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         atomic64_set(&mm->pinned_vm, 0);
1033         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1034         spin_lock_init(&mm->page_table_lock);
1035         spin_lock_init(&mm->arg_lock);
1036         mm_init_cpumask(mm);
1037         mm_init_aio(mm);
1038         mm_init_owner(mm, p);
1039         mm_init_pasid(mm);
1040         RCU_INIT_POINTER(mm->exe_file, NULL);
1041         mmu_notifier_subscriptions_init(mm);
1042         init_tlb_flush_pending(mm);
1043 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1044         mm->pmd_huge_pte = NULL;
1045 #endif
1046         mm_init_uprobes_state(mm);
1047
1048         if (current->mm) {
1049                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1050                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1051         } else {
1052                 mm->flags = default_dump_filter;
1053                 mm->def_flags = 0;
1054         }
1055
1056         if (mm_alloc_pgd(mm))
1057                 goto fail_nopgd;
1058
1059         if (init_new_context(p, mm))
1060                 goto fail_nocontext;
1061
1062         mm->user_ns = get_user_ns(user_ns);
1063         return mm;
1064
1065 fail_nocontext:
1066         mm_free_pgd(mm);
1067 fail_nopgd:
1068         free_mm(mm);
1069         return NULL;
1070 }
1071
1072 /*
1073  * Allocate and initialize an mm_struct.
1074  */
1075 struct mm_struct *mm_alloc(void)
1076 {
1077         struct mm_struct *mm;
1078
1079         mm = allocate_mm();
1080         if (!mm)
1081                 return NULL;
1082
1083         memset(mm, 0, sizeof(*mm));
1084         return mm_init(mm, current, current_user_ns());
1085 }
1086
1087 static inline void __mmput(struct mm_struct *mm)
1088 {
1089         VM_BUG_ON(atomic_read(&mm->mm_users));
1090
1091         uprobe_clear_state(mm);
1092         exit_aio(mm);
1093         ksm_exit(mm);
1094         khugepaged_exit(mm); /* must run before exit_mmap */
1095         exit_mmap(mm);
1096         mm_put_huge_zero_page(mm);
1097         set_mm_exe_file(mm, NULL);
1098         if (!list_empty(&mm->mmlist)) {
1099                 spin_lock(&mmlist_lock);
1100                 list_del(&mm->mmlist);
1101                 spin_unlock(&mmlist_lock);
1102         }
1103         if (mm->binfmt)
1104                 module_put(mm->binfmt->module);
1105         mmdrop(mm);
1106 }
1107
1108 /*
1109  * Decrement the use count and release all resources for an mm.
1110  */
1111 void mmput(struct mm_struct *mm)
1112 {
1113         might_sleep();
1114
1115         if (atomic_dec_and_test(&mm->mm_users))
1116                 __mmput(mm);
1117 }
1118 EXPORT_SYMBOL_GPL(mmput);
1119
1120 #ifdef CONFIG_MMU
1121 static void mmput_async_fn(struct work_struct *work)
1122 {
1123         struct mm_struct *mm = container_of(work, struct mm_struct,
1124                                             async_put_work);
1125
1126         __mmput(mm);
1127 }
1128
1129 void mmput_async(struct mm_struct *mm)
1130 {
1131         if (atomic_dec_and_test(&mm->mm_users)) {
1132                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1133                 schedule_work(&mm->async_put_work);
1134         }
1135 }
1136 #endif
1137
1138 /**
1139  * set_mm_exe_file - change a reference to the mm's executable file
1140  *
1141  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1142  *
1143  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1144  * invocations: in mmput() nobody alive left, in execve task is single
1145  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1146  * mm->exe_file, but does so without using set_mm_exe_file() in order
1147  * to avoid the need for any locks.
1148  */
1149 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1150 {
1151         struct file *old_exe_file;
1152
1153         /*
1154          * It is safe to dereference the exe_file without RCU as
1155          * this function is only called if nobody else can access
1156          * this mm -- see comment above for justification.
1157          */
1158         old_exe_file = rcu_dereference_raw(mm->exe_file);
1159
1160         if (new_exe_file)
1161                 get_file(new_exe_file);
1162         rcu_assign_pointer(mm->exe_file, new_exe_file);
1163         if (old_exe_file)
1164                 fput(old_exe_file);
1165 }
1166
1167 /**
1168  * get_mm_exe_file - acquire a reference to the mm's executable file
1169  *
1170  * Returns %NULL if mm has no associated executable file.
1171  * User must release file via fput().
1172  */
1173 struct file *get_mm_exe_file(struct mm_struct *mm)
1174 {
1175         struct file *exe_file;
1176
1177         rcu_read_lock();
1178         exe_file = rcu_dereference(mm->exe_file);
1179         if (exe_file && !get_file_rcu(exe_file))
1180                 exe_file = NULL;
1181         rcu_read_unlock();
1182         return exe_file;
1183 }
1184 EXPORT_SYMBOL(get_mm_exe_file);
1185
1186 /**
1187  * get_task_exe_file - acquire a reference to the task's executable file
1188  *
1189  * Returns %NULL if task's mm (if any) has no associated executable file or
1190  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1191  * User must release file via fput().
1192  */
1193 struct file *get_task_exe_file(struct task_struct *task)
1194 {
1195         struct file *exe_file = NULL;
1196         struct mm_struct *mm;
1197
1198         task_lock(task);
1199         mm = task->mm;
1200         if (mm) {
1201                 if (!(task->flags & PF_KTHREAD))
1202                         exe_file = get_mm_exe_file(mm);
1203         }
1204         task_unlock(task);
1205         return exe_file;
1206 }
1207 EXPORT_SYMBOL(get_task_exe_file);
1208
1209 /**
1210  * get_task_mm - acquire a reference to the task's mm
1211  *
1212  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1213  * this kernel workthread has transiently adopted a user mm with use_mm,
1214  * to do its AIO) is not set and if so returns a reference to it, after
1215  * bumping up the use count.  User must release the mm via mmput()
1216  * after use.  Typically used by /proc and ptrace.
1217  */
1218 struct mm_struct *get_task_mm(struct task_struct *task)
1219 {
1220         struct mm_struct *mm;
1221
1222         task_lock(task);
1223         mm = task->mm;
1224         if (mm) {
1225                 if (task->flags & PF_KTHREAD)
1226                         mm = NULL;
1227                 else
1228                         mmget(mm);
1229         }
1230         task_unlock(task);
1231         return mm;
1232 }
1233 EXPORT_SYMBOL_GPL(get_task_mm);
1234
1235 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1236 {
1237         struct mm_struct *mm;
1238         int err;
1239
1240         err =  down_read_killable(&task->signal->exec_update_lock);
1241         if (err)
1242                 return ERR_PTR(err);
1243
1244         mm = get_task_mm(task);
1245         if (mm && mm != current->mm &&
1246                         !ptrace_may_access(task, mode)) {
1247                 mmput(mm);
1248                 mm = ERR_PTR(-EACCES);
1249         }
1250         up_read(&task->signal->exec_update_lock);
1251
1252         return mm;
1253 }
1254
1255 static void complete_vfork_done(struct task_struct *tsk)
1256 {
1257         struct completion *vfork;
1258
1259         task_lock(tsk);
1260         vfork = tsk->vfork_done;
1261         if (likely(vfork)) {
1262                 tsk->vfork_done = NULL;
1263                 complete(vfork);
1264         }
1265         task_unlock(tsk);
1266 }
1267
1268 static int wait_for_vfork_done(struct task_struct *child,
1269                                 struct completion *vfork)
1270 {
1271         int killed;
1272
1273         freezer_do_not_count();
1274         cgroup_enter_frozen();
1275         killed = wait_for_completion_killable(vfork);
1276         cgroup_leave_frozen(false);
1277         freezer_count();
1278
1279         if (killed) {
1280                 task_lock(child);
1281                 child->vfork_done = NULL;
1282                 task_unlock(child);
1283         }
1284
1285         put_task_struct(child);
1286         return killed;
1287 }
1288
1289 /* Please note the differences between mmput and mm_release.
1290  * mmput is called whenever we stop holding onto a mm_struct,
1291  * error success whatever.
1292  *
1293  * mm_release is called after a mm_struct has been removed
1294  * from the current process.
1295  *
1296  * This difference is important for error handling, when we
1297  * only half set up a mm_struct for a new process and need to restore
1298  * the old one.  Because we mmput the new mm_struct before
1299  * restoring the old one. . .
1300  * Eric Biederman 10 January 1998
1301  */
1302 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1303 {
1304         uprobe_free_utask(tsk);
1305
1306         /* Get rid of any cached register state */
1307         deactivate_mm(tsk, mm);
1308
1309         /*
1310          * Signal userspace if we're not exiting with a core dump
1311          * because we want to leave the value intact for debugging
1312          * purposes.
1313          */
1314         if (tsk->clear_child_tid) {
1315                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1316                     atomic_read(&mm->mm_users) > 1) {
1317                         /*
1318                          * We don't check the error code - if userspace has
1319                          * not set up a proper pointer then tough luck.
1320                          */
1321                         put_user(0, tsk->clear_child_tid);
1322                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1323                                         1, NULL, NULL, 0, 0);
1324                 }
1325                 tsk->clear_child_tid = NULL;
1326         }
1327
1328         /*
1329          * All done, finally we can wake up parent and return this mm to him.
1330          * Also kthread_stop() uses this completion for synchronization.
1331          */
1332         if (tsk->vfork_done)
1333                 complete_vfork_done(tsk);
1334 }
1335
1336 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1337 {
1338         futex_exit_release(tsk);
1339         mm_release(tsk, mm);
1340 }
1341
1342 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1343 {
1344         futex_exec_release(tsk);
1345         mm_release(tsk, mm);
1346 }
1347
1348 /**
1349  * dup_mm() - duplicates an existing mm structure
1350  * @tsk: the task_struct with which the new mm will be associated.
1351  * @oldmm: the mm to duplicate.
1352  *
1353  * Allocates a new mm structure and duplicates the provided @oldmm structure
1354  * content into it.
1355  *
1356  * Return: the duplicated mm or NULL on failure.
1357  */
1358 static struct mm_struct *dup_mm(struct task_struct *tsk,
1359                                 struct mm_struct *oldmm)
1360 {
1361         struct mm_struct *mm;
1362         int err;
1363
1364         mm = allocate_mm();
1365         if (!mm)
1366                 goto fail_nomem;
1367
1368         memcpy(mm, oldmm, sizeof(*mm));
1369
1370         if (!mm_init(mm, tsk, mm->user_ns))
1371                 goto fail_nomem;
1372
1373         err = dup_mmap(mm, oldmm);
1374         if (err)
1375                 goto free_pt;
1376
1377         mm->hiwater_rss = get_mm_rss(mm);
1378         mm->hiwater_vm = mm->total_vm;
1379
1380         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1381                 goto free_pt;
1382
1383         return mm;
1384
1385 free_pt:
1386         /* don't put binfmt in mmput, we haven't got module yet */
1387         mm->binfmt = NULL;
1388         mm_init_owner(mm, NULL);
1389         mmput(mm);
1390
1391 fail_nomem:
1392         return NULL;
1393 }
1394
1395 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1396 {
1397         struct mm_struct *mm, *oldmm;
1398
1399         tsk->min_flt = tsk->maj_flt = 0;
1400         tsk->nvcsw = tsk->nivcsw = 0;
1401 #ifdef CONFIG_DETECT_HUNG_TASK
1402         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1403         tsk->last_switch_time = 0;
1404 #endif
1405
1406         tsk->mm = NULL;
1407         tsk->active_mm = NULL;
1408
1409         /*
1410          * Are we cloning a kernel thread?
1411          *
1412          * We need to steal a active VM for that..
1413          */
1414         oldmm = current->mm;
1415         if (!oldmm)
1416                 return 0;
1417
1418         /* initialize the new vmacache entries */
1419         vmacache_flush(tsk);
1420
1421         if (clone_flags & CLONE_VM) {
1422                 mmget(oldmm);
1423                 mm = oldmm;
1424         } else {
1425                 mm = dup_mm(tsk, current->mm);
1426                 if (!mm)
1427                         return -ENOMEM;
1428         }
1429
1430         tsk->mm = mm;
1431         tsk->active_mm = mm;
1432         return 0;
1433 }
1434
1435 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1436 {
1437         struct fs_struct *fs = current->fs;
1438         if (clone_flags & CLONE_FS) {
1439                 /* tsk->fs is already what we want */
1440                 spin_lock(&fs->lock);
1441                 if (fs->in_exec) {
1442                         spin_unlock(&fs->lock);
1443                         return -EAGAIN;
1444                 }
1445                 fs->users++;
1446                 spin_unlock(&fs->lock);
1447                 return 0;
1448         }
1449         tsk->fs = copy_fs_struct(fs);
1450         if (!tsk->fs)
1451                 return -ENOMEM;
1452         return 0;
1453 }
1454
1455 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1456 {
1457         struct files_struct *oldf, *newf;
1458         int error = 0;
1459
1460         /*
1461          * A background process may not have any files ...
1462          */
1463         oldf = current->files;
1464         if (!oldf)
1465                 goto out;
1466
1467         if (clone_flags & CLONE_FILES) {
1468                 atomic_inc(&oldf->count);
1469                 goto out;
1470         }
1471
1472         newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1473         if (!newf)
1474                 goto out;
1475
1476         tsk->files = newf;
1477         error = 0;
1478 out:
1479         return error;
1480 }
1481
1482 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1483 {
1484 #ifdef CONFIG_BLOCK
1485         struct io_context *ioc = current->io_context;
1486         struct io_context *new_ioc;
1487
1488         if (!ioc)
1489                 return 0;
1490         /*
1491          * Share io context with parent, if CLONE_IO is set
1492          */
1493         if (clone_flags & CLONE_IO) {
1494                 ioc_task_link(ioc);
1495                 tsk->io_context = ioc;
1496         } else if (ioprio_valid(ioc->ioprio)) {
1497                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1498                 if (unlikely(!new_ioc))
1499                         return -ENOMEM;
1500
1501                 new_ioc->ioprio = ioc->ioprio;
1502                 put_io_context(new_ioc);
1503         }
1504 #endif
1505         return 0;
1506 }
1507
1508 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1509 {
1510         struct sighand_struct *sig;
1511
1512         if (clone_flags & CLONE_SIGHAND) {
1513                 refcount_inc(&current->sighand->count);
1514                 return 0;
1515         }
1516         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1517         RCU_INIT_POINTER(tsk->sighand, sig);
1518         if (!sig)
1519                 return -ENOMEM;
1520
1521         refcount_set(&sig->count, 1);
1522         spin_lock_irq(&current->sighand->siglock);
1523         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1524         spin_unlock_irq(&current->sighand->siglock);
1525
1526         /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1527         if (clone_flags & CLONE_CLEAR_SIGHAND)
1528                 flush_signal_handlers(tsk, 0);
1529
1530         return 0;
1531 }
1532
1533 void __cleanup_sighand(struct sighand_struct *sighand)
1534 {
1535         if (refcount_dec_and_test(&sighand->count)) {
1536                 signalfd_cleanup(sighand);
1537                 /*
1538                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1539                  * without an RCU grace period, see __lock_task_sighand().
1540                  */
1541                 kmem_cache_free(sighand_cachep, sighand);
1542         }
1543 }
1544
1545 /*
1546  * Initialize POSIX timer handling for a thread group.
1547  */
1548 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1549 {
1550         struct posix_cputimers *pct = &sig->posix_cputimers;
1551         unsigned long cpu_limit;
1552
1553         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1554         posix_cputimers_group_init(pct, cpu_limit);
1555 }
1556
1557 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1558 {
1559         struct signal_struct *sig;
1560
1561         if (clone_flags & CLONE_THREAD)
1562                 return 0;
1563
1564         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1565         tsk->signal = sig;
1566         if (!sig)
1567                 return -ENOMEM;
1568
1569         sig->nr_threads = 1;
1570         atomic_set(&sig->live, 1);
1571         refcount_set(&sig->sigcnt, 1);
1572
1573         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1574         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1575         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1576
1577         init_waitqueue_head(&sig->wait_chldexit);
1578         sig->curr_target = tsk;
1579         init_sigpending(&sig->shared_pending);
1580         INIT_HLIST_HEAD(&sig->multiprocess);
1581         seqlock_init(&sig->stats_lock);
1582         prev_cputime_init(&sig->prev_cputime);
1583
1584 #ifdef CONFIG_POSIX_TIMERS
1585         INIT_LIST_HEAD(&sig->posix_timers);
1586         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1587         sig->real_timer.function = it_real_fn;
1588 #endif
1589
1590         task_lock(current->group_leader);
1591         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1592         task_unlock(current->group_leader);
1593
1594         posix_cpu_timers_init_group(sig);
1595
1596         tty_audit_fork(sig);
1597         sched_autogroup_fork(sig);
1598
1599         sig->oom_score_adj = current->signal->oom_score_adj;
1600         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1601
1602         mutex_init(&sig->cred_guard_mutex);
1603         init_rwsem(&sig->exec_update_lock);
1604
1605         return 0;
1606 }
1607
1608 static void copy_seccomp(struct task_struct *p)
1609 {
1610 #ifdef CONFIG_SECCOMP
1611         /*
1612          * Must be called with sighand->lock held, which is common to
1613          * all threads in the group. Holding cred_guard_mutex is not
1614          * needed because this new task is not yet running and cannot
1615          * be racing exec.
1616          */
1617         assert_spin_locked(&current->sighand->siglock);
1618
1619         /* Ref-count the new filter user, and assign it. */
1620         get_seccomp_filter(current);
1621         p->seccomp = current->seccomp;
1622
1623         /*
1624          * Explicitly enable no_new_privs here in case it got set
1625          * between the task_struct being duplicated and holding the
1626          * sighand lock. The seccomp state and nnp must be in sync.
1627          */
1628         if (task_no_new_privs(current))
1629                 task_set_no_new_privs(p);
1630
1631         /*
1632          * If the parent gained a seccomp mode after copying thread
1633          * flags and between before we held the sighand lock, we have
1634          * to manually enable the seccomp thread flag here.
1635          */
1636         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1637                 set_task_syscall_work(p, SECCOMP);
1638 #endif
1639 }
1640
1641 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1642 {
1643         current->clear_child_tid = tidptr;
1644
1645         return task_pid_vnr(current);
1646 }
1647
1648 static void rt_mutex_init_task(struct task_struct *p)
1649 {
1650         raw_spin_lock_init(&p->pi_lock);
1651 #ifdef CONFIG_RT_MUTEXES
1652         p->pi_waiters = RB_ROOT_CACHED;
1653         p->pi_top_task = NULL;
1654         p->pi_blocked_on = NULL;
1655 #endif
1656 }
1657
1658 static inline void init_task_pid_links(struct task_struct *task)
1659 {
1660         enum pid_type type;
1661
1662         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1663                 INIT_HLIST_NODE(&task->pid_links[type]);
1664 }
1665
1666 static inline void
1667 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1668 {
1669         if (type == PIDTYPE_PID)
1670                 task->thread_pid = pid;
1671         else
1672                 task->signal->pids[type] = pid;
1673 }
1674
1675 static inline void rcu_copy_process(struct task_struct *p)
1676 {
1677 #ifdef CONFIG_PREEMPT_RCU
1678         p->rcu_read_lock_nesting = 0;
1679         p->rcu_read_unlock_special.s = 0;
1680         p->rcu_blocked_node = NULL;
1681         INIT_LIST_HEAD(&p->rcu_node_entry);
1682 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1683 #ifdef CONFIG_TASKS_RCU
1684         p->rcu_tasks_holdout = false;
1685         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1686         p->rcu_tasks_idle_cpu = -1;
1687 #endif /* #ifdef CONFIG_TASKS_RCU */
1688 #ifdef CONFIG_TASKS_TRACE_RCU
1689         p->trc_reader_nesting = 0;
1690         p->trc_reader_special.s = 0;
1691         INIT_LIST_HEAD(&p->trc_holdout_list);
1692 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1693 }
1694
1695 struct pid *pidfd_pid(const struct file *file)
1696 {
1697         if (file->f_op == &pidfd_fops)
1698                 return file->private_data;
1699
1700         return ERR_PTR(-EBADF);
1701 }
1702
1703 static int pidfd_release(struct inode *inode, struct file *file)
1704 {
1705         struct pid *pid = file->private_data;
1706
1707         file->private_data = NULL;
1708         put_pid(pid);
1709         return 0;
1710 }
1711
1712 #ifdef CONFIG_PROC_FS
1713 /**
1714  * pidfd_show_fdinfo - print information about a pidfd
1715  * @m: proc fdinfo file
1716  * @f: file referencing a pidfd
1717  *
1718  * Pid:
1719  * This function will print the pid that a given pidfd refers to in the
1720  * pid namespace of the procfs instance.
1721  * If the pid namespace of the process is not a descendant of the pid
1722  * namespace of the procfs instance 0 will be shown as its pid. This is
1723  * similar to calling getppid() on a process whose parent is outside of
1724  * its pid namespace.
1725  *
1726  * NSpid:
1727  * If pid namespaces are supported then this function will also print
1728  * the pid of a given pidfd refers to for all descendant pid namespaces
1729  * starting from the current pid namespace of the instance, i.e. the
1730  * Pid field and the first entry in the NSpid field will be identical.
1731  * If the pid namespace of the process is not a descendant of the pid
1732  * namespace of the procfs instance 0 will be shown as its first NSpid
1733  * entry and no others will be shown.
1734  * Note that this differs from the Pid and NSpid fields in
1735  * /proc/<pid>/status where Pid and NSpid are always shown relative to
1736  * the  pid namespace of the procfs instance. The difference becomes
1737  * obvious when sending around a pidfd between pid namespaces from a
1738  * different branch of the tree, i.e. where no ancestral relation is
1739  * present between the pid namespaces:
1740  * - create two new pid namespaces ns1 and ns2 in the initial pid
1741  *   namespace (also take care to create new mount namespaces in the
1742  *   new pid namespace and mount procfs)
1743  * - create a process with a pidfd in ns1
1744  * - send pidfd from ns1 to ns2
1745  * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1746  *   have exactly one entry, which is 0
1747  */
1748 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1749 {
1750         struct pid *pid = f->private_data;
1751         struct pid_namespace *ns;
1752         pid_t nr = -1;
1753
1754         if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1755                 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1756                 nr = pid_nr_ns(pid, ns);
1757         }
1758
1759         seq_put_decimal_ll(m, "Pid:\t", nr);
1760
1761 #ifdef CONFIG_PID_NS
1762         seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1763         if (nr > 0) {
1764                 int i;
1765
1766                 /* If nr is non-zero it means that 'pid' is valid and that
1767                  * ns, i.e. the pid namespace associated with the procfs
1768                  * instance, is in the pid namespace hierarchy of pid.
1769                  * Start at one below the already printed level.
1770                  */
1771                 for (i = ns->level + 1; i <= pid->level; i++)
1772                         seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1773         }
1774 #endif
1775         seq_putc(m, '\n');
1776 }
1777 #endif
1778
1779 /*
1780  * Poll support for process exit notification.
1781  */
1782 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1783 {
1784         struct pid *pid = file->private_data;
1785         __poll_t poll_flags = 0;
1786
1787         poll_wait(file, &pid->wait_pidfd, pts);
1788
1789         /*
1790          * Inform pollers only when the whole thread group exits.
1791          * If the thread group leader exits before all other threads in the
1792          * group, then poll(2) should block, similar to the wait(2) family.
1793          */
1794         if (thread_group_exited(pid))
1795                 poll_flags = EPOLLIN | EPOLLRDNORM;
1796
1797         return poll_flags;
1798 }
1799
1800 const struct file_operations pidfd_fops = {
1801         .release = pidfd_release,
1802         .poll = pidfd_poll,
1803 #ifdef CONFIG_PROC_FS
1804         .show_fdinfo = pidfd_show_fdinfo,
1805 #endif
1806 };
1807
1808 static void __delayed_free_task(struct rcu_head *rhp)
1809 {
1810         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1811
1812         free_task(tsk);
1813 }
1814
1815 static __always_inline void delayed_free_task(struct task_struct *tsk)
1816 {
1817         if (IS_ENABLED(CONFIG_MEMCG))
1818                 call_rcu(&tsk->rcu, __delayed_free_task);
1819         else
1820                 free_task(tsk);
1821 }
1822
1823 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1824 {
1825         /* Skip if kernel thread */
1826         if (!tsk->mm)
1827                 return;
1828
1829         /* Skip if spawning a thread or using vfork */
1830         if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1831                 return;
1832
1833         /* We need to synchronize with __set_oom_adj */
1834         mutex_lock(&oom_adj_mutex);
1835         set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1836         /* Update the values in case they were changed after copy_signal */
1837         tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1838         tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1839         mutex_unlock(&oom_adj_mutex);
1840 }
1841
1842 /*
1843  * This creates a new process as a copy of the old one,
1844  * but does not actually start it yet.
1845  *
1846  * It copies the registers, and all the appropriate
1847  * parts of the process environment (as per the clone
1848  * flags). The actual kick-off is left to the caller.
1849  */
1850 static __latent_entropy struct task_struct *copy_process(
1851                                         struct pid *pid,
1852                                         int trace,
1853                                         int node,
1854                                         struct kernel_clone_args *args)
1855 {
1856         int pidfd = -1, retval;
1857         struct task_struct *p;
1858         struct multiprocess_signals delayed;
1859         struct file *pidfile = NULL;
1860         u64 clone_flags = args->flags;
1861         struct nsproxy *nsp = current->nsproxy;
1862
1863         /*
1864          * Don't allow sharing the root directory with processes in a different
1865          * namespace
1866          */
1867         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1868                 return ERR_PTR(-EINVAL);
1869
1870         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1871                 return ERR_PTR(-EINVAL);
1872
1873         /*
1874          * Thread groups must share signals as well, and detached threads
1875          * can only be started up within the thread group.
1876          */
1877         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1878                 return ERR_PTR(-EINVAL);
1879
1880         /*
1881          * Shared signal handlers imply shared VM. By way of the above,
1882          * thread groups also imply shared VM. Blocking this case allows
1883          * for various simplifications in other code.
1884          */
1885         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1886                 return ERR_PTR(-EINVAL);
1887
1888         /*
1889          * Siblings of global init remain as zombies on exit since they are
1890          * not reaped by their parent (swapper). To solve this and to avoid
1891          * multi-rooted process trees, prevent global and container-inits
1892          * from creating siblings.
1893          */
1894         if ((clone_flags & CLONE_PARENT) &&
1895                                 current->signal->flags & SIGNAL_UNKILLABLE)
1896                 return ERR_PTR(-EINVAL);
1897
1898         /*
1899          * If the new process will be in a different pid or user namespace
1900          * do not allow it to share a thread group with the forking task.
1901          */
1902         if (clone_flags & CLONE_THREAD) {
1903                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1904                     (task_active_pid_ns(current) != nsp->pid_ns_for_children))
1905                         return ERR_PTR(-EINVAL);
1906         }
1907
1908         /*
1909          * If the new process will be in a different time namespace
1910          * do not allow it to share VM or a thread group with the forking task.
1911          */
1912         if (clone_flags & (CLONE_THREAD | CLONE_VM)) {
1913                 if (nsp->time_ns != nsp->time_ns_for_children)
1914                         return ERR_PTR(-EINVAL);
1915         }
1916
1917         if (clone_flags & CLONE_PIDFD) {
1918                 /*
1919                  * - CLONE_DETACHED is blocked so that we can potentially
1920                  *   reuse it later for CLONE_PIDFD.
1921                  * - CLONE_THREAD is blocked until someone really needs it.
1922                  */
1923                 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1924                         return ERR_PTR(-EINVAL);
1925         }
1926
1927         /*
1928          * Force any signals received before this point to be delivered
1929          * before the fork happens.  Collect up signals sent to multiple
1930          * processes that happen during the fork and delay them so that
1931          * they appear to happen after the fork.
1932          */
1933         sigemptyset(&delayed.signal);
1934         INIT_HLIST_NODE(&delayed.node);
1935
1936         spin_lock_irq(&current->sighand->siglock);
1937         if (!(clone_flags & CLONE_THREAD))
1938                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1939         recalc_sigpending();
1940         spin_unlock_irq(&current->sighand->siglock);
1941         retval = -ERESTARTNOINTR;
1942         if (task_sigpending(current))
1943                 goto fork_out;
1944
1945         retval = -ENOMEM;
1946         p = dup_task_struct(current, node);
1947         if (!p)
1948                 goto fork_out;
1949         if (args->io_thread) {
1950                 /*
1951                  * Mark us an IO worker, and block any signal that isn't
1952                  * fatal or STOP
1953                  */
1954                 p->flags |= PF_IO_WORKER;
1955                 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
1956         }
1957
1958         /*
1959          * This _must_ happen before we call free_task(), i.e. before we jump
1960          * to any of the bad_fork_* labels. This is to avoid freeing
1961          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1962          * kernel threads (PF_KTHREAD).
1963          */
1964         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1965         /*
1966          * Clear TID on mm_release()?
1967          */
1968         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1969
1970         ftrace_graph_init_task(p);
1971
1972         rt_mutex_init_task(p);
1973
1974         lockdep_assert_irqs_enabled();
1975 #ifdef CONFIG_PROVE_LOCKING
1976         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1977 #endif
1978         retval = -EAGAIN;
1979         if (atomic_read(&p->real_cred->user->processes) >=
1980                         task_rlimit(p, RLIMIT_NPROC)) {
1981                 if (p->real_cred->user != INIT_USER &&
1982                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1983                         goto bad_fork_free;
1984         }
1985         current->flags &= ~PF_NPROC_EXCEEDED;
1986
1987         retval = copy_creds(p, clone_flags);
1988         if (retval < 0)
1989                 goto bad_fork_free;
1990
1991         /*
1992          * If multiple threads are within copy_process(), then this check
1993          * triggers too late. This doesn't hurt, the check is only there
1994          * to stop root fork bombs.
1995          */
1996         retval = -EAGAIN;
1997         if (data_race(nr_threads >= max_threads))
1998                 goto bad_fork_cleanup_count;
1999
2000         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
2001         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
2002         p->flags |= PF_FORKNOEXEC;
2003         INIT_LIST_HEAD(&p->children);
2004         INIT_LIST_HEAD(&p->sibling);
2005         rcu_copy_process(p);
2006         p->vfork_done = NULL;
2007         spin_lock_init(&p->alloc_lock);
2008
2009         init_sigpending(&p->pending);
2010
2011         p->utime = p->stime = p->gtime = 0;
2012 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2013         p->utimescaled = p->stimescaled = 0;
2014 #endif
2015         prev_cputime_init(&p->prev_cputime);
2016
2017 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2018         seqcount_init(&p->vtime.seqcount);
2019         p->vtime.starttime = 0;
2020         p->vtime.state = VTIME_INACTIVE;
2021 #endif
2022
2023 #ifdef CONFIG_IO_URING
2024         p->io_uring = NULL;
2025 #endif
2026
2027 #if defined(SPLIT_RSS_COUNTING)
2028         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2029 #endif
2030
2031         p->default_timer_slack_ns = current->timer_slack_ns;
2032
2033 #ifdef CONFIG_PSI
2034         p->psi_flags = 0;
2035 #endif
2036
2037         task_io_accounting_init(&p->ioac);
2038         acct_clear_integrals(p);
2039
2040         posix_cputimers_init(&p->posix_cputimers);
2041
2042         p->io_context = NULL;
2043         audit_set_context(p, NULL);
2044         cgroup_fork(p);
2045 #ifdef CONFIG_NUMA
2046         p->mempolicy = mpol_dup(p->mempolicy);
2047         if (IS_ERR(p->mempolicy)) {
2048                 retval = PTR_ERR(p->mempolicy);
2049                 p->mempolicy = NULL;
2050                 goto bad_fork_cleanup_threadgroup_lock;
2051         }
2052 #endif
2053 #ifdef CONFIG_CPUSETS
2054         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2055         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2056         seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2057 #endif
2058 #ifdef CONFIG_TRACE_IRQFLAGS
2059         memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2060         p->irqtrace.hardirq_disable_ip  = _THIS_IP_;
2061         p->irqtrace.softirq_enable_ip   = _THIS_IP_;
2062         p->softirqs_enabled             = 1;
2063         p->softirq_context              = 0;
2064 #endif
2065
2066         p->pagefault_disabled = 0;
2067
2068 #ifdef CONFIG_LOCKDEP
2069         lockdep_init_task(p);
2070 #endif
2071
2072 #ifdef CONFIG_DEBUG_MUTEXES
2073         p->blocked_on = NULL; /* not blocked yet */
2074 #endif
2075 #ifdef CONFIG_BCACHE
2076         p->sequential_io        = 0;
2077         p->sequential_io_avg    = 0;
2078 #endif
2079 #ifdef CONFIG_BPF_SYSCALL
2080         RCU_INIT_POINTER(p->bpf_storage, NULL);
2081 #endif
2082
2083         /* Perform scheduler related setup. Assign this task to a CPU. */
2084         retval = sched_fork(clone_flags, p);
2085         if (retval)
2086                 goto bad_fork_cleanup_policy;
2087
2088         retval = perf_event_init_task(p, clone_flags);
2089         if (retval)
2090                 goto bad_fork_cleanup_policy;
2091         retval = audit_alloc(p);
2092         if (retval)
2093                 goto bad_fork_cleanup_perf;
2094         /* copy all the process information */
2095         shm_init_task(p);
2096         retval = security_task_alloc(p, clone_flags);
2097         if (retval)
2098                 goto bad_fork_cleanup_audit;
2099         retval = copy_semundo(clone_flags, p);
2100         if (retval)
2101                 goto bad_fork_cleanup_security;
2102         retval = copy_files(clone_flags, p);
2103         if (retval)
2104                 goto bad_fork_cleanup_semundo;
2105         retval = copy_fs(clone_flags, p);
2106         if (retval)
2107                 goto bad_fork_cleanup_files;
2108         retval = copy_sighand(clone_flags, p);
2109         if (retval)
2110                 goto bad_fork_cleanup_fs;
2111         retval = copy_signal(clone_flags, p);
2112         if (retval)
2113                 goto bad_fork_cleanup_sighand;
2114         retval = copy_mm(clone_flags, p);
2115         if (retval)
2116                 goto bad_fork_cleanup_signal;
2117         retval = copy_namespaces(clone_flags, p);
2118         if (retval)
2119                 goto bad_fork_cleanup_mm;
2120         retval = copy_io(clone_flags, p);
2121         if (retval)
2122                 goto bad_fork_cleanup_namespaces;
2123         retval = copy_thread(clone_flags, args->stack, args->stack_size, p, args->tls);
2124         if (retval)
2125                 goto bad_fork_cleanup_io;
2126
2127         stackleak_task_init(p);
2128
2129         if (pid != &init_struct_pid) {
2130                 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2131                                 args->set_tid_size);
2132                 if (IS_ERR(pid)) {
2133                         retval = PTR_ERR(pid);
2134                         goto bad_fork_cleanup_thread;
2135                 }
2136         }
2137
2138         /*
2139          * This has to happen after we've potentially unshared the file
2140          * descriptor table (so that the pidfd doesn't leak into the child
2141          * if the fd table isn't shared).
2142          */
2143         if (clone_flags & CLONE_PIDFD) {
2144                 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2145                 if (retval < 0)
2146                         goto bad_fork_free_pid;
2147
2148                 pidfd = retval;
2149
2150                 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2151                                               O_RDWR | O_CLOEXEC);
2152                 if (IS_ERR(pidfile)) {
2153                         put_unused_fd(pidfd);
2154                         retval = PTR_ERR(pidfile);
2155                         goto bad_fork_free_pid;
2156                 }
2157                 get_pid(pid);   /* held by pidfile now */
2158
2159                 retval = put_user(pidfd, args->pidfd);
2160                 if (retval)
2161                         goto bad_fork_put_pidfd;
2162         }
2163
2164 #ifdef CONFIG_BLOCK
2165         p->plug = NULL;
2166 #endif
2167         futex_init_task(p);
2168
2169         /*
2170          * sigaltstack should be cleared when sharing the same VM
2171          */
2172         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2173                 sas_ss_reset(p);
2174
2175         /*
2176          * Syscall tracing and stepping should be turned off in the
2177          * child regardless of CLONE_PTRACE.
2178          */
2179         user_disable_single_step(p);
2180         clear_task_syscall_work(p, SYSCALL_TRACE);
2181 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2182         clear_task_syscall_work(p, SYSCALL_EMU);
2183 #endif
2184         clear_tsk_latency_tracing(p);
2185
2186         /* ok, now we should be set up.. */
2187         p->pid = pid_nr(pid);
2188         if (clone_flags & CLONE_THREAD) {
2189                 p->group_leader = current->group_leader;
2190                 p->tgid = current->tgid;
2191         } else {
2192                 p->group_leader = p;
2193                 p->tgid = p->pid;
2194         }
2195
2196         p->nr_dirtied = 0;
2197         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2198         p->dirty_paused_when = 0;
2199
2200         p->pdeath_signal = 0;
2201         INIT_LIST_HEAD(&p->thread_group);
2202         p->task_works = NULL;
2203
2204 #ifdef CONFIG_KRETPROBES
2205         p->kretprobe_instances.first = NULL;
2206 #endif
2207
2208         /*
2209          * Ensure that the cgroup subsystem policies allow the new process to be
2210          * forked. It should be noted that the new process's css_set can be changed
2211          * between here and cgroup_post_fork() if an organisation operation is in
2212          * progress.
2213          */
2214         retval = cgroup_can_fork(p, args);
2215         if (retval)
2216                 goto bad_fork_put_pidfd;
2217
2218         /*
2219          * From this point on we must avoid any synchronous user-space
2220          * communication until we take the tasklist-lock. In particular, we do
2221          * not want user-space to be able to predict the process start-time by
2222          * stalling fork(2) after we recorded the start_time but before it is
2223          * visible to the system.
2224          */
2225
2226         p->start_time = ktime_get_ns();
2227         p->start_boottime = ktime_get_boottime_ns();
2228
2229         /*
2230          * Make it visible to the rest of the system, but dont wake it up yet.
2231          * Need tasklist lock for parent etc handling!
2232          */
2233         write_lock_irq(&tasklist_lock);
2234
2235         /* CLONE_PARENT re-uses the old parent */
2236         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2237                 p->real_parent = current->real_parent;
2238                 p->parent_exec_id = current->parent_exec_id;
2239                 if (clone_flags & CLONE_THREAD)
2240                         p->exit_signal = -1;
2241                 else
2242                         p->exit_signal = current->group_leader->exit_signal;
2243         } else {
2244                 p->real_parent = current;
2245                 p->parent_exec_id = current->self_exec_id;
2246                 p->exit_signal = args->exit_signal;
2247         }
2248
2249         klp_copy_process(p);
2250
2251         spin_lock(&current->sighand->siglock);
2252
2253         /*
2254          * Copy seccomp details explicitly here, in case they were changed
2255          * before holding sighand lock.
2256          */
2257         copy_seccomp(p);
2258
2259         rseq_fork(p, clone_flags);
2260
2261         /* Don't start children in a dying pid namespace */
2262         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2263                 retval = -ENOMEM;
2264                 goto bad_fork_cancel_cgroup;
2265         }
2266
2267         /* Let kill terminate clone/fork in the middle */
2268         if (fatal_signal_pending(current)) {
2269                 retval = -EINTR;
2270                 goto bad_fork_cancel_cgroup;
2271         }
2272
2273         /* past the last point of failure */
2274         if (pidfile)
2275                 fd_install(pidfd, pidfile);
2276
2277         init_task_pid_links(p);
2278         if (likely(p->pid)) {
2279                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2280
2281                 init_task_pid(p, PIDTYPE_PID, pid);
2282                 if (thread_group_leader(p)) {
2283                         init_task_pid(p, PIDTYPE_TGID, pid);
2284                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2285                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2286
2287                         if (is_child_reaper(pid)) {
2288                                 ns_of_pid(pid)->child_reaper = p;
2289                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2290                         }
2291                         p->signal->shared_pending.signal = delayed.signal;
2292                         p->signal->tty = tty_kref_get(current->signal->tty);
2293                         /*
2294                          * Inherit has_child_subreaper flag under the same
2295                          * tasklist_lock with adding child to the process tree
2296                          * for propagate_has_child_subreaper optimization.
2297                          */
2298                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2299                                                          p->real_parent->signal->is_child_subreaper;
2300                         list_add_tail(&p->sibling, &p->real_parent->children);
2301                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2302                         attach_pid(p, PIDTYPE_TGID);
2303                         attach_pid(p, PIDTYPE_PGID);
2304                         attach_pid(p, PIDTYPE_SID);
2305                         __this_cpu_inc(process_counts);
2306                 } else {
2307                         current->signal->nr_threads++;
2308                         atomic_inc(&current->signal->live);
2309                         refcount_inc(&current->signal->sigcnt);
2310                         task_join_group_stop(p);
2311                         list_add_tail_rcu(&p->thread_group,
2312                                           &p->group_leader->thread_group);
2313                         list_add_tail_rcu(&p->thread_node,
2314                                           &p->signal->thread_head);
2315                 }
2316                 attach_pid(p, PIDTYPE_PID);
2317                 nr_threads++;
2318         }
2319         total_forks++;
2320         hlist_del_init(&delayed.node);
2321         spin_unlock(&current->sighand->siglock);
2322         syscall_tracepoint_update(p);
2323         write_unlock_irq(&tasklist_lock);
2324
2325         proc_fork_connector(p);
2326         sched_post_fork(p);
2327         cgroup_post_fork(p, args);
2328         perf_event_fork(p);
2329
2330         trace_task_newtask(p, clone_flags);
2331         uprobe_copy_process(p, clone_flags);
2332
2333         copy_oom_score_adj(clone_flags, p);
2334
2335         return p;
2336
2337 bad_fork_cancel_cgroup:
2338         spin_unlock(&current->sighand->siglock);
2339         write_unlock_irq(&tasklist_lock);
2340         cgroup_cancel_fork(p, args);
2341 bad_fork_put_pidfd:
2342         if (clone_flags & CLONE_PIDFD) {
2343                 fput(pidfile);
2344                 put_unused_fd(pidfd);
2345         }
2346 bad_fork_free_pid:
2347         if (pid != &init_struct_pid)
2348                 free_pid(pid);
2349 bad_fork_cleanup_thread:
2350         exit_thread(p);
2351 bad_fork_cleanup_io:
2352         if (p->io_context)
2353                 exit_io_context(p);
2354 bad_fork_cleanup_namespaces:
2355         exit_task_namespaces(p);
2356 bad_fork_cleanup_mm:
2357         if (p->mm) {
2358                 mm_clear_owner(p->mm, p);
2359                 mmput(p->mm);
2360         }
2361 bad_fork_cleanup_signal:
2362         if (!(clone_flags & CLONE_THREAD))
2363                 free_signal_struct(p->signal);
2364 bad_fork_cleanup_sighand:
2365         __cleanup_sighand(p->sighand);
2366 bad_fork_cleanup_fs:
2367         exit_fs(p); /* blocking */
2368 bad_fork_cleanup_files:
2369         exit_files(p); /* blocking */
2370 bad_fork_cleanup_semundo:
2371         exit_sem(p);
2372 bad_fork_cleanup_security:
2373         security_task_free(p);
2374 bad_fork_cleanup_audit:
2375         audit_free(p);
2376 bad_fork_cleanup_perf:
2377         perf_event_free_task(p);
2378 bad_fork_cleanup_policy:
2379         lockdep_free_task(p);
2380 #ifdef CONFIG_NUMA
2381         mpol_put(p->mempolicy);
2382 bad_fork_cleanup_threadgroup_lock:
2383 #endif
2384         delayacct_tsk_free(p);
2385 bad_fork_cleanup_count:
2386         atomic_dec(&p->cred->user->processes);
2387         exit_creds(p);
2388 bad_fork_free:
2389         p->state = TASK_DEAD;
2390         put_task_stack(p);
2391         delayed_free_task(p);
2392 fork_out:
2393         spin_lock_irq(&current->sighand->siglock);
2394         hlist_del_init(&delayed.node);
2395         spin_unlock_irq(&current->sighand->siglock);
2396         return ERR_PTR(retval);
2397 }
2398
2399 static inline void init_idle_pids(struct task_struct *idle)
2400 {
2401         enum pid_type type;
2402
2403         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2404                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2405                 init_task_pid(idle, type, &init_struct_pid);
2406         }
2407 }
2408
2409 struct task_struct *fork_idle(int cpu)
2410 {
2411         struct task_struct *task;
2412         struct kernel_clone_args args = {
2413                 .flags = CLONE_VM,
2414         };
2415
2416         task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2417         if (!IS_ERR(task)) {
2418                 init_idle_pids(task);
2419                 init_idle(task, cpu);
2420         }
2421
2422         return task;
2423 }
2424
2425 struct mm_struct *copy_init_mm(void)
2426 {
2427         return dup_mm(NULL, &init_mm);
2428 }
2429
2430 /*
2431  * This is like kernel_clone(), but shaved down and tailored to just
2432  * creating io_uring workers. It returns a created task, or an error pointer.
2433  * The returned task is inactive, and the caller must fire it up through
2434  * wake_up_new_task(p). All signals are blocked in the created task.
2435  */
2436 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2437 {
2438         unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2439                                 CLONE_IO;
2440         struct kernel_clone_args args = {
2441                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2442                                     CLONE_UNTRACED) & ~CSIGNAL),
2443                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2444                 .stack          = (unsigned long)fn,
2445                 .stack_size     = (unsigned long)arg,
2446                 .io_thread      = 1,
2447         };
2448
2449         return copy_process(NULL, 0, node, &args);
2450 }
2451
2452 /*
2453  *  Ok, this is the main fork-routine.
2454  *
2455  * It copies the process, and if successful kick-starts
2456  * it and waits for it to finish using the VM if required.
2457  *
2458  * args->exit_signal is expected to be checked for sanity by the caller.
2459  */
2460 pid_t kernel_clone(struct kernel_clone_args *args)
2461 {
2462         u64 clone_flags = args->flags;
2463         struct completion vfork;
2464         struct pid *pid;
2465         struct task_struct *p;
2466         int trace = 0;
2467         pid_t nr;
2468
2469         /*
2470          * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2471          * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2472          * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2473          * field in struct clone_args and it still doesn't make sense to have
2474          * them both point at the same memory location. Performing this check
2475          * here has the advantage that we don't need to have a separate helper
2476          * to check for legacy clone().
2477          */
2478         if ((args->flags & CLONE_PIDFD) &&
2479             (args->flags & CLONE_PARENT_SETTID) &&
2480             (args->pidfd == args->parent_tid))
2481                 return -EINVAL;
2482
2483         /*
2484          * Determine whether and which event to report to ptracer.  When
2485          * called from kernel_thread or CLONE_UNTRACED is explicitly
2486          * requested, no event is reported; otherwise, report if the event
2487          * for the type of forking is enabled.
2488          */
2489         if (!(clone_flags & CLONE_UNTRACED)) {
2490                 if (clone_flags & CLONE_VFORK)
2491                         trace = PTRACE_EVENT_VFORK;
2492                 else if (args->exit_signal != SIGCHLD)
2493                         trace = PTRACE_EVENT_CLONE;
2494                 else
2495                         trace = PTRACE_EVENT_FORK;
2496
2497                 if (likely(!ptrace_event_enabled(current, trace)))
2498                         trace = 0;
2499         }
2500
2501         p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2502         add_latent_entropy();
2503
2504         if (IS_ERR(p))
2505                 return PTR_ERR(p);
2506
2507         /*
2508          * Do this prior waking up the new thread - the thread pointer
2509          * might get invalid after that point, if the thread exits quickly.
2510          */
2511         trace_sched_process_fork(current, p);
2512
2513         pid = get_task_pid(p, PIDTYPE_PID);
2514         nr = pid_vnr(pid);
2515
2516         if (clone_flags & CLONE_PARENT_SETTID)
2517                 put_user(nr, args->parent_tid);
2518
2519         if (clone_flags & CLONE_VFORK) {
2520                 p->vfork_done = &vfork;
2521                 init_completion(&vfork);
2522                 get_task_struct(p);
2523         }
2524
2525         wake_up_new_task(p);
2526
2527         /* forking complete and child started to run, tell ptracer */
2528         if (unlikely(trace))
2529                 ptrace_event_pid(trace, pid);
2530
2531         if (clone_flags & CLONE_VFORK) {
2532                 if (!wait_for_vfork_done(p, &vfork))
2533                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2534         }
2535
2536         put_pid(pid);
2537         return nr;
2538 }
2539
2540 /*
2541  * Create a kernel thread.
2542  */
2543 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2544 {
2545         struct kernel_clone_args args = {
2546                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2547                                     CLONE_UNTRACED) & ~CSIGNAL),
2548                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2549                 .stack          = (unsigned long)fn,
2550                 .stack_size     = (unsigned long)arg,
2551         };
2552
2553         return kernel_clone(&args);
2554 }
2555
2556 #ifdef __ARCH_WANT_SYS_FORK
2557 SYSCALL_DEFINE0(fork)
2558 {
2559 #ifdef CONFIG_MMU
2560         struct kernel_clone_args args = {
2561                 .exit_signal = SIGCHLD,
2562         };
2563
2564         return kernel_clone(&args);
2565 #else
2566         /* can not support in nommu mode */
2567         return -EINVAL;
2568 #endif
2569 }
2570 #endif
2571
2572 #ifdef __ARCH_WANT_SYS_VFORK
2573 SYSCALL_DEFINE0(vfork)
2574 {
2575         struct kernel_clone_args args = {
2576                 .flags          = CLONE_VFORK | CLONE_VM,
2577                 .exit_signal    = SIGCHLD,
2578         };
2579
2580         return kernel_clone(&args);
2581 }
2582 #endif
2583
2584 #ifdef __ARCH_WANT_SYS_CLONE
2585 #ifdef CONFIG_CLONE_BACKWARDS
2586 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2587                  int __user *, parent_tidptr,
2588                  unsigned long, tls,
2589                  int __user *, child_tidptr)
2590 #elif defined(CONFIG_CLONE_BACKWARDS2)
2591 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2592                  int __user *, parent_tidptr,
2593                  int __user *, child_tidptr,
2594                  unsigned long, tls)
2595 #elif defined(CONFIG_CLONE_BACKWARDS3)
2596 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2597                 int, stack_size,
2598                 int __user *, parent_tidptr,
2599                 int __user *, child_tidptr,
2600                 unsigned long, tls)
2601 #else
2602 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2603                  int __user *, parent_tidptr,
2604                  int __user *, child_tidptr,
2605                  unsigned long, tls)
2606 #endif
2607 {
2608         struct kernel_clone_args args = {
2609                 .flags          = (lower_32_bits(clone_flags) & ~CSIGNAL),
2610                 .pidfd          = parent_tidptr,
2611                 .child_tid      = child_tidptr,
2612                 .parent_tid     = parent_tidptr,
2613                 .exit_signal    = (lower_32_bits(clone_flags) & CSIGNAL),
2614                 .stack          = newsp,
2615                 .tls            = tls,
2616         };
2617
2618         return kernel_clone(&args);
2619 }
2620 #endif
2621
2622 #ifdef __ARCH_WANT_SYS_CLONE3
2623
2624 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2625                                               struct clone_args __user *uargs,
2626                                               size_t usize)
2627 {
2628         int err;
2629         struct clone_args args;
2630         pid_t *kset_tid = kargs->set_tid;
2631
2632         BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2633                      CLONE_ARGS_SIZE_VER0);
2634         BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2635                      CLONE_ARGS_SIZE_VER1);
2636         BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2637                      CLONE_ARGS_SIZE_VER2);
2638         BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2639
2640         if (unlikely(usize > PAGE_SIZE))
2641                 return -E2BIG;
2642         if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2643                 return -EINVAL;
2644
2645         err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2646         if (err)
2647                 return err;
2648
2649         if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2650                 return -EINVAL;
2651
2652         if (unlikely(!args.set_tid && args.set_tid_size > 0))
2653                 return -EINVAL;
2654
2655         if (unlikely(args.set_tid && args.set_tid_size == 0))
2656                 return -EINVAL;
2657
2658         /*
2659          * Verify that higher 32bits of exit_signal are unset and that
2660          * it is a valid signal
2661          */
2662         if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2663                      !valid_signal(args.exit_signal)))
2664                 return -EINVAL;
2665
2666         if ((args.flags & CLONE_INTO_CGROUP) &&
2667             (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2668                 return -EINVAL;
2669
2670         *kargs = (struct kernel_clone_args){
2671                 .flags          = args.flags,
2672                 .pidfd          = u64_to_user_ptr(args.pidfd),
2673                 .child_tid      = u64_to_user_ptr(args.child_tid),
2674                 .parent_tid     = u64_to_user_ptr(args.parent_tid),
2675                 .exit_signal    = args.exit_signal,
2676                 .stack          = args.stack,
2677                 .stack_size     = args.stack_size,
2678                 .tls            = args.tls,
2679                 .set_tid_size   = args.set_tid_size,
2680                 .cgroup         = args.cgroup,
2681         };
2682
2683         if (args.set_tid &&
2684                 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2685                         (kargs->set_tid_size * sizeof(pid_t))))
2686                 return -EFAULT;
2687
2688         kargs->set_tid = kset_tid;
2689
2690         return 0;
2691 }
2692
2693 /**
2694  * clone3_stack_valid - check and prepare stack
2695  * @kargs: kernel clone args
2696  *
2697  * Verify that the stack arguments userspace gave us are sane.
2698  * In addition, set the stack direction for userspace since it's easy for us to
2699  * determine.
2700  */
2701 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2702 {
2703         if (kargs->stack == 0) {
2704                 if (kargs->stack_size > 0)
2705                         return false;
2706         } else {
2707                 if (kargs->stack_size == 0)
2708                         return false;
2709
2710                 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2711                         return false;
2712
2713 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2714                 kargs->stack += kargs->stack_size;
2715 #endif
2716         }
2717
2718         return true;
2719 }
2720
2721 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2722 {
2723         /* Verify that no unknown flags are passed along. */
2724         if (kargs->flags &
2725             ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2726                 return false;
2727
2728         /*
2729          * - make the CLONE_DETACHED bit reusable for clone3
2730          * - make the CSIGNAL bits reusable for clone3
2731          */
2732         if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2733                 return false;
2734
2735         if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2736             (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2737                 return false;
2738
2739         if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2740             kargs->exit_signal)
2741                 return false;
2742
2743         if (!clone3_stack_valid(kargs))
2744                 return false;
2745
2746         return true;
2747 }
2748
2749 /**
2750  * clone3 - create a new process with specific properties
2751  * @uargs: argument structure
2752  * @size:  size of @uargs
2753  *
2754  * clone3() is the extensible successor to clone()/clone2().
2755  * It takes a struct as argument that is versioned by its size.
2756  *
2757  * Return: On success, a positive PID for the child process.
2758  *         On error, a negative errno number.
2759  */
2760 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2761 {
2762         int err;
2763
2764         struct kernel_clone_args kargs;
2765         pid_t set_tid[MAX_PID_NS_LEVEL];
2766
2767         kargs.set_tid = set_tid;
2768
2769         err = copy_clone_args_from_user(&kargs, uargs, size);
2770         if (err)
2771                 return err;
2772
2773         if (!clone3_args_valid(&kargs))
2774                 return -EINVAL;
2775
2776         return kernel_clone(&kargs);
2777 }
2778 #endif
2779
2780 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2781 {
2782         struct task_struct *leader, *parent, *child;
2783         int res;
2784
2785         read_lock(&tasklist_lock);
2786         leader = top = top->group_leader;
2787 down:
2788         for_each_thread(leader, parent) {
2789                 list_for_each_entry(child, &parent->children, sibling) {
2790                         res = visitor(child, data);
2791                         if (res) {
2792                                 if (res < 0)
2793                                         goto out;
2794                                 leader = child;
2795                                 goto down;
2796                         }
2797 up:
2798                         ;
2799                 }
2800         }
2801
2802         if (leader != top) {
2803                 child = leader;
2804                 parent = child->real_parent;
2805                 leader = parent->group_leader;
2806                 goto up;
2807         }
2808 out:
2809         read_unlock(&tasklist_lock);
2810 }
2811
2812 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2813 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2814 #endif
2815
2816 static void sighand_ctor(void *data)
2817 {
2818         struct sighand_struct *sighand = data;
2819
2820         spin_lock_init(&sighand->siglock);
2821         init_waitqueue_head(&sighand->signalfd_wqh);
2822 }
2823
2824 void __init proc_caches_init(void)
2825 {
2826         unsigned int mm_size;
2827
2828         sighand_cachep = kmem_cache_create("sighand_cache",
2829                         sizeof(struct sighand_struct), 0,
2830                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2831                         SLAB_ACCOUNT, sighand_ctor);
2832         signal_cachep = kmem_cache_create("signal_cache",
2833                         sizeof(struct signal_struct), 0,
2834                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2835                         NULL);
2836         files_cachep = kmem_cache_create("files_cache",
2837                         sizeof(struct files_struct), 0,
2838                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2839                         NULL);
2840         fs_cachep = kmem_cache_create("fs_cache",
2841                         sizeof(struct fs_struct), 0,
2842                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2843                         NULL);
2844
2845         /*
2846          * The mm_cpumask is located at the end of mm_struct, and is
2847          * dynamically sized based on the maximum CPU number this system
2848          * can have, taking hotplug into account (nr_cpu_ids).
2849          */
2850         mm_size = sizeof(struct mm_struct) + cpumask_size();
2851
2852         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2853                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2854                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2855                         offsetof(struct mm_struct, saved_auxv),
2856                         sizeof_field(struct mm_struct, saved_auxv),
2857                         NULL);
2858         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2859         mmap_init();
2860         nsproxy_cache_init();
2861 }
2862
2863 /*
2864  * Check constraints on flags passed to the unshare system call.
2865  */
2866 static int check_unshare_flags(unsigned long unshare_flags)
2867 {
2868         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2869                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2870                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2871                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
2872                                 CLONE_NEWTIME))
2873                 return -EINVAL;
2874         /*
2875          * Not implemented, but pretend it works if there is nothing
2876          * to unshare.  Note that unsharing the address space or the
2877          * signal handlers also need to unshare the signal queues (aka
2878          * CLONE_THREAD).
2879          */
2880         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2881                 if (!thread_group_empty(current))
2882                         return -EINVAL;
2883         }
2884         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2885                 if (refcount_read(&current->sighand->count) > 1)
2886                         return -EINVAL;
2887         }
2888         if (unshare_flags & CLONE_VM) {
2889                 if (!current_is_single_threaded())
2890                         return -EINVAL;
2891         }
2892
2893         return 0;
2894 }
2895
2896 /*
2897  * Unshare the filesystem structure if it is being shared
2898  */
2899 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2900 {
2901         struct fs_struct *fs = current->fs;
2902
2903         if (!(unshare_flags & CLONE_FS) || !fs)
2904                 return 0;
2905
2906         /* don't need lock here; in the worst case we'll do useless copy */
2907         if (fs->users == 1)
2908                 return 0;
2909
2910         *new_fsp = copy_fs_struct(fs);
2911         if (!*new_fsp)
2912                 return -ENOMEM;
2913
2914         return 0;
2915 }
2916
2917 /*
2918  * Unshare file descriptor table if it is being shared
2919  */
2920 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
2921                struct files_struct **new_fdp)
2922 {
2923         struct files_struct *fd = current->files;
2924         int error = 0;
2925
2926         if ((unshare_flags & CLONE_FILES) &&
2927             (fd && atomic_read(&fd->count) > 1)) {
2928                 *new_fdp = dup_fd(fd, max_fds, &error);
2929                 if (!*new_fdp)
2930                         return error;
2931         }
2932
2933         return 0;
2934 }
2935
2936 /*
2937  * unshare allows a process to 'unshare' part of the process
2938  * context which was originally shared using clone.  copy_*
2939  * functions used by kernel_clone() cannot be used here directly
2940  * because they modify an inactive task_struct that is being
2941  * constructed. Here we are modifying the current, active,
2942  * task_struct.
2943  */
2944 int ksys_unshare(unsigned long unshare_flags)
2945 {
2946         struct fs_struct *fs, *new_fs = NULL;
2947         struct files_struct *fd, *new_fd = NULL;
2948         struct cred *new_cred = NULL;
2949         struct nsproxy *new_nsproxy = NULL;
2950         int do_sysvsem = 0;
2951         int err;
2952
2953         /*
2954          * If unsharing a user namespace must also unshare the thread group
2955          * and unshare the filesystem root and working directories.
2956          */
2957         if (unshare_flags & CLONE_NEWUSER)
2958                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2959         /*
2960          * If unsharing vm, must also unshare signal handlers.
2961          */
2962         if (unshare_flags & CLONE_VM)
2963                 unshare_flags |= CLONE_SIGHAND;
2964         /*
2965          * If unsharing a signal handlers, must also unshare the signal queues.
2966          */
2967         if (unshare_flags & CLONE_SIGHAND)
2968                 unshare_flags |= CLONE_THREAD;
2969         /*
2970          * If unsharing namespace, must also unshare filesystem information.
2971          */
2972         if (unshare_flags & CLONE_NEWNS)
2973                 unshare_flags |= CLONE_FS;
2974
2975         err = check_unshare_flags(unshare_flags);
2976         if (err)
2977                 goto bad_unshare_out;
2978         /*
2979          * CLONE_NEWIPC must also detach from the undolist: after switching
2980          * to a new ipc namespace, the semaphore arrays from the old
2981          * namespace are unreachable.
2982          */
2983         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2984                 do_sysvsem = 1;
2985         err = unshare_fs(unshare_flags, &new_fs);
2986         if (err)
2987                 goto bad_unshare_out;
2988         err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
2989         if (err)
2990                 goto bad_unshare_cleanup_fs;
2991         err = unshare_userns(unshare_flags, &new_cred);
2992         if (err)
2993                 goto bad_unshare_cleanup_fd;
2994         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2995                                          new_cred, new_fs);
2996         if (err)
2997                 goto bad_unshare_cleanup_cred;
2998
2999         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3000                 if (do_sysvsem) {
3001                         /*
3002                          * CLONE_SYSVSEM is equivalent to sys_exit().
3003                          */
3004                         exit_sem(current);
3005                 }
3006                 if (unshare_flags & CLONE_NEWIPC) {
3007                         /* Orphan segments in old ns (see sem above). */
3008                         exit_shm(current);
3009                         shm_init_task(current);
3010                 }
3011
3012                 if (new_nsproxy)
3013                         switch_task_namespaces(current, new_nsproxy);
3014
3015                 task_lock(current);
3016
3017                 if (new_fs) {
3018                         fs = current->fs;
3019                         spin_lock(&fs->lock);
3020                         current->fs = new_fs;
3021                         if (--fs->users)
3022                                 new_fs = NULL;
3023                         else
3024                                 new_fs = fs;
3025                         spin_unlock(&fs->lock);
3026                 }
3027
3028                 if (new_fd) {
3029                         fd = current->files;
3030                         current->files = new_fd;
3031                         new_fd = fd;
3032                 }
3033
3034                 task_unlock(current);
3035
3036                 if (new_cred) {
3037                         /* Install the new user namespace */
3038                         commit_creds(new_cred);
3039                         new_cred = NULL;
3040                 }
3041         }
3042
3043         perf_event_namespaces(current);
3044
3045 bad_unshare_cleanup_cred:
3046         if (new_cred)
3047                 put_cred(new_cred);
3048 bad_unshare_cleanup_fd:
3049         if (new_fd)
3050                 put_files_struct(new_fd);
3051
3052 bad_unshare_cleanup_fs:
3053         if (new_fs)
3054                 free_fs_struct(new_fs);
3055
3056 bad_unshare_out:
3057         return err;
3058 }
3059
3060 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3061 {
3062         return ksys_unshare(unshare_flags);
3063 }
3064
3065 /*
3066  *      Helper to unshare the files of the current task.
3067  *      We don't want to expose copy_files internals to
3068  *      the exec layer of the kernel.
3069  */
3070
3071 int unshare_files(void)
3072 {
3073         struct task_struct *task = current;
3074         struct files_struct *old, *copy = NULL;
3075         int error;
3076
3077         error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, &copy);
3078         if (error || !copy)
3079                 return error;
3080
3081         old = task->files;
3082         task_lock(task);
3083         task->files = copy;
3084         task_unlock(task);
3085         put_files_struct(old);
3086         return 0;
3087 }
3088
3089 int sysctl_max_threads(struct ctl_table *table, int write,
3090                        void *buffer, size_t *lenp, loff_t *ppos)
3091 {
3092         struct ctl_table t;
3093         int ret;
3094         int threads = max_threads;
3095         int min = 1;
3096         int max = MAX_THREADS;
3097
3098         t = *table;
3099         t.data = &threads;
3100         t.extra1 = &min;
3101         t.extra2 = &max;
3102
3103         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3104         if (ret || !write)
3105                 return ret;
3106
3107         max_threads = threads;
3108
3109         return 0;
3110 }