup_threadgroup_lock;
1969         }
1970 #endif
1971 #ifdef CONFIG_CPUSETS
1972         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1973         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1974         seqcount_init(&p->mems_allowed_seq);
1975 #endif
1976 #ifdef CONFIG_TRACE_IRQFLAGS
1977         p->irq_events = 0;
1978         p->hardirqs_enabled = 0;
1979         p->hardirq_enable_ip = 0;
1980         p->hardirq_enable_event = 0;
1981         p->hardirq_disable_ip = _THIS_IP_;
1982         p->hardirq_disable_event = 0;
1983         p->softirqs_enabled = 1;
1984         p->softirq_enable_ip = _THIS_IP_;
1985         p->softirq_enable_event = 0;
1986         p->softirq_disable_ip = 0;
1987         p->softirq_disable_event = 0;
1988         p->hardirq_context = 0;
1989         p->softirq_context = 0;
1990 #endif
1991
1992         p->pagefault_disabled = 0;
1993
1994 #ifdef CONFIG_LOCKDEP
1995         lockdep_init_task(p);
1996 #endif
1997
1998 #ifdef CONFIG_DEBUG_MUTEXES
1999         p->blocked_on = NULL; /* not blocked yet */
2000 #endif
2001 #ifdef CONFIG_BCACHE
2002         p->sequential_io        = 0;
2003         p->sequential_io_avg    = 0;
2004 #endif
2005
2006         /* Perform scheduler related setup. Assign this task to a CPU. */
2007         retval = sched_fork(clone_flags, p);
2008         if (retval)
2009                 goto bad_fork_cleanup_policy;
2010
2011         retval = perf_event_init_task(p);
2012         if (retval)
2013                 goto bad_fork_cleanup_policy;
2014         retval = audit_alloc(p);
2015         if (retval)
2016                 goto bad_fork_cleanup_perf;
2017         /* copy all the process information */
2018         shm_init_task(p);
2019         retval = security_task_alloc(p, clone_flags);
2020         if (retval)
2021                 goto bad_fork_cleanup_audit;
2022         retval = copy_semundo(clone_flags, p);
2023         if (retval)
2024                 goto bad_fork_cleanup_security;
2025         retval = copy_files(clone_flags, p);
2026         if (retval)
2027                 goto bad_fork_cleanup_semundo;
2028         retval = copy_fs(clone_flags, p);
2029         if (retval)
2030                 goto bad_fork_cleanup_files;
2031         retval = copy_sighand(clone_flags, p);
2032         if (retval)
2033                 goto bad_fork_cleanup_fs;
2034         retval = copy_signal(clone_flags, p);
2035         if (retval)
2036                 goto bad_fork_cleanup_sighand;
2037         retval = copy_mm(clone_flags, p);
2038         if (retval)
2039                 goto bad_fork_cleanup_signal;
2040         retval = copy_namespaces(clone_flags, p);
2041         if (retval)
2042                 goto bad_fork_cleanup_mm;
2043         retval = copy_io(clone_flags, p);
2044         if (retval)
2045                 goto bad_fork_cleanup_namespaces;
2046         retval = copy_thread_tls(clone_flags, args->stack, args->stack_size, p,
2047                                  args->tls);
2048         if (retval)
2049                 goto bad_fork_cleanup_io;
2050
2051         stackleak_task_init(p);
2052
2053         if (pid != &init_struct_pid) {
2054                 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
2055                 if (IS_ERR(pid)) {
2056                         retval = PTR_ERR(pid);
2057                         goto bad_fork_cleanup_thread;
2058                 }
2059         }
2060
2061         /*
2062          * This has to happen after we've potentially unshared the file
2063          * descriptor table (so that the pidfd doesn't leak into the child
2064          * if the fd table isn't shared).
2065          */
2066         if (clone_flags & CLONE_PIDFD) {
2067                 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2068                 if (retval < 0)
2069                         goto bad_fork_free_pid;
2070
2071                 pidfd = retval;
2072
2073                 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2074                                               O_RDWR | O_CLOEXEC);
2075                 if (IS_ERR(pidfile)) {
2076                         put_unused_fd(pidfd);
2077                         retval = PTR_ERR(pidfile);
2078                         goto bad_fork_free_pid;
2079                 }
2080                 get_pid(pid);   /* held by pidfile now */
2081
2082                 retval = put_user(pidfd, args->pidfd);
2083                 if (retval)
2084                         goto bad_fork_put_pidfd;
2085         }
2086
2087 #ifdef CONFIG_BLOCK
2088         p->plug = NULL;
2089 #endif
2090         futex_init_task(p);
2091
2092         /*
2093          * sigaltstack should be cleared when sharing the same VM
2094          */
2095         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2096                 sas_ss_reset(p);
2097
2098         /*
2099          * Syscall tracing and stepping should be turned off in the
2100          * child regardless of CLONE_PTRACE.
2101          */
2102         user_disable_single_step(p);
2103         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
2104 #ifdef TIF_SYSCALL_EMU
2105         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
2106 #endif
2107         clear_tsk_latency_tracing(p);
2108
2109         /* ok, now we should be set up.. */
2110         p->pid = pid_nr(pid);
2111         if (clone_flags & CLONE_THREAD) {
2112                 p->group_leader = current->group_leader;
2113                 p->tgid = current->tgid;
2114         } else {
2115                 p->group_leader = p;
2116                 p->tgid = p->pid;
2117         }
2118
2119         p->nr_dirtied = 0;
2120         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2121         p->dirty_paused_when = 0;
2122
2123         p->pdeath_signal = 0;
2124         INIT_LIST_HEAD(&p->thread_group);
2125         p->task_works = NULL;
2126
2127         cgroup_threadgroup_change_begin(current);
2128         /*
2129          * Ensure that the cgroup subsystem policies allow the new process to be
2130          * forked. It should be noted the the new process's css_set can be changed
2131          * between here and cgroup_post_fork() if an organisation operation is in
2132          * progress.
2133          */
2134         retval = cgroup_can_fork(p);
2135         if (retval)
2136                 goto bad_fork_cgroup_threadgroup_change_end;
2137
2138         /*
2139          * From this point on we must avoid any synchronous user-space
2140          * communication until we take the tasklist-lock. In particular, we do
2141          * not want user-space to be able to predict the process start-time by
2142          * stalling fork(2) after we recorded the start_time but before it is
2143          * visible to the system.
2144          */
2145
2146         p->start_time = ktime_get_ns();
2147         p->real_start_time = ktime_get_boottime_ns();
2148
2149         /*
2150          * Make it visible to the rest of the system, but dont wake it up yet.
2151          * Need tasklist lock for parent etc handling!
2152          */
2153         write_lock_irq(&tasklist_lock);
2154
2155         /* CLONE_PARENT re-uses the old parent */
2156         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2157                 p->real_parent = current->real_parent;
2158                 p->parent_exec_id = current->parent_exec_id;
2159                 if (clone_flags & CLONE_THREAD)
2160                         p->exit_signal = -1;
2161                 else
2162                         p->exit_signal = current->group_leader->exit_signal;
2163         } else {
2164                 p->real_parent = current;
2165                 p->parent_exec_id = current->self_exec_id;
2166                 p->exit_signal = args->exit_signal;
2167         }
2168
2169         klp_copy_process(p);
2170
2171         spin_lock(&current->sighand->siglock);
2172
2173         /*
2174          * Copy seccomp details explicitly here, in case they were changed
2175          * before holding sighand lock.
2176          */
2177         copy_seccomp(p);
2178
2179         rseq_fork(p, clone_flags);
2180
2181         /* Don't start children in a dying pid namespace */
2182         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2183                 retval = -ENOMEM;
2184                 goto bad_fork_cancel_cgroup;
2185         }
2186
2187         /* Let kill terminate clone/fork in the middle */
2188         if (fatal_signal_pending(current)) {
2189                 retval = -EINTR;
2190                 goto bad_fork_cancel_cgroup;
2191         }
2192
2193         init_task_pid_links(p);
2194         if (likely(p->pid)) {
2195                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2196
2197                 init_task_pid(p, PIDTYPE_PID, pid);
2198                 if (thread_group_leader(p)) {
2199                         init_task_pid(p, PIDTYPE_TGID, pid);
2200                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2201                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2202
2203                         if (is_child_reaper(pid)) {
2204                                 ns_of_pid(pid)->child_reaper = p;
2205                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2206                         }
2207                         p->signal->shared_pending.signal = delayed.signal;
2208                         p->signal->tty = tty_kref_get(current->signal->tty);
2209                         /*
2210                          * Inherit has_child_subreaper flag under the same
2211                          * tasklist_lock with adding child to the process tree
2212                          * for propagate_has_child_subreaper optimization.
2213                          */
2214                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2215                                                          p->real_parent->signal->is_child_subreaper;
2216                         list_add_tail(&p->sibling, &p->real_parent->children);
2217                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2218                         attach_pid(p, PIDTYPE_TGID);
2219                         attach_pid(p, PIDTYPE_PGID);
2220                         attach_pid(p, PIDTYPE_SID);
2221                         __this_cpu_inc(process_counts);
2222                 } else {
2223                         current->signal->nr_threads++;
2224                         atomic_inc(&current->signal->live);
2225                         refcount_inc(&current->signal->sigcnt);
2226                         task_join_group_stop(p);
2227                         list_add_tail_rcu(&p->thread_group,
2228                                           &p->group_leader->thread_group);
2229                         list_add_tail_rcu(&p->thread_node,
2230                                           &p->signal->thread_head);
2231                 }
2232                 attach_pid(p, PIDTYPE_PID);
2233                 nr_threads++;
2234         }
2235         total_forks++;
2236         hlist_del_init(&delayed.node);
2237         spin_unlock(&current->sighand->siglock);
2238         syscall_tracepoint_update(p);
2239         write_unlock_irq(&tasklist_lock);
2240
2241         if (pidfile)
2242                 fd_install(pidfd, pidfile);
2243
2244         proc_fork_connector(p);
2245         cgroup_post_fork(p);
2246         cgroup_threadgroup_change_end(current);
2247         perf_event_fork(p);
2248
2249         trace_task_newtask(p, clone_flags);
2250         uprobe_copy_process(p, clone_flags);
2251
2252         copy_oom_score_adj(clone_flags, p);
2253
2254         return p;
2255
2256 bad_fork_cancel_cgroup:
2257         spin_unlock(&current->sighand->siglock);
2258         write_unlock_irq(&tasklist_lock);
2259         cgroup_cancel_fork(p);
2260 bad_fork_cgroup_threadgroup_change_end:
2261         cgroup_threadgroup_change_end(current);
2262 bad_fork_put_pidfd:
2263         if (clone_flags & CLONE_PIDFD) {
2264                 fput(pidfile);
2265                 put_unused_fd(pidfd);
2266         }
2267 bad_fork_free_pid:
2268         if (pid != &init_struct_pid)
2269                 free_pid(pid);
2270 bad_fork_cleanup_thread:
2271         exit_thread(p);
2272 bad_fork_cleanup_io:
2273         if (p->io_context)
2274                 exit_io_context(p);
2275 bad_fork_cleanup_namespaces:
2276         exit_task_namespaces(p);
2277 bad_fork_cleanup_mm:
2278         if (p->mm) {
2279                 mm_clear_owner(p->mm, p);
2280                 mmput(p->mm);
2281         }
2282 bad_fork_cleanup_signal:
2283         if (!(clone_flags & CLONE_THREAD))
2284                 free_signal_struct(p->signal);
2285 bad_fork_cleanup_sighand:
2286         __cleanup_sighand(p->sighand);
2287 bad_fork_cleanup_fs:
2288         exit_fs(p); /* blocking */
2289 bad_fork_cleanup_files:
2290         exit_files(p); /* blocking */
2291 bad_fork_cleanup_semundo:
2292         exit_sem(p);
2293 bad_fork_cleanup_security:
2294         security_task_free(p);
2295 bad_fork_cleanup_audit:
2296         audit_free(p);
2297 bad_fork_cleanup_perf:
2298         perf_event_free_task(p);
2299 bad_fork_cleanup_policy:
2300         lockdep_free_task(p);
2301 #ifdef CONFIG_NUMA
2302         mpol_put(p->mempolicy);
2303 bad_fork_cleanup_threadgroup_lock:
2304 #endif
2305         delayacct_tsk_free(p);
2306 bad_fork_cleanup_count:
2307         atomic_dec(&p->cred->user->processes);
2308         exit_creds(p);
2309 bad_fork_free:
2310         p->state = TASK_DEAD;
2311         put_task_stack(p);
2312         delayed_free_task(p);
2313 fork_out:
2314         spin_lock_irq(&current->sighand->siglock);
2315         hlist_del_init(&delayed.node);
2316         spin_unlock_irq(&current->sighand->siglock);
2317         return ERR_PTR(retval);
2318 }
2319
2320 static inline void init_idle_pids(struct task_struct *idle)
2321 {
2322         enum pid_type type;
2323
2324         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2325                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2326                 init_task_pid(idle, type, &init_struct_pid);
2327         }
2328 }
2329
2330 struct task_struct *fork_idle(int cpu)
2331 {
2332         struct task_struct *task;
2333         struct kernel_clone_args args = {
2334                 .flags = CLONE_VM,
2335         };
2336
2337         task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2338         if (!IS_ERR(task)) {
2339                 init_idle_pids(task);
2340                 init_idle(task, cpu);
2341         }
2342
2343         return task;
2344 }
2345
2346 /*
2347  *  Ok, this is the main fork-routine.
2348  *
2349  * It copies the process, and if successful kick-starts
2350  * it and waits for it to finish using the VM if required.
2351  *
2352  * args->exit_signal is expected to be checked for sanity by the caller.
2353  */
2354 long _do_fork(struct kernel_clone_args *args)
2355 {
2356         u64 clone_flags = args->flags;
2357         struct completion vfork;
2358         struct pid *pid;
2359         struct task_struct *p;
2360         int trace = 0;
2361         long nr;
2362
2363         /*
2364          * Determine whether and which event to report to ptracer.  When
2365          * called from kernel_thread or CLONE_UNTRACED is explicitly
2366          * requested, no event is reported; otherwise, report if the event
2367          * for the type of forking is enabled.
2368          */
2369         if (!(clone_flags & CLONE_UNTRACED)) {
2370                 if (clone_flags & CLONE_VFORK)
2371                         trace = PTRACE_EVENT_VFORK;
2372                 else if (args->exit_signal != SIGCHLD)
2373                         trace = PTRACE_EVENT_CLONE;
2374                 else
2375                         trace = PTRACE_EVENT_FORK;
2376
2377                 if (likely(!ptrace_event_enabled(current, trace)))
2378                         trace = 0;
2379         }
2380
2381         p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2382         add_latent_entropy();
2383
2384         if (IS_ERR(p))
2385                 return PTR_ERR(p);
2386
2387         /*
2388          * Do this prior waking up the new thread - the thread pointer
2389          * might get invalid after that point, if the thread exits quickly.
2390          */
2391         trace_sched_process_fork(current, p);
2392
2393         pid = get_task_pid(p, PIDTYPE_PID);
2394         nr = pid_vnr(pid);
2395
2396         if (clone_flags & CLONE_PARENT_SETTID)
2397                 put_user(nr, args->parent_tid);
2398
2399         if (clone_flags & CLONE_VFORK) {
2400                 p->vfork_done = &vfork;
2401                 init_completion(&vfork);
2402                 get_task_struct(p);
2403         }
2404
2405         wake_up_new_task(p);
2406
2407         /* forking complete and child started to run, tell ptracer */
2408         if (unlikely(trace))
2409                 ptrace_event_pid(trace, pid);
2410
2411         if (clone_flags & CLONE_VFORK) {
2412                 if (!wait_for_vfork_done(p, &vfork))
2413                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2414         }
2415
2416         put_pid(pid);
2417         return nr;
2418 }
2419
2420 bool legacy_clone_args_valid(const struct kernel_clone_args *kargs)
2421 {
2422         /* clone(CLONE_PIDFD) uses parent_tidptr to return a pidfd */
2423         if ((kargs->flags & CLONE_PIDFD) &&
2424             (kargs->flags & CLONE_PARENT_SETTID))
2425                 return false;
2426
2427         return true;
2428 }
2429
2430 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2431 /* For compatibility with architectures that call do_fork directly rather than
2432  * using the syscall entry points below. */
2433 long do_fork(unsigned long clone_flags,
2434               unsigned long stack_start,
2435               unsigned long stack_size,
2436               int __user *parent_tidptr,
2437               int __user *child_tidptr)
2438 {
2439         struct kernel_clone_args args = {
2440                 .flags          = (lower_32_bits(clone_flags) & ~CSIGNAL),
2441                 .pidfd          = parent_tidptr,
2442                 .child_tid      = child_tidptr,
2443                 .parent_tid     = parent_tidptr,
2444                 .exit_signal    = (lower_32_bits(clone_flags) & CSIGNAL),
2445                 .stack          = stack_start,
2446                 .stack_size     = stack_size,
2447         };
2448
2449         if (!legacy_clone_args_valid(&args))
2450                 return -EINVAL;
2451
2452         return _do_fork(&args);
2453 }
2454 #endif
2455
2456 /*
2457  * Create a kernel thread.
2458  */
2459 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2460 {
2461         struct kernel_clone_args args = {
2462                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2463                                     CLONE_UNTRACED) & ~CSIGNAL),
2464                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2465                 .stack          = (unsigned long)fn,
2466                 .stack_size     = (unsigned long)arg,
2467         };
2468
2469         return _do_fork(&args);
2470 }
2471
2472 #ifdef __ARCH_WANT_SYS_FORK
2473 SYSCALL_DEFINE0(fork)
2474 {
2475 #ifdef CONFIG_MMU
2476         struct kernel_clone_args args = {
2477                 .exit_signal = SIGCHLD,
2478         };
2479
2480         return _do_fork(&args);
2481 #else
2482         /* can not support in nommu mode */
2483         return -EINVAL;
2484 #endif
2485 }
2486 #endif
2487
2488 #ifdef __ARCH_WANT_SYS_VFORK
2489 SYSCALL_DEFINE0(vfork)
2490 {
2491         struct kernel_clone_args args = {
2492                 .flags          = CLONE_VFORK | CLONE_VM,
2493                 .exit_signal    = SIGCHLD,
2494         };
2495
2496         return _do_fork(&args);
2497 }
2498 #endif
2499
2500 #ifdef __ARCH_WANT_SYS_CLONE
2501 #ifdef CONFIG_CLONE_BACKWARDS
2502 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2503                  int __user *, parent_tidptr,
2504                  unsigned long, tls,
2505                  int __user *, child_tidptr)
2506 #elif defined(CONFIG_CLONE_BACKWARDS2)
2507 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2508                  int __user *, parent_tidptr,
2509                  int __user *, child_tidptr,
2510                  unsigned long, tls)
2511 #elif defined(CONFIG_CLONE_BACKWARDS3)
2512 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2513                 int, stack_size,
2514                 int __user *, parent_tidptr,
2515                 int __user *, child_tidptr,
2516                 unsigned long, tls)
2517 #else
2518 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2519                  int __user *, parent_tidptr,
2520                  int __user *, child_tidptr,
2521                  unsigned long, tls)
2522 #endif
2523 {
2524         struct kernel_clone_args args = {
2525                 .flags          = (lower_32_bits(clone_flags) & ~CSIGNAL),
2526                 .pidfd          = parent_tidptr,
2527                 .child_tid      = child_tidptr,
2528                 .parent_tid     = parent_tidptr,
2529                 .exit_signal    = (lower_32_bits(clone_flags) & CSIGNAL),
2530                 .stack          = newsp,
2531                 .tls            = tls,
2532         };
2533
2534         if (!legacy_clone_args_valid(&args))
2535                 return -EINVAL;
2536
2537         return _do_fork(&args);
2538 }
2539 #endif
2540
2541 #ifdef __ARCH_WANT_SYS_CLONE3
2542
2543 /*
2544  * copy_thread implementations handle CLONE_SETTLS by reading the TLS value from
2545  * the registers containing the syscall arguments for clone. This doesn't work
2546  * with clone3 since the TLS value is passed in clone_args instead.
2547  */
2548 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2549 #error clone3 requires copy_thread_tls support in arch
2550 #endif
2551
2552 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2553                                               struct clone_args __user *uargs,
2554                                               size_t usize)
2555 {
2556         int err;
2557         struct clone_args args;
2558
2559         if (unlikely(usize > PAGE_SIZE))
2560                 return -E2BIG;
2561         if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2562                 return -EINVAL;
2563
2564         err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2565         if (err)
2566                 return err;
2567
2568         /*
2569          * Verify that higher 32bits of exit_signal are unset and that
2570          * it is a valid signal
2571          */
2572         if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2573                      !valid_signal(args.exit_signal)))
2574                 return -EINVAL;
2575
2576         *kargs = (struct kernel_clone_args){
2577                 .flags          = args.flags,
2578                 .pidfd          = u64_to_user_ptr(args.pidfd),
2579                 .child_tid      = u64_to_user_ptr(args.child_tid),
2580                 .parent_tid     = u64_to_user_ptr(args.parent_tid),
2581                 .exit_signal    = args.exit_signal,
2582                 .stack          = args.stack,
2583                 .stack_size     = args.stack_size,
2584                 .tls            = args.tls,
2585         };
2586
2587         return 0;
2588 }
2589
2590 /**
2591  * clone3_stack_valid - check and prepare stack
2592  * @kargs: kernel clone args
2593  *
2594  * Verify that the stack arguments userspace gave us are sane.
2595  * In addition, set the stack direction for userspace since it's easy for us to
2596  * determine.
2597  */
2598 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2599 {
2600         if (kargs->stack == 0) {
2601                 if (kargs->stack_size > 0)
2602                         return false;
2603         } else {
2604                 if (kargs->stack_size == 0)
2605                         return false;
2606
2607                 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2608                         return false;
2609
2610 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2611                 kargs->stack += kargs->stack_size;
2612 #endif
2613         }
2614
2615         return true;
2616 }
2617
2618 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2619 {
2620         /*
2621          * All lower bits of the flag word are taken.
2622          * Verify that no other unknown flags are passed along.
2623          */
2624         if (kargs->flags & ~CLONE_LEGACY_FLAGS)
2625                 return false;
2626
2627         /*
2628          * - make the CLONE_DETACHED bit reuseable for clone3
2629          * - make the CSIGNAL bits reuseable for clone3
2630          */
2631         if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2632                 return false;
2633
2634         if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2635             kargs->exit_signal)
2636                 return false;
2637
2638         if (!clone3_stack_valid(kargs))
2639                 return false;
2640
2641         return true;
2642 }
2643
2644 /**
2645  * clone3 - create a new process with specific properties
2646  * @uargs: argument structure
2647  * @size:  size of @uargs
2648  *
2649  * clone3() is the extensible successor to clone()/clone2().
2650  * It takes a struct as argument that is versioned by its size.
2651  *
2652  * Return: On success, a positive PID for the child process.
2653  *         On error, a negative errno number.
2654  */
2655 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2656 {
2657         int err;
2658
2659         struct kernel_clone_args kargs;
2660
2661         err = copy_clone_args_from_user(&kargs, uargs, size);
2662         if (err)
2663                 return err;
2664
2665         if (!clone3_args_valid(&kargs))
2666                 return -EINVAL;
2667
2668         return _do_fork(&kargs);
2669 }
2670 #endif
2671
2672 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2673 {
2674         struct task_struct *leader, *parent, *child;
2675         int res;
2676
2677         read_lock(&tasklist_lock);
2678         leader = top = top->group_leader;
2679 down:
2680         for_each_thread(leader, parent) {
2681                 list_for_each_entry(child, &parent->children, sibling) {
2682                         res = visitor(child, data);
2683                         if (res) {
2684                                 if (res < 0)
2685                                         goto out;
2686                                 leader = child;
2687                                 goto down;
2688                         }
2689 up:
2690                         ;
2691                 }
2692         }
2693
2694         if (leader != top) {
2695                 child = leader;
2696                 parent = child->real_parent;
2697                 leader = parent->group_leader;
2698                 goto up;
2699         }
2700 out:
2701         read_unlock(&tasklist_lock);
2702 }
2703
2704 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2705 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2706 #endif
2707
2708 static void sighand_ctor(void *data)
2709 {
2710         struct sighand_struct *sighand = data;
2711
2712         spin_lock_init(&sighand->siglock);
2713         init_waitqueue_head(&sighand->signalfd_wqh);
2714 }
2715
2716 void __init mm_cache_init(void)
2717 {
2718         unsigned int mm_size;
2719
2720         /*
2721          * The mm_cpumask is located at the end of mm_struct, and is
2722          * dynamically sized based on the maximum CPU number this system
2723          * can have, taking hotplug into account (nr_cpu_ids).
2724          */
2725         mm_size = sizeof(struct mm_struct) + cpumask_size();
2726
2727         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2728                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2729                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2730                         offsetof(struct mm_struct, saved_auxv),
2731                         sizeof_field(struct mm_struct, saved_auxv),
2732                         NULL);
2733 }
2734
2735 void __init proc_caches_init(void)
2736 {
2737         sighand_cachep = kmem_cache_create("sighand_cache",
2738                         sizeof(struct sighand_struct), 0,
2739                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2740                         SLAB_ACCOUNT, sighand_ctor);
2741         signal_cachep = kmem_cache_create("signal_cache",
2742                         sizeof(struct signal_struct), 0,
2743                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2744                         NULL);
2745         files_cachep = kmem_cache_create("files_cache",
2746                         sizeof(struct files_struct), 0,
2747                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2748                         NULL);
2749         fs_cachep = kmem_cache_create("fs_cache",
2750                         sizeof(struct fs_struct), 0,
2751                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2752                         NULL);
2753
2754         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2755         mmap_init();
2756         nsproxy_cache_init();
2757 }
2758
2759 /*
2760  * Check constraints on flags passed to the unshare system call.
2761  */
2762 static int check_unshare_flags(unsigned long unshare_flags)
2763 {
2764         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2765                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2766                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2767                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2768                 return -EINVAL;
2769         /*
2770          * Not implemented, but pretend it works if there is nothing
2771          * to unshare.  Note that unsharing the address space or the
2772          * signal handlers also need to unshare the signal queues (aka
2773          * CLONE_THREAD).
2774          */
2775         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2776                 if (!thread_group_empty(current))
2777                         return -EINVAL;
2778         }
2779         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2780                 if (refcount_read(&current->sighand->count) > 1)
2781                         return -EINVAL;
2782         }
2783         if (unshare_flags & CLONE_VM) {
2784                 if (!current_is_single_threaded())
2785                         return -EINVAL;
2786         }
2787
2788         return 0;
2789 }
2790
2791 /*
2792  * Unshare the filesystem structure if it is being shared
2793  */
2794 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2795 {
2796         struct fs_struct *fs = current->fs;
2797
2798         if (!(unshare_flags & CLONE_FS) || !fs)
2799                 return 0;
2800
2801         /* don't need lock here; in the worst case we'll do useless copy */
2802         if (fs->users == 1)
2803                 return 0;
2804
2805         *new_fsp = copy_fs_struct(fs);
2806         if (!*new_fsp)
2807                 return -ENOMEM;
2808
2809         return 0;
2810 }
2811
2812 /*
2813  * Unshare file descriptor table if it is being shared
2814  */
2815 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2816 {
2817         struct files_struct *fd = current->files;
2818         int error = 0;
2819
2820         if ((unshare_flags & CLONE_FILES) &&
2821             (fd && atomic_read(&fd->count) > 1)) {
2822                 *new_fdp = dup_fd(fd, &error);
2823                 if (!*new_fdp)
2824                         return error;
2825         }
2826
2827         return 0;
2828 }
2829
2830 /*
2831  * unshare allows a process to 'unshare' part of the process
2832  * context which was originally shared using clone.  copy_*
2833  * functions used by do_fork() cannot be used here directly
2834  * because they modify an inactive task_struct that is being
2835  * constructed. Here we are modifying the current, active,
2836  * task_struct.
2837  */
2838 int ksys_unshare(unsigned long unshare_flags)
2839 {
2840         struct fs_struct *fs, *new_fs = NULL;
2841         struct files_struct *fd, *new_fd = NULL;
2842         struct cred *new_cred = NULL;
2843         struct nsproxy *new_nsproxy = NULL;
2844         int do_sysvsem = 0;
2845         int err;
2846
2847         /*
2848          * If unsharing a user namespace must also unshare the thread group
2849          * and unshare the filesystem root and working directories.
2850          */
2851         if (unshare_flags & CLONE_NEWUSER)
2852                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2853         /*
2854          * If unsharing vm, must also unshare signal handlers.
2855          */
2856         if (unshare_flags & CLONE_VM)
2857                 unshare_flags |= CLONE_SIGHAND;
2858         /*
2859          * If unsharing a signal handlers, must also unshare the signal queues.
2860          */
2861         if (unshare_flags & CLONE_SIGHAND)
2862                 unshare_flags |= CLONE_THREAD;
2863         /*
2864          * If unsharing namespace, must also unshare filesystem information.
2865          */
2866         if (unshare_flags & CLONE_NEWNS)
2867                 unshare_flags |= CLONE_FS;
2868
2869         err = check_unshare_flags(unshare_flags);
2870         if (err)
2871                 goto bad_unshare_out;
2872         /*
2873          * CLONE_NEWIPC must also detach from the undolist: after switching
2874          * to a new ipc namespace, the semaphore arrays from the old
2875          * namespace are unreachable.
2876          */
2877         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2878                 do_sysvsem = 1;
2879         err = unshare_fs(unshare_flags, &new_fs);
2880         if (err)
2881                 goto bad_unshare_out;
2882         err = unshare_fd(unshare_flags, &new_fd);
2883         if (err)
2884                 goto bad_unshare_cleanup_fs;
2885         err = unshare_userns(unshare_flags, &new_cred);
2886         if (err)
2887                 goto bad_unshare_cleanup_fd;
2888         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2889                                          new_cred, new_fs);
2890         if (err)
2891                 goto bad_unshare_cleanup_cred;
2892
2893         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2894                 if (do_sysvsem) {
2895                         /*
2896                          * CLONE_SYSVSEM is equivalent to sys_exit().
2897                          */
2898                         exit_sem(current);
2899                 }
2900                 if (unshare_flags & CLONE_NEWIPC) {
2901                         /* Orphan segments in old ns (see sem above). */
2902                         exit_shm(current);
2903                         shm_init_task(current);
2904                 }
2905
2906                 if (new_nsproxy)
2907                         switch_task_namespaces(current, new_nsproxy);
2908
2909                 task_lock(current);
2910
2911                 if (new_fs) {
2912                         fs = current->fs;
2913                         spin_lock(&fs->lock);
2914                         current->fs = new_fs;
2915                         if (--fs->users)
2916                                 new_fs = NULL;
2917                         else
2918                                 new_fs = fs;
2919                         spin_unlock(&fs->lock);
2920                 }
2921
2922                 if (new_fd) {
2923                         fd = current->files;
2924                         current->files = new_fd;
2925                         new_fd = fd;
2926                 }
2927
2928                 task_unlock(current);
2929
2930                 if (new_cred) {
2931                         /* Install the new user namespace */
2932                         commit_creds(new_cred);
2933                         new_cred = NULL;
2934                 }
2935         }
2936
2937         perf_event_namespaces(current);
2938
2939 bad_unshare_cleanup_cred:
2940         if (new_cred)
2941                 put_cred(new_cred);
2942 bad_unshare_cleanup_fd:
2943         if (new_fd)
2944                 put_files_struct(new_fd);
2945
2946 bad_unshare_cleanup_fs:
2947         if (new_fs)
2948                 free_fs_struct(new_fs);
2949
2950 bad_unshare_out:
2951         return err;
2952 }
2953
2954 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2955 {
2956         return ksys_unshare(unshare_flags);
2957 }
2958
2959 /*
2960  *      Helper to unshare the files of the current task.
2961  *      We don't want to expose copy_files internals to
2962  *      the exec layer of the kernel.
2963  */
2964
2965 int unshare_files(struct files_struct **displaced)
2966 {
2967         struct task_struct *task = current;
2968         struct files_struct *copy = NULL;
2969         int error;
2970
2971         error = unshare_fd(CLONE_FILES, &copy);
2972         if (error || !copy) {
2973                 *displaced = NULL;
2974                 return error;
2975         }
2976         *displaced = task->files;
2977         task_lock(task);
2978         task->files = copy;
2979         task_unlock(task);
2980         return 0;
2981 }
2982
2983 int sysctl_max_threads(struct ctl_table *table, int write,
2984                        void __user *buffer, size_t *lenp, loff_t *ppos)
2985 {
2986         struct ctl_table t;
2987         int ret;
2988         int threads = max_threads;
2989         int min = 1;
2990         int max = MAX_THREADS;
2991
2992         t = *table;
2993         t.data = &threads;
2994         t.extra1 = &min;
2995         t.extra2 = &max;
2996
2997         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2998         if (ret || !write)
2999                 return ret;
3000
3001         max_threads = threads;
3002
3003         return 0;
3004 }