kernel/fork.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/kernel/fork.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  */
   7
   8 /*
   9  *  'fork.c' contains the help-routines for the 'fork' system call
  10  * (see also entry.S and others).
  11  * Fork is rather simple, once you get the hang of it, but the memory
  12  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
  13  */
  14
  15 #include <linux/anon_inodes.h>
  16 #include <linux/slab.h>
  17 #include <linux/sched/autogroup.h>
  18 #include <linux/sched/mm.h>
  19 #include <linux/sched/coredump.h>
  20 #include <linux/sched/user.h>
  21 #include <linux/sched/numa_balancing.h>
  22 #include <linux/sched/stat.h>
  23 #include <linux/sched/task.h>
  24 #include <linux/sched/task_stack.h>
  25 #include <linux/sched/cputime.h>
  26 #include <linux/seq_file.h>
  27 #include <linux/rtmutex.h>
  28 #include <linux/init.h>
  29 #include <linux/unistd.h>
  30 #include <linux/module.h>
  31 #include <linux/vmalloc.h>
  32 #include <linux/completion.h>
  33 #include <linux/personality.h>
  34 #include <linux/mempolicy.h>
  35 #include <linux/sem.h>
  36 #include <linux/file.h>
  37 #include <linux/fdtable.h>
  38 #include <linux/iocontext.h>
  39 #include <linux/key.h>
  40 #include <linux/binfmts.h>
  41 #include <linux/mman.h>
  42 #include <linux/mmu_notifier.h>
  43 #include <linux/fs.h>
  44 #include <linux/mm.h>
  45 #include <linux/vmacache.h>
  46 #include <linux/nsproxy.h>
  47 #include <linux/capability.h>
  48 #include <linux/cpu.h>
  49 #include <linux/cgroup.h>
  50 #include <linux/security.h>
  51 #include <linux/hugetlb.h>
  52 #include <linux/seccomp.h>
  53 #include <linux/swap.h>
  54 #include <linux/syscalls.h>
  55 #include <linux/jiffies.h>
  56 #include <linux/futex.h>
  57 #include <linux/compat.h>
  58 #include <linux/kthread.h>
  59 #include <linux/task_io_accounting_ops.h>
  60 #include <linux/rcupdate.h>
  61 #include <linux/ptrace.h>
  62 #include <linux/mount.h>
  63 #include <linux/audit.h>
  64 #include <linux/memcontrol.h>
  65 #include <linux/ftrace.h>
  66 #include <linux/proc_fs.h>
  67 #include <linux/profile.h>
  68 #include <linux/rmap.h>
  69 #include <linux/ksm.h>
  70 #include <linux/acct.h>
  71 #include <linux/userfaultfd_k.h>
  72 #include <linux/tsacct_kern.h>
  73 #include <linux/cn_proc.h>
  74 #include <linux/freezer.h>
  75 #include <linux/delayacct.h>
  76 #include <linux/taskstats_kern.h>
  77 #include <linux/random.h>
  78 #include <linux/tty.h>
  79 #include <linux/blkdev.h>
  80 #include <linux/fs_struct.h>
  81 #include <linux/magic.h>
  82 #include <linux/perf_event.h>
  83 #include <linux/posix-timers.h>
  84 #include <linux/user-return-notifier.h>
  85 #include <linux/oom.h>
  86 #include <linux/khugepaged.h>
  87 #include <linux/signalfd.h>
  88 #include <linux/uprobes.h>
  89 #include <linux/aio.h>
  90 #include <linux/compiler.h>
  91 #include <linux/sysctl.h>
  92 #include <linux/kcov.h>
  93 #include <linux/livepatch.h>
  94 #include <linux/thread_info.h>
  95 #include <linux/stackleak.h>
  96 #include <linux/kasan.h>
  97 #include <linux/scs.h>
  98 #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         }
 370         return new;
 371 }
 372
 373 void vm_area_free(struct vm_area_struct *vma)
 374 {
 375         kmem_cache_free(vm_area_cachep, vma);
 376 }
 377
 378 static void account_kernel_stack(struct task_struct *tsk, int account)
 379 {
 380         void *stack = task_stack_page(tsk);
 381         struct vm_struct *vm = task_stack_vm_area(tsk);
 382
 383         if (vm) {
 384                 int i;
 385
 386                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
 387                         mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
 388                                               account * (PAGE_SIZE / 1024));
 389         } else {
 390                 /* All stack pages are in the same node. */
 391                 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
 392                                       account * (THREAD_SIZE / 1024));
 393         }
 394 }
 395
 396 static int memcg_charge_kernel_stack(struct task_struct *tsk)
 397 {
 398 #ifdef CONFIG_VMAP_STACK
 399         struct vm_struct *vm = task_stack_vm_area(tsk);
 400         int ret;
 401
 402         BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
 403
 404         if (vm) {
 405                 int i;
 406
 407                 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
 408
 409                 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
 410                         /*
 411                          * If memcg_kmem_charge_page() fails, page's
 412                          * memory cgroup pointer is NULL, and
 413                          * memcg_kmem_uncharge_page() in free_thread_stack()
 414                          * will ignore this page.
 415                          */
 416                         ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL,
 417                                                      0);
 418                         if (ret)
 419                                 return ret;
 420                 }
 421         }
 422 #endif
 423         return 0;
 424 }
 425
 426 static void release_task_stack(struct task_struct *tsk)
 427 {
 428         if (WARN_ON(tsk->state != TASK_DEAD))
 429                 return;  /* Better to leak the stack than to free prematurely */
 430
 431         account_kernel_stack(tsk, -1);
 432         free_thread_stack(tsk);
 433         tsk->stack = NULL;
 434 #ifdef CONFIG_VMAP_STACK
 435         tsk->stack_vm_area = NULL;
 436 #endif
 437 }
 438
 439 #ifdef CONFIG_THREAD_INFO_IN_TASK
 440 void put_task_stack(struct task_struct *tsk)
 441 {
 442         if (refcount_dec_and_test(&tsk->stack_refcount))
 443                 release_task_stack(tsk);
 444 }
 445 #endif
 446
 447 void free_task(struct task_struct *tsk)
 448 {
 449         scs_release(tsk);
 450
 451 #ifndef CONFIG_THREAD_INFO_IN_TASK
 452         /*
 453          * The task is finally done with both the stack and thread_info,
 454          * so free both.
 455          */
 456         release_task_stack(tsk);
 457 #else
 458         /*
 459          * If the task had a separate stack allocation, it should be gone
 460          * by now.
 461          */
 462         WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
 463 #endif
 464         rt_mutex_debug_task_free(tsk);
 465         ftrace_graph_exit_task(tsk);
 466         arch_release_task_struct(tsk);
 467         if (tsk->flags & PF_KTHREAD)
 468                 free_kthread_struct(tsk);
 469         free_task_struct(tsk);
 470 }
 471 EXPORT_SYMBOL(free_task);
 472
 473 #ifdef CONFIG_MMU
 474 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 475                                         struct mm_struct *oldmm)
 476 {
 477         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
 478         struct rb_node **rb_link, *rb_parent;
 479         int retval;
 480         unsigned long charge;
 481         LIST_HEAD(uf);
 482
 483         uprobe_start_dup_mmap();
 484         if (mmap_write_lock_killable(oldmm)) {
 485                 retval = -EINTR;
 486                 goto fail_uprobe_end;
 487         }
 488         flush_cache_dup_mm(oldmm);
 489         uprobe_dup_mmap(oldmm, mm);
 490         /*
 491          * Not linked in yet - no deadlock potential:
 492          */
 493         mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
 494
 495         /* No ordering required: file already has been exposed. */
 496         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 497
 498         mm->total_vm = oldmm->total_vm;
 499         mm->data_vm = oldmm->data_vm;
 500         mm->exec_vm = oldmm->exec_vm;
 501         mm->stack_vm = oldmm->stack_vm;
 502
 503         rb_link = &mm->mm_rb.rb_node;
 504         rb_parent = NULL;
 505         pprev = &mm->mmap;
 506         retval = ksm_fork(mm, oldmm);
 507         if (retval)
 508                 goto out;
 509         retval = khugepaged_fork(mm, oldmm);
 510         if (retval)
 511                 goto out;
 512
 513         prev = NULL;
 514         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
 515                 struct file *file;
 516
 517                 if (mpnt->vm_flags & VM_DONTCOPY) {
 518                         vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
 519                         continue;
 520                 }
 521                 charge = 0;
 522                 /*
 523                  * Don't duplicate many vmas if we've been oom-killed (for
 524                  * example)
 525                  */
 526                 if (fatal_signal_pending(current)) {
 527                         retval = -EINTR;
 528                         goto out;
 529                 }
 530                 if (mpnt->vm_flags & VM_ACCOUNT) {
 531                         unsigned long len = vma_pages(mpnt);
 532
 533                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
 534                                 goto fail_nomem;
 535                         charge = len;
 536                 }
 537                 tmp = vm_area_dup(mpnt);
 538                 if (!tmp)
 539                         goto fail_nomem;
 540                 retval = vma_dup_policy(mpnt, tmp);
 541                 if (retval)
 542                         goto fail_nomem_policy;
 543                 tmp->vm_mm = mm;
 544                 retval = dup_userfaultfd(tmp, &uf);
 545                 if (retval)
 546                         goto fail_nomem_anon_vma_fork;
 547                 if (tmp->vm_flags & VM_WIPEONFORK) {
 548                         /*
 549                          * VM_WIPEONFORK gets a clean slate in the child.
 550                          * Don't prepare anon_vma until fault since we don't
 551                          * copy page for current vma.
 552                          */
 553                         tmp->anon_vma = NULL;
 554                 } else if (anon_vma_fork(tmp, mpnt))
 555                         goto fail_nomem_anon_vma_fork;
 556                 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
 557                 file = tmp->vm_file;
 558                 if (file) {
 559                         struct inode *inode = file_inode(file);
 560                         struct address_space *mapping = file->f_mapping;
 561
 562                         get_file(file);
 563                         if (tmp->vm_flags & VM_DENYWRITE)
 564                                 put_write_access(inode);
 565                         i_mmap_lock_write(mapping);
 566                         if (tmp->vm_flags & VM_SHARED)
 567                                 mapping_allow_writable(mapping);
 568                         flush_dcache_mmap_lock(mapping);
 569                         /* insert tmp into the share list, just after mpnt */
 570                         vma_interval_tree_insert_after(tmp, mpnt,
 571                                         &mapping->i_mmap);
 572                         flush_dcache_mmap_unlock(mapping);
 573                         i_mmap_unlock_write(mapping);
 574                 }
 575
 576                 /*
 577                  * Clear hugetlb-related page reserves for children. This only
 578                  * affects MAP_PRIVATE mappings. Faults generated by the child
 579                  * are not guaranteed to succeed, even if read-only
 580                  */
 581                 if (is_vm_hugetlb_page(tmp))
 582                         reset_vma_resv_huge_pages(tmp);
 583
 584                 /*
 585                  * Link in the new vma and copy the page table entries.
 586                  */
 587                 *pprev = tmp;
 588                 pprev = &tmp->vm_next;
 589                 tmp->vm_prev = prev;
 590                 prev = tmp;
 591
 592                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
 593                 rb_link = &tmp->vm_rb.rb_right;
 594                 rb_parent = &tmp->vm_rb;
 595
 596                 mm->map_count++;
 597                 if (!(tmp->vm_flags & VM_WIPEONFORK))
 598                         retval = copy_page_range(tmp, mpnt);
 599
 600                 if (tmp->vm_ops && tmp->vm_ops->open)
 601                         tmp->vm_ops->open(tmp);
 602
 603                 if (retval)
 604                         goto out;
 605         }
 606         /* a new mm has just been created */
 607         retval = arch_dup_mmap(oldmm, mm);
 608 out:
 609         mmap_write_unlock(mm);
 610         flush_tlb_mm(oldmm);
 611         mmap_write_unlock(oldmm);
 612         dup_userfaultfd_complete(&uf);
 613 fail_uprobe_end:
 614         uprobe_end_dup_mmap();
 615         return retval;
 616 fail_nomem_anon_vma_fork:
 617         mpol_put(vma_policy(tmp));
 618 fail_nomem_policy:
 619         vm_area_free(tmp);
 620 fail_nomem:
 621         retval = -ENOMEM;
 622         vm_unacct_memory(charge);
 623         goto out;
 624 }
 625
 626 static inline int mm_alloc_pgd(struct mm_struct *mm)
 627 {
 628         mm->pgd = pgd_alloc(mm);
 629         if (unlikely(!mm->pgd))
 630                 return -ENOMEM;
 631         return 0;
 632 }
 633
 634 static inline void mm_free_pgd(struct mm_struct *mm)
 635 {
 636         pgd_free(mm, mm->pgd);
 637 }
 638 #else
 639 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
 640 {
 641         mmap_write_lock(oldmm);
 642         RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
 643         mmap_write_unlock(oldmm);
 644         return 0;
 645 }
 646 #define mm_alloc_pgd(mm)        (0)
 647 #define mm_free_pgd(mm)
 648 #endif /* CONFIG_MMU */
 649
 650 static void check_mm(struct mm_struct *mm)
 651 {
 652         int i;
 653
 654         BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
 655                          "Please make sure 'struct resident_page_types[]' is updated as well");
 656
 657         for (i = 0; i < NR_MM_COUNTERS; i++) {
 658                 long x = atomic_long_read(&mm->rss_stat.count[i]);
 659
 660                 if (unlikely(x))
 661                         pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
 662                                  mm, resident_page_types[i], x);
 663         }
 664
 665         if (mm_pgtables_bytes(mm))
 666                 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
 667                                 mm_pgtables_bytes(mm));
 668
 669 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
 670         VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
 671 #endif
 672 }
 673
 674 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 675 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
 676
 677 /*
 678  * Called when the last reference to the mm
 679  * is dropped: either by a lazy thread or by
 680  * mmput. Free the page directory and the mm.
 681  */
 682 void __mmdrop(struct mm_struct *mm)
 683 {
 684         BUG_ON(mm == &init_mm);
 685         WARN_ON_ONCE(mm == current->mm);
 686         WARN_ON_ONCE(mm == current->active_mm);
 687         mm_free_pgd(mm);
 688         destroy_context(mm);
 689         mmu_notifier_subscriptions_destroy(mm);
 690         check_mm(mm);
 691         put_user_ns(mm->user_ns);
 692         free_mm(mm);
 693 }
 694 EXPORT_SYMBOL_GPL(__mmdrop);
 695
 696 static void mmdrop_async_fn(struct work_struct *work)
 697 {
 698         struct mm_struct *mm;
 699
 700         mm = container_of(work, struct mm_struct, async_put_work);
 701         __mmdrop(mm);
 702 }
 703
 704 static void mmdrop_async(struct mm_struct *mm)
 705 {
 706         if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
 707                 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
 708                 schedule_work(&mm->async_put_work);
 709         }
 710 }
 711
 712 static inline void free_signal_struct(struct signal_struct *sig)
 713 {
 714         taskstats_tgid_free(sig);
 715         sched_autogroup_exit(sig);
 716         /*
 717          * __mmdrop is not safe to call from softirq context on x86 due to
 718          * pgd_dtor so postpone it to the async context
 719          */
 720         if (sig->oom_mm)
 721                 mmdrop_async(sig->oom_mm);
 722         kmem_cache_free(signal_cachep, sig);
 723 }
 724
 725 static inline void put_signal_struct(struct signal_struct *sig)
 726 {
 727         if (refcount_dec_and_test(&sig->sigcnt))
 728                 free_signal_struct(sig);
 729 }
 730
 731 void __put_task_struct(struct task_struct *tsk)
 732 {
 733         WARN_ON(!tsk->exit_state);
 734         WARN_ON(refcount_read(&tsk->usage));
 735         WARN_ON(tsk == current);
 736
 737         io_uring_free(tsk);
 738         cgroup_free(tsk);
 739         task_numa_free(tsk, true);
 740         security_task_free(tsk);
 741         bpf_task_storage_free(tsk);
 742         exit_creds(tsk);
 743         delayacct_tsk_free(tsk);
 744         put_signal_struct(tsk->signal);
 745
 746         if (!profile_handoff_task(tsk))
 747                 free_task(tsk);
 748 }
 749 EXPORT_SYMBOL_GPL(__put_task_struct);
 750
 751 void __init __weak arch_task_cache_init(void) { }
 752
 753 /*
 754  * set_max_threads
 755  */
 756 static void set_max_threads(unsigned int max_threads_suggested)
 757 {
 758         u64 threads;
 759         unsigned long nr_pages = totalram_pages();
 760
 761         /*
 762          * The number of threads shall be limited such that the thread
 763          * structures may only consume a small part of the available memory.
 764          */
 765         if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
 766                 threads = MAX_THREADS;
 767         else
 768                 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
 769                                     (u64) THREAD_SIZE * 8UL);
 770
 771         if (threads > max_threads_suggested)
 772                 threads = max_threads_suggested;
 773
 774         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 775 }
 776
 777 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 778 /* Initialized by the architecture: */
 779 int arch_task_struct_size __read_mostly;
 780 #endif
 781
 782 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 783 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 784 {
 785         /* Fetch thread_struct whitelist for the architecture. */
 786         arch_thread_struct_whitelist(offset, size);
 787
 788         /*
 789          * Handle zero-sized whitelist or empty thread_struct, otherwise
 790          * adjust offset to position of thread_struct in task_struct.
 791          */
 792         if (unlikely(*size == 0))
 793                 *offset = 0;
 794         else
 795                 *offset += offsetof(struct task_struct, thread);
 796 }
 797 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
 798
 799 void __init fork_init(void)
 800 {
 801         int i;
 802 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 803 #ifndef ARCH_MIN_TASKALIGN
 804 #define ARCH_MIN_TASKALIGN      0
 805 #endif
 806         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 807         unsigned long useroffset, usersize;
 808
 809         /* create a slab on which task_structs can be allocated */
 810         task_struct_whitelist(&useroffset, &usersize);
 811         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 812                         arch_task_struct_size, align,
 813                         SLAB_PANIC|SLAB_ACCOUNT,
 814                         useroffset, usersize, NULL);
 815 #endif
 816
 817         /* do the arch specific task caches init */
 818         arch_task_cache_init();
 819
 820         set_max_threads(MAX_THREADS);
 821
 822         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 823         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 824         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 825                 init_task.signal->rlim[RLIMIT_NPROC];
 826
 827         for (i = 0; i < UCOUNT_COUNTS; i++)
 828                 init_user_ns.ucount_max[i] = max_threads/2;
 829
 830 #ifdef CONFIG_VMAP_STACK
 831         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 832                           NULL, free_vm_stack_cache);
 833 #endif
 834
 835         scs_init();
 836
 837         lockdep_init_task(&init_task);
 838         uprobes_init();
 839 }
 840
 841 int __weak arch_dup_task_struct(struct task_struct *dst,
 842                                                struct task_struct *src)
 843 {
 844         *dst = *src;
 845         return 0;
 846 }
 847
 848 void set_task_stack_end_magic(struct task_struct *tsk)
 849 {
 850         unsigned long *stackend;
 851
 852         stackend = end_of_stack(tsk);
 853         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 854 }
 855
 856 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 857 {
 858         struct task_struct *tsk;
 859         unsigned long *stack;
 860         struct vm_struct *stack_vm_area __maybe_unused;
 861         int err;
 862
 863         if (node == NUMA_NO_NODE)
 864                 node = tsk_fork_get_node(orig);
 865         tsk = alloc_task_struct_node(node);
 866         if (!tsk)
 867                 return NULL;
 868
 869         stack = alloc_thread_stack_node(tsk, node);
 870         if (!stack)
 871                 goto free_tsk;
 872
 873         if (memcg_charge_kernel_stack(tsk))
 874                 goto free_stack;
 875
 876         stack_vm_area = task_stack_vm_area(tsk);
 877
 878         err = arch_dup_task_struct(tsk, orig);
 879
 880         /*
 881          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 882          * sure they're properly initialized before using any stack-related
 883          * functions again.
 884          */
 885         tsk->stack = stack;
 886 #ifdef CONFIG_VMAP_STACK
 887         tsk->stack_vm_area = stack_vm_area;
 888 #endif
 889 #ifdef CONFIG_THREAD_INFO_IN_TASK
 890         refcount_set(&tsk->stack_refcount, 1);
 891 #endif
 892
 893         if (err)
 894                 goto free_stack;
 895
 896         err = scs_prepare(tsk, node);
 897         if (err)
 898                 goto free_stack;
 899
 900 #ifdef CONFIG_SECCOMP
 901         /*
 902          * We must handle setting up seccomp filters once we're under
 903          * the sighand lock in case orig has changed between now and
 904          * then. Until then, filter must be NULL to avoid messing up
 905          * the usage counts on the error path calling free_task.
 906          */
 907         tsk->seccomp.filter = NULL;
 908 #endif
 909
 910         setup_thread_stack(tsk, orig);
 911         clear_user_return_notifier(tsk);
 912         clear_tsk_need_resched(tsk);
 913         set_task_stack_end_magic(tsk);
 914         clear_syscall_work_syscall_user_dispatch(tsk);
 915
 916 #ifdef CONFIG_STACKPROTECTOR
 917         tsk->stack_canary = get_random_canary();
 918 #endif
 919         if (orig->cpus_ptr == &orig->cpus_mask)
 920                 tsk->cpus_ptr = &tsk->cpus_mask;
 921
 922         /*
 923          * One for the user space visible state that goes away when reaped.
 924          * One for the scheduler.
 925          */
 926         refcount_set(&tsk->rcu_users, 2);
 927         /* One for the rcu users */
 928         refcount_set(&tsk->usage, 1);
 929 #ifdef CONFIG_BLK_DEV_IO_TRACE
 930         tsk->btrace_seq = 0;
 931 #endif
 932         tsk->splice_pipe = NULL;
 933         tsk->task_frag.page = NULL;
 934         tsk->wake_q.next = NULL;
 935         tsk->pf_io_worker = NULL;
 936
 937         account_kernel_stack(tsk, 1);
 938
 939         kcov_task_init(tsk);
 940         kmap_local_fork(tsk);
 941
 942 #ifdef CONFIG_FAULT_INJECTION
 943         tsk->fail_nth = 0;
 944 #endif
 945
 946 #ifdef CONFIG_BLK_CGROUP
 947         tsk->throttle_queue = NULL;
 948         tsk->use_memdelay = 0;
 949 #endif
 950
 951 #ifdef CONFIG_MEMCG
 952         tsk->active_memcg = NULL;
 953 #endif
 954         return tsk;
 955
 956 free_stack:
 957         free_thread_stack(tsk);
 958 free_tsk:
 959         free_task_struct(tsk);
 960         return NULL;
 961 }
 962
 963 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 964
 965 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 966
 967 static int __init coredump_filter_setup(char *s)
 968 {
 969         default_dump_filter =
 970                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 971                 MMF_DUMP_FILTER_MASK;
 972         return 1;
 973 }
 974
 975 __setup("coredump_filter=", coredump_filter_setup);
 976
 977 #include <linux/init_task.h>
 978
 979 static void mm_init_aio(struct mm_struct *mm)
 980 {
 981 #ifdef CONFIG_AIO
 982         spin_lock_init(&mm->ioctx_lock);
 983         mm->ioctx_table = NULL;
 984 #endif
 985 }
 986
 987 static __always_inline void mm_clear_owner(struct mm_struct *mm,
 988                                            struct task_struct *p)
 989 {
 990 #ifdef CONFIG_MEMCG
 991         if (mm->owner == p)
 992                 WRITE_ONCE(mm->owner, NULL);
 993 #endif
 994 }
 995
 996 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 997 {
 998 #ifdef CONFIG_MEMCG
 999         mm->owner = p;
1000 #endif
1001 }
1002
1003 static void mm_init_pasid(struct mm_struct *mm)
1004 {
1005 #ifdef CONFIG_IOMMU_SUPPORT
1006         mm->pasid = INIT_PASID;
1007 #endif
1008 }
1009
1010 static void mm_init_uprobes_state(struct mm_struct *mm)
1011 {
1012 #ifdef CONFIG_UPROBES
1013         mm->uprobes_state.xol_area = NULL;
1014 #endif
1015 }
1016
1017 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1018         struct user_namespace *user_ns)
1019 {
1020         mm->mmap = NULL;
1021         mm->mm_rb = RB_ROOT;
1022         mm->vmacache_seqnum = 0;
1023         atomic_set(&mm->mm_users, 1);
1024         atomic_set(&mm->mm_count, 1);
1025         seqcount_init(&mm->write_protect_seq);
1026         mmap_init_lock(mm);
1027         INIT_LIST_HEAD(&mm->mmlist);
1028         mm->core_state = NULL;
1029         mm_pgtables_bytes_init(mm);
1030         mm->map_count = 0;
1031         mm->locked_vm = 0;
1032         atomic_set(&mm->has_pinned, 0);
1033         atomic64_set(&mm->pinned_vm, 0);
1034         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1035         spin_lock_init(&mm->page_table_lock);
1036         spin_lock_init(&mm->arg_lock);
1037         mm_init_cpumask(mm);
1038         mm_init_aio(mm);
1039         mm_init_owner(mm, p);
1040         mm_init_pasid(mm);
1041         RCU_INIT_POINTER(mm->exe_file, NULL);
1042         mmu_notifier_subscriptions_init(mm);
1043         init_tlb_flush_pending(mm);
1044 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1045         mm->pmd_huge_pte = NULL;
1046 #endif
1047         mm_init_uprobes_state(mm);
1048
1049         if (current->mm) {
1050                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1051                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1052         } else {
1053                 mm->flags = default_dump_filter;
1054                 mm->def_flags = 0;
1055         }
1056
1057         if (mm_alloc_pgd(mm))
1058                 goto fail_nopgd;
1059
1060         if (init_new_context(p, mm))
1061                 goto fail_nocontext;
1062
1063         mm->user_ns = get_user_ns(user_ns);
1064         return mm;
1065
1066 fail_nocontext:
1067         mm_free_pgd(mm);
1068 fail_nopgd:
1069         free_mm(mm);
1070         return NULL;
1071 }
1072
1073 /*
1074  * Allocate and initialize an mm_struct.
1075  */
1076 struct mm_struct *mm_alloc(void)
1077 {
1078         struct mm_struct *mm;
1079
1080         mm = allocate_mm();
1081         if (!mm)
1082                 return NULL;
1083
1084         memset(mm, 0, sizeof(*mm));
1085         return mm_init(mm, current, current_user_ns());
1086 }
1087
1088 static inline void __mmput(struct mm_struct *mm)
1089 {
1090         VM_BUG_ON(atomic_read(&mm->mm_users));
1091
1092         uprobe_clear_state(mm);
1093         exit_aio(mm);
1094         ksm_exit(mm);
1095         khugepaged_exit(mm); /* must run before exit_mmap */
1096         exit_mmap(mm);
1097         mm_put_huge_zero_page(mm);
1098         set_mm_exe_file(mm, NULL);
1099         if (!list_empty(&mm->mmlist)) {
1100                 spin_lock(&mmlist_lock);
1101                 list_del(&mm->mmlist);
1102                 spin_unlock(&mmlist_lock);
1103         }
1104         if (mm->binfmt)
1105                 module_put(mm->binfmt->module);
1106         mmdrop(mm);
1107 }
1108
1109 /*
1110  * Decrement the use count and release all resources for an mm.
1111  */
1112 void mmput(struct mm_struct *mm)
1113 {
1114         might_sleep();
1115
1116         if (atomic_dec_and_test(&mm->mm_users))
1117                 __mmput(mm);
1118 }
1119 EXPORT_SYMBOL_GPL(mmput);
1120
1121 #ifdef CONFIG_MMU
1122 static void mmput_async_fn(struct work_struct *work)
1123 {
1124         struct mm_struct *mm = container_of(work, struct mm_struct,
1125                                             async_put_work);
1126
1127         __mmput(mm);
1128 }
1129
1130 void mmput_async(struct mm_struct *mm)
1131 {
1132         if (atomic_dec_and_test(&mm->mm_users)) {
1133                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1134                 schedule_work(&mm->async_put_work);
1135         }
1136 }
1137 #endif
1138
1139 /**
1140  * set_mm_exe_file - change a reference to the mm's executable file
1141  *
1142  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1143  *
1144  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1145  * invocations: in mmput() nobody alive left, in execve task is single
1146  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1147  * mm->exe_file, but does so without using set_mm_exe_file() in order
1148  * to avoid the need for any locks.
1149  */
1150 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1151 {
1152         struct file *old_exe_file;
1153
1154         /*
1155          * It is safe to dereference the exe_file without RCU as
1156          * this function is only called if nobody else can access
1157          * this mm -- see comment above for justification.
1158          */
1159         old_exe_file = rcu_dereference_raw(mm->exe_file);
1160
1161         if (new_exe_file)
1162                 get_file(new_exe_file);
1163         rcu_assign_pointer(mm->exe_file, new_exe_file);
1164         if (old_exe_file)
1165                 fput(old_exe_file);
1166 }
1167
1168 /**
1169  * get_mm_exe_file - acquire a reference to the mm's executable file
1170  *
1171  * Returns %NULL if mm has no associated executable file.
1172  * User must release file via fput().
1173  */
1174 struct file *get_mm_exe_file(struct mm_struct *mm)
1175 {
1176         struct file *exe_file;
1177
1178         rcu_read_lock();
1179         exe_file = rcu_dereference(mm->exe_file);
1180         if (exe_file && !get_file_rcu(exe_file))
1181                 exe_file = NULL;
1182         rcu_read_unlock();
1183         return exe_file;
1184 }
1185 EXPORT_SYMBOL(get_mm_exe_file);
1186
1187 /**
1188  * get_task_exe_file - acquire a reference to the task's executable file
1189  *
1190  * Returns %NULL if task's mm (if any) has no associated executable file or
1191  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1192  * User must release file via fput().
1193  */
1194 struct file *get_task_exe_file(struct task_struct *task)
1195 {
1196         struct file *exe_file = NULL;
1197         struct mm_struct *mm;
1198
1199         task_lock(task);
1200         mm = task->mm;
1201         if (mm) {
1202                 if (!(task->flags & PF_KTHREAD))
1203                         exe_file = get_mm_exe_file(mm);
1204         }
1205         task_unlock(task);
1206         return exe_file;
1207 }
1208 EXPORT_SYMBOL(get_task_exe_file);
1209
1210 /**
1211  * get_task_mm - acquire a reference to the task's mm
1212  *
1213  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1214  * this kernel workthread has transiently adopted a user mm with use_mm,
1215  * to do its AIO) is not set and if so returns a reference to it, after
1216  * bumping up the use count.  User must release the mm via mmput()
1217  * after use.  Typically used by /proc and ptrace.
1218  */
1219 struct mm_struct *get_task_mm(struct task_struct *task)
1220 {
1221         struct mm_struct *mm;
1222
1223         task_lock(task);
1224         mm = task->mm;
1225         if (mm) {
1226                 if (task->flags & PF_KTHREAD)
1227                         mm = NULL;
1228                 else
1229                         mmget(mm);
1230         }
1231         task_unlock(task);
1232         return mm;
1233 }
1234 EXPORT_SYMBOL_GPL(get_task_mm);
1235
1236 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1237 {
1238         struct mm_struct *mm;
1239         int err;
1240
1241         err =  down_read_killable(&task->signal->exec_update_lock);
1242         if (err)
1243                 return ERR_PTR(err);
1244
1245         mm = get_task_mm(task);
1246         if (mm && mm != current->mm &&
1247                         !ptrace_may_access(task, mode)) {
1248                 mmput(mm);
1249                 mm = ERR_PTR(-EACCES);
1250         }
1251         up_read(&task->signal->exec_update_lock);
1252
1253         return mm;
1254 }
1255
1256 static void complete_vfork_done(struct task_struct *tsk)
1257 {
1258         struct completion *vfork;
1259
1260         task_lock(tsk);
1261         vfork = tsk->vfork_done;
1262         if (likely(vfork)) {
1263                 tsk->vfork_done = NULL;
1264                 complete(vfork);
1265         }
1266         task_unlock(tsk);
1267 }
1268
1269 static int wait_for_vfork_done(struct task_struct *child,
1270                                 struct completion *vfork)
1271 {
1272         int killed;
1273
1274         freezer_do_not_count();
1275         cgroup_enter_frozen();
1276         killed = wait_for_completion_killable(vfork);
1277         cgroup_leave_frozen(false);
1278         freezer_count();
1279
1280         if (killed) {
1281                 task_lock(child);
1282                 child->vfork_done = NULL;
1283                 task_unlock(child);
1284         }
1285
1286         put_task_struct(child);
1287         return killed;
1288 }
1289
1290 /* Please note the differences between mmput and mm_release.
1291  * mmput is called whenever we stop holding onto a mm_struct,
1292  * error success whatever.
1293  *
1294  * mm_release is called after a mm_struct has been removed
1295  * from the current process.
1296  *
1297  * This difference is important for error handling, when we
1298  * only half set up a mm_struct for a new process and need to restore
1299  * the old one.  Because we mmput the new mm_struct before
1300  * restoring the old one. . .
1301  * Eric Biederman 10 January 1998
1302  */
1303 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1304 {
1305         uprobe_free_utask(tsk);
1306
1307         /* Get rid of any cached register state */
1308         deactivate_mm(tsk, mm);
1309
1310         /*
1311          * Signal userspace if we're not exiting with a core dump
1312          * because we want to leave the value intact for debugging
1313          * purposes.
1314          */
1315         if (tsk->clear_child_tid) {
1316                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1317                     atomic_read(&mm->mm_users) > 1) {
1318                         /*
1319                          * We don't check the error code - if userspace has
1320                          * not set up a proper pointer then tough luck.
1321                          */
1322                         put_user(0, tsk->clear_child_tid);
1323                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1324                                         1, NULL, NULL, 0, 0);
1325                 }
1326                 tsk->clear_child_tid = NULL;
1327         }
1328
1329         /*
1330          * All done, finally we can wake up parent and return this mm to him.
1331          * Also kthread_stop() uses this completion for synchronization.
1332          */
1333         if (tsk->vfork_done)
1334                 complete_vfork_done(tsk);
1335 }
1336
1337 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1338 {
1339         futex_exit_release(tsk);
1340         mm_release(tsk, mm);
1341 }
1342
1343 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1344 {
1345         futex_exec_release(tsk);
1346         mm_release(tsk, mm);
1347 }
1348
1349 /**
1350  * dup_mm() - duplicates an existing mm structure
1351  * @tsk: the task_struct with which the new mm will be associated.
1352  * @oldmm: the mm to duplicate.
1353  *
1354  * Allocates a new mm structure and duplicates the provided @oldmm structure
1355  * content into it.
1356  *
1357  * Return: the duplicated mm or NULL on failure.
1358  */
1359 static struct mm_struct *dup_mm(struct task_struct *tsk,
1360                                 struct mm_struct *oldmm)
1361 {
1362         struct mm_struct *mm;
1363         int err;
1364
1365         mm = allocate_mm();
1366         if (!mm)
1367                 goto fail_nomem;
1368
1369         memcpy(mm, oldmm, sizeof(*mm));
1370
1371         if (!mm_init(mm, tsk, mm->user_ns))
1372                 goto fail_nomem;
1373
1374         err = dup_mmap(mm, oldmm);
1375         if (err)
1376                 goto free_pt;
1377
1378         mm->hiwater_rss = get_mm_rss(mm);
1379         mm->hiwater_vm = mm->total_vm;
1380
1381         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1382                 goto free_pt;
1383
1384         return mm;
1385
1386 free_pt:
1387         /* don't put binfmt in mmput, we haven't got module yet */
1388         mm->binfmt = NULL;
1389         mm_init_owner(mm, NULL);
1390         mmput(mm);
1391
1392 fail_nomem:
1393         return NULL;
1394 }
1395
1396 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1397 {
1398         struct mm_struct *mm, *oldmm;
1399
1400         tsk->min_flt = tsk->maj_flt = 0;
1401         tsk->nvcsw = tsk->nivcsw = 0;
1402 #ifdef CONFIG_DETECT_HUNG_TASK
1403         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1404         tsk->last_switch_time = 0;
1405 #endif
1406
1407         tsk->mm = NULL;
1408         tsk->active_mm = NULL;
1409
1410         /*
1411          * Are we cloning a kernel thread?
1412          *
1413          * We need to steal a active VM for that..
1414          */
1415         oldmm = current->mm;
1416         if (!oldmm)
1417                 return 0;
1418
1419         /* initialize the new vmacache entries */
1420         vmacache_flush(tsk);
1421
1422         if (clone_flags & CLONE_VM) {
1423                 mmget(oldmm);
1424                 mm = oldmm;
1425         } else {
1426                 mm = dup_mm(tsk, current->mm);
1427                 if (!mm)
1428                         return -ENOMEM;
1429         }
1430
1431         tsk->mm = mm;
1432         tsk->active_mm = mm;
1433         return 0;
1434 }
1435
1436 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1437 {
1438         struct fs_struct *fs = current->fs;
1439         if (clone_flags & CLONE_FS) {
1440                 /* tsk->fs is already what we want */
1441                 spin_lock(&fs->lock);
1442                 if (fs->in_exec) {
1443                         spin_unlock(&fs->lock);
1444                         return -EAGAIN;
1445                 }
1446                 fs->users++;
1447                 spin_unlock(&fs->lock);
1448                 return 0;
1449         }
1450         tsk->fs = copy_fs_struct(fs);
1451         if (!tsk->fs)
1452                 return -ENOMEM;
1453         return 0;
1454 }
1455
1456 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1457 {
1458         struct files_struct *oldf, *newf;
1459         int error = 0;
1460
1461         /*
1462          * A background process may not have any files ...
1463          */
1464         oldf = current->files;
1465         if (!oldf)
1466                 goto out;
1467
1468         if (clone_flags & CLONE_FILES) {
1469                 atomic_inc(&oldf->count);
1470                 goto out;
1471         }
1472
1473         newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1474         if (!newf)
1475                 goto out;
1476
1477         tsk->files = newf;
1478         error = 0;
1479 out:
1480         return error;
1481 }
1482
1483 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1484 {
1485 #ifdef CONFIG_BLOCK
1486         struct io_context *ioc = current->io_context;
1487         struct io_context *new_ioc;
1488
1489         if (!ioc)
1490                 return 0;
1491         /*
1492          * Share io context with parent, if CLONE_IO is set
1493          */
1494         if (clone_flags & CLONE_IO) {
1495                 ioc_task_link(ioc);
1496                 tsk->io_context = ioc;
1497         } else if (ioprio_valid(ioc->ioprio)) {
1498                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1499                 if (unlikely(!new_ioc))
1500                         return -ENOMEM;
1501
1502                 new_ioc->ioprio = ioc->ioprio;
1503                 put_io_context(new_ioc);
1504         }
1505 #endif
1506         return 0;
1507 }
1508
1509 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1510 {
1511         struct sighand_struct *sig;
1512
1513         if (clone_flags & CLONE_SIGHAND) {
1514                 refcount_inc(&current->sighand->count);
1515                 return 0;
1516         }
1517         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1518         RCU_INIT_POINTER(tsk->sighand, sig);
1519         if (!sig)
1520                 return -ENOMEM;
1521
1522         refcount_set(&sig->count, 1);
1523         spin_lock_irq(&current->sighand->siglock);
1524         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1525         spin_unlock_irq(&current->sighand->siglock);
1526
1527         /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1528         if (clone_flags & CLONE_CLEAR_SIGHAND)
1529                 flush_signal_handlers(tsk, 0);
1530
1531         return 0;
1532 }
1533
1534 void __cleanup_sighand(struct sighand_struct *sighand)
1535 {
1536         if (refcount_dec_and_test(&sighand->count)) {
1537                 signalfd_cleanup(sighand);
1538                 /*
1539                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1540                  * without an RCU grace period, see __lock_task_sighand().
1541                  */
1542                 kmem_cache_free(sighand_cachep, sighand);
1543         }
1544 }
1545
1546 /*
1547  * Initialize POSIX timer handling for a thread group.
1548  */
1549 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1550 {
1551         struct posix_cputimers *pct = &sig->posix_cputimers;
1552         unsigned long cpu_limit;
1553
1554         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1555         posix_cputimers_group_init(pct, cpu_limit);
1556 }
1557
1558 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1559 {
1560         struct signal_struct *sig;
1561
1562         if (clone_flags & CLONE_THREAD)
1563                 return 0;
1564
1565         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1566         tsk->signal = sig;
1567         if (!sig)
1568                 return -ENOMEM;
1569
1570         sig->nr_threads = 1;
1571         atomic_set(&sig->live, 1);
1572         refcount_set(&sig->sigcnt, 1);
1573
1574         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1575         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1576         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1577
1578         init_waitqueue_head(&sig->wait_chldexit);
1579         sig->curr_target = tsk;
1580         init_sigpending(&sig->shared_pending);
1581         INIT_HLIST_HEAD(&sig->multiprocess);
1582         seqlock_init(&sig->stats_lock);
1583         prev_cputime_init(&sig->prev_cputime);
1584
1585 #ifdef CONFIG_POSIX_TIMERS
1586         INIT_LIST_HEAD(&sig->posix_timers);
1587         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1588         sig->real_timer.function = it_real_fn;
1589 #endif
1590
1591         task_lock(current->group_leader);
1592         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1593         task_unlock(current->group_leader);
1594
1595         posix_cpu_timers_init_group(sig);
1596
1597         tty_audit_fork(sig);
1598         sched_autogroup_fork(sig);
1599
1600         sig->oom_score_adj = current->signal->oom_score_adj;
1601         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1602
1603         mutex_init(&sig->cred_guard_mutex);
1604         init_rwsem(&sig->exec_update_lock);
1605
1606         return 0;
1607 }
1608
1609 static void copy_seccomp(struct task_struct *p)
1610 {
1611 #ifdef CONFIG_SECCOMP
1612         /*
1613          * Must be called with sighand->lock held, which is common to
1614          * all threads in the group. Holding cred_guard_mutex is not
1615          * needed because this new task is not yet running and cannot
1616          * be racing exec.
1617          */
1618         assert_spin_locked(&current->sighand->siglock);
1619
1620         /* Ref-count the new filter user, and assign it. */
1621         get_seccomp_filter(current);
1622         p->seccomp = current->seccomp;
1623
1624         /*
1625          * Explicitly enable no_new_privs here in case it got set
1626          * between the task_struct being duplicated and holding the
1627          * sighand lock. The seccomp state and nnp must be in sync.
1628          */
1629         if (task_no_new_privs(current))
1630                 task_set_no_new_privs(p);
1631
1632         /*
1633          * If the parent gained a seccomp mode after copying thread
1634          * flags and between before we held the sighand lock, we have
1635          * to manually enable the seccomp thread flag here.
1636          */
1637         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1638                 set_task_syscall_work(p, SECCOMP);
1639 #endif
1640 }
1641
1642 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1643 {
1644         current->clear_child_tid = tidptr;
1645
1646         return task_pid_vnr(current);
1647 }
1648
1649 static void rt_mutex_init_task(struct task_struct *p)
1650 {
1651         raw_spin_lock_init(&p->pi_lock);
1652 #ifdef CONFIG_RT_MUTEXES
1653         p->pi_waiters = RB_ROOT_CACHED;
1654         p->pi_top_task = NULL;
1655         p->pi_blocked_on = NULL;
1656 #endif
1657 }
1658
1659 static inline void init_task_pid_links(struct task_struct *task)
1660 {
1661         enum pid_type type;
1662
1663         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1664                 INIT_HLIST_NODE(&task->pid_links[type]);
1665 }
1666
1667 static inline void
1668 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1669 {
1670         if (type == PIDTYPE_PID)
1671                 task->thread_pid = pid;
1672         else
1673                 task->signal->pids[type] = pid;
1674 }
1675
1676 static inline void rcu_copy_process(struct task_struct *p)
1677 {
1678 #ifdef CONFIG_PREEMPT_RCU
1679         p->rcu_read_lock_nesting = 0;
1680         p->rcu_read_unlock_special.s = 0;
1681         p->rcu_blocked_node = NULL;
1682         INIT_LIST_HEAD(&p->rcu_node_entry);
1683 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1684 #ifdef CONFIG_TASKS_RCU
1685         p->rcu_tasks_holdout = false;
1686         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1687         p->rcu_tasks_idle_cpu = -1;
1688 #endif /* #ifdef CONFIG_TASKS_RCU */
1689 #ifdef CONFIG_TASKS_TRACE_RCU
1690         p->trc_reader_nesting = 0;
1691         p->trc_reader_special.s = 0;
1692         INIT_LIST_HEAD(&p->trc_holdout_list);
1693 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1694 }
1695
1696 struct pid *pidfd_pid(const struct file *file)
1697 {
1698         if (file->f_op == &pidfd_fops)
1699                 return file->private_data;
1700
1701         return ERR_PTR(-EBADF);
1702 }
1703
1704 static int pidfd_release(struct inode *inode, struct file *file)
1705 {
1706         struct pid *pid = file->private_data;
1707
1708         file->private_data = NULL;
1709         put_pid(pid);
1710         return 0;
1711 }
1712
1713 #ifdef CONFIG_PROC_FS
1714 /**
1715  * pidfd_show_fdinfo - print information about a pidfd
1716  * @m: proc fdinfo file
1717  * @f: file referencing a pidfd
1718  *
1719  * Pid:
1720  * This function will print the pid that a given pidfd refers to in the
1721  * pid namespace of the procfs instance.
1722  * If the pid namespace of the process is not a descendant of the pid
1723  * namespace of the procfs instance 0 will be shown as its pid. This is
1724  * similar to calling getppid() on a process whose parent is outside of
1725  * its pid namespace.
1726  *
1727  * NSpid:
1728  * If pid namespaces are supported then this function will also print
1729  * the pid of a given pidfd refers to for all descendant pid namespaces
1730  * starting from the current pid namespace of the instance, i.e. the
1731  * Pid field and the first entry in the NSpid field will be identical.
1732  * If the pid namespace of the process is not a descendant of the pid
1733  * namespace of the procfs instance 0 will be shown as its first NSpid
1734  * entry and no others will be shown.
1735  * Note that this differs from the Pid and NSpid fields in
1736  * /proc/<pid>/status where Pid and NSpid are always shown relative to
1737  * the  pid namespace of the procfs instance. The difference becomes
1738  * obvious when sending around a pidfd between pid namespaces from a
1739  * different branch of the tree, i.e. where no ancestral relation is
1740  * present between the pid namespaces:
1741  * - create two new pid namespaces ns1 and ns2 in the initial pid
1742  *   namespace (also take care to create new mount namespaces in the
1743  *   new pid namespace and mount procfs)
1744  * - create a process with a pidfd in ns1
1745  * - send pidfd from ns1 to ns2
1746  * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1747  *   have exactly one entry, which is 0
1748  */
1749 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1750 {
1751         struct pid *pid = f->private_data;
1752         struct pid_namespace *ns;
1753         pid_t nr = -1;
1754
1755         if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1756                 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1757                 nr = pid_nr_ns(pid, ns);
1758         }
1759
1760         seq_put_decimal_ll(m, "Pid:\t", nr);
1761
1762 #ifdef CONFIG_PID_NS
1763         seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1764         if (nr > 0) {
1765                 int i;
1766
1767                 /* If nr is non-zero it means that 'pid' is valid and that
1768                  * ns, i.e. the pid namespace associated with the procfs
1769                  * instance, is in the pid namespace hierarchy of pid.
1770                  * Start at one below the already printed level.
1771                  */
1772                 for (i = ns->level + 1; i <= pid->level; i++)
1773                         seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1774         }
1775 #endif
1776         seq_putc(m, '\n');
1777 }
1778 #endif
1779
1780 /*
1781  * Poll support for process exit notification.
1782  */
1783 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1784 {
1785         struct pid *pid = file->private_data;
1786         __poll_t poll_flags = 0;
1787
1788         poll_wait(file, &pid->wait_pidfd, pts);
1789
1790         /*
1791          * Inform pollers only when the whole thread group exits.
1792          * If the thread group leader exits before all other threads in the
1793          * group, then poll(2) should block, similar to the wait(2) family.
1794          */
1795         if (thread_group_exited(pid))
1796                 poll_flags = EPOLLIN | EPOLLRDNORM;
1797
1798         return poll_flags;
1799 }
1800
1801 const struct file_operations pidfd_fops = {
1802         .release = pidfd_release,
1803         .poll = pidfd_poll,
1804 #ifdef CONFIG_PROC_FS
1805         .show_fdinfo = pidfd_show_fdinfo,
1806 #endif
1807 };
1808
1809 static void __delayed_free_task(struct rcu_head *rhp)
1810 {
1811         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1812
1813         free_task(tsk);
1814 }
1815
1816 static __always_inline void delayed_free_task(struct task_struct *tsk)
1817 {
1818         if (IS_ENABLED(CONFIG_MEMCG))
1819                 call_rcu(&tsk->rcu, __delayed_free_task);
1820         else
1821                 free_task(tsk);
1822 }
1823
1824 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1825 {
1826         /* Skip if kernel thread */
1827         if (!tsk->mm)
1828                 return;
1829
1830         /* Skip if spawning a thread or using vfork */
1831         if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1832                 return;
1833
1834         /* We need to synchronize with __set_oom_adj */
1835         mutex_lock(&oom_adj_mutex);
1836         set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1837         /* Update the values in case they were changed after copy_signal */
1838         tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1839         tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1840         mutex_unlock(&oom_adj_mutex);
1841 }
1842
1843 /*
1844  * This creates a new process as a copy of the old one,
1845  * but does not actually start it yet.
1846  *
1847  * It copies the registers, and all the appropriate
1848  * parts of the process environment (as per the clone
1849  * flags). The actual kick-off is left to the caller.
1850  */
1851 static __latent_entropy struct task_struct *copy_process(
1852                                         struct pid *pid,
1853                                         int trace,
1854                                         int node,
1855                                         struct kernel_clone_args *args)
1856 {
1857         int pidfd = -1, retval;
1858         struct task_struct *p;
1859         struct multiprocess_signals delayed;
1860         struct file *pidfile = NULL;
1861         u64 clone_flags = args->flags;
1862         struct nsproxy *nsp = current->nsproxy;
1863
1864         /*
1865          * Don't allow sharing the root directory with processes in a different
1866          * namespace
1867          */
1868         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1869                 return ERR_PTR(-EINVAL);
1870
1871         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1872                 return ERR_PTR(-EINVAL);
1873
1874         /*
1875          * Thread groups must share signals as well, and detached threads
1876          * can only be started up within the thread group.
1877          */
1878         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1879                 return ERR_PTR(-EINVAL);
1880
1881         /*
1882          * Shared signal handlers imply shared VM. By way of the above,
1883          * thread groups also imply shared VM. Blocking this case allows
1884          * for various simplifications in other code.
1885          */
1886         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1887                 return ERR_PTR(-EINVAL);
1888
1889         /*
1890          * Siblings of global init remain as zombies on exit since they are
1891          * not reaped by their parent (swapper). To solve this and to avoid
1892          * multi-rooted process trees, prevent global and container-inits
1893          * from creating siblings.
1894          */
1895         if ((clone_flags & CLONE_PARENT) &&
1896                                 current->signal->flags & SIGNAL_UNKILLABLE)
1897                 return ERR_PTR(-EINVAL);
1898
1899         /*
1900          * If the new process will be in a different pid or user namespace
1901          * do not allow it to share a thread group with the forking task.
1902          */
1903         if (clone_flags & CLONE_THREAD) {
1904                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1905                     (task_active_pid_ns(current) != nsp->pid_ns_for_children))
1906                         return ERR_PTR(-EINVAL);
1907         }
1908
1909         /*
1910          * If the new process will be in a different time namespace
1911          * do not allow it to share VM or a thread group with the forking task.
1912          */
1913         if (clone_flags & (CLONE_THREAD | CLONE_VM)) {
1914                 if (nsp->time_ns != nsp->time_ns_for_children)
1915                         return ERR_PTR(-EINVAL);
1916         }
1917
1918         if (clone_flags & CLONE_PIDFD) {
1919                 /*
1920                  * - CLONE_DETACHED is blocked so that we can potentially
1921                  *   reuse it later for CLONE_PIDFD.
1922                  * - CLONE_THREAD is blocked until someone really needs it.
1923                  */
1924                 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1925                         return ERR_PTR(-EINVAL);
1926         }
1927
1928         /*
1929          * Force any signals received before this point to be delivered
1930          * before the fork happens.  Collect up signals sent to multiple
1931          * processes that happen during the fork and delay them so that
1932          * they appear to happen after the fork.
1933          */
1934         sigemptyset(&delayed.signal);
1935         INIT_HLIST_NODE(&delayed.node);
1936
1937         spin_lock_irq(&current->sighand->siglock);
1938         if (!(clone_flags & CLONE_THREAD))
1939                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1940         recalc_sigpending();
1941         spin_unlock_irq(&current->sighand->siglock);
1942         retval = -ERESTARTNOINTR;
1943         if (task_sigpending(current))
1944                 goto fork_out;
1945
1946         retval = -ENOMEM;
1947         p = dup_task_struct(current, node);
1948         if (!p)
1949                 goto fork_out;
1950         if (args->io_thread) {
1951                 /*
1952                  * Mark us an IO worker, and block any signal that isn't
1953                  * fatal or STOP
1954                  */
1955                 p->flags |= PF_IO_WORKER;
1956                 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
1957         }
1958
1959         /*
1960          * This _must_ happen before we call free_task(), i.e. before we jump
1961          * to any of the bad_fork_* labels. This is to avoid freeing
1962          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1963          * kernel threads (PF_KTHREAD).
1964          */
1965         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1966         /*
1967          * Clear TID on mm_release()?
1968          */
1969         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1970
1971         ftrace_graph_init_task(p);
1972
1973         rt_mutex_init_task(p);
1974
1975         lockdep_assert_irqs_enabled();
1976 #ifdef CONFIG_PROVE_LOCKING
1977         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1978 #endif
1979         retval = -EAGAIN;
1980         if (atomic_read(&p->real_cred->user->processes) >=
1981                         task_rlimit(p, RLIMIT_NPROC)) {
1982                 if (p->real_cred->user != INIT_USER &&
1983                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1984                         goto bad_fork_free;
1985         }
1986         current->flags &= ~PF_NPROC_EXCEEDED;
1987
1988         retval = copy_creds(p, clone_flags);
1989         if (retval < 0)
1990                 goto bad_fork_free;
1991
1992         /*
1993          * If multiple threads are within copy_process(), then this check
1994          * triggers too late. This doesn't hurt, the check is only there
1995          * to stop root fork bombs.
1996          */
1997         retval = -EAGAIN;
1998         if (data_race(nr_threads >= max_threads))
1999                 goto bad_fork_cleanup_count;
2000
2001         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
2002         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
2003         p->flags |= PF_FORKNOEXEC;
2004         INIT_LIST_HEAD(&p->children);
2005         INIT_LIST_HEAD(&p->sibling);
2006         rcu_copy_process(p);
2007         p->vfork_done = NULL;
2008         spin_lock_init(&p->alloc_lock);
2009
2010         init_sigpending(&p->pending);
2011         p->sigqueue_cache = NULL;
2012
2013         p->utime = p->stime = p->gtime = 0;
2014 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2015         p->utimescaled = p->stimescaled = 0;
2016 #endif
2017         prev_cputime_init(&p->prev_cputime);
2018
2019 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2020         seqcount_init(&p->vtime.seqcount);
2021         p->vtime.starttime = 0;
2022         p->vtime.state = VTIME_INACTIVE;
2023 #endif
2024
2025 #ifdef CONFIG_IO_URING
2026         p->io_uring = NULL;
2027 #endif
2028
2029 #if defined(SPLIT_RSS_COUNTING)
2030         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2031 #endif
2032
2033         p->default_timer_slack_ns = current->timer_slack_ns;
2034
2035 #ifdef CONFIG_PSI
2036         p->psi_flags = 0;
2037 #endif
2038
2039         task_io_accounting_init(&p->ioac);
2040         acct_clear_integrals(p);
2041
2042         posix_cputimers_init(&p->posix_cputimers);
2043
2044         p->io_context = NULL;
2045         audit_set_context(p, NULL);
2046         cgroup_fork(p);
2047 #ifdef CONFIG_NUMA
2048         p->mempolicy = mpol_dup(p->mempolicy);
2049         if (IS_ERR(p->mempolicy)) {
2050                 retval = PTR_ERR(p->mempolicy);
2051                 p->mempolicy = NULL;
2052                 goto bad_fork_cleanup_threadgroup_lock;
2053         }
2054 #endif
2055 #ifdef CONFIG_CPUSETS
2056         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2057         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2058         seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2059 #endif
2060 #ifdef CONFIG_TRACE_IRQFLAGS
2061         memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2062         p->irqtrace.hardirq_disable_ip  = _THIS_IP_;
2063         p->irqtrace.softirq_enable_ip   = _THIS_IP_;
2064         p->softirqs_enabled             = 1;
2065         p->softirq_context              = 0;
2066 #endif
2067
2068         p->pagefault_disabled = 0;
2069
2070 #ifdef CONFIG_LOCKDEP
2071         lockdep_init_task(p);
2072 #endif
2073
2074 #ifdef CONFIG_DEBUG_MUTEXES
2075         p->blocked_on = NULL; /* not blocked yet */
2076 #endif
2077 #ifdef CONFIG_BCACHE
2078         p->sequential_io        = 0;
2079         p->sequential_io_avg    = 0;
2080 #endif
2081 #ifdef CONFIG_BPF_SYSCALL
2082         RCU_INIT_POINTER(p->bpf_storage, NULL);
2083 #endif
2084
2085         /* Perform scheduler related setup. Assign this task to a CPU. */
2086         retval = sched_fork(clone_flags, p);
2087         if (retval)
2088                 goto bad_fork_cleanup_policy;
2089
2090         retval = perf_event_init_task(p, clone_flags);
2091         if (retval)
2092                 goto bad_fork_cleanup_policy;
2093         retval = audit_alloc(p);
2094         if (retval)
2095                 goto bad_fork_cleanup_perf;
2096         /* copy all the process information */
2097         shm_init_task(p);
2098         retval = security_task_alloc(p, clone_flags);
2099         if (retval)
2100                 goto bad_fork_cleanup_audit;
2101         retval = copy_semundo(clone_flags, p);
2102         if (retval)
2103                 goto bad_fork_cleanup_security;
2104         retval = copy_files(clone_flags, p);
2105         if (retval)
2106                 goto bad_fork_cleanup_semundo;
2107         retval = copy_fs(clone_flags, p);
2108         if (retval)
2109                 goto bad_fork_cleanup_files;
2110         retval = copy_sighand(clone_flags, p);
2111         if (retval)
2112                 goto bad_fork_cleanup_fs;
2113         retval = copy_signal(clone_flags, p);
2114         if (retval)
2115                 goto bad_fork_cleanup_sighand;
2116         retval = copy_mm(clone_flags, p);
2117         if (retval)
2118                 goto bad_fork_cleanup_signal;
2119         retval = copy_namespaces(clone_flags, p);
2120         if (retval)
2121                 goto bad_fork_cleanup_mm;
2122         retval = copy_io(clone_flags, p);
2123         if (retval)
2124                 goto bad_fork_cleanup_namespaces;
2125         retval = copy_thread(clone_flags, args->stack, args->stack_size, p, args->tls);
2126         if (retval)
2127                 goto bad_fork_cleanup_io;
2128
2129         stackleak_task_init(p);
2130
2131         if (pid != &init_struct_pid) {
2132                 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2133                                 args->set_tid_size);
2134                 if (IS_ERR(pid)) {
2135                         retval = PTR_ERR(pid);
2136                         goto bad_fork_cleanup_thread;
2137                 }
2138         }
2139
2140         /*
2141          * This has to happen after we've potentially unshared the file
2142          * descriptor table (so that the pidfd doesn't leak into the child
2143          * if the fd table isn't shared).
2144          */
2145         if (clone_flags & CLONE_PIDFD) {
2146                 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2147                 if (retval < 0)
2148                         goto bad_fork_free_pid;
2149
2150                 pidfd = retval;
2151
2152                 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2153                                               O_RDWR | O_CLOEXEC);
2154                 if (IS_ERR(pidfile)) {
2155                         put_unused_fd(pidfd);
2156                         retval = PTR_ERR(pidfile);
2157                         goto bad_fork_free_pid;
2158                 }
2159                 get_pid(pid);   /* held by pidfile now */
2160
2161                 retval = put_user(pidfd, args->pidfd);
2162                 if (retval)
2163                         goto bad_fork_put_pidfd;
2164         }
2165
2166 #ifdef CONFIG_BLOCK
2167         p->plug = NULL;
2168 #endif
2169         futex_init_task(p);
2170
2171         /*
2172          * sigaltstack should be cleared when sharing the same VM
2173          */
2174         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2175                 sas_ss_reset(p);
2176
2177         /*
2178          * Syscall tracing and stepping should be turned off in the
2179          * child regardless of CLONE_PTRACE.
2180          */
2181         user_disable_single_step(p);
2182         clear_task_syscall_work(p, SYSCALL_TRACE);
2183 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2184         clear_task_syscall_work(p, SYSCALL_EMU);
2185 #endif
2186         clear_tsk_latency_tracing(p);
2187
2188         /* ok, now we should be set up.. */
2189         p->pid = pid_nr(pid);
2190         if (clone_flags & CLONE_THREAD) {
2191                 p->group_leader = current->group_leader;
2192                 p->tgid = current->tgid;
2193         } else {
2194                 p->group_leader = p;
2195                 p->tgid = p->pid;
2196         }
2197
2198         p->nr_dirtied = 0;
2199         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2200         p->dirty_paused_when = 0;
2201
2202         p->pdeath_signal = 0;
2203         INIT_LIST_HEAD(&p->thread_group);
2204         p->task_works = NULL;
2205
2206 #ifdef CONFIG_KRETPROBES
2207         p->kretprobe_instances.first = NULL;
2208 #endif
2209
2210         /*
2211          * Ensure that the cgroup subsystem policies allow the new process to be
2212          * forked. It should be noted that the new process's css_set can be changed
2213          * between here and cgroup_post_fork() if an organisation operation is in
2214          * progress.
2215          */
2216         retval = cgroup_can_fork(p, args);
2217         if (retval)
2218                 goto bad_fork_put_pidfd;
2219
2220         /*
2221          * From this point on we must avoid any synchronous user-space
2222          * communication until we take the tasklist-lock. In particular, we do
2223          * not want user-space to be able to predict the process start-time by
2224          * stalling fork(2) after we recorded the start_time but before it is
2225          * visible to the system.
2226          */
2227
2228         p->start_time = ktime_get_ns();
2229         p->start_boottime = ktime_get_boottime_ns();
2230
2231         /*
2232          * Make it visible to the rest of the system, but dont wake it up yet.
2233          * Need tasklist lock for parent etc handling!
2234          */
2235         write_lock_irq(&tasklist_lock);
2236
2237         /* CLONE_PARENT re-uses the old parent */
2238         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2239                 p->real_parent = current->real_parent;
2240                 p->parent_exec_id = current->parent_exec_id;
2241                 if (clone_flags & CLONE_THREAD)
2242                         p->exit_signal = -1;
2243                 else
2244                         p->exit_signal = current->group_leader->exit_signal;
2245         } else {
2246                 p->real_parent = current;
2247                 p->parent_exec_id = current->self_exec_id;
2248                 p->exit_signal = args->exit_signal;
2249         }
2250
2251         klp_copy_process(p);
2252
2253         spin_lock(&current->sighand->siglock);
2254
2255         /*
2256          * Copy seccomp details explicitly here, in case they were changed
2257          * before holding sighand lock.
2258          */
2259         copy_seccomp(p);
2260
2261         rseq_fork(p, clone_flags);
2262
2263         /* Don't start children in a dying pid namespace */
2264         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2265                 retval = -ENOMEM;
2266                 goto bad_fork_cancel_cgroup;
2267         }
2268
2269         /* Let kill terminate clone/fork in the middle */
2270         if (fatal_signal_pending(current)) {
2271                 retval = -EINTR;
2272                 goto bad_fork_cancel_cgroup;
2273         }
2274
2275         /* past the last point of failure */
2276         if (pidfile)
2277                 fd_install(pidfd, pidfile);
2278
2279         init_task_pid_links(p);
2280         if (likely(p->pid)) {
2281                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2282
2283                 init_task_pid(p, PIDTYPE_PID, pid);
2284                 if (thread_group_leader(p)) {
2285                         init_task_pid(p, PIDTYPE_TGID, pid);
2286                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2287                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2288
2289                         if (is_child_reaper(pid)) {
2290                                 ns_of_pid(pid)->child_reaper = p;
2291                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2292                         }
2293                         p->signal->shared_pending.signal = delayed.signal;
2294                         p->signal->tty = tty_kref_get(current->signal->tty);
2295                         /*
2296                          * Inherit has_child_subreaper flag under the same
2297                          * tasklist_lock with adding child to the process tree
2298                          * for propagate_has_child_subreaper optimization.
2299                          */
2300                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2301                                                          p->real_parent->signal->is_child_subreaper;
2302                         list_add_tail(&p->sibling, &p->real_parent->children);
2303                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2304                         attach_pid(p, PIDTYPE_TGID);
2305                         attach_pid(p, PIDTYPE_PGID);
2306                         attach_pid(p, PIDTYPE_SID);
2307                         __this_cpu_inc(process_counts);
2308                 } else {
2309                         current->signal->nr_threads++;
2310                         atomic_inc(&current->signal->live);
2311                         refcount_inc(&current->signal->sigcnt);
2312                         task_join_group_stop(p);
2313                         list_add_tail_rcu(&p->thread_group,
2314                                           &p->group_leader->thread_group);
2315                         list_add_tail_rcu(&p->thread_node,
2316                                           &p->signal->thread_head);
2317                 }
2318                 attach_pid(p, PIDTYPE_PID);
2319                 nr_threads++;
2320         }
2321         total_forks++;
2322         hlist_del_init(&delayed.node);
2323         spin_unlock(&current->sighand->siglock);
2324         syscall_tracepoint_update(p);
2325         write_unlock_irq(&tasklist_lock);
2326
2327         proc_fork_connector(p);
2328         sched_post_fork(p);
2329         cgroup_post_fork(p, args);
2330         perf_event_fork(p);
2331
2332         trace_task_newtask(p, clone_flags);
2333         uprobe_copy_process(p, clone_flags);
2334
2335         copy_oom_score_adj(clone_flags, p);
2336
2337         return p;
2338
2339 bad_fork_cancel_cgroup:
2340         spin_unlock(&current->sighand->siglock);
2341         write_unlock_irq(&tasklist_lock);
2342         cgroup_cancel_fork(p, args);
2343 bad_fork_put_pidfd:
2344         if (clone_flags & CLONE_PIDFD) {
2345                 fput(pidfile);
2346                 put_unused_fd(pidfd);
2347         }
2348 bad_fork_free_pid:
2349         if (pid != &init_struct_pid)
2350                 free_pid(pid);
2351 bad_fork_cleanup_thread:
2352         exit_thread(p);
2353 bad_fork_cleanup_io:
2354         if (p->io_context)
2355                 exit_io_context(p);
2356 bad_fork_cleanup_namespaces:
2357         exit_task_namespaces(p);
2358 bad_fork_cleanup_mm:
2359         if (p->mm) {
2360                 mm_clear_owner(p->mm, p);
2361                 mmput(p->mm);
2362         }
2363 bad_fork_cleanup_signal:
2364         if (!(clone_flags & CLONE_THREAD))
2365                 free_signal_struct(p->signal);
2366 bad_fork_cleanup_sighand:
2367         __cleanup_sighand(p->sighand);
2368 bad_fork_cleanup_fs:
2369         exit_fs(p); /* blocking */
2370 bad_fork_cleanup_files:
2371         exit_files(p); /* blocking */
2372 bad_fork_cleanup_semundo:
2373         exit_sem(p);
2374 bad_fork_cleanup_security:
2375         security_task_free(p);
2376 bad_fork_cleanup_audit:
2377         audit_free(p);
2378 bad_fork_cleanup_perf:
2379         perf_event_free_task(p);
2380 bad_fork_cleanup_policy:
2381         lockdep_free_task(p);
2382 #ifdef CONFIG_NUMA
2383         mpol_put(p->mempolicy);
2384 bad_fork_cleanup_threadgroup_lock:
2385 #endif
2386         delayacct_tsk_free(p);
2387 bad_fork_cleanup_count:
2388         atomic_dec(&p->cred->user->processes);
2389         exit_creds(p);
2390 bad_fork_free:
2391         p->state = TASK_DEAD;
2392         put_task_stack(p);
2393         delayed_free_task(p);
2394 fork_out:
2395         spin_lock_irq(&current->sighand->siglock);
2396         hlist_del_init(&delayed.node);
2397         spin_unlock_irq(&current->sighand->siglock);
2398         return ERR_PTR(retval);
2399 }
2400
2401 static inline void init_idle_pids(struct task_struct *idle)
2402 {
2403         enum pid_type type;
2404
2405         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2406                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2407                 init_task_pid(idle, type, &init_struct_pid);
2408         }
2409 }
2410
2411 struct task_struct *fork_idle(int cpu)
2412 {
2413         struct task_struct *task;
2414         struct kernel_clone_args args = {
2415                 .flags = CLONE_VM,
2416         };
2417
2418         task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2419         if (!IS_ERR(task)) {
2420                 init_idle_pids(task);
2421                 init_idle(task, cpu);
2422         }
2423
2424         return task;
2425 }
2426
2427 struct mm_struct *copy_init_mm(void)
2428 {
2429         return dup_mm(NULL, &init_mm);
2430 }
2431
2432 /*
2433  * This is like kernel_clone(), but shaved down and tailored to just
2434  * creating io_uring workers. It returns a created task, or an error pointer.
2435  * The returned task is inactive, and the caller must fire it up through
2436  * wake_up_new_task(p). All signals are blocked in the created task.
2437  */
2438 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2439 {
2440         unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2441                                 CLONE_IO;
2442         struct kernel_clone_args args = {
2443                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2444                                     CLONE_UNTRACED) & ~CSIGNAL),
2445                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2446                 .stack          = (unsigned long)fn,
2447                 .stack_size     = (unsigned long)arg,
2448                 .io_thread      = 1,
2449         };
2450
2451         return copy_process(NULL, 0, node, &args);
2452 }
2453
2454 /*
2455  *  Ok, this is the main fork-routine.
2456  *
2457  * It copies the process, and if successful kick-starts
2458  * it and waits for it to finish using the VM if required.
2459  *
2460  * args->exit_signal is expected to be checked for sanity by the caller.
2461  */
2462 pid_t kernel_clone(struct kernel_clone_args *args)
2463 {
2464         u64 clone_flags = args->flags;
2465         struct completion vfork;
2466         struct pid *pid;
2467         struct task_struct *p;
2468         int trace = 0;
2469         pid_t nr;
2470
2471         /*
2472          * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2473          * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2474          * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2475          * field in struct clone_args and it still doesn't make sense to have
2476          * them both point at the same memory location. Performing this check
2477          * here has the advantage that we don't need to have a separate helper
2478          * to check for legacy clone().
2479          */
2480         if ((args->flags & CLONE_PIDFD) &&
2481             (args->flags & CLONE_PARENT_SETTID) &&
2482             (args->pidfd == args->parent_tid))
2483                 return -EINVAL;
2484
2485         /*
2486          * Determine whether and which event to report to ptracer.  When
2487          * called from kernel_thread or CLONE_UNTRACED is explicitly
2488          * requested, no event is reported; otherwise, report if the event
2489          * for the type of forking is enabled.
2490          */
2491         if (!(clone_flags & CLONE_UNTRACED)) {
2492                 if (clone_flags & CLONE_VFORK)
2493                         trace = PTRACE_EVENT_VFORK;
2494                 else if (args->exit_signal != SIGCHLD)
2495                         trace = PTRACE_EVENT_CLONE;
2496                 else
2497                         trace = PTRACE_EVENT_FORK;
2498
2499                 if (likely(!ptrace_event_enabled(current, trace)))
2500                         trace = 0;
2501         }
2502
2503         p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2504         add_latent_entropy();
2505
2506         if (IS_ERR(p))
2507                 return PTR_ERR(p);
2508
2509         /*
2510          * Do this prior waking up the new thread - the thread pointer
2511          * might get invalid after that point, if the thread exits quickly.
2512          */
2513         trace_sched_process_fork(current, p);
2514
2515         pid = get_task_pid(p, PIDTYPE_PID);
2516         nr = pid_vnr(pid);
2517
2518         if (clone_flags & CLONE_PARENT_SETTID)
2519                 put_user(nr, args->parent_tid);
2520
2521         if (clone_flags & CLONE_VFORK) {
2522                 p->vfork_done = &vfork;
2523                 init_completion(&vfork);
2524                 get_task_struct(p);
2525         }
2526
2527         wake_up_new_task(p);
2528
2529         /* forking complete and child started to run, tell ptracer */
2530         if (unlikely(trace))
2531                 ptrace_event_pid(trace, pid);
2532
2533         if (clone_flags & CLONE_VFORK) {
2534                 if (!wait_for_vfork_done(p, &vfork))
2535                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2536         }
2537
2538         put_pid(pid);
2539         return nr;
2540 }
2541
2542 /*
2543  * Create a kernel thread.
2544  */
2545 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2546 {
2547         struct kernel_clone_args args = {
2548                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2549                                     CLONE_UNTRACED) & ~CSIGNAL),
2550                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2551                 .stack          = (unsigned long)fn,
2552                 .stack_size     = (unsigned long)arg,
2553         };
2554
2555         return kernel_clone(&args);
2556 }
2557
2558 #ifdef __ARCH_WANT_SYS_FORK
2559 SYSCALL_DEFINE0(fork)
2560 {
2561 #ifdef CONFIG_MMU
2562         struct kernel_clone_args args = {
2563                 .exit_signal = SIGCHLD,
2564         };
2565
2566         return kernel_clone(&args);
2567 #else
2568         /* can not support in nommu mode */
2569         return -EINVAL;
2570 #endif
2571 }
2572 #endif
2573
2574 #ifdef __ARCH_WANT_SYS_VFORK
2575 SYSCALL_DEFINE0(vfork)
2576 {
2577         struct kernel_clone_args args = {
2578                 .flags          = CLONE_VFORK | CLONE_VM,
2579                 .exit_signal    = SIGCHLD,
2580         };
2581
2582         return kernel_clone(&args);
2583 }
2584 #endif
2585
2586 #ifdef __ARCH_WANT_SYS_CLONE
2587 #ifdef CONFIG_CLONE_BACKWARDS
2588 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2589                  int __user *, parent_tidptr,
2590                  unsigned long, tls,
2591                  int __user *, child_tidptr)
2592 #elif defined(CONFIG_CLONE_BACKWARDS2)
2593 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2594                  int __user *, parent_tidptr,
2595                  int __user *, child_tidptr,
2596                  unsigned long, tls)
2597 #elif defined(CONFIG_CLONE_BACKWARDS3)
2598 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2599                 int, stack_size,
2600                 int __user *, parent_tidptr,
2601                 int __user *, child_tidptr,
2602                 unsigned long, tls)
2603 #else
2604 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2605                  int __user *, parent_tidptr,
2606                  int __user *, child_tidptr,
2607                  unsigned long, tls)
2608 #endif
2609 {
2610         struct kernel_clone_args args = {
2611                 .flags          = (lower_32_bits(clone_flags) & ~CSIGNAL),
2612                 .pidfd          = parent_tidptr,
2613                 .child_tid      = child_tidptr,
2614                 .parent_tid     = parent_tidptr,
2615                 .exit_signal    = (lower_32_bits(clone_flags) & CSIGNAL),
2616                 .stack          = newsp,
2617                 .tls            = tls,
2618         };
2619
2620         return kernel_clone(&args);
2621 }
2622 #endif
2623
2624 #ifdef __ARCH_WANT_SYS_CLONE3
2625
2626 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2627                                               struct clone_args __user *uargs,
2628                                               size_t usize)
2629 {
2630         int err;
2631         struct clone_args args;
2632         pid_t *kset_tid = kargs->set_tid;
2633
2634         BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2635                      CLONE_ARGS_SIZE_VER0);
2636         BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2637                      CLONE_ARGS_SIZE_VER1);
2638         BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2639                      CLONE_ARGS_SIZE_VER2);
2640         BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2641
2642         if (unlikely(usize > PAGE_SIZE))
2643                 return -E2BIG;
2644         if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2645                 return -EINVAL;
2646
2647         err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2648         if (err)
2649                 return err;
2650
2651         if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2652                 return -EINVAL;
2653
2654         if (unlikely(!args.set_tid && args.set_tid_size > 0))
2655                 return -EINVAL;
2656
2657         if (unlikely(args.set_tid && args.set_tid_size == 0))
2658                 return -EINVAL;
2659
2660         /*
2661          * Verify that higher 32bits of exit_signal are unset and that
2662          * it is a valid signal
2663          */
2664         if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2665                      !valid_signal(args.exit_signal)))
2666                 return -EINVAL;
2667
2668         if ((args.flags & CLONE_INTO_CGROUP) &&
2669             (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2670                 return -EINVAL;
2671
2672         *kargs = (struct kernel_clone_args){
2673                 .flags          = args.flags,
2674                 .pidfd          = u64_to_user_ptr(args.pidfd),
2675                 .child_tid      = u64_to_user_ptr(args.child_tid),
2676                 .parent_tid     = u64_to_user_ptr(args.parent_tid),
2677                 .exit_signal    = args.exit_signal,
2678                 .stack          = args.stack,
2679                 .stack_size     = args.stack_size,
2680                 .tls            = args.tls,
2681                 .set_tid_size   = args.set_tid_size,
2682                 .cgroup         = args.cgroup,
2683         };
2684
2685         if (args.set_tid &&
2686                 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2687                         (kargs->set_tid_size * sizeof(pid_t))))
2688                 return -EFAULT;
2689
2690         kargs->set_tid = kset_tid;
2691
2692         return 0;
2693 }
2694
2695 /**
2696  * clone3_stack_valid - check and prepare stack
2697  * @kargs: kernel clone args
2698  *
2699  * Verify that the stack arguments userspace gave us are sane.
2700  * In addition, set the stack direction for userspace since it's easy for us to
2701  * determine.
2702  */
2703 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2704 {
2705         if (kargs->stack == 0) {
2706                 if (kargs->stack_size > 0)
2707                         return false;
2708         } else {
2709                 if (kargs->stack_size == 0)
2710                         return false;
2711
2712                 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2713                         return false;
2714
2715 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2716                 kargs->stack += kargs->stack_size;
2717 #endif
2718         }
2719
2720         return true;
2721 }
2722
2723 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2724 {
2725         /* Verify that no unknown flags are passed along. */
2726         if (kargs->flags &
2727             ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2728                 return false;
2729
2730         /*
2731          * - make the CLONE_DETACHED bit reusable for clone3
2732          * - make the CSIGNAL bits reusable for clone3
2733          */
2734         if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2735                 return false;
2736
2737         if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2738             (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2739                 return false;
2740
2741         if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2742             kargs->exit_signal)
2743                 return false;
2744
2745         if (!clone3_stack_valid(kargs))
2746                 return false;
2747
2748         return true;
2749 }
2750
2751 /**
2752  * clone3 - create a new process with specific properties
2753  * @uargs: argument structure
2754  * @size:  size of @uargs
2755  *
2756  * clone3() is the extensible successor to clone()/clone2().
2757  * It takes a struct as argument that is versioned by its size.
2758  *
2759  * Return: On success, a positive PID for the child process.
2760  *         On error, a negative errno number.
2761  */
2762 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2763 {
2764         int err;
2765
2766         struct kernel_clone_args kargs;
2767         pid_t set_tid[MAX_PID_NS_LEVEL];
2768
2769         kargs.set_tid = set_tid;
2770
2771         err = copy_clone_args_from_user(&kargs, uargs, size);
2772         if (err)
2773                 return err;
2774
2775         if (!clone3_args_valid(&kargs))
2776                 return -EINVAL;
2777
2778         return kernel_clone(&kargs);
2779 }
2780 #endif
2781
2782 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2783 {
2784         struct task_struct *leader, *parent, *child;
2785         int res;
2786
2787         read_lock(&tasklist_lock);
2788         leader = top = top->group_leader;
2789 down:
2790         for_each_thread(leader, parent) {
2791                 list_for_each_entry(child, &parent->children, sibling) {
2792                         res = visitor(child, data);
2793                         if (res) {
2794                                 if (res < 0)
2795                                         goto out;
2796                                 leader = child;
2797                                 goto down;
2798                         }
2799 up:
2800                         ;
2801                 }
2802         }
2803
2804         if (leader != top) {
2805                 child = leader;
2806                 parent = child->real_parent;
2807                 leader = parent->group_leader;
2808                 goto up;
2809         }
2810 out:
2811         read_unlock(&tasklist_lock);
2812 }
2813
2814 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2815 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2816 #endif
2817
2818 static void sighand_ctor(void *data)
2819 {
2820         struct sighand_struct *sighand = data;
2821
2822         spin_lock_init(&sighand->siglock);
2823         init_waitqueue_head(&sighand->signalfd_wqh);
2824 }
2825
2826 void __init proc_caches_init(void)
2827 {
2828         unsigned int mm_size;
2829
2830         sighand_cachep = kmem_cache_create("sighand_cache",
2831                         sizeof(struct sighand_struct), 0,
2832                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2833                         SLAB_ACCOUNT, sighand_ctor);
2834         signal_cachep = kmem_cache_create("signal_cache",
2835                         sizeof(struct signal_struct), 0,
2836                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2837                         NULL);
2838         files_cachep = kmem_cache_create("files_cache",
2839                         sizeof(struct files_struct), 0,
2840                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2841                         NULL);
2842         fs_cachep = kmem_cache_create("fs_cache",
2843                         sizeof(struct fs_struct), 0,
2844                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2845                         NULL);
2846
2847         /*
2848          * The mm_cpumask is located at the end of mm_struct, and is
2849          * dynamically sized based on the maximum CPU number this system
2850          * can have, taking hotplug into account (nr_cpu_ids).
2851          */
2852         mm_size = sizeof(struct mm_struct) + cpumask_size();
2853
2854         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2855                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2856                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2857                         offsetof(struct mm_struct, saved_auxv),
2858                         sizeof_field(struct mm_struct, saved_auxv),
2859                         NULL);
2860         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2861         mmap_init();
2862         nsproxy_cache_init();
2863 }
2864
2865 /*
2866  * Check constraints on flags passed to the unshare system call.
2867  */
2868 static int check_unshare_flags(unsigned long unshare_flags)
2869 {
2870         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2871                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2872                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2873                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
2874                                 CLONE_NEWTIME))
2875                 return -EINVAL;
2876         /*
2877          * Not implemented, but pretend it works if there is nothing
2878          * to unshare.  Note that unsharing the address space or the
2879          * signal handlers also need to unshare the signal queues (aka
2880          * CLONE_THREAD).
2881          */
2882         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2883                 if (!thread_group_empty(current))
2884                         return -EINVAL;
2885         }
2886         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2887                 if (refcount_read(&current->sighand->count) > 1)
2888                         return -EINVAL;
2889         }
2890         if (unshare_flags & CLONE_VM) {
2891                 if (!current_is_single_threaded())
2892                         return -EINVAL;
2893         }
2894
2895         return 0;
2896 }
2897
2898 /*
2899  * Unshare the filesystem structure if it is being shared
2900  */
2901 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2902 {
2903         struct fs_struct *fs = current->fs;
2904
2905         if (!(unshare_flags & CLONE_FS) || !fs)
2906                 return 0;
2907
2908         /* don't need lock here; in the worst case we'll do useless copy */
2909         if (fs->users == 1)
2910                 return 0;
2911
2912         *new_fsp = copy_fs_struct(fs);
2913         if (!*new_fsp)
2914                 return -ENOMEM;
2915
2916         return 0;
2917 }
2918
2919 /*
2920  * Unshare file descriptor table if it is being shared
2921  */
2922 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
2923                struct files_struct **new_fdp)
2924 {
2925         struct files_struct *fd = current->files;
2926         int error = 0;
2927
2928         if ((unshare_flags & CLONE_FILES) &&
2929             (fd && atomic_read(&fd->count) > 1)) {
2930                 *new_fdp = dup_fd(fd, max_fds, &error);
2931                 if (!*new_fdp)
2932                         return error;
2933         }
2934
2935         return 0;
2936 }
2937
2938 /*
2939  * unshare allows a process to 'unshare' part of the process
2940  * context which was originally shared using clone.  copy_*
2941  * functions used by kernel_clone() cannot be used here directly
2942  * because they modify an inactive task_struct that is being
2943  * constructed. Here we are modifying the current, active,
2944  * task_struct.
2945  */
2946 int ksys_unshare(unsigned long unshare_flags)
2947 {
2948         struct fs_struct *fs, *new_fs = NULL;
2949         struct files_struct *fd, *new_fd = NULL;
2950         struct cred *new_cred = NULL;
2951         struct nsproxy *new_nsproxy = NULL;
2952         int do_sysvsem = 0;
2953         int err;
2954
2955         /*
2956          * If unsharing a user namespace must also unshare the thread group
2957          * and unshare the filesystem root and working directories.
2958          */
2959         if (unshare_flags & CLONE_NEWUSER)
2960                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2961         /*
2962          * If unsharing vm, must also unshare signal handlers.
2963          */
2964         if (unshare_flags & CLONE_VM)
2965                 unshare_flags |= CLONE_SIGHAND;
2966         /*
2967          * If unsharing a signal handlers, must also unshare the signal queues.
2968          */
2969         if (unshare_flags & CLONE_SIGHAND)
2970                 unshare_flags |= CLONE_THREAD;
2971         /*
2972          * If unsharing namespace, must also unshare filesystem information.
2973          */
2974         if (unshare_flags & CLONE_NEWNS)
2975                 unshare_flags |= CLONE_FS;
2976
2977         err = check_unshare_flags(unshare_flags);
2978         if (err)
2979                 goto bad_unshare_out;
2980         /*
2981          * CLONE_NEWIPC must also detach from the undolist: after switching
2982          * to a new ipc namespace, the semaphore arrays from the old
2983          * namespace are unreachable.
2984          */
2985         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2986                 do_sysvsem = 1;
2987         err = unshare_fs(unshare_flags, &new_fs);
2988         if (err)
2989                 goto bad_unshare_out;
2990         err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
2991         if (err)
2992                 goto bad_unshare_cleanup_fs;
2993         err = unshare_userns(unshare_flags, &new_cred);
2994         if (err)
2995                 goto bad_unshare_cleanup_fd;
2996         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2997                                          new_cred, new_fs);
2998         if (err)
2999                 goto bad_unshare_cleanup_cred;
3000
3001         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3002                 if (do_sysvsem) {
3003                         /*
3004                          * CLONE_SYSVSEM is equivalent to sys_exit().
3005                          */
3006                         exit_sem(current);
3007                 }
3008                 if (unshare_flags & CLONE_NEWIPC) {
3009                         /* Orphan segments in old ns (see sem above). */
3010                         exit_shm(current);
3011                         shm_init_task(current);
3012                 }
3013
3014                 if (new_nsproxy)
3015                         switch_task_namespaces(current, new_nsproxy);
3016
3017                 task_lock(current);
3018
3019                 if (new_fs) {
3020                         fs = current->fs;
3021                         spin_lock(&fs->lock);
3022                         current->fs = new_fs;
3023                         if (--fs->users)
3024                                 new_fs = NULL;
3025                         else
3026                                 new_fs = fs;
3027                         spin_unlock(&fs->lock);
3028                 }
3029
3030                 if (new_fd) {
3031                         fd = current->files;
3032                         current->files = new_fd;
3033                         new_fd = fd;
3034                 }
3035
3036                 task_unlock(current);
3037
3038                 if (new_cred) {
3039                         /* Install the new user namespace */
3040                         commit_creds(new_cred);
3041                         new_cred = NULL;
3042                 }
3043         }
3044
3045         perf_event_namespaces(current);
3046
3047 bad_unshare_cleanup_cred:
3048         if (new_cred)
3049                 put_cred(new_cred);
3050 bad_unshare_cleanup_fd:
3051         if (new_fd)
3052                 put_files_struct(new_fd);
3053
3054 bad_unshare_cleanup_fs:
3055         if (new_fs)
3056                 free_fs_struct(new_fs);
3057
3058 bad_unshare_out:
3059         return err;
3060 }
3061
3062 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3063 {
3064         return ksys_unshare(unshare_flags);
3065 }
3066
3067 /*
3068  *      Helper to unshare the files of the current task.
3069  *      We don't want to expose copy_files internals to
3070  *      the exec layer of the kernel.
3071  */
3072
3073 int unshare_files(void)
3074 {
3075         struct task_struct *task = current;
3076         struct files_struct *old, *copy = NULL;
3077         int error;
3078
3079         error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, &copy);
3080         if (error || !copy)
3081                 return error;
3082
3083         old = task->files;
3084         task_lock(task);
3085         task->files = copy;
3086         task_unlock(task);
3087         put_files_struct(old);
3088         return 0;
3089 }
3090
3091 int sysctl_max_threads(struct ctl_table *table, int write,
3092                        void *buffer, size_t *lenp, loff_t *ppos)
3093 {
3094         struct ctl_table t;
3095         int ret;
3096         int threads = max_threads;
3097         int min = 1;
3098         int max = MAX_THREADS;
3099
3100         t = *table;
3101         t.data = &threads;
3102         t.extra1 = &min;
3103         t.extra2 = &max;
3104
3105         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3106         if (ret || !write)
3107                 return ret;
3108
3109         max_threads = threads;
3110
3111         return 0;
3112 }