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/mm_inline.h>
  46 #include <linux/vmacache.h>
  47 #include <linux/nsproxy.h>
  48 #include <linux/capability.h>
  49 #include <linux/cpu.h>
  50 #include <linux/cgroup.h>
  51 #include <linux/security.h>
  52 #include <linux/hugetlb.h>
  53 #include <linux/seccomp.h>
  54 #include <linux/swap.h>
  55 #include <linux/syscalls.h>
  56 #include <linux/jiffies.h>
  57 #include <linux/futex.h>
  58 #include <linux/compat.h>
  59 #include <linux/kthread.h>
  60 #include <linux/task_io_accounting_ops.h>
  61 #include <linux/rcupdate.h>
  62 #include <linux/ptrace.h>
  63 #include <linux/mount.h>
  64 #include <linux/audit.h>
  65 #include <linux/memcontrol.h>
  66 #include <linux/ftrace.h>
  67 #include <linux/proc_fs.h>
  68 #include <linux/profile.h>
  69 #include <linux/rmap.h>
  70 #include <linux/ksm.h>
  71 #include <linux/acct.h>
  72 #include <linux/userfaultfd_k.h>
  73 #include <linux/tsacct_kern.h>
  74 #include <linux/cn_proc.h>
  75 #include <linux/freezer.h>
  76 #include <linux/delayacct.h>
  77 #include <linux/taskstats_kern.h>
  78 #include <linux/random.h>
  79 #include <linux/tty.h>
  80 #include <linux/fs_struct.h>
  81 #include <linux/magic.h>
  82 #include <linux/perf_event.h>
  83 #include <linux/posix-timers.h>
  84 #include <linux/user-return-notifier.h>
  85 #include <linux/oom.h>
  86 #include <linux/khugepaged.h>
  87 #include <linux/signalfd.h>
  88 #include <linux/uprobes.h>
  89 #include <linux/aio.h>
  90 #include <linux/compiler.h>
  91 #include <linux/sysctl.h>
  92 #include <linux/kcov.h>
  93 #include <linux/livepatch.h>
  94 #include <linux/thread_info.h>
  95 #include <linux/stackleak.h>
  96 #include <linux/kasan.h>
  97 #include <linux/scs.h>
  98 #include <linux/io_uring.h>
  99 #include <linux/bpf.h>
 100
 101 #include <asm/pgalloc.h>
 102 #include <linux/uaccess.h>
 103 #include <asm/mmu_context.h>
 104 #include <asm/cacheflush.h>
 105 #include <asm/tlbflush.h>
 106
 107 #include <trace/events/sched.h>
 108
 109 #define CREATE_TRACE_POINTS
 110 #include <trace/events/task.h>
 111
 112 /*
 113  * Minimum number of threads to boot the kernel
 114  */
 115 #define MIN_THREADS 20
 116
 117 /*
 118  * Maximum number of threads
 119  */
 120 #define MAX_THREADS FUTEX_TID_MASK
 121
 122 /*
 123  * Protected counters by write_lock_irq(&tasklist_lock)
 124  */
 125 unsigned long total_forks;      /* Handle normal Linux uptimes. */
 126 int nr_threads;                 /* The idle threads do not count.. */
 127
 128 static int max_threads;         /* tunable limit on nr_threads */
 129
 130 #define NAMED_ARRAY_INDEX(x)    [x] = __stringify(x)
 131
 132 static const char * const resident_page_types[] = {
 133         NAMED_ARRAY_INDEX(MM_FILEPAGES),
 134         NAMED_ARRAY_INDEX(MM_ANONPAGES),
 135         NAMED_ARRAY_INDEX(MM_SWAPENTS),
 136         NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
 137 };
 138
 139 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 140
 141 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 142
 143 #ifdef CONFIG_PROVE_RCU
 144 int lockdep_tasklist_lock_is_held(void)
 145 {
 146         return lockdep_is_held(&tasklist_lock);
 147 }
 148 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
 149 #endif /* #ifdef CONFIG_PROVE_RCU */
 150
 151 int nr_processes(void)
 152 {
 153         int cpu;
 154         int total = 0;
 155
 156         for_each_possible_cpu(cpu)
 157                 total += per_cpu(process_counts, cpu);
 158
 159         return total;
 160 }
 161
 162 void __weak arch_release_task_struct(struct task_struct *tsk)
 163 {
 164 }
 165
 166 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 167 static struct kmem_cache *task_struct_cachep;
 168
 169 static inline struct task_struct *alloc_task_struct_node(int node)
 170 {
 171         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
 172 }
 173
 174 static inline void free_task_struct(struct task_struct *tsk)
 175 {
 176         kmem_cache_free(task_struct_cachep, tsk);
 177 }
 178 #endif
 179
 180 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
 181
 182 /*
 183  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
 184  * kmemcache based allocator.
 185  */
 186 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
 187
 188 #ifdef CONFIG_VMAP_STACK
 189 /*
 190  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
 191  * flush.  Try to minimize the number of calls by caching stacks.
 192  */
 193 #define NR_CACHED_STACKS 2
 194 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 195
 196 static int free_vm_stack_cache(unsigned int cpu)
 197 {
 198         struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
 199         int i;
 200
 201         for (i = 0; i < NR_CACHED_STACKS; i++) {
 202                 struct vm_struct *vm_stack = cached_vm_stacks[i];
 203
 204                 if (!vm_stack)
 205                         continue;
 206
 207                 vfree(vm_stack->addr);
 208                 cached_vm_stacks[i] = NULL;
 209         }
 210
 211         return 0;
 212 }
 213 #endif
 214
 215 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 216 {
 217 #ifdef CONFIG_VMAP_STACK
 218         void *stack;
 219         int i;
 220
 221         for (i = 0; i < NR_CACHED_STACKS; i++) {
 222                 struct vm_struct *s;
 223
 224                 s = this_cpu_xchg(cached_stacks[i], NULL);
 225
 226                 if (!s)
 227                         continue;
 228
 229                 /* Mark stack accessible for KASAN. */
 230                 kasan_unpoison_range(s->addr, THREAD_SIZE);
 231
 232                 /* Clear stale pointers from reused stack. */
 233                 memset(s->addr, 0, THREAD_SIZE);
 234
 235                 tsk->stack_vm_area = s;
 236                 tsk->stack = s->addr;
 237                 return s->addr;
 238         }
 239
 240         /*
 241          * Allocated stacks are cached and later reused by new threads,
 242          * so memcg accounting is performed manually on assigning/releasing
 243          * stacks to tasks. Drop __GFP_ACCOUNT.
 244          */
 245         stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
 246                                      VMALLOC_START, VMALLOC_END,
 247                                      THREADINFO_GFP & ~__GFP_ACCOUNT,
 248                                      PAGE_KERNEL,
 249                                      0, node, __builtin_return_address(0));
 250
 251         /*
 252          * We can't call find_vm_area() in interrupt context, and
 253          * free_thread_stack() can be called in interrupt context,
 254          * so cache the vm_struct.
 255          */
 256         if (stack) {
 257                 tsk->stack_vm_area = find_vm_area(stack);
 258                 tsk->stack = stack;
 259         }
 260         return stack;
 261 #else
 262         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
 263                                              THREAD_SIZE_ORDER);
 264
 265         if (likely(page)) {
 266                 tsk->stack = kasan_reset_tag(page_address(page));
 267                 return tsk->stack;
 268         }
 269         return NULL;
 270 #endif
 271 }
 272
 273 static inline void free_thread_stack(struct task_struct *tsk)
 274 {
 275 #ifdef CONFIG_VMAP_STACK
 276         struct vm_struct *vm = task_stack_vm_area(tsk);
 277
 278         if (vm) {
 279                 int i;
 280
 281                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 282                         memcg_kmem_uncharge_page(vm->pages[i], 0);
 283
 284                 for (i = 0; i < NR_CACHED_STACKS; i++) {
 285                         if (this_cpu_cmpxchg(cached_stacks[i],
 286                                         NULL, tsk->stack_vm_area) != NULL)
 287                                 continue;
 288
 289                         return;
 290                 }
 291
 292                 vfree_atomic(tsk->stack);
 293                 return;
 294         }
 295 #endif
 296
 297         __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
 298 }
 299 # else
 300 static struct kmem_cache *thread_stack_cache;
 301
 302 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
 303                                                   int node)
 304 {
 305         unsigned long *stack;
 306         stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
 307         stack = kasan_reset_tag(stack);
 308         tsk->stack = stack;
 309         return stack;
 310 }
 311
 312 static void free_thread_stack(struct task_struct *tsk)
 313 {
 314         kmem_cache_free(thread_stack_cache, tsk->stack);
 315 }
 316
 317 void thread_stack_cache_init(void)
 318 {
 319         thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
 320                                         THREAD_SIZE, THREAD_SIZE, 0, 0,
 321                                         THREAD_SIZE, NULL);
 322         BUG_ON(thread_stack_cache == NULL);
 323 }
 324 # endif
 325 #endif
 326
 327 /* SLAB cache for signal_struct structures (tsk->signal) */
 328 static struct kmem_cache *signal_cachep;
 329
 330 /* SLAB cache for sighand_struct structures (tsk->sighand) */
 331 struct kmem_cache *sighand_cachep;
 332
 333 /* SLAB cache for files_struct structures (tsk->files) */
 334 struct kmem_cache *files_cachep;
 335
 336 /* SLAB cache for fs_struct structures (tsk->fs) */
 337 struct kmem_cache *fs_cachep;
 338
 339 /* SLAB cache for vm_area_struct structures */
 340 static struct kmem_cache *vm_area_cachep;
 341
 342 /* SLAB cache for mm_struct structures (tsk->mm) */
 343 static struct kmem_cache *mm_cachep;
 344
 345 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
 346 {
 347         struct vm_area_struct *vma;
 348
 349         vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 350         if (vma)
 351                 vma_init(vma, mm);
 352         return vma;
 353 }
 354
 355 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
 356 {
 357         struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 358
 359         if (new) {
 360                 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
 361                 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
 362                 /*
 363                  * orig->shared.rb may be modified concurrently, but the clone
 364                  * will be reinitialized.
 365                  */
 366                 *new = data_race(*orig);
 367                 INIT_LIST_HEAD(&new->anon_vma_chain);
 368                 new->vm_next = new->vm_prev = NULL;
 369                 dup_vma_anon_name(orig, new);
 370         }
 371         return new;
 372 }
 373
 374 void vm_area_free(struct vm_area_struct *vma)
 375 {
 376         free_vma_anon_name(vma);
 377         kmem_cache_free(vm_area_cachep, vma);
 378 }
 379
 380 static void account_kernel_stack(struct task_struct *tsk, int account)
 381 {
 382         void *stack = task_stack_page(tsk);
 383         struct vm_struct *vm = task_stack_vm_area(tsk);
 384
 385         if (vm) {
 386                 int i;
 387
 388                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 389                         mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
 390                                               account * (PAGE_SIZE / 1024));
 391         } else {
 392                 /* All stack pages are in the same node. */
 393                 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
 394                                       account * (THREAD_SIZE / 1024));
 395         }
 396 }
 397
 398 static int memcg_charge_kernel_stack(struct task_struct *tsk)
 399 {
 400 #ifdef CONFIG_VMAP_STACK
 401         struct vm_struct *vm = task_stack_vm_area(tsk);
 402         int ret;
 403
 404         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 405
 406         if (vm) {
 407                 int i;
 408
 409                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 410
 411                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 412                         /*
 413                          * If memcg_kmem_charge_page() fails, page's
 414                          * memory cgroup pointer is NULL, and
 415                          * memcg_kmem_uncharge_page() in free_thread_stack()
 416                          * will ignore this page.
 417                          */
 418                         ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL,
 419                                                      0);
 420                         if (ret)
 421                                 return ret;
 422                 }
 423         }
 424 #endif
 425         return 0;
 426 }
 427
 428 static void release_task_stack(struct task_struct *tsk)
 429 {
 430         if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
 431                 return;  /* Better to leak the stack than to free prematurely */
 432
 433         account_kernel_stack(tsk, -1);
 434         free_thread_stack(tsk);
 435         tsk->stack = NULL;
 436 #ifdef CONFIG_VMAP_STACK
 437         tsk->stack_vm_area = NULL;
 438 #endif
 439 }
 440
 441 #ifdef CONFIG_THREAD_INFO_IN_TASK
 442 void put_task_stack(struct task_struct *tsk)
 443 {
 444         if (refcount_dec_and_test(&tsk->stack_refcount))
 445                 release_task_stack(tsk);
 446 }
 447 #endif
 448
 449 void free_task(struct task_struct *tsk)
 450 {
 451         release_user_cpus_ptr(tsk);
 452         scs_release(tsk);
 453
 454 #ifndef CONFIG_THREAD_INFO_IN_TASK
 455         /*
 456          * The task is finally done with both the stack and thread_info,
 457          * so free both.
 458          */
 459         release_task_stack(tsk);
 460 #else
 461         /*
 462          * If the task had a separate stack allocation, it should be gone
 463          * by now.
 464          */
 465         WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
 466 #endif
 467         rt_mutex_debug_task_free(tsk);
 468         ftrace_graph_exit_task(tsk);
 469         arch_release_task_struct(tsk);
 470         if (tsk->flags & PF_KTHREAD)
 471                 free_kthread_struct(tsk);
 472         free_task_struct(tsk);
 473 }
 474 EXPORT_SYMBOL(free_task);
 475
 476 static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
 477 {
 478         struct file *exe_file;
 479
 480         exe_file = get_mm_exe_file(oldmm);
 481         RCU_INIT_POINTER(mm->exe_file, exe_file);
 482         /*
 483          * We depend on the oldmm having properly denied write access to the
 484          * exe_file already.
 485          */
 486         if (exe_file && deny_write_access(exe_file))
 487                 pr_warn_once("deny_write_access() failed in %s\n", __func__);
 488 }
 489
 490 #ifdef CONFIG_MMU
 491 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 492                                         struct mm_struct *oldmm)
 493 {
 494         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 495         struct rb_node **rb_link, *rb_parent;
 496         int retval;
 497         unsigned long charge;
 498         LIST_HEAD(uf);
 499
 500         uprobe_start_dup_mmap();
 501         if (mmap_write_lock_killable(oldmm)) {
 502                 retval = -EINTR;
 503                 goto fail_uprobe_end;
 504         }
 505         flush_cache_dup_mm(oldmm);
 506         uprobe_dup_mmap(oldmm, mm);
 507         /*
 508          * Not linked in yet - no deadlock potential:
 509          */
 510         mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
 511
 512         /* No ordering required: file already has been exposed. */
 513         dup_mm_exe_file(mm, oldmm);
 514
 515         mm->total_vm = oldmm->total_vm;
 516         mm->data_vm = oldmm->data_vm;
 517         mm->exec_vm = oldmm->exec_vm;
 518         mm->stack_vm = oldmm->stack_vm;
 519
 520         rb_link = &mm->mm_rb.rb_node;
 521         rb_parent = NULL;
 522         pprev = &mm->mmap;
 523         retval = ksm_fork(mm, oldmm);
 524         if (retval)
 525                 goto out;
 526         retval = khugepaged_fork(mm, oldmm);
 527         if (retval)
 528                 goto out;
 529
 530         prev = NULL;
 531         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 532                 struct file *file;
 533
 534                 if (mpnt->vm_flags & VM_DONTCOPY) {
 535                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 536                         continue;
 537                 }
 538                 charge = 0;
 539                 /*
 540                  * Don't duplicate many vmas if we've been oom-killed (for
 541                  * example)
 542                  */
 543                 if (fatal_signal_pending(current)) {
 544                         retval = -EINTR;
 545                         goto out;
 546                 }
 547                 if (mpnt->vm_flags & VM_ACCOUNT) {
 548                         unsigned long len = vma_pages(mpnt);
 549
 550                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 551                                 goto fail_nomem;
 552                         charge = len;
 553                 }
 554                 tmp = vm_area_dup(mpnt);
 555                 if (!tmp)
 556                         goto fail_nomem;
 557                 retval = vma_dup_policy(mpnt, tmp);
 558                 if (retval)
 559                         goto fail_nomem_policy;
 560                 tmp->vm_mm = mm;
 561                 retval = dup_userfaultfd(tmp, &uf);
 562                 if (retval)
 563                         goto fail_nomem_anon_vma_fork;
 564                 if (tmp->vm_flags & VM_WIPEONFORK) {
 565                         /*
 566                          * VM_WIPEONFORK gets a clean slate in the child.
 567                          * Don't prepare anon_vma until fault since we don't
 568                          * copy page for current vma.
 569                          */
 570                         tmp->anon_vma = NULL;
 571                 } else if (anon_vma_fork(tmp, mpnt))
 572                         goto fail_nomem_anon_vma_fork;
 573                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 574                 file = tmp->vm_file;
 575                 if (file) {
 576                         struct address_space *mapping = file->f_mapping;
 577
 578                         get_file(file);
 579                         i_mmap_lock_write(mapping);
 580                         if (tmp->vm_flags & VM_SHARED)
 581                                 mapping_allow_writable(mapping);
 582                         flush_dcache_mmap_lock(mapping);
 583                         /* insert tmp into the share list, just after mpnt */
 584                         vma_interval_tree_insert_after(tmp, mpnt,
 585                                         &mapping->i_mmap);
 586                         flush_dcache_mmap_unlock(mapping);
 587                         i_mmap_unlock_write(mapping);
 588                 }
 589
 590                 /*
 591                  * Clear hugetlb-related page reserves for children. This only
 592                  * affects MAP_PRIVATE mappings. Faults generated by the child
 593                  * are not guaranteed to succeed, even if read-only
 594                  */
 595                 if (is_vm_hugetlb_page(tmp))
 596                         reset_vma_resv_huge_pages(tmp);
 597
 598                 /*
 599                  * Link in the new vma and copy the page table entries.
 600                  */
 601                 *pprev = tmp;
 602                 pprev = &tmp->vm_next;
 603                 tmp->vm_prev = prev;
 604                 prev = tmp;
 605
 606                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 607                 rb_link = &tmp->vm_rb.rb_right;
 608                 rb_parent = &tmp->vm_rb;
 609
 610                 mm->map_count++;
 611                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 612                         retval = copy_page_range(tmp, mpnt);
 613
 614                 if (tmp->vm_ops && tmp->vm_ops->open)
 615                         tmp->vm_ops->open(tmp);
 616
 617                 if (retval)
 618                         goto out;
 619         }
 620         /* a new mm has just been created */
 621         retval = arch_dup_mmap(oldmm, mm);
 622 out:
 623         mmap_write_unlock(mm);
 624         flush_tlb_mm(oldmm);
 625         mmap_write_unlock(oldmm);
 626         dup_userfaultfd_complete(&uf);
 627 fail_uprobe_end:
 628         uprobe_end_dup_mmap();
 629         return retval;
 630 fail_nomem_anon_vma_fork:
 631         mpol_put(vma_policy(tmp));
 632 fail_nomem_policy:
 633         vm_area_free(tmp);
 634 fail_nomem:
 635         retval = -ENOMEM;
 636         vm_unacct_memory(charge);
 637         goto out;
 638 }
 639
 640 static inline int mm_alloc_pgd(struct mm_struct *mm)
 641 {
 642         mm->pgd = pgd_alloc(mm);
 643         if (unlikely(!mm->pgd))
 644                 return -ENOMEM;
 645         return 0;
 646 }
 647
 648 static inline void mm_free_pgd(struct mm_struct *mm)
 649 {
 650         pgd_free(mm, mm->pgd);
 651 }
 652 #else
 653 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 654 {
 655         mmap_write_lock(oldmm);
 656         dup_mm_exe_file(mm, oldmm);
 657         mmap_write_unlock(oldmm);
 658         return 0;
 659 }
 660 #define mm_alloc_pgd(mm)        (0)
 661 #define mm_free_pgd(mm)
 662 #endif /* CONFIG_MMU */
 663
 664 static void check_mm(struct mm_struct *mm)
 665 {
 666         int i;
 667
 668         BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
 669                          "Please make sure 'struct resident_page_types[]' is updated as well");
 670
 671         for (i = 0; i < NR_MM_COUNTERS; i++) {
 672                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 673
 674                 if (unlikely(x))
 675                         pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
 676                                  mm, resident_page_types[i], x);
 677         }
 678
 679         if (mm_pgtables_bytes(mm))
 680                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 681                                 mm_pgtables_bytes(mm));
 682
 683 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 684         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 685 #endif
 686 }
 687
 688 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 689 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 690
 691 /*
 692  * Called when the last reference to the mm
 693  * is dropped: either by a lazy thread or by
 694  * mmput. Free the page directory and the mm.
 695  */
 696 void __mmdrop(struct mm_struct *mm)
 697 {
 698         BUG_ON(mm == &init_mm);
 699         WARN_ON_ONCE(mm == current->mm);
 700         WARN_ON_ONCE(mm == current->active_mm);
 701         mm_free_pgd(mm);
 702         destroy_context(mm);
 703         mmu_notifier_subscriptions_destroy(mm);
 704         check_mm(mm);
 705         put_user_ns(mm->user_ns);
 706         free_mm(mm);
 707 }
 708 EXPORT_SYMBOL_GPL(__mmdrop);
 709
 710 static void mmdrop_async_fn(struct work_struct *work)
 711 {
 712         struct mm_struct *mm;
 713
 714         mm = container_of(work, struct mm_struct, async_put_work);
 715         __mmdrop(mm);
 716 }
 717
 718 static void mmdrop_async(struct mm_struct *mm)
 719 {
 720         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 721                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 722                 schedule_work(&mm->async_put_work);
 723         }
 724 }
 725
 726 static inline void free_signal_struct(struct signal_struct *sig)
 727 {
 728         taskstats_tgid_free(sig);
 729         sched_autogroup_exit(sig);
 730         /*
 731          * __mmdrop is not safe to call from softirq context on x86 due to
 732          * pgd_dtor so postpone it to the async context
 733          */
 734         if (sig->oom_mm)
 735                 mmdrop_async(sig->oom_mm);
 736         kmem_cache_free(signal_cachep, sig);
 737 }
 738
 739 static inline void put_signal_struct(struct signal_struct *sig)
 740 {
 741         if (refcount_dec_and_test(&sig->sigcnt))
 742                 free_signal_struct(sig);
 743 }
 744
 745 void __put_task_struct(struct task_struct *tsk)
 746 {
 747         WARN_ON(!tsk->exit_state);
 748         WARN_ON(refcount_read(&tsk->usage));
 749         WARN_ON(tsk == current);
 750
 751         io_uring_free(tsk);
 752         cgroup_free(tsk);
 753         task_numa_free(tsk, true);
 754         security_task_free(tsk);
 755         bpf_task_storage_free(tsk);
 756         exit_creds(tsk);
 757         delayacct_tsk_free(tsk);
 758         put_signal_struct(tsk->signal);
 759         sched_core_free(tsk);
 760         free_task(tsk);
 761 }
 762 EXPORT_SYMBOL_GPL(__put_task_struct);
 763
 764 void __init __weak arch_task_cache_init(void) { }
 765
 766 /*
 767  * set_max_threads
 768  */
 769 static void set_max_threads(unsigned int max_threads_suggested)
 770 {
 771         u64 threads;
 772         unsigned long nr_pages = totalram_pages();
 773
 774         /*
 775          * The number of threads shall be limited such that the thread
 776          * structures may only consume a small part of the available memory.
 777          */
 778         if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
 779                 threads = MAX_THREADS;
 780         else
 781                 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
 782                                     (u64) THREAD_SIZE * 8UL);
 783
 784         if (threads > max_threads_suggested)
 785                 threads = max_threads_suggested;
 786
 787         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 788 }
 789
 790 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 791 /* Initialized by the architecture: */
 792 int arch_task_struct_size __read_mostly;
 793 #endif
 794
 795 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 796 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 797 {
 798         /* Fetch thread_struct whitelist for the architecture. */
 799         arch_thread_struct_whitelist(offset, size);
 800
 801         /*
 802          * Handle zero-sized whitelist or empty thread_struct, otherwise
 803          * adjust offset to position of thread_struct in task_struct.
 804          */
 805         if (unlikely(*size == 0))
 806                 *offset = 0;
 807         else
 808                 *offset += offsetof(struct task_struct, thread);
 809 }
 810 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
 811
 812 void __init fork_init(void)
 813 {
 814         int i;
 815 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 816 #ifndef ARCH_MIN_TASKALIGN
 817 #define ARCH_MIN_TASKALIGN      0
 818 #endif
 819         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 820         unsigned long useroffset, usersize;
 821
 822         /* create a slab on which task_structs can be allocated */
 823         task_struct_whitelist(&useroffset, &usersize);
 824         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 825                         arch_task_struct_size, align,
 826                         SLAB_PANIC|SLAB_ACCOUNT,
 827                         useroffset, usersize, NULL);
 828 #endif
 829
 830         /* do the arch specific task caches init */
 831         arch_task_cache_init();
 832
 833         set_max_threads(MAX_THREADS);
 834
 835         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 836         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 837         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 838                 init_task.signal->rlim[RLIMIT_NPROC];
 839
 840         for (i = 0; i < MAX_PER_NAMESPACE_UCOUNTS; i++)
 841                 init_user_ns.ucount_max[i] = max_threads/2;
 842
 843         set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_NPROC,      RLIM_INFINITY);
 844         set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE,   RLIM_INFINITY);
 845         set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
 846         set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK,    RLIM_INFINITY);
 847
 848 #ifdef CONFIG_VMAP_STACK
 849         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 850                           NULL, free_vm_stack_cache);
 851 #endif
 852
 853         scs_init();
 854
 855         lockdep_init_task(&init_task);
 856         uprobes_init();
 857 }
 858
 859 int __weak arch_dup_task_struct(struct task_struct *dst,
 860                                                struct task_struct *src)
 861 {
 862         *dst = *src;
 863         return 0;
 864 }
 865
 866 void set_task_stack_end_magic(struct task_struct *tsk)
 867 {
 868         unsigned long *stackend;
 869
 870         stackend = end_of_stack(tsk);
 871         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 872 }
 873
 874 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 875 {
 876         struct task_struct *tsk;
 877         unsigned long *stack;
 878         struct vm_struct *stack_vm_area __maybe_unused;
 879         int err;
 880
 881         if (node == NUMA_NO_NODE)
 882                 node = tsk_fork_get_node(orig);
 883         tsk = alloc_task_struct_node(node);
 884         if (!tsk)
 885                 return NULL;
 886
 887         stack = alloc_thread_stack_node(tsk, node);
 888         if (!stack)
 889                 goto free_tsk;
 890
 891         if (memcg_charge_kernel_stack(tsk))
 892                 goto free_stack;
 893
 894         stack_vm_area = task_stack_vm_area(tsk);
 895
 896         err = arch_dup_task_struct(tsk, orig);
 897
 898         /*
 899          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 900          * sure they're properly initialized before using any stack-related
 901          * functions again.
 902          */
 903         tsk->stack = stack;
 904 #ifdef CONFIG_VMAP_STACK
 905         tsk->stack_vm_area = stack_vm_area;
 906 #endif
 907 #ifdef CONFIG_THREAD_INFO_IN_TASK
 908         refcount_set(&tsk->stack_refcount, 1);
 909 #endif
 910
 911         if (err)
 912                 goto free_stack;
 913
 914         err = scs_prepare(tsk, node);
 915         if (err)
 916                 goto free_stack;
 917
 918 #ifdef CONFIG_SECCOMP
 919         /*
 920          * We must handle setting up seccomp filters once we're under
 921          * the sighand lock in case orig has changed between now and
 922          * then. Until then, filter must be NULL to avoid messing up
 923          * the usage counts on the error path calling free_task.
 924          */
 925         tsk->seccomp.filter = NULL;
 926 #endif
 927
 928         setup_thread_stack(tsk, orig);
 929         clear_user_return_notifier(tsk);
 930         clear_tsk_need_resched(tsk);
 931         set_task_stack_end_magic(tsk);
 932         clear_syscall_work_syscall_user_dispatch(tsk);
 933
 934 #ifdef CONFIG_STACKPROTECTOR
 935         tsk->stack_canary = get_random_canary();
 936 #endif
 937         if (orig->cpus_ptr == &orig->cpus_mask)
 938                 tsk->cpus_ptr = &tsk->cpus_mask;
 939         dup_user_cpus_ptr(tsk, orig, node);
 940
 941         /*
 942          * One for the user space visible state that goes away when reaped.
 943          * One for the scheduler.
 944          */
 945         refcount_set(&tsk->rcu_users, 2);
 946         /* One for the rcu users */
 947         refcount_set(&tsk->usage, 1);
 948 #ifdef CONFIG_BLK_DEV_IO_TRACE
 949         tsk->btrace_seq = 0;
 950 #endif
 951         tsk->splice_pipe = NULL;
 952         tsk->task_frag.page = NULL;
 953         tsk->wake_q.next = NULL;
 954         tsk->worker_private = NULL;
 955
 956         account_kernel_stack(tsk, 1);
 957
 958         kcov_task_init(tsk);
 959         kmap_local_fork(tsk);
 960
 961 #ifdef CONFIG_FAULT_INJECTION
 962         tsk->fail_nth = 0;
 963 #endif
 964
 965 #ifdef CONFIG_BLK_CGROUP
 966         tsk->throttle_queue = NULL;
 967         tsk->use_memdelay = 0;
 968 #endif
 969
 970 #ifdef CONFIG_MEMCG
 971         tsk->active_memcg = NULL;
 972 #endif
 973         return tsk;
 974
 975 free_stack:
 976         free_thread_stack(tsk);
 977 free_tsk:
 978         free_task_struct(tsk);
 979         return NULL;
 980 }
 981
 982 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 983
 984 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 985
 986 static int __init coredump_filter_setup(char *s)
 987 {
 988         default_dump_filter =
 989                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 990                 MMF_DUMP_FILTER_MASK;
 991         return 1;
 992 }
 993
 994 __setup("coredump_filter=", coredump_filter_setup);
 995
 996 #include <linux/init_task.h>
 997
 998 static void mm_init_aio(struct mm_struct *mm)
 999 {
1000 #ifdef CONFIG_AIO
1001         spin_lock_init(&mm->ioctx_lock);
1002         mm->ioctx_table = NULL;
1003 #endif
1004 }
1005
1006 static __always_inline void mm_clear_owner(struct mm_struct *mm,
1007                                            struct task_struct *p)
1008 {
1009 #ifdef CONFIG_MEMCG
1010         if (mm->owner == p)
1011                 WRITE_ONCE(mm->owner, NULL);
1012 #endif
1013 }
1014
1015 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1016 {
1017 #ifdef CONFIG_MEMCG
1018         mm->owner = p;
1019 #endif
1020 }
1021
1022 static void mm_init_pasid(struct mm_struct *mm)
1023 {
1024 #ifdef CONFIG_IOMMU_SUPPORT
1025         mm->pasid = INIT_PASID;
1026 #endif
1027 }
1028
1029 static void mm_init_uprobes_state(struct mm_struct *mm)
1030 {
1031 #ifdef CONFIG_UPROBES
1032         mm->uprobes_state.xol_area = NULL;
1033 #endif
1034 }
1035
1036 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1037         struct user_namespace *user_ns)
1038 {
1039         mm->mmap = NULL;
1040         mm->mm_rb = RB_ROOT;
1041         mm->vmacache_seqnum = 0;
1042         atomic_set(&mm->mm_users, 1);
1043         atomic_set(&mm->mm_count, 1);
1044         seqcount_init(&mm->write_protect_seq);
1045         mmap_init_lock(mm);
1046         INIT_LIST_HEAD(&mm->mmlist);
1047         mm_pgtables_bytes_init(mm);
1048         mm->map_count = 0;
1049         mm->locked_vm = 0;
1050         atomic64_set(&mm->pinned_vm, 0);
1051         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1052         spin_lock_init(&mm->page_table_lock);
1053         spin_lock_init(&mm->arg_lock);
1054         mm_init_cpumask(mm);
1055         mm_init_aio(mm);
1056         mm_init_owner(mm, p);
1057         mm_init_pasid(mm);
1058         RCU_INIT_POINTER(mm->exe_file, NULL);
1059         mmu_notifier_subscriptions_init(mm);
1060         init_tlb_flush_pending(mm);
1061 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1062         mm->pmd_huge_pte = NULL;
1063 #endif
1064         mm_init_uprobes_state(mm);
1065         hugetlb_count_init(mm);
1066
1067         if (current->mm) {
1068                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1069                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1070         } else {
1071                 mm->flags = default_dump_filter;
1072                 mm->def_flags = 0;
1073         }
1074
1075         if (mm_alloc_pgd(mm))
1076                 goto fail_nopgd;
1077
1078         if (init_new_context(p, mm))
1079                 goto fail_nocontext;
1080
1081         mm->user_ns = get_user_ns(user_ns);
1082         return mm;
1083
1084 fail_nocontext:
1085         mm_free_pgd(mm);
1086 fail_nopgd:
1087         free_mm(mm);
1088         return NULL;
1089 }
1090
1091 /*
1092  * Allocate and initialize an mm_struct.
1093  */
1094 struct mm_struct *mm_alloc(void)
1095 {
1096         struct mm_struct *mm;
1097
1098         mm = allocate_mm();
1099         if (!mm)
1100                 return NULL;
1101
1102         memset(mm, 0, sizeof(*mm));
1103         return mm_init(mm, current, current_user_ns());
1104 }
1105
1106 static inline void __mmput(struct mm_struct *mm)
1107 {
1108         VM_BUG_ON(atomic_read(&mm->mm_users));
1109
1110         uprobe_clear_state(mm);
1111         exit_aio(mm);
1112         ksm_exit(mm);
1113         khugepaged_exit(mm); /* must run before exit_mmap */
1114         exit_mmap(mm);
1115         mm_put_huge_zero_page(mm);
1116         set_mm_exe_file(mm, NULL);
1117         if (!list_empty(&mm->mmlist)) {
1118                 spin_lock(&mmlist_lock);
1119                 list_del(&mm->mmlist);
1120                 spin_unlock(&mmlist_lock);
1121         }
1122         if (mm->binfmt)
1123                 module_put(mm->binfmt->module);
1124         mmdrop(mm);
1125 }
1126
1127 /*
1128  * Decrement the use count and release all resources for an mm.
1129  */
1130 void mmput(struct mm_struct *mm)
1131 {
1132         might_sleep();
1133
1134         if (atomic_dec_and_test(&mm->mm_users))
1135                 __mmput(mm);
1136 }
1137 EXPORT_SYMBOL_GPL(mmput);
1138
1139 #ifdef CONFIG_MMU
1140 static void mmput_async_fn(struct work_struct *work)
1141 {
1142         struct mm_struct *mm = container_of(work, struct mm_struct,
1143                                             async_put_work);
1144
1145         __mmput(mm);
1146 }
1147
1148 void mmput_async(struct mm_struct *mm)
1149 {
1150         if (atomic_dec_and_test(&mm->mm_users)) {
1151                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1152                 schedule_work(&mm->async_put_work);
1153         }
1154 }
1155 #endif
1156
1157 /**
1158  * set_mm_exe_file - change a reference to the mm's executable file
1159  *
1160  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1161  *
1162  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1163  * invocations: in mmput() nobody alive left, in execve task is single
1164  * threaded.
1165  *
1166  * Can only fail if new_exe_file != NULL.
1167  */
1168 int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1169 {
1170         struct file *old_exe_file;
1171
1172         /*
1173          * It is safe to dereference the exe_file without RCU as
1174          * this function is only called if nobody else can access
1175          * this mm -- see comment above for justification.
1176          */
1177         old_exe_file = rcu_dereference_raw(mm->exe_file);
1178
1179         if (new_exe_file) {
1180                 /*
1181                  * We expect the caller (i.e., sys_execve) to already denied
1182                  * write access, so this is unlikely to fail.
1183                  */
1184                 if (unlikely(deny_write_access(new_exe_file)))
1185                         return -EACCES;
1186                 get_file(new_exe_file);
1187         }
1188         rcu_assign_pointer(mm->exe_file, new_exe_file);
1189         if (old_exe_file) {
1190                 allow_write_access(old_exe_file);
1191                 fput(old_exe_file);
1192         }
1193         return 0;
1194 }
1195
1196 /**
1197  * replace_mm_exe_file - replace a reference to the mm's executable file
1198  *
1199  * This changes mm's executable file (shown as symlink /proc/[pid]/exe),
1200  * dealing with concurrent invocation and without grabbing the mmap lock in
1201  * write mode.
1202  *
1203  * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1204  */
1205 int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1206 {
1207         struct vm_area_struct *vma;
1208         struct file *old_exe_file;
1209         int ret = 0;
1210
1211         /* Forbid mm->exe_file change if old file still mapped. */
1212         old_exe_file = get_mm_exe_file(mm);
1213         if (old_exe_file) {
1214                 mmap_read_lock(mm);
1215                 for (vma = mm->mmap; vma && !ret; vma = vma->vm_next) {
1216                         if (!vma->vm_file)
1217                                 continue;
1218                         if (path_equal(&vma->vm_file->f_path,
1219                                        &old_exe_file->f_path))
1220                                 ret = -EBUSY;
1221                 }
1222                 mmap_read_unlock(mm);
1223                 fput(old_exe_file);
1224                 if (ret)
1225                         return ret;
1226         }
1227
1228         /* set the new file, lockless */
1229         ret = deny_write_access(new_exe_file);
1230         if (ret)
1231                 return -EACCES;
1232         get_file(new_exe_file);
1233
1234         old_exe_file = xchg(&mm->exe_file, new_exe_file);
1235         if (old_exe_file) {
1236                 /*
1237                  * Don't race with dup_mmap() getting the file and disallowing
1238                  * write access while someone might open the file writable.
1239                  */
1240                 mmap_read_lock(mm);
1241                 allow_write_access(old_exe_file);
1242                 fput(old_exe_file);
1243                 mmap_read_unlock(mm);
1244         }
1245         return 0;
1246 }
1247
1248 /**
1249  * get_mm_exe_file - acquire a reference to the mm's executable file
1250  *
1251  * Returns %NULL if mm has no associated executable file.
1252  * User must release file via fput().
1253  */
1254 struct file *get_mm_exe_file(struct mm_struct *mm)
1255 {
1256         struct file *exe_file;
1257
1258         rcu_read_lock();
1259         exe_file = rcu_dereference(mm->exe_file);
1260         if (exe_file && !get_file_rcu(exe_file))
1261                 exe_file = NULL;
1262         rcu_read_unlock();
1263         return exe_file;
1264 }
1265
1266 /**
1267  * get_task_exe_file - acquire a reference to the task's executable file
1268  *
1269  * Returns %NULL if task's mm (if any) has no associated executable file or
1270  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1271  * User must release file via fput().
1272  */
1273 struct file *get_task_exe_file(struct task_struct *task)
1274 {
1275         struct file *exe_file = NULL;
1276         struct mm_struct *mm;
1277
1278         task_lock(task);
1279         mm = task->mm;
1280         if (mm) {
1281                 if (!(task->flags & PF_KTHREAD))
1282                         exe_file = get_mm_exe_file(mm);
1283         }
1284         task_unlock(task);
1285         return exe_file;
1286 }
1287
1288 /**
1289  * get_task_mm - acquire a reference to the task's mm
1290  *
1291  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1292  * this kernel workthread has transiently adopted a user mm with use_mm,
1293  * to do its AIO) is not set and if so returns a reference to it, after
1294  * bumping up the use count.  User must release the mm via mmput()
1295  * after use.  Typically used by /proc and ptrace.
1296  */
1297 struct mm_struct *get_task_mm(struct task_struct *task)
1298 {
1299         struct mm_struct *mm;
1300
1301         task_lock(task);
1302         mm = task->mm;
1303         if (mm) {
1304                 if (task->flags & PF_KTHREAD)
1305                         mm = NULL;
1306                 else
1307                         mmget(mm);
1308         }
1309         task_unlock(task);
1310         return mm;
1311 }
1312 EXPORT_SYMBOL_GPL(get_task_mm);
1313
1314 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1315 {
1316         struct mm_struct *mm;
1317         int err;
1318
1319         err =  down_read_killable(&task->signal->exec_update_lock);
1320         if (err)
1321                 return ERR_PTR(err);
1322
1323         mm = get_task_mm(task);
1324         if (mm && mm != current->mm &&
1325                         !ptrace_may_access(task, mode)) {
1326                 mmput(mm);
1327                 mm = ERR_PTR(-EACCES);
1328         }
1329         up_read(&task->signal->exec_update_lock);
1330
1331         return mm;
1332 }
1333
1334 static void complete_vfork_done(struct task_struct *tsk)
1335 {
1336         struct completion *vfork;
1337
1338         task_lock(tsk);
1339         vfork = tsk->vfork_done;
1340         if (likely(vfork)) {
1341                 tsk->vfork_done = NULL;
1342                 complete(vfork);
1343         }
1344         task_unlock(tsk);
1345 }
1346
1347 static int wait_for_vfork_done(struct task_struct *child,
1348                                 struct completion *vfork)
1349 {
1350         int killed;
1351
1352         freezer_do_not_count();
1353         cgroup_enter_frozen();
1354         killed = wait_for_completion_killable(vfork);
1355         cgroup_leave_frozen(false);
1356         freezer_count();
1357
1358         if (killed) {
1359                 task_lock(child);
1360                 child->vfork_done = NULL;
1361                 task_unlock(child);
1362         }
1363
1364         put_task_struct(child);
1365         return killed;
1366 }
1367
1368 /* Please note the differences between mmput and mm_release.
1369  * mmput is called whenever we stop holding onto a mm_struct,
1370  * error success whatever.
1371  *
1372  * mm_release is called after a mm_struct has been removed
1373  * from the current process.
1374  *
1375  * This difference is important for error handling, when we
1376  * only half set up a mm_struct for a new process and need to restore
1377  * the old one.  Because we mmput the new mm_struct before
1378  * restoring the old one. . .
1379  * Eric Biederman 10 January 1998
1380  */
1381 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1382 {
1383         uprobe_free_utask(tsk);
1384
1385         /* Get rid of any cached register state */
1386         deactivate_mm(tsk, mm);
1387
1388         /*
1389          * Signal userspace if we're not exiting with a core dump
1390          * because we want to leave the value intact for debugging
1391          * purposes.
1392          */
1393         if (tsk->clear_child_tid) {
1394                 if (atomic_read(&mm->mm_users) > 1) {
1395                         /*
1396                          * We don't check the error code - if userspace has
1397                          * not set up a proper pointer then tough luck.
1398                          */
1399                         put_user(0, tsk->clear_child_tid);
1400                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1401                                         1, NULL, NULL, 0, 0);
1402                 }
1403                 tsk->clear_child_tid = NULL;
1404         }
1405
1406         /*
1407          * All done, finally we can wake up parent and return this mm to him.
1408          * Also kthread_stop() uses this completion for synchronization.
1409          */
1410         if (tsk->vfork_done)
1411                 complete_vfork_done(tsk);
1412 }
1413
1414 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1415 {
1416         futex_exit_release(tsk);
1417         mm_release(tsk, mm);
1418 }
1419
1420 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1421 {
1422         futex_exec_release(tsk);
1423         mm_release(tsk, mm);
1424 }
1425
1426 /**
1427  * dup_mm() - duplicates an existing mm structure
1428  * @tsk: the task_struct with which the new mm will be associated.
1429  * @oldmm: the mm to duplicate.
1430  *
1431  * Allocates a new mm structure and duplicates the provided @oldmm structure
1432  * content into it.
1433  *
1434  * Return: the duplicated mm or NULL on failure.
1435  */
1436 static struct mm_struct *dup_mm(struct task_struct *tsk,
1437                                 struct mm_struct *oldmm)
1438 {
1439         struct mm_struct *mm;
1440         int err;
1441
1442         mm = allocate_mm();
1443         if (!mm)
1444                 goto fail_nomem;
1445
1446         memcpy(mm, oldmm, sizeof(*mm));
1447
1448         if (!mm_init(mm, tsk, mm->user_ns))
1449                 goto fail_nomem;
1450
1451         err = dup_mmap(mm, oldmm);
1452         if (err)
1453                 goto free_pt;
1454
1455         mm->hiwater_rss = get_mm_rss(mm);
1456         mm->hiwater_vm = mm->total_vm;
1457
1458         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1459                 goto free_pt;
1460
1461         return mm;
1462
1463 free_pt:
1464         /* don't put binfmt in mmput, we haven't got module yet */
1465         mm->binfmt = NULL;
1466         mm_init_owner(mm, NULL);
1467         mmput(mm);
1468
1469 fail_nomem:
1470         return NULL;
1471 }
1472
1473 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1474 {
1475         struct mm_struct *mm, *oldmm;
1476
1477         tsk->min_flt = tsk->maj_flt = 0;
1478         tsk->nvcsw = tsk->nivcsw = 0;
1479 #ifdef CONFIG_DETECT_HUNG_TASK
1480         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1481         tsk->last_switch_time = 0;
1482 #endif
1483
1484         tsk->mm = NULL;
1485         tsk->active_mm = NULL;
1486
1487         /*
1488          * Are we cloning a kernel thread?
1489          *
1490          * We need to steal a active VM for that..
1491          */
1492         oldmm = current->mm;
1493         if (!oldmm)
1494                 return 0;
1495
1496         /* initialize the new vmacache entries */
1497         vmacache_flush(tsk);
1498
1499         if (clone_flags & CLONE_VM) {
1500                 mmget(oldmm);
1501                 mm = oldmm;
1502         } else {
1503                 mm = dup_mm(tsk, current->mm);
1504                 if (!mm)
1505                         return -ENOMEM;
1506         }
1507
1508         tsk->mm = mm;
1509         tsk->active_mm = mm;
1510         return 0;
1511 }
1512
1513 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1514 {
1515         struct fs_struct *fs = current->fs;
1516         if (clone_flags & CLONE_FS) {
1517                 /* tsk->fs is already what we want */
1518                 spin_lock(&fs->lock);
1519                 if (fs->in_exec) {
1520                         spin_unlock(&fs->lock);
1521                         return -EAGAIN;
1522                 }
1523                 fs->users++;
1524                 spin_unlock(&fs->lock);
1525                 return 0;
1526         }
1527         tsk->fs = copy_fs_struct(fs);
1528         if (!tsk->fs)
1529                 return -ENOMEM;
1530         return 0;
1531 }
1532
1533 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1534 {
1535         struct files_struct *oldf, *newf;
1536         int error = 0;
1537
1538         /*
1539          * A background process may not have any files ...
1540          */
1541         oldf = current->files;
1542         if (!oldf)
1543                 goto out;
1544
1545         if (clone_flags & CLONE_FILES) {
1546                 atomic_inc(&oldf->count);
1547                 goto out;
1548         }
1549
1550         newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1551         if (!newf)
1552                 goto out;
1553
1554         tsk->files = newf;
1555         error = 0;
1556 out:
1557         return error;
1558 }
1559
1560 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1561 {
1562         struct sighand_struct *sig;
1563
1564         if (clone_flags & CLONE_SIGHAND) {
1565                 refcount_inc(&current->sighand->count);
1566                 return 0;
1567         }
1568         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1569         RCU_INIT_POINTER(tsk->sighand, sig);
1570         if (!sig)
1571                 return -ENOMEM;
1572
1573         refcount_set(&sig->count, 1);
1574         spin_lock_irq(&current->sighand->siglock);
1575         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1576         spin_unlock_irq(&current->sighand->siglock);
1577
1578         /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1579         if (clone_flags & CLONE_CLEAR_SIGHAND)
1580                 flush_signal_handlers(tsk, 0);
1581
1582         return 0;
1583 }
1584
1585 void __cleanup_sighand(struct sighand_struct *sighand)
1586 {
1587         if (refcount_dec_and_test(&sighand->count)) {
1588                 signalfd_cleanup(sighand);
1589                 /*
1590                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1591                  * without an RCU grace period, see __lock_task_sighand().
1592                  */
1593                 kmem_cache_free(sighand_cachep, sighand);
1594         }
1595 }
1596
1597 /*
1598  * Initialize POSIX timer handling for a thread group.
1599  */
1600 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1601 {
1602         struct posix_cputimers *pct = &sig->posix_cputimers;
1603         unsigned long cpu_limit;
1604
1605         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1606         posix_cputimers_group_init(pct, cpu_limit);
1607 }
1608
1609 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1610 {
1611         struct signal_struct *sig;
1612
1613         if (clone_flags & CLONE_THREAD)
1614                 return 0;
1615
1616         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1617         tsk->signal = sig;
1618         if (!sig)
1619                 return -ENOMEM;
1620
1621         sig->nr_threads = 1;
1622         atomic_set(&sig->live, 1);
1623         refcount_set(&sig->sigcnt, 1);
1624
1625         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1626         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1627         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1628
1629         init_waitqueue_head(&sig->wait_chldexit);
1630         sig->curr_target = tsk;
1631         init_sigpending(&sig->shared_pending);
1632         INIT_HLIST_HEAD(&sig->multiprocess);
1633         seqlock_init(&sig->stats_lock);
1634         prev_cputime_init(&sig->prev_cputime);
1635
1636 #ifdef CONFIG_POSIX_TIMERS
1637         INIT_LIST_HEAD(&sig->posix_timers);
1638         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1639         sig->real_timer.function = it_real_fn;
1640 #endif
1641
1642         task_lock(current->group_leader);
1643         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1644         task_unlock(current->group_leader);
1645
1646         posix_cpu_timers_init_group(sig);
1647
1648         tty_audit_fork(sig);
1649         sched_autogroup_fork(sig);
1650
1651         sig->oom_score_adj = current->signal->oom_score_adj;
1652         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1653
1654         mutex_init(&sig->cred_guard_mutex);
1655         init_rwsem(&sig->exec_update_lock);
1656
1657         return 0;
1658 }
1659
1660 static void copy_seccomp(struct task_struct *p)
1661 {
1662 #ifdef CONFIG_SECCOMP
1663         /*
1664          * Must be called with sighand->lock held, which is common to
1665          * all threads in the group. Holding cred_guard_mutex is not
1666          * needed because this new task is not yet running and cannot
1667          * be racing exec.
1668          */
1669         assert_spin_locked(&current->sighand->siglock);
1670
1671         /* Ref-count the new filter user, and assign it. */
1672         get_seccomp_filter(current);
1673         p->seccomp = current->seccomp;
1674
1675         /*
1676          * Explicitly enable no_new_privs here in case it got set
1677          * between the task_struct being duplicated and holding the
1678          * sighand lock. The seccomp state and nnp must be in sync.
1679          */
1680         if (task_no_new_privs(current))
1681                 task_set_no_new_privs(p);
1682
1683         /*
1684          * If the parent gained a seccomp mode after copying thread
1685          * flags and between before we held the sighand lock, we have
1686          * to manually enable the seccomp thread flag here.
1687          */
1688         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1689                 set_task_syscall_work(p, SECCOMP);
1690 #endif
1691 }
1692
1693 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1694 {
1695         current->clear_child_tid = tidptr;
1696
1697         return task_pid_vnr(current);
1698 }
1699
1700 static void rt_mutex_init_task(struct task_struct *p)
1701 {
1702         raw_spin_lock_init(&p->pi_lock);
1703 #ifdef CONFIG_RT_MUTEXES
1704         p->pi_waiters = RB_ROOT_CACHED;
1705         p->pi_top_task = NULL;
1706         p->pi_blocked_on = NULL;
1707 #endif
1708 }
1709
1710 static inline void init_task_pid_links(struct task_struct *task)
1711 {
1712         enum pid_type type;
1713
1714         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1715                 INIT_HLIST_NODE(&task->pid_links[type]);
1716 }
1717
1718 static inline void
1719 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1720 {
1721         if (type == PIDTYPE_PID)
1722                 task->thread_pid = pid;
1723         else
1724                 task->signal->pids[type] = pid;
1725 }
1726
1727 static inline void rcu_copy_process(struct task_struct *p)
1728 {
1729 #ifdef CONFIG_PREEMPT_RCU
1730         p->rcu_read_lock_nesting = 0;
1731         p->rcu_read_unlock_special.s = 0;
1732         p->rcu_blocked_node = NULL;
1733         INIT_LIST_HEAD(&p->rcu_node_entry);
1734 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1735 #ifdef CONFIG_TASKS_RCU
1736         p->rcu_tasks_holdout = false;
1737         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1738         p->rcu_tasks_idle_cpu = -1;
1739 #endif /* #ifdef CONFIG_TASKS_RCU */
1740 #ifdef CONFIG_TASKS_TRACE_RCU
1741         p->trc_reader_nesting = 0;
1742         p->trc_reader_special.s = 0;
1743         INIT_LIST_HEAD(&p->trc_holdout_list);
1744 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1745 }
1746
1747 struct pid *pidfd_pid(const struct file *file)
1748 {
1749         if (file->f_op == &pidfd_fops)
1750                 return file->private_data;
1751
1752         return ERR_PTR(-EBADF);
1753 }
1754
1755 static int pidfd_release(struct inode *inode, struct file *file)
1756 {
1757         struct pid *pid = file->private_data;
1758
1759         file->private_data = NULL;
1760         put_pid(pid);
1761         return 0;
1762 }
1763
1764 #ifdef CONFIG_PROC_FS
1765 /**
1766  * pidfd_show_fdinfo - print information about a pidfd
1767  * @m: proc fdinfo file
1768  * @f: file referencing a pidfd
1769  *
1770  * Pid:
1771  * This function will print the pid that a given pidfd refers to in the
1772  * pid namespace of the procfs instance.
1773  * If the pid namespace of the process is not a descendant of the pid
1774  * namespace of the procfs instance 0 will be shown as its pid. This is
1775  * similar to calling getppid() on a process whose parent is outside of
1776  * its pid namespace.
1777  *
1778  * NSpid:
1779  * If pid namespaces are supported then this function will also print
1780  * the pid of a given pidfd refers to for all descendant pid namespaces
1781  * starting from the current pid namespace of the instance, i.e. the
1782  * Pid field and the first entry in the NSpid field will be identical.
1783  * If the pid namespace of the process is not a descendant of the pid
1784  * namespace of the procfs instance 0 will be shown as its first NSpid
1785  * entry and no others will be shown.
1786  * Note that this differs from the Pid and NSpid fields in
1787  * /proc/<pid>/status where Pid and NSpid are always shown relative to
1788  * the  pid namespace of the procfs instance. The difference becomes
1789  * obvious when sending around a pidfd between pid namespaces from a
1790  * different branch of the tree, i.e. where no ancestral relation is
1791  * present between the pid namespaces:
1792  * - create two new pid namespaces ns1 and ns2 in the initial pid
1793  *   namespace (also take care to create new mount namespaces in the
1794  *   new pid namespace and mount procfs)
1795  * - create a process with a pidfd in ns1
1796  * - send pidfd from ns1 to ns2
1797  * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1798  *   have exactly one entry, which is 0
1799  */
1800 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1801 {
1802         struct pid *pid = f->private_data;
1803         struct pid_namespace *ns;
1804         pid_t nr = -1;
1805
1806         if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1807                 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1808                 nr = pid_nr_ns(pid, ns);
1809         }
1810
1811         seq_put_decimal_ll(m, "Pid:\t", nr);
1812
1813 #ifdef CONFIG_PID_NS
1814         seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1815         if (nr > 0) {
1816                 int i;
1817
1818                 /* If nr is non-zero it means that 'pid' is valid and that
1819                  * ns, i.e. the pid namespace associated with the procfs
1820                  * instance, is in the pid namespace hierarchy of pid.
1821                  * Start at one below the already printed level.
1822                  */
1823                 for (i = ns->level + 1; i <= pid->level; i++)
1824                         seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1825         }
1826 #endif
1827         seq_putc(m, '\n');
1828 }
1829 #endif
1830
1831 /*
1832  * Poll support for process exit notification.
1833  */
1834 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1835 {
1836         struct pid *pid = file->private_data;
1837         __poll_t poll_flags = 0;
1838
1839         poll_wait(file, &pid->wait_pidfd, pts);
1840
1841         /*
1842          * Inform pollers only when the whole thread group exits.
1843          * If the thread group leader exits before all other threads in the
1844          * group, then poll(2) should block, similar to the wait(2) family.
1845          */
1846         if (thread_group_exited(pid))
1847                 poll_flags = EPOLLIN | EPOLLRDNORM;
1848
1849         return poll_flags;
1850 }
1851
1852 const struct file_operations pidfd_fops = {
1853         .release = pidfd_release,
1854         .poll = pidfd_poll,
1855 #ifdef CONFIG_PROC_FS
1856         .show_fdinfo = pidfd_show_fdinfo,
1857 #endif
1858 };
1859
1860 static void __delayed_free_task(struct rcu_head *rhp)
1861 {
1862         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1863
1864         free_task(tsk);
1865 }
1866
1867 static __always_inline void delayed_free_task(struct task_struct *tsk)
1868 {
1869         if (IS_ENABLED(CONFIG_MEMCG))
1870                 call_rcu(&tsk->rcu, __delayed_free_task);
1871         else
1872                 free_task(tsk);
1873 }
1874
1875 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1876 {
1877         /* Skip if kernel thread */
1878         if (!tsk->mm)
1879                 return;
1880
1881         /* Skip if spawning a thread or using vfork */
1882         if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1883                 return;
1884
1885         /* We need to synchronize with __set_oom_adj */
1886         mutex_lock(&oom_adj_mutex);
1887         set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1888         /* Update the values in case they were changed after copy_signal */
1889         tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1890         tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1891         mutex_unlock(&oom_adj_mutex);
1892 }
1893
1894 /*
1895  * This creates a new process as a copy of the old one,
1896  * but does not actually start it yet.
1897  *
1898  * It copies the registers, and all the appropriate
1899  * parts of the process environment (as per the clone
1900  * flags). The actual kick-off is left to the caller.
1901  */
1902 static __latent_entropy struct task_struct *copy_process(
1903                                         struct pid *pid,
1904                                         int trace,
1905                                         int node,
1906                                         struct kernel_clone_args *args)
1907 {
1908         int pidfd = -1, retval;
1909         struct task_struct *p;
1910         struct multiprocess_signals delayed;
1911         struct file *pidfile = NULL;
1912         u64 clone_flags = args->flags;
1913         struct nsproxy *nsp = current->nsproxy;
1914
1915         /*
1916          * Don't allow sharing the root directory with processes in a different
1917          * namespace
1918          */
1919         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1920                 return ERR_PTR(-EINVAL);
1921
1922         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1923                 return ERR_PTR(-EINVAL);
1924
1925         /*
1926          * Thread groups must share signals as well, and detached threads
1927          * can only be started up within the thread group.
1928          */
1929         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1930                 return ERR_PTR(-EINVAL);
1931
1932         /*
1933          * Shared signal handlers imply shared VM. By way of the above,
1934          * thread groups also imply shared VM. Blocking this case allows
1935          * for various simplifications in other code.
1936          */
1937         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1938                 return ERR_PTR(-EINVAL);
1939
1940         /*
1941          * Siblings of global init remain as zombies on exit since they are
1942          * not reaped by their parent (swapper). To solve this and to avoid
1943          * multi-rooted process trees, prevent global and container-inits
1944          * from creating siblings.
1945          */
1946         if ((clone_flags & CLONE_PARENT) &&
1947                                 current->signal->flags & SIGNAL_UNKILLABLE)
1948                 return ERR_PTR(-EINVAL);
1949
1950         /*
1951          * If the new process will be in a different pid or user namespace
1952          * do not allow it to share a thread group with the forking task.
1953          */
1954         if (clone_flags & CLONE_THREAD) {
1955                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1956                     (task_active_pid_ns(current) != nsp->pid_ns_for_children))
1957                         return ERR_PTR(-EINVAL);
1958         }
1959
1960         /*
1961          * If the new process will be in a different time namespace
1962          * do not allow it to share VM or a thread group with the forking task.
1963          */
1964         if (clone_flags & (CLONE_THREAD | CLONE_VM)) {
1965                 if (nsp->time_ns != nsp->time_ns_for_children)
1966                         return ERR_PTR(-EINVAL);
1967         }
1968
1969         if (clone_flags & CLONE_PIDFD) {
1970                 /*
1971                  * - CLONE_DETACHED is blocked so that we can potentially
1972                  *   reuse it later for CLONE_PIDFD.
1973                  * - CLONE_THREAD is blocked until someone really needs it.
1974                  */
1975                 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1976                         return ERR_PTR(-EINVAL);
1977         }
1978
1979         /*
1980          * Force any signals received before this point to be delivered
1981          * before the fork happens.  Collect up signals sent to multiple
1982          * processes that happen during the fork and delay them so that
1983          * they appear to happen after the fork.
1984          */
1985         sigemptyset(&delayed.signal);
1986         INIT_HLIST_NODE(&delayed.node);
1987
1988         spin_lock_irq(&current->sighand->siglock);
1989         if (!(clone_flags & CLONE_THREAD))
1990                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1991         recalc_sigpending();
1992         spin_unlock_irq(&current->sighand->siglock);
1993         retval = -ERESTARTNOINTR;
1994         if (task_sigpending(current))
1995                 goto fork_out;
1996
1997         retval = -ENOMEM;
1998         p = dup_task_struct(current, node);
1999         if (!p)
2000                 goto fork_out;
2001         if (args->io_thread) {
2002                 /*
2003                  * Mark us an IO worker, and block any signal that isn't
2004                  * fatal or STOP
2005                  */
2006                 p->flags |= PF_IO_WORKER;
2007                 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2008         }
2009
2010         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2011         /*
2012          * Clear TID on mm_release()?
2013          */
2014         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2015
2016         ftrace_graph_init_task(p);
2017
2018         rt_mutex_init_task(p);
2019
2020         lockdep_assert_irqs_enabled();
2021 #ifdef CONFIG_PROVE_LOCKING
2022         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2023 #endif
2024         retval = copy_creds(p, clone_flags);
2025         if (retval < 0)
2026                 goto bad_fork_free;
2027
2028         retval = -EAGAIN;
2029         if (is_ucounts_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2030                 if (p->real_cred->user != INIT_USER &&
2031                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2032                         goto bad_fork_cleanup_count;
2033         }
2034         current->flags &= ~PF_NPROC_EXCEEDED;
2035
2036         /*
2037          * If multiple threads are within copy_process(), then this check
2038          * triggers too late. This doesn't hurt, the check is only there
2039          * to stop root fork bombs.
2040          */
2041         retval = -EAGAIN;
2042         if (data_race(nr_threads >= max_threads))
2043                 goto bad_fork_cleanup_count;
2044
2045         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
2046         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2047         p->flags |= PF_FORKNOEXEC;
2048         INIT_LIST_HEAD(&p->children);
2049         INIT_LIST_HEAD(&p->sibling);
2050         rcu_copy_process(p);
2051         p->vfork_done = NULL;
2052         spin_lock_init(&p->alloc_lock);
2053
2054         init_sigpending(&p->pending);
2055
2056         p->utime = p->stime = p->gtime = 0;
2057 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2058         p->utimescaled = p->stimescaled = 0;
2059 #endif
2060         prev_cputime_init(&p->prev_cputime);
2061
2062 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2063         seqcount_init(&p->vtime.seqcount);
2064         p->vtime.starttime = 0;
2065         p->vtime.state = VTIME_INACTIVE;
2066 #endif
2067
2068 #ifdef CONFIG_IO_URING
2069         p->io_uring = NULL;
2070 #endif
2071
2072 #if defined(SPLIT_RSS_COUNTING)
2073         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2074 #endif
2075
2076         p->default_timer_slack_ns = current->timer_slack_ns;
2077
2078 #ifdef CONFIG_PSI
2079         p->psi_flags = 0;
2080 #endif
2081
2082         task_io_accounting_init(&p->ioac);
2083         acct_clear_integrals(p);
2084
2085         posix_cputimers_init(&p->posix_cputimers);
2086
2087         p->io_context = NULL;
2088         audit_set_context(p, NULL);
2089         cgroup_fork(p);
2090         if (p->flags & PF_KTHREAD) {
2091                 if (!set_kthread_struct(p))
2092                         goto bad_fork_cleanup_delayacct;
2093         }
2094 #ifdef CONFIG_NUMA
2095         p->mempolicy = mpol_dup(p->mempolicy);
2096         if (IS_ERR(p->mempolicy)) {
2097                 retval = PTR_ERR(p->mempolicy);
2098                 p->mempolicy = NULL;
2099                 goto bad_fork_cleanup_delayacct;
2100         }
2101 #endif
2102 #ifdef CONFIG_CPUSETS
2103         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2104         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2105         seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2106 #endif
2107 #ifdef CONFIG_TRACE_IRQFLAGS
2108         memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2109         p->irqtrace.hardirq_disable_ip  = _THIS_IP_;
2110         p->irqtrace.softirq_enable_ip   = _THIS_IP_;
2111         p->softirqs_enabled             = 1;
2112         p->softirq_context              = 0;
2113 #endif
2114
2115         p->pagefault_disabled = 0;
2116
2117 #ifdef CONFIG_LOCKDEP
2118         lockdep_init_task(p);
2119 #endif
2120
2121 #ifdef CONFIG_DEBUG_MUTEXES
2122         p->blocked_on = NULL; /* not blocked yet */
2123 #endif
2124 #ifdef CONFIG_BCACHE
2125         p->sequential_io        = 0;
2126         p->sequential_io_avg    = 0;
2127 #endif
2128 #ifdef CONFIG_BPF_SYSCALL
2129         RCU_INIT_POINTER(p->bpf_storage, NULL);
2130         p->bpf_ctx = NULL;
2131 #endif
2132
2133         /* Perform scheduler related setup. Assign this task to a CPU. */
2134         retval = sched_fork(clone_flags, p);
2135         if (retval)
2136                 goto bad_fork_cleanup_policy;
2137
2138         retval = perf_event_init_task(p, clone_flags);
2139         if (retval)
2140                 goto bad_fork_cleanup_policy;
2141         retval = audit_alloc(p);
2142         if (retval)
2143                 goto bad_fork_cleanup_perf;
2144         /* copy all the process information */
2145         shm_init_task(p);
2146         retval = security_task_alloc(p, clone_flags);
2147         if (retval)
2148                 goto bad_fork_cleanup_audit;
2149         retval = copy_semundo(clone_flags, p);
2150         if (retval)
2151                 goto bad_fork_cleanup_security;
2152         retval = copy_files(clone_flags, p);
2153         if (retval)
2154                 goto bad_fork_cleanup_semundo;
2155         retval = copy_fs(clone_flags, p);
2156         if (retval)
2157                 goto bad_fork_cleanup_files;
2158         retval = copy_sighand(clone_flags, p);
2159         if (retval)
2160                 goto bad_fork_cleanup_fs;
2161         retval = copy_signal(clone_flags, p);
2162         if (retval)
2163                 goto bad_fork_cleanup_sighand;
2164         retval = copy_mm(clone_flags, p);
2165         if (retval)
2166                 goto bad_fork_cleanup_signal;
2167         retval = copy_namespaces(clone_flags, p);
2168         if (retval)
2169                 goto bad_fork_cleanup_mm;
2170         retval = copy_io(clone_flags, p);
2171         if (retval)
2172                 goto bad_fork_cleanup_namespaces;
2173         retval = copy_thread(clone_flags, args->stack, args->stack_size, p, args->tls);
2174         if (retval)
2175                 goto bad_fork_cleanup_io;
2176
2177         stackleak_task_init(p);
2178
2179         if (pid != &init_struct_pid) {
2180                 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2181                                 args->set_tid_size);
2182                 if (IS_ERR(pid)) {
2183                         retval = PTR_ERR(pid);
2184                         goto bad_fork_cleanup_thread;
2185                 }
2186         }
2187
2188         /*
2189          * This has to happen after we've potentially unshared the file
2190          * descriptor table (so that the pidfd doesn't leak into the child
2191          * if the fd table isn't shared).
2192          */
2193         if (clone_flags & CLONE_PIDFD) {
2194                 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2195                 if (retval < 0)
2196                         goto bad_fork_free_pid;
2197
2198                 pidfd = retval;
2199
2200                 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2201                                               O_RDWR | O_CLOEXEC);
2202                 if (IS_ERR(pidfile)) {
2203                         put_unused_fd(pidfd);
2204                         retval = PTR_ERR(pidfile);
2205                         goto bad_fork_free_pid;
2206                 }
2207                 get_pid(pid);   /* held by pidfile now */
2208
2209                 retval = put_user(pidfd, args->pidfd);
2210                 if (retval)
2211                         goto bad_fork_put_pidfd;
2212         }
2213
2214 #ifdef CONFIG_BLOCK
2215         p->plug = NULL;
2216 #endif
2217         futex_init_task(p);
2218
2219         /*
2220          * sigaltstack should be cleared when sharing the same VM
2221          */
2222         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2223                 sas_ss_reset(p);
2224
2225         /*
2226          * Syscall tracing and stepping should be turned off in the
2227          * child regardless of CLONE_PTRACE.
2228          */
2229         user_disable_single_step(p);
2230         clear_task_syscall_work(p, SYSCALL_TRACE);
2231 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2232         clear_task_syscall_work(p, SYSCALL_EMU);
2233 #endif
2234         clear_tsk_latency_tracing(p);
2235
2236         /* ok, now we should be set up.. */
2237         p->pid = pid_nr(pid);
2238         if (clone_flags & CLONE_THREAD) {
2239                 p->group_leader = current->group_leader;
2240                 p->tgid = current->tgid;
2241         } else {
2242                 p->group_leader = p;
2243                 p->tgid = p->pid;
2244         }
2245
2246         p->nr_dirtied = 0;
2247         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2248         p->dirty_paused_when = 0;
2249
2250         p->pdeath_signal = 0;
2251         INIT_LIST_HEAD(&p->thread_group);
2252         p->task_works = NULL;
2253         clear_posix_cputimers_work(p);
2254
2255 #ifdef CONFIG_KRETPROBES
2256         p->kretprobe_instances.first = NULL;
2257 #endif
2258
2259         /*
2260          * Ensure that the cgroup subsystem policies allow the new process to be
2261          * forked. It should be noted that the new process's css_set can be changed
2262          * between here and cgroup_post_fork() if an organisation operation is in
2263          * progress.
2264          */
2265         retval = cgroup_can_fork(p, args);
2266         if (retval)
2267                 goto bad_fork_put_pidfd;
2268
2269         /*
2270          * Now that the cgroups are pinned, re-clone the parent cgroup and put
2271          * the new task on the correct runqueue. All this *before* the task
2272          * becomes visible.
2273          *
2274          * This isn't part of ->can_fork() because while the re-cloning is
2275          * cgroup specific, it unconditionally needs to place the task on a
2276          * runqueue.
2277          */
2278         sched_cgroup_fork(p, args);
2279
2280         /*
2281          * From this point on we must avoid any synchronous user-space
2282          * communication until we take the tasklist-lock. In particular, we do
2283          * not want user-space to be able to predict the process start-time by
2284          * stalling fork(2) after we recorded the start_time but before it is
2285          * visible to the system.
2286          */
2287
2288         p->start_time = ktime_get_ns();
2289         p->start_boottime = ktime_get_boottime_ns();
2290
2291         /*
2292          * Make it visible to the rest of the system, but dont wake it up yet.
2293          * Need tasklist lock for parent etc handling!
2294          */
2295         write_lock_irq(&tasklist_lock);
2296
2297         /* CLONE_PARENT re-uses the old parent */
2298         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2299                 p->real_parent = current->real_parent;
2300                 p->parent_exec_id = current->parent_exec_id;
2301                 if (clone_flags & CLONE_THREAD)
2302                         p->exit_signal = -1;
2303                 else
2304                         p->exit_signal = current->group_leader->exit_signal;
2305         } else {
2306                 p->real_parent = current;
2307                 p->parent_exec_id = current->self_exec_id;
2308                 p->exit_signal = args->exit_signal;
2309         }
2310
2311         klp_copy_process(p);
2312
2313         sched_core_fork(p);
2314
2315         spin_lock(&current->sighand->siglock);
2316
2317         /*
2318          * Copy seccomp details explicitly here, in case they were changed
2319          * before holding sighand lock.
2320          */
2321         copy_seccomp(p);
2322
2323         rseq_fork(p, clone_flags);
2324
2325         /* Don't start children in a dying pid namespace */
2326         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2327                 retval = -ENOMEM;
2328                 goto bad_fork_cancel_cgroup;
2329         }
2330
2331         /* Let kill terminate clone/fork in the middle */
2332         if (fatal_signal_pending(current)) {
2333                 retval = -EINTR;
2334                 goto bad_fork_cancel_cgroup;
2335         }
2336
2337         init_task_pid_links(p);
2338         if (likely(p->pid)) {
2339                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2340
2341                 init_task_pid(p, PIDTYPE_PID, pid);
2342                 if (thread_group_leader(p)) {
2343                         init_task_pid(p, PIDTYPE_TGID, pid);
2344                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2345                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2346
2347                         if (is_child_reaper(pid)) {
2348                                 ns_of_pid(pid)->child_reaper = p;
2349                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2350                         }
2351                         p->signal->shared_pending.signal = delayed.signal;
2352                         p->signal->tty = tty_kref_get(current->signal->tty);
2353                         /*
2354                          * Inherit has_child_subreaper flag under the same
2355                          * tasklist_lock with adding child to the process tree
2356                          * for propagate_has_child_subreaper optimization.
2357                          */
2358                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2359                                                          p->real_parent->signal->is_child_subreaper;
2360                         list_add_tail(&p->sibling, &p->real_parent->children);
2361                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2362                         attach_pid(p, PIDTYPE_TGID);
2363                         attach_pid(p, PIDTYPE_PGID);
2364                         attach_pid(p, PIDTYPE_SID);
2365                         __this_cpu_inc(process_counts);
2366                 } else {
2367                         current->signal->nr_threads++;
2368                         atomic_inc(&current->signal->live);
2369                         refcount_inc(&current->signal->sigcnt);
2370                         task_join_group_stop(p);
2371                         list_add_tail_rcu(&p->thread_group,
2372                                           &p->group_leader->thread_group);
2373                         list_add_tail_rcu(&p->thread_node,
2374                                           &p->signal->thread_head);
2375                 }
2376                 attach_pid(p, PIDTYPE_PID);
2377                 nr_threads++;
2378         }
2379         total_forks++;
2380         hlist_del_init(&delayed.node);
2381         spin_unlock(&current->sighand->siglock);
2382         syscall_tracepoint_update(p);
2383         write_unlock_irq(&tasklist_lock);
2384
2385         if (pidfile)
2386                 fd_install(pidfd, pidfile);
2387
2388         proc_fork_connector(p);
2389         sched_post_fork(p);
2390         cgroup_post_fork(p, args);
2391         perf_event_fork(p);
2392
2393         trace_task_newtask(p, clone_flags);
2394         uprobe_copy_process(p, clone_flags);
2395
2396         copy_oom_score_adj(clone_flags, p);
2397
2398         return p;
2399
2400 bad_fork_cancel_cgroup:
2401         sched_core_free(p);
2402         spin_unlock(&current->sighand->siglock);
2403         write_unlock_irq(&tasklist_lock);
2404         cgroup_cancel_fork(p, args);
2405 bad_fork_put_pidfd:
2406         if (clone_flags & CLONE_PIDFD) {
2407                 fput(pidfile);
2408                 put_unused_fd(pidfd);
2409         }
2410 bad_fork_free_pid:
2411         if (pid != &init_struct_pid)
2412                 free_pid(pid);
2413 bad_fork_cleanup_thread:
2414         exit_thread(p);
2415 bad_fork_cleanup_io:
2416         if (p->io_context)
2417                 exit_io_context(p);
2418 bad_fork_cleanup_namespaces:
2419         exit_task_namespaces(p);
2420 bad_fork_cleanup_mm:
2421         if (p->mm) {
2422                 mm_clear_owner(p->mm, p);
2423                 mmput(p->mm);
2424         }
2425 bad_fork_cleanup_signal:
2426         if (!(clone_flags & CLONE_THREAD))
2427                 free_signal_struct(p->signal);
2428 bad_fork_cleanup_sighand:
2429         __cleanup_sighand(p->sighand);
2430 bad_fork_cleanup_fs:
2431         exit_fs(p); /* blocking */
2432 bad_fork_cleanup_files:
2433         exit_files(p); /* blocking */
2434 bad_fork_cleanup_semundo:
2435         exit_sem(p);
2436 bad_fork_cleanup_security:
2437         security_task_free(p);
2438 bad_fork_cleanup_audit:
2439         audit_free(p);
2440 bad_fork_cleanup_perf:
2441         perf_event_free_task(p);
2442 bad_fork_cleanup_policy:
2443         lockdep_free_task(p);
2444 #ifdef CONFIG_NUMA
2445         mpol_put(p->mempolicy);
2446 #endif
2447 bad_fork_cleanup_delayacct:
2448         delayacct_tsk_free(p);
2449 bad_fork_cleanup_count:
2450         dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2451         exit_creds(p);
2452 bad_fork_free:
2453         WRITE_ONCE(p->__state, TASK_DEAD);
2454         put_task_stack(p);
2455         delayed_free_task(p);
2456 fork_out:
2457         spin_lock_irq(&current->sighand->siglock);
2458         hlist_del_init(&delayed.node);
2459         spin_unlock_irq(&current->sighand->siglock);
2460         return ERR_PTR(retval);
2461 }
2462
2463 static inline void init_idle_pids(struct task_struct *idle)
2464 {
2465         enum pid_type type;
2466
2467         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2468                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2469                 init_task_pid(idle, type, &init_struct_pid);
2470         }
2471 }
2472
2473 struct task_struct * __init fork_idle(int cpu)
2474 {
2475         struct task_struct *task;
2476         struct kernel_clone_args args = {
2477                 .flags = CLONE_VM,
2478         };
2479
2480         task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2481         if (!IS_ERR(task)) {
2482                 init_idle_pids(task);
2483                 init_idle(task, cpu);
2484         }
2485
2486         return task;
2487 }
2488
2489 struct mm_struct *copy_init_mm(void)
2490 {
2491         return dup_mm(NULL, &init_mm);
2492 }
2493
2494 /*
2495  * This is like kernel_clone(), but shaved down and tailored to just
2496  * creating io_uring workers. It returns a created task, or an error pointer.
2497  * The returned task is inactive, and the caller must fire it up through
2498  * wake_up_new_task(p). All signals are blocked in the created task.
2499  */
2500 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2501 {
2502         unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2503                                 CLONE_IO;
2504         struct kernel_clone_args args = {
2505                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2506                                     CLONE_UNTRACED) & ~CSIGNAL),
2507                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2508                 .stack          = (unsigned long)fn,
2509                 .stack_size     = (unsigned long)arg,
2510                 .io_thread      = 1,
2511         };
2512
2513         return copy_process(NULL, 0, node, &args);
2514 }
2515
2516 /*
2517  *  Ok, this is the main fork-routine.
2518  *
2519  * It copies the process, and if successful kick-starts
2520  * it and waits for it to finish using the VM if required.
2521  *
2522  * args->exit_signal is expected to be checked for sanity by the caller.
2523  */
2524 pid_t kernel_clone(struct kernel_clone_args *args)
2525 {
2526         u64 clone_flags = args->flags;
2527         struct completion vfork;
2528         struct pid *pid;
2529         struct task_struct *p;
2530         int trace = 0;
2531         pid_t nr;
2532
2533         /*
2534          * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2535          * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2536          * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2537          * field in struct clone_args and it still doesn't make sense to have
2538          * them both point at the same memory location. Performing this check
2539          * here has the advantage that we don't need to have a separate helper
2540          * to check for legacy clone().
2541          */
2542         if ((args->flags & CLONE_PIDFD) &&
2543             (args->flags & CLONE_PARENT_SETTID) &&
2544             (args->pidfd == args->parent_tid))
2545                 return -EINVAL;
2546
2547         /*
2548          * Determine whether and which event to report to ptracer.  When
2549          * called from kernel_thread or CLONE_UNTRACED is explicitly
2550          * requested, no event is reported; otherwise, report if the event
2551          * for the type of forking is enabled.
2552          */
2553         if (!(clone_flags & CLONE_UNTRACED)) {
2554                 if (clone_flags & CLONE_VFORK)
2555                         trace = PTRACE_EVENT_VFORK;
2556                 else if (args->exit_signal != SIGCHLD)
2557                         trace = PTRACE_EVENT_CLONE;
2558                 else
2559                         trace = PTRACE_EVENT_FORK;
2560
2561                 if (likely(!ptrace_event_enabled(current, trace)))
2562                         trace = 0;
2563         }
2564
2565         p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2566         add_latent_entropy();
2567
2568         if (IS_ERR(p))
2569                 return PTR_ERR(p);
2570
2571         /*
2572          * Do this prior waking up the new thread - the thread pointer
2573          * might get invalid after that point, if the thread exits quickly.
2574          */
2575         trace_sched_process_fork(current, p);
2576
2577         pid = get_task_pid(p, PIDTYPE_PID);
2578         nr = pid_vnr(pid);
2579
2580         if (clone_flags & CLONE_PARENT_SETTID)
2581                 put_user(nr, args->parent_tid);
2582
2583         if (clone_flags & CLONE_VFORK) {
2584                 p->vfork_done = &vfork;
2585                 init_completion(&vfork);
2586                 get_task_struct(p);
2587         }
2588
2589         wake_up_new_task(p);
2590
2591         /* forking complete and child started to run, tell ptracer */
2592         if (unlikely(trace))
2593                 ptrace_event_pid(trace, pid);
2594
2595         if (clone_flags & CLONE_VFORK) {
2596                 if (!wait_for_vfork_done(p, &vfork))
2597                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2598         }
2599
2600         put_pid(pid);
2601         return nr;
2602 }
2603
2604 /*
2605  * Create a kernel thread.
2606  */
2607 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2608 {
2609         struct kernel_clone_args args = {
2610                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2611                                     CLONE_UNTRACED) & ~CSIGNAL),
2612                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2613                 .stack          = (unsigned long)fn,
2614                 .stack_size     = (unsigned long)arg,
2615         };
2616
2617         return kernel_clone(&args);
2618 }
2619
2620 #ifdef __ARCH_WANT_SYS_FORK
2621 SYSCALL_DEFINE0(fork)
2622 {
2623 #ifdef CONFIG_MMU
2624         struct kernel_clone_args args = {
2625                 .exit_signal = SIGCHLD,
2626         };
2627
2628         return kernel_clone(&args);
2629 #else
2630         /* can not support in nommu mode */
2631         return -EINVAL;
2632 #endif
2633 }
2634 #endif
2635
2636 #ifdef __ARCH_WANT_SYS_VFORK
2637 SYSCALL_DEFINE0(vfork)
2638 {
2639         struct kernel_clone_args args = {
2640                 .flags          = CLONE_VFORK | CLONE_VM,
2641                 .exit_signal    = SIGCHLD,
2642         };
2643
2644         return kernel_clone(&args);
2645 }
2646 #endif
2647
2648 #ifdef __ARCH_WANT_SYS_CLONE
2649 #ifdef CONFIG_CLONE_BACKWARDS
2650 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2651                  int __user *, parent_tidptr,
2652                  unsigned long, tls,
2653                  int __user *, child_tidptr)
2654 #elif defined(CONFIG_CLONE_BACKWARDS2)
2655 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2656                  int __user *, parent_tidptr,
2657                  int __user *, child_tidptr,
2658                  unsigned long, tls)
2659 #elif defined(CONFIG_CLONE_BACKWARDS3)
2660 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2661                 int, stack_size,
2662                 int __user *, parent_tidptr,
2663                 int __user *, child_tidptr,
2664                 unsigned long, tls)
2665 #else
2666 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2667                  int __user *, parent_tidptr,
2668                  int __user *, child_tidptr,
2669                  unsigned long, tls)
2670 #endif
2671 {
2672         struct kernel_clone_args args = {
2673                 .flags          = (lower_32_bits(clone_flags) & ~CSIGNAL),
2674                 .pidfd          = parent_tidptr,
2675                 .child_tid      = child_tidptr,
2676                 .parent_tid     = parent_tidptr,
2677                 .exit_signal    = (lower_32_bits(clone_flags) & CSIGNAL),
2678                 .stack          = newsp,
2679                 .tls            = tls,
2680         };
2681
2682         return kernel_clone(&args);
2683 }
2684 #endif
2685
2686 #ifdef __ARCH_WANT_SYS_CLONE3
2687
2688 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2689                                               struct clone_args __user *uargs,
2690                                               size_t usize)
2691 {
2692         int err;
2693         struct clone_args args;
2694         pid_t *kset_tid = kargs->set_tid;
2695
2696         BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2697                      CLONE_ARGS_SIZE_VER0);
2698         BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2699                      CLONE_ARGS_SIZE_VER1);
2700         BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2701                      CLONE_ARGS_SIZE_VER2);
2702         BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2703
2704         if (unlikely(usize > PAGE_SIZE))
2705                 return -E2BIG;
2706         if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2707                 return -EINVAL;
2708
2709         err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2710         if (err)
2711                 return err;
2712
2713         if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2714                 return -EINVAL;
2715
2716         if (unlikely(!args.set_tid && args.set_tid_size > 0))
2717                 return -EINVAL;
2718
2719         if (unlikely(args.set_tid && args.set_tid_size == 0))
2720                 return -EINVAL;
2721
2722         /*
2723          * Verify that higher 32bits of exit_signal are unset and that
2724          * it is a valid signal
2725          */
2726         if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2727                      !valid_signal(args.exit_signal)))
2728                 return -EINVAL;
2729
2730         if ((args.flags & CLONE_INTO_CGROUP) &&
2731             (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2732                 return -EINVAL;
2733
2734         *kargs = (struct kernel_clone_args){
2735                 .flags          = args.flags,
2736                 .pidfd          = u64_to_user_ptr(args.pidfd),
2737                 .child_tid      = u64_to_user_ptr(args.child_tid),
2738                 .parent_tid     = u64_to_user_ptr(args.parent_tid),
2739                 .exit_signal    = args.exit_signal,
2740                 .stack          = args.stack,
2741                 .stack_size     = args.stack_size,
2742                 .tls            = args.tls,
2743                 .set_tid_size   = args.set_tid_size,
2744                 .cgroup         = args.cgroup,
2745         };
2746
2747         if (args.set_tid &&
2748                 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2749                         (kargs->set_tid_size * sizeof(pid_t))))
2750                 return -EFAULT;
2751
2752         kargs->set_tid = kset_tid;
2753
2754         return 0;
2755 }
2756
2757 /**
2758  * clone3_stack_valid - check and prepare stack
2759  * @kargs: kernel clone args
2760  *
2761  * Verify that the stack arguments userspace gave us are sane.
2762  * In addition, set the stack direction for userspace since it's easy for us to
2763  * determine.
2764  */
2765 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2766 {
2767         if (kargs->stack == 0) {
2768                 if (kargs->stack_size > 0)
2769                         return false;
2770         } else {
2771                 if (kargs->stack_size == 0)
2772                         return false;
2773
2774                 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2775                         return false;
2776
2777 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2778                 kargs->stack += kargs->stack_size;
2779 #endif
2780         }
2781
2782         return true;
2783 }
2784
2785 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2786 {
2787         /* Verify that no unknown flags are passed along. */
2788         if (kargs->flags &
2789             ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2790                 return false;
2791
2792         /*
2793          * - make the CLONE_DETACHED bit reusable for clone3
2794          * - make the CSIGNAL bits reusable for clone3
2795          */
2796         if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2797                 return false;
2798
2799         if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2800             (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2801                 return false;
2802
2803         if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2804             kargs->exit_signal)
2805                 return false;
2806
2807         if (!clone3_stack_valid(kargs))
2808                 return false;
2809
2810         return true;
2811 }
2812
2813 /**
2814  * clone3 - create a new process with specific properties
2815  * @uargs: argument structure
2816  * @size:  size of @uargs
2817  *
2818  * clone3() is the extensible successor to clone()/clone2().
2819  * It takes a struct as argument that is versioned by its size.
2820  *
2821  * Return: On success, a positive PID for the child process.
2822  *         On error, a negative errno number.
2823  */
2824 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2825 {
2826         int err;
2827
2828         struct kernel_clone_args kargs;
2829         pid_t set_tid[MAX_PID_NS_LEVEL];
2830
2831         kargs.set_tid = set_tid;
2832
2833         err = copy_clone_args_from_user(&kargs, uargs, size);
2834         if (err)
2835                 return err;
2836
2837         if (!clone3_args_valid(&kargs))
2838                 return -EINVAL;
2839
2840         return kernel_clone(&kargs);
2841 }
2842 #endif
2843
2844 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2845 {
2846         struct task_struct *leader, *parent, *child;
2847         int res;
2848
2849         read_lock(&tasklist_lock);
2850         leader = top = top->group_leader;
2851 down:
2852         for_each_thread(leader, parent) {
2853                 list_for_each_entry(child, &parent->children, sibling) {
2854                         res = visitor(child, data);
2855                         if (res) {
2856                                 if (res < 0)
2857                                         goto out;
2858                                 leader = child;
2859                                 goto down;
2860                         }
2861 up:
2862                         ;
2863                 }
2864         }
2865
2866         if (leader != top) {
2867                 child = leader;
2868                 parent = child->real_parent;
2869                 leader = parent->group_leader;
2870                 goto up;
2871         }
2872 out:
2873         read_unlock(&tasklist_lock);
2874 }
2875
2876 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2877 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2878 #endif
2879
2880 static void sighand_ctor(void *data)
2881 {
2882         struct sighand_struct *sighand = data;
2883
2884         spin_lock_init(&sighand->siglock);
2885         init_waitqueue_head(&sighand->signalfd_wqh);
2886 }
2887
2888 void __init proc_caches_init(void)
2889 {
2890         unsigned int mm_size;
2891
2892         sighand_cachep = kmem_cache_create("sighand_cache",
2893                         sizeof(struct sighand_struct), 0,
2894                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2895                         SLAB_ACCOUNT, sighand_ctor);
2896         signal_cachep = kmem_cache_create("signal_cache",
2897                         sizeof(struct signal_struct), 0,
2898                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2899                         NULL);
2900         files_cachep = kmem_cache_create("files_cache",
2901                         sizeof(struct files_struct), 0,
2902                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2903                         NULL);
2904         fs_cachep = kmem_cache_create("fs_cache",
2905                         sizeof(struct fs_struct), 0,
2906                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2907                         NULL);
2908
2909         /*
2910          * The mm_cpumask is located at the end of mm_struct, and is
2911          * dynamically sized based on the maximum CPU number this system
2912          * can have, taking hotplug into account (nr_cpu_ids).
2913          */
2914         mm_size = sizeof(struct mm_struct) + cpumask_size();
2915
2916         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2917                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2918                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2919                         offsetof(struct mm_struct, saved_auxv),
2920                         sizeof_field(struct mm_struct, saved_auxv),
2921                         NULL);
2922         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2923         mmap_init();
2924         nsproxy_cache_init();
2925 }
2926
2927 /*
2928  * Check constraints on flags passed to the unshare system call.
2929  */
2930 static int check_unshare_flags(unsigned long unshare_flags)
2931 {
2932         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2933                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2934                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2935                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
2936                                 CLONE_NEWTIME))
2937                 return -EINVAL;
2938         /*
2939          * Not implemented, but pretend it works if there is nothing
2940          * to unshare.  Note that unsharing the address space or the
2941          * signal handlers also need to unshare the signal queues (aka
2942          * CLONE_THREAD).
2943          */
2944         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2945                 if (!thread_group_empty(current))
2946                         return -EINVAL;
2947         }
2948         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2949                 if (refcount_read(&current->sighand->count) > 1)
2950                         return -EINVAL;
2951         }
2952         if (unshare_flags & CLONE_VM) {
2953                 if (!current_is_single_threaded())
2954                         return -EINVAL;
2955         }
2956
2957         return 0;
2958 }
2959
2960 /*
2961  * Unshare the filesystem structure if it is being shared
2962  */
2963 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2964 {
2965         struct fs_struct *fs = current->fs;
2966
2967         if (!(unshare_flags & CLONE_FS) || !fs)
2968                 return 0;
2969
2970         /* don't need lock here; in the worst case we'll do useless copy */
2971         if (fs->users == 1)
2972                 return 0;
2973
2974         *new_fsp = copy_fs_struct(fs);
2975         if (!*new_fsp)
2976                 return -ENOMEM;
2977
2978         return 0;
2979 }
2980
2981 /*
2982  * Unshare file descriptor table if it is being shared
2983  */
2984 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
2985                struct files_struct **new_fdp)
2986 {
2987         struct files_struct *fd = current->files;
2988         int error = 0;
2989
2990         if ((unshare_flags & CLONE_FILES) &&
2991             (fd && atomic_read(&fd->count) > 1)) {
2992                 *new_fdp = dup_fd(fd, max_fds, &error);
2993                 if (!*new_fdp)
2994                         return error;
2995         }
2996
2997         return 0;
2998 }
2999
3000 /*
3001  * unshare allows a process to 'unshare' part of the process
3002  * context which was originally shared using clone.  copy_*
3003  * functions used by kernel_clone() cannot be used here directly
3004  * because they modify an inactive task_struct that is being
3005  * constructed. Here we are modifying the current, active,
3006  * task_struct.
3007  */
3008 int ksys_unshare(unsigned long unshare_flags)
3009 {
3010         struct fs_struct *fs, *new_fs = NULL;
3011         struct files_struct *new_fd = NULL;
3012         struct cred *new_cred = NULL;
3013         struct nsproxy *new_nsproxy = NULL;
3014         int do_sysvsem = 0;
3015         int err;
3016
3017         /*
3018          * If unsharing a user namespace must also unshare the thread group
3019          * and unshare the filesystem root and working directories.
3020          */
3021         if (unshare_flags & CLONE_NEWUSER)
3022                 unshare_flags |= CLONE_THREAD | CLONE_FS;
3023         /*
3024          * If unsharing vm, must also unshare signal handlers.
3025          */
3026         if (unshare_flags & CLONE_VM)
3027                 unshare_flags |= CLONE_SIGHAND;
3028         /*
3029          * If unsharing a signal handlers, must also unshare the signal queues.
3030          */
3031         if (unshare_flags & CLONE_SIGHAND)
3032                 unshare_flags |= CLONE_THREAD;
3033         /*
3034          * If unsharing namespace, must also unshare filesystem information.
3035          */
3036         if (unshare_flags & CLONE_NEWNS)
3037                 unshare_flags |= CLONE_FS;
3038
3039         err = check_unshare_flags(unshare_flags);
3040         if (err)
3041                 goto bad_unshare_out;
3042         /*
3043          * CLONE_NEWIPC must also detach from the undolist: after switching
3044          * to a new ipc namespace, the semaphore arrays from the old
3045          * namespace are unreachable.
3046          */
3047         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3048                 do_sysvsem = 1;
3049         err = unshare_fs(unshare_flags, &new_fs);
3050         if (err)
3051                 goto bad_unshare_out;
3052         err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3053         if (err)
3054                 goto bad_unshare_cleanup_fs;
3055         err = unshare_userns(unshare_flags, &new_cred);
3056         if (err)
3057                 goto bad_unshare_cleanup_fd;
3058         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3059                                          new_cred, new_fs);
3060         if (err)
3061                 goto bad_unshare_cleanup_cred;
3062
3063         if (new_cred) {
3064                 err = set_cred_ucounts(new_cred);
3065                 if (err)
3066                         goto bad_unshare_cleanup_cred;
3067         }
3068
3069         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3070                 if (do_sysvsem) {
3071                         /*
3072                          * CLONE_SYSVSEM is equivalent to sys_exit().
3073                          */
3074                         exit_sem(current);
3075                 }
3076                 if (unshare_flags & CLONE_NEWIPC) {
3077                         /* Orphan segments in old ns (see sem above). */
3078                         exit_shm(current);
3079                         shm_init_task(current);
3080                 }
3081
3082                 if (new_nsproxy)
3083                         switch_task_namespaces(current, new_nsproxy);
3084
3085                 task_lock(current);
3086
3087                 if (new_fs) {
3088                         fs = current->fs;
3089                         spin_lock(&fs->lock);
3090                         current->fs = new_fs;
3091                         if (--fs->users)
3092                                 new_fs = NULL;
3093                         else
3094                                 new_fs = fs;
3095                         spin_unlock(&fs->lock);
3096                 }
3097
3098                 if (new_fd)
3099                         swap(current->files, new_fd);
3100
3101                 task_unlock(current);
3102
3103                 if (new_cred) {
3104                         /* Install the new user namespace */
3105                         commit_creds(new_cred);
3106                         new_cred = NULL;
3107                 }
3108         }
3109
3110         perf_event_namespaces(current);
3111
3112 bad_unshare_cleanup_cred:
3113         if (new_cred)
3114                 put_cred(new_cred);
3115 bad_unshare_cleanup_fd:
3116         if (new_fd)
3117                 put_files_struct(new_fd);
3118
3119 bad_unshare_cleanup_fs:
3120         if (new_fs)
3121                 free_fs_struct(new_fs);
3122
3123 bad_unshare_out:
3124         return err;
3125 }
3126
3127 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3128 {
3129         return ksys_unshare(unshare_flags);
3130 }
3131
3132 /*
3133  *      Helper to unshare the files of the current task.
3134  *      We don't want to expose copy_files internals to
3135  *      the exec layer of the kernel.
3136  */
3137
3138 int unshare_files(void)
3139 {
3140         struct task_struct *task = current;
3141         struct files_struct *old, *copy = NULL;
3142         int error;
3143
3144         error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, &copy);
3145         if (error || !copy)
3146                 return error;
3147
3148         old = task->files;
3149         task_lock(task);
3150         task->files = copy;
3151         task_unlock(task);
3152         put_files_struct(old);
3153         return 0;
3154 }
3155
3156 int sysctl_max_threads(struct ctl_table *table, int write,
3157                        void *buffer, size_t *lenp, loff_t *ppos)
3158 {
3159         struct ctl_table t;
3160         int ret;
3161         int threads = max_threads;
3162         int min = 1;
3163         int max = MAX_THREADS;
3164
3165         t = *table;
3166         t.data = &threads;
3167         t.extra1 = &min;
3168         t.extra2 = &max;
3169
3170         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3171         if (ret || !write)
3172                 return ret;
3173
3174         max_threads = threads;
3175
3176         return 0;
3177 }