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         sched_core_free(tsk);
 746
 747         if (!profile_handoff_task(tsk))
 748                 free_task(tsk);
 749 }
 750 EXPORT_SYMBOL_GPL(__put_task_struct);
 751
 752 void __init __weak arch_task_cache_init(void) { }
 753
 754 /*
 755  * set_max_threads
 756  */
 757 static void set_max_threads(unsigned int max_threads_suggested)
 758 {
 759         u64 threads;
 760         unsigned long nr_pages = totalram_pages();
 761
 762         /*
 763          * The number of threads shall be limited such that the thread
 764          * structures may only consume a small part of the available memory.
 765          */
 766         if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
 767                 threads = MAX_THREADS;
 768         else
 769                 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
 770                                     (u64) THREAD_SIZE * 8UL);
 771
 772         if (threads > max_threads_suggested)
 773                 threads = max_threads_suggested;
 774
 775         max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
 776 }
 777
 778 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
 779 /* Initialized by the architecture: */
 780 int arch_task_struct_size __read_mostly;
 781 #endif
 782
 783 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 784 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
 785 {
 786         /* Fetch thread_struct whitelist for the architecture. */
 787         arch_thread_struct_whitelist(offset, size);
 788
 789         /*
 790          * Handle zero-sized whitelist or empty thread_struct, otherwise
 791          * adjust offset to position of thread_struct in task_struct.
 792          */
 793         if (unlikely(*size == 0))
 794                 *offset = 0;
 795         else
 796                 *offset += offsetof(struct task_struct, thread);
 797 }
 798 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
 799
 800 void __init fork_init(void)
 801 {
 802         int i;
 803 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
 804 #ifndef ARCH_MIN_TASKALIGN
 805 #define ARCH_MIN_TASKALIGN      0
 806 #endif
 807         int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
 808         unsigned long useroffset, usersize;
 809
 810         /* create a slab on which task_structs can be allocated */
 811         task_struct_whitelist(&useroffset, &usersize);
 812         task_struct_cachep = kmem_cache_create_usercopy("task_struct",
 813                         arch_task_struct_size, align,
 814                         SLAB_PANIC|SLAB_ACCOUNT,
 815                         useroffset, usersize, NULL);
 816 #endif
 817
 818         /* do the arch specific task caches init */
 819         arch_task_cache_init();
 820
 821         set_max_threads(MAX_THREADS);
 822
 823         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
 824         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
 825         init_task.signal->rlim[RLIMIT_SIGPENDING] =
 826                 init_task.signal->rlim[RLIMIT_NPROC];
 827
 828         for (i = 0; i < UCOUNT_COUNTS; i++)
 829                 init_user_ns.ucount_max[i] = max_threads/2;
 830
 831 #ifdef CONFIG_VMAP_STACK
 832         cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
 833                           NULL, free_vm_stack_cache);
 834 #endif
 835
 836         scs_init();
 837
 838         lockdep_init_task(&init_task);
 839         uprobes_init();
 840 }
 841
 842 int __weak arch_dup_task_struct(struct task_struct *dst,
 843                                                struct task_struct *src)
 844 {
 845         *dst = *src;
 846         return 0;
 847 }
 848
 849 void set_task_stack_end_magic(struct task_struct *tsk)
 850 {
 851         unsigned long *stackend;
 852
 853         stackend = end_of_stack(tsk);
 854         *stackend = STACK_END_MAGIC;    /* for overflow detection */
 855 }
 856
 857 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
 858 {
 859         struct task_struct *tsk;
 860         unsigned long *stack;
 861         struct vm_struct *stack_vm_area __maybe_unused;
 862         int err;
 863
 864         if (node == NUMA_NO_NODE)
 865                 node = tsk_fork_get_node(orig);
 866         tsk = alloc_task_struct_node(node);
 867         if (!tsk)
 868                 return NULL;
 869
 870         stack = alloc_thread_stack_node(tsk, node);
 871         if (!stack)
 872                 goto free_tsk;
 873
 874         if (memcg_charge_kernel_stack(tsk))
 875                 goto free_stack;
 876
 877         stack_vm_area = task_stack_vm_area(tsk);
 878
 879         err = arch_dup_task_struct(tsk, orig);
 880
 881         /*
 882          * arch_dup_task_struct() clobbers the stack-related fields.  Make
 883          * sure they're properly initialized before using any stack-related
 884          * functions again.
 885          */
 886         tsk->stack = stack;
 887 #ifdef CONFIG_VMAP_STACK
 888         tsk->stack_vm_area = stack_vm_area;
 889 #endif
 890 #ifdef CONFIG_THREAD_INFO_IN_TASK
 891         refcount_set(&tsk->stack_refcount, 1);
 892 #endif
 893
 894         if (err)
 895                 goto free_stack;
 896
 897         err = scs_prepare(tsk, node);
 898         if (err)
 899                 goto free_stack;
 900
 901 #ifdef CONFIG_SECCOMP
 902         /*
 903          * We must handle setting up seccomp filters once we're under
 904          * the sighand lock in case orig has changed between now and
 905          * then. Until then, filter must be NULL to avoid messing up
 906          * the usage counts on the error path calling free_task.
 907          */
 908         tsk->seccomp.filter = NULL;
 909 #endif
 910
 911         setup_thread_stack(tsk, orig);
 912         clear_user_return_notifier(tsk);
 913         clear_tsk_need_resched(tsk);
 914         set_task_stack_end_magic(tsk);
 915         clear_syscall_work_syscall_user_dispatch(tsk);
 916
 917 #ifdef CONFIG_STACKPROTECTOR
 918         tsk->stack_canary = get_random_canary();
 919 #endif
 920         if (orig->cpus_ptr == &orig->cpus_mask)
 921                 tsk->cpus_ptr = &tsk->cpus_mask;
 922
 923         /*
 924          * One for the user space visible state that goes away when reaped.
 925          * One for the scheduler.
 926          */
 927         refcount_set(&tsk->rcu_users, 2);
 928         /* One for the rcu users */
 929         refcount_set(&tsk->usage, 1);
 930 #ifdef CONFIG_BLK_DEV_IO_TRACE
 931         tsk->btrace_seq = 0;
 932 #endif
 933         tsk->splice_pipe = NULL;
 934         tsk->task_frag.page = NULL;
 935         tsk->wake_q.next = NULL;
 936         tsk->pf_io_worker = NULL;
 937
 938         account_kernel_stack(tsk, 1);
 939
 940         kcov_task_init(tsk);
 941         kmap_local_fork(tsk);
 942
 943 #ifdef CONFIG_FAULT_INJECTION
 944         tsk->fail_nth = 0;
 945 #endif
 946
 947 #ifdef CONFIG_BLK_CGROUP
 948         tsk->throttle_queue = NULL;
 949         tsk->use_memdelay = 0;
 950 #endif
 951
 952 #ifdef CONFIG_MEMCG
 953         tsk->active_memcg = NULL;
 954 #endif
 955         return tsk;
 956
 957 free_stack:
 958         free_thread_stack(tsk);
 959 free_tsk:
 960         free_task_struct(tsk);
 961         return NULL;
 962 }
 963
 964 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 965
 966 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
 967
 968 static int __init coredump_filter_setup(char *s)
 969 {
 970         default_dump_filter =
 971                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
 972                 MMF_DUMP_FILTER_MASK;
 973         return 1;
 974 }
 975
 976 __setup("coredump_filter=", coredump_filter_setup);
 977
 978 #include <linux/init_task.h>
 979
 980 static void mm_init_aio(struct mm_struct *mm)
 981 {
 982 #ifdef CONFIG_AIO
 983         spin_lock_init(&mm->ioctx_lock);
 984         mm->ioctx_table = NULL;
 985 #endif
 986 }
 987
 988 static __always_inline void mm_clear_owner(struct mm_struct *mm,
 989                                            struct task_struct *p)
 990 {
 991 #ifdef CONFIG_MEMCG
 992         if (mm->owner == p)
 993                 WRITE_ONCE(mm->owner, NULL);
 994 #endif
 995 }
 996
 997 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
 998 {
 999 #ifdef CONFIG_MEMCG
1000         mm->owner = p;
1001 #endif
1002 }
1003
1004 static void mm_init_pasid(struct mm_struct *mm)
1005 {
1006 #ifdef CONFIG_IOMMU_SUPPORT
1007         mm->pasid = INIT_PASID;
1008 #endif
1009 }
1010
1011 static void mm_init_uprobes_state(struct mm_struct *mm)
1012 {
1013 #ifdef CONFIG_UPROBES
1014         mm->uprobes_state.xol_area = NULL;
1015 #endif
1016 }
1017
1018 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1019         struct user_namespace *user_ns)
1020 {
1021         mm->mmap = NULL;
1022         mm->mm_rb = RB_ROOT;
1023         mm->vmacache_seqnum = 0;
1024         atomic_set(&mm->mm_users, 1);
1025         atomic_set(&mm->mm_count, 1);
1026         seqcount_init(&mm->write_protect_seq);
1027         mmap_init_lock(mm);
1028         INIT_LIST_HEAD(&mm->mmlist);
1029         mm->core_state = NULL;
1030         mm_pgtables_bytes_init(mm);
1031         mm->map_count = 0;
1032         mm->locked_vm = 0;
1033         atomic_set(&mm->has_pinned, 0);
1034         atomic64_set(&mm->pinned_vm, 0);
1035         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1036         spin_lock_init(&mm->page_table_lock);
1037         spin_lock_init(&mm->arg_lock);
1038         mm_init_cpumask(mm);
1039         mm_init_aio(mm);
1040         mm_init_owner(mm, p);
1041         mm_init_pasid(mm);
1042         RCU_INIT_POINTER(mm->exe_file, NULL);
1043         mmu_notifier_subscriptions_init(mm);
1044         init_tlb_flush_pending(mm);
1045 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1046         mm->pmd_huge_pte = NULL;
1047 #endif
1048         mm_init_uprobes_state(mm);
1049
1050         if (current->mm) {
1051                 mm->flags = current->mm->flags & MMF_INIT_MASK;
1052                 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1053         } else {
1054                 mm->flags = default_dump_filter;
1055                 mm->def_flags = 0;
1056         }
1057
1058         if (mm_alloc_pgd(mm))
1059                 goto fail_nopgd;
1060
1061         if (init_new_context(p, mm))
1062                 goto fail_nocontext;
1063
1064         mm->user_ns = get_user_ns(user_ns);
1065         return mm;
1066
1067 fail_nocontext:
1068         mm_free_pgd(mm);
1069 fail_nopgd:
1070         free_mm(mm);
1071         return NULL;
1072 }
1073
1074 /*
1075  * Allocate and initialize an mm_struct.
1076  */
1077 struct mm_struct *mm_alloc(void)
1078 {
1079         struct mm_struct *mm;
1080
1081         mm = allocate_mm();
1082         if (!mm)
1083                 return NULL;
1084
1085         memset(mm, 0, sizeof(*mm));
1086         return mm_init(mm, current, current_user_ns());
1087 }
1088
1089 static inline void __mmput(struct mm_struct *mm)
1090 {
1091         VM_BUG_ON(atomic_read(&mm->mm_users));
1092
1093         uprobe_clear_state(mm);
1094         exit_aio(mm);
1095         ksm_exit(mm);
1096         khugepaged_exit(mm); /* must run before exit_mmap */
1097         exit_mmap(mm);
1098         mm_put_huge_zero_page(mm);
1099         set_mm_exe_file(mm, NULL);
1100         if (!list_empty(&mm->mmlist)) {
1101                 spin_lock(&mmlist_lock);
1102                 list_del(&mm->mmlist);
1103                 spin_unlock(&mmlist_lock);
1104         }
1105         if (mm->binfmt)
1106                 module_put(mm->binfmt->module);
1107         mmdrop(mm);
1108 }
1109
1110 /*
1111  * Decrement the use count and release all resources for an mm.
1112  */
1113 void mmput(struct mm_struct *mm)
1114 {
1115         might_sleep();
1116
1117         if (atomic_dec_and_test(&mm->mm_users))
1118                 __mmput(mm);
1119 }
1120 EXPORT_SYMBOL_GPL(mmput);
1121
1122 #ifdef CONFIG_MMU
1123 static void mmput_async_fn(struct work_struct *work)
1124 {
1125         struct mm_struct *mm = container_of(work, struct mm_struct,
1126                                             async_put_work);
1127
1128         __mmput(mm);
1129 }
1130
1131 void mmput_async(struct mm_struct *mm)
1132 {
1133         if (atomic_dec_and_test(&mm->mm_users)) {
1134                 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1135                 schedule_work(&mm->async_put_work);
1136         }
1137 }
1138 #endif
1139
1140 /**
1141  * set_mm_exe_file - change a reference to the mm's executable file
1142  *
1143  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1144  *
1145  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1146  * invocations: in mmput() nobody alive left, in execve task is single
1147  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1148  * mm->exe_file, but does so without using set_mm_exe_file() in order
1149  * to avoid the need for any locks.
1150  */
1151 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1152 {
1153         struct file *old_exe_file;
1154
1155         /*
1156          * It is safe to dereference the exe_file without RCU as
1157          * this function is only called if nobody else can access
1158          * this mm -- see comment above for justification.
1159          */
1160         old_exe_file = rcu_dereference_raw(mm->exe_file);
1161
1162         if (new_exe_file)
1163                 get_file(new_exe_file);
1164         rcu_assign_pointer(mm->exe_file, new_exe_file);
1165         if (old_exe_file)
1166                 fput(old_exe_file);
1167 }
1168
1169 /**
1170  * get_mm_exe_file - acquire a reference to the mm's executable file
1171  *
1172  * Returns %NULL if mm has no associated executable file.
1173  * User must release file via fput().
1174  */
1175 struct file *get_mm_exe_file(struct mm_struct *mm)
1176 {
1177         struct file *exe_file;
1178
1179         rcu_read_lock();
1180         exe_file = rcu_dereference(mm->exe_file);
1181         if (exe_file && !get_file_rcu(exe_file))
1182                 exe_file = NULL;
1183         rcu_read_unlock();
1184         return exe_file;
1185 }
1186 EXPORT_SYMBOL(get_mm_exe_file);
1187
1188 /**
1189  * get_task_exe_file - acquire a reference to the task's executable file
1190  *
1191  * Returns %NULL if task's mm (if any) has no associated executable file or
1192  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1193  * User must release file via fput().
1194  */
1195 struct file *get_task_exe_file(struct task_struct *task)
1196 {
1197         struct file *exe_file = NULL;
1198         struct mm_struct *mm;
1199
1200         task_lock(task);
1201         mm = task->mm;
1202         if (mm) {
1203                 if (!(task->flags & PF_KTHREAD))
1204                         exe_file = get_mm_exe_file(mm);
1205         }
1206         task_unlock(task);
1207         return exe_file;
1208 }
1209 EXPORT_SYMBOL(get_task_exe_file);
1210
1211 /**
1212  * get_task_mm - acquire a reference to the task's mm
1213  *
1214  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1215  * this kernel workthread has transiently adopted a user mm with use_mm,
1216  * to do its AIO) is not set and if so returns a reference to it, after
1217  * bumping up the use count.  User must release the mm via mmput()
1218  * after use.  Typically used by /proc and ptrace.
1219  */
1220 struct mm_struct *get_task_mm(struct task_struct *task)
1221 {
1222         struct mm_struct *mm;
1223
1224         task_lock(task);
1225         mm = task->mm;
1226         if (mm) {
1227                 if (task->flags & PF_KTHREAD)
1228                         mm = NULL;
1229                 else
1230                         mmget(mm);
1231         }
1232         task_unlock(task);
1233         return mm;
1234 }
1235 EXPORT_SYMBOL_GPL(get_task_mm);
1236
1237 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1238 {
1239         struct mm_struct *mm;
1240         int err;
1241
1242         err =  down_read_killable(&task->signal->exec_update_lock);
1243         if (err)
1244                 return ERR_PTR(err);
1245
1246         mm = get_task_mm(task);
1247         if (mm && mm != current->mm &&
1248                         !ptrace_may_access(task, mode)) {
1249                 mmput(mm);
1250                 mm = ERR_PTR(-EACCES);
1251         }
1252         up_read(&task->signal->exec_update_lock);
1253
1254         return mm;
1255 }
1256
1257 static void complete_vfork_done(struct task_struct *tsk)
1258 {
1259         struct completion *vfork;
1260
1261         task_lock(tsk);
1262         vfork = tsk->vfork_done;
1263         if (likely(vfork)) {
1264                 tsk->vfork_done = NULL;
1265                 complete(vfork);
1266         }
1267         task_unlock(tsk);
1268 }
1269
1270 static int wait_for_vfork_done(struct task_struct *child,
1271                                 struct completion *vfork)
1272 {
1273         int killed;
1274
1275         freezer_do_not_count();
1276         cgroup_enter_frozen();
1277         killed = wait_for_completion_killable(vfork);
1278         cgroup_leave_frozen(false);
1279         freezer_count();
1280
1281         if (killed) {
1282                 task_lock(child);
1283                 child->vfork_done = NULL;
1284                 task_unlock(child);
1285         }
1286
1287         put_task_struct(child);
1288         return killed;
1289 }
1290
1291 /* Please note the differences between mmput and mm_release.
1292  * mmput is called whenever we stop holding onto a mm_struct,
1293  * error success whatever.
1294  *
1295  * mm_release is called after a mm_struct has been removed
1296  * from the current process.
1297  *
1298  * This difference is important for error handling, when we
1299  * only half set up a mm_struct for a new process and need to restore
1300  * the old one.  Because we mmput the new mm_struct before
1301  * restoring the old one. . .
1302  * Eric Biederman 10 January 1998
1303  */
1304 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1305 {
1306         uprobe_free_utask(tsk);
1307
1308         /* Get rid of any cached register state */
1309         deactivate_mm(tsk, mm);
1310
1311         /*
1312          * Signal userspace if we're not exiting with a core dump
1313          * because we want to leave the value intact for debugging
1314          * purposes.
1315          */
1316         if (tsk->clear_child_tid) {
1317                 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1318                     atomic_read(&mm->mm_users) > 1) {
1319                         /*
1320                          * We don't check the error code - if userspace has
1321                          * not set up a proper pointer then tough luck.
1322                          */
1323                         put_user(0, tsk->clear_child_tid);
1324                         do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1325                                         1, NULL, NULL, 0, 0);
1326                 }
1327                 tsk->clear_child_tid = NULL;
1328         }
1329
1330         /*
1331          * All done, finally we can wake up parent and return this mm to him.
1332          * Also kthread_stop() uses this completion for synchronization.
1333          */
1334         if (tsk->vfork_done)
1335                 complete_vfork_done(tsk);
1336 }
1337
1338 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1339 {
1340         futex_exit_release(tsk);
1341         mm_release(tsk, mm);
1342 }
1343
1344 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1345 {
1346         futex_exec_release(tsk);
1347         mm_release(tsk, mm);
1348 }
1349
1350 /**
1351  * dup_mm() - duplicates an existing mm structure
1352  * @tsk: the task_struct with which the new mm will be associated.
1353  * @oldmm: the mm to duplicate.
1354  *
1355  * Allocates a new mm structure and duplicates the provided @oldmm structure
1356  * content into it.
1357  *
1358  * Return: the duplicated mm or NULL on failure.
1359  */
1360 static struct mm_struct *dup_mm(struct task_struct *tsk,
1361                                 struct mm_struct *oldmm)
1362 {
1363         struct mm_struct *mm;
1364         int err;
1365
1366         mm = allocate_mm();
1367         if (!mm)
1368                 goto fail_nomem;
1369
1370         memcpy(mm, oldmm, sizeof(*mm));
1371
1372         if (!mm_init(mm, tsk, mm->user_ns))
1373                 goto fail_nomem;
1374
1375         err = dup_mmap(mm, oldmm);
1376         if (err)
1377                 goto free_pt;
1378
1379         mm->hiwater_rss = get_mm_rss(mm);
1380         mm->hiwater_vm = mm->total_vm;
1381
1382         if (mm->binfmt && !try_module_get(mm->binfmt->module))
1383                 goto free_pt;
1384
1385         return mm;
1386
1387 free_pt:
1388         /* don't put binfmt in mmput, we haven't got module yet */
1389         mm->binfmt = NULL;
1390         mm_init_owner(mm, NULL);
1391         mmput(mm);
1392
1393 fail_nomem:
1394         return NULL;
1395 }
1396
1397 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1398 {
1399         struct mm_struct *mm, *oldmm;
1400
1401         tsk->min_flt = tsk->maj_flt = 0;
1402         tsk->nvcsw = tsk->nivcsw = 0;
1403 #ifdef CONFIG_DETECT_HUNG_TASK
1404         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1405         tsk->last_switch_time = 0;
1406 #endif
1407
1408         tsk->mm = NULL;
1409         tsk->active_mm = NULL;
1410
1411         /*
1412          * Are we cloning a kernel thread?
1413          *
1414          * We need to steal a active VM for that..
1415          */
1416         oldmm = current->mm;
1417         if (!oldmm)
1418                 return 0;
1419
1420         /* initialize the new vmacache entries */
1421         vmacache_flush(tsk);
1422
1423         if (clone_flags & CLONE_VM) {
1424                 mmget(oldmm);
1425                 mm = oldmm;
1426         } else {
1427                 mm = dup_mm(tsk, current->mm);
1428                 if (!mm)
1429                         return -ENOMEM;
1430         }
1431
1432         tsk->mm = mm;
1433         tsk->active_mm = mm;
1434         return 0;
1435 }
1436
1437 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1438 {
1439         struct fs_struct *fs = current->fs;
1440         if (clone_flags & CLONE_FS) {
1441                 /* tsk->fs is already what we want */
1442                 spin_lock(&fs->lock);
1443                 if (fs->in_exec) {
1444                         spin_unlock(&fs->lock);
1445                         return -EAGAIN;
1446                 }
1447                 fs->users++;
1448                 spin_unlock(&fs->lock);
1449                 return 0;
1450         }
1451         tsk->fs = copy_fs_struct(fs);
1452         if (!tsk->fs)
1453                 return -ENOMEM;
1454         return 0;
1455 }
1456
1457 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1458 {
1459         struct files_struct *oldf, *newf;
1460         int error = 0;
1461
1462         /*
1463          * A background process may not have any files ...
1464          */
1465         oldf = current->files;
1466         if (!oldf)
1467                 goto out;
1468
1469         if (clone_flags & CLONE_FILES) {
1470                 atomic_inc(&oldf->count);
1471                 goto out;
1472         }
1473
1474         newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1475         if (!newf)
1476                 goto out;
1477
1478         tsk->files = newf;
1479         error = 0;
1480 out:
1481         return error;
1482 }
1483
1484 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1485 {
1486 #ifdef CONFIG_BLOCK
1487         struct io_context *ioc = current->io_context;
1488         struct io_context *new_ioc;
1489
1490         if (!ioc)
1491                 return 0;
1492         /*
1493          * Share io context with parent, if CLONE_IO is set
1494          */
1495         if (clone_flags & CLONE_IO) {
1496                 ioc_task_link(ioc);
1497                 tsk->io_context = ioc;
1498         } else if (ioprio_valid(ioc->ioprio)) {
1499                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1500                 if (unlikely(!new_ioc))
1501                         return -ENOMEM;
1502
1503                 new_ioc->ioprio = ioc->ioprio;
1504                 put_io_context(new_ioc);
1505         }
1506 #endif
1507         return 0;
1508 }
1509
1510 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1511 {
1512         struct sighand_struct *sig;
1513
1514         if (clone_flags & CLONE_SIGHAND) {
1515                 refcount_inc(&current->sighand->count);
1516                 return 0;
1517         }
1518         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1519         RCU_INIT_POINTER(tsk->sighand, sig);
1520         if (!sig)
1521                 return -ENOMEM;
1522
1523         refcount_set(&sig->count, 1);
1524         spin_lock_irq(&current->sighand->siglock);
1525         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1526         spin_unlock_irq(&current->sighand->siglock);
1527
1528         /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1529         if (clone_flags & CLONE_CLEAR_SIGHAND)
1530                 flush_signal_handlers(tsk, 0);
1531
1532         return 0;
1533 }
1534
1535 void __cleanup_sighand(struct sighand_struct *sighand)
1536 {
1537         if (refcount_dec_and_test(&sighand->count)) {
1538                 signalfd_cleanup(sighand);
1539                 /*
1540                  * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1541                  * without an RCU grace period, see __lock_task_sighand().
1542                  */
1543                 kmem_cache_free(sighand_cachep, sighand);
1544         }
1545 }
1546
1547 /*
1548  * Initialize POSIX timer handling for a thread group.
1549  */
1550 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1551 {
1552         struct posix_cputimers *pct = &sig->posix_cputimers;
1553         unsigned long cpu_limit;
1554
1555         cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1556         posix_cputimers_group_init(pct, cpu_limit);
1557 }
1558
1559 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1560 {
1561         struct signal_struct *sig;
1562
1563         if (clone_flags & CLONE_THREAD)
1564                 return 0;
1565
1566         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1567         tsk->signal = sig;
1568         if (!sig)
1569                 return -ENOMEM;
1570
1571         sig->nr_threads = 1;
1572         atomic_set(&sig->live, 1);
1573         refcount_set(&sig->sigcnt, 1);
1574
1575         /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1576         sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1577         tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1578
1579         init_waitqueue_head(&sig->wait_chldexit);
1580         sig->curr_target = tsk;
1581         init_sigpending(&sig->shared_pending);
1582         INIT_HLIST_HEAD(&sig->multiprocess);
1583         seqlock_init(&sig->stats_lock);
1584         prev_cputime_init(&sig->prev_cputime);
1585
1586 #ifdef CONFIG_POSIX_TIMERS
1587         INIT_LIST_HEAD(&sig->posix_timers);
1588         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1589         sig->real_timer.function = it_real_fn;
1590 #endif
1591
1592         task_lock(current->group_leader);
1593         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1594         task_unlock(current->group_leader);
1595
1596         posix_cpu_timers_init_group(sig);
1597
1598         tty_audit_fork(sig);
1599         sched_autogroup_fork(sig);
1600
1601         sig->oom_score_adj = current->signal->oom_score_adj;
1602         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1603
1604         mutex_init(&sig->cred_guard_mutex);
1605         init_rwsem(&sig->exec_update_lock);
1606
1607         return 0;
1608 }
1609
1610 static void copy_seccomp(struct task_struct *p)
1611 {
1612 #ifdef CONFIG_SECCOMP
1613         /*
1614          * Must be called with sighand->lock held, which is common to
1615          * all threads in the group. Holding cred_guard_mutex is not
1616          * needed because this new task is not yet running and cannot
1617          * be racing exec.
1618          */
1619         assert_spin_locked(&current->sighand->siglock);
1620
1621         /* Ref-count the new filter user, and assign it. */
1622         get_seccomp_filter(current);
1623         p->seccomp = current->seccomp;
1624
1625         /*
1626          * Explicitly enable no_new_privs here in case it got set
1627          * between the task_struct being duplicated and holding the
1628          * sighand lock. The seccomp state and nnp must be in sync.
1629          */
1630         if (task_no_new_privs(current))
1631                 task_set_no_new_privs(p);
1632
1633         /*
1634          * If the parent gained a seccomp mode after copying thread
1635          * flags and between before we held the sighand lock, we have
1636          * to manually enable the seccomp thread flag here.
1637          */
1638         if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1639                 set_task_syscall_work(p, SECCOMP);
1640 #endif
1641 }
1642
1643 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1644 {
1645         current->clear_child_tid = tidptr;
1646
1647         return task_pid_vnr(current);
1648 }
1649
1650 static void rt_mutex_init_task(struct task_struct *p)
1651 {
1652         raw_spin_lock_init(&p->pi_lock);
1653 #ifdef CONFIG_RT_MUTEXES
1654         p->pi_waiters = RB_ROOT_CACHED;
1655         p->pi_top_task = NULL;
1656         p->pi_blocked_on = NULL;
1657 #endif
1658 }
1659
1660 static inline void init_task_pid_links(struct task_struct *task)
1661 {
1662         enum pid_type type;
1663
1664         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1665                 INIT_HLIST_NODE(&task->pid_links[type]);
1666 }
1667
1668 static inline void
1669 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1670 {
1671         if (type == PIDTYPE_PID)
1672                 task->thread_pid = pid;
1673         else
1674                 task->signal->pids[type] = pid;
1675 }
1676
1677 static inline void rcu_copy_process(struct task_struct *p)
1678 {
1679 #ifdef CONFIG_PREEMPT_RCU
1680         p->rcu_read_lock_nesting = 0;
1681         p->rcu_read_unlock_special.s = 0;
1682         p->rcu_blocked_node = NULL;
1683         INIT_LIST_HEAD(&p->rcu_node_entry);
1684 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1685 #ifdef CONFIG_TASKS_RCU
1686         p->rcu_tasks_holdout = false;
1687         INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1688         p->rcu_tasks_idle_cpu = -1;
1689 #endif /* #ifdef CONFIG_TASKS_RCU */
1690 #ifdef CONFIG_TASKS_TRACE_RCU
1691         p->trc_reader_nesting = 0;
1692         p->trc_reader_special.s = 0;
1693         INIT_LIST_HEAD(&p->trc_holdout_list);
1694 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1695 }
1696
1697 struct pid *pidfd_pid(const struct file *file)
1698 {
1699         if (file->f_op == &pidfd_fops)
1700                 return file->private_data;
1701
1702         return ERR_PTR(-EBADF);
1703 }
1704
1705 static int pidfd_release(struct inode *inode, struct file *file)
1706 {
1707         struct pid *pid = file->private_data;
1708
1709         file->private_data = NULL;
1710         put_pid(pid);
1711         return 0;
1712 }
1713
1714 #ifdef CONFIG_PROC_FS
1715 /**
1716  * pidfd_show_fdinfo - print information about a pidfd
1717  * @m: proc fdinfo file
1718  * @f: file referencing a pidfd
1719  *
1720  * Pid:
1721  * This function will print the pid that a given pidfd refers to in the
1722  * pid namespace of the procfs instance.
1723  * If the pid namespace of the process is not a descendant of the pid
1724  * namespace of the procfs instance 0 will be shown as its pid. This is
1725  * similar to calling getppid() on a process whose parent is outside of
1726  * its pid namespace.
1727  *
1728  * NSpid:
1729  * If pid namespaces are supported then this function will also print
1730  * the pid of a given pidfd refers to for all descendant pid namespaces
1731  * starting from the current pid namespace of the instance, i.e. the
1732  * Pid field and the first entry in the NSpid field will be identical.
1733  * If the pid namespace of the process is not a descendant of the pid
1734  * namespace of the procfs instance 0 will be shown as its first NSpid
1735  * entry and no others will be shown.
1736  * Note that this differs from the Pid and NSpid fields in
1737  * /proc/<pid>/status where Pid and NSpid are always shown relative to
1738  * the  pid namespace of the procfs instance. The difference becomes
1739  * obvious when sending around a pidfd between pid namespaces from a
1740  * different branch of the tree, i.e. where no ancestral relation is
1741  * present between the pid namespaces:
1742  * - create two new pid namespaces ns1 and ns2 in the initial pid
1743  *   namespace (also take care to create new mount namespaces in the
1744  *   new pid namespace and mount procfs)
1745  * - create a process with a pidfd in ns1
1746  * - send pidfd from ns1 to ns2
1747  * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1748  *   have exactly one entry, which is 0
1749  */
1750 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1751 {
1752         struct pid *pid = f->private_data;
1753         struct pid_namespace *ns;
1754         pid_t nr = -1;
1755
1756         if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1757                 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1758                 nr = pid_nr_ns(pid, ns);
1759         }
1760
1761         seq_put_decimal_ll(m, "Pid:\t", nr);
1762
1763 #ifdef CONFIG_PID_NS
1764         seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1765         if (nr > 0) {
1766                 int i;
1767
1768                 /* If nr is non-zero it means that 'pid' is valid and that
1769                  * ns, i.e. the pid namespace associated with the procfs
1770                  * instance, is in the pid namespace hierarchy of pid.
1771                  * Start at one below the already printed level.
1772                  */
1773                 for (i = ns->level + 1; i <= pid->level; i++)
1774                         seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1775         }
1776 #endif
1777         seq_putc(m, '\n');
1778 }
1779 #endif
1780
1781 /*
1782  * Poll support for process exit notification.
1783  */
1784 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1785 {
1786         struct pid *pid = file->private_data;
1787         __poll_t poll_flags = 0;
1788
1789         poll_wait(file, &pid->wait_pidfd, pts);
1790
1791         /*
1792          * Inform pollers only when the whole thread group exits.
1793          * If the thread group leader exits before all other threads in the
1794          * group, then poll(2) should block, similar to the wait(2) family.
1795          */
1796         if (thread_group_exited(pid))
1797                 poll_flags = EPOLLIN | EPOLLRDNORM;
1798
1799         return poll_flags;
1800 }
1801
1802 const struct file_operations pidfd_fops = {
1803         .release = pidfd_release,
1804         .poll = pidfd_poll,
1805 #ifdef CONFIG_PROC_FS
1806         .show_fdinfo = pidfd_show_fdinfo,
1807 #endif
1808 };
1809
1810 static void __delayed_free_task(struct rcu_head *rhp)
1811 {
1812         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1813
1814         free_task(tsk);
1815 }
1816
1817 static __always_inline void delayed_free_task(struct task_struct *tsk)
1818 {
1819         if (IS_ENABLED(CONFIG_MEMCG))
1820                 call_rcu(&tsk->rcu, __delayed_free_task);
1821         else
1822                 free_task(tsk);
1823 }
1824
1825 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1826 {
1827         /* Skip if kernel thread */
1828         if (!tsk->mm)
1829                 return;
1830
1831         /* Skip if spawning a thread or using vfork */
1832         if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1833                 return;
1834
1835         /* We need to synchronize with __set_oom_adj */
1836         mutex_lock(&oom_adj_mutex);
1837         set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1838         /* Update the values in case they were changed after copy_signal */
1839         tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1840         tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1841         mutex_unlock(&oom_adj_mutex);
1842 }
1843
1844 /*
1845  * This creates a new process as a copy of the old one,
1846  * but does not actually start it yet.
1847  *
1848  * It copies the registers, and all the appropriate
1849  * parts of the process environment (as per the clone
1850  * flags). The actual kick-off is left to the caller.
1851  */
1852 static __latent_entropy struct task_struct *copy_process(
1853                                         struct pid *pid,
1854                                         int trace,
1855                                         int node,
1856                                         struct kernel_clone_args *args)
1857 {
1858         int pidfd = -1, retval;
1859         struct task_struct *p;
1860         struct multiprocess_signals delayed;
1861         struct file *pidfile = NULL;
1862         u64 clone_flags = args->flags;
1863         struct nsproxy *nsp = current->nsproxy;
1864
1865         /*
1866          * Don't allow sharing the root directory with processes in a different
1867          * namespace
1868          */
1869         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1870                 return ERR_PTR(-EINVAL);
1871
1872         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1873                 return ERR_PTR(-EINVAL);
1874
1875         /*
1876          * Thread groups must share signals as well, and detached threads
1877          * can only be started up within the thread group.
1878          */
1879         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1880                 return ERR_PTR(-EINVAL);
1881
1882         /*
1883          * Shared signal handlers imply shared VM. By way of the above,
1884          * thread groups also imply shared VM. Blocking this case allows
1885          * for various simplifications in other code.
1886          */
1887         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1888                 return ERR_PTR(-EINVAL);
1889
1890         /*
1891          * Siblings of global init remain as zombies on exit since they are
1892          * not reaped by their parent (swapper). To solve this and to avoid
1893          * multi-rooted process trees, prevent global and container-inits
1894          * from creating siblings.
1895          */
1896         if ((clone_flags & CLONE_PARENT) &&
1897                                 current->signal->flags & SIGNAL_UNKILLABLE)
1898                 return ERR_PTR(-EINVAL);
1899
1900         /*
1901          * If the new process will be in a different pid or user namespace
1902          * do not allow it to share a thread group with the forking task.
1903          */
1904         if (clone_flags & CLONE_THREAD) {
1905                 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1906                     (task_active_pid_ns(current) != nsp->pid_ns_for_children))
1907                         return ERR_PTR(-EINVAL);
1908         }
1909
1910         /*
1911          * If the new process will be in a different time namespace
1912          * do not allow it to share VM or a thread group with the forking task.
1913          */
1914         if (clone_flags & (CLONE_THREAD | CLONE_VM)) {
1915                 if (nsp->time_ns != nsp->time_ns_for_children)
1916                         return ERR_PTR(-EINVAL);
1917         }
1918
1919         if (clone_flags & CLONE_PIDFD) {
1920                 /*
1921                  * - CLONE_DETACHED is blocked so that we can potentially
1922                  *   reuse it later for CLONE_PIDFD.
1923                  * - CLONE_THREAD is blocked until someone really needs it.
1924                  */
1925                 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1926                         return ERR_PTR(-EINVAL);
1927         }
1928
1929         /*
1930          * Force any signals received before this point to be delivered
1931          * before the fork happens.  Collect up signals sent to multiple
1932          * processes that happen during the fork and delay them so that
1933          * they appear to happen after the fork.
1934          */
1935         sigemptyset(&delayed.signal);
1936         INIT_HLIST_NODE(&delayed.node);
1937
1938         spin_lock_irq(&current->sighand->siglock);
1939         if (!(clone_flags & CLONE_THREAD))
1940                 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1941         recalc_sigpending();
1942         spin_unlock_irq(&current->sighand->siglock);
1943         retval = -ERESTARTNOINTR;
1944         if (task_sigpending(current))
1945                 goto fork_out;
1946
1947         retval = -ENOMEM;
1948         p = dup_task_struct(current, node);
1949         if (!p)
1950                 goto fork_out;
1951         if (args->io_thread) {
1952                 /*
1953                  * Mark us an IO worker, and block any signal that isn't
1954                  * fatal or STOP
1955                  */
1956                 p->flags |= PF_IO_WORKER;
1957                 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
1958         }
1959
1960         /*
1961          * This _must_ happen before we call free_task(), i.e. before we jump
1962          * to any of the bad_fork_* labels. This is to avoid freeing
1963          * p->set_child_tid which is (ab)used as a kthread's data pointer for
1964          * kernel threads (PF_KTHREAD).
1965          */
1966         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1967         /*
1968          * Clear TID on mm_release()?
1969          */
1970         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1971
1972         ftrace_graph_init_task(p);
1973
1974         rt_mutex_init_task(p);
1975
1976         lockdep_assert_irqs_enabled();
1977 #ifdef CONFIG_PROVE_LOCKING
1978         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1979 #endif
1980         retval = -EAGAIN;
1981         if (atomic_read(&p->real_cred->user->processes) >=
1982                         task_rlimit(p, RLIMIT_NPROC)) {
1983                 if (p->real_cred->user != INIT_USER &&
1984                     !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1985                         goto bad_fork_free;
1986         }
1987         current->flags &= ~PF_NPROC_EXCEEDED;
1988
1989         retval = copy_creds(p, clone_flags);
1990         if (retval < 0)
1991                 goto bad_fork_free;
1992
1993         /*
1994          * If multiple threads are within copy_process(), then this check
1995          * triggers too late. This doesn't hurt, the check is only there
1996          * to stop root fork bombs.
1997          */
1998         retval = -EAGAIN;
1999         if (data_race(nr_threads >= max_threads))
2000                 goto bad_fork_cleanup_count;
2001
2002         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
2003         p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2004         p->flags |= PF_FORKNOEXEC;
2005         INIT_LIST_HEAD(&p->children);
2006         INIT_LIST_HEAD(&p->sibling);
2007         rcu_copy_process(p);
2008         p->vfork_done = NULL;
2009         spin_lock_init(&p->alloc_lock);
2010
2011         init_sigpending(&p->pending);
2012         p->sigqueue_cache = NULL;
2013
2014         p->utime = p->stime = p->gtime = 0;
2015 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2016         p->utimescaled = p->stimescaled = 0;
2017 #endif
2018         prev_cputime_init(&p->prev_cputime);
2019
2020 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2021         seqcount_init(&p->vtime.seqcount);
2022         p->vtime.starttime = 0;
2023         p->vtime.state = VTIME_INACTIVE;
2024 #endif
2025
2026 #ifdef CONFIG_IO_URING
2027         p->io_uring = NULL;
2028 #endif
2029
2030 #if defined(SPLIT_RSS_COUNTING)
2031         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2032 #endif
2033
2034         p->default_timer_slack_ns = current->timer_slack_ns;
2035
2036 #ifdef CONFIG_PSI
2037         p->psi_flags = 0;
2038 #endif
2039
2040         task_io_accounting_init(&p->ioac);
2041         acct_clear_integrals(p);
2042
2043         posix_cputimers_init(&p->posix_cputimers);
2044
2045         p->io_context = NULL;
2046         audit_set_context(p, NULL);
2047         cgroup_fork(p);
2048 #ifdef CONFIG_NUMA
2049         p->mempolicy = mpol_dup(p->mempolicy);
2050         if (IS_ERR(p->mempolicy)) {
2051                 retval = PTR_ERR(p->mempolicy);
2052                 p->mempolicy = NULL;
2053                 goto bad_fork_cleanup_threadgroup_lock;
2054         }
2055 #endif
2056 #ifdef CONFIG_CPUSETS
2057         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2058         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2059         seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2060 #endif
2061 #ifdef CONFIG_TRACE_IRQFLAGS
2062         memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2063         p->irqtrace.hardirq_disable_ip  = _THIS_IP_;
2064         p->irqtrace.softirq_enable_ip   = _THIS_IP_;
2065         p->softirqs_enabled             = 1;
2066         p->softirq_context              = 0;
2067 #endif
2068
2069         p->pagefault_disabled = 0;
2070
2071 #ifdef CONFIG_LOCKDEP
2072         lockdep_init_task(p);
2073 #endif
2074
2075 #ifdef CONFIG_DEBUG_MUTEXES
2076         p->blocked_on = NULL; /* not blocked yet */
2077 #endif
2078 #ifdef CONFIG_BCACHE
2079         p->sequential_io        = 0;
2080         p->sequential_io_avg    = 0;
2081 #endif
2082 #ifdef CONFIG_BPF_SYSCALL
2083         RCU_INIT_POINTER(p->bpf_storage, NULL);
2084 #endif
2085
2086         /* Perform scheduler related setup. Assign this task to a CPU. */
2087         retval = sched_fork(clone_flags, p);
2088         if (retval)
2089                 goto bad_fork_cleanup_policy;
2090
2091         retval = perf_event_init_task(p, clone_flags);
2092         if (retval)
2093                 goto bad_fork_cleanup_policy;
2094         retval = audit_alloc(p);
2095         if (retval)
2096                 goto bad_fork_cleanup_perf;
2097         /* copy all the process information */
2098         shm_init_task(p);
2099         retval = security_task_alloc(p, clone_flags);
2100         if (retval)
2101                 goto bad_fork_cleanup_audit;
2102         retval = copy_semundo(clone_flags, p);
2103         if (retval)
2104                 goto bad_fork_cleanup_security;
2105         retval = copy_files(clone_flags, p);
2106         if (retval)
2107                 goto bad_fork_cleanup_semundo;
2108         retval = copy_fs(clone_flags, p);
2109         if (retval)
2110                 goto bad_fork_cleanup_files;
2111         retval = copy_sighand(clone_flags, p);
2112         if (retval)
2113                 goto bad_fork_cleanup_fs;
2114         retval = copy_signal(clone_flags, p);
2115         if (retval)
2116                 goto bad_fork_cleanup_sighand;
2117         retval = copy_mm(clone_flags, p);
2118         if (retval)
2119                 goto bad_fork_cleanup_signal;
2120         retval = copy_namespaces(clone_flags, p);
2121         if (retval)
2122                 goto bad_fork_cleanup_mm;
2123         retval = copy_io(clone_flags, p);
2124         if (retval)
2125                 goto bad_fork_cleanup_namespaces;
2126         retval = copy_thread(clone_flags, args->stack, args->stack_size, p, args->tls);
2127         if (retval)
2128                 goto bad_fork_cleanup_io;
2129
2130         stackleak_task_init(p);
2131
2132         if (pid != &init_struct_pid) {
2133                 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2134                                 args->set_tid_size);
2135                 if (IS_ERR(pid)) {
2136                         retval = PTR_ERR(pid);
2137                         goto bad_fork_cleanup_thread;
2138                 }
2139         }
2140
2141         /*
2142          * This has to happen after we've potentially unshared the file
2143          * descriptor table (so that the pidfd doesn't leak into the child
2144          * if the fd table isn't shared).
2145          */
2146         if (clone_flags & CLONE_PIDFD) {
2147                 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2148                 if (retval < 0)
2149                         goto bad_fork_free_pid;
2150
2151                 pidfd = retval;
2152
2153                 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2154                                               O_RDWR | O_CLOEXEC);
2155                 if (IS_ERR(pidfile)) {
2156                         put_unused_fd(pidfd);
2157                         retval = PTR_ERR(pidfile);
2158                         goto bad_fork_free_pid;
2159                 }
2160                 get_pid(pid);   /* held by pidfile now */
2161
2162                 retval = put_user(pidfd, args->pidfd);
2163                 if (retval)
2164                         goto bad_fork_put_pidfd;
2165         }
2166
2167 #ifdef CONFIG_BLOCK
2168         p->plug = NULL;
2169 #endif
2170         futex_init_task(p);
2171
2172         /*
2173          * sigaltstack should be cleared when sharing the same VM
2174          */
2175         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2176                 sas_ss_reset(p);
2177
2178         /*
2179          * Syscall tracing and stepping should be turned off in the
2180          * child regardless of CLONE_PTRACE.
2181          */
2182         user_disable_single_step(p);
2183         clear_task_syscall_work(p, SYSCALL_TRACE);
2184 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2185         clear_task_syscall_work(p, SYSCALL_EMU);
2186 #endif
2187         clear_tsk_latency_tracing(p);
2188
2189         /* ok, now we should be set up.. */
2190         p->pid = pid_nr(pid);
2191         if (clone_flags & CLONE_THREAD) {
2192                 p->group_leader = current->group_leader;
2193                 p->tgid = current->tgid;
2194         } else {
2195                 p->group_leader = p;
2196                 p->tgid = p->pid;
2197         }
2198
2199         p->nr_dirtied = 0;
2200         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2201         p->dirty_paused_when = 0;
2202
2203         p->pdeath_signal = 0;
2204         INIT_LIST_HEAD(&p->thread_group);
2205         p->task_works = NULL;
2206
2207 #ifdef CONFIG_KRETPROBES
2208         p->kretprobe_instances.first = NULL;
2209 #endif
2210
2211         /*
2212          * Ensure that the cgroup subsystem policies allow the new process to be
2213          * forked. It should be noted that the new process's css_set can be changed
2214          * between here and cgroup_post_fork() if an organisation operation is in
2215          * progress.
2216          */
2217         retval = cgroup_can_fork(p, args);
2218         if (retval)
2219                 goto bad_fork_put_pidfd;
2220
2221         /*
2222          * From this point on we must avoid any synchronous user-space
2223          * communication until we take the tasklist-lock. In particular, we do
2224          * not want user-space to be able to predict the process start-time by
2225          * stalling fork(2) after we recorded the start_time but before it is
2226          * visible to the system.
2227          */
2228
2229         p->start_time = ktime_get_ns();
2230         p->start_boottime = ktime_get_boottime_ns();
2231
2232         /*
2233          * Make it visible to the rest of the system, but dont wake it up yet.
2234          * Need tasklist lock for parent etc handling!
2235          */
2236         write_lock_irq(&tasklist_lock);
2237
2238         /* CLONE_PARENT re-uses the old parent */
2239         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2240                 p->real_parent = current->real_parent;
2241                 p->parent_exec_id = current->parent_exec_id;
2242                 if (clone_flags & CLONE_THREAD)
2243                         p->exit_signal = -1;
2244                 else
2245                         p->exit_signal = current->group_leader->exit_signal;
2246         } else {
2247                 p->real_parent = current;
2248                 p->parent_exec_id = current->self_exec_id;
2249                 p->exit_signal = args->exit_signal;
2250         }
2251
2252         klp_copy_process(p);
2253
2254         sched_core_fork(p);
2255
2256         spin_lock(&current->sighand->siglock);
2257
2258         /*
2259          * Copy seccomp details explicitly here, in case they were changed
2260          * before holding sighand lock.
2261          */
2262         copy_seccomp(p);
2263
2264         rseq_fork(p, clone_flags);
2265
2266         /* Don't start children in a dying pid namespace */
2267         if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2268                 retval = -ENOMEM;
2269                 goto bad_fork_cancel_cgroup;
2270         }
2271
2272         /* Let kill terminate clone/fork in the middle */
2273         if (fatal_signal_pending(current)) {
2274                 retval = -EINTR;
2275                 goto bad_fork_cancel_cgroup;
2276         }
2277
2278         /* past the last point of failure */
2279         if (pidfile)
2280                 fd_install(pidfd, pidfile);
2281
2282         init_task_pid_links(p);
2283         if (likely(p->pid)) {
2284                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2285
2286                 init_task_pid(p, PIDTYPE_PID, pid);
2287                 if (thread_group_leader(p)) {
2288                         init_task_pid(p, PIDTYPE_TGID, pid);
2289                         init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2290                         init_task_pid(p, PIDTYPE_SID, task_session(current));
2291
2292                         if (is_child_reaper(pid)) {
2293                                 ns_of_pid(pid)->child_reaper = p;
2294                                 p->signal->flags |= SIGNAL_UNKILLABLE;
2295                         }
2296                         p->signal->shared_pending.signal = delayed.signal;
2297                         p->signal->tty = tty_kref_get(current->signal->tty);
2298                         /*
2299                          * Inherit has_child_subreaper flag under the same
2300                          * tasklist_lock with adding child to the process tree
2301                          * for propagate_has_child_subreaper optimization.
2302                          */
2303                         p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2304                                                          p->real_parent->signal->is_child_subreaper;
2305                         list_add_tail(&p->sibling, &p->real_parent->children);
2306                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
2307                         attach_pid(p, PIDTYPE_TGID);
2308                         attach_pid(p, PIDTYPE_PGID);
2309                         attach_pid(p, PIDTYPE_SID);
2310                         __this_cpu_inc(process_counts);
2311                 } else {
2312                         current->signal->nr_threads++;
2313                         atomic_inc(&current->signal->live);
2314                         refcount_inc(&current->signal->sigcnt);
2315                         task_join_group_stop(p);
2316                         list_add_tail_rcu(&p->thread_group,
2317                                           &p->group_leader->thread_group);
2318                         list_add_tail_rcu(&p->thread_node,
2319                                           &p->signal->thread_head);
2320                 }
2321                 attach_pid(p, PIDTYPE_PID);
2322                 nr_threads++;
2323         }
2324         total_forks++;
2325         hlist_del_init(&delayed.node);
2326         spin_unlock(&current->sighand->siglock);
2327         syscall_tracepoint_update(p);
2328         write_unlock_irq(&tasklist_lock);
2329
2330         proc_fork_connector(p);
2331         sched_post_fork(p);
2332         cgroup_post_fork(p, args);
2333         perf_event_fork(p);
2334
2335         trace_task_newtask(p, clone_flags);
2336         uprobe_copy_process(p, clone_flags);
2337
2338         copy_oom_score_adj(clone_flags, p);
2339
2340         return p;
2341
2342 bad_fork_cancel_cgroup:
2343         sched_core_free(p);
2344         spin_unlock(&current->sighand->siglock);
2345         write_unlock_irq(&tasklist_lock);
2346         cgroup_cancel_fork(p, args);
2347 bad_fork_put_pidfd:
2348         if (clone_flags & CLONE_PIDFD) {
2349                 fput(pidfile);
2350                 put_unused_fd(pidfd);
2351         }
2352 bad_fork_free_pid:
2353         if (pid != &init_struct_pid)
2354                 free_pid(pid);
2355 bad_fork_cleanup_thread:
2356         exit_thread(p);
2357 bad_fork_cleanup_io:
2358         if (p->io_context)
2359                 exit_io_context(p);
2360 bad_fork_cleanup_namespaces:
2361         exit_task_namespaces(p);
2362 bad_fork_cleanup_mm:
2363         if (p->mm) {
2364                 mm_clear_owner(p->mm, p);
2365                 mmput(p->mm);
2366         }
2367 bad_fork_cleanup_signal:
2368         if (!(clone_flags & CLONE_THREAD))
2369                 free_signal_struct(p->signal);
2370 bad_fork_cleanup_sighand:
2371         __cleanup_sighand(p->sighand);
2372 bad_fork_cleanup_fs:
2373         exit_fs(p); /* blocking */
2374 bad_fork_cleanup_files:
2375         exit_files(p); /* blocking */
2376 bad_fork_cleanup_semundo:
2377         exit_sem(p);
2378 bad_fork_cleanup_security:
2379         security_task_free(p);
2380 bad_fork_cleanup_audit:
2381         audit_free(p);
2382 bad_fork_cleanup_perf:
2383         perf_event_free_task(p);
2384 bad_fork_cleanup_policy:
2385         lockdep_free_task(p);
2386 #ifdef CONFIG_NUMA
2387         mpol_put(p->mempolicy);
2388 bad_fork_cleanup_threadgroup_lock:
2389 #endif
2390         delayacct_tsk_free(p);
2391 bad_fork_cleanup_count:
2392         atomic_dec(&p->cred->user->processes);
2393         exit_creds(p);
2394 bad_fork_free:
2395         p->state = TASK_DEAD;
2396         put_task_stack(p);
2397         delayed_free_task(p);
2398 fork_out:
2399         spin_lock_irq(&current->sighand->siglock);
2400         hlist_del_init(&delayed.node);
2401         spin_unlock_irq(&current->sighand->siglock);
2402         return ERR_PTR(retval);
2403 }
2404
2405 static inline void init_idle_pids(struct task_struct *idle)
2406 {
2407         enum pid_type type;
2408
2409         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2410                 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2411                 init_task_pid(idle, type, &init_struct_pid);
2412         }
2413 }
2414
2415 struct task_struct * __init fork_idle(int cpu)
2416 {
2417         struct task_struct *task;
2418         struct kernel_clone_args args = {
2419                 .flags = CLONE_VM,
2420         };
2421
2422         task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2423         if (!IS_ERR(task)) {
2424                 init_idle_pids(task);
2425                 init_idle(task, cpu);
2426         }
2427
2428         return task;
2429 }
2430
2431 struct mm_struct *copy_init_mm(void)
2432 {
2433         return dup_mm(NULL, &init_mm);
2434 }
2435
2436 /*
2437  * This is like kernel_clone(), but shaved down and tailored to just
2438  * creating io_uring workers. It returns a created task, or an error pointer.
2439  * The returned task is inactive, and the caller must fire it up through
2440  * wake_up_new_task(p). All signals are blocked in the created task.
2441  */
2442 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2443 {
2444         unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2445                                 CLONE_IO;
2446         struct kernel_clone_args args = {
2447                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2448                                     CLONE_UNTRACED) & ~CSIGNAL),
2449                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2450                 .stack          = (unsigned long)fn,
2451                 .stack_size     = (unsigned long)arg,
2452                 .io_thread      = 1,
2453         };
2454
2455         return copy_process(NULL, 0, node, &args);
2456 }
2457
2458 /*
2459  *  Ok, this is the main fork-routine.
2460  *
2461  * It copies the process, and if successful kick-starts
2462  * it and waits for it to finish using the VM if required.
2463  *
2464  * args->exit_signal is expected to be checked for sanity by the caller.
2465  */
2466 pid_t kernel_clone(struct kernel_clone_args *args)
2467 {
2468         u64 clone_flags = args->flags;
2469         struct completion vfork;
2470         struct pid *pid;
2471         struct task_struct *p;
2472         int trace = 0;
2473         pid_t nr;
2474
2475         /*
2476          * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2477          * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2478          * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2479          * field in struct clone_args and it still doesn't make sense to have
2480          * them both point at the same memory location. Performing this check
2481          * here has the advantage that we don't need to have a separate helper
2482          * to check for legacy clone().
2483          */
2484         if ((args->flags & CLONE_PIDFD) &&
2485             (args->flags & CLONE_PARENT_SETTID) &&
2486             (args->pidfd == args->parent_tid))
2487                 return -EINVAL;
2488
2489         /*
2490          * Determine whether and which event to report to ptracer.  When
2491          * called from kernel_thread or CLONE_UNTRACED is explicitly
2492          * requested, no event is reported; otherwise, report if the event
2493          * for the type of forking is enabled.
2494          */
2495         if (!(clone_flags & CLONE_UNTRACED)) {
2496                 if (clone_flags & CLONE_VFORK)
2497                         trace = PTRACE_EVENT_VFORK;
2498                 else if (args->exit_signal != SIGCHLD)
2499                         trace = PTRACE_EVENT_CLONE;
2500                 else
2501                         trace = PTRACE_EVENT_FORK;
2502
2503                 if (likely(!ptrace_event_enabled(current, trace)))
2504                         trace = 0;
2505         }
2506
2507         p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2508         add_latent_entropy();
2509
2510         if (IS_ERR(p))
2511                 return PTR_ERR(p);
2512
2513         /*
2514          * Do this prior waking up the new thread - the thread pointer
2515          * might get invalid after that point, if the thread exits quickly.
2516          */
2517         trace_sched_process_fork(current, p);
2518
2519         pid = get_task_pid(p, PIDTYPE_PID);
2520         nr = pid_vnr(pid);
2521
2522         if (clone_flags & CLONE_PARENT_SETTID)
2523                 put_user(nr, args->parent_tid);
2524
2525         if (clone_flags & CLONE_VFORK) {
2526                 p->vfork_done = &vfork;
2527                 init_completion(&vfork);
2528                 get_task_struct(p);
2529         }
2530
2531         wake_up_new_task(p);
2532
2533         /* forking complete and child started to run, tell ptracer */
2534         if (unlikely(trace))
2535                 ptrace_event_pid(trace, pid);
2536
2537         if (clone_flags & CLONE_VFORK) {
2538                 if (!wait_for_vfork_done(p, &vfork))
2539                         ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2540         }
2541
2542         put_pid(pid);
2543         return nr;
2544 }
2545
2546 /*
2547  * Create a kernel thread.
2548  */
2549 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2550 {
2551         struct kernel_clone_args args = {
2552                 .flags          = ((lower_32_bits(flags) | CLONE_VM |
2553                                     CLONE_UNTRACED) & ~CSIGNAL),
2554                 .exit_signal    = (lower_32_bits(flags) & CSIGNAL),
2555                 .stack          = (unsigned long)fn,
2556                 .stack_size     = (unsigned long)arg,
2557         };
2558
2559         return kernel_clone(&args);
2560 }
2561
2562 #ifdef __ARCH_WANT_SYS_FORK
2563 SYSCALL_DEFINE0(fork)
2564 {
2565 #ifdef CONFIG_MMU
2566         struct kernel_clone_args args = {
2567                 .exit_signal = SIGCHLD,
2568         };
2569
2570         return kernel_clone(&args);
2571 #else
2572         /* can not support in nommu mode */
2573         return -EINVAL;
2574 #endif
2575 }
2576 #endif
2577
2578 #ifdef __ARCH_WANT_SYS_VFORK
2579 SYSCALL_DEFINE0(vfork)
2580 {
2581         struct kernel_clone_args args = {
2582                 .flags          = CLONE_VFORK | CLONE_VM,
2583                 .exit_signal    = SIGCHLD,
2584         };
2585
2586         return kernel_clone(&args);
2587 }
2588 #endif
2589
2590 #ifdef __ARCH_WANT_SYS_CLONE
2591 #ifdef CONFIG_CLONE_BACKWARDS
2592 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2593                  int __user *, parent_tidptr,
2594                  unsigned long, tls,
2595                  int __user *, child_tidptr)
2596 #elif defined(CONFIG_CLONE_BACKWARDS2)
2597 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2598                  int __user *, parent_tidptr,
2599                  int __user *, child_tidptr,
2600                  unsigned long, tls)
2601 #elif defined(CONFIG_CLONE_BACKWARDS3)
2602 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2603                 int, stack_size,
2604                 int __user *, parent_tidptr,
2605                 int __user *, child_tidptr,
2606                 unsigned long, tls)
2607 #else
2608 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2609                  int __user *, parent_tidptr,
2610                  int __user *, child_tidptr,
2611                  unsigned long, tls)
2612 #endif
2613 {
2614         struct kernel_clone_args args = {
2615                 .flags          = (lower_32_bits(clone_flags) & ~CSIGNAL),
2616                 .pidfd          = parent_tidptr,
2617                 .child_tid      = child_tidptr,
2618                 .parent_tid     = parent_tidptr,
2619                 .exit_signal    = (lower_32_bits(clone_flags) & CSIGNAL),
2620                 .stack          = newsp,
2621                 .tls            = tls,
2622         };
2623
2624         return kernel_clone(&args);
2625 }
2626 #endif
2627
2628 #ifdef __ARCH_WANT_SYS_CLONE3
2629
2630 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2631                                               struct clone_args __user *uargs,
2632                                               size_t usize)
2633 {
2634         int err;
2635         struct clone_args args;
2636         pid_t *kset_tid = kargs->set_tid;
2637
2638         BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2639                      CLONE_ARGS_SIZE_VER0);
2640         BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2641                      CLONE_ARGS_SIZE_VER1);
2642         BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2643                      CLONE_ARGS_SIZE_VER2);
2644         BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2645
2646         if (unlikely(usize > PAGE_SIZE))
2647                 return -E2BIG;
2648         if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2649                 return -EINVAL;
2650
2651         err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2652         if (err)
2653                 return err;
2654
2655         if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2656                 return -EINVAL;
2657
2658         if (unlikely(!args.set_tid && args.set_tid_size > 0))
2659                 return -EINVAL;
2660
2661         if (unlikely(args.set_tid && args.set_tid_size == 0))
2662                 return -EINVAL;
2663
2664         /*
2665          * Verify that higher 32bits of exit_signal are unset and that
2666          * it is a valid signal
2667          */
2668         if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2669                      !valid_signal(args.exit_signal)))
2670                 return -EINVAL;
2671
2672         if ((args.flags & CLONE_INTO_CGROUP) &&
2673             (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2674                 return -EINVAL;
2675
2676         *kargs = (struct kernel_clone_args){
2677                 .flags          = args.flags,
2678                 .pidfd          = u64_to_user_ptr(args.pidfd),
2679                 .child_tid      = u64_to_user_ptr(args.child_tid),
2680                 .parent_tid     = u64_to_user_ptr(args.parent_tid),
2681                 .exit_signal    = args.exit_signal,
2682                 .stack          = args.stack,
2683                 .stack_size     = args.stack_size,
2684                 .tls            = args.tls,
2685                 .set_tid_size   = args.set_tid_size,
2686                 .cgroup         = args.cgroup,
2687         };
2688
2689         if (args.set_tid &&
2690                 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2691                         (kargs->set_tid_size * sizeof(pid_t))))
2692                 return -EFAULT;
2693
2694         kargs->set_tid = kset_tid;
2695
2696         return 0;
2697 }
2698
2699 /**
2700  * clone3_stack_valid - check and prepare stack
2701  * @kargs: kernel clone args
2702  *
2703  * Verify that the stack arguments userspace gave us are sane.
2704  * In addition, set the stack direction for userspace since it's easy for us to
2705  * determine.
2706  */
2707 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2708 {
2709         if (kargs->stack == 0) {
2710                 if (kargs->stack_size > 0)
2711                         return false;
2712         } else {
2713                 if (kargs->stack_size == 0)
2714                         return false;
2715
2716                 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2717                         return false;
2718
2719 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2720                 kargs->stack += kargs->stack_size;
2721 #endif
2722         }
2723
2724         return true;
2725 }
2726
2727 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2728 {
2729         /* Verify that no unknown flags are passed along. */
2730         if (kargs->flags &
2731             ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2732                 return false;
2733
2734         /*
2735          * - make the CLONE_DETACHED bit reusable for clone3
2736          * - make the CSIGNAL bits reusable for clone3
2737          */
2738         if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2739                 return false;
2740
2741         if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2742             (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2743                 return false;
2744
2745         if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2746             kargs->exit_signal)
2747                 return false;
2748
2749         if (!clone3_stack_valid(kargs))
2750                 return false;
2751
2752         return true;
2753 }
2754
2755 /**
2756  * clone3 - create a new process with specific properties
2757  * @uargs: argument structure
2758  * @size:  size of @uargs
2759  *
2760  * clone3() is the extensible successor to clone()/clone2().
2761  * It takes a struct as argument that is versioned by its size.
2762  *
2763  * Return: On success, a positive PID for the child process.
2764  *         On error, a negative errno number.
2765  */
2766 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2767 {
2768         int err;
2769
2770         struct kernel_clone_args kargs;
2771         pid_t set_tid[MAX_PID_NS_LEVEL];
2772
2773         kargs.set_tid = set_tid;
2774
2775         err = copy_clone_args_from_user(&kargs, uargs, size);
2776         if (err)
2777                 return err;
2778
2779         if (!clone3_args_valid(&kargs))
2780                 return -EINVAL;
2781
2782         return kernel_clone(&kargs);
2783 }
2784 #endif
2785
2786 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2787 {
2788         struct task_struct *leader, *parent, *child;
2789         int res;
2790
2791         read_lock(&tasklist_lock);
2792         leader = top = top->group_leader;
2793 down:
2794         for_each_thread(leader, parent) {
2795                 list_for_each_entry(child, &parent->children, sibling) {
2796                         res = visitor(child, data);
2797                         if (res) {
2798                                 if (res < 0)
2799                                         goto out;
2800                                 leader = child;
2801                                 goto down;
2802                         }
2803 up:
2804                         ;
2805                 }
2806         }
2807
2808         if (leader != top) {
2809                 child = leader;
2810                 parent = child->real_parent;
2811                 leader = parent->group_leader;
2812                 goto up;
2813         }
2814 out:
2815         read_unlock(&tasklist_lock);
2816 }
2817
2818 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2819 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2820 #endif
2821
2822 static void sighand_ctor(void *data)
2823 {
2824         struct sighand_struct *sighand = data;
2825
2826         spin_lock_init(&sighand->siglock);
2827         init_waitqueue_head(&sighand->signalfd_wqh);
2828 }
2829
2830 void __init proc_caches_init(void)
2831 {
2832         unsigned int mm_size;
2833
2834         sighand_cachep = kmem_cache_create("sighand_cache",
2835                         sizeof(struct sighand_struct), 0,
2836                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2837                         SLAB_ACCOUNT, sighand_ctor);
2838         signal_cachep = kmem_cache_create("signal_cache",
2839                         sizeof(struct signal_struct), 0,
2840                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2841                         NULL);
2842         files_cachep = kmem_cache_create("files_cache",
2843                         sizeof(struct files_struct), 0,
2844                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2845                         NULL);
2846         fs_cachep = kmem_cache_create("fs_cache",
2847                         sizeof(struct fs_struct), 0,
2848                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2849                         NULL);
2850
2851         /*
2852          * The mm_cpumask is located at the end of mm_struct, and is
2853          * dynamically sized based on the maximum CPU number this system
2854          * can have, taking hotplug into account (nr_cpu_ids).
2855          */
2856         mm_size = sizeof(struct mm_struct) + cpumask_size();
2857
2858         mm_cachep = kmem_cache_create_usercopy("mm_struct",
2859                         mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2860                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2861                         offsetof(struct mm_struct, saved_auxv),
2862                         sizeof_field(struct mm_struct, saved_auxv),
2863                         NULL);
2864         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2865         mmap_init();
2866         nsproxy_cache_init();
2867 }
2868
2869 /*
2870  * Check constraints on flags passed to the unshare system call.
2871  */
2872 static int check_unshare_flags(unsigned long unshare_flags)
2873 {
2874         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2875                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2876                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2877                                 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
2878                                 CLONE_NEWTIME))
2879                 return -EINVAL;
2880         /*
2881          * Not implemented, but pretend it works if there is nothing
2882          * to unshare.  Note that unsharing the address space or the
2883          * signal handlers also need to unshare the signal queues (aka
2884          * CLONE_THREAD).
2885          */
2886         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2887                 if (!thread_group_empty(current))
2888                         return -EINVAL;
2889         }
2890         if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2891                 if (refcount_read(&current->sighand->count) > 1)
2892                         return -EINVAL;
2893         }
2894         if (unshare_flags & CLONE_VM) {
2895                 if (!current_is_single_threaded())
2896                         return -EINVAL;
2897         }
2898
2899         return 0;
2900 }
2901
2902 /*
2903  * Unshare the filesystem structure if it is being shared
2904  */
2905 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2906 {
2907         struct fs_struct *fs = current->fs;
2908
2909         if (!(unshare_flags & CLONE_FS) || !fs)
2910                 return 0;
2911
2912         /* don't need lock here; in the worst case we'll do useless copy */
2913         if (fs->users == 1)
2914                 return 0;
2915
2916         *new_fsp = copy_fs_struct(fs);
2917         if (!*new_fsp)
2918                 return -ENOMEM;
2919
2920         return 0;
2921 }
2922
2923 /*
2924  * Unshare file descriptor table if it is being shared
2925  */
2926 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
2927                struct files_struct **new_fdp)
2928 {
2929         struct files_struct *fd = current->files;
2930         int error = 0;
2931
2932         if ((unshare_flags & CLONE_FILES) &&
2933             (fd && atomic_read(&fd->count) > 1)) {
2934                 *new_fdp = dup_fd(fd, max_fds, &error);
2935                 if (!*new_fdp)
2936                         return error;
2937         }
2938
2939         return 0;
2940 }
2941
2942 /*
2943  * unshare allows a process to 'unshare' part of the process
2944  * context which was originally shared using clone.  copy_*
2945  * functions used by kernel_clone() cannot be used here directly
2946  * because they modify an inactive task_struct that is being
2947  * constructed. Here we are modifying the current, active,
2948  * task_struct.
2949  */
2950 int ksys_unshare(unsigned long unshare_flags)
2951 {
2952         struct fs_struct *fs, *new_fs = NULL;
2953         struct files_struct *fd, *new_fd = NULL;
2954         struct cred *new_cred = NULL;
2955         struct nsproxy *new_nsproxy = NULL;
2956         int do_sysvsem = 0;
2957         int err;
2958
2959         /*
2960          * If unsharing a user namespace must also unshare the thread group
2961          * and unshare the filesystem root and working directories.
2962          */
2963         if (unshare_flags & CLONE_NEWUSER)
2964                 unshare_flags |= CLONE_THREAD | CLONE_FS;
2965         /*
2966          * If unsharing vm, must also unshare signal handlers.
2967          */
2968         if (unshare_flags & CLONE_VM)
2969                 unshare_flags |= CLONE_SIGHAND;
2970         /*
2971          * If unsharing a signal handlers, must also unshare the signal queues.
2972          */
2973         if (unshare_flags & CLONE_SIGHAND)
2974                 unshare_flags |= CLONE_THREAD;
2975         /*
2976          * If unsharing namespace, must also unshare filesystem information.
2977          */
2978         if (unshare_flags & CLONE_NEWNS)
2979                 unshare_flags |= CLONE_FS;
2980
2981         err = check_unshare_flags(unshare_flags);
2982         if (err)
2983                 goto bad_unshare_out;
2984         /*
2985          * CLONE_NEWIPC must also detach from the undolist: after switching
2986          * to a new ipc namespace, the semaphore arrays from the old
2987          * namespace are unreachable.
2988          */
2989         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2990                 do_sysvsem = 1;
2991         err = unshare_fs(unshare_flags, &new_fs);
2992         if (err)
2993                 goto bad_unshare_out;
2994         err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
2995         if (err)
2996                 goto bad_unshare_cleanup_fs;
2997         err = unshare_userns(unshare_flags, &new_cred);
2998         if (err)
2999                 goto bad_unshare_cleanup_fd;
3000         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3001                                          new_cred, new_fs);
3002         if (err)
3003                 goto bad_unshare_cleanup_cred;
3004
3005         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3006                 if (do_sysvsem) {
3007                         /*
3008                          * CLONE_SYSVSEM is equivalent to sys_exit().
3009                          */
3010                         exit_sem(current);
3011                 }
3012                 if (unshare_flags & CLONE_NEWIPC) {
3013                         /* Orphan segments in old ns (see sem above). */
3014                         exit_shm(current);
3015                         shm_init_task(current);
3016                 }
3017
3018                 if (new_nsproxy)
3019                         switch_task_namespaces(current, new_nsproxy);
3020
3021                 task_lock(current);
3022
3023                 if (new_fs) {
3024                         fs = current->fs;
3025                         spin_lock(&fs->lock);
3026                         current->fs = new_fs;
3027                         if (--fs->users)
3028                                 new_fs = NULL;
3029                         else
3030                                 new_fs = fs;
3031                         spin_unlock(&fs->lock);
3032                 }
3033
3034                 if (new_fd) {
3035                         fd = current->files;
3036                         current->files = new_fd;
3037                         new_fd = fd;
3038                 }
3039
3040                 task_unlock(current);
3041
3042                 if (new_cred) {
3043                         /* Install the new user namespace */
3044                         commit_creds(new_cred);
3045                         new_cred = NULL;
3046                 }
3047         }
3048
3049         perf_event_namespaces(current);
3050
3051 bad_unshare_cleanup_cred:
3052         if (new_cred)
3053                 put_cred(new_cred);
3054 bad_unshare_cleanup_fd:
3055         if (new_fd)
3056                 put_files_struct(new_fd);
3057
3058 bad_unshare_cleanup_fs:
3059         if (new_fs)
3060                 free_fs_struct(new_fs);
3061
3062 bad_unshare_out:
3063         return err;
3064 }
3065
3066 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3067 {
3068         return ksys_unshare(unshare_flags);
3069 }
3070
3071 /*
3072  *      Helper to unshare the files of the current task.
3073  *      We don't want to expose copy_files internals to
3074  *      the exec layer of the kernel.
3075  */
3076
3077 int unshare_files(void)
3078 {
3079         struct task_struct *task = current;
3080         struct files_struct *old, *copy = NULL;
3081         int error;
3082
3083         error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, &copy);
3084         if (error || !copy)
3085                 return error;
3086
3087         old = task->files;
3088         task_lock(task);
3089         task->files = copy;
3090         task_unlock(task);
3091         put_files_struct(old);
3092         return 0;
3093 }
3094
3095 int sysctl_max_threads(struct ctl_table *table, int write,
3096                        void *buffer, size_t *lenp, loff_t *ppos)
3097 {
3098         struct ctl_table t;
3099         int ret;
3100         int threads = max_threads;
3101         int min = 1;
3102         int max = MAX_THREADS;
3103
3104         t = *table;
3105         t.data = &threads;
3106         t.extra1 = &min;
3107         t.extra2 = &max;
3108
3109         ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3110         if (ret || !write)
3111                 return ret;
3112
3113         max_threads = threads;
3114
3115         return 0;
3116 }