kernel/fork.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/kernel/fork.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  */
   7
   8 /*
   9  *  'fork.c' contains the help-routines for the 'fork' system call
  10  * (see also entry.S and others).
  11  * Fork is rather simple, once you get the hang of it, but the memory
  12  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  13  */
  14
  15 #include <linux/anon_inodes.h>
  16 #include <linux/slab.h>
  17 #include <linux/sched/autogroup.h>
  18 #include <linux/sched/mm.h>
  19 #include <linux/sched/coredump.h>
  20 #include <linux/sched/user.h>
  21 #include <linux/sched/numa_balancing.h>
  22 #include <linux/sched/stat.h>
  23 #include <linux/sched/task.h>
  24 #include <linux/sched/task_stack.h>
  25 #include <linux/sched/cputime.h>
  26 #include <linux/seq_file.h>
  27 #include <linux/rtmutex.h>
  28 #include <linux/init.h>
  29 #include <linux/unistd.h>
  30 #include <linux/module.h>
  31 #include <linux/vmalloc.h>
  32 #include <linux/completion.h>
  33 #include <linux/personality.h>
  34 #include <linux/mempolicy.h>
  35 #include <linux/sem.h>
  36 #include <linux/file.h>
  37 #include <linux/fdtable.h>
  38 #include <linux/iocontext.h>
  39 #include <linux/key.h>
  40 #include <linux/binfmts.h>
  41 #include <linux/mman.h>
  42 #include <linux/mmu_notifier.h>
  43 #include <linux/fs.h>
  44 #include <linux/mm.h>
  45 #include <linux/vmacache.h>
  46 #include <linux/nsproxy.h>
  47 #include <linux/capability.h>
  48 #include <linux/cpu.h>
  49 #include <linux/cgroup.h>
  50 #include <linux/security.h>
  51 #include <linux/hugetlb.h>
  52 #include <linux/seccomp.h>
  53 #include <linux/swap.h>
  54 #include <linux/syscalls.h>
  55 #include <linux/jiffies.h>
  56 #include <linux/futex.h>
  57 #include <linux/compat.h>
  58 #include <linux/kthread.h>
  59 #include <linux/task_io_accounting_ops.h>
  60 #include <linux/rcupdate.h>
  61 #include <linux/ptrace.h>
  62 #include <linux/mount.h>
  63 #include <linux/audit.h>
  64 #include <linux/memcontrol.h>
  65 #include <linux/ftrace.h>
  66 #include <linux/proc_fs.h>
  67 #include <linux/profile.h>
  68 #include <linux/rmap.h>
  69 #include <linux/ksm.h>
  70 #include <linux/acct.h>
  71 #include <linux/userfaultfd_k.h>
  72 #include <linux/tsacct_kern.h>
  73 #include <linux/cn_proc.h>
  74 #include <linux/freezer.h>
  75 #include <linux/delayacct.h>
  76 #include <linux/taskstats_kern.h>
  77 #include <linux/random.h>
  78 #include <linux/tty.h>
  79 #include <linux/blkdev.h>
  80 #include <linux/fs_struct.h>
  81 #include <linux/magic.h>
  82 #include <linux/perf_event.h>
  83 #include <linux/posix-timers.h>
  84 #include <linux/user-return-notifier.h>
  85 #include <linux/oom.h>
  86 #include <linux/khugepaged.h>
  87 #include <linux/signalfd.h>
  88 #include <linux/uprobes.h>
  89 #include <linux/aio.h>
  90 #include <linux/compiler.h>
  91 #include <linux/sysctl.h>
  92 #include <linux/kcov.h>
  93 #include <linux/livepatch.h>
  94 #include <linux/thread_info.h>
  95 #include <linux/stackleak.h>
  96 #include <linux/kasan.h>
  97 #include <linux/scs.h>
  98
  99 #include <asm/pgalloc.h>
 100 #include <linux/uaccess.h>
 101 #include <asm/mmu_context.h>
 102 #include <asm/cacheflush.h>
 103 #include <asm/tlbflush.h>
 104
 105 #include <trace/events/sched.h>
 106
 107 #define CREATE_TRACE_POINTS
 108 #include <trace/events/task.h>
 109
 110 /*
 111  * Minimum number of threads to boot the kernel
 112  */
 113 #define MIN_THREADS 20
 114
 115 /*
 116  * Maximum number of threads
 117  */
 118 #define MAX_THREADS FUTEX_TID_MASK
 119
 120 /*
 121  * Protected counters by write_lock_irq(&tasklist_lock)
 122  */
 123 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 124 int nr_threads;                 /* The idle threads do not count.. */
 125
 126 static int max_threads;         /* tunable limit on nr_threads */
 127
 128 #define NAMED_ARRAY_INDEX(x)    [x] = __stringify(x)
 129
 130 static const char * const resident_page_types[] = {
 131         NAMED_ARRAY_INDEX(MM_FILEPAGES),
 132         NAMED_ARRAY_INDEX(MM_ANONPAGES),
 133         NAMED_ARRAY_INDEX(MM_SWAPENTS),
 134         NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
 135 };
 136
 137 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 138
 139 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 140
 141 #ifdef CONFIG_PROVE_RCU
 142 int lockdep_tasklist_lock_is_held(void)
 143 {
 144         return lockdep_is_held(&tasklist_lock);
 145 }
 146 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 147 #endif /* #ifdef CONFIG_PROVE_RCU */
 148
 149 int nr_processes(void)
 150 {
 151         int cpu;
 152         int total = 0;
 153
 154         for_each_possible_cpu(cpu)
 155                 total += per_cpu(process_counts, cpu);
 156
 157         return total;
 158 }
 159
 160 void __weak arch_release_task_struct(struct task_struct *tsk)
 161 {
 162 }
 163
 164 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 165 static struct kmem_cache *task_struct_cachep;
 166
 167 static inline struct task_struct *alloc_task_struct_node(int node)
 168 {
 169         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 170 }
 171
 172 static inline void free_task_struct(struct task_struct *tsk)
 173 {
 174         kmem_cache_free(task_struct_cachep, tsk);
 175 }
 176 #endif
 177
 178 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 179
 180 /*
 181  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 182  * kmemcache based allocator.
 183  */
 184 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 185
 186 #ifdef CONFIG_VMAP_STACK
 187 /*
 188  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 189  * flush.  Try to minimize the number of calls by caching stacks.
 190  */
 191 #define NR_CACHED_STACKS 2
 192 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 193
 194 static int free_vm_stack_cache(unsigned int cpu)
 195 {
 196         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 197         int i;
 198
 199         for (i = 0; i < NR_CACHED_STACKS; i++) {
 200                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 201
 202                 if (!vm_stack)
 203                         continue;
 204
 205                 vfree(vm_stack->addr);
 206                 cached_vm_stacks[i] = NULL;
 207         }
 208
 209         return 0;
 210 }
 211 #endif
 212
 213 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 214 {
 215 #ifdef CONFIG_VMAP_STACK
 216         void *stack;
 217         int i;
 218
 219         for (i = 0; i < NR_CACHED_STACKS; i++) {
 220                 struct vm_struct *s;
 221
 222                 s = this_cpu_xchg(cached_stacks[i], NULL);
 223
 224                 if (!s)
 225                         continue;
 226
 227                 /* Clear the KASAN shadow of the stack. */
 228                 kasan_unpoison_shadow(s->addr, THREAD_SIZE);
 229
 230                 /* Clear stale pointers from reused stack. */
 231                 memset(s->addr, 0, THREAD_SIZE);
 232
 233                 tsk->stack_vm_area = s;
 234                 tsk->stack = s->addr;
 235                 return s->addr;
 236         }
 237
 238         /*
 239          * Allocated stacks are cached and later reused by new threads,
 240          * so memcg accounting is performed manually on assigning/releasing
 241          * stacks to tasks. Drop __GFP_ACCOUNT.
 242          */
 243         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 244                                      VMALLOC_START, VMALLOC_END,
 245                                      THREADINFO_GFP & ~__GFP_ACCOUNT,
 246                                      PAGE_KERNEL,
 247                                      0, node, __builtin_return_address(0));
 248
 249         /*
 250          * We can't call find_vm_area() in interrupt context, and
 251          * free_thread_stack() can be called in interrupt context,
 252          * so cache the vm_struct.
 253          */
 254         if (stack) {
 255                 tsk->stack_vm_area = find_vm_area(stack);
 256                 tsk->stack = stack;
 257         }
 258         return stack;
 259 #else
 260         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 261                                              THREAD_SIZE_ORDER);
 262
 263         if (likely(page)) {
 264                 tsk->stack = page_address(page);
 265                 return tsk->stack;
 266         }
 267         return NULL;
 268 #endif
 269 }
 270
 271 static inline void free_thread_stack(struct task_struct *tsk)
 272 {
 273 #ifdef CONFIG_VMAP_STACK
 274         struct vm_struct *vm = task_stack_vm_area(tsk);
 275
 276         if (vm) {
 277                 int i;
 278
 279                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 280                         mod_memcg_page_state(vm->pages[i],
 281                                              MEMCG_KERNEL_STACK_KB,
 282                                              -(int)(PAGE_SIZE / 1024));
 283
 284                         memcg_kmem_uncharge_page(vm->pages[i], 0);
 285                 }
 286
 287                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 288                         if (this_cpu_cmpxchg(cached_stacks[i],
 289                                         NULL, tsk->stack_vm_area) != NULL)
 290                                 continue;
 291
 292                         return;
 293                 }
 294
 295                 vfree_atomic(tsk->stack);
 296                 return;
 297         }
 298 #endif
 299
 300         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 301 }
 302 # else
 303 static struct kmem_cache *thread_stack_cache;
 304
 305 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 306                                                   int node)
 307 {
 308         unsigned long *stack;
 309         stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 310         tsk->stack = stack;
 311         return stack;
 312 }
 313
 314 static void free_thread_stack(struct task_struct *tsk)
 315 {
 316         kmem_cache_free(thread_stack_cache, tsk->stack);
 317 }
 318
 319 void thread_stack_cache_init(void)
 320 {
 321         thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 322                                         THREAD_SIZE, THREAD_SIZE, 0, 0,
 323                                         THREAD_SIZE, NULL);
 324         BUG_ON(thread_stack_cache == NULL);
 325 }
 326 # endif
 327 #endif
 328
 329 /* SLAB cache for signal_struct structures (tsk->signal) */
 330 static struct kmem_cache *signal_cachep;
 331
 332 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 333 struct kmem_cache *sighand_cachep;
 334
 335 /* SLAB cache for files_struct structures (tsk->files) */
 336 struct kmem_cache *files_cachep;
 337
 338 /* SLAB cache for fs_struct structures (tsk->fs) */
 339 struct kmem_cache *fs_cachep;
 340
 341 /* SLAB cache for vm_area_struct structures */
 342 static struct kmem_cache *vm_area_cachep;
 343
 344 /* SLAB cache for mm_struct structures (tsk->mm) */
 345 static struct kmem_cache *mm_cachep;
 346
 347 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 348 {
 349         struct vm_area_struct *vma;
 350
 351         vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 352         if (vma)
 353                 vma_init(vma, mm);
 354         return vma;
 355 }
 356
 357 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 358 {
 359         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 360
 361         if (new) {
 362                 *new = *orig;
 363                 INIT_LIST_HEAD(&new->anon_vma_chain);
 364                 new->vm_next = new->vm_prev = NULL;
 365         }
 366         return new;
 367 }
 368
 369 void vm_area_free(struct vm_area_struct *vma)
 370 {
 371         kmem_cache_free(vm_area_cachep, vma);
 372 }
 373
 374 static void account_kernel_stack(struct task_struct *tsk, int account)
 375 {
 376         void *stack = task_stack_page(tsk);
 377         struct vm_struct *vm = task_stack_vm_area(tsk);
 378
 379         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 380
 381         if (vm) {
 382                 int i;
 383
 384                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 385
 386                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 387                         mod_zone_page_state(page_zone(vm->pages[i]),
 388                                             NR_KERNEL_STACK_KB,
 389                                             PAGE_SIZE / 1024 * account);
 390                 }
 391         } else {
 392                 /*
 393                  * All stack pages are in the same zone and belong to the
 394                  * same memcg.
 395                  */
 396                 struct page *first_page = virt_to_page(stack);
 397
 398                 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
 399                                     THREAD_SIZE / 1024 * account);
 400
 401                 mod_memcg_obj_state(stack, MEMCG_KERNEL_STACK_KB,
 402                                     account * (THREAD_SIZE / 1024));
 403         }
 404 }
 405
 406 static int memcg_charge_kernel_stack(struct task_struct *tsk)
 407 {
 408 #ifdef CONFIG_VMAP_STACK
 409         struct vm_struct *vm = task_stack_vm_area(tsk);
 410         int ret;
 411
 412         if (vm) {
 413                 int i;
 414
 415                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 416                         /*
 417                          * If memcg_kmem_charge_page() fails, page->mem_cgroup
 418                          * pointer is NULL, and both memcg_kmem_uncharge_page()
 419                          * and mod_memcg_page_state() in free_thread_stack()
 420                          * will ignore this page. So it's safe.
 421                          */
 422                         ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL,
 423                                                      0);
 424                         if (ret)
 425                                 return ret;
 426
 427                         mod_memcg_page_state(vm->pages[i],
 428                                              MEMCG_KERNEL_STACK_KB,
 429                                              PAGE_SIZE / 1024);
 430                 }
 431         }
 432 #endif
 433         return 0;
 434 }
 435
 436 static void release_task_stack(struct task_struct *tsk)
 437 {
 438         if (WARN_ON(tsk->state != TASK_DEAD))
 439                 return;  /* Better to leak the stack than to free prematurely */
 440
 441         account_kernel_stack(tsk, -1);
 442         free_thread_stack(tsk);
 443         tsk->stack = NULL;
 444 #ifdef CONFIG_VMAP_STACK
 445         tsk->stack_vm_area = NULL;
 446 #endif
 447 }
 448
 449 #ifdef CONFIG_THREAD_INFO_IN_TASK
 450 void put_task_stack(struct task_struct *tsk)
 451 {
 452         if (refcount_dec_and_test(&tsk->stack_refcount))
 453                 release_task_stack(tsk);
 454 }
 455 #endif
 456
 457 void free_task(struct task_struct *tsk)
 458 {
 459         scs_release(tsk);
 460
 461 #ifndef CONFIG_THREAD_INFO_IN_TASK
 462         /*
 463          * The task is finally done with both the stack and thread_info,
 464          * so free both.
 465          */
 466         release_task_stack(tsk);
 467 #else
 468         /*
 469          * If the task had a separate stack allocation, it should be gone
 470          * by now.
 471          */
 472         WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
 473 #endif
 474         rt_mutex_debug_task_free(tsk);
 475         ftrace_graph_exit_task(tsk);
 476         put_seccomp_filter(tsk);
 477         arch_release_task_struct(tsk);
 478         if (tsk->flags & PF_KTHREAD)
 479                 free_kthread_struct(tsk);
 480         free_task_struct(tsk);
 481 }
 482 EXPORT_SYMBOL(free_task);
 483
 484 #ifdef CONFIG_MMU
 485 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 486                                         struct mm_struct *oldmm)
 487 {
 488         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 489         struct rb_node **rb_link, *rb_parent;
 490         int retval;
 491         unsigned long charge;
 492         LIST_HEAD(uf);
 493
 494         uprobe_start_dup_mmap();
 495         if (mmap_write_lock_killable(oldmm)) {
 496                 retval = -EINTR;
 497                 goto fail_uprobe_end;
 498         }
 499         flush_cache_dup_mm(oldmm);
 500         uprobe_dup_mmap(oldmm, mm);
 501         /*
 502          * Not linked in yet - no deadlock potential:
 503          */
 504         mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
 505
 506         /* No ordering required: file already has been exposed. */
 507         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 508
 509         mm->total_vm = oldmm->total_vm;
 510         mm->data_vm = oldmm->data_vm;
 511         mm->exec_vm = oldmm->exec_vm;
 512         mm->stack_vm = oldmm->stack_vm;
 513
 514         rb_link = &mm->mm_rb.rb_node;
 515         rb_parent = NULL;
 516         pprev = &mm->mmap;
 517         retval = ksm_fork(mm, oldmm);
 518         if (retval)
 519                 goto out;
 520         retval = khugepaged_fork(mm, oldmm);
 521         if (retval)
 522                 goto out;
 523
 524         prev = NULL;
 525         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 526                 struct file *file;
 527
 528                 if (mpnt->vm_flags & VM_DONTCOPY) {
 529                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 530                         continue;
 531                 }
 532                 charge = 0;
 533                 /*
 534                  * Don't duplicate many vmas if we've been oom-killed (for
 535                  * example)
 536                  */
 537                 if (fatal_signal_pending(current)) {
 538                         retval = -EINTR;
 539                         goto out;
 540                 }
 541                 if (mpnt->vm_flags & VM_ACCOUNT) {
 542                         unsigned long len = vma_pages(mpnt);
 543
 544                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 545                                 goto fail_nomem;
 546                         charge = len;
 547                 }
 548                 tmp = vm_area_dup(mpnt);
 549                 if (!tmp)
 550                         goto fail_nomem;
 551                 retval = vma_dup_policy(mpnt, tmp);
 552                 if (retval)
 553                         goto fail_nomem_policy;
 554                 tmp->vm_mm = mm;
 555                 retval = dup_userfaultfd(tmp, &uf);
 556                 if (retval)
 557                         goto fail_nomem_anon_vma_fork;
 558                 if (tmp->vm_flags & VM_WIPEONFORK) {
 559                         /*
 560                          * VM_WIPEONFORK gets a clean slate in the child.
 561                          * Don't prepare anon_vma until fault since we don't
 562                          * copy page for current vma.
 563                          */
 564                         tmp->anon_vma = NULL;
 565                 } else if (anon_vma_fork(tmp, mpnt))
 566                         goto fail_nomem_anon_vma_fork;
 567                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 568                 file = tmp->vm_file;
 569                 if (file) {
 570                         struct inode *inode = file_inode(file);
 571                         struct address_space *mapping = file->f_mapping;
 572
 573                         get_file(file);
 574                         if (tmp->vm_flags & VM_DENYWRITE)
 575                                 atomic_dec(&inode->i_writecount);
 576                         i_mmap_lock_write(mapping);
 577                         if (tmp->vm_flags & VM_SHARED)
 578                                 atomic_inc(&mapping->i_mmap_writable);
 579                         flush_dcache_mmap_lock(mapping);
 580                         /* insert tmp into the share list, just after mpnt */
 581                         vma_interval_tree_insert_after(tmp, mpnt,
 582                                         &mapping->i_mmap);
 583                         flush_dcache_mmap_unlock(mapping);
 584                         i_mmap_unlock_write(mapping);
 585                 }
 586
 587                 /*
 588                  * Clear hugetlb-related page reserves for children. This only
 589                  * affects MAP_PRIVATE mappings. Faults generated by the child
 590                  * are not guaranteed to succeed, even if read-only
 591                  */
 592                 if (is_vm_hugetlb_page(tmp))
 593                         reset_vma_resv_huge_pages(tmp);
 594
 595                 /*
 596                  * Link in the new vma and copy the page table entries.
 597                  */
 598                 *pprev = tmp;
 599                 pprev = &tmp->vm_next;
 600                 tmp->vm_prev = prev;
 601                 prev = tmp;
 602
 603                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 604                 rb_link = &tmp->vm_rb.rb_right;
 605                 rb_parent = &tmp->vm_rb;
 606
 607                 mm->map_count++;
 608                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 609                         retval = copy_page_range(mm, oldmm, mpnt);
 610
 611                 if (tmp->vm_ops && tmp->vm_ops->open)
 612                         tmp->vm_ops->open(tmp);
 613
 614                 if (retval)
 615                         goto out;
 616         }
 617         /* a new mm has just been created */
 618         retval = arch_dup_mmap(oldmm, mm);
 619 out:
 620         mmap_write_unlock(mm);
 621         flush_tlb_mm(oldmm);
 622         mmap_write_unlock(oldmm);
 623         dup_userfaultfd_complete(&uf);
 624 fail_uprobe_end:
 625         uprobe_end_dup_mmap();
 626         return retval;
 627 fail_nomem_anon_vma_fork:
 628         mpol_put(vma_policy(tmp));
 629 fail_nomem_policy:
 630         vm_area_free(tmp);
 631 fail_nomem:
 632         retval = -ENOMEM;
 633         vm_unacct_memory(charge);
 634         goto out;
 635 }
 636
 637 static inline int mm_alloc_pgd(struct mm_struct *mm)
 638 {
 639         mm->pgd = pgd_alloc(mm);
 640         if (unlikely(!mm->pgd))
 641                 return -ENOMEM;
 642         return 0;
 643 }
 644
 645 static inline void mm_free_pgd(struct mm_struct *mm)
 646 {
 647         pgd_free(mm, mm->pgd);
 648 }
 649 #else
 650 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 651 {
 652         mmap_write_lock(oldmm);
 653         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 654         mmap_write_unlock(oldmm);
 655         return 0;
 656 }
 657 #define mm_alloc_pgd(mm)        (0)
 658 #define mm_free_pgd(mm)
 659 #endif /* CONFIG_MMU */
 660
 661 static void check_mm(struct mm_struct *mm)
 662 {
 663         int i;
 664
 665         BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
 666                          "Please make sure 'struct resident_page_types[]' is updated as well");
 667
 668         for (i = 0; i < NR_MM_COUNTERS; i++) {
 669                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 670
 671                 if (unlikely(x))
 672                         pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
 673                                  mm, resident_page_types[i], x);
 674         }
 675
 676         if (mm_pgtables_bytes(mm))
 677                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 678                                 mm_pgtables_bytes(mm));
 679
 680 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 681         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 682 #endif
 683 }
 684
 685 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 686 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 687
 688 /*
 689  * Called when the last reference to the mm
 690  * is dropped: either by a lazy thread or by
 691  * mmput. Free the page directory and the mm.
 692  */
 693 void __mmdrop(struct mm_struct *mm)
 694 {
 695         BUG_ON(mm == &init_mm);
 696         WARN_ON_ONCE(mm == current->mm);
 697         WARN_ON_ONCE(mm == current->active_mm);
 698         mm_free_pgd(mm);
 699         destroy_context(mm);
 700         mmu_notifier_subscriptions_destroy(mm);
 701         check_mm(mm);
 702         put_user_ns(mm->user_ns);
 703         free_mm(mm);
 704 }
 705 EXPORT_SYMBOL_GPL(__mmdrop);
 706
 707 static void mmdrop_async_fn(struct work_struct *work)
 708 {
 709         struct mm_struct *mm;
 710
 711         mm = container_of(work, struct mm_struct, async_put_work);
 712         __mmdrop(mm);
 713 }
 714
 715 static void mmdrop_async(struct mm_struct *mm)
 716 {
 717         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 718                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 719                 schedule_work(&mm->async_put_work);
 720         }
 721 }
 722
 723 static inline void free_signal_struct(struct signal_struct *sig)
 724 {
 725         taskstats_tgid_free(sig);
 726         sched_autogroup_exit(sig);
 727         /*
 728          * __mmdrop is not safe to call from softirq context on x86 due to
 729          * pgd_dtor so postpone it to the async context
 730          */
 731         if (sig->oom_mm)
 732                 mmdrop_async(sig->oom_mm);
 733         kmem_cache_free(signal_cachep, sig);
 734 }
 735
 736 static inline void put_signal_struct(struct signal_struct *sig)
 737 {
 738         if (refcount_dec_and_test(&sig->sigcnt))
 739                 free_signal_struct(sig);
 740 }
 741
 742 void __put_task_struct(struct task_struct *tsk)
 743 {
 744         WARN_ON(!tsk->exit_state);
 745         WARN_ON(refcount_read(&tsk->usage));
 746         WARN_ON(tsk == current);
 747
 748         cgroup_free(tsk);
 749         task_numa_free(tsk, true);
 750         security_task_free(tsk);
 751         exit_creds(tsk);
 752         delayacct_tsk_free(tsk);
 753         put_signal_struct(tsk->signal);
 754
 755         if (!profile_handoff_task(tsk))
 756                 free_task(tsk);
 757 }
 758 EXPORT_SYMBOL_GPL(__put_task_struct);
 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         scs_init();
 846
 847         lockdep_init_task(&init_task);
 848         uprobes_init();
 849 }
 850
 851 int __weak arch_dup_task_struct(struct task_struct *dst,
 852                                                struct task_struct *src)
 853 {
 854         *dst = *src;
 855         return 0;
 856 }
 857
 858 void set_task_stack_end_magic(struct task_struct *tsk)
 859 {
 860         unsigned long *stackend;
 861
 862         stackend = end_of_stack(tsk);
 863         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 864 }
 865
 866 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 867 {
 868         struct task_struct *tsk;
 869         unsigned long *stack;
 870         struct vm_struct *stack_vm_area __maybe_unused;
 871         int err;
 872
 873         if (node == NUMA_NO_NODE)
 874                 node = tsk_fork_get_node(orig);
 875         tsk = alloc_task_struct_node(node);
 876         if (!tsk)
 877                 return NULL;
 878
 879         stack = alloc_thread_stack_node(tsk, node);
 880         if (!stack)
 881                 goto free_tsk;
 882
 883         if (memcg_charge_kernel_stack(tsk))
 884                 goto free_stack;
 885
 886         stack_vm_area = task_stack_vm_area(tsk);
 887
 888         err = arch_dup_task_struct(tsk, orig);
 889
 890         /*
 891          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 892          * sure they're properly initialized before using any stack-related
 893          * functions again.
 894          */
 895         tsk->stack = stack;
 896 #ifdef CONFIG_VMAP_STACK
 897         tsk->stack_vm_area = stack_vm_area;
 898 #endif
 899 #ifdef CONFIG_THREAD_INFO_IN_TASK
 900         refcount_set(&tsk->stack_refcount, 1);
 901 #endif
 902
 903         if (err)
 904                 goto free_stack;
 905
 906         err = scs_prepare(tsk, node);
 907         if (err)
 908                 goto free_stack;
 909
 910 #ifdef CONFIG_SECCOMP
 911         /*
 912          * We must handle setting up seccomp filters once we're under
 913          * the sighand lock in case orig has changed between now and
 914          * then. Until then, filter must be NULL to avoid messing up
 915          * the usage counts on the error path calling free_task.
 916          */
 917         tsk->seccomp.filter = NULL;
 918 #endif
 919
 920         setup_thread_stack(tsk, orig);
 921         clear_user_return_notifier(tsk);
 922         clear_tsk_need_resched(tsk);
 923         set_task_stack_end_magic(tsk);
 924
 925 #ifdef CONFIG_STACKPROTECTOR
 926         tsk->stack_canary = get_random_canary();
 927 #endif
 928         if (orig->cpus_ptr == &orig->cpus_mask)
 929                 tsk->cpus_ptr = &tsk->cpus_mask;
 930
 931         /*
 932          * One for the user space visible state that goes away when reaped.
 933          * One for the scheduler.
 934          */
 935         refcount_set(&tsk->rcu_users, 2);
 936         /* One for the rcu users */
 937         refcount_set(&tsk->usage, 1);
 938 #ifdef CONFIG_BLK_DEV_IO_TRACE
 939         tsk->btrace_seq = 0;
 940 #endif
 941         tsk->splice_pipe = NULL;
 942         tsk->task_frag.page = NULL;
 943         tsk->wake_q.next = NULL;
 944
 945         account_kernel_stack(tsk, 1);
 946
 947         kcov_task_init(tsk);
 948
 949 #ifdef CONFIG_FAULT_INJECTION
 950         tsk->fail_nth = 0;
 951 #endif
 952
 953 #ifdef CONFIG_BLK_CGROUP
 954         tsk->throttle_queue = NULL;
 955         tsk->use_memdelay = 0;
 956 #endif
 957
 958 #ifdef CONFIG_MEMCG
 959         tsk->active_memcg = NULL;
 960 #endif
 961         return tsk;
 962
 963 free_stack:
 964         free_thread_stack(tsk);
 965 free_tsk:
 966         free_task_struct(tsk);
 967         return NULL;
 968 }
 969
 970 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 971
 972 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 973
 974 static int __init coredump_filter_setup(char *s)
 975 {
 976         default_dump_filter =
 977                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 978                 MMF_DUMP_FILTER_MASK;
 979         return 1;
 980 }
 981
 982 __setup("coredump_filter=", coredump_filter_setup);
 983
 984 #include <linux/init_task.h>
 985
 986 static void mm_init_aio(struct mm_struct *mm)
 987 {
 988 #ifdef CONFIG_AIO
 989         spin_lock_init(&mm->ioctx_lock);
 990         mm->ioctx_table = NULL;
 991 #endif
 992 }
 993
 994 static __always_inline void mm_clear_owner(struct mm_struct *mm,
 995                                            struct task_struct *p)
 996 {
 997 #ifdef CONFIG_MEMCG
 998         if (mm->owner == p)
 999                 WRITE_ONCE(mm->owner, NULL);
1000 #endif
1001 }
1002
1003 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1004 {
1005 #ifdef CONFIG_MEMCG
1006         mm->owner = p;
1007 #endif
1008 }
1009
1010 static void mm_init_uprobes_state(struct mm_struct *mm)
1011 {
1012 #ifdef CONFIG_UPROBES
1013         mm->uprobes_state.xol_area = NULL;
1014 #endif
1015 }
1016
1017 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1018         struct user_namespace *user_ns)
1019 {
1020         mm->mmap = NULL;
1021         mm->mm_rb = RB_ROOT;
1022         mm->vmacache_seqnum = 0;
1023         atomic_set(&mm->mm_users, 1);
1024         atomic_set(&mm->mm_count, 1);
1025         mmap_init_lock(mm);
1026         INIT_LIST_HEAD(&mm->mmlist);
1027         mm->core_state = NULL;
1028         mm_pgtables_bytes_init(mm);
1029         mm->map_count = 0;
1030         mm->locked_vm = 0;
1031         atomic64_set(&mm->pinned_vm, 0);
1032         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1033         spin_lock_init(&mm->page_table_lock);
1034         spin_lock_init(&mm->arg_lock);
1035         mm_init_cpumask(mm);
1036         mm_init_aio(mm);
1037         mm_init_owner(mm, p);
1038         RCU_INIT_POINTER(mm->exe_file, NULL);
1039         mmu_notifier_subscriptions_init(mm);
1040         init_tlb_flush_pending(mm);
1041 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1042         mm->pmd_huge_pte = NULL;
1043 #endif
1044         mm_init_uprobes_state(mm);
1045
1046         if (current->mm) {
1047                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1048                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1049         } else {
1050                 mm->flags = default_dump_filter;
1051                 mm->def_flags = 0;
1052         }
1053
1054         if (mm_alloc_pgd(mm))
1055                 goto fail_nopgd;
1056
1057         if (init_new_context(p, mm))
1058                 goto fail_nocontext;
1059
1060         mm->user_ns = get_user_ns(user_ns);
1061         return mm;
1062
1063 fail_nocontext:
1064         mm_free_pgd(mm);
1065 fail_nopgd:
1066         free_mm(mm);
1067         return NULL;
1068 }
1069
1070 /*
1071  * Allocate and initialize an mm_struct.
1072  */
1073 struct mm_struct *mm_alloc(void)
1074 {
1075         struct mm_struct *mm;
1076
1077         mm = allocate_mm();
1078         if (!mm)
1079                 return NULL;
1080
1081         memset(mm, 0, sizeof(*mm));
1082         return mm_init(mm, current, current_user_ns());
1083 }
1084
1085 static inline void __mmput(struct mm_struct *mm)
1086 {
1087         VM_BUG_ON(atomic_read(&mm->mm_users));
1088
1089         uprobe_clear_state(mm);
1090         exit_aio(mm);
1091         ksm_exit(mm);
1092         khugepaged_exit(mm); /* must run before exit_mmap */
1093         exit_mmap(mm);
1094         mm_put_huge_zero_page(mm);
1095         set_mm_exe_file(mm, NULL);
1096         if (!list_empty(&mm->mmlist)) {
1097                 spin_lock(&mmlist_lock);
1098                 list_del(&mm->mmlist);
1099                 spin_unlock(&mmlist_lock);
1100         }
1101         if (mm->binfmt)
1102                 module_put(mm->binfmt->module);
1103         mmdrop(mm);
1104 }
1105
1106 /*
1107  * Decrement the use count and release all resources for an mm.
1108  */
1109 void mmput(struct mm_struct *mm)
1110 {
1111         might_sleep();
1112
1113         if (atomic_dec_and_test(&mm->mm_users))
1114                 __mmput(mm);
1115 }
1116 EXPORT_SYMBOL_GPL(mmput);
1117
1118 #ifdef CONFIG_MMU
1119 static void mmput_async_fn(struct work_struct *work)
1120 {
1121         struct mm_struct *mm = container_of(work, struct mm_struct,
1122                                             async_put_work);
1123
1124         __mmput(mm);
1125 }
1126
1127 void mmput_async(struct mm_struct *mm)
1128 {
1129         if (atomic_dec_and_test(&mm->mm_users)) {
1130                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1131                 schedule_work(&mm->async_put_work);
1132         }
1133 }
1134 #endif
1135
1136 /**
1137  * set_mm_exe_file - change a reference to the mm's executable file
1138  *
1139  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1140  *
1141  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1142  * invocations: in mmput() nobody alive left, in execve task is single
1143  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1144  * mm->exe_file, but does so without using set_mm_exe_file() in order
1145  * to do avoid the need for any locks.
1146  */
1147 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1148 {
1149         struct file *old_exe_file;
1150
1151         /*
1152          * It is safe to dereference the exe_file without RCU as
1153          * this function is only called if nobody else can access
1154          * this mm -- see comment above for justification.
1155          */
1156         old_exe_file = rcu_dereference_raw(mm->exe_file);
1157
1158         if (new_exe_file)
1159                 get_file(new_exe_file);
1160         rcu_assign_pointer(mm->exe_file, new_exe_file);
1161         if (old_exe_file)
1162                 fput(old_exe_file);
1163 }
1164
1165 /**
1166  * get_mm_exe_file - acquire a reference to the mm's executable file
1167  *
1168  * Returns %NULL if mm has no associated executable file.
1169  * User must release file via fput().
1170  */
1171 struct file *get_mm_exe_file(struct mm_struct *mm)
1172 {
1173         struct file *exe_file;
1174
1175         rcu_read_lock();
1176         exe_file = rcu_dereference(mm->exe_file);
1177         if (exe_file && !get_file_rcu(exe_file))
1178                 exe_file = NULL;
1179         rcu_read_unlock();
1180         return exe_file;
1181 }
1182 EXPORT_SYMBOL(get_mm_exe_file);
1183
1184 /**
1185  * get_task_exe_file - acquire a reference to the task's executable file
1186  *
1187  * Returns %NULL if task's mm (if any) has no associated executable file or
1188  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1189  * User must release file via fput().
1190  */
1191 struct file *get_task_exe_file(struct task_struct *task)
1192 {
1193         struct file *exe_file = NULL;
1194         struct mm_struct *mm;
1195
1196         task_lock(task);
1197         mm = task->mm;
1198         if (mm) {
1199                 if (!(task->flags & PF_KTHREAD))
1200                         exe_file = get_mm_exe_file(mm);
1201         }
1202         task_unlock(task);
1203         return exe_file;
1204 }
1205 EXPORT_SYMBOL(get_task_exe_file);
1206
1207 /**
1208  * get_task_mm - acquire a reference to the task's mm
1209  *
1210  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1211  * this kernel workthread has transiently adopted a user mm with use_mm,
1212  * to do its AIO) is not set and if so returns a reference to it, after
1213  * bumping up the use count.  User must release the mm via mmput()
1214  * after use.  Typically used by /proc and ptrace.
1215  */
1216 struct mm_struct *get_task_mm(struct task_struct *task)
1217 {
1218         struct mm_struct *mm;
1219
1220         task_lock(task);
1221         mm = task->mm;
1222         if (mm) {
1223                 if (task->flags & PF_KTHREAD)
1224                         mm = NULL;
1225                 else
1226                         mmget(mm);
1227         }
1228         task_unlock(task);
1229         return mm;
1230 }
1231 EXPORT_SYMBOL_GPL(get_task_mm);
1232
1233 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1234 {
1235         struct mm_struct *mm;
1236         int err;
1237
1238         err =  mutex_lock_killable(&task->signal->exec_update_mutex);
1239         if (err)
1240                 return ERR_PTR(err);
1241
1242         mm = get_task_mm(task);
1243         if (mm && mm != current->mm &&
1244                         !ptrace_may_access(task, mode)) {
1245                 mmput(mm);
1246                 mm = ERR_PTR(-EACCES);
1247         }
1248         mutex_unlock(&task->signal->exec_update_mutex);
1249
1250         return mm;
1251 }
1252
1253 static void complete_vfork_done(struct task_struct *tsk)
1254 {
1255         struct completion *vfork;
1256
1257         task_lock(tsk);
1258         vfork = tsk->vfork_done;
1259         if (likely(vfork)) {
1260                 tsk->vfork_done = NULL;
1261                 complete(vfork);
1262         }
1263         task_unlock(tsk);
1264 }
1265
1266 static int wait_for_vfork_done(struct task_struct *child,
1267                                 struct completion *vfork)
1268 {
1269         int killed;
1270
1271         freezer_do_not_count();
1272         cgroup_enter_frozen();
1273         killed = wait_for_completion_killable(vfork);
1274         cgroup_leave_frozen(false);
1275         freezer_count();
1276
1277         if (killed) {
1278                 task_lock(child);
1279                 child->vfork_done = NULL;
1280                 task_unlock(child);
1281         }
1282
1283         put_task_struct(child);
1284         return killed;
1285 }
1286
1287 /* Please note the differences between mmput and mm_release.
1288  * mmput is called whenever we stop holding onto a mm_struct,
1289  * error success whatever.
1290  *
1291  * mm_release is called after a mm_struct has been removed
1292  * from the current process.
1293  *
1294  * This difference is important for error handling, when we
1295  * only half set up a mm_struct for a new process and need to restore
1296  * the old one.  Because we mmput the new mm_struct before
1297  * restoring the old one. . .
1298  * Eric Biederman 10 January 1998
1299  */
1300 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1301 {
1302         uprobe_free_utask(tsk);
1303
1304         /* Get rid of any cached register state */
1305         deactivate_mm(tsk, mm);
1306
1307         /*
1308          * Signal userspace if we're not exiting with a core dump
1309          * because we want to leave the value intact for debugging
1310          * purposes.
1311          */
1312         if (tsk->clear_child_tid) {
1313                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1314                     atomic_read(&mm->mm_users) > 1) {
1315                         /*
1316                          * We don't check the error code - if userspace has
1317                          * not set up a proper pointer then tough luck.
1318                          */
1319                         put_user(0, tsk->clear_child_tid);
1320                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1321                                         1, NULL, NULL, 0, 0);
1322                 }
1323                 tsk->clear_child_tid = NULL;
1324         }
1325
1326         /*
1327          * All done, finally we can wake up parent and return this mm to him.
1328          * Also kthread_stop() uses this completion for synchronization.
1329          */
1330         if (tsk->vfork_done)
1331                 complete_vfork_done(tsk);
1332 }
1333
1334 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1335 {
1336         futex_exit_release(tsk);
1337         mm_release(tsk, mm);
1338 }
1339
1340 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1341 {
1342         futex_exec_release(tsk);
1343         mm_release(tsk, mm);
1344 }
1345
1346 /**
1347  * dup_mm() - duplicates an existing mm structure
1348  * @tsk: the task_struct with which the new mm will be associated.
1349  * @oldmm: the mm to duplicate.
1350  *
1351  * Allocates a new mm structure and duplicates the provided @oldmm structure
1352  * content into it.
1353  *
1354  * Return: the duplicated mm or NULL on failure.
1355  */
1356 static struct mm_struct *dup_mm(struct task_struct *tsk,
1357                                 struct mm_struct *oldmm)
1358 {
1359         struct mm_struct *mm;
1360         int err;
1361
1362         mm = allocate_mm();
1363         if (!mm)
1364                 goto fail_nomem;
1365
1366         memcpy(mm, oldmm, sizeof(*mm));
1367
1368         if (!mm_init(mm, tsk, mm->user_ns))
1369                 goto fail_nomem;
1370
1371         err = dup_mmap(mm, oldmm);
1372         if (err)
1373                 goto free_pt;
1374
1375         mm->hiwater_rss = get_mm_rss(mm);
1376         mm->hiwater_vm = mm->total_vm;
1377
1378         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1379                 goto free_pt;
1380
1381         return mm;
1382
1383 free_pt:
1384         /* don't put binfmt in mmput, we haven't got module yet */
1385         mm->binfmt = NULL;
1386         mm_init_owner(mm, NULL);
1387         mmput(mm);
1388
1389 fail_nomem:
1390         return NULL;
1391 }
1392
1393 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1394 {
1395         struct mm_struct *mm, *oldmm;
1396         int retval;
1397
1398         tsk->min_flt = tsk->maj_flt = 0;
1399         tsk->nvcsw = tsk->nivcsw = 0;
1400 #ifdef CONFIG_DETECT_HUNG_TASK
1401         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1402         tsk->last_switch_time = 0;
1403 #endif
1404
1405         tsk->mm = NULL;
1406         tsk->active_mm = NULL;
1407
1408         /*
1409          * Are we cloning a kernel thread?
1410          *
1411          * We need to steal a active VM for that..
1412          */
1413         oldmm = current->mm;
1414         if (!oldmm)
1415                 return 0;
1416
1417         /* initialize the new vmacache entries */
1418         vmacache_flush(tsk);
1419
1420         if (clone_flags & CLONE_VM) {
1421                 mmget(oldmm);
1422                 mm = oldmm;
1423                 goto good_mm;
1424         }
1425
1426         retval = -ENOMEM;
1427         mm = dup_mm(tsk, current->mm);
1428         if (!mm)
1429                 goto fail_nomem;
1430
1431 good_mm:
1432         tsk->mm = mm;
1433         tsk->active_mm = mm;
1434         return 0;
1435
1436 fail_nomem:
1437         return retval;
1438 }
1439
1440 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1441 {
1442         struct fs_struct *fs = current->fs;
1443         if (clone_flags & CLONE_FS) {
1444                 /* tsk->fs is already what we want */
1445                 spin_lock(&fs->lock);
1446                 if (fs->in_exec) {
1447                         spin_unlock(&fs->lock);
1448                         return -EAGAIN;
1449                 }
1450                 fs->users++;
1451                 spin_unlock(&fs->lock);
1452                 return 0;
1453         }
1454         tsk->fs = copy_fs_struct(fs);
1455         if (!tsk->fs)
1456                 return -ENOMEM;
1457         return 0;
1458 }
1459
1460 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1461 {
1462         struct files_struct *oldf, *newf;
1463         int error = 0;
1464
1465         /*
1466          * A background process may not have any files ...
1467          */
1468         oldf = current->files;
1469         if (!oldf)
1470                 goto out;
1471
1472         if (clone_flags & CLONE_FILES) {
1473                 atomic_inc(&oldf->count);
1474                 goto out;
1475         }
1476
1477         newf = dup_fd(oldf, &error);
1478         if (!newf)
1479                 goto out;
1480
1481         tsk->files = newf;
1482         error = 0;
1483 out:
1484         return error;
1485 }
1486
1487 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1488 {
1489 #ifdef CONFIG_BLOCK
1490         struct io_context *ioc = current->io_context;
1491         struct io_context *new_ioc;
1492
1493         if (!ioc)
1494                 return 0;
1495         /*
1496          * Share io context with parent, if CLONE_IO is set
1497          */
1498         if (clone_flags & CLONE_IO) {
1499                 ioc_task_link(ioc);
1500                 tsk->io_context = ioc;
1501         } else if (ioprio_valid(ioc->ioprio)) {
1502                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1503                 if (unlikely(!new_ioc))
1504                         return -ENOMEM;
1505
1506                 new_ioc->ioprio = ioc->ioprio;
1507                 put_io_context(new_ioc);
1508         }
1509 #endif
1510         return 0;
1511 }
1512
1513 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1514 {
1515         struct sighand_struct *sig;
1516
1517         if (clone_flags & CLONE_SIGHAND) {
1518                 refcount_inc(&current->sighand->count);
1519                 return 0;
1520         }
1521         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1522         RCU_INIT_POINTER(tsk->sighand, sig);
1523         if (!sig)
1524                 return -ENOMEM;
1525
1526         refcount_set(&sig->count, 1);
1527         spin_lock_irq(&current->sighand->siglock);
1528         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1529         spin_unlock_irq(&current->sighand->siglock);
1530
1531         /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1532         if (clone_flags & CLONE_CLEAR_SIGHAND)
1533                 flush_signal_handlers(tsk, 0);
1534
1535         return 0;
1536 }
1537
1538 void __cleanup_sighand(struct sighand_struct *sighand)
1539 {
1540         if (refcount_dec_and_test(&sighand->count)) {
1541                 signalfd_cleanup(sighand);
1542                 /*
1543                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1544                  * without an RCU grace period, see __lock_task_sighand().
1545                  */
1546                 kmem_cache_free(sighand_cachep, sighand);
1547         }
1548 }
1549
1550 /*
1551  * Initialize POSIX timer handling for a thread group.
1552  */
1553 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1554 {
1555         struct posix_cputimers *pct = &sig->posix_cputimers;
1556         unsigned long cpu_limit;
1557
1558         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1559         posix_cputimers_group_init(pct, cpu_limit);
1560 }
1561
1562 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1563 {
1564         struct signal_struct *sig;
1565
1566         if (clone_flags & CLONE_THREAD)
1567                 return 0;
1568
1569         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1570         tsk->signal = sig;
1571         if (!sig)
1572                 return -ENOMEM;
1573
1574         sig->nr_threads = 1;
1575         atomic_set(&sig->live, 1);
1576         refcount_set(&sig->sigcnt, 1);
1577
1578         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1579         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1580         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1581
1582         init_waitqueue_head(&sig->wait_chldexit);
1583         sig->curr_target = tsk;
1584         init_sigpending(&sig->shared_pending);
1585         INIT_HLIST_HEAD(&sig->multiprocess);
1586         seqlock_init(&sig->stats_lock);
1587         prev_cputime_init(&sig->prev_cputime);
1588
1589 #ifdef CONFIG_POSIX_TIMERS
1590         INIT_LIST_HEAD(&sig->posix_timers);
1591         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1592         sig->real_timer.function = it_real_fn;
1593 #endif
1594
1595         task_lock(current->group_leader);
1596         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1597         task_unlock(current->group_leader);
1598
1599         posix_cpu_timers_init_group(sig);
1600
1601         tty_audit_fork(sig);
1602         sched_autogroup_fork(sig);
1603
1604         sig->oom_score_adj = current->signal->oom_score_adj;
1605         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1606
1607         mutex_init(&sig->cred_guard_mutex);
1608         mutex_init(&sig->exec_update_mutex);
1609
1610         return 0;
1611 }
1612
1613 static void copy_seccomp(struct task_struct *p)
1614 {
1615 #ifdef CONFIG_SECCOMP
1616         /*
1617          * Must be called with sighand->lock held, which is common to
1618          * all threads in the group. Holding cred_guard_mutex is not
1619          * needed because this new task is not yet running and cannot
1620          * be racing exec.
1621          */
1622         assert_spin_locked(&current->sighand->siglock);
1623
1624         /* Ref-count the new filter user, and assign it. */
1625         get_seccomp_filter(current);
1626         p->seccomp = current->seccomp;
1627
1628         /*
1629          * Explicitly enable no_new_privs here in case it got set
1630          * between the task_struct being duplicated and holding the
1631          * sighand lock. The seccomp state and nnp must be in sync.
1632          */
1633         if (task_no_new_privs(current))
1634                 task_set_no_new_privs(p);
1635
1636         /*
1637          * If the parent gained a seccomp mode after copying thread
1638          * flags and between before we held the sighand lock, we have
1639          * to manually enable the seccomp thread flag here.
1640          */
1641         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1642                 set_tsk_thread_flag(p, TIF_SECCOMP);
1643 #endif
1644 }
1645
1646 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1647 {
1648         current->clear_child_tid = tidptr;
1649
1650         return task_pid_vnr(current);
1651 }
1652
1653 static void rt_mutex_init_task(struct task_struct *p)
1654 {
1655         raw_spin_lock_init(&p->pi_lock);
1656 #ifdef CONFIG_RT_MUTEXES
1657         p->pi_waiters = RB_ROOT_CACHED;
1658         p->pi_top_task = NULL;
1659         p->pi_blocked_on = NULL;
1660 #endif
1661 }
1662
1663 static inline void init_task_pid_links(struct task_struct *task)
1664 {
1665         enum pid_type type;
1666
1667         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1668                 INIT_HLIST_NODE(&task->pid_links[type]);
1669         }
1670 }
1671
1672 static inline void
1673 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1674 {
1675         if (type == PIDTYPE_PID)
1676                 task->thread_pid = pid;
1677         else
1678                 task->signal->pids[type] = pid;
1679 }
1680
1681 static inline void rcu_copy_process(struct task_struct *p)
1682 {
1683 #ifdef CONFIG_PREEMPT_RCU
1684         p->rcu_read_lock_nesting = 0;
1685         p->rcu_read_unlock_special.s = 0;
1686         p->rcu_blocked_node = NULL;
1687         INIT_LIST_HEAD(&p->rcu_node_entry);
1688 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1689 #ifdef CONFIG_TASKS_RCU
1690         p->rcu_tasks_holdout = false;
1691         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1692         p->rcu_tasks_idle_cpu = -1;
1693 #endif /* #ifdef CONFIG_TASKS_RCU */
1694 #ifdef CONFIG_TASKS_TRACE_RCU
1695         p->trc_reader_nesting = 0;
1696         p->trc_reader_special.s = 0;
1697         INIT_LIST_HEAD(&p->trc_holdout_list);
1698 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1699 }
1700
1701 struct pid *pidfd_pid(const struct file *file)
1702 {
1703         if (file->f_op == &pidfd_fops)
1704                 return file->private_data;
1705
1706         return ERR_PTR(-EBADF);
1707 }
1708
1709 static int pidfd_release(struct inode *inode, struct file *file)
1710 {
1711         struct pid *pid = file->private_data;
1712
1713         file->private_data = NULL;
1714         put_pid(pid);
1715         return 0;
1716 }
1717
1718 #ifdef CONFIG_PROC_FS
1719 /**
1720  * pidfd_show_fdinfo - print information about a pidfd
1721  * @m: proc fdinfo file
1722  * @f: file referencing a pidfd
1723  *
1724  * Pid:
1725  * This function will print the pid that a given pidfd refers to in the
1726  * pid namespace of the procfs instance.
1727  * If the pid namespace of the process is not a descendant of the pid
1728  * namespace of the procfs instance 0 will be shown as its pid. This is
1729  * similar to calling getppid() on a process whose parent is outside of
1730  * its pid namespace.
1731  *
1732  * NSpid:
1733  * If pid namespaces are supported then this function will also print
1734  * the pid of a given pidfd refers to for all descendant pid namespaces
1735  * starting from the current pid namespace of the instance, i.e. the
1736  * Pid field and the first entry in the NSpid field will be identical.
1737  * If the pid namespace of the process is not a descendant of the pid
1738  * namespace of the procfs instance 0 will be shown as its first NSpid
1739  * entry and no others will be shown.
1740  * Note that this differs from the Pid and NSpid fields in
1741  * /proc/<pid>/status where Pid and NSpid are always shown relative to
1742  * the  pid namespace of the procfs instance. The difference becomes
1743  * obvious when sending around a pidfd between pid namespaces from a
1744  * different branch of the tree, i.e. where no ancestoral relation is
1745  * present between the pid namespaces:
1746  * - create two new pid namespaces ns1 and ns2 in the initial pid
1747  *   namespace (also take care to create new mount namespaces in the
1748  *   new pid namespace and mount procfs)
1749  * - create a process with a pidfd in ns1
1750  * - send pidfd from ns1 to ns2
1751  * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1752  *   have exactly one entry, which is 0
1753  */
1754 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1755 {
1756         struct pid *pid = f->private_data;
1757         struct pid_namespace *ns;
1758         pid_t nr = -1;
1759
1760         if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1761                 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1762                 nr = pid_nr_ns(pid, ns);
1763         }
1764
1765         seq_put_decimal_ll(m, "Pid:\t", nr);
1766
1767 #ifdef CONFIG_PID_NS
1768         seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1769         if (nr > 0) {
1770                 int i;
1771
1772                 /* If nr is non-zero it means that 'pid' is valid and that
1773                  * ns, i.e. the pid namespace associated with the procfs
1774                  * instance, is in the pid namespace hierarchy of pid.
1775                  * Start at one below the already printed level.
1776                  */
1777                 for (i = ns->level + 1; i <= pid->level; i++)
1778                         seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1779         }
1780 #endif
1781         seq_putc(m, '\n');
1782 }
1783 #endif
1784
1785 /*
1786  * Poll support for process exit notification.
1787  */
1788 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1789 {
1790         struct task_struct *task;
1791         struct pid *pid = file->private_data;
1792         __poll_t poll_flags = 0;
1793
1794         poll_wait(file, &pid->wait_pidfd, pts);
1795
1796         rcu_read_lock();
1797         task = pid_task(pid, PIDTYPE_PID);
1798         /*
1799          * Inform pollers only when the whole thread group exits.
1800          * If the thread group leader exits before all other threads in the
1801          * group, then poll(2) should block, similar to the wait(2) family.
1802          */
1803         if (!task || (task->exit_state && thread_group_empty(task)))
1804                 poll_flags = EPOLLIN | EPOLLRDNORM;
1805         rcu_read_unlock();
1806
1807         return poll_flags;
1808 }
1809
1810 const struct file_operations pidfd_fops = {
1811         .release = pidfd_release,
1812         .poll = pidfd_poll,
1813 #ifdef CONFIG_PROC_FS
1814         .show_fdinfo = pidfd_show_fdinfo,
1815 #endif
1816 };
1817
1818 static void __delayed_free_task(struct rcu_head *rhp)
1819 {
1820         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1821
1822         free_task(tsk);
1823 }
1824
1825 static __always_inline void delayed_free_task(struct task_struct *tsk)
1826 {
1827         if (IS_ENABLED(CONFIG_MEMCG))
1828                 call_rcu(&tsk->rcu, __delayed_free_task);
1829         else
1830                 free_task(tsk);
1831 }
1832
1833 /*
1834  * This creates a new process as a copy of the old one,
1835  * but does not actually start it yet.
1836  *
1837  * It copies the registers, and all the appropriate
1838  * parts of the process environment (as per the clone
1839  * flags). The actual kick-off is left to the caller.
1840  */
1841 static __latent_entropy struct task_struct *copy_process(
1842                                         struct pid *pid,
1843                                         int trace,
1844                                         int node,
1845                                         struct kernel_clone_args *args)
1846 {
1847         int pidfd = -1, retval;
1848         struct task_struct *p;
1849         struct multiprocess_signals delayed;
1850         struct file *pidfile = NULL;
1851         u64 clone_flags = args->flags;
1852         struct nsproxy *nsp = current->nsproxy;
1853
1854         /*
1855          * Don't allow sharing the root directory with processes in a different
1856          * namespace
1857          */
1858         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1859                 return ERR_PTR(-EINVAL);
1860
1861         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1862                 return ERR_PTR(-EINVAL);
1863
1864         /*
1865          * Thread groups must share signals as well, and detached threads
1866          * can only be started up within the thread group.
1867          */
1868         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1869                 return ERR_PTR(-EINVAL);
1870
1871         /*
1872          * Shared signal handlers imply shared VM. By way of the above,
1873          * thread groups also imply shared VM. Blocking this case allows
1874          * for various simplifications in other code.
1875          */
1876         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1877                 return ERR_PTR(-EINVAL);
1878
1879         /*
1880          * Siblings of global init remain as zombies on exit since they are
1881          * not reaped by their parent (swapper). To solve this and to avoid
1882          * multi-rooted process trees, prevent global and container-inits
1883          * from creating siblings.
1884          */
1885         if ((clone_flags & CLONE_PARENT) &&
1886                                 current->signal->flags & SIGNAL_UNKILLABLE)
1887                 return ERR_PTR(-EINVAL);
1888
1889         /*
1890          * If the new process will be in a different pid or user namespace
1891          * do not allow it to share a thread group with the forking task.
1892          */
1893         if (clone_flags & CLONE_THREAD) {
1894                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1895                     (task_active_pid_ns(current) != nsp->pid_ns_for_children))
1896                         return ERR_PTR(-EINVAL);
1897         }
1898
1899         /*
1900          * If the new process will be in a different time namespace
1901          * do not allow it to share VM or a thread group with the forking task.
1902          */
1903         if (clone_flags & (CLONE_THREAD | CLONE_VM)) {
1904                 if (nsp->time_ns != nsp->time_ns_for_children)
1905                         return ERR_PTR(-EINVAL);
1906         }
1907
1908         if (clone_flags & CLONE_PIDFD) {
1909                 /*
1910                  * - CLONE_DETACHED is blocked so that we can potentially
1911                  *   reuse it later for CLONE_PIDFD.
1912                  * - CLONE_THREAD is blocked until someone really needs it.
1913                  */
1914                 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1915                         return ERR_PTR(-EINVAL);
1916         }
1917
1918         /*
1919          * Force any signals received before this point to be delivered
1920          * before the fork happens.  Collect up signals sent to multiple
1921          * processes that happen during the fork and delay them so that
1922          * they appear to happen after the fork.
1923          */
1924         sigemptyset(&delayed.signal);
1925         INIT_HLIST_NODE(&delayed.node);
1926
1927         spin_lock_irq(&current->sighand->siglock);
1928         if (!(clone_flags & CLONE_THREAD))
1929                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1930         recalc_sigpending();
1931         spin_unlock_irq(&current->sighand->siglock);
1932         retval = -ERESTARTNOINTR;
1933         if (signal_pending(current))
1934                 goto fork_out;
1935
1936         retval = -ENOMEM;
1937         p = dup_task_struct(current, node);
1938         if (!p)
1939                 goto fork_out;
1940
1941         /*
1942          * This _must_ happen before we call free_task(), i.e. before we jump
1943          * to any of the bad_fork_* labels. This is to avoid freeing
1944          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1945          * kernel threads (PF_KTHREAD).
1946          */
1947         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1948         /*
1949          * Clear TID on mm_release()?
1950          */
1951         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1952
1953         ftrace_graph_init_task(p);
1954
1955         rt_mutex_init_task(p);
1956
1957 #ifdef CONFIG_PROVE_LOCKING
1958         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1959         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1960 #endif
1961         retval = -EAGAIN;
1962         if (atomic_read(&p->real_cred->user->processes) >=
1963                         task_rlimit(p, RLIMIT_NPROC)) {
1964                 if (p->real_cred->user != INIT_USER &&
1965                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1966                         goto bad_fork_free;
1967         }
1968         current->flags &= ~PF_NPROC_EXCEEDED;
1969
1970         retval = copy_creds(p, clone_flags);
1971         if (retval < 0)
1972                 goto bad_fork_free;
1973
1974         /*
1975          * If multiple threads are within copy_process(), then this check
1976          * triggers too late. This doesn't hurt, the check is only there
1977          * to stop root fork bombs.
1978          */
1979         retval = -EAGAIN;
1980         if (data_race(nr_threads >= max_threads))
1981                 goto bad_fork_cleanup_count;
1982
1983         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1984         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1985         p->flags |= PF_FORKNOEXEC;
1986         INIT_LIST_HEAD(&p->children);
1987         INIT_LIST_HEAD(&p->sibling);
1988         rcu_copy_process(p);
1989         p->vfork_done = NULL;
1990         spin_lock_init(&p->alloc_lock);
1991
1992         init_sigpending(&p->pending);
1993
1994         p->utime = p->stime = p->gtime = 0;
1995 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1996         p->utimescaled = p->stimescaled = 0;
1997 #endif
1998         prev_cputime_init(&p->prev_cputime);
1999
2000 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2001         seqcount_init(&p->vtime.seqcount);
2002         p->vtime.starttime = 0;
2003         p->vtime.state = VTIME_INACTIVE;
2004 #endif
2005
2006 #if defined(SPLIT_RSS_COUNTING)
2007         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2008 #endif
2009
2010         p->default_timer_slack_ns = current->timer_slack_ns;
2011
2012 #ifdef CONFIG_PSI
2013         p->psi_flags = 0;
2014 #endif
2015
2016         task_io_accounting_init(&p->ioac);
2017         acct_clear_integrals(p);
2018
2019         posix_cputimers_init(&p->posix_cputimers);
2020
2021         p->io_context = NULL;
2022         audit_set_context(p, NULL);
2023         cgroup_fork(p);
2024 #ifdef CONFIG_NUMA
2025         p->mempolicy = mpol_dup(p->mempolicy);
2026         if (IS_ERR(p->mempolicy)) {
2027                 retval = PTR_ERR(p->mempolicy);
2028                 p->mempolicy = NULL;
2029                 goto bad_fork_cleanup_threadgroup_lock;
2030         }
2031 #endif
2032 #ifdef CONFIG_CPUSETS
2033         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2034         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2035         seqcount_init(&p->mems_allowed_seq);
2036 #endif
2037 #ifdef CONFIG_TRACE_IRQFLAGS
2038         p->irq_events = 0;
2039         p->hardirqs_enabled = 0;
2040         p->hardirq_enable_ip = 0;
2041         p->hardirq_enable_event = 0;
2042         p->hardirq_disable_ip = _THIS_IP_;
2043         p->hardirq_disable_event = 0;
2044         p->softirqs_enabled = 1;
2045         p->softirq_enable_ip = _THIS_IP_;
2046         p->softirq_enable_event = 0;
2047         p->softirq_disable_ip = 0;
2048         p->softirq_disable_event = 0;
2049         p->hardirq_context = 0;
2050         p->softirq_context = 0;
2051 #endif
2052
2053         p->pagefault_disabled = 0;
2054
2055 #ifdef CONFIG_LOCKDEP
2056         lockdep_init_task(p);
2057 #endif
2058
2059 #ifdef CONFIG_DEBUG_MUTEXES
2060         p->blocked_on = NULL; /* not blocked yet */
2061 #endif
2062 #ifdef CONFIG_BCACHE
2063         p->sequential_io        = 0;
2064         p->sequential_io_avg    = 0;
2065 #endif
2066
2067         /* Perform scheduler related setup. Assign this task to a CPU. */
2068         retval = sched_fork(clone_flags, p);
2069         if (retval)
2070                 goto bad_fork_cleanup_policy;
2071
2072         retval = perf_event_init_task(p);
2073         if (retval)
2074                 goto bad_fork_cleanup_policy;
2075         retval = audit_alloc(p);
2076         if (retval)
2077                 goto bad_fork_cleanup_perf;
2078         /* copy all the process information */
2079         shm_init_task(p);
2080         retval = security_task_alloc(p, clone_flags);
2081         if (retval)
2082                 goto bad_fork_cleanup_audit;
2083         retval = copy_semundo(clone_flags, p);
2084         if (retval)
2085                 goto bad_fork_cleanup_security;
2086         retval = copy_files(clone_flags, p);
2087         if (retval)
2088                 goto bad_fork_cleanup_semundo;
2089         retval = copy_fs(clone_flags, p);
2090         if (retval)
2091                 goto bad_fork_cleanup_files;
2092         retval = copy_sighand(clone_flags, p);
2093         if (retval)
2094                 goto bad_fork_cleanup_fs;
2095         retval = copy_signal(clone_flags, p);
2096         if (retval)
2097                 goto bad_fork_cleanup_sighand;
2098         retval = copy_mm(clone_flags, p);
2099         if (retval)
2100                 goto bad_fork_cleanup_signal;
2101         retval = copy_namespaces(clone_flags, p);
2102         if (retval)
2103                 goto bad_fork_cleanup_mm;
2104         retval = copy_io(clone_flags, p);
2105         if (retval)
2106                 goto bad_fork_cleanup_namespaces;
2107         retval = copy_thread_tls(clone_flags, args->stack, args->stack_size, p,
2108                                  args->tls);
2109         if (retval)
2110                 goto bad_fork_cleanup_io;
2111
2112         stackleak_task_init(p);
2113
2114         if (pid != &init_struct_pid) {
2115                 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2116                                 args->set_tid_size);
2117                 if (IS_ERR(pid)) {
2118                         retval = PTR_ERR(pid);
2119                         goto bad_fork_cleanup_thread;
2120                 }
2121         }
2122
2123         /*
2124          * This has to happen after we've potentially unshared the file
2125          * descriptor table (so that the pidfd doesn't leak into the child
2126          * if the fd table isn't shared).
2127          */
2128         if (clone_flags & CLONE_PIDFD) {
2129                 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2130                 if (retval < 0)
2131                         goto bad_fork_free_pid;
2132
2133                 pidfd = retval;
2134
2135                 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2136                                               O_RDWR | O_CLOEXEC);
2137                 if (IS_ERR(pidfile)) {
2138                         put_unused_fd(pidfd);
2139                         retval = PTR_ERR(pidfile);
2140                         goto bad_fork_free_pid;
2141                 }
2142                 get_pid(pid);   /* held by pidfile now */
2143
2144                 retval = put_user(pidfd, args->pidfd);
2145                 if (retval)
2146                         goto bad_fork_put_pidfd;
2147         }
2148
2149 #ifdef CONFIG_BLOCK
2150         p->plug = NULL;
2151 #endif
2152         futex_init_task(p);
2153
2154         /*
2155          * sigaltstack should be cleared when sharing the same VM
2156          */
2157         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2158                 sas_ss_reset(p);
2159
2160         /*
2161          * Syscall tracing and stepping should be turned off in the
2162          * child regardless of CLONE_PTRACE.
2163          */
2164         user_disable_single_step(p);
2165         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
2166 #ifdef TIF_SYSCALL_EMU
2167         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
2168 #endif
2169         clear_tsk_latency_tracing(p);
2170
2171         /* ok, now we should be set up.. */
2172         p->pid = pid_nr(pid);
2173         if (clone_flags & CLONE_THREAD) {
2174                 p->exit_signal = -1;
2175                 p->group_leader = current->group_leader;
2176                 p->tgid = current->tgid;
2177         } else {
2178                 if (clone_flags & CLONE_PARENT)
2179                         p->exit_signal = current->group_leader->exit_signal;
2180                 else
2181                         p->exit_signal = args->exit_signal;
2182                 p->group_leader = p;
2183                 p->tgid = p->pid;
2184         }
2185
2186         p->nr_dirtied = 0;
2187         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2188         p->dirty_paused_when = 0;
2189
2190         p->pdeath_signal = 0;
2191         INIT_LIST_HEAD(&p->thread_group);
2192         p->task_works = NULL;
2193
2194         /*
2195          * Ensure that the cgroup subsystem policies allow the new process to be
2196          * forked. It should be noted the the new process's css_set can be changed
2197          * between here and cgroup_post_fork() if an organisation operation is in
2198          * progress.
2199          */
2200         retval = cgroup_can_fork(p, args);
2201         if (retval)
2202                 goto bad_fork_put_pidfd;
2203
2204         /*
2205          * From this point on we must avoid any synchronous user-space
2206          * communication until we take the tasklist-lock. In particular, we do
2207          * not want user-space to be able to predict the process start-time by
2208          * stalling fork(2) after we recorded the start_time but before it is
2209          * visible to the system.
2210          */
2211
2212         p->start_time = ktime_get_ns();
2213         p->start_boottime = ktime_get_boottime_ns();
2214
2215         /*
2216          * Make it visible to the rest of the system, but dont wake it up yet.
2217          * Need tasklist lock for parent etc handling!
2218          */
2219         write_lock_irq(&tasklist_lock);
2220
2221         /* CLONE_PARENT re-uses the old parent */
2222         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2223                 p->real_parent = current->real_parent;
2224                 p->parent_exec_id = current->parent_exec_id;
2225         } else {
2226                 p->real_parent = current;
2227                 p->parent_exec_id = current->self_exec_id;
2228         }
2229
2230         klp_copy_process(p);
2231
2232         spin_lock(&current->sighand->siglock);
2233
2234         /*
2235          * Copy seccomp details explicitly here, in case they were changed
2236          * before holding sighand lock.
2237          */
2238         copy_seccomp(p);
2239
2240         rseq_fork(p, clone_flags);
2241
2242         /* Don't start children in a dying pid namespace */
2243         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2244                 retval = -ENOMEM;
2245                 goto bad_fork_cancel_cgroup;
2246         }
2247
2248         /* Let kill terminate clone/fork in the middle */
2249         if (fatal_signal_pending(current)) {
2250                 retval = -EINTR;
2251                 goto bad_fork_cancel_cgroup;
2252         }
2253
2254         /* past the last point of failure */
2255         if (pidfile)
2256                 fd_install(pidfd, pidfile);
2257
2258         init_task_pid_links(p);
2259         if (likely(p->pid)) {
2260                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2261
2262                 init_task_pid(p, PIDTYPE_PID, pid);
2263                 if (thread_group_leader(p)) {
2264                         init_task_pid(p, PIDTYPE_TGID, pid);
2265                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2266                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2267
2268                         if (is_child_reaper(pid)) {
2269                                 ns_of_pid(pid)->child_reaper = p;
2270                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2271                         }
2272                         p->signal->shared_pending.signal = delayed.signal;
2273                         p->signal->tty = tty_kref_get(current->signal->tty);
2274                         /*
2275                          * Inherit has_child_subreaper flag under the same
2276                          * tasklist_lock with adding child to the process tree
2277                          * for propagate_has_child_subreaper optimization.
2278                          */
2279                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2280                                                          p->real_parent->signal->is_child_subreaper;
2281                         list_add_tail(&p->sibling, &p->real_parent->children);
2282                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2283                         attach_pid(p, PIDTYPE_TGID);
2284                         attach_pid(p, PIDTYPE_PGID);
2285                         attach_pid(p, PIDTYPE_SID);
2286                         __this_cpu_inc(process_counts);
2287                 } else {
2288                         current->signal->nr_threads++;
2289                         atomic_inc(&current->signal->live);
2290                         refcount_inc(&current->signal->sigcnt);
2291                         task_join_group_stop(p);
2292                         list_add_tail_rcu(&p->thread_group,
2293                                           &p->group_leader->thread_group);
2294                         list_add_tail_rcu(&p->thread_node,
2295                                           &p->signal->thread_head);
2296                 }
2297                 attach_pid(p, PIDTYPE_PID);
2298                 nr_threads++;
2299         }
2300         total_forks++;
2301         hlist_del_init(&delayed.node);
2302         spin_unlock(&current->sighand->siglock);
2303         syscall_tracepoint_update(p);
2304         write_unlock_irq(&tasklist_lock);
2305
2306         proc_fork_connector(p);
2307         cgroup_post_fork(p, args);
2308         perf_event_fork(p);
2309
2310         trace_task_newtask(p, clone_flags);
2311         uprobe_copy_process(p, clone_flags);
2312
2313         return p;
2314
2315 bad_fork_cancel_cgroup:
2316         spin_unlock(&current->sighand->siglock);
2317         write_unlock_irq(&tasklist_lock);
2318         cgroup_cancel_fork(p, args);
2319 bad_fork_put_pidfd:
2320         if (clone_flags & CLONE_PIDFD) {
2321                 fput(pidfile);
2322                 put_unused_fd(pidfd);
2323         }
2324 bad_fork_free_pid:
2325         if (pid != &init_struct_pid)
2326                 free_pid(pid);
2327 bad_fork_cleanup_thread:
2328         exit_thread(p);
2329 bad_fork_cleanup_io:
2330         if (p->io_context)
2331                 exit_io_context(p);
2332 bad_fork_cleanup_namespaces:
2333         exit_task_namespaces(p);
2334 bad_fork_cleanup_mm:
2335         if (p->mm) {
2336                 mm_clear_owner(p->mm, p);
2337                 mmput(p->mm);
2338         }
2339 bad_fork_cleanup_signal:
2340         if (!(clone_flags & CLONE_THREAD))
2341                 free_signal_struct(p->signal);
2342 bad_fork_cleanup_sighand:
2343         __cleanup_sighand(p->sighand);
2344 bad_fork_cleanup_fs:
2345         exit_fs(p); /* blocking */
2346 bad_fork_cleanup_files:
2347         exit_files(p); /* blocking */
2348 bad_fork_cleanup_semundo:
2349         exit_sem(p);
2350 bad_fork_cleanup_security:
2351         security_task_free(p);
2352 bad_fork_cleanup_audit:
2353         audit_free(p);
2354 bad_fork_cleanup_perf:
2355         perf_event_free_task(p);
2356 bad_fork_cleanup_policy:
2357         lockdep_free_task(p);
2358 #ifdef CONFIG_NUMA
2359         mpol_put(p->mempolicy);
2360 bad_fork_cleanup_threadgroup_lock:
2361 #endif
2362         delayacct_tsk_free(p);
2363 bad_fork_cleanup_count:
2364         atomic_dec(&p->cred->user->processes);
2365         exit_creds(p);
2366 bad_fork_free:
2367         p->state = TASK_DEAD;
2368         put_task_stack(p);
2369         delayed_free_task(p);
2370 fork_out:
2371         spin_lock_irq(&current->sighand->siglock);
2372         hlist_del_init(&delayed.node);
2373         spin_unlock_irq(&current->sighand->siglock);
2374         return ERR_PTR(retval);
2375 }
2376
2377 static inline void init_idle_pids(struct task_struct *idle)
2378 {
2379         enum pid_type type;
2380
2381         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2382                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2383                 init_task_pid(idle, type, &init_struct_pid);
2384         }
2385 }
2386
2387 struct task_struct *fork_idle(int cpu)
2388 {
2389         struct task_struct *task;
2390         struct kernel_clone_args args = {
2391                 .flags = CLONE_VM,
2392         };
2393
2394         task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2395         if (!IS_ERR(task)) {
2396                 init_idle_pids(task);
2397                 init_idle(task, cpu);
2398         }
2399
2400         return task;
2401 }
2402
2403 struct mm_struct *copy_init_mm(void)
2404 {
2405         return dup_mm(NULL, &init_mm);
2406 }
2407
2408 /*
2409  *  Ok, this is the main fork-routine.
2410  *
2411  * It copies the process, and if successful kick-starts
2412  * it and waits for it to finish using the VM if required.
2413  *
2414  * args->exit_signal is expected to be checked for sanity by the caller.
2415  */
2416 long _do_fork(struct kernel_clone_args *args)
2417 {
2418         u64 clone_flags = args->flags;
2419         struct completion vfork;
2420         struct pid *pid;
2421         struct task_struct *p;
2422         int trace = 0;
2423         long nr;
2424
2425         /*
2426          * Determine whether and which event to report to ptracer.  When
2427          * called from kernel_thread or CLONE_UNTRACED is explicitly
2428          * requested, no event is reported; otherwise, report if the event
2429          * for the type of forking is enabled.
2430          */
2431         if (!(clone_flags & CLONE_UNTRACED)) {
2432                 if (clone_flags & CLONE_VFORK)
2433                         trace = PTRACE_EVENT_VFORK;
2434                 else if (args->exit_signal != SIGCHLD)
2435                         trace = PTRACE_EVENT_CLONE;
2436                 else
2437                         trace = PTRACE_EVENT_FORK;
2438
2439                 if (likely(!ptrace_event_enabled(current, trace)))
2440                         trace = 0;
2441         }
2442
2443         p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2444         add_latent_entropy();
2445
2446         if (IS_ERR(p))
2447                 return PTR_ERR(p);
2448
2449         /*
2450          * Do this prior waking up the new thread - the thread pointer
2451          * might get invalid after that point, if the thread exits quickly.
2452          */
2453         trace_sched_process_fork(current, p);
2454
2455         pid = get_task_pid(p, PIDTYPE_PID);
2456         nr = pid_vnr(pid);
2457
2458         if (clone_flags & CLONE_PARENT_SETTID)
2459                 put_user(nr, args->parent_tid);
2460
2461         if (clone_flags & CLONE_VFORK) {
2462                 p->vfork_done = &vfork;
2463                 init_completion(&vfork);
2464                 get_task_struct(p);
2465         }
2466
2467         wake_up_new_task(p);
2468
2469         /* forking complete and child started to run, tell ptracer */
2470         if (unlikely(trace))
2471                 ptrace_event_pid(trace, pid);
2472
2473         if (clone_flags & CLONE_VFORK) {
2474                 if (!wait_for_vfork_done(p, &vfork))
2475                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2476         }
2477
2478         put_pid(pid);
2479         return nr;
2480 }
2481
2482 bool legacy_clone_args_valid(const struct kernel_clone_args *kargs)
2483 {
2484         /* clone(CLONE_PIDFD) uses parent_tidptr to return a pidfd */
2485         if ((kargs->flags & CLONE_PIDFD) &&
2486             (kargs->flags & CLONE_PARENT_SETTID))
2487                 return false;
2488
2489         return true;
2490 }
2491
2492 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2493 /* For compatibility with architectures that call do_fork directly rather than
2494  * using the syscall entry points below. */
2495 long do_fork(unsigned long clone_flags,
2496               unsigned long stack_start,
2497               unsigned long stack_size,
2498               int __user *parent_tidptr,
2499               int __user *child_tidptr)
2500 {
2501         struct kernel_clone_args args = {
2502                 .flags          = (lower_32_bits(clone_flags) & ~CSIGNAL),
2503                 .pidfd          = parent_tidptr,
2504                 .child_tid      = child_tidptr,
2505                 .parent_tid     = parent_tidptr,
2506                 .exit_signal    = (lower_32_bits(clone_flags) & CSIGNAL),
2507                 .stack          = stack_start,
2508                 .stack_size     = stack_size,
2509         };
2510
2511         if (!legacy_clone_args_valid(&args))
2512                 return -EINVAL;
2513
2514         return _do_fork(&args);
2515 }
2516 #endif
2517
2518 /*
2519  * Create a kernel thread.
2520  */
2521 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2522 {
2523         struct kernel_clone_args args = {
2524                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2525                                     CLONE_UNTRACED) & ~CSIGNAL),
2526                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2527                 .stack          = (unsigned long)fn,
2528                 .stack_size     = (unsigned long)arg,
2529         };
2530
2531         return _do_fork(&args);
2532 }
2533
2534 #ifdef __ARCH_WANT_SYS_FORK
2535 SYSCALL_DEFINE0(fork)
2536 {
2537 #ifdef CONFIG_MMU
2538         struct kernel_clone_args args = {
2539                 .exit_signal = SIGCHLD,
2540         };
2541
2542         return _do_fork(&args);
2543 #else
2544         /* can not support in nommu mode */
2545         return -EINVAL;
2546 #endif
2547 }
2548 #endif
2549
2550 #ifdef __ARCH_WANT_SYS_VFORK
2551 SYSCALL_DEFINE0(vfork)
2552 {
2553         struct kernel_clone_args args = {
2554                 .flags          = CLONE_VFORK | CLONE_VM,
2555                 .exit_signal    = SIGCHLD,
2556         };
2557
2558         return _do_fork(&args);
2559 }
2560 #endif
2561
2562 #ifdef __ARCH_WANT_SYS_CLONE
2563 #ifdef CONFIG_CLONE_BACKWARDS
2564 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2565                  int __user *, parent_tidptr,
2566                  unsigned long, tls,
2567                  int __user *, child_tidptr)
2568 #elif defined(CONFIG_CLONE_BACKWARDS2)
2569 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2570                  int __user *, parent_tidptr,
2571                  int __user *, child_tidptr,
2572                  unsigned long, tls)
2573 #elif defined(CONFIG_CLONE_BACKWARDS3)
2574 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2575                 int, stack_size,
2576                 int __user *, parent_tidptr,
2577                 int __user *, child_tidptr,
2578                 unsigned long, tls)
2579 #else
2580 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2581                  int __user *, parent_tidptr,
2582                  int __user *, child_tidptr,
2583                  unsigned long, tls)
2584 #endif
2585 {
2586         struct kernel_clone_args args = {
2587                 .flags          = (lower_32_bits(clone_flags) & ~CSIGNAL),
2588                 .pidfd          = parent_tidptr,
2589                 .child_tid      = child_tidptr,
2590                 .parent_tid     = parent_tidptr,
2591                 .exit_signal    = (lower_32_bits(clone_flags) & CSIGNAL),
2592                 .stack          = newsp,
2593                 .tls            = tls,
2594         };
2595
2596         if (!legacy_clone_args_valid(&args))
2597                 return -EINVAL;
2598
2599         return _do_fork(&args);
2600 }
2601 #endif
2602
2603 #ifdef __ARCH_WANT_SYS_CLONE3
2604
2605 /*
2606  * copy_thread implementations handle CLONE_SETTLS by reading the TLS value from
2607  * the registers containing the syscall arguments for clone. This doesn't work
2608  * with clone3 since the TLS value is passed in clone_args instead.
2609  */
2610 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2611 #error clone3 requires copy_thread_tls support in arch
2612 #endif
2613
2614 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2615                                               struct clone_args __user *uargs,
2616                                               size_t usize)
2617 {
2618         int err;
2619         struct clone_args args;
2620         pid_t *kset_tid = kargs->set_tid;
2621
2622         BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2623                      CLONE_ARGS_SIZE_VER0);
2624         BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2625                      CLONE_ARGS_SIZE_VER1);
2626         BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2627                      CLONE_ARGS_SIZE_VER2);
2628         BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2629
2630         if (unlikely(usize > PAGE_SIZE))
2631                 return -E2BIG;
2632         if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2633                 return -EINVAL;
2634
2635         err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2636         if (err)
2637                 return err;
2638
2639         if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2640                 return -EINVAL;
2641
2642         if (unlikely(!args.set_tid && args.set_tid_size > 0))
2643                 return -EINVAL;
2644
2645         if (unlikely(args.set_tid && args.set_tid_size == 0))
2646                 return -EINVAL;
2647
2648         /*
2649          * Verify that higher 32bits of exit_signal are unset and that
2650          * it is a valid signal
2651          */
2652         if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2653                      !valid_signal(args.exit_signal)))
2654                 return -EINVAL;
2655
2656         if ((args.flags & CLONE_INTO_CGROUP) &&
2657             (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2658                 return -EINVAL;
2659
2660         *kargs = (struct kernel_clone_args){
2661                 .flags          = args.flags,
2662                 .pidfd          = u64_to_user_ptr(args.pidfd),
2663                 .child_tid      = u64_to_user_ptr(args.child_tid),
2664                 .parent_tid     = u64_to_user_ptr(args.parent_tid),
2665                 .exit_signal    = args.exit_signal,
2666                 .stack          = args.stack,
2667                 .stack_size     = args.stack_size,
2668                 .tls            = args.tls,
2669                 .set_tid_size   = args.set_tid_size,
2670                 .cgroup         = args.cgroup,
2671         };
2672
2673         if (args.set_tid &&
2674                 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2675                         (kargs->set_tid_size * sizeof(pid_t))))
2676                 return -EFAULT;
2677
2678         kargs->set_tid = kset_tid;
2679
2680         return 0;
2681 }
2682
2683 /**
2684  * clone3_stack_valid - check and prepare stack
2685  * @kargs: kernel clone args
2686  *
2687  * Verify that the stack arguments userspace gave us are sane.
2688  * In addition, set the stack direction for userspace since it's easy for us to
2689  * determine.
2690  */
2691 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2692 {
2693         if (kargs->stack == 0) {
2694                 if (kargs->stack_size > 0)
2695                         return false;
2696         } else {
2697                 if (kargs->stack_size == 0)
2698                         return false;
2699
2700                 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2701                         return false;
2702
2703 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2704                 kargs->stack += kargs->stack_size;
2705 #endif
2706         }
2707
2708         return true;
2709 }
2710
2711 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2712 {
2713         /* Verify that no unknown flags are passed along. */
2714         if (kargs->flags &
2715             ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2716                 return false;
2717
2718         /*
2719          * - make the CLONE_DETACHED bit reuseable for clone3
2720          * - make the CSIGNAL bits reuseable for clone3
2721          */
2722         if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2723                 return false;
2724
2725         if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2726             (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2727                 return false;
2728
2729         if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2730             kargs->exit_signal)
2731                 return false;
2732
2733         if (!clone3_stack_valid(kargs))
2734                 return false;
2735
2736         return true;
2737 }
2738
2739 /**
2740  * clone3 - create a new process with specific properties
2741  * @uargs: argument structure
2742  * @size:  size of @uargs
2743  *
2744  * clone3() is the extensible successor to clone()/clone2().
2745  * It takes a struct as argument that is versioned by its size.
2746  *
2747  * Return: On success, a positive PID for the child process.
2748  *         On error, a negative errno number.
2749  */
2750 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2751 {
2752         int err;
2753
2754         struct kernel_clone_args kargs;
2755         pid_t set_tid[MAX_PID_NS_LEVEL];
2756
2757         kargs.set_tid = set_tid;
2758
2759         err = copy_clone_args_from_user(&kargs, uargs, size);
2760         if (err)
2761                 return err;
2762
2763         if (!clone3_args_valid(&kargs))
2764                 return -EINVAL;
2765
2766         return _do_fork(&kargs);
2767 }
2768 #endif
2769
2770 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2771 {
2772         struct task_struct *leader, *parent, *child;
2773         int res;
2774
2775         read_lock(&tasklist_lock);
2776         leader = top = top->group_leader;
2777 down:
2778         for_each_thread(leader, parent) {
2779                 list_for_each_entry(child, &parent->children, sibling) {
2780                         res = visitor(child, data);
2781                         if (res) {
2782                                 if (res < 0)
2783                                         goto out;
2784                                 leader = child;
2785                                 goto down;
2786                         }
2787 up:
2788                         ;
2789                 }
2790         }
2791
2792         if (leader != top) {
2793                 child = leader;
2794                 parent = child->real_parent;
2795                 leader = parent->group_leader;
2796                 goto up;
2797         }
2798 out:
2799         read_unlock(&tasklist_lock);
2800 }
2801
2802 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2803 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2804 #endif
2805
2806 static void sighand_ctor(void *data)
2807 {
2808         struct sighand_struct *sighand = data;
2809
2810         spin_lock_init(&sighand->siglock);
2811         init_waitqueue_head(&sighand->signalfd_wqh);
2812 }
2813
2814 void __init proc_caches_init(void)
2815 {
2816         unsigned int mm_size;
2817
2818         sighand_cachep = kmem_cache_create("sighand_cache",
2819                         sizeof(struct sighand_struct), 0,
2820                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2821                         SLAB_ACCOUNT, sighand_ctor);
2822         signal_cachep = kmem_cache_create("signal_cache",
2823                         sizeof(struct signal_struct), 0,
2824                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2825                         NULL);
2826         files_cachep = kmem_cache_create("files_cache",
2827                         sizeof(struct files_struct), 0,
2828                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2829                         NULL);
2830         fs_cachep = kmem_cache_create("fs_cache",
2831                         sizeof(struct fs_struct), 0,
2832                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2833                         NULL);
2834
2835         /*
2836          * The mm_cpumask is located at the end of mm_struct, and is
2837          * dynamically sized based on the maximum CPU number this system
2838          * can have, taking hotplug into account (nr_cpu_ids).
2839          */
2840         mm_size = sizeof(struct mm_struct) + cpumask_size();
2841
2842         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2843                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2844                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2845                         offsetof(struct mm_struct, saved_auxv),
2846                         sizeof_field(struct mm_struct, saved_auxv),
2847                         NULL);
2848         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2849         mmap_init();
2850         nsproxy_cache_init();
2851 }
2852
2853 /*
2854  * Check constraints on flags passed to the unshare system call.
2855  */
2856 static int check_unshare_flags(unsigned long unshare_flags)
2857 {
2858         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2859                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2860                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2861                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
2862                                 CLONE_NEWTIME))
2863                 return -EINVAL;
2864         /*
2865          * Not implemented, but pretend it works if there is nothing
2866          * to unshare.  Note that unsharing the address space or the
2867          * signal handlers also need to unshare the signal queues (aka
2868          * CLONE_THREAD).
2869          */
2870         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2871                 if (!thread_group_empty(current))
2872                         return -EINVAL;
2873         }
2874         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2875                 if (refcount_read(&current->sighand->count) > 1)
2876                         return -EINVAL;
2877         }
2878         if (unshare_flags & CLONE_VM) {
2879                 if (!current_is_single_threaded())
2880                         return -EINVAL;
2881         }
2882
2883         return 0;
2884 }
2885
2886 /*
2887  * Unshare the filesystem structure if it is being shared
2888  */
2889 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2890 {
2891         struct fs_struct *fs = current->fs;
2892
2893         if (!(unshare_flags & CLONE_FS) || !fs)
2894                 return 0;
2895
2896         /* don't need lock here; in the worst case we'll do useless copy */
2897         if (fs->users == 1)
2898                 return 0;
2899
2900         *new_fsp = copy_fs_struct(fs);
2901         if (!*new_fsp)
2902                 return -ENOMEM;
2903
2904         return 0;
2905 }
2906
2907 /*
2908  * Unshare file descriptor table if it is being shared
2909  */
2910 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2911 {
2912         struct files_struct *fd = current->files;
2913         int error = 0;
2914
2915         if ((unshare_flags & CLONE_FILES) &&
2916             (fd && atomic_read(&fd->count) > 1)) {
2917                 *new_fdp = dup_fd(fd, &error);
2918                 if (!*new_fdp)
2919                         return error;
2920         }
2921
2922         return 0;
2923 }
2924
2925 /*
2926  * unshare allows a process to 'unshare' part of the process
2927  * context which was originally shared using clone.  copy_*
2928  * functions used by do_fork() cannot be used here directly
2929  * because they modify an inactive task_struct that is being
2930  * constructed. Here we are modifying the current, active,
2931  * task_struct.
2932  */
2933 int ksys_unshare(unsigned long unshare_flags)
2934 {
2935         struct fs_struct *fs, *new_fs = NULL;
2936         struct files_struct *fd, *new_fd = NULL;
2937         struct cred *new_cred = NULL;
2938         struct nsproxy *new_nsproxy = NULL;
2939         int do_sysvsem = 0;
2940         int err;
2941
2942         /*
2943          * If unsharing a user namespace must also unshare the thread group
2944          * and unshare the filesystem root and working directories.
2945          */
2946         if (unshare_flags & CLONE_NEWUSER)
2947                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2948         /*
2949          * If unsharing vm, must also unshare signal handlers.
2950          */
2951         if (unshare_flags & CLONE_VM)
2952                 unshare_flags |= CLONE_SIGHAND;
2953         /*
2954          * If unsharing a signal handlers, must also unshare the signal queues.
2955          */
2956         if (unshare_flags & CLONE_SIGHAND)
2957                 unshare_flags |= CLONE_THREAD;
2958         /*
2959          * If unsharing namespace, must also unshare filesystem information.
2960          */
2961         if (unshare_flags & CLONE_NEWNS)
2962                 unshare_flags |= CLONE_FS;
2963
2964         err = check_unshare_flags(unshare_flags);
2965         if (err)
2966                 goto bad_unshare_out;
2967         /*
2968          * CLONE_NEWIPC must also detach from the undolist: after switching
2969          * to a new ipc namespace, the semaphore arrays from the old
2970          * namespace are unreachable.
2971          */
2972         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2973                 do_sysvsem = 1;
2974         err = unshare_fs(unshare_flags, &new_fs);
2975         if (err)
2976                 goto bad_unshare_out;
2977         err = unshare_fd(unshare_flags, &new_fd);
2978         if (err)
2979                 goto bad_unshare_cleanup_fs;
2980         err = unshare_userns(unshare_flags, &new_cred);
2981         if (err)
2982                 goto bad_unshare_cleanup_fd;
2983         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2984                                          new_cred, new_fs);
2985         if (err)
2986                 goto bad_unshare_cleanup_cred;
2987
2988         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2989                 if (do_sysvsem) {
2990                         /*
2991                          * CLONE_SYSVSEM is equivalent to sys_exit().
2992                          */
2993                         exit_sem(current);
2994                 }
2995                 if (unshare_flags & CLONE_NEWIPC) {
2996                         /* Orphan segments in old ns (see sem above). */
2997                         exit_shm(current);
2998                         shm_init_task(current);
2999                 }
3000
3001                 if (new_nsproxy)
3002                         switch_task_namespaces(current, new_nsproxy);
3003
3004                 task_lock(current);
3005
3006                 if (new_fs) {
3007                         fs = current->fs;
3008                         spin_lock(&fs->lock);
3009                         current->fs = new_fs;
3010                         if (--fs->users)
3011                                 new_fs = NULL;
3012                         else
3013                                 new_fs = fs;
3014                         spin_unlock(&fs->lock);
3015                 }
3016
3017                 if (new_fd) {
3018                         fd = current->files;
3019                         current->files = new_fd;
3020                         new_fd = fd;
3021                 }
3022
3023                 task_unlock(current);
3024
3025                 if (new_cred) {
3026                         /* Install the new user namespace */
3027                         commit_creds(new_cred);
3028                         new_cred = NULL;
3029                 }
3030         }
3031
3032         perf_event_namespaces(current);
3033
3034 bad_unshare_cleanup_cred:
3035         if (new_cred)
3036                 put_cred(new_cred);
3037 bad_unshare_cleanup_fd:
3038         if (new_fd)
3039                 put_files_struct(new_fd);
3040
3041 bad_unshare_cleanup_fs:
3042         if (new_fs)
3043                 free_fs_struct(new_fs);
3044
3045 bad_unshare_out:
3046         return err;
3047 }
3048
3049 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3050 {
3051         return ksys_unshare(unshare_flags);
3052 }
3053
3054 /*
3055  *      Helper to unshare the files of the current task.
3056  *      We don't want to expose copy_files internals to
3057  *      the exec layer of the kernel.
3058  */
3059
3060 int unshare_files(struct files_struct **displaced)
3061 {
3062         struct task_struct *task = current;
3063         struct files_struct *copy = NULL;
3064         int error;
3065
3066         error = unshare_fd(CLONE_FILES, &copy);
3067         if (error || !copy) {
3068                 *displaced = NULL;
3069                 return error;
3070         }
3071         *displaced = task->files;
3072         task_lock(task);
3073         task->files = copy;
3074         task_unlock(task);
3075         return 0;
3076 }
3077
3078 int sysctl_max_threads(struct ctl_table *table, int write,
3079                        void __user *buffer, size_t *lenp, loff_t *ppos)
3080 {
3081         struct ctl_table t;
3082         int ret;
3083         int threads = max_threads;
3084         int min = 1;
3085         int max = MAX_THREADS;
3086
3087         t = *table;
3088         t.data = &threads;
3089         t.extra1 = &min;
3090         t.extra2 = &max;
3091
3092         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3093         if (ret || !write)
3094                 return ret;
3095
3096         max_threads = threads;
3097
3098         return 0;
3099 }