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/mm_inline.h>
   5 #include <linux/hugetlb.h>
   6 #include <linux/huge_mm.h>
   7 #include <linux/mount.h>
   8 #include <linux/seq_file.h>
   9 #include <linux/highmem.h>
  10 #include <linux/ptrace.h>
  11 #include <linux/slab.h>
  12 #include <linux/pagemap.h>
  13 #include <linux/mempolicy.h>
  14 #include <linux/rmap.h>
  15 #include <linux/swap.h>
  16 #include <linux/sched/mm.h>
  17 #include <linux/swapops.h>
  18 #include <linux/mmu_notifier.h>
  19 #include <linux/page_idle.h>
  20 #include <linux/shmem_fs.h>
  21 #include <linux/uaccess.h>
  22 #include <linux/pkeys.h>
  23
  24 #include <asm/elf.h>
  25 #include <asm/tlb.h>
  26 #include <asm/tlbflush.h>
  27 #include "internal.h"
  28
  29 #define SEQ_PUT_DEC(str, val) \
  30                 seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8)
  31 void task_mem(struct seq_file *m, struct mm_struct *mm)
  32 {
  33         unsigned long text, lib, swap, anon, file, shmem;
  34         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  35
  36         anon = get_mm_counter(mm, MM_ANONPAGES);
  37         file = get_mm_counter(mm, MM_FILEPAGES);
  38         shmem = get_mm_counter(mm, MM_SHMEMPAGES);
  39
  40         /*
  41          * Note: to minimize their overhead, mm maintains hiwater_vm and
  42          * hiwater_rss only when about to *lower* total_vm or rss.  Any
  43          * collector of these hiwater stats must therefore get total_vm
  44          * and rss too, which will usually be the higher.  Barriers? not
  45          * worth the effort, such snapshots can always be inconsistent.
  46          */
  47         hiwater_vm = total_vm = mm->total_vm;
  48         if (hiwater_vm < mm->hiwater_vm)
  49                 hiwater_vm = mm->hiwater_vm;
  50         hiwater_rss = total_rss = anon + file + shmem;
  51         if (hiwater_rss < mm->hiwater_rss)
  52                 hiwater_rss = mm->hiwater_rss;
  53
  54         /* split executable areas between text and lib */
  55         text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK);
  56         text = min(text, mm->exec_vm << PAGE_SHIFT);
  57         lib = (mm->exec_vm << PAGE_SHIFT) - text;
  58
  59         swap = get_mm_counter(mm, MM_SWAPENTS);
  60         SEQ_PUT_DEC("VmPeak:\t", hiwater_vm);
  61         SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm);
  62         SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm);
  63         SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm->pinned_vm));
  64         SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss);
  65         SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss);
  66         SEQ_PUT_DEC(" kB\nRssAnon:\t", anon);
  67         SEQ_PUT_DEC(" kB\nRssFile:\t", file);
  68         SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem);
  69         SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm);
  70         SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm);
  71         seq_put_decimal_ull_width(m,
  72                     " kB\nVmExe:\t", text >> 10, 8);
  73         seq_put_decimal_ull_width(m,
  74                     " kB\nVmLib:\t", lib >> 10, 8);
  75         seq_put_decimal_ull_width(m,
  76                     " kB\nVmPTE:\t", mm_pgtables_bytes(mm) >> 10, 8);
  77         SEQ_PUT_DEC(" kB\nVmSwap:\t", swap);
  78         seq_puts(m, " kB\n");
  79         hugetlb_report_usage(m, mm);
  80 }
  81 #undef SEQ_PUT_DEC
  82
  83 unsigned long task_vsize(struct mm_struct *mm)
  84 {
  85         return PAGE_SIZE * mm->total_vm;
  86 }
  87
  88 unsigned long task_statm(struct mm_struct *mm,
  89                          unsigned long *shared, unsigned long *text,
  90                          unsigned long *data, unsigned long *resident)
  91 {
  92         *shared = get_mm_counter(mm, MM_FILEPAGES) +
  93                         get_mm_counter(mm, MM_SHMEMPAGES);
  94         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
  95                                                                 >> PAGE_SHIFT;
  96         *data = mm->data_vm + mm->stack_vm;
  97         *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
  98         return mm->total_vm;
  99 }
 100
 101 #ifdef CONFIG_NUMA
 102 /*
 103  * Save get_task_policy() for show_numa_map().
 104  */
 105 static void hold_task_mempolicy(struct proc_maps_private *priv)
 106 {
 107         struct task_struct *task = priv->task;
 108
 109         task_lock(task);
 110         priv->task_mempolicy = get_task_policy(task);
 111         mpol_get(priv->task_mempolicy);
 112         task_unlock(task);
 113 }
 114 static void release_task_mempolicy(struct proc_maps_private *priv)
 115 {
 116         mpol_put(priv->task_mempolicy);
 117 }
 118 #else
 119 static void hold_task_mempolicy(struct proc_maps_private *priv)
 120 {
 121 }
 122 static void release_task_mempolicy(struct proc_maps_private *priv)
 123 {
 124 }
 125 #endif
 126
 127 static void *m_start(struct seq_file *m, loff_t *ppos)
 128 {
 129         struct proc_maps_private *priv = m->private;
 130         unsigned long last_addr = *ppos;
 131         struct mm_struct *mm;
 132         struct vm_area_struct *vma;
 133
 134         /* See m_next(). Zero at the start or after lseek. */
 135         if (last_addr == -1UL)
 136                 return NULL;
 137
 138         priv->task = get_proc_task(priv->inode);
 139         if (!priv->task)
 140                 return ERR_PTR(-ESRCH);
 141
 142         mm = priv->mm;
 143         if (!mm || !mmget_not_zero(mm)) {
 144                 put_task_struct(priv->task);
 145                 priv->task = NULL;
 146                 return NULL;
 147         }
 148
 149         if (mmap_read_lock_killable(mm)) {
 150                 mmput(mm);
 151                 put_task_struct(priv->task);
 152                 priv->task = NULL;
 153                 return ERR_PTR(-EINTR);
 154         }
 155
 156         hold_task_mempolicy(priv);
 157         priv->tail_vma = get_gate_vma(mm);
 158
 159         vma = find_vma(mm, last_addr);
 160         if (vma)
 161                 return vma;
 162
 163         return priv->tail_vma;
 164 }
 165
 166 static void *m_next(struct seq_file *m, void *v, loff_t *ppos)
 167 {
 168         struct proc_maps_private *priv = m->private;
 169         struct vm_area_struct *next, *vma = v;
 170
 171         if (vma == priv->tail_vma)
 172                 next = NULL;
 173         else if (vma->vm_next)
 174                 next = vma->vm_next;
 175         else
 176                 next = priv->tail_vma;
 177
 178         *ppos = next ? next->vm_start : -1UL;
 179
 180         return next;
 181 }
 182
 183 static void m_stop(struct seq_file *m, void *v)
 184 {
 185         struct proc_maps_private *priv = m->private;
 186         struct mm_struct *mm = priv->mm;
 187
 188         if (!priv->task)
 189                 return;
 190
 191         release_task_mempolicy(priv);
 192         mmap_read_unlock(mm);
 193         mmput(mm);
 194         put_task_struct(priv->task);
 195         priv->task = NULL;
 196 }
 197
 198 static int proc_maps_open(struct inode *inode, struct file *file,
 199                         const struct seq_operations *ops, int psize)
 200 {
 201         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
 202
 203         if (!priv)
 204                 return -ENOMEM;
 205
 206         priv->inode = inode;
 207         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 208         if (IS_ERR(priv->mm)) {
 209                 int err = PTR_ERR(priv->mm);
 210
 211                 seq_release_private(inode, file);
 212                 return err;
 213         }
 214
 215         return 0;
 216 }
 217
 218 static int proc_map_release(struct inode *inode, struct file *file)
 219 {
 220         struct seq_file *seq = file->private_data;
 221         struct proc_maps_private *priv = seq->private;
 222
 223         if (priv->mm)
 224                 mmdrop(priv->mm);
 225
 226         return seq_release_private(inode, file);
 227 }
 228
 229 static int do_maps_open(struct inode *inode, struct file *file,
 230                         const struct seq_operations *ops)
 231 {
 232         return proc_maps_open(inode, file, ops,
 233                                 sizeof(struct proc_maps_private));
 234 }
 235
 236 /*
 237  * Indicate if the VMA is a stack for the given task; for
 238  * /proc/PID/maps that is the stack of the main task.
 239  */
 240 static int is_stack(struct vm_area_struct *vma)
 241 {
 242         /*
 243          * We make no effort to guess what a given thread considers to be
 244          * its "stack".  It's not even well-defined for programs written
 245          * languages like Go.
 246          */
 247         return vma->vm_start <= vma->vm_mm->start_stack &&
 248                 vma->vm_end >= vma->vm_mm->start_stack;
 249 }
 250
 251 static void show_vma_header_prefix(struct seq_file *m,
 252                                    unsigned long start, unsigned long end,
 253                                    vm_flags_t flags, unsigned long long pgoff,
 254                                    dev_t dev, unsigned long ino)
 255 {
 256         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 257         seq_put_hex_ll(m, NULL, start, 8);
 258         seq_put_hex_ll(m, "-", end, 8);
 259         seq_putc(m, ' ');
 260         seq_putc(m, flags & VM_READ ? 'r' : '-');
 261         seq_putc(m, flags & VM_WRITE ? 'w' : '-');
 262         seq_putc(m, flags & VM_EXEC ? 'x' : '-');
 263         seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
 264         seq_put_hex_ll(m, " ", pgoff, 8);
 265         seq_put_hex_ll(m, " ", MAJOR(dev), 2);
 266         seq_put_hex_ll(m, ":", MINOR(dev), 2);
 267         seq_put_decimal_ull(m, " ", ino);
 268         seq_putc(m, ' ');
 269 }
 270
 271 static void
 272 show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
 273 {
 274         struct mm_struct *mm = vma->vm_mm;
 275         struct file *file = vma->vm_file;
 276         vm_flags_t flags = vma->vm_flags;
 277         unsigned long ino = 0;
 278         unsigned long long pgoff = 0;
 279         unsigned long start, end;
 280         dev_t dev = 0;
 281         const char *name = NULL;
 282
 283         if (file) {
 284                 struct inode *inode = file_inode(vma->vm_file);
 285                 dev = inode->i_sb->s_dev;
 286                 ino = inode->i_ino;
 287                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 288         }
 289
 290         start = vma->vm_start;
 291         end = vma->vm_end;
 292         show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
 293
 294         /*
 295          * Print the dentry name for named mappings, and a
 296          * special [heap] marker for the heap:
 297          */
 298         if (file) {
 299                 seq_pad(m, ' ');
 300                 seq_file_path(m, file, "\n");
 301                 goto done;
 302         }
 303
 304         if (vma->vm_ops && vma->vm_ops->name) {
 305                 name = vma->vm_ops->name(vma);
 306                 if (name)
 307                         goto done;
 308         }
 309
 310         name = arch_vma_name(vma);
 311         if (!name) {
 312                 struct anon_vma_name *anon_name;
 313
 314                 if (!mm) {
 315                         name = "[vdso]";
 316                         goto done;
 317                 }
 318
 319                 if (vma->vm_start <= mm->brk &&
 320                     vma->vm_end >= mm->start_brk) {
 321                         name = "[heap]";
 322                         goto done;
 323                 }
 324
 325                 if (is_stack(vma)) {
 326                         name = "[stack]";
 327                         goto done;
 328                 }
 329
 330                 anon_name = anon_vma_name(vma);
 331                 if (anon_name) {
 332                         seq_pad(m, ' ');
 333                         seq_printf(m, "[anon:%s]", anon_name->name);
 334                 }
 335         }
 336
 337 done:
 338         if (name) {
 339                 seq_pad(m, ' ');
 340                 seq_puts(m, name);
 341         }
 342         seq_putc(m, '\n');
 343 }
 344
 345 static int show_map(struct seq_file *m, void *v)
 346 {
 347         show_map_vma(m, v);
 348         return 0;
 349 }
 350
 351 static const struct seq_operations proc_pid_maps_op = {
 352         .start  = m_start,
 353         .next   = m_next,
 354         .stop   = m_stop,
 355         .show   = show_map
 356 };
 357
 358 static int pid_maps_open(struct inode *inode, struct file *file)
 359 {
 360         return do_maps_open(inode, file, &proc_pid_maps_op);
 361 }
 362
 363 const struct file_operations proc_pid_maps_operations = {
 364         .open           = pid_maps_open,
 365         .read           = seq_read,
 366         .llseek         = seq_lseek,
 367         .release        = proc_map_release,
 368 };
 369
 370 /*
 371  * Proportional Set Size(PSS): my share of RSS.
 372  *
 373  * PSS of a process is the count of pages it has in memory, where each
 374  * page is divided by the number of processes sharing it.  So if a
 375  * process has 1000 pages all to itself, and 1000 shared with one other
 376  * process, its PSS will be 1500.
 377  *
 378  * To keep (accumulated) division errors low, we adopt a 64bit
 379  * fixed-point pss counter to minimize division errors. So (pss >>
 380  * PSS_SHIFT) would be the real byte count.
 381  *
 382  * A shift of 12 before division means (assuming 4K page size):
 383  *      - 1M 3-user-pages add up to 8KB errors;
 384  *      - supports mapcount up to 2^24, or 16M;
 385  *      - supports PSS up to 2^52 bytes, or 4PB.
 386  */
 387 #define PSS_SHIFT 12
 388
 389 #ifdef CONFIG_PROC_PAGE_MONITOR
 390 struct mem_size_stats {
 391         unsigned long resident;
 392         unsigned long shared_clean;
 393         unsigned long shared_dirty;
 394         unsigned long private_clean;
 395         unsigned long private_dirty;
 396         unsigned long referenced;
 397         unsigned long anonymous;
 398         unsigned long lazyfree;
 399         unsigned long anonymous_thp;
 400         unsigned long shmem_thp;
 401         unsigned long file_thp;
 402         unsigned long swap;
 403         unsigned long shared_hugetlb;
 404         unsigned long private_hugetlb;
 405         u64 pss;
 406         u64 pss_anon;
 407         u64 pss_file;
 408         u64 pss_shmem;
 409         u64 pss_dirty;
 410         u64 pss_locked;
 411         u64 swap_pss;
 412 };
 413
 414 static void smaps_page_accumulate(struct mem_size_stats *mss,
 415                 struct page *page, unsigned long size, unsigned long pss,
 416                 bool dirty, bool locked, bool private)
 417 {
 418         mss->pss += pss;
 419
 420         if (PageAnon(page))
 421                 mss->pss_anon += pss;
 422         else if (PageSwapBacked(page))
 423                 mss->pss_shmem += pss;
 424         else
 425                 mss->pss_file += pss;
 426
 427         if (locked)
 428                 mss->pss_locked += pss;
 429
 430         if (dirty || PageDirty(page)) {
 431                 mss->pss_dirty += pss;
 432                 if (private)
 433                         mss->private_dirty += size;
 434                 else
 435                         mss->shared_dirty += size;
 436         } else {
 437                 if (private)
 438                         mss->private_clean += size;
 439                 else
 440                         mss->shared_clean += size;
 441         }
 442 }
 443
 444 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 445                 bool compound, bool young, bool dirty, bool locked,
 446                 bool migration)
 447 {
 448         int i, nr = compound ? compound_nr(page) : 1;
 449         unsigned long size = nr * PAGE_SIZE;
 450
 451         /*
 452          * First accumulate quantities that depend only on |size| and the type
 453          * of the compound page.
 454          */
 455         if (PageAnon(page)) {
 456                 mss->anonymous += size;
 457                 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
 458                         mss->lazyfree += size;
 459         }
 460
 461         mss->resident += size;
 462         /* Accumulate the size in pages that have been accessed. */
 463         if (young || page_is_young(page) || PageReferenced(page))
 464                 mss->referenced += size;
 465
 466         /*
 467          * Then accumulate quantities that may depend on sharing, or that may
 468          * differ page-by-page.
 469          *
 470          * page_count(page) == 1 guarantees the page is mapped exactly once.
 471          * If any subpage of the compound page mapped with PTE it would elevate
 472          * page_count().
 473          *
 474          * The page_mapcount() is called to get a snapshot of the mapcount.
 475          * Without holding the page lock this snapshot can be slightly wrong as
 476          * we cannot always read the mapcount atomically.  It is not safe to
 477          * call page_mapcount() even with PTL held if the page is not mapped,
 478          * especially for migration entries.  Treat regular migration entries
 479          * as mapcount == 1.
 480          */
 481         if ((page_count(page) == 1) || migration) {
 482                 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
 483                         locked, true);
 484                 return;
 485         }
 486         for (i = 0; i < nr; i++, page++) {
 487                 int mapcount = page_mapcount(page);
 488                 unsigned long pss = PAGE_SIZE << PSS_SHIFT;
 489                 if (mapcount >= 2)
 490                         pss /= mapcount;
 491                 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
 492                                       mapcount < 2);
 493         }
 494 }
 495
 496 #ifdef CONFIG_SHMEM
 497 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 498                           __always_unused int depth, struct mm_walk *walk)
 499 {
 500         struct mem_size_stats *mss = walk->private;
 501         struct vm_area_struct *vma = walk->vma;
 502
 503         mss->swap += shmem_partial_swap_usage(walk->vma->vm_file->f_mapping,
 504                                               linear_page_index(vma, addr),
 505                                               linear_page_index(vma, end));
 506
 507         return 0;
 508 }
 509 #else
 510 #define smaps_pte_hole          NULL
 511 #endif /* CONFIG_SHMEM */
 512
 513 static void smaps_pte_hole_lookup(unsigned long addr, struct mm_walk *walk)
 514 {
 515 #ifdef CONFIG_SHMEM
 516         if (walk->ops->pte_hole) {
 517                 /* depth is not used */
 518                 smaps_pte_hole(addr, addr + PAGE_SIZE, 0, walk);
 519         }
 520 #endif
 521 }
 522
 523 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 524                 struct mm_walk *walk)
 525 {
 526         struct mem_size_stats *mss = walk->private;
 527         struct vm_area_struct *vma = walk->vma;
 528         bool locked = !!(vma->vm_flags & VM_LOCKED);
 529         struct page *page = NULL;
 530         bool migration = false, young = false, dirty = false;
 531
 532         if (pte_present(*pte)) {
 533                 page = vm_normal_page(vma, addr, *pte);
 534                 young = pte_young(*pte);
 535                 dirty = pte_dirty(*pte);
 536         } else if (is_swap_pte(*pte)) {
 537                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 538
 539                 if (!non_swap_entry(swpent)) {
 540                         int mapcount;
 541
 542                         mss->swap += PAGE_SIZE;
 543                         mapcount = swp_swapcount(swpent);
 544                         if (mapcount >= 2) {
 545                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 546
 547                                 do_div(pss_delta, mapcount);
 548                                 mss->swap_pss += pss_delta;
 549                         } else {
 550                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 551                         }
 552                 } else if (is_pfn_swap_entry(swpent)) {
 553                         if (is_migration_entry(swpent))
 554                                 migration = true;
 555                         page = pfn_swap_entry_to_page(swpent);
 556                 }
 557         } else {
 558                 smaps_pte_hole_lookup(addr, walk);
 559                 return;
 560         }
 561
 562         if (!page)
 563                 return;
 564
 565         smaps_account(mss, page, false, young, dirty, locked, migration);
 566 }
 567
 568 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 569 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 570                 struct mm_walk *walk)
 571 {
 572         struct mem_size_stats *mss = walk->private;
 573         struct vm_area_struct *vma = walk->vma;
 574         bool locked = !!(vma->vm_flags & VM_LOCKED);
 575         struct page *page = NULL;
 576         bool migration = false;
 577
 578         if (pmd_present(*pmd)) {
 579                 /* FOLL_DUMP will return -EFAULT on huge zero page */
 580                 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 581         } else if (unlikely(thp_migration_supported() && is_swap_pmd(*pmd))) {
 582                 swp_entry_t entry = pmd_to_swp_entry(*pmd);
 583
 584                 if (is_migration_entry(entry)) {
 585                         migration = true;
 586                         page = pfn_swap_entry_to_page(entry);
 587                 }
 588         }
 589         if (IS_ERR_OR_NULL(page))
 590                 return;
 591         if (PageAnon(page))
 592                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 593         else if (PageSwapBacked(page))
 594                 mss->shmem_thp += HPAGE_PMD_SIZE;
 595         else if (is_zone_device_page(page))
 596                 /* pass */;
 597         else
 598                 mss->file_thp += HPAGE_PMD_SIZE;
 599
 600         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd),
 601                       locked, migration);
 602 }
 603 #else
 604 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 605                 struct mm_walk *walk)
 606 {
 607 }
 608 #endif
 609
 610 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 611                            struct mm_walk *walk)
 612 {
 613         struct vm_area_struct *vma = walk->vma;
 614         pte_t *pte;
 615         spinlock_t *ptl;
 616
 617         ptl = pmd_trans_huge_lock(pmd, vma);
 618         if (ptl) {
 619                 smaps_pmd_entry(pmd, addr, walk);
 620                 spin_unlock(ptl);
 621                 goto out;
 622         }
 623
 624         if (pmd_trans_unstable(pmd))
 625                 goto out;
 626         /*
 627          * The mmap_lock held all the way back in m_start() is what
 628          * keeps khugepaged out of here and from collapsing things
 629          * in here.
 630          */
 631         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 632         for (; addr != end; pte++, addr += PAGE_SIZE)
 633                 smaps_pte_entry(pte, addr, walk);
 634         pte_unmap_unlock(pte - 1, ptl);
 635 out:
 636         cond_resched();
 637         return 0;
 638 }
 639
 640 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 641 {
 642         /*
 643          * Don't forget to update Documentation/ on changes.
 644          */
 645         static const char mnemonics[BITS_PER_LONG][2] = {
 646                 /*
 647                  * In case if we meet a flag we don't know about.
 648                  */
 649                 [0 ... (BITS_PER_LONG-1)] = "??",
 650
 651                 [ilog2(VM_READ)]        = "rd",
 652                 [ilog2(VM_WRITE)]       = "wr",
 653                 [ilog2(VM_EXEC)]        = "ex",
 654                 [ilog2(VM_SHARED)]      = "sh",
 655                 [ilog2(VM_MAYREAD)]     = "mr",
 656                 [ilog2(VM_MAYWRITE)]    = "mw",
 657                 [ilog2(VM_MAYEXEC)]     = "me",
 658                 [ilog2(VM_MAYSHARE)]    = "ms",
 659                 [ilog2(VM_GROWSDOWN)]   = "gd",
 660                 [ilog2(VM_PFNMAP)]      = "pf",
 661                 [ilog2(VM_LOCKED)]      = "lo",
 662                 [ilog2(VM_IO)]          = "io",
 663                 [ilog2(VM_SEQ_READ)]    = "sr",
 664                 [ilog2(VM_RAND_READ)]   = "rr",
 665                 [ilog2(VM_DONTCOPY)]    = "dc",
 666                 [ilog2(VM_DONTEXPAND)]  = "de",
 667                 [ilog2(VM_ACCOUNT)]     = "ac",
 668                 [ilog2(VM_NORESERVE)]   = "nr",
 669                 [ilog2(VM_HUGETLB)]     = "ht",
 670                 [ilog2(VM_SYNC)]        = "sf",
 671                 [ilog2(VM_ARCH_1)]      = "ar",
 672                 [ilog2(VM_WIPEONFORK)]  = "wf",
 673                 [ilog2(VM_DONTDUMP)]    = "dd",
 674 #ifdef CONFIG_ARM64_BTI
 675                 [ilog2(VM_ARM64_BTI)]   = "bt",
 676 #endif
 677 #ifdef CONFIG_MEM_SOFT_DIRTY
 678                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 679 #endif
 680                 [ilog2(VM_MIXEDMAP)]    = "mm",
 681                 [ilog2(VM_HUGEPAGE)]    = "hg",
 682                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 683                 [ilog2(VM_MERGEABLE)]   = "mg",
 684                 [ilog2(VM_UFFD_MISSING)]= "um",
 685                 [ilog2(VM_UFFD_WP)]     = "uw",
 686 #ifdef CONFIG_ARM64_MTE
 687                 [ilog2(VM_MTE)]         = "mt",
 688                 [ilog2(VM_MTE_ALLOWED)] = "",
 689 #endif
 690 #ifdef CONFIG_ARCH_HAS_PKEYS
 691                 /* These come out via ProtectionKey: */
 692                 [ilog2(VM_PKEY_BIT0)]   = "",
 693                 [ilog2(VM_PKEY_BIT1)]   = "",
 694                 [ilog2(VM_PKEY_BIT2)]   = "",
 695                 [ilog2(VM_PKEY_BIT3)]   = "",
 696 #if VM_PKEY_BIT4
 697                 [ilog2(VM_PKEY_BIT4)]   = "",
 698 #endif
 699 #endif /* CONFIG_ARCH_HAS_PKEYS */
 700 #ifdef CONFIG_HAVE_ARCH_USERFAULTFD_MINOR
 701                 [ilog2(VM_UFFD_MINOR)]  = "ui",
 702 #endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */
 703         };
 704         size_t i;
 705
 706         seq_puts(m, "VmFlags: ");
 707         for (i = 0; i < BITS_PER_LONG; i++) {
 708                 if (!mnemonics[i][0])
 709                         continue;
 710                 if (vma->vm_flags & (1UL << i)) {
 711                         seq_putc(m, mnemonics[i][0]);
 712                         seq_putc(m, mnemonics[i][1]);
 713                         seq_putc(m, ' ');
 714                 }
 715         }
 716         seq_putc(m, '\n');
 717 }
 718
 719 #ifdef CONFIG_HUGETLB_PAGE
 720 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 721                                  unsigned long addr, unsigned long end,
 722                                  struct mm_walk *walk)
 723 {
 724         struct mem_size_stats *mss = walk->private;
 725         struct vm_area_struct *vma = walk->vma;
 726         struct page *page = NULL;
 727
 728         if (pte_present(*pte)) {
 729                 page = vm_normal_page(vma, addr, *pte);
 730         } else if (is_swap_pte(*pte)) {
 731                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 732
 733                 if (is_pfn_swap_entry(swpent))
 734                         page = pfn_swap_entry_to_page(swpent);
 735         }
 736         if (page) {
 737                 int mapcount = page_mapcount(page);
 738
 739                 if (mapcount >= 2)
 740                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 741                 else
 742                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 743         }
 744         return 0;
 745 }
 746 #else
 747 #define smaps_hugetlb_range     NULL
 748 #endif /* HUGETLB_PAGE */
 749
 750 static const struct mm_walk_ops smaps_walk_ops = {
 751         .pmd_entry              = smaps_pte_range,
 752         .hugetlb_entry          = smaps_hugetlb_range,
 753 };
 754
 755 static const struct mm_walk_ops smaps_shmem_walk_ops = {
 756         .pmd_entry              = smaps_pte_range,
 757         .hugetlb_entry          = smaps_hugetlb_range,
 758         .pte_hole               = smaps_pte_hole,
 759 };
 760
 761 /*
 762  * Gather mem stats from @vma with the indicated beginning
 763  * address @start, and keep them in @mss.
 764  *
 765  * Use vm_start of @vma as the beginning address if @start is 0.
 766  */
 767 static void smap_gather_stats(struct vm_area_struct *vma,
 768                 struct mem_size_stats *mss, unsigned long start)
 769 {
 770         const struct mm_walk_ops *ops = &smaps_walk_ops;
 771
 772         /* Invalid start */
 773         if (start >= vma->vm_end)
 774                 return;
 775
 776 #ifdef CONFIG_SHMEM
 777         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 778                 /*
 779                  * For shared or readonly shmem mappings we know that all
 780                  * swapped out pages belong to the shmem object, and we can
 781                  * obtain the swap value much more efficiently. For private
 782                  * writable mappings, we might have COW pages that are
 783                  * not affected by the parent swapped out pages of the shmem
 784                  * object, so we have to distinguish them during the page walk.
 785                  * Unless we know that the shmem object (or the part mapped by
 786                  * our VMA) has no swapped out pages at all.
 787                  */
 788                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 789
 790                 if (!start && (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 791                                         !(vma->vm_flags & VM_WRITE))) {
 792                         mss->swap += shmem_swapped;
 793                 } else {
 794                         ops = &smaps_shmem_walk_ops;
 795                 }
 796         }
 797 #endif
 798         /* mmap_lock is held in m_start */
 799         if (!start)
 800                 walk_page_vma(vma, ops, mss);
 801         else
 802                 walk_page_range(vma->vm_mm, start, vma->vm_end, ops, mss);
 803 }
 804
 805 #define SEQ_PUT_DEC(str, val) \
 806                 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
 807
 808 /* Show the contents common for smaps and smaps_rollup */
 809 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
 810         bool rollup_mode)
 811 {
 812         SEQ_PUT_DEC("Rss:            ", mss->resident);
 813         SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
 814         SEQ_PUT_DEC(" kB\nPss_Dirty:      ", mss->pss_dirty >> PSS_SHIFT);
 815         if (rollup_mode) {
 816                 /*
 817                  * These are meaningful only for smaps_rollup, otherwise two of
 818                  * them are zero, and the other one is the same as Pss.
 819                  */
 820                 SEQ_PUT_DEC(" kB\nPss_Anon:       ",
 821                         mss->pss_anon >> PSS_SHIFT);
 822                 SEQ_PUT_DEC(" kB\nPss_File:       ",
 823                         mss->pss_file >> PSS_SHIFT);
 824                 SEQ_PUT_DEC(" kB\nPss_Shmem:      ",
 825                         mss->pss_shmem >> PSS_SHIFT);
 826         }
 827         SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
 828         SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
 829         SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
 830         SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
 831         SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
 832         SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
 833         SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
 834         SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
 835         SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
 836         SEQ_PUT_DEC(" kB\nFilePmdMapped:  ", mss->file_thp);
 837         SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
 838         seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
 839                                   mss->private_hugetlb >> 10, 7);
 840         SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
 841         SEQ_PUT_DEC(" kB\nSwapPss:        ",
 842                                         mss->swap_pss >> PSS_SHIFT);
 843         SEQ_PUT_DEC(" kB\nLocked:         ",
 844                                         mss->pss_locked >> PSS_SHIFT);
 845         seq_puts(m, " kB\n");
 846 }
 847
 848 static int show_smap(struct seq_file *m, void *v)
 849 {
 850         struct vm_area_struct *vma = v;
 851         struct mem_size_stats mss;
 852
 853         memset(&mss, 0, sizeof(mss));
 854
 855         smap_gather_stats(vma, &mss, 0);
 856
 857         show_map_vma(m, vma);
 858
 859         SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
 860         SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
 861         SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
 862         seq_puts(m, " kB\n");
 863
 864         __show_smap(m, &mss, false);
 865
 866         seq_printf(m, "THPeligible:    %d\n",
 867                    hugepage_vma_check(vma, vma->vm_flags, true, false));
 868
 869         if (arch_pkeys_enabled())
 870                 seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
 871         show_smap_vma_flags(m, vma);
 872
 873         return 0;
 874 }
 875
 876 static int show_smaps_rollup(struct seq_file *m, void *v)
 877 {
 878         struct proc_maps_private *priv = m->private;
 879         struct mem_size_stats mss;
 880         struct mm_struct *mm;
 881         struct vm_area_struct *vma;
 882         unsigned long last_vma_end = 0;
 883         int ret = 0;
 884
 885         priv->task = get_proc_task(priv->inode);
 886         if (!priv->task)
 887                 return -ESRCH;
 888
 889         mm = priv->mm;
 890         if (!mm || !mmget_not_zero(mm)) {
 891                 ret = -ESRCH;
 892                 goto out_put_task;
 893         }
 894
 895         memset(&mss, 0, sizeof(mss));
 896
 897         ret = mmap_read_lock_killable(mm);
 898         if (ret)
 899                 goto out_put_mm;
 900
 901         hold_task_mempolicy(priv);
 902
 903         for (vma = priv->mm->mmap; vma;) {
 904                 smap_gather_stats(vma, &mss, 0);
 905                 last_vma_end = vma->vm_end;
 906
 907                 /*
 908                  * Release mmap_lock temporarily if someone wants to
 909                  * access it for write request.
 910                  */
 911                 if (mmap_lock_is_contended(mm)) {
 912                         mmap_read_unlock(mm);
 913                         ret = mmap_read_lock_killable(mm);
 914                         if (ret) {
 915                                 release_task_mempolicy(priv);
 916                                 goto out_put_mm;
 917                         }
 918
 919                         /*
 920                          * After dropping the lock, there are four cases to
 921                          * consider. See the following example for explanation.
 922                          *
 923                          *   +------+------+-----------+
 924                          *   | VMA1 | VMA2 | VMA3      |
 925                          *   +------+------+-----------+
 926                          *   |      |      |           |
 927                          *  4k     8k     16k         400k
 928                          *
 929                          * Suppose we drop the lock after reading VMA2 due to
 930                          * contention, then we get:
 931                          *
 932                          *      last_vma_end = 16k
 933                          *
 934                          * 1) VMA2 is freed, but VMA3 exists:
 935                          *
 936                          *    find_vma(mm, 16k - 1) will return VMA3.
 937                          *    In this case, just continue from VMA3.
 938                          *
 939                          * 2) VMA2 still exists:
 940                          *
 941                          *    find_vma(mm, 16k - 1) will return VMA2.
 942                          *    Iterate the loop like the original one.
 943                          *
 944                          * 3) No more VMAs can be found:
 945                          *
 946                          *    find_vma(mm, 16k - 1) will return NULL.
 947                          *    No more things to do, just break.
 948                          *
 949                          * 4) (last_vma_end - 1) is the middle of a vma (VMA'):
 950                          *
 951                          *    find_vma(mm, 16k - 1) will return VMA' whose range
 952                          *    contains last_vma_end.
 953                          *    Iterate VMA' from last_vma_end.
 954                          */
 955                         vma = find_vma(mm, last_vma_end - 1);
 956                         /* Case 3 above */
 957                         if (!vma)
 958                                 break;
 959
 960                         /* Case 1 above */
 961                         if (vma->vm_start >= last_vma_end)
 962                                 continue;
 963
 964                         /* Case 4 above */
 965                         if (vma->vm_end > last_vma_end)
 966                                 smap_gather_stats(vma, &mss, last_vma_end);
 967                 }
 968                 /* Case 2 above */
 969                 vma = vma->vm_next;
 970         }
 971
 972         show_vma_header_prefix(m, priv->mm->mmap->vm_start,
 973                                last_vma_end, 0, 0, 0, 0);
 974         seq_pad(m, ' ');
 975         seq_puts(m, "[rollup]\n");
 976
 977         __show_smap(m, &mss, true);
 978
 979         release_task_mempolicy(priv);
 980         mmap_read_unlock(mm);
 981
 982 out_put_mm:
 983         mmput(mm);
 984 out_put_task:
 985         put_task_struct(priv->task);
 986         priv->task = NULL;
 987
 988         return ret;
 989 }
 990 #undef SEQ_PUT_DEC
 991
 992 static const struct seq_operations proc_pid_smaps_op = {
 993         .start  = m_start,
 994         .next   = m_next,
 995         .stop   = m_stop,
 996         .show   = show_smap
 997 };
 998
 999 static int pid_smaps_open(struct inode *inode, struct file *file)
1000 {
1001         return do_maps_open(inode, file, &proc_pid_smaps_op);
1002 }
1003
1004 static int smaps_rollup_open(struct inode *inode, struct file *file)
1005 {
1006         int ret;
1007         struct proc_maps_private *priv;
1008
1009         priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
1010         if (!priv)
1011                 return -ENOMEM;
1012
1013         ret = single_open(file, show_smaps_rollup, priv);
1014         if (ret)
1015                 goto out_free;
1016
1017         priv->inode = inode;
1018         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
1019         if (IS_ERR(priv->mm)) {
1020                 ret = PTR_ERR(priv->mm);
1021
1022                 single_release(inode, file);
1023                 goto out_free;
1024         }
1025
1026         return 0;
1027
1028 out_free:
1029         kfree(priv);
1030         return ret;
1031 }
1032
1033 static int smaps_rollup_release(struct inode *inode, struct file *file)
1034 {
1035         struct seq_file *seq = file->private_data;
1036         struct proc_maps_private *priv = seq->private;
1037
1038         if (priv->mm)
1039                 mmdrop(priv->mm);
1040
1041         kfree(priv);
1042         return single_release(inode, file);
1043 }
1044
1045 const struct file_operations proc_pid_smaps_operations = {
1046         .open           = pid_smaps_open,
1047         .read           = seq_read,
1048         .llseek         = seq_lseek,
1049         .release        = proc_map_release,
1050 };
1051
1052 const struct file_operations proc_pid_smaps_rollup_operations = {
1053         .open           = smaps_rollup_open,
1054         .read           = seq_read,
1055         .llseek         = seq_lseek,
1056         .release        = smaps_rollup_release,
1057 };
1058
1059 enum clear_refs_types {
1060         CLEAR_REFS_ALL = 1,
1061         CLEAR_REFS_ANON,
1062         CLEAR_REFS_MAPPED,
1063         CLEAR_REFS_SOFT_DIRTY,
1064         CLEAR_REFS_MM_HIWATER_RSS,
1065         CLEAR_REFS_LAST,
1066 };
1067
1068 struct clear_refs_private {
1069         enum clear_refs_types type;
1070 };
1071
1072 #ifdef CONFIG_MEM_SOFT_DIRTY
1073
1074 static inline bool pte_is_pinned(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1075 {
1076         struct page *page;
1077
1078         if (!pte_write(pte))
1079                 return false;
1080         if (!is_cow_mapping(vma->vm_flags))
1081                 return false;
1082         if (likely(!test_bit(MMF_HAS_PINNED, &vma->vm_mm->flags)))
1083                 return false;
1084         page = vm_normal_page(vma, addr, pte);
1085         if (!page)
1086                 return false;
1087         return page_maybe_dma_pinned(page);
1088 }
1089
1090 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1091                 unsigned long addr, pte_t *pte)
1092 {
1093         /*
1094          * The soft-dirty tracker uses #PF-s to catch writes
1095          * to pages, so write-protect the pte as well. See the
1096          * Documentation/admin-guide/mm/soft-dirty.rst for full description
1097          * of how soft-dirty works.
1098          */
1099         pte_t ptent = *pte;
1100
1101         if (pte_present(ptent)) {
1102                 pte_t old_pte;
1103
1104                 if (pte_is_pinned(vma, addr, ptent))
1105                         return;
1106                 old_pte = ptep_modify_prot_start(vma, addr, pte);
1107                 ptent = pte_wrprotect(old_pte);
1108                 ptent = pte_clear_soft_dirty(ptent);
1109                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1110         } else if (is_swap_pte(ptent)) {
1111                 ptent = pte_swp_clear_soft_dirty(ptent);
1112                 set_pte_at(vma->vm_mm, addr, pte, ptent);
1113         }
1114 }
1115 #else
1116 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1117                 unsigned long addr, pte_t *pte)
1118 {
1119 }
1120 #endif
1121
1122 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1123 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1124                 unsigned long addr, pmd_t *pmdp)
1125 {
1126         pmd_t old, pmd = *pmdp;
1127
1128         if (pmd_present(pmd)) {
1129                 /* See comment in change_huge_pmd() */
1130                 old = pmdp_invalidate(vma, addr, pmdp);
1131                 if (pmd_dirty(old))
1132                         pmd = pmd_mkdirty(pmd);
1133                 if (pmd_young(old))
1134                         pmd = pmd_mkyoung(pmd);
1135
1136                 pmd = pmd_wrprotect(pmd);
1137                 pmd = pmd_clear_soft_dirty(pmd);
1138
1139                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1140         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1141                 pmd = pmd_swp_clear_soft_dirty(pmd);
1142                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1143         }
1144 }
1145 #else
1146 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1147                 unsigned long addr, pmd_t *pmdp)
1148 {
1149 }
1150 #endif
1151
1152 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1153                                 unsigned long end, struct mm_walk *walk)
1154 {
1155         struct clear_refs_private *cp = walk->private;
1156         struct vm_area_struct *vma = walk->vma;
1157         pte_t *pte, ptent;
1158         spinlock_t *ptl;
1159         struct page *page;
1160
1161         ptl = pmd_trans_huge_lock(pmd, vma);
1162         if (ptl) {
1163                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1164                         clear_soft_dirty_pmd(vma, addr, pmd);
1165                         goto out;
1166                 }
1167
1168                 if (!pmd_present(*pmd))
1169                         goto out;
1170
1171                 page = pmd_page(*pmd);
1172
1173                 /* Clear accessed and referenced bits. */
1174                 pmdp_test_and_clear_young(vma, addr, pmd);
1175                 test_and_clear_page_young(page);
1176                 ClearPageReferenced(page);
1177 out:
1178                 spin_unlock(ptl);
1179                 return 0;
1180         }
1181
1182         if (pmd_trans_unstable(pmd))
1183                 return 0;
1184
1185         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1186         for (; addr != end; pte++, addr += PAGE_SIZE) {
1187                 ptent = *pte;
1188
1189                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1190                         clear_soft_dirty(vma, addr, pte);
1191                         continue;
1192                 }
1193
1194                 if (!pte_present(ptent))
1195                         continue;
1196
1197                 page = vm_normal_page(vma, addr, ptent);
1198                 if (!page)
1199                         continue;
1200
1201                 /* Clear accessed and referenced bits. */
1202                 ptep_test_and_clear_young(vma, addr, pte);
1203                 test_and_clear_page_young(page);
1204                 ClearPageReferenced(page);
1205         }
1206         pte_unmap_unlock(pte - 1, ptl);
1207         cond_resched();
1208         return 0;
1209 }
1210
1211 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1212                                 struct mm_walk *walk)
1213 {
1214         struct clear_refs_private *cp = walk->private;
1215         struct vm_area_struct *vma = walk->vma;
1216
1217         if (vma->vm_flags & VM_PFNMAP)
1218                 return 1;
1219
1220         /*
1221          * Writing 1 to /proc/pid/clear_refs affects all pages.
1222          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1223          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1224          * Writing 4 to /proc/pid/clear_refs affects all pages.
1225          */
1226         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1227                 return 1;
1228         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1229                 return 1;
1230         return 0;
1231 }
1232
1233 static const struct mm_walk_ops clear_refs_walk_ops = {
1234         .pmd_entry              = clear_refs_pte_range,
1235         .test_walk              = clear_refs_test_walk,
1236 };
1237
1238 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1239                                 size_t count, loff_t *ppos)
1240 {
1241         struct task_struct *task;
1242         char buffer[PROC_NUMBUF];
1243         struct mm_struct *mm;
1244         struct vm_area_struct *vma;
1245         enum clear_refs_types type;
1246         int itype;
1247         int rv;
1248
1249         memset(buffer, 0, sizeof(buffer));
1250         if (count > sizeof(buffer) - 1)
1251                 count = sizeof(buffer) - 1;
1252         if (copy_from_user(buffer, buf, count))
1253                 return -EFAULT;
1254         rv = kstrtoint(strstrip(buffer), 10, &itype);
1255         if (rv < 0)
1256                 return rv;
1257         type = (enum clear_refs_types)itype;
1258         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1259                 return -EINVAL;
1260
1261         task = get_proc_task(file_inode(file));
1262         if (!task)
1263                 return -ESRCH;
1264         mm = get_task_mm(task);
1265         if (mm) {
1266                 struct mmu_notifier_range range;
1267                 struct clear_refs_private cp = {
1268                         .type = type,
1269                 };
1270
1271                 if (mmap_write_lock_killable(mm)) {
1272                         count = -EINTR;
1273                         goto out_mm;
1274                 }
1275                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1276                         /*
1277                          * Writing 5 to /proc/pid/clear_refs resets the peak
1278                          * resident set size to this mm's current rss value.
1279                          */
1280                         reset_mm_hiwater_rss(mm);
1281                         goto out_unlock;
1282                 }
1283
1284                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1285                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1286                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1287                                         continue;
1288                                 vma->vm_flags &= ~VM_SOFTDIRTY;
1289                                 vma_set_page_prot(vma);
1290                         }
1291
1292                         inc_tlb_flush_pending(mm);
1293                         mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1294                                                 0, NULL, mm, 0, -1UL);
1295                         mmu_notifier_invalidate_range_start(&range);
1296                 }
1297                 walk_page_range(mm, 0, mm->highest_vm_end, &clear_refs_walk_ops,
1298                                 &cp);
1299                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1300                         mmu_notifier_invalidate_range_end(&range);
1301                         flush_tlb_mm(mm);
1302                         dec_tlb_flush_pending(mm);
1303                 }
1304 out_unlock:
1305                 mmap_write_unlock(mm);
1306 out_mm:
1307                 mmput(mm);
1308         }
1309         put_task_struct(task);
1310
1311         return count;
1312 }
1313
1314 const struct file_operations proc_clear_refs_operations = {
1315         .write          = clear_refs_write,
1316         .llseek         = noop_llseek,
1317 };
1318
1319 typedef struct {
1320         u64 pme;
1321 } pagemap_entry_t;
1322
1323 struct pagemapread {
1324         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1325         pagemap_entry_t *buffer;
1326         bool show_pfn;
1327 };
1328
1329 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1330 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1331
1332 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1333 #define PM_PFRAME_BITS          55
1334 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1335 #define PM_SOFT_DIRTY           BIT_ULL(55)
1336 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1337 #define PM_UFFD_WP              BIT_ULL(57)
1338 #define PM_FILE                 BIT_ULL(61)
1339 #define PM_SWAP                 BIT_ULL(62)
1340 #define PM_PRESENT              BIT_ULL(63)
1341
1342 #define PM_END_OF_BUFFER    1
1343
1344 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1345 {
1346         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1347 }
1348
1349 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1350                           struct pagemapread *pm)
1351 {
1352         pm->buffer[pm->pos++] = *pme;
1353         if (pm->pos >= pm->len)
1354                 return PM_END_OF_BUFFER;
1355         return 0;
1356 }
1357
1358 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1359                             __always_unused int depth, struct mm_walk *walk)
1360 {
1361         struct pagemapread *pm = walk->private;
1362         unsigned long addr = start;
1363         int err = 0;
1364
1365         while (addr < end) {
1366                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1367                 pagemap_entry_t pme = make_pme(0, 0);
1368                 /* End of address space hole, which we mark as non-present. */
1369                 unsigned long hole_end;
1370
1371                 if (vma)
1372                         hole_end = min(end, vma->vm_start);
1373                 else
1374                         hole_end = end;
1375
1376                 for (; addr < hole_end; addr += PAGE_SIZE) {
1377                         err = add_to_pagemap(addr, &pme, pm);
1378                         if (err)
1379                                 goto out;
1380                 }
1381
1382                 if (!vma)
1383                         break;
1384
1385                 /* Addresses in the VMA. */
1386                 if (vma->vm_flags & VM_SOFTDIRTY)
1387                         pme = make_pme(0, PM_SOFT_DIRTY);
1388                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1389                         err = add_to_pagemap(addr, &pme, pm);
1390                         if (err)
1391                                 goto out;
1392                 }
1393         }
1394 out:
1395         return err;
1396 }
1397
1398 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1399                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1400 {
1401         u64 frame = 0, flags = 0;
1402         struct page *page = NULL;
1403         bool migration = false;
1404
1405         if (pte_present(pte)) {
1406                 if (pm->show_pfn)
1407                         frame = pte_pfn(pte);
1408                 flags |= PM_PRESENT;
1409                 page = vm_normal_page(vma, addr, pte);
1410                 if (pte_soft_dirty(pte))
1411                         flags |= PM_SOFT_DIRTY;
1412                 if (pte_uffd_wp(pte))
1413                         flags |= PM_UFFD_WP;
1414         } else if (is_swap_pte(pte)) {
1415                 swp_entry_t entry;
1416                 if (pte_swp_soft_dirty(pte))
1417                         flags |= PM_SOFT_DIRTY;
1418                 if (pte_swp_uffd_wp(pte))
1419                         flags |= PM_UFFD_WP;
1420                 entry = pte_to_swp_entry(pte);
1421                 if (pm->show_pfn)
1422                         frame = swp_type(entry) |
1423                                 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1424                 flags |= PM_SWAP;
1425                 migration = is_migration_entry(entry);
1426                 if (is_pfn_swap_entry(entry))
1427                         page = pfn_swap_entry_to_page(entry);
1428                 if (pte_marker_entry_uffd_wp(entry))
1429                         flags |= PM_UFFD_WP;
1430         }
1431
1432         if (page && !PageAnon(page))
1433                 flags |= PM_FILE;
1434         if (page && !migration && page_mapcount(page) == 1)
1435                 flags |= PM_MMAP_EXCLUSIVE;
1436         if (vma->vm_flags & VM_SOFTDIRTY)
1437                 flags |= PM_SOFT_DIRTY;
1438
1439         return make_pme(frame, flags);
1440 }
1441
1442 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1443                              struct mm_walk *walk)
1444 {
1445         struct vm_area_struct *vma = walk->vma;
1446         struct pagemapread *pm = walk->private;
1447         spinlock_t *ptl;
1448         pte_t *pte, *orig_pte;
1449         int err = 0;
1450 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1451         bool migration = false;
1452
1453         ptl = pmd_trans_huge_lock(pmdp, vma);
1454         if (ptl) {
1455                 u64 flags = 0, frame = 0;
1456                 pmd_t pmd = *pmdp;
1457                 struct page *page = NULL;
1458
1459                 if (vma->vm_flags & VM_SOFTDIRTY)
1460                         flags |= PM_SOFT_DIRTY;
1461
1462                 if (pmd_present(pmd)) {
1463                         page = pmd_page(pmd);
1464
1465                         flags |= PM_PRESENT;
1466                         if (pmd_soft_dirty(pmd))
1467                                 flags |= PM_SOFT_DIRTY;
1468                         if (pmd_uffd_wp(pmd))
1469                                 flags |= PM_UFFD_WP;
1470                         if (pm->show_pfn)
1471                                 frame = pmd_pfn(pmd) +
1472                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1473                 }
1474 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1475                 else if (is_swap_pmd(pmd)) {
1476                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1477                         unsigned long offset;
1478
1479                         if (pm->show_pfn) {
1480                                 offset = swp_offset(entry) +
1481                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1482                                 frame = swp_type(entry) |
1483                                         (offset << MAX_SWAPFILES_SHIFT);
1484                         }
1485                         flags |= PM_SWAP;
1486                         if (pmd_swp_soft_dirty(pmd))
1487                                 flags |= PM_SOFT_DIRTY;
1488                         if (pmd_swp_uffd_wp(pmd))
1489                                 flags |= PM_UFFD_WP;
1490                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1491                         migration = is_migration_entry(entry);
1492                         page = pfn_swap_entry_to_page(entry);
1493                 }
1494 #endif
1495
1496                 if (page && !migration && page_mapcount(page) == 1)
1497                         flags |= PM_MMAP_EXCLUSIVE;
1498
1499                 for (; addr != end; addr += PAGE_SIZE) {
1500                         pagemap_entry_t pme = make_pme(frame, flags);
1501
1502                         err = add_to_pagemap(addr, &pme, pm);
1503                         if (err)
1504                                 break;
1505                         if (pm->show_pfn) {
1506                                 if (flags & PM_PRESENT)
1507                                         frame++;
1508                                 else if (flags & PM_SWAP)
1509                                         frame += (1 << MAX_SWAPFILES_SHIFT);
1510                         }
1511                 }
1512                 spin_unlock(ptl);
1513                 return err;
1514         }
1515
1516         if (pmd_trans_unstable(pmdp))
1517                 return 0;
1518 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1519
1520         /*
1521          * We can assume that @vma always points to a valid one and @end never
1522          * goes beyond vma->vm_end.
1523          */
1524         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1525         for (; addr < end; pte++, addr += PAGE_SIZE) {
1526                 pagemap_entry_t pme;
1527
1528                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1529                 err = add_to_pagemap(addr, &pme, pm);
1530                 if (err)
1531                         break;
1532         }
1533         pte_unmap_unlock(orig_pte, ptl);
1534
1535         cond_resched();
1536
1537         return err;
1538 }
1539
1540 #ifdef CONFIG_HUGETLB_PAGE
1541 /* This function walks within one hugetlb entry in the single call */
1542 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1543                                  unsigned long addr, unsigned long end,
1544                                  struct mm_walk *walk)
1545 {
1546         struct pagemapread *pm = walk->private;
1547         struct vm_area_struct *vma = walk->vma;
1548         u64 flags = 0, frame = 0;
1549         int err = 0;
1550         pte_t pte;
1551
1552         if (vma->vm_flags & VM_SOFTDIRTY)
1553                 flags |= PM_SOFT_DIRTY;
1554
1555         pte = huge_ptep_get(ptep);
1556         if (pte_present(pte)) {
1557                 struct page *page = pte_page(pte);
1558
1559                 if (!PageAnon(page))
1560                         flags |= PM_FILE;
1561
1562                 if (page_mapcount(page) == 1)
1563                         flags |= PM_MMAP_EXCLUSIVE;
1564
1565                 if (huge_pte_uffd_wp(pte))
1566                         flags |= PM_UFFD_WP;
1567
1568                 flags |= PM_PRESENT;
1569                 if (pm->show_pfn)
1570                         frame = pte_pfn(pte) +
1571                                 ((addr & ~hmask) >> PAGE_SHIFT);
1572         } else if (pte_swp_uffd_wp_any(pte)) {
1573                 flags |= PM_UFFD_WP;
1574         }
1575
1576         for (; addr != end; addr += PAGE_SIZE) {
1577                 pagemap_entry_t pme = make_pme(frame, flags);
1578
1579                 err = add_to_pagemap(addr, &pme, pm);
1580                 if (err)
1581                         return err;
1582                 if (pm->show_pfn && (flags & PM_PRESENT))
1583                         frame++;
1584         }
1585
1586         cond_resched();
1587
1588         return err;
1589 }
1590 #else
1591 #define pagemap_hugetlb_range   NULL
1592 #endif /* HUGETLB_PAGE */
1593
1594 static const struct mm_walk_ops pagemap_ops = {
1595         .pmd_entry      = pagemap_pmd_range,
1596         .pte_hole       = pagemap_pte_hole,
1597         .hugetlb_entry  = pagemap_hugetlb_range,
1598 };
1599
1600 /*
1601  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1602  *
1603  * For each page in the address space, this file contains one 64-bit entry
1604  * consisting of the following:
1605  *
1606  * Bits 0-54  page frame number (PFN) if present
1607  * Bits 0-4   swap type if swapped
1608  * Bits 5-54  swap offset if swapped
1609  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1610  * Bit  56    page exclusively mapped
1611  * Bit  57    pte is uffd-wp write-protected
1612  * Bits 58-60 zero
1613  * Bit  61    page is file-page or shared-anon
1614  * Bit  62    page swapped
1615  * Bit  63    page present
1616  *
1617  * If the page is not present but in swap, then the PFN contains an
1618  * encoding of the swap file number and the page's offset into the
1619  * swap. Unmapped pages return a null PFN. This allows determining
1620  * precisely which pages are mapped (or in swap) and comparing mapped
1621  * pages between processes.
1622  *
1623  * Efficient users of this interface will use /proc/pid/maps to
1624  * determine which areas of memory are actually mapped and llseek to
1625  * skip over unmapped regions.
1626  */
1627 static ssize_t pagemap_read(struct file *file, char __user *buf,
1628                             size_t count, loff_t *ppos)
1629 {
1630         struct mm_struct *mm = file->private_data;
1631         struct pagemapread pm;
1632         unsigned long src;
1633         unsigned long svpfn;
1634         unsigned long start_vaddr;
1635         unsigned long end_vaddr;
1636         int ret = 0, copied = 0;
1637
1638         if (!mm || !mmget_not_zero(mm))
1639                 goto out;
1640
1641         ret = -EINVAL;
1642         /* file position must be aligned */
1643         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1644                 goto out_mm;
1645
1646         ret = 0;
1647         if (!count)
1648                 goto out_mm;
1649
1650         /* do not disclose physical addresses: attack vector */
1651         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1652
1653         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1654         pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1655         ret = -ENOMEM;
1656         if (!pm.buffer)
1657                 goto out_mm;
1658
1659         src = *ppos;
1660         svpfn = src / PM_ENTRY_BYTES;
1661         end_vaddr = mm->task_size;
1662
1663         /* watch out for wraparound */
1664         start_vaddr = end_vaddr;
1665         if (svpfn <= (ULONG_MAX >> PAGE_SHIFT))
1666                 start_vaddr = untagged_addr(svpfn << PAGE_SHIFT);
1667
1668         /* Ensure the address is inside the task */
1669         if (start_vaddr > mm->task_size)
1670                 start_vaddr = end_vaddr;
1671
1672         /*
1673          * The odds are that this will stop walking way
1674          * before end_vaddr, because the length of the
1675          * user buffer is tracked in "pm", and the walk
1676          * will stop when we hit the end of the buffer.
1677          */
1678         ret = 0;
1679         while (count && (start_vaddr < end_vaddr)) {
1680                 int len;
1681                 unsigned long end;
1682
1683                 pm.pos = 0;
1684                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1685                 /* overflow ? */
1686                 if (end < start_vaddr || end > end_vaddr)
1687                         end = end_vaddr;
1688                 ret = mmap_read_lock_killable(mm);
1689                 if (ret)
1690                         goto out_free;
1691                 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1692                 mmap_read_unlock(mm);
1693                 start_vaddr = end;
1694
1695                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1696                 if (copy_to_user(buf, pm.buffer, len)) {
1697                         ret = -EFAULT;
1698                         goto out_free;
1699                 }
1700                 copied += len;
1701                 buf += len;
1702                 count -= len;
1703         }
1704         *ppos += copied;
1705         if (!ret || ret == PM_END_OF_BUFFER)
1706                 ret = copied;
1707
1708 out_free:
1709         kfree(pm.buffer);
1710 out_mm:
1711         mmput(mm);
1712 out:
1713         return ret;
1714 }
1715
1716 static int pagemap_open(struct inode *inode, struct file *file)
1717 {
1718         struct mm_struct *mm;
1719
1720         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1721         if (IS_ERR(mm))
1722                 return PTR_ERR(mm);
1723         file->private_data = mm;
1724         return 0;
1725 }
1726
1727 static int pagemap_release(struct inode *inode, struct file *file)
1728 {
1729         struct mm_struct *mm = file->private_data;
1730
1731         if (mm)
1732                 mmdrop(mm);
1733         return 0;
1734 }
1735
1736 const struct file_operations proc_pagemap_operations = {
1737         .llseek         = mem_lseek, /* borrow this */
1738         .read           = pagemap_read,
1739         .open           = pagemap_open,
1740         .release        = pagemap_release,
1741 };
1742 #endif /* CONFIG_PROC_PAGE_MONITOR */
1743
1744 #ifdef CONFIG_NUMA
1745
1746 struct numa_maps {
1747         unsigned long pages;
1748         unsigned long anon;
1749         unsigned long active;
1750         unsigned long writeback;
1751         unsigned long mapcount_max;
1752         unsigned long dirty;
1753         unsigned long swapcache;
1754         unsigned long node[MAX_NUMNODES];
1755 };
1756
1757 struct numa_maps_private {
1758         struct proc_maps_private proc_maps;
1759         struct numa_maps md;
1760 };
1761
1762 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1763                         unsigned long nr_pages)
1764 {
1765         int count = page_mapcount(page);
1766
1767         md->pages += nr_pages;
1768         if (pte_dirty || PageDirty(page))
1769                 md->dirty += nr_pages;
1770
1771         if (PageSwapCache(page))
1772                 md->swapcache += nr_pages;
1773
1774         if (PageActive(page) || PageUnevictable(page))
1775                 md->active += nr_pages;
1776
1777         if (PageWriteback(page))
1778                 md->writeback += nr_pages;
1779
1780         if (PageAnon(page))
1781                 md->anon += nr_pages;
1782
1783         if (count > md->mapcount_max)
1784                 md->mapcount_max = count;
1785
1786         md->node[page_to_nid(page)] += nr_pages;
1787 }
1788
1789 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1790                 unsigned long addr)
1791 {
1792         struct page *page;
1793         int nid;
1794
1795         if (!pte_present(pte))
1796                 return NULL;
1797
1798         page = vm_normal_page(vma, addr, pte);
1799         if (!page || is_zone_device_page(page))
1800                 return NULL;
1801
1802         if (PageReserved(page))
1803                 return NULL;
1804
1805         nid = page_to_nid(page);
1806         if (!node_isset(nid, node_states[N_MEMORY]))
1807                 return NULL;
1808
1809         return page;
1810 }
1811
1812 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1813 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1814                                               struct vm_area_struct *vma,
1815                                               unsigned long addr)
1816 {
1817         struct page *page;
1818         int nid;
1819
1820         if (!pmd_present(pmd))
1821                 return NULL;
1822
1823         page = vm_normal_page_pmd(vma, addr, pmd);
1824         if (!page)
1825                 return NULL;
1826
1827         if (PageReserved(page))
1828                 return NULL;
1829
1830         nid = page_to_nid(page);
1831         if (!node_isset(nid, node_states[N_MEMORY]))
1832                 return NULL;
1833
1834         return page;
1835 }
1836 #endif
1837
1838 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1839                 unsigned long end, struct mm_walk *walk)
1840 {
1841         struct numa_maps *md = walk->private;
1842         struct vm_area_struct *vma = walk->vma;
1843         spinlock_t *ptl;
1844         pte_t *orig_pte;
1845         pte_t *pte;
1846
1847 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1848         ptl = pmd_trans_huge_lock(pmd, vma);
1849         if (ptl) {
1850                 struct page *page;
1851
1852                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1853                 if (page)
1854                         gather_stats(page, md, pmd_dirty(*pmd),
1855                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1856                 spin_unlock(ptl);
1857                 return 0;
1858         }
1859
1860         if (pmd_trans_unstable(pmd))
1861                 return 0;
1862 #endif
1863         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1864         do {
1865                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1866                 if (!page)
1867                         continue;
1868                 gather_stats(page, md, pte_dirty(*pte), 1);
1869
1870         } while (pte++, addr += PAGE_SIZE, addr != end);
1871         pte_unmap_unlock(orig_pte, ptl);
1872         cond_resched();
1873         return 0;
1874 }
1875 #ifdef CONFIG_HUGETLB_PAGE
1876 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1877                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1878 {
1879         pte_t huge_pte = huge_ptep_get(pte);
1880         struct numa_maps *md;
1881         struct page *page;
1882
1883         if (!pte_present(huge_pte))
1884                 return 0;
1885
1886         page = pte_page(huge_pte);
1887
1888         md = walk->private;
1889         gather_stats(page, md, pte_dirty(huge_pte), 1);
1890         return 0;
1891 }
1892
1893 #else
1894 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1895                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1896 {
1897         return 0;
1898 }
1899 #endif
1900
1901 static const struct mm_walk_ops show_numa_ops = {
1902         .hugetlb_entry = gather_hugetlb_stats,
1903         .pmd_entry = gather_pte_stats,
1904 };
1905
1906 /*
1907  * Display pages allocated per node and memory policy via /proc.
1908  */
1909 static int show_numa_map(struct seq_file *m, void *v)
1910 {
1911         struct numa_maps_private *numa_priv = m->private;
1912         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1913         struct vm_area_struct *vma = v;
1914         struct numa_maps *md = &numa_priv->md;
1915         struct file *file = vma->vm_file;
1916         struct mm_struct *mm = vma->vm_mm;
1917         struct mempolicy *pol;
1918         char buffer[64];
1919         int nid;
1920
1921         if (!mm)
1922                 return 0;
1923
1924         /* Ensure we start with an empty set of numa_maps statistics. */
1925         memset(md, 0, sizeof(*md));
1926
1927         pol = __get_vma_policy(vma, vma->vm_start);
1928         if (pol) {
1929                 mpol_to_str(buffer, sizeof(buffer), pol);
1930                 mpol_cond_put(pol);
1931         } else {
1932                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1933         }
1934
1935         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1936
1937         if (file) {
1938                 seq_puts(m, " file=");
1939                 seq_file_path(m, file, "\n\t= ");
1940         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1941                 seq_puts(m, " heap");
1942         } else if (is_stack(vma)) {
1943                 seq_puts(m, " stack");
1944         }
1945
1946         if (is_vm_hugetlb_page(vma))
1947                 seq_puts(m, " huge");
1948
1949         /* mmap_lock is held by m_start */
1950         walk_page_vma(vma, &show_numa_ops, md);
1951
1952         if (!md->pages)
1953                 goto out;
1954
1955         if (md->anon)
1956                 seq_printf(m, " anon=%lu", md->anon);
1957
1958         if (md->dirty)
1959                 seq_printf(m, " dirty=%lu", md->dirty);
1960
1961         if (md->pages != md->anon && md->pages != md->dirty)
1962                 seq_printf(m, " mapped=%lu", md->pages);
1963
1964         if (md->mapcount_max > 1)
1965                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1966
1967         if (md->swapcache)
1968                 seq_printf(m, " swapcache=%lu", md->swapcache);
1969
1970         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1971                 seq_printf(m, " active=%lu", md->active);
1972
1973         if (md->writeback)
1974                 seq_printf(m, " writeback=%lu", md->writeback);
1975
1976         for_each_node_state(nid, N_MEMORY)
1977                 if (md->node[nid])
1978                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1979
1980         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1981 out:
1982         seq_putc(m, '\n');
1983         return 0;
1984 }
1985
1986 static const struct seq_operations proc_pid_numa_maps_op = {
1987         .start  = m_start,
1988         .next   = m_next,
1989         .stop   = m_stop,
1990         .show   = show_numa_map,
1991 };
1992
1993 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1994 {
1995         return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
1996                                 sizeof(struct numa_maps_private));
1997 }
1998
1999 const struct file_operations proc_pid_numa_maps_operations = {
2000         .open           = pid_numa_maps_open,
2001         .read           = seq_read,
2002         .llseek         = seq_lseek,
2003         .release        = proc_map_release,
2004 };
2005
2006 #endif /* CONFIG_NUMA */