fs/proc/task_mmu.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/pagewalk.h>
   3 #include <linux/vmacache.h>
   4 #include <linux/hugetlb.h>
   5 #include <linux/huge_mm.h>
   6 #include <linux/mount.h>
   7 #include <linux/seq_file.h>
   8 #include <linux/highmem.h>
   9 #include <linux/ptrace.h>
  10 #include <linux/slab.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/mempolicy.h>
  13 #include <linux/rmap.h>
  14 #include <linux/swap.h>
  15 #include <linux/sched/mm.h>
  16 #include <linux/swapops.h>
  17 #include <linux/mmu_notifier.h>
  18 #include <linux/page_idle.h>
  19 #include <linux/shmem_fs.h>
  20 #include <linux/uaccess.h>
  21 #include <linux/pkeys.h>
  22
  23 #include <asm/elf.h>
  24 #include <asm/tlb.h>
  25 #include <asm/tlbflush.h>
  26 #include "internal.h"
  27
  28 #define SEQ_PUT_DEC(str, val) \
  29                 seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8)
  30 void task_mem(struct seq_file *m, struct mm_struct *mm)
  31 {
  32         unsigned long text, lib, swap, anon, file, shmem;
  33         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  34
  35         anon = get_mm_counter(mm, MM_ANONPAGES);
  36         file = get_mm_counter(mm, MM_FILEPAGES);
  37         shmem = get_mm_counter(mm, MM_SHMEMPAGES);
  38
  39         /*
  40          * Note: to minimize their overhead, mm maintains hiwater_vm and
  41          * hiwater_rss only when about to *lower* total_vm or rss.  Any
  42          * collector of these hiwater stats must therefore get total_vm
  43          * and rss too, which will usually be the higher.  Barriers? not
  44          * worth the effort, such snapshots can always be inconsistent.
  45          */
  46         hiwater_vm = total_vm = mm->total_vm;
  47         if (hiwater_vm < mm->hiwater_vm)
  48                 hiwater_vm = mm->hiwater_vm;
  49         hiwater_rss = total_rss = anon + file + shmem;
  50         if (hiwater_rss < mm->hiwater_rss)
  51                 hiwater_rss = mm->hiwater_rss;
  52
  53         /* split executable areas between text and lib */
  54         text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK);
  55         text = min(text, mm->exec_vm << PAGE_SHIFT);
  56         lib = (mm->exec_vm << PAGE_SHIFT) - text;
  57
  58         swap = get_mm_counter(mm, MM_SWAPENTS);
  59         SEQ_PUT_DEC("VmPeak:\t", hiwater_vm);
  60         SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm);
  61         SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm);
  62         SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm->pinned_vm));
  63         SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss);
  64         SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss);
  65         SEQ_PUT_DEC(" kB\nRssAnon:\t", anon);
  66         SEQ_PUT_DEC(" kB\nRssFile:\t", file);
  67         SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem);
  68         SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm);
  69         SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm);
  70         seq_put_decimal_ull_width(m,
  71                     " kB\nVmExe:\t", text >> 10, 8);
  72         seq_put_decimal_ull_width(m,
  73                     " kB\nVmLib:\t", lib >> 10, 8);
  74         seq_put_decimal_ull_width(m,
  75                     " kB\nVmPTE:\t", mm_pgtables_bytes(mm) >> 10, 8);
  76         SEQ_PUT_DEC(" kB\nVmSwap:\t", swap);
  77         seq_puts(m, " kB\n");
  78         hugetlb_report_usage(m, mm);
  79 }
  80 #undef SEQ_PUT_DEC
  81
  82 unsigned long task_vsize(struct mm_struct *mm)
  83 {
  84         return PAGE_SIZE * mm->total_vm;
  85 }
  86
  87 unsigned long task_statm(struct mm_struct *mm,
  88                          unsigned long *shared, unsigned long *text,
  89                          unsigned long *data, unsigned long *resident)
  90 {
  91         *shared = get_mm_counter(mm, MM_FILEPAGES) +
  92                         get_mm_counter(mm, MM_SHMEMPAGES);
  93         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
  94                                                                 >> PAGE_SHIFT;
  95         *data = mm->data_vm + mm->stack_vm;
  96         *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
  97         return mm->total_vm;
  98 }
  99
 100 #ifdef CONFIG_NUMA
 101 /*
 102  * Save get_task_policy() for show_numa_map().
 103  */
 104 static void hold_task_mempolicy(struct proc_maps_private *priv)
 105 {
 106         struct task_struct *task = priv->task;
 107
 108         task_lock(task);
 109         priv->task_mempolicy = get_task_policy(task);
 110         mpol_get(priv->task_mempolicy);
 111         task_unlock(task);
 112 }
 113 static void release_task_mempolicy(struct proc_maps_private *priv)
 114 {
 115         mpol_put(priv->task_mempolicy);
 116 }
 117 #else
 118 static void hold_task_mempolicy(struct proc_maps_private *priv)
 119 {
 120 }
 121 static void release_task_mempolicy(struct proc_maps_private *priv)
 122 {
 123 }
 124 #endif
 125
 126 static void *m_start(struct seq_file *m, loff_t *ppos)
 127 {
 128         struct proc_maps_private *priv = m->private;
 129         unsigned long last_addr = *ppos;
 130         struct mm_struct *mm;
 131         struct vm_area_struct *vma;
 132
 133         /* See m_next(). Zero at the start or after lseek. */
 134         if (last_addr == -1UL)
 135                 return NULL;
 136
 137         priv->task = get_proc_task(priv->inode);
 138         if (!priv->task)
 139                 return ERR_PTR(-ESRCH);
 140
 141         mm = priv->mm;
 142         if (!mm || !mmget_not_zero(mm)) {
 143                 put_task_struct(priv->task);
 144                 priv->task = NULL;
 145                 return NULL;
 146         }
 147
 148         if (mmap_read_lock_killable(mm)) {
 149                 mmput(mm);
 150                 put_task_struct(priv->task);
 151                 priv->task = NULL;
 152                 return ERR_PTR(-EINTR);
 153         }
 154
 155         hold_task_mempolicy(priv);
 156         priv->tail_vma = get_gate_vma(mm);
 157
 158         vma = find_vma(mm, last_addr);
 159         if (vma)
 160                 return vma;
 161
 162         return priv->tail_vma;
 163 }
 164
 165 static void *m_next(struct seq_file *m, void *v, loff_t *ppos)
 166 {
 167         struct proc_maps_private *priv = m->private;
 168         struct vm_area_struct *next, *vma = v;
 169
 170         if (vma == priv->tail_vma)
 171                 next = NULL;
 172         else if (vma->vm_next)
 173                 next = vma->vm_next;
 174         else
 175                 next = priv->tail_vma;
 176
 177         *ppos = next ? next->vm_start : -1UL;
 178
 179         return next;
 180 }
 181
 182 static void m_stop(struct seq_file *m, void *v)
 183 {
 184         struct proc_maps_private *priv = m->private;
 185         struct mm_struct *mm = priv->mm;
 186
 187         if (!priv->task)
 188                 return;
 189
 190         release_task_mempolicy(priv);
 191         mmap_read_unlock(mm);
 192         mmput(mm);
 193         put_task_struct(priv->task);
 194         priv->task = NULL;
 195 }
 196
 197 static int proc_maps_open(struct inode *inode, struct file *file,
 198                         const struct seq_operations *ops, int psize)
 199 {
 200         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
 201
 202         if (!priv)
 203                 return -ENOMEM;
 204
 205         priv->inode = inode;
 206         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 207         if (IS_ERR(priv->mm)) {
 208                 int err = PTR_ERR(priv->mm);
 209
 210                 seq_release_private(inode, file);
 211                 return err;
 212         }
 213
 214         return 0;
 215 }
 216
 217 static int proc_map_release(struct inode *inode, struct file *file)
 218 {
 219         struct seq_file *seq = file->private_data;
 220         struct proc_maps_private *priv = seq->private;
 221
 222         if (priv->mm)
 223                 mmdrop(priv->mm);
 224
 225         return seq_release_private(inode, file);
 226 }
 227
 228 static int do_maps_open(struct inode *inode, struct file *file,
 229                         const struct seq_operations *ops)
 230 {
 231         return proc_maps_open(inode, file, ops,
 232                                 sizeof(struct proc_maps_private));
 233 }
 234
 235 /*
 236  * Indicate if the VMA is a stack for the given task; for
 237  * /proc/PID/maps that is the stack of the main task.
 238  */
 239 static int is_stack(struct vm_area_struct *vma)
 240 {
 241         /*
 242          * We make no effort to guess what a given thread considers to be
 243          * its "stack".  It's not even well-defined for programs written
 244          * languages like Go.
 245          */
 246         return vma->vm_start <= vma->vm_mm->start_stack &&
 247                 vma->vm_end >= vma->vm_mm->start_stack;
 248 }
 249
 250 static void show_vma_header_prefix(struct seq_file *m,
 251                                    unsigned long start, unsigned long end,
 252                                    vm_flags_t flags, unsigned long long pgoff,
 253                                    dev_t dev, unsigned long ino)
 254 {
 255         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 256         seq_put_hex_ll(m, NULL, start, 8);
 257         seq_put_hex_ll(m, "-", end, 8);
 258         seq_putc(m, ' ');
 259         seq_putc(m, flags & VM_READ ? 'r' : '-');
 260         seq_putc(m, flags & VM_WRITE ? 'w' : '-');
 261         seq_putc(m, flags & VM_EXEC ? 'x' : '-');
 262         seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
 263         seq_put_hex_ll(m, " ", pgoff, 8);
 264         seq_put_hex_ll(m, " ", MAJOR(dev), 2);
 265         seq_put_hex_ll(m, ":", MINOR(dev), 2);
 266         seq_put_decimal_ull(m, " ", ino);
 267         seq_putc(m, ' ');
 268 }
 269
 270 static void
 271 show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
 272 {
 273         struct mm_struct *mm = vma->vm_mm;
 274         struct file *file = vma->vm_file;
 275         vm_flags_t flags = vma->vm_flags;
 276         unsigned long ino = 0;
 277         unsigned long long pgoff = 0;
 278         unsigned long start, end;
 279         dev_t dev = 0;
 280         const char *name = NULL;
 281
 282         if (file) {
 283                 struct inode *inode = file_inode(vma->vm_file);
 284                 dev = inode->i_sb->s_dev;
 285                 ino = inode->i_ino;
 286                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 287         }
 288
 289         start = vma->vm_start;
 290         end = vma->vm_end;
 291         show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
 292
 293         /*
 294          * Print the dentry name for named mappings, and a
 295          * special [heap] marker for the heap:
 296          */
 297         if (file) {
 298                 seq_pad(m, ' ');
 299                 seq_file_path(m, file, "\n");
 300                 goto done;
 301         }
 302
 303         if (vma->vm_ops && vma->vm_ops->name) {
 304                 name = vma->vm_ops->name(vma);
 305                 if (name)
 306                         goto done;
 307         }
 308
 309         name = arch_vma_name(vma);
 310         if (!name) {
 311                 if (!mm) {
 312                         name = "[vdso]";
 313                         goto done;
 314                 }
 315
 316                 if (vma->vm_start <= mm->brk &&
 317                     vma->vm_end >= mm->start_brk) {
 318                         name = "[heap]";
 319                         goto done;
 320                 }
 321
 322                 if (is_stack(vma))
 323                         name = "[stack]";
 324         }
 325
 326 done:
 327         if (name) {
 328                 seq_pad(m, ' ');
 329                 seq_puts(m, name);
 330         }
 331         seq_putc(m, '\n');
 332 }
 333
 334 static int show_map(struct seq_file *m, void *v)
 335 {
 336         show_map_vma(m, v);
 337         return 0;
 338 }
 339
 340 static const struct seq_operations proc_pid_maps_op = {
 341         .start  = m_start,
 342         .next   = m_next,
 343         .stop   = m_stop,
 344         .show   = show_map
 345 };
 346
 347 static int pid_maps_open(struct inode *inode, struct file *file)
 348 {
 349         return do_maps_open(inode, file, &proc_pid_maps_op);
 350 }
 351
 352 const struct file_operations proc_pid_maps_operations = {
 353         .open           = pid_maps_open,
 354         .read           = seq_read,
 355         .llseek         = seq_lseek,
 356         .release        = proc_map_release,
 357 };
 358
 359 /*
 360  * Proportional Set Size(PSS): my share of RSS.
 361  *
 362  * PSS of a process is the count of pages it has in memory, where each
 363  * page is divided by the number of processes sharing it.  So if a
 364  * process has 1000 pages all to itself, and 1000 shared with one other
 365  * process, its PSS will be 1500.
 366  *
 367  * To keep (accumulated) division errors low, we adopt a 64bit
 368  * fixed-point pss counter to minimize division errors. So (pss >>
 369  * PSS_SHIFT) would be the real byte count.
 370  *
 371  * A shift of 12 before division means (assuming 4K page size):
 372  *      - 1M 3-user-pages add up to 8KB errors;
 373  *      - supports mapcount up to 2^24, or 16M;
 374  *      - supports PSS up to 2^52 bytes, or 4PB.
 375  */
 376 #define PSS_SHIFT 12
 377
 378 #ifdef CONFIG_PROC_PAGE_MONITOR
 379 struct mem_size_stats {
 380         unsigned long resident;
 381         unsigned long shared_clean;
 382         unsigned long shared_dirty;
 383         unsigned long private_clean;
 384         unsigned long private_dirty;
 385         unsigned long referenced;
 386         unsigned long anonymous;
 387         unsigned long lazyfree;
 388         unsigned long anonymous_thp;
 389         unsigned long shmem_thp;
 390         unsigned long file_thp;
 391         unsigned long swap;
 392         unsigned long shared_hugetlb;
 393         unsigned long private_hugetlb;
 394         u64 pss;
 395         u64 pss_anon;
 396         u64 pss_file;
 397         u64 pss_shmem;
 398         u64 pss_locked;
 399         u64 swap_pss;
 400         bool check_shmem_swap;
 401 };
 402
 403 static void smaps_page_accumulate(struct mem_size_stats *mss,
 404                 struct page *page, unsigned long size, unsigned long pss,
 405                 bool dirty, bool locked, bool private)
 406 {
 407         mss->pss += pss;
 408
 409         if (PageAnon(page))
 410                 mss->pss_anon += pss;
 411         else if (PageSwapBacked(page))
 412                 mss->pss_shmem += pss;
 413         else
 414                 mss->pss_file += pss;
 415
 416         if (locked)
 417                 mss->pss_locked += pss;
 418
 419         if (dirty || PageDirty(page)) {
 420                 if (private)
 421                         mss->private_dirty += size;
 422                 else
 423                         mss->shared_dirty += size;
 424         } else {
 425                 if (private)
 426                         mss->private_clean += size;
 427                 else
 428                         mss->shared_clean += size;
 429         }
 430 }
 431
 432 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 433                 bool compound, bool young, bool dirty, bool locked)
 434 {
 435         int i, nr = compound ? compound_nr(page) : 1;
 436         unsigned long size = nr * PAGE_SIZE;
 437
 438         /*
 439          * First accumulate quantities that depend only on |size| and the type
 440          * of the compound page.
 441          */
 442         if (PageAnon(page)) {
 443                 mss->anonymous += size;
 444                 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
 445                         mss->lazyfree += size;
 446         }
 447
 448         mss->resident += size;
 449         /* Accumulate the size in pages that have been accessed. */
 450         if (young || page_is_young(page) || PageReferenced(page))
 451                 mss->referenced += size;
 452
 453         /*
 454          * Then accumulate quantities that may depend on sharing, or that may
 455          * differ page-by-page.
 456          *
 457          * page_count(page) == 1 guarantees the page is mapped exactly once.
 458          * If any subpage of the compound page mapped with PTE it would elevate
 459          * page_count().
 460          */
 461         if (page_count(page) == 1) {
 462                 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
 463                         locked, true);
 464                 return;
 465         }
 466         for (i = 0; i < nr; i++, page++) {
 467                 int mapcount = page_mapcount(page);
 468                 unsigned long pss = PAGE_SIZE << PSS_SHIFT;
 469                 if (mapcount >= 2)
 470                         pss /= mapcount;
 471                 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
 472                                       mapcount < 2);
 473         }
 474 }
 475
 476 #ifdef CONFIG_SHMEM
 477 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 478                           __always_unused int depth, struct mm_walk *walk)
 479 {
 480         struct mem_size_stats *mss = walk->private;
 481
 482         mss->swap += shmem_partial_swap_usage(
 483                         walk->vma->vm_file->f_mapping, addr, end);
 484
 485         return 0;
 486 }
 487 #else
 488 #define smaps_pte_hole          NULL
 489 #endif /* CONFIG_SHMEM */
 490
 491 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 492                 struct mm_walk *walk)
 493 {
 494         struct mem_size_stats *mss = walk->private;
 495         struct vm_area_struct *vma = walk->vma;
 496         bool locked = !!(vma->vm_flags & VM_LOCKED);
 497         struct page *page = NULL;
 498
 499         if (pte_present(*pte)) {
 500                 page = vm_normal_page(vma, addr, *pte);
 501         } else if (is_swap_pte(*pte)) {
 502                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 503
 504                 if (!non_swap_entry(swpent)) {
 505                         int mapcount;
 506
 507                         mss->swap += PAGE_SIZE;
 508                         mapcount = swp_swapcount(swpent);
 509                         if (mapcount >= 2) {
 510                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 511
 512                                 do_div(pss_delta, mapcount);
 513                                 mss->swap_pss += pss_delta;
 514                         } else {
 515                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 516                         }
 517                 } else if (is_migration_entry(swpent))
 518                         page = migration_entry_to_page(swpent);
 519                 else if (is_device_private_entry(swpent))
 520                         page = device_private_entry_to_page(swpent);
 521         } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
 522                                                         && pte_none(*pte))) {
 523                 page = xa_load(&vma->vm_file->f_mapping->i_pages,
 524                                                 linear_page_index(vma, addr));
 525                 if (xa_is_value(page))
 526                         mss->swap += PAGE_SIZE;
 527                 return;
 528         }
 529
 530         if (!page)
 531                 return;
 532
 533         smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte), locked);
 534 }
 535
 536 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 537 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 538                 struct mm_walk *walk)
 539 {
 540         struct mem_size_stats *mss = walk->private;
 541         struct vm_area_struct *vma = walk->vma;
 542         bool locked = !!(vma->vm_flags & VM_LOCKED);
 543         struct page *page = NULL;
 544
 545         if (pmd_present(*pmd)) {
 546                 /* FOLL_DUMP will return -EFAULT on huge zero page */
 547                 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 548         } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) {
 549                 swp_entry_t entry = pmd_to_swp_entry(*pmd);
 550
 551                 if (is_migration_entry(entry))
 552                         page = migration_entry_to_page(entry);
 553         }
 554         if (IS_ERR_OR_NULL(page))
 555                 return;
 556         if (PageAnon(page))
 557                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 558         else if (PageSwapBacked(page))
 559                 mss->shmem_thp += HPAGE_PMD_SIZE;
 560         else if (is_zone_device_page(page))
 561                 /* pass */;
 562         else
 563                 mss->file_thp += HPAGE_PMD_SIZE;
 564         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd), locked);
 565 }
 566 #else
 567 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 568                 struct mm_walk *walk)
 569 {
 570 }
 571 #endif
 572
 573 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 574                            struct mm_walk *walk)
 575 {
 576         struct vm_area_struct *vma = walk->vma;
 577         pte_t *pte;
 578         spinlock_t *ptl;
 579
 580         ptl = pmd_trans_huge_lock(pmd, vma);
 581         if (ptl) {
 582                 smaps_pmd_entry(pmd, addr, walk);
 583                 spin_unlock(ptl);
 584                 goto out;
 585         }
 586
 587         if (pmd_trans_unstable(pmd))
 588                 goto out;
 589         /*
 590          * The mmap_lock held all the way back in m_start() is what
 591          * keeps khugepaged out of here and from collapsing things
 592          * in here.
 593          */
 594         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 595         for (; addr != end; pte++, addr += PAGE_SIZE)
 596                 smaps_pte_entry(pte, addr, walk);
 597         pte_unmap_unlock(pte - 1, ptl);
 598 out:
 599         cond_resched();
 600         return 0;
 601 }
 602
 603 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 604 {
 605         /*
 606          * Don't forget to update Documentation/ on changes.
 607          */
 608         static const char mnemonics[BITS_PER_LONG][2] = {
 609                 /*
 610                  * In case if we meet a flag we don't know about.
 611                  */
 612                 [0 ... (BITS_PER_LONG-1)] = "??",
 613
 614                 [ilog2(VM_READ)]        = "rd",
 615                 [ilog2(VM_WRITE)]       = "wr",
 616                 [ilog2(VM_EXEC)]        = "ex",
 617                 [ilog2(VM_SHARED)]      = "sh",
 618                 [ilog2(VM_MAYREAD)]     = "mr",
 619                 [ilog2(VM_MAYWRITE)]    = "mw",
 620                 [ilog2(VM_MAYEXEC)]     = "me",
 621                 [ilog2(VM_MAYSHARE)]    = "ms",
 622                 [ilog2(VM_GROWSDOWN)]   = "gd",
 623                 [ilog2(VM_PFNMAP)]      = "pf",
 624                 [ilog2(VM_DENYWRITE)]   = "dw",
 625                 [ilog2(VM_LOCKED)]      = "lo",
 626                 [ilog2(VM_IO)]          = "io",
 627                 [ilog2(VM_SEQ_READ)]    = "sr",
 628                 [ilog2(VM_RAND_READ)]   = "rr",
 629                 [ilog2(VM_DONTCOPY)]    = "dc",
 630                 [ilog2(VM_DONTEXPAND)]  = "de",
 631                 [ilog2(VM_ACCOUNT)]     = "ac",
 632                 [ilog2(VM_NORESERVE)]   = "nr",
 633                 [ilog2(VM_HUGETLB)]     = "ht",
 634                 [ilog2(VM_SYNC)]        = "sf",
 635                 [ilog2(VM_ARCH_1)]      = "ar",
 636                 [ilog2(VM_WIPEONFORK)]  = "wf",
 637                 [ilog2(VM_DONTDUMP)]    = "dd",
 638 #ifdef CONFIG_ARM64_BTI
 639                 [ilog2(VM_ARM64_BTI)]   = "bt",
 640 #endif
 641 #ifdef CONFIG_MEM_SOFT_DIRTY
 642                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 643 #endif
 644                 [ilog2(VM_MIXEDMAP)]    = "mm",
 645                 [ilog2(VM_HUGEPAGE)]    = "hg",
 646                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 647                 [ilog2(VM_MERGEABLE)]   = "mg",
 648                 [ilog2(VM_UFFD_MISSING)]= "um",
 649                 [ilog2(VM_UFFD_WP)]     = "uw",
 650 #ifdef CONFIG_ARM64_MTE
 651                 [ilog2(VM_MTE)]         = "mt",
 652                 [ilog2(VM_MTE_ALLOWED)] = "",
 653 #endif
 654 #ifdef CONFIG_ARCH_HAS_PKEYS
 655                 /* These come out via ProtectionKey: */
 656                 [ilog2(VM_PKEY_BIT0)]   = "",
 657                 [ilog2(VM_PKEY_BIT1)]   = "",
 658                 [ilog2(VM_PKEY_BIT2)]   = "",
 659                 [ilog2(VM_PKEY_BIT3)]   = "",
 660 #if VM_PKEY_BIT4
 661                 [ilog2(VM_PKEY_BIT4)]   = "",
 662 #endif
 663 #endif /* CONFIG_ARCH_HAS_PKEYS */
 664         };
 665         size_t i;
 666
 667         seq_puts(m, "VmFlags: ");
 668         for (i = 0; i < BITS_PER_LONG; i++) {
 669                 if (!mnemonics[i][0])
 670                         continue;
 671                 if (vma->vm_flags & (1UL << i)) {
 672                         seq_putc(m, mnemonics[i][0]);
 673                         seq_putc(m, mnemonics[i][1]);
 674                         seq_putc(m, ' ');
 675                 }
 676         }
 677         seq_putc(m, '\n');
 678 }
 679
 680 #ifdef CONFIG_HUGETLB_PAGE
 681 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 682                                  unsigned long addr, unsigned long end,
 683                                  struct mm_walk *walk)
 684 {
 685         struct mem_size_stats *mss = walk->private;
 686         struct vm_area_struct *vma = walk->vma;
 687         struct page *page = NULL;
 688
 689         if (pte_present(*pte)) {
 690                 page = vm_normal_page(vma, addr, *pte);
 691         } else if (is_swap_pte(*pte)) {
 692                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 693
 694                 if (is_migration_entry(swpent))
 695                         page = migration_entry_to_page(swpent);
 696                 else if (is_device_private_entry(swpent))
 697                         page = device_private_entry_to_page(swpent);
 698         }
 699         if (page) {
 700                 int mapcount = page_mapcount(page);
 701
 702                 if (mapcount >= 2)
 703                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 704                 else
 705                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 706         }
 707         return 0;
 708 }
 709 #else
 710 #define smaps_hugetlb_range     NULL
 711 #endif /* HUGETLB_PAGE */
 712
 713 static const struct mm_walk_ops smaps_walk_ops = {
 714         .pmd_entry              = smaps_pte_range,
 715         .hugetlb_entry          = smaps_hugetlb_range,
 716 };
 717
 718 static const struct mm_walk_ops smaps_shmem_walk_ops = {
 719         .pmd_entry              = smaps_pte_range,
 720         .hugetlb_entry          = smaps_hugetlb_range,
 721         .pte_hole               = smaps_pte_hole,
 722 };
 723
 724 /*
 725  * Gather mem stats from @vma with the indicated beginning
 726  * address @start, and keep them in @mss.
 727  *
 728  * Use vm_start of @vma as the beginning address if @start is 0.
 729  */
 730 static void smap_gather_stats(struct vm_area_struct *vma,
 731                 struct mem_size_stats *mss, unsigned long start)
 732 {
 733         const struct mm_walk_ops *ops = &smaps_walk_ops;
 734
 735         /* Invalid start */
 736         if (start >= vma->vm_end)
 737                 return;
 738
 739 #ifdef CONFIG_SHMEM
 740         /* In case of smaps_rollup, reset the value from previous vma */
 741         mss->check_shmem_swap = false;
 742         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 743                 /*
 744                  * For shared or readonly shmem mappings we know that all
 745                  * swapped out pages belong to the shmem object, and we can
 746                  * obtain the swap value much more efficiently. For private
 747                  * writable mappings, we might have COW pages that are
 748                  * not affected by the parent swapped out pages of the shmem
 749                  * object, so we have to distinguish them during the page walk.
 750                  * Unless we know that the shmem object (or the part mapped by
 751                  * our VMA) has no swapped out pages at all.
 752                  */
 753                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 754
 755                 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 756                                         !(vma->vm_flags & VM_WRITE))) {
 757                         mss->swap += shmem_swapped;
 758                 } else {
 759                         mss->check_shmem_swap = true;
 760                         ops = &smaps_shmem_walk_ops;
 761                 }
 762         }
 763 #endif
 764         /* mmap_lock is held in m_start */
 765         if (!start)
 766                 walk_page_vma(vma, ops, mss);
 767         else
 768                 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss);
 769 }
 770
 771 #define SEQ_PUT_DEC(str, val) \
 772                 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
 773
 774 /* Show the contents common for smaps and smaps_rollup */
 775 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
 776         bool rollup_mode)
 777 {
 778         SEQ_PUT_DEC("Rss:            ", mss->resident);
 779         SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
 780         if (rollup_mode) {
 781                 /*
 782                  * These are meaningful only for smaps_rollup, otherwise two of
 783                  * them are zero, and the other one is the same as Pss.
 784                  */
 785                 SEQ_PUT_DEC(" kB\nPss_Anon:       ",
 786                         mss->pss_anon >> PSS_SHIFT);
 787                 SEQ_PUT_DEC(" kB\nPss_File:       ",
 788                         mss->pss_file >> PSS_SHIFT);
 789                 SEQ_PUT_DEC(" kB\nPss_Shmem:      ",
 790                         mss->pss_shmem >> PSS_SHIFT);
 791         }
 792         SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
 793         SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
 794         SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
 795         SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
 796         SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
 797         SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
 798         SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
 799         SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
 800         SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
 801         SEQ_PUT_DEC(" kB\nFilePmdMapped:  ", mss->file_thp);
 802         SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
 803         seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
 804                                   mss->private_hugetlb >> 10, 7);
 805         SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
 806         SEQ_PUT_DEC(" kB\nSwapPss:        ",
 807                                         mss->swap_pss >> PSS_SHIFT);
 808         SEQ_PUT_DEC(" kB\nLocked:         ",
 809                                         mss->pss_locked >> PSS_SHIFT);
 810         seq_puts(m, " kB\n");
 811 }
 812
 813 static int show_smap(struct seq_file *m, void *v)
 814 {
 815         struct vm_area_struct *vma = v;
 816         struct mem_size_stats mss;
 817
 818         memset(&mss, 0, sizeof(mss));
 819
 820         smap_gather_stats(vma, &mss, 0);
 821
 822         show_map_vma(m, vma);
 823
 824         SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
 825         SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
 826         SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
 827         seq_puts(m, " kB\n");
 828
 829         __show_smap(m, &mss, false);
 830
 831         seq_printf(m, "THPeligible:    %d\n",
 832                    transparent_hugepage_enabled(vma));
 833
 834         if (arch_pkeys_enabled())
 835                 seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
 836         show_smap_vma_flags(m, vma);
 837
 838         return 0;
 839 }
 840
 841 static int show_smaps_rollup(struct seq_file *m, void *v)
 842 {
 843         struct proc_maps_private *priv = m->private;
 844         struct mem_size_stats mss;
 845         struct mm_struct *mm;
 846         struct vm_area_struct *vma;
 847         unsigned long last_vma_end = 0;
 848         int ret = 0;
 849
 850         priv->task = get_proc_task(priv->inode);
 851         if (!priv->task)
 852                 return -ESRCH;
 853
 854         mm = priv->mm;
 855         if (!mm || !mmget_not_zero(mm)) {
 856                 ret = -ESRCH;
 857                 goto out_put_task;
 858         }
 859
 860         memset(&mss, 0, sizeof(mss));
 861
 862         ret = mmap_read_lock_killable(mm);
 863         if (ret)
 864                 goto out_put_mm;
 865
 866         hold_task_mempolicy(priv);
 867
 868         for (vma = priv->mm->mmap; vma;) {
 869                 smap_gather_stats(vma, &mss, 0);
 870                 last_vma_end = vma->vm_end;
 871
 872                 /*
 873                  * Release mmap_lock temporarily if someone wants to
 874                  * access it for write request.
 875                  */
 876                 if (mmap_lock_is_contended(mm)) {
 877                         mmap_read_unlock(mm);
 878                         ret = mmap_read_lock_killable(mm);
 879                         if (ret) {
 880                                 release_task_mempolicy(priv);
 881                                 goto out_put_mm;
 882                         }
 883
 884                         /*
 885                          * After dropping the lock, there are four cases to
 886                          * consider. See the following example for explanation.
 887                          *
 888                          *   +------+------+-----------+
 889                          *   | VMA1 | VMA2 | VMA3      |
 890                          *   +------+------+-----------+
 891                          *   |      |      |           |
 892                          *  4k     8k     16k         400k
 893                          *
 894                          * Suppose we drop the lock after reading VMA2 due to
 895                          * contention, then we get:
 896                          *
 897                          *      last_vma_end = 16k
 898                          *
 899                          * 1) VMA2 is freed, but VMA3 exists:
 900                          *
 901                          *    find_vma(mm, 16k - 1) will return VMA3.
 902                          *    In this case, just continue from VMA3.
 903                          *
 904                          * 2) VMA2 still exists:
 905                          *
 906                          *    find_vma(mm, 16k - 1) will return VMA2.
 907                          *    Iterate the loop like the original one.
 908                          *
 909                          * 3) No more VMAs can be found:
 910                          *
 911                          *    find_vma(mm, 16k - 1) will return NULL.
 912                          *    No more things to do, just break.
 913                          *
 914                          * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
 915                          *
 916                          *    find_vma(mm, 16k - 1) will return VMA' whose range
 917                          *    contains last_vma_end.
 918                          *    Iterate VMA' from last_vma_end.
 919                          */
 920                         vma = find_vma(mm, last_vma_end - 1);
 921                         /* Case 3 above */
 922                         if (!vma)
 923                                 break;
 924
 925                         /* Case 1 above */
 926                         if (vma->vm_start >= last_vma_end)
 927                                 continue;
 928
 929                         /* Case 4 above */
 930                         if (vma->vm_end > last_vma_end)
 931                                 smap_gather_stats(vma, &mss, last_vma_end);
 932                 }
 933                 /* Case 2 above */
 934                 vma = vma->vm_next;
 935         }
 936
 937         show_vma_header_prefix(m, priv->mm->mmap->vm_start,
 938                                last_vma_end, 0, 0, 0, 0);
 939         seq_pad(m, ' ');
 940         seq_puts(m, "[rollup]\n");
 941
 942         __show_smap(m, &mss, true);
 943
 944         release_task_mempolicy(priv);
 945         mmap_read_unlock(mm);
 946
 947 out_put_mm:
 948         mmput(mm);
 949 out_put_task:
 950         put_task_struct(priv->task);
 951         priv->task = NULL;
 952
 953         return ret;
 954 }
 955 #undef SEQ_PUT_DEC
 956
 957 static const struct seq_operations proc_pid_smaps_op = {
 958         .start  = m_start,
 959         .next   = m_next,
 960         .stop   = m_stop,
 961         .show   = show_smap
 962 };
 963
 964 static int pid_smaps_open(struct inode *inode, struct file *file)
 965 {
 966         return do_maps_open(inode, file, &proc_pid_smaps_op);
 967 }
 968
 969 static int smaps_rollup_open(struct inode *inode, struct file *file)
 970 {
 971         int ret;
 972         struct proc_maps_private *priv;
 973
 974         priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
 975         if (!priv)
 976                 return -ENOMEM;
 977
 978         ret = single_open(file, show_smaps_rollup, priv);
 979         if (ret)
 980                 goto out_free;
 981
 982         priv->inode = inode;
 983         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 984         if (IS_ERR(priv->mm)) {
 985                 ret = PTR_ERR(priv->mm);
 986
 987                 single_release(inode, file);
 988                 goto out_free;
 989         }
 990
 991         return 0;
 992
 993 out_free:
 994         kfree(priv);
 995         return ret;
 996 }
 997
 998 static int smaps_rollup_release(struct inode *inode, struct file *file)
 999 {
1000         struct seq_file *seq = file->private_data;
1001         struct proc_maps_private *priv = seq->private;
1002
1003         if (priv->mm)
1004                 mmdrop(priv->mm);
1005
1006         kfree(priv);
1007         return single_release(inode, file);
1008 }
1009
1010 const struct file_operations proc_pid_smaps_operations = {
1011         .open           = pid_smaps_open,
1012         .read           = seq_read,
1013         .llseek         = seq_lseek,
1014         .release        = proc_map_release,
1015 };
1016
1017 const struct file_operations proc_pid_smaps_rollup_operations = {
1018         .open           = smaps_rollup_open,
1019         .read           = seq_read,
1020         .llseek         = seq_lseek,
1021         .release        = smaps_rollup_release,
1022 };
1023
1024 enum clear_refs_types {
1025         CLEAR_REFS_ALL = 1,
1026         CLEAR_REFS_ANON,
1027         CLEAR_REFS_MAPPED,
1028         CLEAR_REFS_SOFT_DIRTY,
1029         CLEAR_REFS_MM_HIWATER_RSS,
1030         CLEAR_REFS_LAST,
1031 };
1032
1033 struct clear_refs_private {
1034         enum clear_refs_types type;
1035 };
1036
1037 #ifdef CONFIG_MEM_SOFT_DIRTY
1038 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1039                 unsigned long addr, pte_t *pte)
1040 {
1041         /*
1042          * The soft-dirty tracker uses #PF-s to catch writes
1043          * to pages, so write-protect the pte as well. See the
1044          * Documentation/admin-guide/mm/soft-dirty.rst for full description
1045          * of how soft-dirty works.
1046          */
1047         pte_t ptent = *pte;
1048
1049         if (pte_present(ptent)) {
1050                 pte_t old_pte;
1051
1052                 old_pte = ptep_modify_prot_start(vma, addr, pte);
1053                 ptent = pte_wrprotect(old_pte);
1054                 ptent = pte_clear_soft_dirty(ptent);
1055                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1056         } else if (is_swap_pte(ptent)) {
1057                 ptent = pte_swp_clear_soft_dirty(ptent);
1058                 set_pte_at(vma->vm_mm, addr, pte, ptent);
1059         }
1060 }
1061 #else
1062 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1063                 unsigned long addr, pte_t *pte)
1064 {
1065 }
1066 #endif
1067
1068 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1069 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1070                 unsigned long addr, pmd_t *pmdp)
1071 {
1072         pmd_t old, pmd = *pmdp;
1073
1074         if (pmd_present(pmd)) {
1075                 /* See comment in change_huge_pmd() */
1076                 old = pmdp_invalidate(vma, addr, pmdp);
1077                 if (pmd_dirty(old))
1078                         pmd = pmd_mkdirty(pmd);
1079                 if (pmd_young(old))
1080                         pmd = pmd_mkyoung(pmd);
1081
1082                 pmd = pmd_wrprotect(pmd);
1083                 pmd = pmd_clear_soft_dirty(pmd);
1084
1085                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1086         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1087                 pmd = pmd_swp_clear_soft_dirty(pmd);
1088                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1089         }
1090 }
1091 #else
1092 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1093                 unsigned long addr, pmd_t *pmdp)
1094 {
1095 }
1096 #endif
1097
1098 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1099                                 unsigned long end, struct mm_walk *walk)
1100 {
1101         struct clear_refs_private *cp = walk->private;
1102         struct vm_area_struct *vma = walk->vma;
1103         pte_t *pte, ptent;
1104         spinlock_t *ptl;
1105         struct page *page;
1106
1107         ptl = pmd_trans_huge_lock(pmd, vma);
1108         if (ptl) {
1109                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1110                         clear_soft_dirty_pmd(vma, addr, pmd);
1111                         goto out;
1112                 }
1113
1114                 if (!pmd_present(*pmd))
1115                         goto out;
1116
1117                 page = pmd_page(*pmd);
1118
1119                 /* Clear accessed and referenced bits. */
1120                 pmdp_test_and_clear_young(vma, addr, pmd);
1121                 test_and_clear_page_young(page);
1122                 ClearPageReferenced(page);
1123 out:
1124                 spin_unlock(ptl);
1125                 return 0;
1126         }
1127
1128         if (pmd_trans_unstable(pmd))
1129                 return 0;
1130
1131         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1132         for (; addr != end; pte++, addr += PAGE_SIZE) {
1133                 ptent = *pte;
1134
1135                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1136                         clear_soft_dirty(vma, addr, pte);
1137                         continue;
1138                 }
1139
1140                 if (!pte_present(ptent))
1141                         continue;
1142
1143                 page = vm_normal_page(vma, addr, ptent);
1144                 if (!page)
1145                         continue;
1146
1147                 /* Clear accessed and referenced bits. */
1148                 ptep_test_and_clear_young(vma, addr, pte);
1149                 test_and_clear_page_young(page);
1150                 ClearPageReferenced(page);
1151         }
1152         pte_unmap_unlock(pte - 1, ptl);
1153         cond_resched();
1154         return 0;
1155 }
1156
1157 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1158                                 struct mm_walk *walk)
1159 {
1160         struct clear_refs_private *cp = walk->private;
1161         struct vm_area_struct *vma = walk->vma;
1162
1163         if (vma->vm_flags & VM_PFNMAP)
1164                 return 1;
1165
1166         /*
1167          * Writing 1 to /proc/pid/clear_refs affects all pages.
1168          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1169          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1170          * Writing 4 to /proc/pid/clear_refs affects all pages.
1171          */
1172         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1173                 return 1;
1174         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1175                 return 1;
1176         return 0;
1177 }
1178
1179 static const struct mm_walk_ops clear_refs_walk_ops = {
1180         .pmd_entry              = clear_refs_pte_range,
1181         .test_walk              = clear_refs_test_walk,
1182 };
1183
1184 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1185                                 size_t count, loff_t *ppos)
1186 {
1187         struct task_struct *task;
1188         char buffer[PROC_NUMBUF];
1189         struct mm_struct *mm;
1190         struct vm_area_struct *vma;
1191         enum clear_refs_types type;
1192         struct mmu_gather tlb;
1193         int itype;
1194         int rv;
1195
1196         memset(buffer, 0, sizeof(buffer));
1197         if (count > sizeof(buffer) - 1)
1198                 count = sizeof(buffer) - 1;
1199         if (copy_from_user(buffer, buf, count))
1200                 return -EFAULT;
1201         rv = kstrtoint(strstrip(buffer), 10, &itype);
1202         if (rv < 0)
1203                 return rv;
1204         type = (enum clear_refs_types)itype;
1205         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1206                 return -EINVAL;
1207
1208         task = get_proc_task(file_inode(file));
1209         if (!task)
1210                 return -ESRCH;
1211         mm = get_task_mm(task);
1212         if (mm) {
1213                 struct mmu_notifier_range range;
1214                 struct clear_refs_private cp = {
1215                         .type = type,
1216                 };
1217
1218                 if (mmap_write_lock_killable(mm)) {
1219                         count = -EINTR;
1220                         goto out_mm;
1221                 }
1222                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1223                         /*
1224                          * Writing 5 to /proc/pid/clear_refs resets the peak
1225                          * resident set size to this mm's current rss value.
1226                          */
1227                         reset_mm_hiwater_rss(mm);
1228                         goto out_unlock;
1229                 }
1230
1231                 tlb_gather_mmu(&tlb, mm, 0, -1);
1232                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1233                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1234                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1235                                         continue;
1236                                 vma->vm_flags &= ~VM_SOFTDIRTY;
1237                                 vma_set_page_prot(vma);
1238                         }
1239
1240                         mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1241                                                 0, NULL, mm, 0, -1UL);
1242                         mmu_notifier_invalidate_range_start(&range);
1243                 }
1244                 walk_page_range(mm, 0, mm->highest_vm_end, &clear_refs_walk_ops,
1245                                 &cp);
1246                 if (type == CLEAR_REFS_SOFT_DIRTY)
1247                         mmu_notifier_invalidate_range_end(&range);
1248                 tlb_finish_mmu(&tlb, 0, -1);
1249 out_unlock:
1250                 mmap_write_unlock(mm);
1251 out_mm:
1252                 mmput(mm);
1253         }
1254         put_task_struct(task);
1255
1256         return count;
1257 }
1258
1259 const struct file_operations proc_clear_refs_operations = {
1260         .write          = clear_refs_write,
1261         .llseek         = noop_llseek,
1262 };
1263
1264 typedef struct {
1265         u64 pme;
1266 } pagemap_entry_t;
1267
1268 struct pagemapread {
1269         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1270         pagemap_entry_t *buffer;
1271         bool show_pfn;
1272 };
1273
1274 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1275 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1276
1277 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1278 #define PM_PFRAME_BITS          55
1279 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1280 #define PM_SOFT_DIRTY           BIT_ULL(55)
1281 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1282 #define PM_FILE                 BIT_ULL(61)
1283 #define PM_SWAP                 BIT_ULL(62)
1284 #define PM_PRESENT              BIT_ULL(63)
1285
1286 #define PM_END_OF_BUFFER    1
1287
1288 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1289 {
1290         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1291 }
1292
1293 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1294                           struct pagemapread *pm)
1295 {
1296         pm->buffer[pm->pos++] = *pme;
1297         if (pm->pos >= pm->len)
1298                 return PM_END_OF_BUFFER;
1299         return 0;
1300 }
1301
1302 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1303                             __always_unused int depth, struct mm_walk *walk)
1304 {
1305         struct pagemapread *pm = walk->private;
1306         unsigned long addr = start;
1307         int err = 0;
1308
1309         while (addr < end) {
1310                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1311                 pagemap_entry_t pme = make_pme(0, 0);
1312                 /* End of address space hole, which we mark as non-present. */
1313                 unsigned long hole_end;
1314
1315                 if (vma)
1316                         hole_end = min(end, vma->vm_start);
1317                 else
1318                         hole_end = end;
1319
1320                 for (; addr < hole_end; addr += PAGE_SIZE) {
1321                         err = add_to_pagemap(addr, &pme, pm);
1322                         if (err)
1323                                 goto out;
1324                 }
1325
1326                 if (!vma)
1327                         break;
1328
1329                 /* Addresses in the VMA. */
1330                 if (vma->vm_flags & VM_SOFTDIRTY)
1331                         pme = make_pme(0, PM_SOFT_DIRTY);
1332                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1333                         err = add_to_pagemap(addr, &pme, pm);
1334                         if (err)
1335                                 goto out;
1336                 }
1337         }
1338 out:
1339         return err;
1340 }
1341
1342 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1343                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1344 {
1345         u64 frame = 0, flags = 0;
1346         struct page *page = NULL;
1347
1348         if (pte_present(pte)) {
1349                 if (pm->show_pfn)
1350                         frame = pte_pfn(pte);
1351                 flags |= PM_PRESENT;
1352                 page = vm_normal_page(vma, addr, pte);
1353                 if (pte_soft_dirty(pte))
1354                         flags |= PM_SOFT_DIRTY;
1355         } else if (is_swap_pte(pte)) {
1356                 swp_entry_t entry;
1357                 if (pte_swp_soft_dirty(pte))
1358                         flags |= PM_SOFT_DIRTY;
1359                 entry = pte_to_swp_entry(pte);
1360                 if (pm->show_pfn)
1361                         frame = swp_type(entry) |
1362                                 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1363                 flags |= PM_SWAP;
1364                 if (is_migration_entry(entry))
1365                         page = migration_entry_to_page(entry);
1366
1367                 if (is_device_private_entry(entry))
1368                         page = device_private_entry_to_page(entry);
1369         }
1370
1371         if (page && !PageAnon(page))
1372                 flags |= PM_FILE;
1373         if (page && page_mapcount(page) == 1)
1374                 flags |= PM_MMAP_EXCLUSIVE;
1375         if (vma->vm_flags & VM_SOFTDIRTY)
1376                 flags |= PM_SOFT_DIRTY;
1377
1378         return make_pme(frame, flags);
1379 }
1380
1381 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1382                              struct mm_walk *walk)
1383 {
1384         struct vm_area_struct *vma = walk->vma;
1385         struct pagemapread *pm = walk->private;
1386         spinlock_t *ptl;
1387         pte_t *pte, *orig_pte;
1388         int err = 0;
1389
1390 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1391         ptl = pmd_trans_huge_lock(pmdp, vma);
1392         if (ptl) {
1393                 u64 flags = 0, frame = 0;
1394                 pmd_t pmd = *pmdp;
1395                 struct page *page = NULL;
1396
1397                 if (vma->vm_flags & VM_SOFTDIRTY)
1398                         flags |= PM_SOFT_DIRTY;
1399
1400                 if (pmd_present(pmd)) {
1401                         page = pmd_page(pmd);
1402
1403                         flags |= PM_PRESENT;
1404                         if (pmd_soft_dirty(pmd))
1405                                 flags |= PM_SOFT_DIRTY;
1406                         if (pm->show_pfn)
1407                                 frame = pmd_pfn(pmd) +
1408                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1409                 }
1410 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1411                 else if (is_swap_pmd(pmd)) {
1412                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1413                         unsigned long offset;
1414
1415                         if (pm->show_pfn) {
1416                                 offset = swp_offset(entry) +
1417                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1418                                 frame = swp_type(entry) |
1419                                         (offset << MAX_SWAPFILES_SHIFT);
1420                         }
1421                         flags |= PM_SWAP;
1422                         if (pmd_swp_soft_dirty(pmd))
1423                                 flags |= PM_SOFT_DIRTY;
1424                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1425                         page = migration_entry_to_page(entry);
1426                 }
1427 #endif
1428
1429                 if (page && page_mapcount(page) == 1)
1430                         flags |= PM_MMAP_EXCLUSIVE;
1431
1432                 for (; addr != end; addr += PAGE_SIZE) {
1433                         pagemap_entry_t pme = make_pme(frame, flags);
1434
1435                         err = add_to_pagemap(addr, &pme, pm);
1436                         if (err)
1437                                 break;
1438                         if (pm->show_pfn) {
1439                                 if (flags & PM_PRESENT)
1440                                         frame++;
1441                                 else if (flags & PM_SWAP)
1442                                         frame += (1 << MAX_SWAPFILES_SHIFT);
1443                         }
1444                 }
1445                 spin_unlock(ptl);
1446                 return err;
1447         }
1448
1449         if (pmd_trans_unstable(pmdp))
1450                 return 0;
1451 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1452
1453         /*
1454          * We can assume that @vma always points to a valid one and @end never
1455          * goes beyond vma->vm_end.
1456          */
1457         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1458         for (; addr < end; pte++, addr += PAGE_SIZE) {
1459                 pagemap_entry_t pme;
1460
1461                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1462                 err = add_to_pagemap(addr, &pme, pm);
1463                 if (err)
1464                         break;
1465         }
1466         pte_unmap_unlock(orig_pte, ptl);
1467
1468         cond_resched();
1469
1470         return err;
1471 }
1472
1473 #ifdef CONFIG_HUGETLB_PAGE
1474 /* This function walks within one hugetlb entry in the single call */
1475 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1476                                  unsigned long addr, unsigned long end,
1477                                  struct mm_walk *walk)
1478 {
1479         struct pagemapread *pm = walk->private;
1480         struct vm_area_struct *vma = walk->vma;
1481         u64 flags = 0, frame = 0;
1482         int err = 0;
1483         pte_t pte;
1484
1485         if (vma->vm_flags & VM_SOFTDIRTY)
1486                 flags |= PM_SOFT_DIRTY;
1487
1488         pte = huge_ptep_get(ptep);
1489         if (pte_present(pte)) {
1490                 struct page *page = pte_page(pte);
1491
1492                 if (!PageAnon(page))
1493                         flags |= PM_FILE;
1494
1495                 if (page_mapcount(page) == 1)
1496                         flags |= PM_MMAP_EXCLUSIVE;
1497
1498                 flags |= PM_PRESENT;
1499                 if (pm->show_pfn)
1500                         frame = pte_pfn(pte) +
1501                                 ((addr & ~hmask) >> PAGE_SHIFT);
1502         }
1503
1504         for (; addr != end; addr += PAGE_SIZE) {
1505                 pagemap_entry_t pme = make_pme(frame, flags);
1506
1507                 err = add_to_pagemap(addr, &pme, pm);
1508                 if (err)
1509                         return err;
1510                 if (pm->show_pfn && (flags & PM_PRESENT))
1511                         frame++;
1512         }
1513
1514         cond_resched();
1515
1516         return err;
1517 }
1518 #else
1519 #define pagemap_hugetlb_range   NULL
1520 #endif /* HUGETLB_PAGE */
1521
1522 static const struct mm_walk_ops pagemap_ops = {
1523         .pmd_entry      = pagemap_pmd_range,
1524         .pte_hole       = pagemap_pte_hole,
1525         .hugetlb_entry  = pagemap_hugetlb_range,
1526 };
1527
1528 /*
1529  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1530  *
1531  * For each page in the address space, this file contains one 64-bit entry
1532  * consisting of the following:
1533  *
1534  * Bits 0-54  page frame number (PFN) if present
1535  * Bits 0-4   swap type if swapped
1536  * Bits 5-54  swap offset if swapped
1537  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1538  * Bit  56    page exclusively mapped
1539  * Bits 57-60 zero
1540  * Bit  61    page is file-page or shared-anon
1541  * Bit  62    page swapped
1542  * Bit  63    page present
1543  *
1544  * If the page is not present but in swap, then the PFN contains an
1545  * encoding of the swap file number and the page's offset into the
1546  * swap. Unmapped pages return a null PFN. This allows determining
1547  * precisely which pages are mapped (or in swap) and comparing mapped
1548  * pages between processes.
1549  *
1550  * Efficient users of this interface will use /proc/pid/maps to
1551  * determine which areas of memory are actually mapped and llseek to
1552  * skip over unmapped regions.
1553  */
1554 static ssize_t pagemap_read(struct file *file, char __user *buf,
1555                             size_t count, loff_t *ppos)
1556 {
1557         struct mm_struct *mm = file->private_data;
1558         struct pagemapread pm;
1559         unsigned long src;
1560         unsigned long svpfn;
1561         unsigned long start_vaddr;
1562         unsigned long end_vaddr;
1563         int ret = 0, copied = 0;
1564
1565         if (!mm || !mmget_not_zero(mm))
1566                 goto out;
1567
1568         ret = -EINVAL;
1569         /* file position must be aligned */
1570         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1571                 goto out_mm;
1572
1573         ret = 0;
1574         if (!count)
1575                 goto out_mm;
1576
1577         /* do not disclose physical addresses: attack vector */
1578         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1579
1580         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1581         pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1582         ret = -ENOMEM;
1583         if (!pm.buffer)
1584                 goto out_mm;
1585
1586         src = *ppos;
1587         svpfn = src / PM_ENTRY_BYTES;
1588         end_vaddr = mm->task_size;
1589
1590         /* watch out for wraparound */
1591         start_vaddr = end_vaddr;
1592         if (svpfn <= (ULONG_MAX >> PAGE_SHIFT))
1593                 start_vaddr = untagged_addr(svpfn << PAGE_SHIFT);
1594
1595         /* Ensure the address is inside the task */
1596         if (start_vaddr > mm->task_size)
1597                 start_vaddr = end_vaddr;
1598
1599         /*
1600          * The odds are that this will stop walking way
1601          * before end_vaddr, because the length of the
1602          * user buffer is tracked in "pm", and the walk
1603          * will stop when we hit the end of the buffer.
1604          */
1605         ret = 0;
1606         while (count && (start_vaddr < end_vaddr)) {
1607                 int len;
1608                 unsigned long end;
1609
1610                 pm.pos = 0;
1611                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1612                 /* overflow ? */
1613                 if (end < start_vaddr || end > end_vaddr)
1614                         end = end_vaddr;
1615                 ret = mmap_read_lock_killable(mm);
1616                 if (ret)
1617                         goto out_free;
1618                 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1619                 mmap_read_unlock(mm);
1620                 start_vaddr = end;
1621
1622                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1623                 if (copy_to_user(buf, pm.buffer, len)) {
1624                         ret = -EFAULT;
1625                         goto out_free;
1626                 }
1627                 copied += len;
1628                 buf += len;
1629                 count -= len;
1630         }
1631         *ppos += copied;
1632         if (!ret || ret == PM_END_OF_BUFFER)
1633                 ret = copied;
1634
1635 out_free:
1636         kfree(pm.buffer);
1637 out_mm:
1638         mmput(mm);
1639 out:
1640         return ret;
1641 }
1642
1643 static int pagemap_open(struct inode *inode, struct file *file)
1644 {
1645         struct mm_struct *mm;
1646
1647         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1648         if (IS_ERR(mm))
1649                 return PTR_ERR(mm);
1650         file->private_data = mm;
1651         return 0;
1652 }
1653
1654 static int pagemap_release(struct inode *inode, struct file *file)
1655 {
1656         struct mm_struct *mm = file->private_data;
1657
1658         if (mm)
1659                 mmdrop(mm);
1660         return 0;
1661 }
1662
1663 const struct file_operations proc_pagemap_operations = {
1664         .llseek         = mem_lseek, /* borrow this */
1665         .read           = pagemap_read,
1666         .open           = pagemap_open,
1667         .release        = pagemap_release,
1668 };
1669 #endif /* CONFIG_PROC_PAGE_MONITOR */
1670
1671 #ifdef CONFIG_NUMA
1672
1673 struct numa_maps {
1674         unsigned long pages;
1675         unsigned long anon;
1676         unsigned long active;
1677         unsigned long writeback;
1678         unsigned long mapcount_max;
1679         unsigned long dirty;
1680         unsigned long swapcache;
1681         unsigned long node[MAX_NUMNODES];
1682 };
1683
1684 struct numa_maps_private {
1685         struct proc_maps_private proc_maps;
1686         struct numa_maps md;
1687 };
1688
1689 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1690                         unsigned long nr_pages)
1691 {
1692         int count = page_mapcount(page);
1693
1694         md->pages += nr_pages;
1695         if (pte_dirty || PageDirty(page))
1696                 md->dirty += nr_pages;
1697
1698         if (PageSwapCache(page))
1699                 md->swapcache += nr_pages;
1700
1701         if (PageActive(page) || PageUnevictable(page))
1702                 md->active += nr_pages;
1703
1704         if (PageWriteback(page))
1705                 md->writeback += nr_pages;
1706
1707         if (PageAnon(page))
1708                 md->anon += nr_pages;
1709
1710         if (count > md->mapcount_max)
1711                 md->mapcount_max = count;
1712
1713         md->node[page_to_nid(page)] += nr_pages;
1714 }
1715
1716 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1717                 unsigned long addr)
1718 {
1719         struct page *page;
1720         int nid;
1721
1722         if (!pte_present(pte))
1723                 return NULL;
1724
1725         page = vm_normal_page(vma, addr, pte);
1726         if (!page)
1727                 return NULL;
1728
1729         if (PageReserved(page))
1730                 return NULL;
1731
1732         nid = page_to_nid(page);
1733         if (!node_isset(nid, node_states[N_MEMORY]))
1734                 return NULL;
1735
1736         return page;
1737 }
1738
1739 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1740 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1741                                               struct vm_area_struct *vma,
1742                                               unsigned long addr)
1743 {
1744         struct page *page;
1745         int nid;
1746
1747         if (!pmd_present(pmd))
1748                 return NULL;
1749
1750         page = vm_normal_page_pmd(vma, addr, pmd);
1751         if (!page)
1752                 return NULL;
1753
1754         if (PageReserved(page))
1755                 return NULL;
1756
1757         nid = page_to_nid(page);
1758         if (!node_isset(nid, node_states[N_MEMORY]))
1759                 return NULL;
1760
1761         return page;
1762 }
1763 #endif
1764
1765 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1766                 unsigned long end, struct mm_walk *walk)
1767 {
1768         struct numa_maps *md = walk->private;
1769         struct vm_area_struct *vma = walk->vma;
1770         spinlock_t *ptl;
1771         pte_t *orig_pte;
1772         pte_t *pte;
1773
1774 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1775         ptl = pmd_trans_huge_lock(pmd, vma);
1776         if (ptl) {
1777                 struct page *page;
1778
1779                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1780                 if (page)
1781                         gather_stats(page, md, pmd_dirty(*pmd),
1782                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1783                 spin_unlock(ptl);
1784                 return 0;
1785         }
1786
1787         if (pmd_trans_unstable(pmd))
1788                 return 0;
1789 #endif
1790         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1791         do {
1792                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1793                 if (!page)
1794                         continue;
1795                 gather_stats(page, md, pte_dirty(*pte), 1);
1796
1797         } while (pte++, addr += PAGE_SIZE, addr != end);
1798         pte_unmap_unlock(orig_pte, ptl);
1799         cond_resched();
1800         return 0;
1801 }
1802 #ifdef CONFIG_HUGETLB_PAGE
1803 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1804                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1805 {
1806         pte_t huge_pte = huge_ptep_get(pte);
1807         struct numa_maps *md;
1808         struct page *page;
1809
1810         if (!pte_present(huge_pte))
1811                 return 0;
1812
1813         page = pte_page(huge_pte);
1814         if (!page)
1815                 return 0;
1816
1817         md = walk->private;
1818         gather_stats(page, md, pte_dirty(huge_pte), 1);
1819         return 0;
1820 }
1821
1822 #else
1823 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1824                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1825 {
1826         return 0;
1827 }
1828 #endif
1829
1830 static const struct mm_walk_ops show_numa_ops = {
1831         .hugetlb_entry = gather_hugetlb_stats,
1832         .pmd_entry = gather_pte_stats,
1833 };
1834
1835 /*
1836  * Display pages allocated per node and memory policy via /proc.
1837  */
1838 static int show_numa_map(struct seq_file *m, void *v)
1839 {
1840         struct numa_maps_private *numa_priv = m->private;
1841         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1842         struct vm_area_struct *vma = v;
1843         struct numa_maps *md = &numa_priv->md;
1844         struct file *file = vma->vm_file;
1845         struct mm_struct *mm = vma->vm_mm;
1846         struct mempolicy *pol;
1847         char buffer[64];
1848         int nid;
1849
1850         if (!mm)
1851                 return 0;
1852
1853         /* Ensure we start with an empty set of numa_maps statistics. */
1854         memset(md, 0, sizeof(*md));
1855
1856         pol = __get_vma_policy(vma, vma->vm_start);
1857         if (pol) {
1858                 mpol_to_str(buffer, sizeof(buffer), pol);
1859                 mpol_cond_put(pol);
1860         } else {
1861                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1862         }
1863
1864         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1865
1866         if (file) {
1867                 seq_puts(m, " file=");
1868                 seq_file_path(m, file, "\n\t= ");
1869         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1870                 seq_puts(m, " heap");
1871         } else if (is_stack(vma)) {
1872                 seq_puts(m, " stack");
1873         }
1874
1875         if (is_vm_hugetlb_page(vma))
1876                 seq_puts(m, " huge");
1877
1878         /* mmap_lock is held by m_start */
1879         walk_page_vma(vma, &show_numa_ops, md);
1880
1881         if (!md->pages)
1882                 goto out;
1883
1884         if (md->anon)
1885                 seq_printf(m, " anon=%lu", md->anon);
1886
1887         if (md->dirty)
1888                 seq_printf(m, " dirty=%lu", md->dirty);
1889
1890         if (md->pages != md->anon && md->pages != md->dirty)
1891                 seq_printf(m, " mapped=%lu", md->pages);
1892
1893         if (md->mapcount_max > 1)
1894                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1895
1896         if (md->swapcache)
1897                 seq_printf(m, " swapcache=%lu", md->swapcache);
1898
1899         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1900                 seq_printf(m, " active=%lu", md->active);
1901
1902         if (md->writeback)
1903                 seq_printf(m, " writeback=%lu", md->writeback);
1904
1905         for_each_node_state(nid, N_MEMORY)
1906                 if (md->node[nid])
1907                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1908
1909         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1910 out:
1911         seq_putc(m, '\n');
1912         return 0;
1913 }
1914
1915 static const struct seq_operations proc_pid_numa_maps_op = {
1916         .start  = m_start,
1917         .next   = m_next,
1918         .stop   = m_stop,
1919         .show   = show_numa_map,
1920 };
1921
1922 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1923 {
1924         return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
1925                                 sizeof(struct numa_maps_private));
1926 }
1927
1928 const struct file_operations proc_pid_numa_maps_operations = {
1929         .open           = pid_numa_maps_open,
1930         .read           = seq_read,
1931         .llseek         = seq_lseek,
1932         .release        = proc_map_release,
1933 };
1934
1935 #endif /* CONFIG_NUMA */